d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
4177525533
freebsd-dev/sys/dev/isp/isp.c

7011 lines
187 KiB
C
Raw Normal View History

Start each of the license/copyright comments with /*-, minor shuffle of lines
2005-01-06 01:43:34 +00:00
/*-
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
* Copyright (c) 1997-2006 by Matthew Jacob
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
some copyright cleanups
2000-09-21 20:16:04 +00:00
* 2. The name of the author may not be used to endorse or promote products
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
/*
* Machine and OS Independent (well, as best as possible)
* code for the Qlogic ISP SCSI adapters.
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Inspiration and ideas about this driver are from Erik Moe's Linux driver
* (qlogicisp.c) and Dave Miller's SBus version of same (qlogicisp.c). Some
* ideas dredged from the Solaris driver.
*/
/*
* Include header file appropriate for platform we're building on.
*/
#ifdef __NetBSD__
#include <dev/ic/isp_netbsd.h>
#endif
#ifdef __FreeBSD__
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
#include <dev/isp/isp_freebsd.h>
#endif
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
#ifdef __OpenBSD__
#include <dev/ic/isp_openbsd.h>
#endif
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
#ifdef __linux__
clarify headers;move uninit to outer layer;remove watchdog
1998-12-28 19:22:27 +00:00
#include "isp_linux.h"
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
#endif
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
#ifdef __svr4__
#include "isp_solaris.h"
#endif
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* General defines
*/
#define MBOX_DELAY_COUNT 1000000 / 100
/*
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* Local static data
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char portshift[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Target %d Loop ID 0x%x (Port 0x%x) => Loop 0x%x (Port 0x%x)";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char portdup[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Target %d duplicates Target %d- killing off both";
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
static const char retained[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Retaining Loop ID 0x%x for Target %d (Port 0x%x)";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char lretained[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Retained login of Target %d (Loop ID 0x%x) Port 0x%x";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char plogout[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Logging out Target %d at Loop ID 0x%x (Port 0x%x)";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char plogierr[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Command Error in PLOGI for Port 0x%x (0x%x)";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char nopdb[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Could not get PDB for Device @ Port 0x%x";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char pdbmfail1[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"PDB Loop ID info for Device @ Port 0x%x does not match up (0x%x)";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char pdbmfail2[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"PDB Port info for Device @ Port 0x%x does not match up (0x%x)";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char ldumped[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Target %d (Loop ID 0x%x) Port 0x%x dumped after login info mismatch";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char notresp[] =
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
"Not RESPONSE in RESPONSE Queue (type 0x%x) @ idx %d (next %d) nlooked %d";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char xact1[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"HBA attempted queued transaction with disconnect not set for %d.%d.%d";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char xact2[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"HBA attempted queued transaction to target routine %d on target %d bus %d";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char xact3[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"HBA attempted queued cmd for %d.%d.%d when queueing disabled";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char pskip[] =
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"SCSI phase skipped for target %d.%d.%d";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char topology[] =
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
"Loop ID %d, Port ID 0x%x, Loop State 0x%x, Topology '%s'";
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
static const char swrej[] =
"Fabric Nameserver rejected %s (Reason=0x%x Expl=0x%x) for Port ID 0x%x";
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char finmsg[] =
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
"%d.%d.%d: FIN dl%d resid %d STS 0x%x SKEY %c XS_ERR=0x%x";
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
static const char sc0[] =
"%s CHAN %d FTHRSH %d IID %d RESETD %d RETRYC %d RETRYD %d ASD 0x%x";
static const char sc1[] =
"%s RAAN 0x%x DLAN 0x%x DDMAB 0x%x CDMAB 0x%x SELTIME %d MQD %d";
static const char sc2[] = "%s CHAN %d TGT %d FLAGS 0x%x 0x%x/0x%x";
static const char sc3[] = "Generated";
static const char sc4[] = "NVRAM";
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
static const char bun[] =
"bad underrun for %d.%d (count %d, resid %d, status %s)";
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* Local function prototypes.
*/
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
static int isp_parse_async(ispsoftc_t *, uint16_t);
static int isp_handle_other_response(ispsoftc_t *, int, isphdr_t *,
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_status(ispsoftc_t *, ispstatusreq_t *, XS_T *);
static void isp_fastpost_complete(ispsoftc_t *, uint16_t);
static int isp_mbox_continue(ispsoftc_t *);
static void isp_scsi_init(ispsoftc_t *);
static void isp_scsi_channel_init(ispsoftc_t *, int);
static void isp_fibre_init(ispsoftc_t *);
static void isp_mark_getpdb_all(ispsoftc_t *);
static int isp_getmap(ispsoftc_t *, fcpos_map_t *);
static int isp_getpdb(ispsoftc_t *, int, isp_pdb_t *);
static uint64_t isp_get_portname(ispsoftc_t *, int, int);
static int isp_fclink_test(ispsoftc_t *, int);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
static const char *isp2100_fw_statename(int);
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
static int isp_pdb_sync(ispsoftc_t *);
static int isp_scan_loop(ispsoftc_t *);
static int isp_fabric_mbox_cmd(ispsoftc_t *, mbreg_t *);
static int isp_scan_fabric(ispsoftc_t *, int);
static void isp_register_fc4_type(ispsoftc_t *);
static void isp_fw_state(ispsoftc_t *);
static void isp_mboxcmd_qnw(ispsoftc_t *, mbreg_t *, int);
static void isp_mboxcmd(ispsoftc_t *, mbreg_t *, int);
static void isp_update(ispsoftc_t *);
static void isp_update_bus(ispsoftc_t *, int);
static void isp_setdfltparm(ispsoftc_t *, int);
static int isp_read_nvram(ispsoftc_t *);
static void isp_rdnvram_word(ispsoftc_t *, int, uint16_t *);
static void isp_parse_nvram_1020(ispsoftc_t *, uint8_t *);
static void isp_parse_nvram_1080(ispsoftc_t *, int, uint8_t *);
static void isp_parse_nvram_12160(ispsoftc_t *, int, uint8_t *);
static void isp_parse_nvram_2100(ispsoftc_t *, uint8_t *);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Reset Hardware.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*
Use pointer to f/w in md structure as to whether f/w exists or not. If firmware length isn't specified, extract from the 4th short into the firmware.
1999-10-30 19:32:44 +00:00
* Hit the chip over the head, download new f/w if available and set it running.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*
* Locking done elsewhere.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_reset(ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
mbreg_t mbs;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t code_org;
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
int loops, i, dodnld = 1;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
char *btype = "????";
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_state = ISP_NILSTATE;
/*
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* Basic types (SCSI, FibreChannel and PCI or SBus)
* have been set in the MD code. We figure out more
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
* here. Possibly more refined types based upon PCI
* identification. Chip revision has been gathered.
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
*
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
* After we've fired this chip up, zero out the conf1 register
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
* for SCSI adapters and do other settings for the 2100.
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
*/
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
* Get the current running firmware revision out of the
* chip before we hit it over the head (if this is our
* first time through). Note that we store this as the
* 'ROM' firmware revision- which it may not be. In any
* case, we don't really use this yet, but we may in
* the future.
*/
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (isp->isp_touched == 0) {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
* First see whether or not we're sitting in the ISP PROM.
* If we've just been reset, we'll have the string "ISP "
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
* spread through outgoing mailbox registers 1-3. We do
* this for PCI cards because otherwise we really don't
* know what state the card is in and we could hang if
* we try this command otherwise.
*
* For SBus cards, we just do this because they almost
* certainly will be running firmware by now.
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (ISP_READ(isp, OUTMAILBOX1) != 0x4953 ||
ISP_READ(isp, OUTMAILBOX2) != 0x5020 ||
ISP_READ(isp, OUTMAILBOX3) != 0x2020) {
/*
* Just in case it was paused...
*/
ISP_WRITE(isp, HCCR, HCCR_CMD_RELEASE);
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
mbs.param[0] = MBOX_ABOUT_FIRMWARE;
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
isp->isp_romfw_rev[0] = mbs.param[1];
isp->isp_romfw_rev[1] = mbs.param[2];
isp->isp_romfw_rev[2] = mbs.param[3];
}
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
}
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
isp->isp_touched = 1;
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
}
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
DISABLE_INTS(isp);
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
/*
* Set up default request/response queue in-pointer/out-pointer
* register indices.
*/
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_23XX(isp)) {
When calling isp_reset, set the request/response in/out pointers all at once so there isn't a window with the ones for the 23XX cards being wrong. When being verbose, print out some more FC NVRAM values (like framesize). MFC after: 1 week
2001-09-29 19:37:49 +00:00
isp->isp_rqstinrp = BIU_REQINP;
isp->isp_rqstoutrp = BIU_REQOUTP;
isp->isp_respinrp = BIU_RSPINP;
isp->isp_respoutrp = BIU_RSPOUTP;
Initialize 2300 request/response pointers in isp_reset- not in isp_fibre_init.
2006-08-04 20:14:52 +00:00
ISP_WRITE(isp, isp->isp_rqstinrp, 0);
ISP_WRITE(isp, isp->isp_rqstoutrp, 0);
ISP_WRITE(isp, isp->isp_respinrp, 0);
ISP_WRITE(isp, isp->isp_respoutrp, 0);
When calling isp_reset, set the request/response in/out pointers all at once so there isn't a window with the ones for the 23XX cards being wrong. When being verbose, print out some more FC NVRAM values (like framesize). MFC after: 1 week
2001-09-29 19:37:49 +00:00
} else {
isp->isp_rqstinrp = INMAILBOX4;
isp->isp_rqstoutrp = OUTMAILBOX4;
isp->isp_respinrp = OUTMAILBOX5;
isp->isp_respoutrp = INMAILBOX5;
}
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
* Put the board into PAUSE mode (so we can read the SXP registers
* or write FPM/FBM registers).
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
*/
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_FC(isp)) {
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
switch (isp->isp_type) {
case ISP_HA_FC_2100:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "2100";
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
break;
case ISP_HA_FC_2200:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "2200";
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
break;
Spring MegaChange #1. ---- Make a device for each ISP- really usable only with devfs and add an ioctl entry point (this can be used to (re)set debug levels, reset the HBA, rescan the fabric, issue lips, etc). ---- Add in a kernel thread for Fibre Channel cards. The purpose of this thread is to be woken up to clean up after Fibre Channel events block things. Basically, any FC event that casts doubt on the location or identify of FC devices blocks the queues. When, and if, we get the PORT DATABASE CHANGED or NAME SERVER DATABASE CHANGED async event, we activate the kthread which will then, in full thread context, re-evaluate the local loop and/or the fabric. When it's satisfied that things are stable, it can then release the blocked queues and let commands flow again. The prior mechanism was a lazy evaluation. That is, the next command to come down the pipe after change events would pay the full price for re-evaluation. And if this was done off of a softcall, it really could hang up the system. These changes brings the FreeBSD port more in line with the Solaris, Linux and NetBSD ports. It also, more importantly, gets us being more proactive about topology changes which could then be reflected upwards to CAM so that the periph driver can be informed sooner rather than later when things arrive or depart. --- Add in the (correct) usage of locking macros- we now have lock transition macros which allow us to transition from holding the CAM lock (Giant) and grabbing the softc lock and vice versa. Switch over to having this HBA do real locking. Some folks claim this won't be a win. They're right. But you have to start somewhere, and this will begin to teach us how to DTRT for HBAs, etc. -- Start putting in prototype 2300 support. Add back in LIP and Loop Reset as async events that each platform will handle. Add in another int_bogus instrumentation point. Do some more substantial target mode cleanups. MFC after: 8 weeks
2001-05-28 21:20:43 +00:00
case ISP_HA_FC_2300:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "2300";
Spring MegaChange #1. ---- Make a device for each ISP- really usable only with devfs and add an ioctl entry point (this can be used to (re)set debug levels, reset the HBA, rescan the fabric, issue lips, etc). ---- Add in a kernel thread for Fibre Channel cards. The purpose of this thread is to be woken up to clean up after Fibre Channel events block things. Basically, any FC event that casts doubt on the location or identify of FC devices blocks the queues. When, and if, we get the PORT DATABASE CHANGED or NAME SERVER DATABASE CHANGED async event, we activate the kthread which will then, in full thread context, re-evaluate the local loop and/or the fabric. When it's satisfied that things are stable, it can then release the blocked queues and let commands flow again. The prior mechanism was a lazy evaluation. That is, the next command to come down the pipe after change events would pay the full price for re-evaluation. And if this was done off of a softcall, it really could hang up the system. These changes brings the FreeBSD port more in line with the Solaris, Linux and NetBSD ports. It also, more importantly, gets us being more proactive about topology changes which could then be reflected upwards to CAM so that the periph driver can be informed sooner rather than later when things arrive or depart. --- Add in the (correct) usage of locking macros- we now have lock transition macros which allow us to transition from holding the CAM lock (Giant) and grabbing the softc lock and vice versa. Switch over to having this HBA do real locking. Some folks claim this won't be a win. They're right. But you have to start somewhere, and this will begin to teach us how to DTRT for HBAs, etc. -- Start putting in prototype 2300 support. Add back in LIP and Loop Reset as async events that each platform will handle. Add in another int_bogus instrumentation point. Do some more substantial target mode cleanups. MFC after: 8 weeks
2001-05-28 21:20:43 +00:00
break;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
case ISP_HA_FC_2312:
btype = "2312";
break;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
case ISP_HA_FC_2322:
btype = "2322";
break;
case ISP_HA_FC_2422:
btype = "2422";
break;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
default:
break;
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* While we're paused, reset the FPM module and FBM fifos.
*/
ISP_WRITE(isp, BIU2100_CSR, BIU2100_FPM0_REGS);
ISP_WRITE(isp, FPM_DIAG_CONFIG, FPM_SOFT_RESET);
ISP_WRITE(isp, BIU2100_CSR, BIU2100_FB_REGS);
ISP_WRITE(isp, FBM_CMD, FBMCMD_FIFO_RESET_ALL);
ISP_WRITE(isp, BIU2100_CSR, BIU2100_RISC_REGS);
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
} else if (IS_1240(isp)) {
sdparam *sdp = isp->isp_param;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1240";
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 60;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
sdp->isp_ultramode = 1;
sdp++;
sdp->isp_ultramode = 1;
/*
* XXX: Should probably do some bus sensing.
*/
} else if (IS_ULTRA2(isp)) {
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
static const char m[] = "bus %d is in %s Mode";
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t l;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
sdparam *sdp = isp->isp_param;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 100;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
if (IS_1280(isp))
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1280";
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
else if (IS_1080(isp))
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1080";
This should enable 10160 support. As best as I can tell, the same f/w as 12160 is used, and otherwise, this is just a single channel variant of the 10160. MFC after: 0 days
2002-10-11 17:28:01 +00:00
else if (IS_10160(isp))
btype = "10160";
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
else if (IS_12160(isp))
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "12160";
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
else
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "<UNKLVD>";
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
l = ISP_READ(isp, SXP_PINS_DIFF) & ISP1080_MODE_MASK;
switch (l) {
case ISP1080_LVD_MODE:
sdp->isp_lvdmode = 1;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, m, 0, "LVD");
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
case ISP1080_HVD_MODE:
sdp->isp_diffmode = 1;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, m, 0, "Differential");
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
case ISP1080_SE_MODE:
sdp->isp_ultramode = 1;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, m, 0, "Single-Ended");
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"unknown mode on bus %d (0x%x)", 0, l);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
}
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
if (IS_DUALBUS(isp)) {
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
sdp++;
l = ISP_READ(isp, SXP_PINS_DIFF|SXP_BANK1_SELECT);
l &= ISP1080_MODE_MASK;
switch(l) {
case ISP1080_LVD_MODE:
sdp->isp_lvdmode = 1;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, m, 1, "LVD");
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
break;
case ISP1080_HVD_MODE:
sdp->isp_diffmode = 1;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG,
m, 1, "Differential");
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
break;
case ISP1080_SE_MODE:
sdp->isp_ultramode = 1;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG,
m, 1, "Single-Ended");
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
break;
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"unknown mode on bus %d (0x%x)", 1, l);
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
break;
}
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
} else {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
sdparam *sdp = isp->isp_param;
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
i = ISP_READ(isp, BIU_CONF0) & BIU_CONF0_HW_MASK;
switch (i) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGALL, "Unknown Chip Type 0x%x", i);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/* FALLTHROUGH */
case 1:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1020";
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_type = ISP_HA_SCSI_1020;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 40;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case 2:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* Some 1020A chips are Ultra Capable, but don't
* run the clock rate up for that unless told to
* do so by the Ultra Capable bits being set.
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1020A";
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
isp->isp_type = ISP_HA_SCSI_1020A;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 40;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case 3:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1040";
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
isp->isp_type = ISP_HA_SCSI_1040;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 60;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case 4:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1040A";
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_type = ISP_HA_SCSI_1040A;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 60;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case 5:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1040B";
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_type = ISP_HA_SCSI_1040B;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 60;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
case 6:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
btype = "1040C";
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp->isp_type = ISP_HA_SCSI_1040C;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 60;
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
* Now, while we're at it, gather info about ultra
* and/or differential mode.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
if (ISP_READ(isp, SXP_PINS_DIFF) & SXP_PINS_DIFF_MODE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, "Differential Mode");
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
sdp->isp_diffmode = 1;
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
} else {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
sdp->isp_diffmode = 0;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
i = ISP_READ(isp, RISC_PSR);
if (isp->isp_bustype == ISP_BT_SBUS) {
i &= RISC_PSR_SBUS_ULTRA;
} else {
i &= RISC_PSR_PCI_ULTRA;
}
if (i != 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, "Ultra Mode Capable");
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
sdp->isp_ultramode = 1;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
Revert previous commit. Violates Maintainer (O'Brien knows how to reach me directly), but more importantly, breaks compiles on non-FreeBSD platforms.
2003-08-25 17:58:23 +00:00
* If we're in Ultra Mode, we have to be 60MHz clock-
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
* even for the SBus version.
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_clock = 60;
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
} else {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
sdp->isp_ultramode = 0;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
* Clock is known. Gronk.
*/
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* Machine dependent clock (if set) overrides
* our generic determinations.
*/
if (isp->isp_mdvec->dv_clock) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (isp->isp_mdvec->dv_clock < isp->isp_clock) {
isp->isp_clock = isp->isp_mdvec->dv_clock;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
/*
* Clear instrumentation
*/
isp->isp_intcnt = isp->isp_intbogus = 0;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Do MD specific pre initialization
*/
ISP_RESET0(isp);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
again:
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Hit the chip over the head with hammer,
* and give the ISP a chance to recover.
*/
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_SCSI(isp)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_WRITE(isp, BIU_ICR, BIU_ICR_SOFT_RESET);
/*
* A slight delay...
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
USEC_DELAY(100);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Clear data && control DMA engines.
*/
ISP_WRITE(isp, CDMA_CONTROL,
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
DMA_CNTRL_CLEAR_CHAN | DMA_CNTRL_RESET_INT);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_WRITE(isp, DDMA_CONTROL,
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
DMA_CNTRL_CLEAR_CHAN | DMA_CNTRL_RESET_INT);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
} else {
ISP_WRITE(isp, BIU2100_CSR, BIU2100_SOFT_RESET);
/*
* A slight delay...
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
USEC_DELAY(100);
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
* Clear data && control DMA engines.
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_WRITE(isp, CDMA2100_CONTROL,
DMA_CNTRL2100_CLEAR_CHAN | DMA_CNTRL2100_RESET_INT);
ISP_WRITE(isp, TDMA2100_CONTROL,
DMA_CNTRL2100_CLEAR_CHAN | DMA_CNTRL2100_RESET_INT);
ISP_WRITE(isp, RDMA2100_CONTROL,
DMA_CNTRL2100_CLEAR_CHAN | DMA_CNTRL2100_RESET_INT);
}
/*
* Wait for ISP to be ready to go...
*/
loops = MBOX_DELAY_COUNT;
for (;;) {
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_SCSI(isp)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (!(ISP_READ(isp, BIU_ICR) & BIU_ICR_SOFT_RESET))
break;
} else {
if (!(ISP_READ(isp, BIU2100_CSR) & BIU2100_SOFT_RESET))
break;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
USEC_DELAY(100);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (--loops < 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
ISP_DUMPREGS(isp, "chip reset timed out");
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
}
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
* After we've fired this chip up, zero out the conf1 register
* for SCSI adapters and other settings for the 2100.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_SCSI(isp)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_WRITE(isp, BIU_CONF1, 0);
} else {
ISP_WRITE(isp, BIU2100_CSR, 0);
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
* Reset RISC Processor
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_WRITE(isp, HCCR, HCCR_CMD_RESET);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
USEC_DELAY(100);
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/* Clear semaphore register (just to be sure) */
ISP_WRITE(isp, BIU_SEMA, 0);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
/*
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
* Establish some initial burst rate stuff.
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
* (only for the 1XX0 boards). This really should
* be done later after fetching from NVRAM.
*/
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_SCSI(isp)) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t tmp = isp->isp_mdvec->dv_conf1;
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
/*
* Busted FIFO. Turn off all but burst enables.
*/
if (isp->isp_type == ISP_HA_SCSI_1040A) {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
tmp &= BIU_BURST_ENABLE;
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
ISP_SETBITS(isp, BIU_CONF1, tmp);
if (tmp & BIU_BURST_ENABLE) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_SETBITS(isp, CDMA_CONF, DMA_ENABLE_BURST);
ISP_SETBITS(isp, DDMA_CONF, DMA_ENABLE_BURST);
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
#ifdef PTI_CARDS
if (((sdparam *) isp->isp_param)->isp_ultramode) {
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
while (ISP_READ(isp, RISC_MTR) != 0x1313) {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
ISP_WRITE(isp, RISC_MTR, 0x1313);
ISP_WRITE(isp, HCCR, HCCR_CMD_STEP);
}
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
} else {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
ISP_WRITE(isp, RISC_MTR, 0x1212);
}
/*
* PTI specific register
*/
ISP_WRITE(isp, RISC_EMB, DUAL_BANK)
#else
ISP_WRITE(isp, RISC_MTR, 0x1212);
#endif
} else {
ISP_WRITE(isp, RISC_MTR2100, 0x1212);
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_2200(isp) || IS_23XX(isp)) {
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
ISP_WRITE(isp, HCCR, HCCR_2X00_DISABLE_PARITY_PAUSE);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
ISP_WRITE(isp, HCCR, HCCR_CMD_RELEASE); /* release paused processor */
/*
* Do MD specific post initialization
*/
ISP_RESET1(isp);
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
Misunderstanding documentation caused me to try and set 1Gbps/2Gps/Auto connection speed for the 2300 in the wrong offset in the ICB. Oops. Respect some QLogic errat wrt PCI errors on certain shared host/RISC registers.
2001-10-06 20:41:18 +00:00
* Wait for everything to finish firing up.
*
* Avoid doing this on the 2312 because you can generate a PCI
* parity error (chip breakage).
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
*/
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_23XX(isp)) {
Misunderstanding documentation caused me to try and set 1Gbps/2Gps/Auto connection speed for the 2300 in the wrong offset in the ICB. Oops. Respect some QLogic errat wrt PCI errors on certain shared host/RISC registers.
2001-10-06 20:41:18 +00:00
USEC_DELAY(5);
} else {
loops = MBOX_DELAY_COUNT;
while (ISP_READ(isp, OUTMAILBOX0) == MBOX_BUSY) {
USEC_DELAY(100);
if (--loops < 0) {
isp_prt(isp, ISP_LOGERR,
"MBOX_BUSY never cleared on reset");
return;
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
}
}
/*
* Up until this point we've done everything by just reading or
* setting registers. From this point on we rely on at least *some*
* kind of firmware running in the card.
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Do some sanity checking.
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_NO_OP;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
if (IS_SCSI(isp)) {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_MAILBOX_REG_TEST;
mbs.param[1] = 0xdead;
mbs.param[2] = 0xbeef;
mbs.param[3] = 0xffff;
mbs.param[4] = 0x1111;
mbs.param[5] = 0xa5a5;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
if (mbs.param[1] != 0xdead || mbs.param[2] != 0xbeef ||
mbs.param[3] != 0xffff || mbs.param[4] != 0x1111 ||
mbs.param[5] != 0xa5a5) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"Register Test Failed (0x%x 0x%x 0x%x 0x%x 0x%x)",
mbs.param[1], mbs.param[2], mbs.param[3],
mbs.param[4], mbs.param[5]);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
}
}
/*
* Download new Firmware, unless requested not to do so.
* This is made slightly trickier in some cases where the
* firmware of the ROM revision is newer than the revision
* compiled into the driver. So, where we used to compare
* versions of our f/w and the ROM f/w, now we just see
* whether we have f/w at all and whether a config flag
* has disabled our download.
*/
correct moronic typo
1999-11-01 04:39:52 +00:00
if ((isp->isp_mdvec->dv_ispfw == NULL) ||
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
(isp->isp_confopts & ISP_CFG_NORELOAD)) {
dodnld = 0;
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_23XX(isp)) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
code_org = ISP_CODE_ORG_2300;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
} else {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
code_org = ISP_CODE_ORG;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
correct moronic typo
1999-11-01 04:39:52 +00:00
if (dodnld) {
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
uint16_t *ptr = isp->isp_mdvec->dv_ispfw;
isp->isp_mbxworkp = &ptr[1];
isp->isp_mbxwrk0 = ptr[3] - 1;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_mbxwrk1 = code_org + 1;
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
mbs.param[0] = MBOX_WRITE_RAM_WORD;
mbs.param[1] = code_org;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
mbs.param[2] = ptr[0];
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGERR,
"F/W download failed at word %d",
isp->isp_mbxwrk1 - code_org);
dodnld = 0;
goto again;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
* Verify that it downloaded correctly.
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
mbs.param[0] = MBOX_VERIFY_CHECKSUM;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[1] = code_org;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Ram Checksum Failure");
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
return;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
/*
* If we're a 2322 or 2422, the firmware actually comes
* in three chunks. We loaded the first at the code_org
* address. The other two chunks, which follow right
* after each other in memory here, get loaded at addresses
* specfied at offset 0x9..0xB.
*/
if (IS_2322(isp)) {
ptr = &ptr[ptr[3]];
isp->isp_mbxworkp = &ptr[1];
isp->isp_mbxwrk0 = ptr[3] - 1;
isp->isp_mbxwrk1 = ptr[5] + 1;
isp->isp_mbxwrk8 = ptr[4];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
mbs.param[0] = MBOX_WRITE_RAM_WORD_EXTENDED;
mbs.param[1] = ptr[5];
mbs.param[2] = ptr[0];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
mbs.param[8] = ptr[4];
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGERR,
"Receive Sequencer F/W Load Failed");
return;
}
ptr = &ptr[ptr[3]];
isp->isp_mbxworkp = &ptr[1];
isp->isp_mbxwrk0 = ptr[3] - 1;
isp->isp_mbxwrk1 = ptr[5] + 1;
isp->isp_mbxwrk8 = ptr[4];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
mbs.param[0] = MBOX_WRITE_RAM_WORD_EXTENDED;
mbs.param[1] = ptr[5];
mbs.param[2] = ptr[0];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
mbs.param[8] = ptr[4];
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGERR,
"Transmit Sequencer F/W Load Failed");
return;
}
}
various fixes
2000-08-27 23:38:44 +00:00
isp->isp_loaded_fw = 1;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
} else {
various fixes
2000-08-27 23:38:44 +00:00
isp->isp_loaded_fw = 0;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGDEBUG2, "skipping f/w download");
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
/*
* Now start it rolling.
*
* If we didn't actually download f/w,
* we still need to (re)start it.
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_EXEC_FIRMWARE;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[1] = code_org;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2322(isp) || IS_24XX(isp)) {
if (isp->isp_loaded_fw) {
mbs.param[2] = 0;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
} else {
mbs.param[2] = 1;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
}
mbs.obits |= 2;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
/*
* Give it a chance to start.
*/
USEC_DELAY(500);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_SCSI(isp)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* Set CLOCK RATE, but only if asked to.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (isp->isp_clock) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_SET_CLOCK_RATE;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[1] = isp->isp_clock;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
/* we will try not to care if this fails */
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_ABOUT_FIRMWARE;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
/*
* The SBus firmware that we are using apparently does not return
* major, minor, micro revisions in the mailbox registers, which
* is really, really, annoying.
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (ISP_SBUS_SUPPORTED && isp->isp_bustype == ISP_BT_SBUS) {
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
if (dodnld) {
#ifdef ISP_TARGET_MODE
isp->isp_fwrev[0] = 7;
isp->isp_fwrev[1] = 55;
#else
isp->isp_fwrev[0] = 1;
isp->isp_fwrev[1] = 37;
#endif
isp->isp_fwrev[2] = 0;
}
} else {
isp->isp_fwrev[0] = mbs.param[1];
isp->isp_fwrev[1] = mbs.param[2];
isp->isp_fwrev[2] = mbs.param[3];
}
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG,
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
"Board Type %s, Chip Revision 0x%x, %s F/W Revision %d.%d.%d",
btype, isp->isp_revision, dodnld? "loaded" : "resident",
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
isp->isp_fwrev[0], isp->isp_fwrev[1], isp->isp_fwrev[2]);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
if (IS_FC(isp)) {
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
/*
* We do not believe firmware attributes for 2100 code less
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
* than 1.17.0, unless it's the firmware we specifically
* are loading.
*
* Note that all 22XX and 23XX f/w is greater than 1.X.0.
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
*/
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
if (!(ISP_FW_NEWER_THAN(isp, 1, 17, 0))) {
#ifdef USE_SMALLER_2100_FIRMWARE
FCPARAM(isp)->isp_fwattr = ISP_FW_ATTR_SCCLUN;
#else
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
FCPARAM(isp)->isp_fwattr = 0;
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
#endif
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
} else {
FCPARAM(isp)->isp_fwattr = mbs.param[6];
isp_prt(isp, ISP_LOGDEBUG0,
"Firmware Attributes = 0x%x", mbs.param[6]);
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
isp_prt(isp, ISP_LOGCONFIG, "2K Logins Supported");
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
}
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
if (isp->isp_romfw_rev[0] || isp->isp_romfw_rev[1] ||
isp->isp_romfw_rev[2]) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGCONFIG, "Last F/W revision was %d.%d.%d",
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp->isp_romfw_rev[0], isp->isp_romfw_rev[1],
isp->isp_romfw_rev[2]);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
mbs.param[0] = MBOX_GET_FIRMWARE_STATUS;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
isp->isp_maxcmds = mbs.param[2];
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO,
"%d max I/O commands supported", mbs.param[2]);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp_fw_state(isp);
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/*
* Set up DMA for the request and result mailboxes.
*/
if (ISP_MBOXDMASETUP(isp) != 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Cannot setup DMA");
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
return;
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_state = ISP_RESETSTATE;
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
/*
* Okay- now that we have new firmware running, we now (re)set our
* notion of how many luns we support. This is somewhat tricky because
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
* if we haven't loaded firmware, we sometimes do not have an easy way
* of knowing how many luns we support.
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
*
* Expanded lun firmware gives you 32 luns for SCSI cards and
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
* 16384 luns for Fibre Channel cards.
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
*
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
* It turns out that even for QLogic 2100s with ROM 1.10 and above
* we do get a firmware attributes word returned in mailbox register 6.
*
Correct typos, mostly s/ a / an / where appropriate. Some whitespace cleanup, especially in troff files.
2003-01-01 18:49:04 +00:00
* Because the lun is in a different position in the Request Queue
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
* Entry structure for Fibre Channel with expanded lun firmware, we
* can only support one lun (lun zero) when we don't know what kind
* of firmware we're running.
*/
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (IS_SCSI(isp)) {
if (dodnld) {
if (IS_ULTRA2(isp) || IS_ULTRA3(isp)) {
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
isp->isp_maxluns = 32;
} else {
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
isp->isp_maxluns = 8;
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
}
} else {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
isp->isp_maxluns = 8;
}
} else {
if (FCPARAM(isp)->isp_fwattr & ISP_FW_ATTR_SCCLUN) {
isp->isp_maxluns = 16384;
} else {
isp->isp_maxluns = 16;
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
}
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
/*
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
* Initialize Parameters of Hardware to a known state.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*
* Locks are held before coming here.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_init(ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Must do this first to get defaults established.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp_setdfltparm(isp, 0);
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if (IS_DUALBUS(isp)) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp_setdfltparm(isp, 1);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
if (IS_FC(isp)) {
isp_fibre_init(isp);
} else {
isp_scsi_init(isp);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_scsi_init(ispsoftc_t *isp)
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
{
sdparam *sdp_chan0, *sdp_chan1;
mbreg_t mbs;
sdp_chan0 = isp->isp_param;
sdp_chan1 = sdp_chan0;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if (IS_DUALBUS(isp)) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
sdp_chan1++;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* If we have no role (neither target nor initiator), return.
*/
if (isp->isp_role == ISP_ROLE_NONE) {
return;
}
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/* First do overall per-card settings. */
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
/*
* If we have fast memory timing enabled, turn it on.
*/
various fixes
2000-08-27 23:38:44 +00:00
if (sdp_chan0->isp_fast_mttr) {
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
ISP_WRITE(isp, RISC_MTR, 0x1313);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Set Retry Delay and Count.
* You set both channels at the same time.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[0] = MBOX_SET_RETRY_COUNT;
mbs.param[1] = sdp_chan0->isp_retry_count;
mbs.param[2] = sdp_chan0->isp_retry_delay;
mbs.param[6] = sdp_chan1->isp_retry_count;
mbs.param[7] = sdp_chan1->isp_retry_delay;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Set ASYNC DATA SETUP time. This is very important.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[0] = MBOX_SET_ASYNC_DATA_SETUP_TIME;
mbs.param[1] = sdp_chan0->isp_async_data_setup;
mbs.param[2] = sdp_chan1->isp_async_data_setup;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Set ACTIVE Negation State.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[0] = MBOX_SET_ACT_NEG_STATE;
mbs.param[1] =
(sdp_chan0->isp_req_ack_active_neg << 4) |
(sdp_chan0->isp_data_line_active_neg << 5);
mbs.param[2] =
(sdp_chan1->isp_req_ack_active_neg << 4) |
(sdp_chan1->isp_data_line_active_neg << 5);
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"failed to set active negation state (%d,%d), (%d,%d)",
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
sdp_chan0->isp_req_ack_active_neg,
sdp_chan0->isp_data_line_active_neg,
sdp_chan1->isp_req_ack_active_neg,
sdp_chan1->isp_data_line_active_neg);
/*
* But don't return.
*/
}
/*
* Set the Tag Aging limit
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[0] = MBOX_SET_TAG_AGE_LIMIT;
mbs.param[1] = sdp_chan0->isp_tag_aging;
mbs.param[2] = sdp_chan1->isp_tag_aging;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "failed to set tag age limit (%d,%d)",
sdp_chan0->isp_tag_aging, sdp_chan1->isp_tag_aging);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Set selection timeout.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[0] = MBOX_SET_SELECT_TIMEOUT;
mbs.param[1] = sdp_chan0->isp_selection_timeout;
mbs.param[2] = sdp_chan1->isp_selection_timeout;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/* now do per-channel settings */
isp_scsi_channel_init(isp, 0);
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if (IS_DUALBUS(isp))
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp_scsi_channel_init(isp, 1);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Now enable request/response queues
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
If we have a 1240 or an ULTRA2 or better card, use MBOX_INIT_RES_QUEUE_A64 (preparation for DAC/A64 support)
2002-09-23 04:59:42 +00:00
if (IS_ULTRA2(isp) || IS_1240(isp)) {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
If we have a 1240 or an ULTRA2 or better card, use MBOX_INIT_RES_QUEUE_A64 (preparation for DAC/A64 support)
2002-09-23 04:59:42 +00:00
mbs.param[0] = MBOX_INIT_RES_QUEUE_A64;
mbs.param[1] = RESULT_QUEUE_LEN(isp);
mbs.param[2] = DMA_WD1(isp->isp_result_dma);
mbs.param[3] = DMA_WD0(isp->isp_result_dma);
mbs.param[4] = 0;
mbs.param[6] = DMA_WD3(isp->isp_result_dma);
mbs.param[7] = DMA_WD2(isp->isp_result_dma);
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
isp->isp_residx = mbs.param[5];
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
If we have a 1240 or an ULTRA2 or better card, use MBOX_INIT_RES_QUEUE_A64 (preparation for DAC/A64 support)
2002-09-23 04:59:42 +00:00
mbs.param[0] = MBOX_INIT_REQ_QUEUE_A64;
mbs.param[1] = RQUEST_QUEUE_LEN(isp);
mbs.param[2] = DMA_WD1(isp->isp_rquest_dma);
mbs.param[3] = DMA_WD0(isp->isp_rquest_dma);
mbs.param[5] = 0;
mbs.param[6] = DMA_WD3(isp->isp_result_dma);
mbs.param[7] = DMA_WD2(isp->isp_result_dma);
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
isp->isp_reqidx = isp->isp_reqodx = mbs.param[4];
} else {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
If we have a 1240 or an ULTRA2 or better card, use MBOX_INIT_RES_QUEUE_A64 (preparation for DAC/A64 support)
2002-09-23 04:59:42 +00:00
mbs.param[0] = MBOX_INIT_RES_QUEUE;
mbs.param[1] = RESULT_QUEUE_LEN(isp);
mbs.param[2] = DMA_WD1(isp->isp_result_dma);
mbs.param[3] = DMA_WD0(isp->isp_result_dma);
mbs.param[4] = 0;
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
isp->isp_residx = mbs.param[5];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
If we have a 1240 or an ULTRA2 or better card, use MBOX_INIT_RES_QUEUE_A64 (preparation for DAC/A64 support)
2002-09-23 04:59:42 +00:00
mbs.param[0] = MBOX_INIT_REQ_QUEUE;
mbs.param[1] = RQUEST_QUEUE_LEN(isp);
mbs.param[2] = DMA_WD1(isp->isp_rquest_dma);
mbs.param[3] = DMA_WD0(isp->isp_rquest_dma);
mbs.param[5] = 0;
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
isp->isp_reqidx = isp->isp_reqodx = mbs.param[4];
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
* Turn on Fast Posting, LVD transitions
*
* Ultra2 F/W always has had fast posting (and LVD transitions)
*
Clean out residual bogosity for fast posting stuff- ISP_NO_FASTPOST_SCSI is gone as a define. We just don't support fast posting for anything less than the 1240/1080/1280/12160 or Fibre Channel cards. Put in support for CDB's larger than 12 bytes for parallel SCSI (up to 44 bytes are allowed). Approved: jkh
2000-02-15 00:35:00 +00:00
* Ultra and older (i.e., SBus) cards may not. It's just safer
* to assume not for them.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
mbs.param[0] = MBOX_SET_FW_FEATURES;
mbs.param[1] = 0;
if (IS_ULTRA2(isp))
mbs.param[1] |= FW_FEATURE_LVD_NOTIFY;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
#ifndef ISP_NO_RIO
if (IS_ULTRA2(isp) || IS_1240(isp))
mbs.param[1] |= FW_FEATURE_RIO_16BIT;
#else
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
if (IS_ULTRA2(isp) || IS_1240(isp))
mbs.param[1] |= FW_FEATURE_FAST_POST;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
#endif
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
if (mbs.param[1] != 0) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t sfeat = mbs.param[1];
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGINFO,
"Enabled FW features (0x%x)", sfeat);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
}
/*
* Let the outer layers decide whether to issue a SCSI bus reset.
*/
isp->isp_state = ISP_INITSTATE;
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_scsi_channel_init(ispsoftc_t *isp, int channel)
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
{
sdparam *sdp;
mbreg_t mbs;
int tgt;
sdp = isp->isp_param;
sdp += channel;
/*
* Set (possibly new) Initiator ID.
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[0] = MBOX_SET_INIT_SCSI_ID;
mbs.param[1] = (channel << 7) | sdp->isp_initiator_id;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
}
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
isp_prt(isp, ISP_LOGINFO, "Initiator ID is %d on Channel %d",
sdp->isp_initiator_id, channel);
Inintialize the queue index stuff from what the f/w sends back- just in case it's insane enough to not do what you tell it to. Print out (LOGINFO level) initiator ID.
2000-09-21 17:06:45 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
* Set current per-target parameters to an initial safe minimum.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
int lun;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t sdf;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (sdp->isp_devparam[tgt].dev_enable == 0) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
continue;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
#ifndef ISP_TARGET_MODE
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdf = sdp->isp_devparam[tgt].goal_flags;
sdf &= DPARM_SAFE_DFLT;
/*
* It is not quite clear when this changed over so that
* we could force narrow and async for 1000/1020 cards,
* but assume that this is only the case for loaded
* firmware.
*/
if (isp->isp_loaded_fw) {
sdf |= DPARM_NARROW | DPARM_ASYNC;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
}
#else
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/*
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
* The !$*!)$!$)* f/w uses the same index into some
* internal table to decide how to respond to negotiations,
* so if we've said "let's be safe" for ID X, and ID X
* selects *us*, the negotiations will back to 'safe'
* (as in narrow/async). What the f/w *should* do is
* use the initiator id settings to decide how to respond.
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
*/
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].goal_flags = sdf = DPARM_DEFAULT;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
#endif
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
mbs.param[0] = MBOX_SET_TARGET_PARAMS;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
mbs.param[1] = (channel << 15) | (tgt << 8);
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
mbs.param[2] = sdf;
various fixes
2000-08-27 23:38:44 +00:00
if ((sdf & DPARM_SYNC) == 0) {
mbs.param[3] = 0;
} else {
mbs.param[3] =
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
(sdp->isp_devparam[tgt].goal_offset << 8) |
(sdp->isp_devparam[tgt].goal_period);
various fixes
2000-08-27 23:38:44 +00:00
}
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
"Initial Settings bus%d tgt%d flags 0x%x off 0x%x per 0x%x",
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
channel, tgt, mbs.param[2], mbs.param[3] >> 8,
mbs.param[3] & 0xff);
isp_mboxcmd(isp, &mbs, MBLOGNONE);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
sdf = DPARM_SAFE_DFLT;
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_SET_TARGET_PARAMS;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[1] = (tgt << 8) | (channel << 15);
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
mbs.param[2] = sdf;
various fixes
2000-08-27 23:38:44 +00:00
mbs.param[3] = 0;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
continue;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
Annoying little nigglet- apparently *some* Qlogic temporarily ignore settings you've just sent them and return random values if you follow the set by a get. This causes problems when you latter run a Tag-enabled command when you've command tagged mode off.
1999-03-26 00:33:13 +00:00
/*
various fixes
2000-08-27 23:38:44 +00:00
* We don't update any information directly from the f/w
* because we need to run at least one command to cause a
* new state to be latched up. So, we just assume that we
* converge to the values we just had set.
*
Annoying little nigglet- apparently *some* Qlogic temporarily ignore settings you've just sent them and return random values if you follow the set by a get. This causes problems when you latter run a Tag-enabled command when you've command tagged mode off.
1999-03-26 00:33:13 +00:00
* Ensure that we don't believe tagged queuing is enabled yet.
* It turns out that sometimes the ISP just ignores our
* attempts to set parameters for devices that it hasn't
* seen yet.
*/
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].actv_flags = sdf & ~DPARM_TQING;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
for (lun = 0; lun < (int) isp->isp_maxluns; lun++) {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_SET_DEV_QUEUE_PARAMS;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[1] = (channel << 15) | (tgt << 8) | lun;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[2] = sdp->isp_max_queue_depth;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
mbs.param[3] = sdp->isp_devparam[tgt].exc_throttle;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
if (sdp->isp_devparam[tgt].dev_refresh) {
isp->isp_sendmarker |= (1 << channel);
isp->isp_update |= (1 << channel);
break;
}
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* Fibre Channel specific initialization.
*
* Locks are held before coming here.
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_fibre_init(ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
fcparam *fcp;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_icb_t local, *icbp = &local;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbreg_t mbs;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
int loopid;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint64_t nwwn, pwwn;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
fcp = isp->isp_param;
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* Do this *before* initializing the firmware.
*/
isp_mark_getpdb_all(isp);
fcp->isp_fwstate = FW_CONFIG_WAIT;
fcp->isp_loopstate = LOOP_NIL;
/*
* If we have no role (neither target nor initiator), return.
*/
if (isp->isp_role == ISP_ROLE_NONE) {
return;
}
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
loopid = fcp->isp_loopid;
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
MEMZERO(icbp, sizeof (*icbp));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
icbp->icb_version = ICB_VERSION1;
various fixes
2000-08-27 23:38:44 +00:00
/*
* Firmware Options are either retrieved from NVRAM or
* are patched elsewhere. We check them for sanity here
* and make changes based on board revision, but otherwise
* let others decide policy.
*/
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
/*
* If this is a 2100 < revision 5, we have to turn off FAIRNESS.
*/
if ((isp->isp_type == ISP_HA_FC_2100) && isp->isp_revision < 5) {
fcp->isp_fwoptions &= ~ICBOPT_FAIRNESS;
}
various fixes
2000-08-27 23:38:44 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
* We have to use FULL LOGIN even though it resets the loop too much
* because otherwise port database entries don't get updated after
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
* a LIP- this is a known f/w bug for 2100 f/w less than 1.17.0.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
if (!ISP_FW_NEWER_THAN(isp, 1, 17, 0)) {
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
fcp->isp_fwoptions |= ICBOPT_FULL_LOGIN;
}
various fixes
2000-08-27 23:38:44 +00:00
/*
* Insist on Port Database Update Async notifications
*/
fcp->isp_fwoptions |= ICBOPT_PDBCHANGE_AE;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* Make sure that target role reflects into fwoptions.
*/
if (isp->isp_role & ISP_ROLE_TARGET) {
fcp->isp_fwoptions |= ICBOPT_TGT_ENABLE;
} else {
fcp->isp_fwoptions &= ~ICBOPT_TGT_ENABLE;
}
If we have ISP_ROLE_INITIATOR set, make sure that we clear ICBOPT_INI_DISABLE from the fwoptions. Likewise, we *set* ICBOPT_INI_DISABLE if we don't have initiator role.
2004-01-23 23:23:31 +00:00
if (isp->isp_role & ISP_ROLE_INITIATOR) {
fcp->isp_fwoptions &= ~ICBOPT_INI_DISABLE;
} else {
fcp->isp_fwoptions |= ICBOPT_INI_DISABLE;
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* Propagate all of this into the ICB structure.
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
icbp->icb_fwoptions = fcp->isp_fwoptions;
icbp->icb_maxfrmlen = fcp->isp_maxfrmlen;
if (icbp->icb_maxfrmlen < ICB_MIN_FRMLEN ||
icbp->icb_maxfrmlen > ICB_MAX_FRMLEN) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"bad frame length (%d) from NVRAM- using %d",
fcp->isp_maxfrmlen, ICB_DFLT_FRMLEN);
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
icbp->icb_maxfrmlen = ICB_DFLT_FRMLEN;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
icbp->icb_maxalloc = fcp->isp_maxalloc;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
if (icbp->icb_maxalloc < 1) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"bad maximum allocation (%d)- using 16", fcp->isp_maxalloc);
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
icbp->icb_maxalloc = 16;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
icbp->icb_execthrottle = fcp->isp_execthrottle;
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
if (icbp->icb_execthrottle < 1) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"bad execution throttle of %d- using 16",
fcp->isp_execthrottle);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
icbp->icb_execthrottle = ICB_DFLT_THROTTLE;
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
icbp->icb_retry_delay = fcp->isp_retry_delay;
icbp->icb_retry_count = fcp->isp_retry_count;
oops on last
1998-12-05 01:46:40 +00:00
icbp->icb_hardaddr = loopid;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (icbp->icb_hardaddr > 125) {
On reset, make sure that we have some parameters set correctly. This fixes a longstanding issue WRT resetting the chip after startup- it would fail if we were connected as an F-port to a switch. If we were connected as an F-port, we got assigned a hard loop ID of 255, which is really a bogus loop id. Then when we turned around to reset ourselves, the firmware would reject the ICB_INIT request because the loop id was bogus. *sputter* Minor fixlet from somebody in NetBSD with too much time on their hands (dma -> DMA).
2003-09-13 01:55:44 +00:00
/*
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
* We end up with these Loop IDs for F-Port topologies
On reset, make sure that we have some parameters set correctly. This fixes a longstanding issue WRT resetting the chip after startup- it would fail if we were connected as an F-port to a switch. If we were connected as an F-port, we got assigned a hard loop ID of 255, which is really a bogus loop id. Then when we turned around to reset ourselves, the firmware would reject the ICB_INIT request because the loop id was bogus. *sputter* Minor fixlet from somebody in NetBSD with too much time on their hands (dma -> DMA).
2003-09-13 01:55:44 +00:00
*/
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
if (icbp->icb_hardaddr != 0xff && icbp->icb_hardaddr != 0x800) {
On reset, make sure that we have some parameters set correctly. This fixes a longstanding issue WRT resetting the chip after startup- it would fail if we were connected as an F-port to a switch. If we were connected as an F-port, we got assigned a hard loop ID of 255, which is really a bogus loop id. Then when we turned around to reset ourselves, the firmware would reject the ICB_INIT request because the loop id was bogus. *sputter* Minor fixlet from somebody in NetBSD with too much time on their hands (dma -> DMA).
2003-09-13 01:55:44 +00:00
isp_prt(isp, ISP_LOGERR,
"bad hard address %u- resetting to zero",
icbp->icb_hardaddr);
}
icbp->icb_hardaddr = 0;
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
What the heck - make the last (most recent) 2200 f/w also do Hard Loop acquisition.
2006-07-03 20:56:48 +00:00
/*
* Our life seems so much better with 2200s and later with
* the latest f/w if we set Hard Address.
*/
if (ISP_FW_NEWER_THAN(isp, 2, 2, 5)) {
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
icbp->icb_fwoptions |= ICBOPT_HARD_ADDRESS;
}
various fixes
2000-08-27 23:38:44 +00:00
/*
* Right now we just set extended options to prefer point-to-point
* over loop based upon some soft config options.
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
*
* NB: for the 2300, ICBOPT_EXTENDED is required.
various fixes
2000-08-27 23:38:44 +00:00
*/
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_2200(isp) || IS_23XX(isp)) {
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
icbp->icb_fwoptions |= ICBOPT_EXTENDED;
/*
* Prefer or force Point-To-Point instead Loop?
*/
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
switch(isp->isp_confopts & ISP_CFG_PORT_PREF) {
case ISP_CFG_NPORT:
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
icbp->icb_xfwoptions |= ICBXOPT_PTP_2_LOOP;
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
break;
case ISP_CFG_NPORT_ONLY:
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
icbp->icb_xfwoptions |= ICBXOPT_PTP_ONLY;
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
break;
case ISP_CFG_LPORT_ONLY:
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
icbp->icb_xfwoptions |= ICBXOPT_LOOP_ONLY;
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
break;
default:
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
icbp->icb_xfwoptions |= ICBXOPT_LOOP_2_PTP;
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
break;
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_2200(isp)) {
if (ISP_FW_NEWER_THAN(isp, 1, 17, 0)) {
icbp->icb_xfwoptions |= ICBXOPT_RIO_16BIT;
icbp->icb_racctimer = 4;
icbp->icb_idelaytimer = 8;
}
icbp->icb_fwoptions |= ICBOPT_FAST_POST;
} else {
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
/*
* QLogic recommends that FAST Posting be turned
* off for 23XX cards and instead allow the HBA
* to write response queue entries and interrupt
* after a delay (ZIO).
*/
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
icbp->icb_fwoptions &= ~ICBOPT_FAST_POST;
if ((fcp->isp_xfwoptions & ICBXOPT_TIMER_MASK) ==
ICBXOPT_ZIO) {
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
icbp->icb_xfwoptions |= ICBXOPT_ZIO;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
icbp->icb_idelaytimer = 10;
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
}
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
if (isp->isp_confopts & ISP_CFG_ONEGB) {
Misunderstanding documentation caused me to try and set 1Gbps/2Gps/Auto connection speed for the 2300 in the wrong offset in the ICB. Oops. Respect some QLogic errat wrt PCI errors on certain shared host/RISC registers.
2001-10-06 20:41:18 +00:00
icbp->icb_zfwoptions |= ICBZOPT_RATE_ONEGB;
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
} else if (isp->isp_confopts & ISP_CFG_TWOGB) {
Misunderstanding documentation caused me to try and set 1Gbps/2Gps/Auto connection speed for the 2300 in the wrong offset in the ICB. Oops. Respect some QLogic errat wrt PCI errors on certain shared host/RISC registers.
2001-10-06 20:41:18 +00:00
icbp->icb_zfwoptions |= ICBZOPT_RATE_TWOGB;
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
} else {
Misunderstanding documentation caused me to try and set 1Gbps/2Gps/Auto connection speed for the 2300 in the wrong offset in the ICB. Oops. Respect some QLogic errat wrt PCI errors on certain shared host/RISC registers.
2001-10-06 20:41:18 +00:00
icbp->icb_zfwoptions |= ICBZOPT_RATE_AUTO;
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (fcp->isp_zfwoptions & ICBZOPT_50_OHM) {
icbp->icb_zfwoptions |= ICBZOPT_50_OHM;
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
}
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
/*
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
* For 22XX > 2.1.26 && 23XX, set some options.
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
* XXX: Probably okay for newer 2100 f/w too.
*/
if (ISP_FW_NEWER_THAN(isp, 2, 26, 0)) {
Enable LIP F8, LIP Reset async events. Be more chatty about SNS failures. Fix typo for skipped phase mesage. Correct MBOX_GET_PORT_QUEUE_PARAMS options in table. MFC after: 2 weeks
2001-08-16 17:25:08 +00:00
/*
* Turn on LIP F8 async event (1)
* Turn on generate AE 8013 on all LIP Resets (2)
* Disable LIP F7 switching (8)
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Enable LIP F8, LIP Reset async events. Be more chatty about SNS failures. Fix typo for skipped phase mesage. Correct MBOX_GET_PORT_QUEUE_PARAMS options in table. MFC after: 2 weeks
2001-08-16 17:25:08 +00:00
mbs.param[0] = MBOX_SET_FIRMWARE_OPTIONS;
mbs.param[1] = 0xb;
mbs.param[2] = 0;
mbs.param[3] = 0;
isp_mboxcmd(isp, &mbs, MBLOGALL);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
icbp->icb_logintime = ICB_LOGIN_TOV;
icbp->icb_lunetimeout = ICB_LUN_ENABLE_TOV;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
nwwn = ISP_NODEWWN(isp);
pwwn = ISP_PORTWWN(isp);
if (nwwn && pwwn) {
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
icbp->icb_fwoptions |= ICBOPT_BOTH_WWNS;
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
MAKE_NODE_NAME_FROM_WWN(icbp->icb_nodename, nwwn);
MAKE_NODE_NAME_FROM_WWN(icbp->icb_portname, pwwn);
isp_prt(isp, ISP_LOGDEBUG1,
"Setting ICB Node 0x%08x%08x Port 0x%08x%08x",
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
((uint32_t) (nwwn >> 32)),
((uint32_t) (nwwn & 0xffffffff)),
((uint32_t) (pwwn >> 32)),
((uint32_t) (pwwn & 0xffffffff)));
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
} else {
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
isp_prt(isp, ISP_LOGDEBUG1, "Not using any WWNs");
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
icbp->icb_fwoptions &= ~(ICBOPT_BOTH_WWNS|ICBOPT_FULL_LOGIN);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
icbp->icb_rqstqlen = RQUEST_QUEUE_LEN(isp);
On reset, make sure that we have some parameters set correctly. This fixes a longstanding issue WRT resetting the chip after startup- it would fail if we were connected as an F-port to a switch. If we were connected as an F-port, we got assigned a hard loop ID of 255, which is really a bogus loop id. Then when we turned around to reset ourselves, the firmware would reject the ICB_INIT request because the loop id was bogus. *sputter* Minor fixlet from somebody in NetBSD with too much time on their hands (dma -> DMA).
2003-09-13 01:55:44 +00:00
if (icbp->icb_rqstqlen < 1) {
isp_prt(isp, ISP_LOGERR, "bad request queue length");
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
icbp->icb_rsltqlen = RESULT_QUEUE_LEN(isp);
On reset, make sure that we have some parameters set correctly. This fixes a longstanding issue WRT resetting the chip after startup- it would fail if we were connected as an F-port to a switch. If we were connected as an F-port, we got assigned a hard loop ID of 255, which is really a bogus loop id. Then when we turned around to reset ourselves, the firmware would reject the ICB_INIT request because the loop id was bogus. *sputter* Minor fixlet from somebody in NetBSD with too much time on their hands (dma -> DMA).
2003-09-13 01:55:44 +00:00
if (icbp->icb_rsltqlen < 1) {
isp_prt(isp, ISP_LOGERR, "bad result queue length");
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
icbp->icb_rqstaddr[RQRSP_ADDR0015] = DMA_WD0(isp->isp_rquest_dma);
icbp->icb_rqstaddr[RQRSP_ADDR1631] = DMA_WD1(isp->isp_rquest_dma);
icbp->icb_rqstaddr[RQRSP_ADDR3247] = DMA_WD2(isp->isp_rquest_dma);
icbp->icb_rqstaddr[RQRSP_ADDR4863] = DMA_WD3(isp->isp_rquest_dma);
icbp->icb_respaddr[RQRSP_ADDR0015] = DMA_WD0(isp->isp_result_dma);
icbp->icb_respaddr[RQRSP_ADDR1631] = DMA_WD1(isp->isp_result_dma);
icbp->icb_respaddr[RQRSP_ADDR3247] = DMA_WD2(isp->isp_result_dma);
icbp->icb_respaddr[RQRSP_ADDR4863] = DMA_WD3(isp->isp_result_dma);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
"isp_fibre_init: fwopt 0x%x xfwopt 0x%x zfwopt 0x%x",
icbp->icb_fwoptions, icbp->icb_xfwoptions, icbp->icb_zfwoptions);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_ACQUIRE(isp);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_put_icb(isp, icbp, (isp_icb_t *)fcp->isp_scratch);
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
/*
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
* Init the firmware
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
mbs.param[0] = MBOX_INIT_FIRMWARE;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "INIT F/W from %p (%08x%08x)",
fcp->isp_scratch, (uint32_t) ((uint64_t)fcp->isp_scdma >> 32),
(uint32_t) fcp->isp_scdma);
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, sizeof (*icbp));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
isp->isp_reqidx = isp->isp_reqodx = 0;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_residx = 0;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
isp->isp_sendmarker = 1;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
* Whatever happens, we're now committed to being here.
*/
isp->isp_state = ISP_INITSTATE;
}
/*
* Fibre Channel Support- get the port database for the id.
*
* Locks are held before coming here. Return 0 if success,
* else failure.
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_getmap(ispsoftc_t *isp, fcpos_map_t *map)
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
{
fcparam *fcp = (fcparam *) isp->isp_param;
mbreg_t mbs;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbs.param[0] = MBOX_GET_FC_AL_POSITION_MAP;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_ACQUIRE(isp);
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, sizeof (fcpos_map_t));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL & ~MBOX_COMMAND_PARAM_ERROR);
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
MEMCPY(map, fcp->isp_scratch, sizeof (fcpos_map_t));
map->fwmap = mbs.param[1] != 0;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
return (0);
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
return (-1);
}
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_mark_getpdb_all(ispsoftc_t *isp)
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
{
fcparam *fcp = (fcparam *) isp->isp_param;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
int i;
for (i = 0; i < MAX_FC_TARG; i++) {
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
fcp->portdb[i].valid = fcp->portdb[i].fabric_dev = 0;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
}
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_getpdb(ispsoftc_t *isp, int id, isp_pdb_t *pdbp)
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
{
fcparam *fcp = (fcparam *) isp->isp_param;
mbreg_t mbs;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
mbs.param[0] = MBOX_GET_PORT_DB;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
mbs.param[1] = id;
mbs.obits |= (1 << 10);
} else {
mbs.param[1] = id << 8;
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_ACQUIRE(isp);
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, sizeof (isp_pdb_t));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL & ~MBOX_COMMAND_PARAM_ERROR);
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_get_pdb(isp, (isp_pdb_t *)fcp->isp_scratch, pdbp);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
return (0);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
return (-1);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
static uint64_t
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_get_portname(ispsoftc_t *isp, int loopid, int nodename)
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint64_t wwn = 0;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
mbreg_t mbs;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
mbs.param[0] = MBOX_GET_PORT_NAME;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
mbs.param[1] = loopid;
mbs.obits |= (1 << 10);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (nodename) {
mbs.param[10] = 1;
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
} else {
mbs.param[1] = loopid << 8;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (nodename) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
mbs.param[1] |= 1;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL & ~MBOX_COMMAND_PARAM_ERROR);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
wwn =
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
(((uint64_t)(mbs.param[2] & 0xff)) << 56) |
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)(mbs.param[2] >> 8)) << 48) |
(((uint64_t)(mbs.param[3] & 0xff)) << 40) |
(((uint64_t)(mbs.param[3] >> 8)) << 32) |
(((uint64_t)(mbs.param[6] & 0xff)) << 24) |
(((uint64_t)(mbs.param[6] >> 8)) << 16) |
(((uint64_t)(mbs.param[7] & 0xff)) << 8) |
(((uint64_t)(mbs.param[7] >> 8)));
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
return (wwn);
}
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
* Make sure we have good FC link and know our Loop ID.
*/
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_fclink_test(ispsoftc_t *isp, int usdelay)
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
{
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
static char *toponames[] = {
"Private Loop",
"FL Port",
"N-Port to N-Port",
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
"F Port",
"F Port (no FLOGI_ACC response)"
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
};
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
mbreg_t mbs;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
int count, check_for_fabric;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t lwfs;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
fcparam *fcp;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
struct lportdb *lp;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
isp_pdb_t pdb;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
fcp = isp->isp_param;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "FC Link Test Entry");
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
/*
* XXX: Here is where we would start a 'loop dead' timeout
*/
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
* Wait up to N microseconds for F/W to go to a ready state.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
lwfs = FW_CONFIG_WAIT;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
count = 0;
while (count < usdelay) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint64_t enano;
uint32_t wrk;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
NANOTIME_T hra, hrb;
GET_NANOTIME(&hra);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp_fw_state(isp);
if (lwfs != fcp->isp_fwstate) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO, "Firmware State <%s->%s>",
isp2100_fw_statename((int)lwfs),
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
isp2100_fw_statename((int)fcp->isp_fwstate));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
lwfs = fcp->isp_fwstate;
}
if (fcp->isp_fwstate == FW_READY) {
break;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
GET_NANOTIME(&hrb);
/*
* Get the elapsed time in nanoseconds.
* Always guaranteed to be non-zero.
*/
enano = NANOTIME_SUB(&hrb, &hra);
In order to save ourselves grief with the SUNPRO compiler under Solaris (which, for reasons unknown to me, chokes on u_int16_t as a typedef of unsigned short if used in a transitional (mixed K&R and ANSI) way), we'll go the extra mile and fully ANSIfy things.
2001-03-14 04:11:56 +00:00
isp_prt(isp, ISP_LOGDEBUG1,
"usec%d: 0x%lx->0x%lx enano 0x%x%08x",
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
count, (long) GET_NANOSEC(&hra), (long) GET_NANOSEC(&hrb),
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(uint32_t)(enano >> 32), (uint32_t)(enano & 0xffffffff));
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
/*
* If the elapsed time is less than 1 millisecond,
* delay a period of time up to that millisecond of
* waiting.
Add in Bill Sommerfelds -Wformat changes. Set up default node && port WWNs correctly (Again!) - this time for the case that we're not going to fully init the adapter if isp_init is called (with ISP_CFG_NOINIT set in options). The pupose for this is to bring the adapter up to almost ready to go, get info out of NVRAM, but to not start it up- leaving it until later to actually start things up if wanted (and possibly with different roles selected).
2000-12-29 19:12:44 +00:00
*
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
* This peculiar code is an attempt to try and avoid
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
* invoking uint64_t math support functions for some
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
* platforms where linkage is a problem.
*/
if (enano < (1000 * 1000)) {
count += 1000;
enano = (1000 * 1000) - enano;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
while (enano > (uint64_t) 4000000000U) {
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
USEC_SLEEP(isp, 4000000);
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
enano -= (uint64_t) 4000000000U;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
wrk = enano;
In order to save ourselves grief with the SUNPRO compiler under Solaris (which, for reasons unknown to me, chokes on u_int16_t as a typedef of unsigned short if used in a transitional (mixed K&R and ANSI) way), we'll go the extra mile and fully ANSIfy things.
2001-03-14 04:11:56 +00:00
wrk /= 1000;
USEC_SLEEP(isp, wrk);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
} else {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
while (enano > (uint64_t) 4000000000U) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
count += 4000000;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
enano -= (uint64_t) 4000000000U;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
wrk = enano;
count += (wrk / 1000);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
/*
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
* If we haven't gone to 'ready' state, return.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (fcp->isp_fwstate != FW_READY) {
return (-1);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
* Get our Loop ID (if possible). We really need to have it.
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
mbs.param[0] = MBOX_GET_LOOP_ID;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
return (-1);
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_2KLOGIN(isp)) {
fcp->isp_loopid = mbs.param[1];
} else {
fcp->isp_loopid = mbs.param[1] & 0xff;
}
Make sure we don't pick up a loopid that's larger than our current portdb max (MAX_FC_TARG == 256) now that we support 2K Login f/w. MFC after: 3 days
2006-02-02 09:02:16 +00:00
if (fcp->isp_loopid == 0xffff) { /* happens with 2k login f/w */
fcp->isp_loopid = MAX_FC_TARG-1;
} else if (fcp->isp_loopid >= MAX_FC_TARG) {
isp_prt(isp, ISP_LOGWARN, "bad initiator loopid (0x%x)", fcp->isp_loopid);
fcp->isp_loopid = MAX_FC_TARG-1;
}
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_2200(isp) || IS_23XX(isp)) {
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
int topo = (int) mbs.param[6];
if (topo < TOPO_NL_PORT || topo > TOPO_PTP_STUB)
topo = TOPO_PTP_STUB;
fcp->isp_topo = topo;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
} else {
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
fcp->isp_topo = TOPO_NL_PORT;
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
/*
Make sure we don't pick up a loopid that's larger than our current portdb max (MAX_FC_TARG == 256) now that we support 2K Login f/w. MFC after: 3 days
2006-02-02 09:02:16 +00:00
* Get the port id.
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
*/
Make sure we don't pick up a loopid that's larger than our current portdb max (MAX_FC_TARG == 256) now that we support 2K Login f/w. MFC after: 3 days
2006-02-02 09:02:16 +00:00
fcp->isp_portid = mbs.param[2] | (mbs.param[3] << 16);
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* Check to see if we're on a fabric by trying to see if we
* can talk to the fabric name server. This can be a bit
* tricky because if we're a 2100, we should check always
Fix typos, mostly s/ an / a / where appropriate and a few s/an/and/ Add FreeBSD Id tag where missing.
2002-12-30 21:18:15 +00:00
* (in case we're connected to a server doing aliasing).
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
*/
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
fcp->isp_onfabric = 0;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
if (IS_2100(isp)) {
/*
* Don't bother with fabric if we are using really old
* 2100 firmware. It's just not worth it.
*/
if (ISP_FW_NEWER_THAN(isp, 1, 15, 37)) {
check_for_fabric = 1;
} else {
check_for_fabric = 0;
}
} else if (fcp->isp_topo == TOPO_FL_PORT ||
fcp->isp_topo == TOPO_F_PORT) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
check_for_fabric = 1;
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
} else
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
check_for_fabric = 0;
if (check_for_fabric && isp_getpdb(isp, FL_PORT_ID, &pdb) == 0) {
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
int loopid = FL_PORT_ID;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
if (IS_2100(isp)) {
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
fcp->isp_topo = TOPO_FL_PORT;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (BITS2WORD(pdb.pdb_portid_bits) == 0) {
/*
* Crock.
*/
fcp->isp_topo = TOPO_NL_PORT;
goto not_on_fabric;
}
fcp->isp_portid = mbs.param[2] | ((int) mbs.param[3] << 16);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
* Save the Fabric controller's port database entry.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
lp = &fcp->portdb[loopid];
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
lp->node_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_nodename[0]) << 56) |
(((uint64_t)pdb.pdb_nodename[1]) << 48) |
(((uint64_t)pdb.pdb_nodename[2]) << 40) |
(((uint64_t)pdb.pdb_nodename[3]) << 32) |
(((uint64_t)pdb.pdb_nodename[4]) << 24) |
(((uint64_t)pdb.pdb_nodename[5]) << 16) |
(((uint64_t)pdb.pdb_nodename[6]) << 8) |
(((uint64_t)pdb.pdb_nodename[7]));
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
lp->port_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_portname[0]) << 56) |
(((uint64_t)pdb.pdb_portname[1]) << 48) |
(((uint64_t)pdb.pdb_portname[2]) << 40) |
(((uint64_t)pdb.pdb_portname[3]) << 32) |
(((uint64_t)pdb.pdb_portname[4]) << 24) |
(((uint64_t)pdb.pdb_portname[5]) << 16) |
(((uint64_t)pdb.pdb_portname[6]) << 8) |
(((uint64_t)pdb.pdb_portname[7]));
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
lp->roles =
(pdb.pdb_prli_svc3 & SVC3_ROLE_MASK) >> SVC3_ROLE_SHIFT;
lp->portid = BITS2WORD(pdb.pdb_portid_bits);
lp->loopid = pdb.pdb_loopid;
lp->loggedin = lp->valid = 1;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_onfabric = 1;
(void) isp_async(isp, ISPASYNC_PROMENADE, &loopid);
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
isp_register_fc4_type(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
} else {
not_on_fabric:
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
fcp->isp_onfabric = 0;
fcp->portdb[FL_PORT_ID].valid = 0;
}
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
fcp->isp_gbspeed = 1;
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_23XX(isp)) {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
mbs.param[0] = MBOX_GET_SET_DATA_RATE;
mbs.param[1] = MBGSD_GET_RATE;
/* mbs.param[2] undefined if we're just getting rate */
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
if (mbs.param[1] == MBGSD_TWOGB) {
isp_prt(isp, ISP_LOGINFO, "2Gb link speed/s");
fcp->isp_gbspeed = 2;
}
}
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
isp_prt(isp, ISP_LOGCONFIG, topology, fcp->isp_loopid,
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
fcp->isp_portid, fcp->isp_loopstate, toponames[fcp->isp_topo]);
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* Announce ourselves, too. This involves synthesizing an entry.
*/
if (fcp->isp_iid_set == 0) {
fcp->isp_iid_set = 1;
fcp->isp_iid = fcp->isp_loopid;
lp = &fcp->portdb[fcp->isp_iid];
} else {
lp = &fcp->portdb[fcp->isp_iid];
if (fcp->isp_portid != lp->portid ||
fcp->isp_loopid != lp->loopid ||
fcp->isp_nodewwn != ISP_NODEWWN(isp) ||
fcp->isp_portwwn != ISP_PORTWWN(isp)) {
lp->valid = 0;
count = fcp->isp_iid;
(void) isp_async(isp, ISPASYNC_PROMENADE, &count);
}
}
lp->loopid = fcp->isp_loopid;
lp->portid = fcp->isp_portid;
lp->node_wwn = ISP_NODEWWN(isp);
lp->port_wwn = ISP_PORTWWN(isp);
switch (isp->isp_role) {
case ISP_ROLE_NONE:
lp->roles = 0;
break;
case ISP_ROLE_TARGET:
lp->roles = SVC3_TGT_ROLE >> SVC3_ROLE_SHIFT;
break;
case ISP_ROLE_INITIATOR:
lp->roles = SVC3_INI_ROLE >> SVC3_ROLE_SHIFT;
break;
case ISP_ROLE_BOTH:
lp->roles = (SVC3_INI_ROLE|SVC3_TGT_ROLE) >> SVC3_ROLE_SHIFT;
break;
}
lp->loggedin = lp->valid = 1;
count = fcp->isp_iid;
(void) isp_async(isp, ISPASYNC_PROMENADE, &count);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "FC Link Test Complete");
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
return (0);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
static const char *
In order to save ourselves grief with the SUNPRO compiler under Solaris (which, for reasons unknown to me, chokes on u_int16_t as a typedef of unsigned short if used in a transitional (mixed K&R and ANSI) way), we'll go the extra mile and fully ANSIfy things.
2001-03-14 04:11:56 +00:00
isp2100_fw_statename(int state)
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
{
switch(state) {
case FW_CONFIG_WAIT: return "Config Wait";
case FW_WAIT_AL_PA: return "Waiting for AL_PA";
case FW_WAIT_LOGIN: return "Wait Login";
case FW_READY: return "Ready";
case FW_LOSS_OF_SYNC: return "Loss Of Sync";
case FW_ERROR: return "Error";
case FW_REINIT: return "Re-Init";
case FW_NON_PART: return "Nonparticipating";
default: return "?????";
}
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
* Synchronize our soft copy of the port database with what the f/w thinks
* (with a view toward possibly for a specific target....)
*/
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_pdb_sync(ispsoftc_t *isp)
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
{
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
struct lportdb *lp;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
fcparam *fcp = isp->isp_param;
isp_pdb_t pdb;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
int loopid, base, lim;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Make sure we're okay for doing this right now.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_loopstate != LOOP_PDB_RCVD &&
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
fcp->isp_loopstate != LOOP_FSCAN_DONE &&
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate != LOOP_LSCAN_DONE) {
return (-1);
}
if (fcp->isp_topo == TOPO_FL_PORT || fcp->isp_topo == TOPO_NL_PORT ||
fcp->isp_topo == TOPO_N_PORT) {
if (fcp->isp_loopstate < LOOP_LSCAN_DONE) {
if (isp_scan_loop(isp) != 0) {
return (-1);
}
}
}
fcp->isp_loopstate = LOOP_SYNCING_PDB;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If we get this far, we've settled our differences with the f/w
* (for local loop device) and we can say that the loop state is ready.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_topo == TOPO_NL_PORT) {
fcp->loop_seen_once = 1;
fcp->isp_loopstate = LOOP_READY;
return (0);
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Find all Fabric Entities that didn't make it from one scan to the
* next and let the world know they went away. Scan the whole database.
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
for (lp = &fcp->portdb[0]; lp < &fcp->portdb[MAX_FC_TARG]; lp++) {
if (lp->was_fabric_dev && lp->fabric_dev == 0) {
loopid = lp - fcp->portdb;
lp->valid = 0; /* should already be set */
(void) isp_async(isp, ISPASYNC_PROMENADE, &loopid);
MEMZERO((void *) lp, sizeof (*lp));
continue;
}
lp->was_fabric_dev = lp->fabric_dev;
}
if (fcp->isp_topo == TOPO_FL_PORT)
base = FC_SNS_ID+1;
else
base = 0;
if (fcp->isp_topo == TOPO_N_PORT)
lim = 1;
else
lim = MAX_FC_TARG;
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Now log in any fabric devices that the outer layer has
* left for us to see. This seems the most sane policy
* for the moment.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
for (lp = &fcp->portdb[base]; lp < &fcp->portdb[lim]; lp++) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint32_t portid;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbreg_t mbs;
loopid = lp - fcp->portdb;
if (loopid >= FL_PORT_ID && loopid <= FC_SNS_ID) {
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
continue;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
}
/*
* Anything here?
*/
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
if (lp->port_wwn == 0) {
continue;
}
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Don't try to log into yourself.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if ((portid = lp->portid) == fcp->isp_portid) {
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
continue;
}
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If we'd been logged in- see if we still are and we haven't
* changed. If so, no need to log ourselves out, etc..
*
* Unfortunately, our charming Qlogic f/w has decided to
* return a valid port database entry for a fabric device
* that has, in fact, gone away. And it hangs trying to
* log it out.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
if (lp->loggedin && lp->force_logout == 0 &&
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_getpdb(isp, lp->loopid, &pdb) == 0) {
int nrole;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint64_t nwwnn, nwwpn;
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
nwwnn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_nodename[0]) << 56) |
(((uint64_t)pdb.pdb_nodename[1]) << 48) |
(((uint64_t)pdb.pdb_nodename[2]) << 40) |
(((uint64_t)pdb.pdb_nodename[3]) << 32) |
(((uint64_t)pdb.pdb_nodename[4]) << 24) |
(((uint64_t)pdb.pdb_nodename[5]) << 16) |
(((uint64_t)pdb.pdb_nodename[6]) << 8) |
(((uint64_t)pdb.pdb_nodename[7]));
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
nwwpn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_portname[0]) << 56) |
(((uint64_t)pdb.pdb_portname[1]) << 48) |
(((uint64_t)pdb.pdb_portname[2]) << 40) |
(((uint64_t)pdb.pdb_portname[3]) << 32) |
(((uint64_t)pdb.pdb_portname[4]) << 24) |
(((uint64_t)pdb.pdb_portname[5]) << 16) |
(((uint64_t)pdb.pdb_portname[6]) << 8) |
(((uint64_t)pdb.pdb_portname[7]));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
nrole = (pdb.pdb_prli_svc3 & SVC3_ROLE_MASK) >>
SVC3_ROLE_SHIFT;
if (pdb.pdb_loopid == lp->loopid && lp->portid ==
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(uint32_t) BITS2WORD(pdb.pdb_portid_bits) &&
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
nwwnn == lp->node_wwn && nwwpn == lp->port_wwn &&
Fix botch for state levels. Role minor release. Start adding code for a 'force logout' path. MFC after: 4 weeks
2001-06-05 17:11:06 +00:00
lp->roles == nrole && lp->force_logout == 0) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
lp->loggedin = lp->valid = 1;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp_prt(isp, ISP_LOGCONFIG, lretained,
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
(int) (lp - fcp->portdb),
(int) lp->loopid, lp->portid);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
continue;
}
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_SYNCING_PDB) {
return (-1);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Force a logout if we were logged in.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (lp->loggedin) {
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
if (lp->force_logout ||
isp_getpdb(isp, lp->loopid, &pdb) == 0) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbs.param[0] = MBOX_FABRIC_LOGOUT;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
mbs.param[1] = lp->loopid;
mbs.obits |= (1 << 10);
} else {
mbs.param[1] = lp->loopid << 8;
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
isp_prt(isp, ISP_LOGINFO, plogout,
(int) (lp - fcp->portdb), lp->loopid,
lp->portid);
}
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
lp->force_logout = lp->loggedin = 0;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_SYNCING_PDB) {
return (-1);
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* And log in....
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
loopid = lp - fcp->portdb;
lp->loopid = FL_PORT_ID;
do {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbs.param[0] = MBOX_FABRIC_LOGIN;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
mbs.param[1] = loopid;
mbs.obits |= (1 << 10);
} else {
mbs.param[1] = loopid << 8;
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbs.param[2] = portid >> 16;
mbs.param[3] = portid & 0xffff;
isp_mboxcmd(isp, &mbs, MBLOGALL & ~(MBOX_LOOP_ID_USED |
MBOX_PORT_ID_USED | MBOX_COMMAND_ERROR));
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_SYNCING_PDB) {
return (-1);
}
switch (mbs.param[0]) {
case MBOX_LOOP_ID_USED:
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Try the next available loop id.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
loopid++;
break;
case MBOX_PORT_ID_USED:
/*
* This port is already logged in.
* Snaffle the loop id it's using if it's
* nonzero, otherwise we're hosed.
*/
if (mbs.param[1] != 0) {
loopid = mbs.param[1];
Make sure we don't pick up a loopid that's larger than our current portdb max (MAX_FC_TARG == 256) now that we support 2K Login f/w. MFC after: 3 days
2006-02-02 09:02:16 +00:00
if (loopid >= MAX_FC_TARG) {
loopid = MAX_FC_TARG;
break;
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_prt(isp, ISP_LOGINFO, retained,
loopid, (int) (lp - fcp->portdb),
lp->portid);
} else {
loopid = MAX_FC_TARG;
break;
}
/* FALLTHROUGH */
case MBOX_COMMAND_COMPLETE:
lp->loggedin = 1;
lp->loopid = loopid;
break;
case MBOX_COMMAND_ERROR:
isp_prt(isp, ISP_LOGINFO, plogierr,
portid, mbs.param[1]);
/* FALLTHROUGH */
case MBOX_ALL_IDS_USED: /* We're outta IDs */
default:
loopid = MAX_FC_TARG;
break;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
} while (lp->loopid == FL_PORT_ID && loopid < MAX_FC_TARG);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If we get here and we haven't set a Loop ID,
* we failed to log into this device.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (lp->loopid == FL_PORT_ID) {
lp->loopid = 0;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
continue;
}
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Make sure we can get the approriate port information.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (isp_getpdb(isp, lp->loopid, &pdb) != 0) {
isp_prt(isp, ISP_LOGWARN, nopdb, lp->portid);
goto dump_em;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_SYNCING_PDB) {
return (-1);
}
if (pdb.pdb_loopid != lp->loopid) {
isp_prt(isp, ISP_LOGWARN, pdbmfail1,
lp->portid, pdb.pdb_loopid);
goto dump_em;
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if (lp->portid != (uint32_t) BITS2WORD(pdb.pdb_portid_bits)) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_prt(isp, ISP_LOGWARN, pdbmfail2,
lp->portid, BITS2WORD(pdb.pdb_portid_bits));
goto dump_em;
}
lp->roles =
(pdb.pdb_prli_svc3 & SVC3_ROLE_MASK) >> SVC3_ROLE_SHIFT;
lp->node_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_nodename[0]) << 56) |
(((uint64_t)pdb.pdb_nodename[1]) << 48) |
(((uint64_t)pdb.pdb_nodename[2]) << 40) |
(((uint64_t)pdb.pdb_nodename[3]) << 32) |
(((uint64_t)pdb.pdb_nodename[4]) << 24) |
(((uint64_t)pdb.pdb_nodename[5]) << 16) |
(((uint64_t)pdb.pdb_nodename[6]) << 8) |
(((uint64_t)pdb.pdb_nodename[7]));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
lp->port_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_portname[0]) << 56) |
(((uint64_t)pdb.pdb_portname[1]) << 48) |
(((uint64_t)pdb.pdb_portname[2]) << 40) |
(((uint64_t)pdb.pdb_portname[3]) << 32) |
(((uint64_t)pdb.pdb_portname[4]) << 24) |
(((uint64_t)pdb.pdb_portname[5]) << 16) |
(((uint64_t)pdb.pdb_portname[6]) << 8) |
(((uint64_t)pdb.pdb_portname[7]));
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Check to make sure this all makes sense.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (lp->node_wwn && lp->port_wwn) {
lp->valid = 1;
loopid = lp - fcp->portdb;
(void) isp_async(isp, ISPASYNC_PROMENADE, &loopid);
continue;
}
dump_em:
lp->valid = 0;
isp_prt(isp, ISP_LOGINFO,
ldumped, loopid, lp->loopid, lp->portid);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbs.param[0] = MBOX_FABRIC_LOGOUT;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
mbs.param[1] = lp->loopid;
mbs.obits |= (1 << 10);
} else {
mbs.param[1] = lp->loopid << 8;
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_SYNCING_PDB) {
return (-1);
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If we get here, we've for sure seen not only a valid loop
* but know what is or isn't on it, so mark this for usage
* in isp_start.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->loop_seen_once = 1;
fcp->isp_loopstate = LOOP_READY;
return (0);
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_scan_loop(ispsoftc_t *isp)
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
{
struct lportdb *lp;
fcparam *fcp = isp->isp_param;
isp_pdb_t pdb;
int loopid, lim, hival;
switch (fcp->isp_topo) {
case TOPO_NL_PORT:
hival = FL_PORT_ID;
break;
case TOPO_N_PORT:
hival = 2;
break;
case TOPO_FL_PORT:
hival = FC_PORT_ID;
break;
default:
What the heck - make the last (most recent) 2200 f/w also do Hard Loop acquisition.
2006-07-03 20:56:48 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "no loop topology to scan");
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate = LOOP_LSCAN_DONE;
return (0);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate = LOOP_SCANNING_LOOP;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "scanning local loop 0..%d", hival);
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* make sure the temp port database is clean...
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
MEMZERO((void *)fcp->tport, sizeof (fcp->tport));
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Run through the local loop ports and get port database info
* for each loop ID.
*
* There's a somewhat unexplained situation where the f/w passes back
* the wrong database entity- if that happens, just restart (up to
* FL_PORT_ID times).
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
for (lim = loopid = 0; loopid < hival; loopid++) {
lp = &fcp->tport[loopid];
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* Don't even try for ourselves...
*/
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (loopid == fcp->isp_loopid) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
continue;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_2100(isp) || IS_2200(isp)) {
lp->node_wwn = isp_get_portname(isp, loopid, 1);
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP) {
return (-1);
}
if (lp->node_wwn == 0) {
continue;
}
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Get an entry....
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (isp_getpdb(isp, loopid, &pdb) != 0) {
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
return (-1);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
continue;
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP) {
return (-1);
}
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If the returned database element doesn't match what we
* asked for, restart the process entirely (up to a point...).
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (pdb.pdb_loopid != loopid) {
loopid = 0;
if (lim++ < hival) {
continue;
}
isp_prt(isp, ISP_LOGWARN,
"giving up on synchronizing the port database");
return (-1);
}
/*
* Save the pertinent info locally.
*/
lp->node_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_nodename[0]) << 56) |
(((uint64_t)pdb.pdb_nodename[1]) << 48) |
(((uint64_t)pdb.pdb_nodename[2]) << 40) |
(((uint64_t)pdb.pdb_nodename[3]) << 32) |
(((uint64_t)pdb.pdb_nodename[4]) << 24) |
(((uint64_t)pdb.pdb_nodename[5]) << 16) |
(((uint64_t)pdb.pdb_nodename[6]) << 8) |
(((uint64_t)pdb.pdb_nodename[7]));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
lp->port_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)pdb.pdb_portname[0]) << 56) |
(((uint64_t)pdb.pdb_portname[1]) << 48) |
(((uint64_t)pdb.pdb_portname[2]) << 40) |
(((uint64_t)pdb.pdb_portname[3]) << 32) |
(((uint64_t)pdb.pdb_portname[4]) << 24) |
(((uint64_t)pdb.pdb_portname[5]) << 16) |
(((uint64_t)pdb.pdb_portname[6]) << 8) |
(((uint64_t)pdb.pdb_portname[7]));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
lp->roles =
(pdb.pdb_prli_svc3 & SVC3_ROLE_MASK) >> SVC3_ROLE_SHIFT;
lp->portid = BITS2WORD(pdb.pdb_portid_bits);
lp->loopid = pdb.pdb_loopid;
}
/*
* Mark all of the permanent local loop database entries as invalid
* (except our own entry).
*/
for (loopid = 0; loopid < hival; loopid++) {
if (loopid == fcp->isp_iid) {
fcp->portdb[loopid].valid = 1;
fcp->portdb[loopid].loopid = fcp->isp_loopid;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
continue;
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->portdb[loopid].valid = 0;
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* Now merge our local copy of the port database into our saved copy.
* Notify the outer layers of new devices arriving.
*/
for (loopid = 0; loopid < hival; loopid++) {
int i;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If we don't have a non-zero Port WWN, we're not here.
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->tport[loopid].port_wwn == 0) {
continue;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Skip ourselves.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (loopid == fcp->isp_iid) {
continue;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* For the purposes of deciding whether this is the
* 'same' device or not, we only search for an identical
* Port WWN. Node WWNs may or may not be the same as
* the Port WWN, and there may be multiple different
* Port WWNs with the same Node WWN. It would be chaos
* to have multiple identical Port WWNs, so we don't
* allow that.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
for (i = 0; i < hival; i++) {
int j;
if (fcp->portdb[i].port_wwn == 0)
continue;
if (fcp->portdb[i].port_wwn !=
fcp->tport[loopid].port_wwn)
continue;
/*
* We found this WWN elsewhere- it's changed
* loopids then. We don't change it's actual
* position in our cached port database- we
* just change the actual loop ID we'd use.
*/
if (fcp->portdb[i].loopid != loopid) {
isp_prt(isp, ISP_LOGINFO, portshift, i,
fcp->portdb[i].loopid,
fcp->portdb[i].portid, loopid,
fcp->tport[loopid].portid);
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->portdb[i].portid = fcp->tport[loopid].portid;
fcp->portdb[i].loopid = loopid;
fcp->portdb[i].valid = 1;
fcp->portdb[i].roles = fcp->tport[loopid].roles;
/*
* Now make sure this Port WWN doesn't exist elsewhere
* in the port database.
*/
for (j = i+1; j < hival; j++) {
if (fcp->portdb[i].port_wwn !=
fcp->portdb[j].port_wwn) {
continue;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_prt(isp, ISP_LOGWARN, portdup, j, i);
/*
* Invalidate the 'old' *and* 'new' ones.
* This is really harsh and not quite right,
* but if this happens, we really don't know
* who is what at this point.
*/
fcp->portdb[i].valid = 0;
fcp->portdb[j].valid = 0;
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
break;
}
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* If we didn't traverse the entire port database,
* then we found (and remapped) an existing entry.
* No need to notify anyone- go for the next one.
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (i < hival) {
Fix botch for state levels. Role minor release. Start adding code for a 'force logout' path. MFC after: 4 weeks
2001-06-05 17:11:06 +00:00
isp_prt(isp, ISP_LOGINFO, retained,
fcp->portdb[i].loopid, i, fcp->portdb[i].portid);
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
continue;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
}
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* We've not found this Port WWN anywhere. It's a new entry.
* See if we can leave it where it is (with target == loopid).
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->portdb[loopid].port_wwn != 0) {
for (lim = 0; lim < hival; lim++) {
if (fcp->portdb[lim].port_wwn == 0)
break;
}
/* "Cannot Happen" */
if (lim == hival) {
isp_prt(isp, ISP_LOGWARN, "Remap Overflow");
continue;
}
i = lim;
} else {
i = loopid;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* NB: The actual loopid we use here is loopid- we may
* in fact be at a completely different index (target).
*/
fcp->portdb[i].loopid = loopid;
fcp->portdb[i].port_wwn = fcp->tport[loopid].port_wwn;
fcp->portdb[i].node_wwn = fcp->tport[loopid].node_wwn;
fcp->portdb[i].roles = fcp->tport[loopid].roles;
fcp->portdb[i].portid = fcp->tport[loopid].portid;
fcp->portdb[i].valid = 1;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Tell the outside world we've arrived.
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
(void) isp_async(isp, ISPASYNC_PROMENADE, &i);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Now find all previously used targets that are now invalid and
* notify the outer layers that they're gone.
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
for (lp = &fcp->portdb[0]; lp < &fcp->portdb[hival]; lp++) {
if (lp->valid || lp->port_wwn == 0) {
continue;
}
/*
* Tell the outside world we've gone
* away and erase our pdb entry.
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
*/
loopid = lp - fcp->portdb;
(void) isp_async(isp, ISPASYNC_PROMENADE, &loopid);
MEMZERO((void *) lp, sizeof (*lp));
}
fcp->isp_loopstate = LOOP_LSCAN_DONE;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
return (0);
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_fabric_mbox_cmd(ispsoftc_t *isp, mbreg_t *mbp)
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
{
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
/* the caller sets up the mailbox */
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_mboxcmd(isp, mbp, MBLOGNONE);
if (mbp->param[0] != MBOX_COMMAND_COMPLETE) {
if (FCPARAM(isp)->isp_loopstate == LOOP_SCANNING_FABRIC) {
FCPARAM(isp)->isp_loopstate = LOOP_PDB_RCVD;
}
if (mbp->param[0] == MBOX_COMMAND_ERROR) {
char tbuf[16];
char *m;
switch (mbp->param[1]) {
case 1:
m = "No Loop";
break;
case 2:
m = "Failed to allocate IOCB buffer";
break;
case 3:
m = "Failed to allocate XCB buffer";
break;
case 4:
m = "timeout or transmit failed";
break;
case 5:
m = "no fabric loop";
break;
case 6:
m = "remote device not a target";
break;
default:
SNPRINTF(tbuf, sizeof tbuf, "%x",
mbp->param[1]);
m = tbuf;
break;
}
isp_prt(isp, ISP_LOGERR, "SNS Failed- %s", m);
}
return (-1);
}
if (FCPARAM(isp)->isp_fwstate != FW_READY ||
FCPARAM(isp)->isp_loopstate < LOOP_SCANNING_FABRIC) {
return (-1);
}
return(0);
}
#ifdef ISP_USE_GA_NXT
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_scan_fabric(ispsoftc_t *isp, int ftype)
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
{
fcparam *fcp = isp->isp_param;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint32_t portid, first_portid, last_portid;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
int hicap, last_port_same;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_onfabric == 0) {
fcp->isp_loopstate = LOOP_FSCAN_DONE;
return (0);
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "scanning fabric (GA_NXT)");
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
FC_SCRATCH_ACQUIRE(isp);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* Since Port IDs are 24 bits, we can check against having seen
* anything yet with this value.
*/
Protect against deranged fabric nameservers that spit out 10000 identical port numbers. MFC after: 1 day
2001-10-18 17:26:52 +00:00
last_port_same = 0;
last_portid = 0xffffffff; /* not a port */
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
first_portid = portid = fcp->isp_portid;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate = LOOP_SCANNING_FABRIC;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
for (hicap = 0; hicap < GA_NXT_MAX; hicap++) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
mbreg_t mbs;
sns_screq_t *rq;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
sns_ga_nxt_rsp_t *rs0, *rs1;
struct lportdb lcl;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t sc[SNS_GA_NXT_RESP_SIZE];
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
rq = (sns_screq_t *)sc;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMZERO((void *) rq, SNS_GA_NXT_REQ_SIZE);
rq->snscb_rblen = SNS_GA_NXT_RESP_SIZE >> 1;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
rq->snscb_addr[RQRSP_ADDR0015] = DMA_WD0(fcp->isp_scdma+0x100);
rq->snscb_addr[RQRSP_ADDR1631] = DMA_WD1(fcp->isp_scdma+0x100);
rq->snscb_addr[RQRSP_ADDR3247] = DMA_WD2(fcp->isp_scdma+0x100);
rq->snscb_addr[RQRSP_ADDR4863] = DMA_WD3(fcp->isp_scdma+0x100);
rq->snscb_sblen = 6;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
rq->snscb_data[0] = SNS_GA_NXT;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
rq->snscb_data[4] = portid & 0xffff;
rq->snscb_data[5] = (portid >> 16) & 0xff;
isp_put_sns_request(isp, rq, (sns_screq_t *) fcp->isp_scratch);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_GA_NXT_REQ_SIZE);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
mbs.param[0] = MBOX_SEND_SNS;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
mbs.param[1] = SNS_GA_NXT_REQ_SIZE >> 1;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
/*
* Leave 4 and 5 alone
*/
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
if (isp_fabric_mbox_cmd(isp, &mbs)) {
if (fcp->isp_loopstate >= LOOP_SCANNING_FABRIC) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate = LOOP_PDB_RCVD;
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
FC_SCRATCH_RELEASE(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
return (-1);
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMORYBARRIER(isp, SYNC_SFORCPU, 0x100, SNS_GA_NXT_RESP_SIZE);
rs1 = (sns_ga_nxt_rsp_t *) sc;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
rs0 = (sns_ga_nxt_rsp_t *) ((uint8_t *)fcp->isp_scratch+0x100);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_get_ga_nxt_response(isp, rs0, rs1);
if (rs1->snscb_cthdr.ct_response != FS_ACC) {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
int level;
if (rs1->snscb_cthdr.ct_reason == 9 &&
rs1->snscb_cthdr.ct_explanation == 7)
level = ISP_LOGDEBUG0;
else
level = ISP_LOGWARN;
isp_prt(isp, level, swrej, "GA_NXT",
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
rs1->snscb_cthdr.ct_reason,
rs1->snscb_cthdr.ct_explanation, portid);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
fcp->isp_loopstate = LOOP_FSCAN_DONE;
return (0);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
portid =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint32_t) rs1->snscb_port_id[0]) << 16) |
(((uint32_t) rs1->snscb_port_id[1]) << 8) |
(((uint32_t) rs1->snscb_port_id[2]));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
/*
* XXX: We should check to make sure that this entry
* XXX: supports the type(s) we are interested in.
*/
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
/*
* Okay, we now have information about a fabric object.
* If it is the type we're interested in, tell the outer layers
* about it. The outer layer needs to know: Port ID, WWNN,
* WWPN, FC4 type, and port type.
*
* The lportdb structure is adequate for this.
*/
MEMZERO(&lcl, sizeof (lcl));
lcl.port_type = rs1->snscb_port_type;
lcl.fc4_type = ftype;
lcl.portid = portid;
lcl.node_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)rs1->snscb_nodename[0]) << 56) |
(((uint64_t)rs1->snscb_nodename[1]) << 48) |
(((uint64_t)rs1->snscb_nodename[2]) << 40) |
(((uint64_t)rs1->snscb_nodename[3]) << 32) |
(((uint64_t)rs1->snscb_nodename[4]) << 24) |
(((uint64_t)rs1->snscb_nodename[5]) << 16) |
(((uint64_t)rs1->snscb_nodename[6]) << 8) |
(((uint64_t)rs1->snscb_nodename[7]));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
lcl.port_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)rs1->snscb_portname[0]) << 56) |
(((uint64_t)rs1->snscb_portname[1]) << 48) |
(((uint64_t)rs1->snscb_portname[2]) << 40) |
(((uint64_t)rs1->snscb_portname[3]) << 32) |
(((uint64_t)rs1->snscb_portname[4]) << 24) |
(((uint64_t)rs1->snscb_portname[5]) << 16) |
(((uint64_t)rs1->snscb_portname[6]) << 8) |
(((uint64_t)rs1->snscb_portname[7]));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
/*
* Does this fabric object support the type we want?
* If not, skip it.
*/
if (rs1->snscb_fc4_types[ftype >> 5] & (1 << (ftype & 0x1f))) {
if (first_portid == portid) {
lcl.last_fabric_dev = 1;
} else {
lcl.last_fabric_dev = 0;
}
(void) isp_async(isp, ISPASYNC_FABRIC_DEV, &lcl);
} else {
isp_prt(isp, ISP_LOGDEBUG0,
"PortID 0x%x doesn't support FC4 type 0x%x",
portid, ftype);
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
if (first_portid == portid) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate = LOOP_FSCAN_DONE;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
FC_SCRATCH_RELEASE(isp);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
return (0);
}
Protect against deranged fabric nameservers that spit out 10000 identical port numbers. MFC after: 1 day
2001-10-18 17:26:52 +00:00
if (portid == last_portid) {
if (last_port_same++ > 20) {
isp_prt(isp, ISP_LOGWARN,
"tangled fabric database detected");
break;
}
} else {
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
last_port_same = 0 ;
Protect against deranged fabric nameservers that spit out 10000 identical port numbers. MFC after: 1 day
2001-10-18 17:26:52 +00:00
last_portid = portid;
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
FC_SCRATCH_RELEASE(isp);
if (hicap >= GA_NXT_MAX) {
isp_prt(isp, ISP_LOGWARN, "fabric too big (> %d)", GA_NXT_MAX);
}
fcp->isp_loopstate = LOOP_FSCAN_DONE;
return (0);
}
#else
#define GIDLEN ((ISP2100_SCRLEN >> 1) + 16)
#define NGENT ((GIDLEN - 16) >> 2)
#define IGPOFF (ISP2100_SCRLEN - GIDLEN)
#define GXOFF (256)
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_scan_fabric(ispsoftc_t *isp, int ftype)
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
{
fcparam *fcp = FCPARAM(isp);
mbreg_t mbs;
int i;
sns_gid_ft_req_t *rq;
sns_gid_ft_rsp_t *rs0, *rs1;
if (fcp->isp_onfabric == 0) {
fcp->isp_loopstate = LOOP_FSCAN_DONE;
return (0);
Protect against deranged fabric nameservers that spit out 10000 identical port numbers. MFC after: 1 day
2001-10-18 17:26:52 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "scanning fabric (GID_FT)");
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
FC_SCRATCH_ACQUIRE(isp);
fcp->isp_loopstate = LOOP_SCANNING_FABRIC;
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
rq = (sns_gid_ft_req_t *)fcp->tport;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMZERO((void *) rq, SNS_GID_FT_REQ_SIZE);
rq->snscb_rblen = GIDLEN >> 1;
rq->snscb_addr[RQRSP_ADDR0015] = DMA_WD0(fcp->isp_scdma+IGPOFF);
rq->snscb_addr[RQRSP_ADDR1631] = DMA_WD1(fcp->isp_scdma+IGPOFF);
rq->snscb_addr[RQRSP_ADDR3247] = DMA_WD2(fcp->isp_scdma+IGPOFF);
rq->snscb_addr[RQRSP_ADDR4863] = DMA_WD3(fcp->isp_scdma+IGPOFF);
rq->snscb_sblen = 6;
rq->snscb_cmd = SNS_GID_FT;
rq->snscb_mword_div_2 = NGENT;
rq->snscb_fc4_type = ftype;
isp_put_gid_ft_request(isp, rq, (sns_gid_ft_req_t *) fcp->isp_scratch);
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_GID_FT_REQ_SIZE);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
mbs.param[0] = MBOX_SEND_SNS;
mbs.param[1] = SNS_GID_FT_REQ_SIZE >> 1;
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
* Leave 4 and 5 alone
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
if (isp_fabric_mbox_cmd(isp, &mbs)) {
if (fcp->isp_loopstate >= LOOP_SCANNING_FABRIC) {
fcp->isp_loopstate = LOOP_PDB_RCVD;
}
FC_SCRATCH_RELEASE(isp);
return (-1);
}
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
MEMORYBARRIER(isp, SYNC_SFORCPU, IGPOFF, GIDLEN);
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
rs1 = (sns_gid_ft_rsp_t *) fcp->tport;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
rs0 = (sns_gid_ft_rsp_t *) ((uint8_t *)fcp->isp_scratch+IGPOFF);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_get_gid_ft_response(isp, rs0, rs1, NGENT);
if (rs1->snscb_cthdr.ct_response != FS_ACC) {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
int level;
if (rs1->snscb_cthdr.ct_reason == 9 &&
rs1->snscb_cthdr.ct_explanation == 7)
level = ISP_LOGDEBUG0;
else
level = ISP_LOGWARN;
isp_prt(isp, level, swrej, "GID_FT",
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
rs1->snscb_cthdr.ct_reason,
rs1->snscb_cthdr.ct_explanation, 0);
FC_SCRATCH_RELEASE(isp);
fcp->isp_loopstate = LOOP_FSCAN_DONE;
return (0);
}
/*
* Okay, we now have a list of Port IDs for this class of device.
* Go through the list and for each one get the WWPN/WWNN for it
* and tell the outer layers about it. The outer layer needs to
* know: Port ID, WWNN, WWPN, FC4 type, and (possibly) port type.
*
* The lportdb structure is adequate for this.
*/
i = -1;
do {
sns_gxn_id_req_t grqbuf, *gq = &grqbuf;
sns_gxn_id_rsp_t *gs0, grsbuf, *gs1 = &grsbuf;
struct lportdb lcl;
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
#if 0
sns_gff_id_rsp_t *fs0, ffsbuf, *fs1 = &ffsbuf;
#endif
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
i++;
MEMZERO(&lcl, sizeof (lcl));
lcl.fc4_type = ftype;
lcl.portid =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint32_t) rs1->snscb_ports[i].portid[0]) << 16) |
(((uint32_t) rs1->snscb_ports[i].portid[1]) << 8) |
(((uint32_t) rs1->snscb_ports[i].portid[2]));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMZERO((void *) gq, sizeof (sns_gxn_id_req_t));
gq->snscb_rblen = SNS_GXN_ID_RESP_SIZE >> 1;
gq->snscb_addr[RQRSP_ADDR0015] = DMA_WD0(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR1631] = DMA_WD1(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR3247] = DMA_WD2(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR4863] = DMA_WD3(fcp->isp_scdma+GXOFF);
gq->snscb_sblen = 6;
gq->snscb_cmd = SNS_GPN_ID;
gq->snscb_portid = lcl.portid;
isp_put_gxn_id_request(isp, gq,
(sns_gxn_id_req_t *) fcp->isp_scratch);
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_GXN_ID_REQ_SIZE);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
mbs.param[0] = MBOX_SEND_SNS;
mbs.param[1] = SNS_GXN_ID_REQ_SIZE >> 1;
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
/*
* Leave 4 and 5 alone
*/
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
if (isp_fabric_mbox_cmd(isp, &mbs)) {
if (fcp->isp_loopstate >= LOOP_SCANNING_FABRIC) {
fcp->isp_loopstate = LOOP_PDB_RCVD;
}
FC_SCRATCH_RELEASE(isp);
return (-1);
}
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
MEMORYBARRIER(isp, SYNC_SFORCPU, GXOFF, SNS_GXN_ID_RESP_SIZE);
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
gs0 = (sns_gxn_id_rsp_t *) ((uint8_t *)fcp->isp_scratch+GXOFF);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_get_gxn_id_response(isp, gs0, gs1);
if (gs1->snscb_cthdr.ct_response != FS_ACC) {
isp_prt(isp, ISP_LOGWARN, swrej, "GPN_ID",
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
gs1->snscb_cthdr.ct_reason,
gs1->snscb_cthdr.ct_explanation, lcl.portid);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
continue;
}
lcl.port_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)gs1->snscb_wwn[0]) << 56) |
(((uint64_t)gs1->snscb_wwn[1]) << 48) |
(((uint64_t)gs1->snscb_wwn[2]) << 40) |
(((uint64_t)gs1->snscb_wwn[3]) << 32) |
(((uint64_t)gs1->snscb_wwn[4]) << 24) |
(((uint64_t)gs1->snscb_wwn[5]) << 16) |
(((uint64_t)gs1->snscb_wwn[6]) << 8) |
(((uint64_t)gs1->snscb_wwn[7]));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMZERO((void *) gq, sizeof (sns_gxn_id_req_t));
gq->snscb_rblen = SNS_GXN_ID_RESP_SIZE >> 1;
gq->snscb_addr[RQRSP_ADDR0015] = DMA_WD0(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR1631] = DMA_WD1(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR3247] = DMA_WD2(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR4863] = DMA_WD3(fcp->isp_scdma+GXOFF);
gq->snscb_sblen = 6;
gq->snscb_cmd = SNS_GNN_ID;
gq->snscb_portid = lcl.portid;
isp_put_gxn_id_request(isp, gq,
(sns_gxn_id_req_t *) fcp->isp_scratch);
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_GXN_ID_REQ_SIZE);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
mbs.param[0] = MBOX_SEND_SNS;
mbs.param[1] = SNS_GXN_ID_REQ_SIZE >> 1;
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
/*
* Leave 4 and 5 alone
*/
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
if (isp_fabric_mbox_cmd(isp, &mbs)) {
if (fcp->isp_loopstate >= LOOP_SCANNING_FABRIC) {
fcp->isp_loopstate = LOOP_PDB_RCVD;
}
FC_SCRATCH_RELEASE(isp);
return (-1);
}
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
MEMORYBARRIER(isp, SYNC_SFORCPU, GXOFF, SNS_GXN_ID_RESP_SIZE);
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
gs0 = (sns_gxn_id_rsp_t *) ((uint8_t *)fcp->isp_scratch+GXOFF);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_get_gxn_id_response(isp, gs0, gs1);
if (gs1->snscb_cthdr.ct_response != FS_ACC) {
isp_prt(isp, ISP_LOGWARN, swrej, "GNN_ID",
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
gs1->snscb_cthdr.ct_reason,
gs1->snscb_cthdr.ct_explanation, lcl.portid);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
continue;
}
lcl.node_wwn =
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(((uint64_t)gs1->snscb_wwn[0]) << 56) |
(((uint64_t)gs1->snscb_wwn[1]) << 48) |
(((uint64_t)gs1->snscb_wwn[2]) << 40) |
(((uint64_t)gs1->snscb_wwn[3]) << 32) |
(((uint64_t)gs1->snscb_wwn[4]) << 24) |
(((uint64_t)gs1->snscb_wwn[5]) << 16) |
(((uint64_t)gs1->snscb_wwn[6]) << 8) |
(((uint64_t)gs1->snscb_wwn[7]));
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
/*
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
* The QLogic f/w is bouncing this with a parameter error.
*/
#if 0
/*
* Try and get FC4 Features (FC-GS-3 only).
* We can use the sns_gxn_id_req_t for this request.
*/
MEMZERO((void *) gq, sizeof (sns_gxn_id_req_t));
gq->snscb_rblen = SNS_GFF_ID_RESP_SIZE >> 1;
gq->snscb_addr[RQRSP_ADDR0015] = DMA_WD0(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR1631] = DMA_WD1(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR3247] = DMA_WD2(fcp->isp_scdma+GXOFF);
gq->snscb_addr[RQRSP_ADDR4863] = DMA_WD3(fcp->isp_scdma+GXOFF);
gq->snscb_sblen = 6;
gq->snscb_cmd = SNS_GFF_ID;
gq->snscb_portid = lcl.portid;
isp_put_gxn_id_request(isp, gq,
(sns_gxn_id_req_t *) fcp->isp_scratch);
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_GXN_ID_REQ_SIZE);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
mbs.param[0] = MBOX_SEND_SNS;
mbs.param[1] = SNS_GXN_ID_REQ_SIZE >> 1;
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
/*
* Leave 4 and 5 alone
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
*/
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
if (isp_fabric_mbox_cmd(isp, &mbs)) {
if (fcp->isp_loopstate >= LOOP_SCANNING_FABRIC) {
fcp->isp_loopstate = LOOP_PDB_RCVD;
}
FC_SCRATCH_RELEASE(isp);
return (-1);
}
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
MEMORYBARRIER(isp, SYNC_SFORCPU, GXOFF, SNS_GFF_ID_RESP_SIZE);
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
fs0 = (sns_gff_id_rsp_t *) ((uint8_t *)fcp->isp_scratch+GXOFF);
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
isp_get_gff_id_response(isp, fs0, fs1);
if (fs1->snscb_cthdr.ct_response != FS_ACC) {
isp_prt(isp, /* ISP_LOGDEBUG0 */ ISP_LOGWARN,
swrej, "GFF_ID",
fs1->snscb_cthdr.ct_reason,
fs1->snscb_cthdr.ct_explanation, lcl.portid);
if (fcp->isp_loopstate != LOOP_SCANNING_FABRIC) {
FC_SCRATCH_RELEASE(isp);
return (-1);
}
} else {
int index = (ftype >> 3);
int bshft = (ftype & 0x7) * 4;
int fc4_fval =
(fs1->snscb_fc4_features[index] >> bshft) & 0xf;
if (fc4_fval & 0x1) {
lcl.roles |=
(SVC3_INI_ROLE >> SVC3_ROLE_SHIFT);
}
if (fc4_fval & 0x2) {
lcl.roles |=
(SVC3_TGT_ROLE >> SVC3_ROLE_SHIFT);
}
}
#endif
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
/*
* If we really want to know what kind of port type this is,
* we have to run another CT command. Otherwise, we'll leave
* it as undefined.
*
lcl.port_type = 0;
*/
if (rs1->snscb_ports[i].control & 0x80) {
lcl.last_fabric_dev = 1;
} else {
lcl.last_fabric_dev = 0;
}
(void) isp_async(isp, ISPASYNC_FABRIC_DEV, &lcl);
} while ((rs1->snscb_ports[i].control & 0x80) == 0 && i < NGENT-1);
/*
* If we're not at the last entry, our list isn't big enough.
*/
if ((rs1->snscb_ports[i].control & 0x80) == 0) {
isp_prt(isp, ISP_LOGWARN, "fabric too big for scratch area");
}
FC_SCRATCH_RELEASE(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
fcp->isp_loopstate = LOOP_FSCAN_DONE;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
return (0);
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
#endif
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_register_fc4_type(ispsoftc_t *isp)
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
{
fcparam *fcp = isp->isp_param;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t local[SNS_RFT_ID_REQ_SIZE];
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
sns_screq_t *reqp = (sns_screq_t *) local;
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
mbreg_t mbs;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
MEMZERO((void *) reqp, SNS_RFT_ID_REQ_SIZE);
reqp->snscb_rblen = SNS_RFT_ID_RESP_SIZE >> 1;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
reqp->snscb_addr[RQRSP_ADDR0015] = DMA_WD0(fcp->isp_scdma + 0x100);
reqp->snscb_addr[RQRSP_ADDR1631] = DMA_WD1(fcp->isp_scdma + 0x100);
reqp->snscb_addr[RQRSP_ADDR3247] = DMA_WD2(fcp->isp_scdma + 0x100);
reqp->snscb_addr[RQRSP_ADDR4863] = DMA_WD3(fcp->isp_scdma + 0x100);
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
reqp->snscb_sblen = 22;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
reqp->snscb_data[0] = SNS_RFT_ID;
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
reqp->snscb_data[4] = fcp->isp_portid & 0xffff;
reqp->snscb_data[5] = (fcp->isp_portid >> 16) & 0xff;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
reqp->snscb_data[6] = (1 << FC4_SCSI);
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
#if 0
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
reqp->snscb_data[6] |= (1 << FC4_IP); /* ISO/IEC 8802-2 LLC/SNAP */
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
#endif
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_ACQUIRE(isp);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_put_sns_request(isp, reqp, (sns_screq_t *) fcp->isp_scratch);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
mbs.param[0] = MBOX_SEND_SNS;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
mbs.param[1] = SNS_RFT_ID_REQ_SIZE >> 1;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
/*
* Leave 4 and 5 alone
*/
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_RFT_ID_REQ_SIZE);
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
FC_SCRATCH_RELEASE(isp);
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "Register FC4 types succeeded");
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
}
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* Start a command. Locking is assumed done in the caller.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
int
In order to save ourselves grief with the SUNPRO compiler under Solaris (which, for reasons unknown to me, chokes on u_int16_t as a typedef of unsigned short if used in a transitional (mixed K&R and ANSI) way), we'll go the extra mile and fully ANSIfy things.
2001-03-14 04:11:56 +00:00
isp_start(XS_T *xs)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
ispsoftc_t *isp;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t nxti, optr, handle;
uint8_t local[QENTRY_LEN];
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ispreq_t *reqp, *qep;
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
int target, i;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
XS_INITERR(xs);
isp = XS_ISP(xs);
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* Check to make sure we're supporting initiator role.
*/
if ((isp->isp_role & ISP_ROLE_INITIATOR) == 0) {
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
/*
* Now make sure we're running.
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
if (isp->isp_state != ISP_RUNSTATE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Adapter not at RUNSTATE");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
XS_SETERR(xs, HBA_BOTCH);
return (CMD_COMPLETE);
}
/*
Clean out residual bogosity for fast posting stuff- ISP_NO_FASTPOST_SCSI is gone as a define. We just don't support fast posting for anything less than the 1240/1080/1280/12160 or Fibre Channel cards. Put in support for CDB's larger than 12 bytes for parallel SCSI (up to 44 bytes are allowed). Approved: jkh
2000-02-15 00:35:00 +00:00
* Check command CDB length, etc.. We really are limited to 16 bytes
* for Fibre Channel, but can do up to 44 bytes in parallel SCSI,
* but probably only if we're running fairly new firmware (we'll
* let the old f/w choke on an extended command queue entry).
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
Clean out residual bogosity for fast posting stuff- ISP_NO_FASTPOST_SCSI is gone as a define. We just don't support fast posting for anything less than the 1240/1080/1280/12160 or Fibre Channel cards. Put in support for CDB's larger than 12 bytes for parallel SCSI (up to 44 bytes are allowed). Approved: jkh
2000-02-15 00:35:00 +00:00
if (XS_CDBLEN(xs) > (IS_FC(isp)? 16 : 44) || XS_CDBLEN(xs) == 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"unsupported cdb length (%d, CDB[0]=0x%x)",
XS_CDBLEN(xs), XS_CDBP(xs)[0] & 0xff);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
XS_SETERR(xs, HBA_BOTCH);
return (CMD_COMPLETE);
}
/*
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
* Check to see whether we have good firmware state still or
* need to refresh our port database for this target.
*/
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
target = XS_TGT(xs);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_FC(isp)) {
fcparam *fcp = isp->isp_param;
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
struct lportdb *lp;
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
#ifdef HANDLE_LOOPSTATE_IN_OUTER_LAYERS
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_READY) {
return (CMD_RQLATER);
}
/*
* If we're not on a Fabric, we can't have a target
* above FL_PORT_ID-1.
*
* If we're on a fabric and *not* connected as an F-port,
* we can't have a target less than FC_SNS_ID+1. This
* keeps us from having to sort out the difference between
* local public loop devices and those which we might get
* from a switch's database.
*/
if (fcp->isp_onfabric == 0) {
if (target >= FL_PORT_ID) {
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
} else {
if (target >= FL_PORT_ID && target <= FC_SNS_ID) {
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
Hmm. Let's try this on for size... We originally had it such that if the connection topology was FL-loop (public loop), we never looked at any local loop addresses. The reason for not doing that was fear or concern that we'd see the same local loop disks reflected from the name server and we'd attach them twice. However, when I recently hooked up a JBOD and a system to an ANCOR SA-8 switch, the disks did *not* show up on the fabric. So at least the ANCOR is screening those disks from appearing on the fabric. Now, it's possible this is a 'feature' of the ANCOR. When I get a chance, I'll check the Brocade (it's hard to do this on a low budget). In any case, if they *do* also show up on the fabric, we should simply elect to not log into them because we already have an entry for the local loop. There is relatively unexercised code just for this case. MFC after: 2 weeks
2001-07-11 02:34:21 +00:00
/*
* We used to exclude having local loop ports
* at the same time that we have fabric ports.
* That is, we used to exclude having ports
* at < FL_PORT_ID if we're FL-port.
*
* That's wrong. The only thing that could be
* dicey is if the switch you're connected to
* has these local loop ports appear on the
* fabric and we somehow attach them twice.
*/
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
}
#else
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* Check for f/w being in ready state. If the f/w
* isn't in ready state, then we don't know our
* loop ID and the f/w hasn't completed logging
* into all targets on the loop. If this is the
* case, then bounce the command. We pretend this is
* a SELECTION TIMEOUT error if we've never gone to
* FW_READY state at all- in this case we may not
* be hooked to a loop at all and we shouldn't hang
* the machine for this. Otherwise, defer this command
* until later.
*/
if (fcp->isp_fwstate != FW_READY) {
/*
* Give ourselves at most a 250ms delay.
*/
if (isp_fclink_test(isp, 250000)) {
XS_SETERR(xs, HBA_SELTIMEOUT);
if (fcp->loop_seen_once) {
return (CMD_RQLATER);
} else {
return (CMD_COMPLETE);
}
}
}
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
/*
* If we're not on a Fabric, we can't have a target
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
* above FL_PORT_ID-1.
*
* If we're on a fabric and *not* connected as an F-port,
* we can't have a target less than FC_SNS_ID+1. This
* keeps us from having to sort out the difference between
* local public loop devices and those which we might get
* from a switch's database.
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
*/
if (fcp->isp_onfabric == 0) {
if (target >= FL_PORT_ID) {
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
} else {
if (target >= FL_PORT_ID && target <= FC_SNS_ID) {
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
if (fcp->isp_topo != TOPO_F_PORT &&
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
target < FL_PORT_ID) {
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
}
/*
* If our loop state is such that we haven't yet received
* a "Port Database Changed" notification (after a LIP or
* a Loop Reset or firmware initialization), then defer
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
* sending commands for a little while, but only if we've
* seen a valid loop at one point (otherwise we can get
* stuck at initialization time).
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
if (fcp->isp_loopstate < LOOP_PDB_RCVD) {
XS_SETERR(xs, HBA_SELTIMEOUT);
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
if (fcp->loop_seen_once) {
return (CMD_RQLATER);
} else {
return (CMD_COMPLETE);
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/*
* If we're in the middle of loop or fabric scanning
* or merging the port databases, retry this command later.
*/
if (fcp->isp_loopstate == LOOP_SCANNING_FABRIC ||
fcp->isp_loopstate == LOOP_SCANNING_LOOP ||
fcp->isp_loopstate == LOOP_SYNCING_PDB) {
return (CMD_RQLATER);
}
/*
* If our loop state is now such that we've just now
* received a Port Database Change notification, then
* we have to go off and (re)scan the fabric. We back
* out and try again later if this doesn't work.
*/
if (fcp->isp_loopstate == LOOP_PDB_RCVD && fcp->isp_onfabric) {
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
if (isp_scan_fabric(isp, FC4_SCSI)) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
return (CMD_RQLATER);
}
if (fcp->isp_fwstate != FW_READY ||
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
fcp->isp_loopstate < LOOP_FSCAN_DONE) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
return (CMD_RQLATER);
}
}
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
* If our loop state is now such that we've just now
* received a Port Database Change notification, then
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
* we have to go off and (re)synchronize our port
* database.
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
*/
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (fcp->isp_loopstate < LOOP_READY) {
if (isp_pdb_sync(isp)) {
return (CMD_RQLATER);
}
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate != LOOP_READY) {
return (CMD_RQLATER);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
}
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
/*
* XXX: Here's were we would cancel any loop_dead flag
* XXX: also cancel in dead_loop timeout that's running
*/
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
#endif
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
/*
* Now check whether we should even think about pursuing this.
*/
lp = &fcp->portdb[target];
if (lp->valid == 0) {
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
if ((lp->roles & (SVC3_TGT_ROLE >> SVC3_ROLE_SHIFT)) == 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGDEBUG2,
"Target %d does not have target service", target);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
XS_SETERR(xs, HBA_SELTIMEOUT);
return (CMD_COMPLETE);
}
/*
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
* Now turn target into what the actual Loop ID is.
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
*/
target = lp->loopid;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
/*
* Next check to see if any HBA or Device
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* parameters need to be updated.
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (isp->isp_update != 0) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
isp_update(isp);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Pick up some compilation warning fixes from NetBSD. If we don't have ISP_FW_CRASH_DUMP defined, we have to do a isp_reinit in the core code- not the platform code- so fix the ISP_CONN_FATAL case.
2003-02-16 01:32:52 +00:00
if (isp_getrqentry(isp, &nxti, &optr, (void *)&qep)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "Request Queue Overflow");
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
XS_SETERR(xs, HBA_BOTCH);
return (CMD_EAGAIN);
}
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/*
* Now see if we need to synchronize the ISP with respect to anything.
* We do dual duty here (cough) for synchronizing for busses other
* than which we got here to send a command to.
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
reqp = (ispreq_t *) local;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (isp->isp_sendmarker) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t n = (IS_DUALBUS(isp)? 2: 1);
ISP_DMASETUP now returns a value to be possibly punted to outer layers. Turn request queue overflow messages into debug messages. Ensure on isp_restarts that we nullify the xflist array.
1998-09-17 21:03:45 +00:00
/*
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
* Check ports to send markers for...
ISP_DMASETUP now returns a value to be possibly punted to outer layers. Turn request queue overflow messages into debug messages. Ensure on isp_restarts that we nullify the xflist array.
1998-09-17 21:03:45 +00:00
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
for (i = 0; i < n; i++) {
if ((isp->isp_sendmarker & (1 << i)) == 0) {
continue;
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO((void *) reqp, QENTRY_LEN);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
reqp->req_header.rqs_entry_count = 1;
reqp->req_header.rqs_entry_type = RQSTYPE_MARKER;
reqp->req_modifier = SYNC_ALL;
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
reqp->req_target = i << 7; /* insert bus number */
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_put_request(isp, reqp, qep);
ISP_ADD_REQUEST(isp, nxti);
isp->isp_sendmarker &= ~(1 << i);
Pick up some compilation warning fixes from NetBSD. If we don't have ISP_FW_CRASH_DUMP defined, we have to do a isp_reinit in the core code- not the platform code- so fix the ISP_CONN_FATAL case.
2003-02-16 01:32:52 +00:00
if (isp_getrqentry(isp, &nxti, &optr, (void *) &qep)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
"Request Queue Overflow+");
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
XS_SETERR(xs, HBA_BOTCH);
return (CMD_EAGAIN);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO((void *)reqp, QENTRY_LEN);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
reqp->req_header.rqs_entry_count = 1;
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_FC(isp)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
reqp->req_header.rqs_entry_type = RQSTYPE_T2RQS;
} else {
Clean out residual bogosity for fast posting stuff- ISP_NO_FASTPOST_SCSI is gone as a define. We just don't support fast posting for anything less than the 1240/1080/1280/12160 or Fibre Channel cards. Put in support for CDB's larger than 12 bytes for parallel SCSI (up to 44 bytes are allowed). Approved: jkh
2000-02-15 00:35:00 +00:00
if (XS_CDBLEN(xs) > 12)
reqp->req_header.rqs_entry_type = RQSTYPE_CMDONLY;
else
reqp->req_header.rqs_entry_type = RQSTYPE_REQUEST;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
/* reqp->req_header.rqs_flags = 0; */
/* reqp->req_header.rqs_seqno = 0; */
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_FC(isp)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* See comment in isp_intr
*/
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
/* XS_RESID(xs) = 0; */
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
I was misinformed. I cannot get away from specifying tags for FC. Some devices are happy w/o them- some are unhappy (IBM drives).
1999-10-28 02:48:42 +00:00
* Fibre Channel always requires some kind of tag.
* The Qlogic drivers seem be happy not to use a tag,
* but this breaks for some devices (IBM drives).
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_TAG_P(xs)) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
((ispreqt2_t *)reqp)->req_flags = XS_TAG_TYPE(xs);
I was misinformed. I cannot get away from specifying tags for FC. Some devices are happy w/o them- some are unhappy (IBM drives).
1999-10-28 02:48:42 +00:00
} else {
various fixes
2000-08-27 23:38:44 +00:00
/*
* If we don't know what tag to use, use HEAD OF QUEUE
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
* for Request Sense or Simple.
various fixes
2000-08-27 23:38:44 +00:00
*/
I was misinformed. I cannot get away from specifying tags for FC. Some devices are happy w/o them- some are unhappy (IBM drives).
1999-10-28 02:48:42 +00:00
if (XS_CDBP(xs)[0] == 0x3) /* REQUEST SENSE */
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
((ispreqt2_t *)reqp)->req_flags = REQFLAG_HTAG;
I was misinformed. I cannot get away from specifying tags for FC. Some devices are happy w/o them- some are unhappy (IBM drives).
1999-10-28 02:48:42 +00:00
else
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
((ispreqt2_t *)reqp)->req_flags = REQFLAG_STAG;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
} else {
sdparam *sdp = (sdparam *)isp->isp_param;
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
sdp += XS_CHANNEL(xs);
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
if ((sdp->isp_devparam[target].actv_flags & DPARM_TQING) &&
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_TAG_P(xs)) {
reqp->req_flags = XS_TAG_TYPE(xs);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_SCSI(isp)) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
reqp->req_target = target | (XS_CHANNEL(xs) << 7);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
reqp->req_lun_trn = XS_LUN(xs);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
reqp->req_cdblen = XS_CDBLEN(xs);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
} else if (IS_2KLOGIN(isp)) {
((ispreqt2e_t *)reqp)->req_target = target;
((ispreqt2e_t *)reqp)->req_scclun = XS_LUN(xs);
} else if (FCPARAM(isp)->isp_fwattr & ISP_FW_ATTR_SCCLUN) {
((ispreqt2_t *)reqp)->req_target = target;
((ispreqt2_t *)reqp)->req_scclun = XS_LUN(xs);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
} else {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
((ispreqt2_t *)reqp)->req_target = target;
((ispreqt2_t *)reqp)->req_lun_trn = XS_LUN(xs);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Suggested by bde@freebsd.org- memcpy not necessarily good to use. D'oh- not in the BSD DKI. Stop being lazy and finish the defines so MEMCPY becomes bzero for FreeBSD.
1999-01-10 11:15:23 +00:00
MEMCPY(reqp->req_cdb, XS_CDBP(xs), XS_CDBLEN(xs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
reqp->req_time = XS_TIME(xs) / 1000;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (reqp->req_time == 0 && XS_TIME(xs)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
reqp->req_time = 1;
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
if (isp_save_xs(isp, xs, &handle)) {
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
isp_prt(isp, ISP_LOGDEBUG0, "out of xflist pointers");
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
XS_SETERR(xs, HBA_BOTCH);
return (CMD_EAGAIN);
}
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
reqp->req_handle = handle;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
/*
* Set up DMA and/or do any bus swizzling of the request entry
* so that the Qlogic F/W understands what is being asked of it.
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
*/
i = ISP_DMASETUP(isp, xs, reqp, &nxti, optr);
ISP_DMASETUP now returns a value to be possibly punted to outer layers. Turn request queue overflow messages into debug messages. Ensure on isp_restarts that we nullify the xflist array.
1998-09-17 21:03:45 +00:00
if (i != CMD_QUEUED) {
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
isp_destroy_handle(isp, handle);
ISP_DMASETUP now returns a value to be possibly punted to outer layers. Turn request queue overflow messages into debug messages. Ensure on isp_restarts that we nullify the xflist array.
1998-09-17 21:03:45 +00:00
/*
* dmasetup sets actual error in packet, and
* return what we were given to return.
*/
return (i);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
XS_SETERR(xs, HBA_NOERROR);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGDEBUG2,
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
"START cmd for %d.%d.%d cmd 0x%x datalen %ld",
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs), XS_CDBP(xs)[0],
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
(long) XS_XFRLEN(xs));
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ISP_ADD_REQUEST(isp, nxti);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
isp->isp_nactive++;
return (CMD_QUEUED);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
/*
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* isp control
* Locks (ints blocked) assumed held.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_control(ispsoftc_t *isp, ispctl_t ctl, void *arg)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_T *xs;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
mbreg_t mbs;
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
int bus, tgt;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t handle;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
switch (ctl) {
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Unknown Control Opcode 0x%x", ctl);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ISPCTL_RESET_BUS:
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
* Issue a bus reset.
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
mbs.param[0] = MBOX_BUS_RESET;
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_SCSI(isp)) {
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
mbs.param[1] =
((sdparam *) isp->isp_param)->isp_bus_reset_delay;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (mbs.param[1] < 2) {
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
mbs.param[1] = 2;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
bus = *((int *) arg);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_DUALBUS(isp)) {
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
mbs.param[2] = bus;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
} else {
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
mbs.param[1] = 10;
bus = 0;
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
}
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker |= (1 << bus);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
break;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO,
"driver initiated bus reset of bus %d", bus);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return (0);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
case ISPCTL_RESET_DEV:
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
tgt = (*((int *) arg)) & 0xffff;
bus = (*((int *) arg)) >> 16;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
mbs.param[0] = MBOX_ABORT_TARGET;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_SCSI(isp)) {
mbs.param[1] = (tgt << 8) | (bus << 15);
} else {
if (IS_2KLOGIN(isp)) {
mbs.param[1] = tgt;
mbs.obits |= (1 << 10);
} else {
mbs.param[1] = (tgt << 8);
}
}
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
mbs.param[2] = 3; /* 'delay', in seconds */
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
break;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO,
"Target %d on Bus %d Reset Succeeded", tgt, bus);
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker |= (1 << bus);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return (0);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
case ISPCTL_ABORT_CMD:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xs = (XS_T *) arg;
Clean up ISPCTL_ABORT_CMD function to not be too chatty if it succeeds, or even if it fails with INVALID_PARM (which just means that the handle doesn't refer to an active commane).
2000-07-05 06:41:36 +00:00
tgt = XS_TGT(xs);
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
handle = isp_find_handle(isp, xs);
if (handle == 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"cannot find handle for command to abort");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
bus = XS_CHANNEL(xs);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
mbs.param[0] = MBOX_ABORT;
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_FC(isp)) {
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
if (FCPARAM(isp)->isp_fwattr & ISP_FW_ATTR_SCCLUN) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
if (IS_2KLOGIN(isp)) {
mbs.param[1] = tgt;
} else {
mbs.param[1] = tgt << 8;
}
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
mbs.param[6] = XS_LUN(xs);
} else {
Clean up ISPCTL_ABORT_CMD function to not be too chatty if it succeeds, or even if it fails with INVALID_PARM (which just means that the handle doesn't refer to an active commane).
2000-07-05 06:41:36 +00:00
mbs.param[1] = tgt << 8 | XS_LUN(xs);
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
} else {
mbs.param[1] =
(bus << 15) | (XS_TGT(xs) << 8) | XS_LUN(xs);
}
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
mbs.param[2] = handle;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL & ~MBOX_COMMAND_ERROR);
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
return (0);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
/*
* XXX: Look for command in the REQUEST QUEUE. That is,
* XXX: It hasen't been picked up by firmware yet.
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
break;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case ISPCTL_UPDATE_PARAMS:
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
isp_update(isp);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
return (0);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
case ISPCTL_FCLINK_TEST:
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if (IS_FC(isp)) {
Redo some code based upon issues found by Coverity.
2006-04-21 18:46:35 +00:00
int usdelay = *((int *) arg);
if (usdelay == 0) {
usdelay = 250000;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
return (isp_fclink_test(isp, usdelay));
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
}
break;
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
case ISPCTL_SCAN_FABRIC:
if (IS_FC(isp)) {
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
int ftype = (arg)? *((int *) arg) : FC4_SCSI;
return (isp_scan_fabric(isp, ftype));
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
}
break;
case ISPCTL_SCAN_LOOP:
if (IS_FC(isp)) {
return (isp_scan_loop(isp));
}
break;
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
case ISPCTL_PDB_SYNC:
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (IS_FC(isp)) {
return (isp_pdb_sync(isp));
}
break;
case ISPCTL_SEND_LIP:
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if (IS_FC(isp)) {
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
mbs.param[0] = MBOX_INIT_LIP;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_2KLOGIN(isp)) {
mbs.obits |= (1 << 10);
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
return (0);
}
}
break;
case ISPCTL_GET_POSMAP:
if (IS_FC(isp) && arg) {
return (isp_getmap(isp, arg));
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
}
break;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
Add case to handle ISPCTL_GET_PDB. MFC after: 1 week
2004-02-07 03:42:17 +00:00
case ISPCTL_GET_PDB:
if (IS_FC(isp) && arg) {
int id = *((int *)arg);
isp_pdb_t *pdb = arg;
return (isp_getpdb(isp, id, pdb));
}
break;
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
case ISPCTL_RUN_MBOXCMD:
isp_mboxcmd(isp, arg, MBLOGALL);
return(0);
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#ifdef ISP_TARGET_MODE
case ISPCTL_TOGGLE_TMODE:
various fixes
2000-08-27 23:38:44 +00:00
{
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
/*
* We don't check/set against role here- that's the
* responsibility for the outer layer to coordinate.
*/
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
if (IS_SCSI(isp)) {
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
int param = *(int *)arg;
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
mbs.param[0] = MBOX_ENABLE_TARGET_MODE;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
mbs.param[1] = param & 0xffff;
mbs.param[2] = param >> 16;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
break;
}
}
return (0);
various fixes
2000-08-27 23:38:44 +00:00
}
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
return (-1);
}
/*
* Interrupt Service Routine(s).
*
* External (OS) framework has done the appropriate locking,
* and the locking will be held throughout this function.
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
/*
* Limit our stack depth by sticking with the max likely number
* of completions on a request queue at any one time.
*/
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
#ifndef MAX_REQUESTQ_COMPLETIONS
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
#define MAX_REQUESTQ_COMPLETIONS 64
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
#endif
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_intr(ispsoftc_t *isp, uint16_t isr, uint16_t sema, uint16_t mbox)
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
{
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_T *complist[MAX_REQUESTQ_COMPLETIONS], *xs;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t iptr, optr, junk;
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
int i, nlooked = 0, ndone = 0;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
again:
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
/*
* Is this a mailbox related interrupt?
* The mailbox semaphore will be nonzero if so.
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (sema) {
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (mbox & 0x4000) {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_intmboxc++;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (isp->isp_mboxbsy) {
int i = 0, obits = isp->isp_obits;
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
isp->isp_mboxtmp[i++] = mbox;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 1; i < MAX_MAILBOX(isp); i++) {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if ((obits & (1 << i)) == 0) {
continue;
}
isp->isp_mboxtmp[i] =
ISP_READ(isp, MBOX_OFF(i));
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (isp->isp_mbxwrk0) {
if (isp_mbox_continue(isp) == 0) {
return;
}
}
Change delay loop in new isp_mboxcmd to the use of the new MBOX_WAIT_COMPLETE macro. Change notification of completion of a mailbox command in isp_intr to MBOX_NOTIFY_COMPLETE macro.
2000-07-04 01:02:38 +00:00
MBOX_NOTIFY_COMPLETE(isp);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
} else {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"Mbox Command Async (0x%x) with no waiters",
mbox);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
}
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
} else if (isp_parse_async(isp, mbox) < 0) {
return;
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
}
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
if ((IS_FC(isp) && mbox != ASYNC_RIO_RESP) ||
isp->isp_state != ISP_RUNSTATE) {
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
ISP_WRITE(isp, BIU_SEMA, 0);
return;
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
/*
* We can't be getting this now.
*/
if (isp->isp_state != ISP_RUNSTATE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
"interrupt (ISR=%x SEMA=%x) when not ready", isr, sema);
/*
* Thank you very much! *Burrrp*!
*/
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp,
READ_RESPONSE_QUEUE_IN_POINTER(isp));
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
ISP_WRITE(isp, BIU_SEMA, 0);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
return;
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
}
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
/*
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
* Get the current Response Queue Out Pointer.
*
* If we're a 2300, we can ask what hardware what it thinks.
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
*/
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_23XX(isp)) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
optr = ISP_READ(isp, isp->isp_respoutrp);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
/*
* Debug: to be taken out eventually
*/
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
if (isp->isp_residx != optr) {
isp_prt(isp, ISP_LOGWARN, "optr %x soft optr %x",
optr, isp->isp_residx);
}
} else {
optr = isp->isp_residx;
}
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
/*
* You *must* read the Response Queue In Pointer
* prior to clearing the RISC interrupt.
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
*
* Debounce the 2300 if revision less than 2.
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
*/
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_2100(isp) || (IS_2300(isp) && isp->isp_revision < 2)) {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
i = 0;
do {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
iptr = READ_RESPONSE_QUEUE_IN_POINTER(isp);
junk = READ_RESPONSE_QUEUE_IN_POINTER(isp);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
} while (junk != iptr && ++i < 1000);
if (iptr != junk) {
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
"Response Queue Out Pointer Unstable (%x, %x)",
iptr, junk);
return;
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
}
} else {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
iptr = READ_RESPONSE_QUEUE_IN_POINTER(isp);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
}
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
isp->isp_resodx = iptr;
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
Once we have firmware running (if isp_reset) and this is the first time through, establish what our LUN width is. Unfortunately, we can't ask the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs (SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We have to assume only one LUN for Fibre Channel because the LUN setting in Request Queue entries is in different places whether we have SCCLUN firmware or not, so the only LUN guaranteed to work for both is LUN 0. Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used- instead use run time determinants based upon isp->isp_maxluns. After starting firmware, delay 500us to give it a chance to get rolling. Fix the interrupt service routine to check for both isr && sema being zero before thinking this was a spurious interrupt. Following the manuals, allow for both Mailbox as well as Queue Reponse type interrupts for regular SCSI.
2000-06-18 04:56:17 +00:00
if (optr == iptr && sema == 0) {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
/*
* There are a lot of these- reasons unknown- mostly on
* faster Alpha machines.
*
* I tried delaying after writing HCCR_CMD_CLEAR_RISC_INT to
* make sure the old interrupt went away (to avoid 'ringing'
* effects), but that didn't stop this from occurring.
*/
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (IS_23XX(isp)) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
USEC_DELAY(100);
iptr = READ_RESPONSE_QUEUE_IN_POINTER(isp);
junk = ISP_READ(isp, BIU_R2HSTSLO);
} else {
junk = ISP_READ(isp, BIU_ISR);
}
if (optr == iptr) {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (IS_23XX(isp)) {
;
} else {
sema = ISP_READ(isp, BIU_SEMA);
mbox = ISP_READ(isp, OUTMAILBOX0);
if ((sema & 0x3) && (mbox & 0x8000)) {
goto again;
}
}
isp->isp_intbogus++;
isp_prt(isp, ISP_LOGDEBUG1,
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
"bogus intr- isr %x (%x) iptr %x optr %x",
isr, junk, iptr, optr);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
isp->isp_resodx = iptr;
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
ISP_WRITE(isp, BIU_SEMA, 0);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
if (isp->isp_rspbsy) {
return;
}
isp->isp_rspbsy = 1;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
while (optr != iptr) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
ispstatusreq_t local, *sp = &local;
isphdr_t *hp;
int type;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t oop;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
int buddaboom = 0;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
hp = (isphdr_t *) ISP_QUEUE_ENTRY(isp->isp_result, optr);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
oop = optr;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
optr = ISP_NXT_QENTRY(optr, RESULT_QUEUE_LEN(isp));
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
nlooked++;
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
read_again:
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
/*
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
* Synchronize our view of this response queue entry.
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMORYBARRIER(isp, SYNC_RESULT, oop, QENTRY_LEN);
type = isp_get_response_type(isp, hp);
if (type == RQSTYPE_RESPONSE) {
isp_get_response(isp, (ispstatusreq_t *) hp, sp);
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
} else if (type == RQSTYPE_RIO2) {
isp_rio2_t rio;
isp_get_rio2(isp, (isp_rio2_t *) hp, &rio);
for (i = 0; i < rio.req_header.rqs_seqno; i++) {
isp_fastpost_complete(isp, rio.req_handles[i]);
}
if (isp->isp_fpcchiwater < rio.req_header.rqs_seqno)
isp->isp_fpcchiwater = rio.req_header.rqs_seqno;
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
MEMZERO(hp, QENTRY_LEN); /* PERF */
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
continue;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Disable RIO (reduced interrupt operation) for 2200 boards- it seemed like it worked- but I ran into a case with a 2204 where commands were being lost right and left. Best be safe. For target mode, or things called if we call isp_handle_other response- note that we might have dropped locks by changing the output pointer so we bail from the loop. It's the responsibility of the entity dropping the lock to make sure that we let the f/w know we've read thus far into the response queue (else we begin processing the same entries again- blech!). MFC after: 1 day
2002-03-07 17:32:45 +00:00
/*
* Somebody reachable via isp_handle_other_response
* may have updated the response queue pointers for
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
* us, so we reload our goal index.
Disable RIO (reduced interrupt operation) for 2200 boards- it seemed like it worked- but I ran into a case with a 2204 where commands were being lost right and left. Best be safe. For target mode, or things called if we call isp_handle_other response- note that we might have dropped locks by changing the output pointer so we bail from the loop. It's the responsibility of the entity dropping the lock to make sure that we let the f/w know we've read thus far into the response queue (else we begin processing the same entries again- blech!). MFC after: 1 day
2002-03-07 17:32:45 +00:00
*/
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
int i = isp_handle_other_response(isp, type, hp, &optr);
if (i < 0) {
goto read_again;
} else if (i > 0) {
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
iptr = isp->isp_resodx;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO(hp, QENTRY_LEN); /* PERF */
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
continue;
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
/*
* After this point, we'll just look at the header as
* we don't know how to deal with the rest of the
* response.
*/
isp_get_response(isp, (ispstatusreq_t *) hp, sp);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* It really has to be a bounced request just copied
temp fix for internal queue overflow problem
1999-04-14 17:37:36 +00:00
* from the request queue to the response queue. If
* not, something bad has happened.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (sp->req_header.rqs_entry_type != RQSTYPE_REQUEST) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, notresp,
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
sp->req_header.rqs_entry_type, oop, optr,
nlooked);
if (isp->isp_dblev & ISP_LOGDEBUG0) {
isp_print_bytes(isp, "Queue Entry",
QENTRY_LEN, sp);
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO(hp, QENTRY_LEN); /* PERF */
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
continue;
}
buddaboom = 1;
}
if (sp->req_header.rqs_flags & 0xf) {
temp fix for internal queue overflow problem
1999-04-14 17:37:36 +00:00
#define _RQS_OFLAGS \
~(RQSFLAG_CONTINUATION|RQSFLAG_FULL|RQSFLAG_BADHEADER|RQSFLAG_BADPACKET)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (sp->req_header.rqs_flags & RQSFLAG_CONTINUATION) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"continuation segment");
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
continue;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
if (sp->req_header.rqs_flags & RQSFLAG_FULL) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGDEBUG1,
"internal queues full");
temp fix for internal queue overflow problem
1999-04-14 17:37:36 +00:00
/*
* We'll synthesize a QUEUE FULL message below.
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
if (sp->req_header.rqs_flags & RQSFLAG_BADHEADER) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "bad header flag");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
buddaboom++;
}
if (sp->req_header.rqs_flags & RQSFLAG_BADPACKET) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "bad request packet");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
buddaboom++;
}
temp fix for internal queue overflow problem
1999-04-14 17:37:36 +00:00
if (sp->req_header.rqs_flags & _RQS_OFLAGS) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"unknown flags (0x%x) in response",
sp->req_header.rqs_flags);
temp fix for internal queue overflow problem
1999-04-14 17:37:36 +00:00
buddaboom++;
}
#undef _RQS_OFLAGS
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
if (sp->req_handle > isp->isp_maxcmds || sp->req_handle < 1) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO(hp, QENTRY_LEN); /* PERF */
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
"bad request handle %d (type 0x%x, flags 0x%x)",
sp->req_handle, sp->req_header.rqs_entry_type,
sp->req_header.rqs_flags);
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
continue;
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
xs = isp_find_xs(isp, sp->req_handle);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (xs == NULL) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t ts = sp->req_completion_status & 0xff;
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO(hp, QENTRY_LEN); /* PERF */
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
/*
* Only whine if this isn't the expected fallout of
* aborting the command.
*/
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (sp->req_header.rqs_entry_type != RQSTYPE_RESPONSE) {
isp_prt(isp, ISP_LOGERR,
"cannot find handle 0x%x (type 0x%x)",
sp->req_handle,
sp->req_header.rqs_entry_type);
} else if (ts != RQCS_ABORTED) {
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
isp_prt(isp, ISP_LOGERR,
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
"cannot find handle 0x%x (status 0x%x)",
sp->req_handle, ts);
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
}
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
continue;
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
isp_destroy_handle(isp, sp->req_handle);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (sp->req_status_flags & RQSTF_BUS_RESET) {
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
XS_SETERR(xs, HBA_BUSRESET);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_sendmarker |= (1 << XS_CHANNEL(xs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
if (buddaboom) {
XS_SETERR(xs, HBA_BOTCH);
}
various fixes
2000-08-27 23:38:44 +00:00
if (IS_FC(isp) && (sp->req_scsi_status & RQCS_SV)) {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
/*
various fixes
2000-08-27 23:38:44 +00:00
* Fibre Channel F/W doesn't say we got status
* if there's Sense Data instead. I guess they
* think it goes w/o saying.
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
*/
various fixes
2000-08-27 23:38:44 +00:00
sp->req_state_flags |= RQSF_GOT_STATUS;
}
if (sp->req_state_flags & RQSF_GOT_STATUS) {
*XS_STSP(xs) = sp->req_scsi_status & 0xff;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
switch (sp->req_header.rqs_entry_type) {
case RQSTYPE_RESPONSE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_SET_STATE_STAT(isp, xs, sp);
isp_parse_status(isp, sp, xs);
if ((XS_NOERR(xs) || XS_ERR(xs) == HBA_NOERROR) &&
(*XS_STSP(xs) == SCSI_BUSY)) {
XS_SETERR(xs, HBA_TGTBSY);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
if (IS_SCSI(isp)) {
XS_RESID(xs) = sp->req_resid;
if ((sp->req_state_flags & RQSF_GOT_STATUS) &&
(*XS_STSP(xs) == SCSI_CHECK) &&
(sp->req_state_flags & RQSF_GOT_SENSE)) {
XS_SAVE_SENSE(xs, sp);
}
/*
* A new synchronous rate was negotiated for
* this target. Mark state such that we'll go
* look up that which has changed later.
*/
if (sp->req_status_flags & RQSTF_NEGOTIATION) {
int t = XS_TGT(xs);
sdparam *sdp = isp->isp_param;
sdp += XS_CHANNEL(xs);
sdp->isp_devparam[t].dev_refresh = 1;
isp->isp_update |=
(1 << XS_CHANNEL(xs));
}
} else {
if (sp->req_status_flags & RQSF_XFER_COMPLETE) {
XS_RESID(xs) = 0;
} else if (sp->req_scsi_status & RQCS_RESID) {
XS_RESID(xs) = sp->req_resid;
} else {
XS_RESID(xs) = 0;
}
if ((sp->req_state_flags & RQSF_GOT_STATUS) &&
(*XS_STSP(xs) == SCSI_CHECK) &&
(sp->req_scsi_status & RQCS_SV)) {
XS_SAVE_SENSE(xs, sp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
/* solely for the benefit of debug */
sp->req_state_flags |= RQSF_GOT_SENSE;
various fixes
2000-08-27 23:38:44 +00:00
}
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
isp_prt(isp, ISP_LOGDEBUG2,
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
"asked for %ld got raw resid %ld settled for %ld",
(long) XS_XFRLEN(xs), (long) sp->req_resid,
(long) XS_RESID(xs));
various fixes
2000-08-27 23:38:44 +00:00
break;
case RQSTYPE_REQUEST:
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (sp->req_header.rqs_flags & RQSFLAG_FULL) {
/*
* Force Queue Full status.
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
*XS_STSP(xs) = SCSI_QFULL;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
XS_SETERR(xs, HBA_NOERROR);
} else if (XS_NOERR(xs)) {
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
/*
* ????
*/
isp_prt(isp, ISP_LOGDEBUG0,
"Request Queue Entry bounced back");
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
XS_SETERR(xs, HBA_BOTCH);
}
various fixes
2000-08-27 23:38:44 +00:00
XS_RESID(xs) = XS_XFRLEN(xs);
break;
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
"unhandled response queue type 0x%x",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
sp->req_header.rqs_entry_type);
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
if (XS_NOERR(xs)) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
XS_SETERR(xs, HBA_BOTCH);
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
/*
On reset, make sure that we have some parameters set correctly. This fixes a longstanding issue WRT resetting the chip after startup- it would fail if we were connected as an F-port to a switch. If we were connected as an F-port, we got assigned a hard loop ID of 255, which is really a bogus loop id. Then when we turned around to reset ourselves, the firmware would reject the ICB_INIT request because the loop id was bogus. *sputter* Minor fixlet from somebody in NetBSD with too much time on their hands (dma -> DMA).
2003-09-13 01:55:44 +00:00
* Free any DMA resources. As a side effect, this may
various fixes
2000-08-27 23:38:44 +00:00
* also do any cache flushing necessary for data coherence. */
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (XS_XFRLEN(xs)) {
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
ISP_DMAFREE(isp, xs, sp->req_handle);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (((isp->isp_dblev & (ISP_LOGDEBUG2|ISP_LOGDEBUG3))) ||
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
((isp->isp_dblev & ISP_LOGDEBUG1) && ((!XS_NOERR(xs)) ||
(*XS_STSP(xs) != SCSI_GOOD)))) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
char skey;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
if (sp->req_state_flags & RQSF_GOT_SENSE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
skey = XS_SNSKEY(xs) & 0xf;
if (skey < 10)
skey += '0';
else
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
skey += 'a' - 10;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
} else if (*XS_STSP(xs) == SCSI_CHECK) {
skey = '?';
} else {
skey = '.';
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGALL, finmsg, XS_CHANNEL(xs),
XS_TGT(xs), XS_LUN(xs), XS_XFRLEN(xs), XS_RESID(xs),
*XS_STSP(xs), skey, XS_ERR(xs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
if (isp->isp_nactive > 0)
isp->isp_nactive--;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
complist[ndone++] = xs; /* defer completion call until later */
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
MEMZERO(hp, QENTRY_LEN); /* PERF */
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (ndone == MAX_REQUESTQ_COMPLETIONS) {
break;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
* If we looked at any commands, then it's valid to find out
* what the outpointer is. It also is a trigger to update the
* ISP's notion of what we've seen so far.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (nlooked) {
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
/*
Disable RIO (reduced interrupt operation) for 2200 boards- it seemed like it worked- but I ran into a case with a 2204 where commands were being lost right and left. Best be safe. For target mode, or things called if we call isp_handle_other response- note that we might have dropped locks by changing the output pointer so we bail from the loop. It's the responsibility of the entity dropping the lock to make sure that we let the f/w know we've read thus far into the response queue (else we begin processing the same entries again- blech!). MFC after: 1 day
2002-03-07 17:32:45 +00:00
* While we're at it, read the requst queue out pointer.
Add 2 Gigabit Fibre Channel support (2300 && 2312 cards). This required some reworking (and consequent cleanup) of the interrupt service code. Also begin to start a cleanup of target mode support that will (eventually) not require more inforamtion routed with the ATIO to come back with the CTIO other than tag. MFC after: 4 weeks
2001-08-31 21:39:04 +00:00
*/
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
isp->isp_reqodx = READ_REQUEST_QUEUE_OUT_POINTER(isp);
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (isp->isp_rscchiwater < ndone)
isp->isp_rscchiwater = ndone;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp->isp_residx = optr;
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
isp->isp_rspbsy = 0;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
for (i = 0; i < ndone; i++) {
xs = complist[i];
if (xs) {
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
isp->isp_rsltccmplt++;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_done(xs);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
/*
* Support routines.
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_async(ispsoftc_t *isp, uint16_t mbox)
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
{
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
int rval = 0;
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
int bus;
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
if (IS_DUALBUS(isp)) {
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
bus = ISP_READ(isp, OUTMAILBOX6);
} else {
bus = 0;
}
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
isp_prt(isp, ISP_LOGDEBUG2, "Async Mbox 0x%x", mbox);
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
switch (mbox) {
case ASYNC_BUS_RESET:
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker |= (1 << bus);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#ifdef ISP_TARGET_MODE
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (isp_target_async(isp, bus, mbox))
rval = -1;
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#endif
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
isp_async(isp, ISPASYNC_BUS_RESET, &bus);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_SYSTEM_ERROR:
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp->isp_state = ISP_CRASHED;
if (IS_FC(isp)) {
FCPARAM(isp)->isp_loopstate = LOOP_NIL;
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
}
/*
* Were we waiting for a mailbox command to complete?
* If so, it's dead, so wake up the waiter.
*/
if (isp->isp_mboxbsy) {
isp->isp_obits = 1;
isp->isp_mboxtmp[0] = MBOX_HOST_INTERFACE_ERROR;
MBOX_NOTIFY_COMPLETE(isp);
}
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
#ifdef ISP_FW_CRASH_DUMP
/*
* If we have crash dumps enabled, it's up to the handler
* for isp_async to reinit stuff and restart the firmware
* after performing the crash dump. The reason we do things
* this way is that we may need to activate a kernel thread
* to do all the crash dump goop.
*/
isp_async(isp, ISPASYNC_FW_CRASH, NULL);
#else
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
isp_async(isp, ISPASYNC_FW_CRASH, NULL);
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
isp_reinit(isp);
isp_async(isp, ISPASYNC_FW_RESTARTED, NULL);
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
#endif
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
rval = -1;
break;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case ASYNC_RQS_XFER_ERR:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Request Queue Transfer Error");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_RSP_XFER_ERR:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Response Queue Transfer Error");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_QWAKEUP:
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
/*
* We've just been notified that the Queue has woken up.
* We don't need to be chatty about this- just unlatch things
* and move on.
*/
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
mbox = READ_REQUEST_QUEUE_OUT_POINTER(isp);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_TIMEOUT_RESET:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
Remove a superfluous newline in a string (isp_prt adds this). Fix a missed conversion of 32 to 16 bit handles.
2001-03-04 18:41:23 +00:00
"timeout initiated SCSI bus reset of bus %d", bus);
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker |= (1 << bus);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#ifdef ISP_TARGET_MODE
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (isp_target_async(isp, bus, mbox))
rval = -1;
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
case ASYNC_DEVICE_RESET:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO, "device reset on bus %d", bus);
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker |= (1 << bus);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#ifdef ISP_TARGET_MODE
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (isp_target_async(isp, bus, mbox))
rval = -1;
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_EXTMSG_UNDERRUN:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "extended message underrun");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_SCAM_INT:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO, "SCAM interrupt");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_HUNG_SCSI:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"stalled SCSI Bus after DATA Overrun");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/* XXX: Need to issue SCSI reset at this point */
break;
case ASYNC_KILLED_BUS:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "SCSI Bus reset after DATA Overrun");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_BUS_TRANSIT:
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
mbox = ISP_READ(isp, OUTMAILBOX2);
switch (mbox & 0x1c00) {
case SXP_PINS_LVD_MODE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO, "Transition to LVD mode");
various fixes
2000-08-27 23:38:44 +00:00
SDPARAM(isp)->isp_diffmode = 0;
SDPARAM(isp)->isp_ultramode = 0;
SDPARAM(isp)->isp_lvdmode = 1;
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
case SXP_PINS_HVD_MODE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO,
"Transition to Differential mode");
various fixes
2000-08-27 23:38:44 +00:00
SDPARAM(isp)->isp_diffmode = 1;
SDPARAM(isp)->isp_ultramode = 0;
SDPARAM(isp)->isp_lvdmode = 0;
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
case SXP_PINS_SE_MODE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO,
"Transition to Single Ended mode");
various fixes
2000-08-27 23:38:44 +00:00
SDPARAM(isp)->isp_diffmode = 0;
SDPARAM(isp)->isp_ultramode = 1;
SDPARAM(isp)->isp_lvdmode = 0;
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"Transition to Unknown Mode 0x%x", mbox);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
break;
}
/*
* XXX: Set up to renegotiate again!
*/
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/* Can only be for a 1080... */
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker |= (1 << bus);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
/*
* We can use bus, which will always be zero for FC cards,
* as a mailbox pattern accumulator to be checked below.
*/
case ASYNC_RIO5:
bus = 0x1ce; /* outgoing mailbox regs 1-3, 6-7 */
break;
case ASYNC_RIO4:
bus = 0x14e; /* outgoing mailbox regs 1-3, 6 */
break;
case ASYNC_RIO3:
bus = 0x10e; /* outgoing mailbox regs 1-3 */
break;
case ASYNC_RIO2:
bus = 0x106; /* outgoing mailbox regs 1-2 */
break;
case ASYNC_RIO1:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case ASYNC_CMD_CMPLT:
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
bus = 0x102; /* outgoing mailbox regs 1 */
break;
case ASYNC_RIO_RESP:
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
return (rval);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case ASYNC_CTIO_DONE:
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
{
various fixes
2000-08-27 23:38:44 +00:00
#ifdef ISP_TARGET_MODE
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
int handle =
(ISP_READ(isp, OUTMAILBOX2) << 16) |
(ISP_READ(isp, OUTMAILBOX1));
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
if (isp_target_async(isp, handle, mbox)) {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
rval = -1;
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
} else {
/* count it as a fast posting intr */
isp->isp_fphccmplt++;
}
various fixes
2000-08-27 23:38:44 +00:00
#else
isp_prt(isp, ISP_LOGINFO, "Fast Posting CTIO done");
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_fphccmplt++; /* count it as a fast posting intr */
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
#endif
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
break;
}
Enable LIP F8, LIP Reset async events. Be more chatty about SNS failures. Fix typo for skipped phase mesage. Correct MBOX_GET_PORT_QUEUE_PARAMS options in table. MFC after: 2 weeks
2001-08-16 17:25:08 +00:00
case ASYNC_LIP_F8:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case ASYNC_LIP_OCCURRED:
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
isp->isp_sendmarker = 1;
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp_mark_getpdb_all(isp);
Spring MegaChange #1. ---- Make a device for each ISP- really usable only with devfs and add an ioctl entry point (this can be used to (re)set debug levels, reset the HBA, rescan the fabric, issue lips, etc). ---- Add in a kernel thread for Fibre Channel cards. The purpose of this thread is to be woken up to clean up after Fibre Channel events block things. Basically, any FC event that casts doubt on the location or identify of FC devices blocks the queues. When, and if, we get the PORT DATABASE CHANGED or NAME SERVER DATABASE CHANGED async event, we activate the kthread which will then, in full thread context, re-evaluate the local loop and/or the fabric. When it's satisfied that things are stable, it can then release the blocked queues and let commands flow again. The prior mechanism was a lazy evaluation. That is, the next command to come down the pipe after change events would pay the full price for re-evaluation. And if this was done off of a softcall, it really could hang up the system. These changes brings the FreeBSD port more in line with the Solaris, Linux and NetBSD ports. It also, more importantly, gets us being more proactive about topology changes which could then be reflected upwards to CAM so that the periph driver can be informed sooner rather than later when things arrive or depart. --- Add in the (correct) usage of locking macros- we now have lock transition macros which allow us to transition from holding the CAM lock (Giant) and grabbing the softc lock and vice versa. Switch over to having this HBA do real locking. Some folks claim this won't be a win. They're right. But you have to start somewhere, and this will begin to teach us how to DTRT for HBAs, etc. -- Start putting in prototype 2300 support. Add back in LIP and Loop Reset as async events that each platform will handle. Add in another int_bogus instrumentation point. Do some more substantial target mode cleanups. MFC after: 8 weeks
2001-05-28 21:20:43 +00:00
isp_async(isp, ISPASYNC_LIP, NULL);
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#ifdef ISP_TARGET_MODE
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (isp_target_async(isp, bus, mbox))
rval = -1;
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#endif
We've had problems with data corruption occuring on commands that complete (with no apparent error) after we receive a LIP. This has been observed mostly on Local Loop topologies. To be safe, let's just mark all active commands as dead if we get a LIP and we're on a private or public loop. MFC after: 4 weeks
2001-06-14 17:13:24 +00:00
/*
* We've had problems with data corruption occuring on
* commands that complete (with no apparent error) after
* we receive a LIP. This has been observed mostly on
* Local Loop topologies. To be safe, let's just mark
* all active commands as dead.
*/
if (FCPARAM(isp)->isp_topo == TOPO_NL_PORT ||
FCPARAM(isp)->isp_topo == TOPO_FL_PORT) {
int i, j;
for (i = j = 0; i < isp->isp_maxcmds; i++) {
XS_T *xs;
xs = isp->isp_xflist[i];
if (xs != NULL) {
j++;
XS_SETERR(xs, HBA_BUSRESET);
}
}
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
if (j) {
isp_prt(isp, ISP_LOGERR,
"LIP destroyed %d active commands", j);
}
We've had problems with data corruption occuring on commands that complete (with no apparent error) after we receive a LIP. This has been observed mostly on Local Loop topologies. To be safe, let's just mark all active commands as dead if we get a LIP and we're on a private or public loop. MFC after: 4 weeks
2001-06-14 17:13:24 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_LOOP_UP:
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
isp->isp_sendmarker = 1;
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp_mark_getpdb_all(isp);
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
isp_async(isp, ISPASYNC_LOOP_UP, NULL);
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#ifdef ISP_TARGET_MODE
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (isp_target_async(isp, bus, mbox)) {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
rval = -1;
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_LOOP_DOWN:
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp->isp_sendmarker = 1;
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
FCPARAM(isp)->isp_loopstate = LOOP_NIL;
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp_mark_getpdb_all(isp);
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
isp_async(isp, ISPASYNC_LOOP_DOWN, NULL);
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#ifdef ISP_TARGET_MODE
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (isp_target_async(isp, bus, mbox)) {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
rval = -1;
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_LOOP_RESET:
Pick up topology more sanely at f/w startup. Change the restrictions of where we can have targets (based on topology). Much more importantly, make sure all mods to isp_sendmarker or |= so we don't lose the marking of a bus that needs to have a marker sent for it.
2000-04-21 02:04:34 +00:00
isp->isp_sendmarker = 1;
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
FCPARAM(isp)->isp_loopstate = LOOP_NIL;
Make firmware revision a triple. Clean up some FC init stuff for board versions with no BIOS. Separate mailbox interrupts from IOCB interrupts. Read OUTMAILBOX5 while RISC_INT is active- not after you clear it (potential race condition). Clear out older broken BIG_ENDIAN goop. Don't negotiate narrow/async for LVD busses at startup if already in LVD mode. Note usage of presumptive 1040C revision. For all the LIP, PDB Changed, Loop UP/DOWN async events, mark fw state as unknown as well as marking the need to do a getpdb on targets- after a LIP for certain the f/w has to do PRLI/PLOGI for all targets again and marking f/w state as unknown gives us a fighting chance to (start to) hold up for that to complete.
1999-04-04 02:28:29 +00:00
isp_mark_getpdb_all(isp);
Spring MegaChange #1. ---- Make a device for each ISP- really usable only with devfs and add an ioctl entry point (this can be used to (re)set debug levels, reset the HBA, rescan the fabric, issue lips, etc). ---- Add in a kernel thread for Fibre Channel cards. The purpose of this thread is to be woken up to clean up after Fibre Channel events block things. Basically, any FC event that casts doubt on the location or identify of FC devices blocks the queues. When, and if, we get the PORT DATABASE CHANGED or NAME SERVER DATABASE CHANGED async event, we activate the kthread which will then, in full thread context, re-evaluate the local loop and/or the fabric. When it's satisfied that things are stable, it can then release the blocked queues and let commands flow again. The prior mechanism was a lazy evaluation. That is, the next command to come down the pipe after change events would pay the full price for re-evaluation. And if this was done off of a softcall, it really could hang up the system. These changes brings the FreeBSD port more in line with the Solaris, Linux and NetBSD ports. It also, more importantly, gets us being more proactive about topology changes which could then be reflected upwards to CAM so that the periph driver can be informed sooner rather than later when things arrive or depart. --- Add in the (correct) usage of locking macros- we now have lock transition macros which allow us to transition from holding the CAM lock (Giant) and grabbing the softc lock and vice versa. Switch over to having this HBA do real locking. Some folks claim this won't be a win. They're right. But you have to start somewhere, and this will begin to teach us how to DTRT for HBAs, etc. -- Start putting in prototype 2300 support. Add back in LIP and Loop Reset as async events that each platform will handle. Add in another int_bogus instrumentation point. Do some more substantial target mode cleanups. MFC after: 8 weeks
2001-05-28 21:20:43 +00:00
isp_async(isp, ISPASYNC_LOOP_RESET, NULL);
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#ifdef ISP_TARGET_MODE
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (isp_target_async(isp, bus, mbox)) {
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
rval = -1;
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_PDB_CHANGED:
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
isp->isp_sendmarker = 1;
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_loopstate = LOOP_PDB_RCVD;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
isp_mark_getpdb_all(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_PDB);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case ASYNC_CHANGE_NOTIFY:
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
/*
* Not correct, but it will force us to rescan the loop.
*/
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_loopstate = LOOP_PDB_RCVD;
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
isp_mark_getpdb_all(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_SNS);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
case ASYNC_PTPMODE:
various fixes
2000-08-27 23:38:44 +00:00
if (FCPARAM(isp)->isp_onfabric)
FCPARAM(isp)->isp_topo = TOPO_F_PORT;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
else
FCPARAM(isp)->isp_topo = TOPO_N_PORT;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
isp_mark_getpdb_all(isp);
isp->isp_sendmarker = 1;
various fixes
2000-08-27 23:38:44 +00:00
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_OTHER);
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
#ifdef ISP_TARGET_MODE
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if (isp_target_async(isp, bus, mbox))
rval = -1;
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
#endif
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGINFO, "Point-to-Point mode");
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
break;
case ASYNC_CONNMODE:
mbox = ISP_READ(isp, OUTMAILBOX1);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_mark_getpdb_all(isp);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
switch (mbox) {
case ISP_CONN_LOOP:
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_prt(isp, ISP_LOGINFO,
"Point-to-Point -> Loop mode");
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
break;
case ISP_CONN_PTP:
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_prt(isp, ISP_LOGINFO,
"Loop -> Point-to-Point mode");
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
break;
case ISP_CONN_BADLIP:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
"Point-to-Point -> Loop mode (BAD LIP)");
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
break;
case ISP_CONN_FATAL:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "FATAL CONNECTION ERROR");
Pick up some compilation warning fixes from NetBSD. If we don't have ISP_FW_CRASH_DUMP defined, we have to do a isp_reinit in the core code- not the platform code- so fix the ISP_CONN_FATAL case.
2003-02-16 01:32:52 +00:00
#ifdef ISP_FW_CRASH_DUMP
isp_async(isp, ISPASYNC_FW_CRASH, NULL);
#else
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
isp_async(isp, ISPASYNC_FW_CRASH, NULL);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_reinit(isp);
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
isp_async(isp, ISPASYNC_FW_RESTARTED, NULL);
Pick up some compilation warning fixes from NetBSD. If we don't have ISP_FW_CRASH_DUMP defined, we have to do a isp_reinit in the core code- not the platform code- so fix the ISP_CONN_FATAL case.
2003-02-16 01:32:52 +00:00
#endif
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
return (-1);
case ISP_CONN_LOOPBACK:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"Looped Back in Point-to-Point mode");
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
break;
default:
isp_prt(isp, ISP_LOGWARN,
"Unknown connection mode (0x%x)", mbox);
break;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_OTHER);
isp->isp_sendmarker = 1;
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
break;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "Unknown Async Code 0x%x", mbox);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
}
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if (bus & 0x100) {
int i, nh;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t handles[16];
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (nh = 0, i = 1; i < MAX_MAILBOX(isp); i++) {
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
if ((bus & (1 << i)) == 0) {
continue;
}
handles[nh++] = ISP_READ(isp, MBOX_OFF(i));
}
for (i = 0; i < nh; i++) {
isp_fastpost_complete(isp, handles[i]);
isp_prt(isp, ISP_LOGDEBUG3,
"fast post completion of %u", handles[i]);
}
if (isp->isp_fpcchiwater < nh)
isp->isp_fpcchiwater = nh;
} else {
isp->isp_intoasync++;
}
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
return (rval);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Clean up stupidity in the isp_handle_other_response function- indexes of queue entries have to be at least 16 bits now! If we're running a 2100 less than rev 5, turn off loop fairness (per Qlogic errata). Fix typo in checking against 2200 F/W revision. Slightly fix/reorder fabric login stuff. Change to usage of isp_getrqentry for code clarity. Add some defensive dual bus assumptions. Various cleanups, etc...
1999-11-21 03:18:22 +00:00
/*
* Handle other response entries. A pointer to the request queue output
* index is here in case we want to eat several entries at once, although
* this is not used currently.
*/
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_handle_other_response(ispsoftc_t *isp, int type,
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isphdr_t *hp, uint16_t *optrp)
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
{
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
switch (type) {
various fixes
2000-08-27 23:38:44 +00:00
case RQSTYPE_STATUS_CONT:
isp_prt(isp, ISP_LOGINFO, "Ignored Continuation Response");
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
return (1);
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
case RQSTYPE_ATIO:
Raise default FCP logintime to 60 seconds. Move the position of where we could have seen the loop up at least once so it makes sense. Change some stuff in ispscsicmd so we don't get stuck there if the loop has never come up yet. Add in some target mode support code.
2000-01-03 23:52:41 +00:00
case RQSTYPE_CTIO:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQSTYPE_ENABLE_LUN:
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
case RQSTYPE_MODIFY_LUN:
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
case RQSTYPE_NOTIFY:
case RQSTYPE_NOTIFY_ACK:
case RQSTYPE_CTIO1:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQSTYPE_ATIO2:
case RQSTYPE_CTIO2:
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
case RQSTYPE_CTIO3:
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_rsltccmplt++; /* count as a response completion */
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
#ifdef ISP_TARGET_MODE
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
if (isp_target_notify(isp, (ispstatusreq_t *) hp, optrp)) {
return (1);
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
#endif
Revert previous commit. Violates Maintainer (O'Brien knows how to reach me directly), but more importantly, breaks compiles on non-FreeBSD platforms.
2003-08-25 17:58:23 +00:00
/* FALLTHROUGH */
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
case RQSTYPE_REQUEST:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
default:
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
USEC_DELAY(100);
if (type != isp_get_response_type(isp, hp)) {
/*
* This is questionable- we're just papering over
* something we've seen on SMP linux in target
* mode- we don't really know what's happening
* here that causes us to think we've gotten
* an entry, but that either the entry isn't
* filled out yet or our CPU read data is stale.
*/
isp_prt(isp, ISP_LOGINFO,
"unstable type in response queue");
return (-1);
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "Unhandled Response Type 0x%x",
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp_get_response_type(isp, hp));
Don't set ZIO for 23XX for target mode (use fast posting instead). Use the correct number of handles for multihandle returns. Very, very, rarely on some SMP systems we've seen an 'unstable' type in the response queue. I dunno whether or not it's a bug in our handling, or whether there's a cache incoherency issue, but try to guard against it. MFC after: 2 weeks
2005-01-23 06:26:45 +00:00
if (isp_async(isp, ISPASYNC_UNHANDLED_RESPONSE, hp)) {
return (1);
}
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
return (0);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
}
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_status(ispsoftc_t *isp, ispstatusreq_t *sp, XS_T *xs)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
switch (sp->req_completion_status & 0xff) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case RQCS_COMPLETE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_NOERROR);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
case RQCS_INCOMPLETE:
if ((sp->req_state_flags & RQSF_GOT_TARGET) == 0) {
various fixes
2000-08-27 23:38:44 +00:00
isp_prt(isp, ISP_LOGDEBUG1,
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"Selection Timeout for %d.%d.%d",
Make the Not RESPONSE in RESPONSE QUEUE message have a bit more info (specifically, how many entries we've looked at so far). Maintain interrupt instrumentation. Use USEC_SLEEP instead of USEC_DELAY in a number of places (this allows us to drop locks and sleep instead of spin). Track changes to configuration options for topology preference. Fix botched order of printout for Channel, Target, Lun.
2000-12-02 18:08:35 +00:00
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_SELTIMEOUT);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"command incomplete for %d.%d.%d, state 0x%x",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs),
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
sp->req_state_flags);
break;
case RQCS_DMA_ERROR:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "DMA error for command on %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case RQCS_TRANSPORT_ERROR:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
{
char buf[172];
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "states=>");
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (sp->req_state_flags & RQSF_GOT_BUS) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s GOT_BUS", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_state_flags & RQSF_GOT_TARGET) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s GOT_TGT", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_state_flags & RQSF_SENT_CDB) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s SENT_CDB", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_state_flags & RQSF_XFRD_DATA) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s XFRD_DATA", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_state_flags & RQSF_GOT_STATUS) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s GOT_STS", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_state_flags & RQSF_GOT_SENSE) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s GOT_SNS", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_state_flags & RQSF_XFER_COMPLETE) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s XFR_CMPLT", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s\nstatus=>", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (sp->req_status_flags & RQSTF_DISCONNECT) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Disconnect", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_SYNCHRONOUS) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Sync_xfr", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_PARITY_ERROR) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Parity", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_BUS_RESET) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Bus_Reset", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_DEVICE_RESET) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Device_Reset", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_ABORTED) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Aborted", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_TIMEOUT) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Timeout", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
if (sp->req_status_flags & RQSTF_NEGOTIATION) {
The size argument to snprintf does not have to be backed off by one to account for a NULL byte. Submitted by: Jacques A. Vidrine <nectar@celabo.org>
2002-09-07 16:12:52 +00:00
SNPRINTF(buf, sizeof (buf), "%s Negotiation", buf);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
isp_prt(isp, ISP_LOGERR, "%s", buf);
isp_prt(isp, ISP_LOGERR, "transport error for %d.%d.%d:\n%s",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs), buf);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_RESET_OCCURRED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"bus reset destroyed command for %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
isp->isp_sendmarker |= (1 << XS_CHANNEL(xs));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_BUSRESET);
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return;
case RQCS_ABORTED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "command aborted for %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
isp->isp_sendmarker |= (1 << XS_CHANNEL(xs));
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_ABORTED);
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return;
case RQCS_TIMEOUT:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "command timed out for %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
/*
* Check to see if we logged out the device.
*/
if (IS_FC(isp)) {
if ((sp->req_completion_status & RQSTF_LOGOUT) &&
FCPARAM(isp)->portdb[XS_TGT(xs)].valid &&
FCPARAM(isp)->portdb[XS_TGT(xs)].fabric_dev) {
FCPARAM(isp)->portdb[XS_TGT(xs)].relogin = 1;
}
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_CMDTIMEOUT);
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case RQCS_DATA_OVERRUN:
various fixes
2000-08-27 23:38:44 +00:00
XS_RESID(xs) = sp->req_resid;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "data overrun for command on %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_DATAOVR);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_COMMAND_OVERRUN:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"command overrun for command on %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_STATUS_OVERRUN:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"status overrun for command on %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_BAD_MESSAGE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"msg not COMMAND COMPLETE after status %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_NO_MESSAGE_OUT:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"No MESSAGE OUT phase after selection on %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_EXT_ID_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "EXTENDED IDENTIFY failed %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_IDE_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"INITIATOR DETECTED ERROR rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_ABORT_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "ABORT OPERATION rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_REJECT_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "MESSAGE REJECT rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_NOP_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "NOP rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_PARITY_ERROR_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"MESSAGE PARITY ERROR rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_DEVICE_RESET_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"BUS DEVICE RESET rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_ID_MSG_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "IDENTIFY rejected by %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_UNEXP_BUS_FREE:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "%d.%d.%d had an unexpected bus free",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case RQCS_DATA_UNDERRUN:
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
{
if (IS_FC(isp)) {
int ru_marked = (sp->req_scsi_status & RQCS_RU) != 0;
if (!ru_marked || sp->req_resid > XS_XFRLEN(xs)) {
isp_prt(isp, ISP_LOGWARN, bun, XS_TGT(xs),
XS_LUN(xs), XS_XFRLEN(xs), sp->req_resid,
(ru_marked)? "marked" : "not marked");
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_BOTCH);
}
return;
}
}
various fixes
2000-08-27 23:38:44 +00:00
XS_RESID(xs) = sp->req_resid;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_NOERROR);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_XACT_ERR1:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, xact1, XS_CHANNEL(xs),
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_XACT_ERR2:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, xact2,
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
XS_LUN(xs), XS_TGT(xs), XS_CHANNEL(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_XACT_ERR3:
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
isp_prt(isp, ISP_LOGERR, xact3,
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_BAD_ENTRY:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Invalid IOCB entry type detected");
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_QUEUE_FULL:
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
"internal queues full for %d.%d.%d status 0x%x",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs), *XS_STSP(xs));
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
/*
* If QFULL or some other status byte is set, then this
* isn't an error, per se.
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
*
* Unfortunately, some QLogic f/w writers have, in
* some cases, ommitted to *set* status to QFULL.
*
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (*XS_STSP(xs) != SCSI_GOOD && XS_NOERR(xs)) {
Add some prototype deadchip detection. Set FIFO bursting (1XX0 only- it's already on for the 2XX0) and detect the broken 1040A FIFO. Change bzero to MEMZERO (portability with **nux). Use memcpy for same reason. Finally detect QUEUE FULL conditions and return this as an error that will get cam_periph_error to do it's 'tagged openings now XXX' dance.
1999-01-10 02:55:10 +00:00
XS_SETERR(xs, HBA_NOERROR);
return;
}
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
*
*
*/
*XS_STSP(xs) = SCSI_QFULL;
XS_SETERR(xs, HBA_NOERROR);
return;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_PHASE_SKIPPED:
oops- typo in a previous commit
2001-08-16 17:39:45 +00:00
isp_prt(isp, ISP_LOGERR, pskip, XS_CHANNEL(xs),
Enable LIP F8, LIP Reset async events. Be more chatty about SNS failures. Fix typo for skipped phase mesage. Correct MBOX_GET_PORT_QUEUE_PARAMS options in table. MFC after: 2 weeks
2001-08-16 17:25:08 +00:00
XS_TGT(xs), XS_LUN(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
case RQCS_ARQS_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"Auto Request Sense failed for %d.%d.%d",
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_ARQFAIL);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_WIDE_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"Wide Negotiation failed for %d.%d.%d",
XS_TGT(xs), XS_LUN(xs), XS_CHANNEL(xs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (IS_SCSI(isp)) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
sdparam *sdp = isp->isp_param;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
sdp += XS_CHANNEL(xs);
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[XS_TGT(xs)].goal_flags &= ~DPARM_WIDE;
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
sdp->isp_devparam[XS_TGT(xs)].dev_update = 1;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
isp->isp_update |= (1 << XS_CHANNEL(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_NOERROR);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
case RQCS_SYNCXFER_FAILED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"SDTR Message failed for target %d.%d.%d",
XS_TGT(xs), XS_LUN(xs), XS_CHANNEL(xs));
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (IS_SCSI(isp)) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
sdparam *sdp = isp->isp_param;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
sdp += XS_CHANNEL(xs);
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[XS_TGT(xs)].goal_flags &= ~DPARM_SYNC;
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
sdp->isp_devparam[XS_TGT(xs)].dev_update = 1;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
isp->isp_update |= (1 << XS_CHANNEL(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
break;
case RQCS_LVD_BUSERR:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR,
"Bad LVD condition while talking to %d.%d.%d",
XS_TGT(xs), XS_LUN(xs), XS_CHANNEL(xs));
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case RQCS_PORT_UNAVAILABLE:
/*
* No such port on the loop. Moral equivalent of SELTIMEO
*/
case RQCS_PORT_LOGGED_OUT:
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* It was there (maybe)- treat as a selection timeout.
*/
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
if ((sp->req_completion_status & 0xff) == RQCS_PORT_UNAVAILABLE)
isp_prt(isp, ISP_LOGINFO,
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
"port unavailable for target %d", XS_TGT(xs));
Limit fabric search to a default 256 entries. This will all go away soon because it's just getting harder and harder to find switches that correctly implement the GET ALL NEXT subcommands for the SNS protocol. Latch up result out pointer and set a busy flag when we're looking at the response queue. This allows for a cleaner way to make sure we don't get multiple CPUs trying to read the same response queue entries. Change how isp_handle_other_response returns values (clarity). Make PORT UNAVAILABLE the same as PORT LOGOUT (force a LIP). Do some formatting changes. MFC after: 0 days
2002-03-21 21:10:16 +00:00
else
isp_prt(isp, ISP_LOGINFO,
"port logout for target %d", XS_TGT(xs));
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
/*
* If we're on a local loop, force a LIP (which is overkill)
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
* to force a re-login of this unit. If we're on fabric,
* then we'll have to relogin as a matter of course.
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
*/
if (FCPARAM(isp)->isp_topo == TOPO_NL_PORT ||
FCPARAM(isp)->isp_topo == TOPO_FL_PORT) {
mbreg_t mbs;
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
mbs.param[0] = MBOX_INIT_LIP;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (IS_2KLOGIN(isp)) {
mbs.obits |= (1 << 10);
}
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
isp_mboxcmd_qnw(isp, &mbs, 1);
}
/*
* Probably overkill.
*/
isp->isp_sendmarker = 1;
FCPARAM(isp)->isp_loopstate = LOOP_PDB_RCVD;
isp_mark_getpdb_all(isp);
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_OTHER);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_SELTIMEOUT);
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case RQCS_PORT_CHANGED:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"port changed for target %d", XS_TGT(xs));
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_SELTIMEOUT);
}
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
return;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case RQCS_PORT_BUSY:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"port busy for target %d", XS_TGT(xs));
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_TGTBSY);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Unknown Completion Status 0x%x",
sp->req_completion_status);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_BOTCH);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_fastpost_complete(ispsoftc_t *isp, uint16_t fph)
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
{
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_T *xs;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
Switch to using 16 bit handles instead of 32 bit handles. This is a pretty invasive change, but there are three good reasons to do this: 1. We'll never have > 16 bits of handle. 2. We can (eventually) enable the RIO (Reduced Interrupt Operation) bits which return multiple completing 16 bit handles in mailbox registers. 3. The !)$*)$*~)@$*~)$* Qlogic target mode for parallel SCSI spec changed such that at_reserved (which was 32 bits) was split into two pieces- and one of which was a 16 bit handle id that functions like the at_rxid for Fibre Channel (a tag for the f/w to correlate CTIOs with a particular command). Since we had to muck with that and this changed the whole handler architecture, we might as well... Propagate new at_handle on through int ct_fwhandle. Follow implications of changing to 16 bit handles. These above changes at least get Qlogic 1040 cards working in target mode again. 1080/12160 cards don't work yet. In isp.c: Prepare for doing all loop management in outer layers.
2001-03-02 06:28:55 +00:00
if (fph == 0) {
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
return;
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
}
xs = isp_find_xs(isp, fph);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (xs == NULL) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN,
"Command for fast post handle 0x%x not found", fph);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
return;
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
isp_destroy_handle(isp, fph);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
* Since we don't have a result queue entry item,
* we must believe that SCSI status is zero and
* that all data transferred.
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_SET_STATE_STAT(isp, xs, NULL);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
XS_RESID(xs) = 0;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
*XS_STSP(xs) = SCSI_GOOD;
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (XS_XFRLEN(xs)) {
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
ISP_DMAFREE(isp, xs, fph);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
if (isp->isp_nactive)
isp->isp_nactive--;
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
isp->isp_fphccmplt++;
various fixes
2000-08-27 23:38:44 +00:00
isp_done(xs);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_mbox_continue(ispsoftc_t *isp)
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
{
mbreg_t mbs;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *ptr;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
switch (isp->isp_lastmbxcmd) {
case MBOX_WRITE_RAM_WORD:
case MBOX_READ_RAM_WORD:
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
case MBOX_WRITE_RAM_WORD_EXTENDED:
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
case MBOX_READ_RAM_WORD_EXTENDED:
break;
default:
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
return (1);
}
if (isp->isp_mboxtmp[0] != MBOX_COMMAND_COMPLETE) {
isp->isp_mbxwrk0 = 0;
return (-1);
}
/*
* Clear the previous interrupt.
*/
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
ISP_WRITE(isp, BIU_SEMA, 0);
/*
* Continue with next word.
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
ptr = isp->isp_mbxworkp;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
switch (isp->isp_lastmbxcmd) {
case MBOX_WRITE_RAM_WORD:
mbs.param[2] = *ptr++;
mbs.param[1] = isp->isp_mbxwrk1++;
break;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
case MBOX_WRITE_RAM_WORD_EXTENDED:
mbs.param[2] = *ptr++;
mbs.param[1] = isp->isp_mbxwrk1++;
if (isp->isp_mbxwrk1 == 0) {
isp->isp_mbxwrk8++;
}
mbs.param[8] = isp->isp_mbxwrk8;
break;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
case MBOX_READ_RAM_WORD:
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
*ptr++ = isp->isp_mboxtmp[2];
mbs.param[1] = isp->isp_mbxwrk1++;
break;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
case MBOX_READ_RAM_WORD_EXTENDED:
*ptr++ = isp->isp_mboxtmp[2];
mbs.param[1] = isp->isp_mbxwrk1++;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (isp->isp_mbxwrk1 == 0) {
isp->isp_mbxwrk8++;
}
mbs.param[8] = isp->isp_mbxwrk8;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
break;
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_mbxworkp = ptr;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
mbs.param[0] = isp->isp_lastmbxcmd;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
isp->isp_mbxwrk0 -= 1;
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
isp_mboxcmd_qnw(isp, &mbs, 0);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
return (0);
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
#define HIWRD(x) ((x) >> 16)
#define LOWRD(x) ((x) & 0xffff)
#define ISPOPMAP(a, b) (((a) << 16) | (b))
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
static const uint32_t mbpscsi[] = {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x01, 0x01), /* 0x00: MBOX_NO_OP */
ISPOPMAP(0x1f, 0x01), /* 0x01: MBOX_LOAD_RAM */
ISPOPMAP(0x03, 0x01), /* 0x02: MBOX_EXEC_FIRMWARE */
ISPOPMAP(0x1f, 0x01), /* 0x03: MBOX_DUMP_RAM */
ISPOPMAP(0x07, 0x07), /* 0x04: MBOX_WRITE_RAM_WORD */
ISPOPMAP(0x03, 0x07), /* 0x05: MBOX_READ_RAM_WORD */
ISPOPMAP(0x3f, 0x3f), /* 0x06: MBOX_MAILBOX_REG_TEST */
ISPOPMAP(0x03, 0x07), /* 0x07: MBOX_VERIFY_CHECKSUM */
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
ISPOPMAP(0x01, 0x0f), /* 0x08: MBOX_ABOUT_FIRMWARE */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x00, 0x00), /* 0x09: */
ISPOPMAP(0x00, 0x00), /* 0x0a: */
ISPOPMAP(0x00, 0x00), /* 0x0b: */
ISPOPMAP(0x00, 0x00), /* 0x0c: */
ISPOPMAP(0x00, 0x00), /* 0x0d: */
ISPOPMAP(0x01, 0x05), /* 0x0e: MBOX_CHECK_FIRMWARE */
ISPOPMAP(0x00, 0x00), /* 0x0f: */
ISPOPMAP(0x1f, 0x1f), /* 0x10: MBOX_INIT_REQ_QUEUE */
ISPOPMAP(0x3f, 0x3f), /* 0x11: MBOX_INIT_RES_QUEUE */
ISPOPMAP(0x0f, 0x0f), /* 0x12: MBOX_EXECUTE_IOCB */
ISPOPMAP(0x03, 0x03), /* 0x13: MBOX_WAKE_UP */
ISPOPMAP(0x01, 0x3f), /* 0x14: MBOX_STOP_FIRMWARE */
ISPOPMAP(0x0f, 0x0f), /* 0x15: MBOX_ABORT */
ISPOPMAP(0x03, 0x03), /* 0x16: MBOX_ABORT_DEVICE */
ISPOPMAP(0x07, 0x07), /* 0x17: MBOX_ABORT_TARGET */
ISPOPMAP(0x07, 0x07), /* 0x18: MBOX_BUS_RESET */
ISPOPMAP(0x03, 0x07), /* 0x19: MBOX_STOP_QUEUE */
ISPOPMAP(0x03, 0x07), /* 0x1a: MBOX_START_QUEUE */
ISPOPMAP(0x03, 0x07), /* 0x1b: MBOX_SINGLE_STEP_QUEUE */
ISPOPMAP(0x03, 0x07), /* 0x1c: MBOX_ABORT_QUEUE */
ISPOPMAP(0x03, 0x4f), /* 0x1d: MBOX_GET_DEV_QUEUE_STATUS */
ISPOPMAP(0x00, 0x00), /* 0x1e: */
ISPOPMAP(0x01, 0x07), /* 0x1f: MBOX_GET_FIRMWARE_STATUS */
ISPOPMAP(0x01, 0x07), /* 0x20: MBOX_GET_INIT_SCSI_ID */
ISPOPMAP(0x01, 0x07), /* 0x21: MBOX_GET_SELECT_TIMEOUT */
ISPOPMAP(0x01, 0xc7), /* 0x22: MBOX_GET_RETRY_COUNT */
ISPOPMAP(0x01, 0x07), /* 0x23: MBOX_GET_TAG_AGE_LIMIT */
ISPOPMAP(0x01, 0x03), /* 0x24: MBOX_GET_CLOCK_RATE */
ISPOPMAP(0x01, 0x07), /* 0x25: MBOX_GET_ACT_NEG_STATE */
ISPOPMAP(0x01, 0x07), /* 0x26: MBOX_GET_ASYNC_DATA_SETUP_TIME */
ISPOPMAP(0x01, 0x07), /* 0x27: MBOX_GET_PCI_PARAMS */
ISPOPMAP(0x03, 0x4f), /* 0x28: MBOX_GET_TARGET_PARAMS */
ISPOPMAP(0x03, 0x0f), /* 0x29: MBOX_GET_DEV_QUEUE_PARAMS */
ISPOPMAP(0x01, 0x07), /* 0x2a: MBOX_GET_RESET_DELAY_PARAMS */
ISPOPMAP(0x00, 0x00), /* 0x2b: */
ISPOPMAP(0x00, 0x00), /* 0x2c: */
ISPOPMAP(0x00, 0x00), /* 0x2d: */
ISPOPMAP(0x00, 0x00), /* 0x2e: */
ISPOPMAP(0x00, 0x00), /* 0x2f: */
ISPOPMAP(0x03, 0x03), /* 0x30: MBOX_SET_INIT_SCSI_ID */
ISPOPMAP(0x07, 0x07), /* 0x31: MBOX_SET_SELECT_TIMEOUT */
ISPOPMAP(0xc7, 0xc7), /* 0x32: MBOX_SET_RETRY_COUNT */
ISPOPMAP(0x07, 0x07), /* 0x33: MBOX_SET_TAG_AGE_LIMIT */
ISPOPMAP(0x03, 0x03), /* 0x34: MBOX_SET_CLOCK_RATE */
ISPOPMAP(0x07, 0x07), /* 0x35: MBOX_SET_ACT_NEG_STATE */
ISPOPMAP(0x07, 0x07), /* 0x36: MBOX_SET_ASYNC_DATA_SETUP_TIME */
ISPOPMAP(0x07, 0x07), /* 0x37: MBOX_SET_PCI_CONTROL_PARAMS */
ISPOPMAP(0x4f, 0x4f), /* 0x38: MBOX_SET_TARGET_PARAMS */
ISPOPMAP(0x0f, 0x0f), /* 0x39: MBOX_SET_DEV_QUEUE_PARAMS */
ISPOPMAP(0x07, 0x07), /* 0x3a: MBOX_SET_RESET_DELAY_PARAMS */
ISPOPMAP(0x00, 0x00), /* 0x3b: */
ISPOPMAP(0x00, 0x00), /* 0x3c: */
ISPOPMAP(0x00, 0x00), /* 0x3d: */
ISPOPMAP(0x00, 0x00), /* 0x3e: */
ISPOPMAP(0x00, 0x00), /* 0x3f: */
ISPOPMAP(0x01, 0x03), /* 0x40: MBOX_RETURN_BIOS_BLOCK_ADDR */
ISPOPMAP(0x3f, 0x01), /* 0x41: MBOX_WRITE_FOUR_RAM_WORDS */
ISPOPMAP(0x03, 0x07), /* 0x42: MBOX_EXEC_BIOS_IOCB */
ISPOPMAP(0x00, 0x00), /* 0x43: */
ISPOPMAP(0x00, 0x00), /* 0x44: */
ISPOPMAP(0x03, 0x03), /* 0x45: SET SYSTEM PARAMETER */
ISPOPMAP(0x01, 0x03), /* 0x46: GET SYSTEM PARAMETER */
ISPOPMAP(0x00, 0x00), /* 0x47: */
ISPOPMAP(0x01, 0xcf), /* 0x48: GET SCAM CONFIGURATION */
ISPOPMAP(0xcf, 0xcf), /* 0x49: SET SCAM CONFIGURATION */
ISPOPMAP(0x03, 0x03), /* 0x4a: MBOX_SET_FIRMWARE_FEATURES */
ISPOPMAP(0x01, 0x03), /* 0x4b: MBOX_GET_FIRMWARE_FEATURES */
ISPOPMAP(0x00, 0x00), /* 0x4c: */
ISPOPMAP(0x00, 0x00), /* 0x4d: */
ISPOPMAP(0x00, 0x00), /* 0x4e: */
ISPOPMAP(0x00, 0x00), /* 0x4f: */
ISPOPMAP(0xdf, 0xdf), /* 0x50: LOAD RAM A64 */
ISPOPMAP(0xdf, 0xdf), /* 0x51: DUMP RAM A64 */
If we have a 1240 or an ULTRA2 or better card, use MBOX_INIT_RES_QUEUE_A64 (preparation for DAC/A64 support)
2002-09-23 04:59:42 +00:00
ISPOPMAP(0xdf, 0xff), /* 0x52: INITIALIZE REQUEST QUEUE A64 */
ISPOPMAP(0xef, 0xff), /* 0x53: INITIALIZE RESPONSE QUEUE A64 */
Tra-La, another QLogic f/w funny- this time with the 2300. If we get a completion status of RQCS_QUEUE_FULL, it means that the internal queues are full. Other QLogic boards set the QFULL SCSI status. But *nooooooooooo*, not the 2300. MFC after: 1 day
2001-10-23 23:05:20 +00:00
ISPOPMAP(0xcf, 0x01), /* 0x54: EXECUTE IOCB A64 */
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
ISPOPMAP(0x07, 0x01), /* 0x55: ENABLE TARGET MODE */
ISPOPMAP(0x03, 0x0f), /* 0x56: GET TARGET STATUS */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x00, 0x00), /* 0x57: */
ISPOPMAP(0x00, 0x00), /* 0x58: */
ISPOPMAP(0x00, 0x00), /* 0x59: */
ISPOPMAP(0x03, 0x03), /* 0x5a: SET DATA OVERRUN RECOVERY MODE */
ISPOPMAP(0x01, 0x03), /* 0x5b: GET DATA OVERRUN RECOVERY MODE */
ISPOPMAP(0x0f, 0x0f), /* 0x5c: SET HOST DATA */
ISPOPMAP(0x01, 0x01) /* 0x5d: GET NOST DATA */
};
In order to save ourselves grief with the SUNPRO compiler under Solaris (which, for reasons unknown to me, chokes on u_int16_t as a typedef of unsigned short if used in a transitional (mixed K&R and ANSI) way), we'll go the extra mile and fully ANSIfy things.
2001-03-14 04:11:56 +00:00
#ifndef ISP_STRIPPED
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
static char *scsi_mbcmd_names[] = {
"NO-OP",
"LOAD RAM",
"EXEC FIRMWARE",
"DUMP RAM",
"WRITE RAM WORD",
"READ RAM WORD",
"MAILBOX REG TEST",
"VERIFY CHECKSUM",
"ABOUT FIRMWARE",
NULL,
NULL,
NULL,
NULL,
NULL,
"CHECK FIRMWARE",
NULL,
"INIT REQUEST QUEUE",
"INIT RESULT QUEUE",
"EXECUTE IOCB",
"WAKE UP",
"STOP FIRMWARE",
"ABORT",
"ABORT DEVICE",
"ABORT TARGET",
"BUS RESET",
"STOP QUEUE",
"START QUEUE",
"SINGLE STEP QUEUE",
"ABORT QUEUE",
"GET DEV QUEUE STATUS",
NULL,
"GET FIRMWARE STATUS",
"GET INIT SCSI ID",
"GET SELECT TIMEOUT",
"GET RETRY COUNT",
"GET TAG AGE LIMIT",
"GET CLOCK RATE",
"GET ACT NEG STATE",
"GET ASYNC DATA SETUP TIME",
"GET PCI PARAMS",
"GET TARGET PARAMS",
"GET DEV QUEUE PARAMS",
"GET RESET DELAY PARAMS",
NULL,
NULL,
NULL,
NULL,
NULL,
"SET INIT SCSI ID",
"SET SELECT TIMEOUT",
"SET RETRY COUNT",
"SET TAG AGE LIMIT",
"SET CLOCK RATE",
"SET ACT NEG STATE",
"SET ASYNC DATA SETUP TIME",
"SET PCI CONTROL PARAMS",
"SET TARGET PARAMS",
"SET DEV QUEUE PARAMS",
"SET RESET DELAY PARAMS",
NULL,
NULL,
NULL,
NULL,
NULL,
"RETURN BIOS BLOCK ADDR",
"WRITE FOUR RAM WORDS",
"EXEC BIOS IOCB",
NULL,
NULL,
"SET SYSTEM PARAMETER",
"GET SYSTEM PARAMETER",
NULL,
"GET SCAM CONFIGURATION",
"SET SCAM CONFIGURATION",
"SET FIRMWARE FEATURES",
"GET FIRMWARE FEATURES",
NULL,
NULL,
NULL,
NULL,
"LOAD RAM A64",
"DUMP RAM A64",
"INITIALIZE REQUEST QUEUE A64",
"INITIALIZE RESPONSE QUEUE A64",
"EXECUTE IOCB A64",
"ENABLE TARGET MODE",
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
"GET TARGET MODE STATE",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
NULL,
NULL,
NULL,
"SET DATA OVERRUN RECOVERY MODE",
"GET DATA OVERRUN RECOVERY MODE",
"SET HOST DATA",
"GET NOST DATA",
};
#endif
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
static const uint32_t mbpfc[] = {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x01, 0x01), /* 0x00: MBOX_NO_OP */
ISPOPMAP(0x1f, 0x01), /* 0x01: MBOX_LOAD_RAM */
ISPOPMAP(0x03, 0x01), /* 0x02: MBOX_EXEC_FIRMWARE */
Fix a longstanding bug- we had the sense of what bit 14 for the ICB firmware options meant- *I* had taken it to mean that if you set it, Node Name would be ignored and derived from Port Name. Actually, it meant the opposite. As a consequence- change ICBOPT_USE_PORTNAME to the define ICBOPT_BOTH_WWNS- makes more sense. Fix wrong input bitmap for MBOX_DUMP_RAM command. Call ISP_DUMPREGS if we get a f/w crash. Add ISPCTL_RUN_MBOXCMD control command (so outer layers can run a mailbox command directly) and add a ISPASYNC_UNHANDLED_RESPONSE hook so outer layers can understand response queue entries we might not know about.
2001-02-23 05:35:50 +00:00
ISPOPMAP(0xdf, 0x01), /* 0x03: MBOX_DUMP_RAM */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x07, 0x07), /* 0x04: MBOX_WRITE_RAM_WORD */
ISPOPMAP(0x03, 0x07), /* 0x05: MBOX_READ_RAM_WORD */
ISPOPMAP(0xff, 0xff), /* 0x06: MBOX_MAILBOX_REG_TEST */
ISPOPMAP(0x03, 0x05), /* 0x07: MBOX_VERIFY_CHECKSUM */
Clarify issues about whether we have SCCLUN (65535 luns) or non-SCCLUN (16 luns) firmware for the Fibre Channel cards. We used to assume that if we didn't download firmware, we couldn't know what the firmware capability with respect to SCCLUNs is- and it's important because the lun field changes in the request queue entry based upon which firmware it is. At any rate, we *do* get back firmware attributes in mailbox register 6 when we do ABOUT FIRMWARE for all 2200/2300 cards- and for 2100 cards with at least 1.17.0 firmware. So- we now assume non-SCCLUN behaviour for 2100 cards with firmware < 1.17.0- and we check the firmware attributes for other cards (loaded firmware or not). This also allows us to get rid of the crappy test of isp_maxluns > 16- we simply can check firmware attributes for SCCLUN behaviour. This required an 'oops' fix to the outgoing mailbox count field for ABOUT FIRMWARE for FC cards. Also- while here, hardwire firmware revisions for loaded code for SBus cards. Apparently the 1.35 or 1.37 f/w we've been loading into isp1000 just doesn't report firmware revisions out to mailbox regs 1, 2 and 3 like everyone else. Grumble. Not that this fix hardly matters for FreeBSD. MFC after: 4 weeks
2001-09-03 03:09:48 +00:00
ISPOPMAP(0x01, 0x4f), /* 0x08: MBOX_ABOUT_FIRMWARE */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0xdf, 0x01), /* 0x09: LOAD RAM */
ISPOPMAP(0xdf, 0x01), /* 0x0a: DUMP RAM */
ISPOPMAP(0x00, 0x00), /* 0x0b: */
ISPOPMAP(0x00, 0x00), /* 0x0c: */
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
ISPOPMAP(0x13, 0x01), /* 0x0d: MBOX_WRITE_RAM_WORD_EXTENDED) */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x01, 0x05), /* 0x0e: MBOX_CHECK_FIRMWARE */
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
ISPOPMAP(0x13, 0x05), /* 0x0f: MBOX_READ_RAM_WORD_EXTENDED */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x1f, 0x11), /* 0x10: MBOX_INIT_REQ_QUEUE */
ISPOPMAP(0x2f, 0x21), /* 0x11: MBOX_INIT_RES_QUEUE */
ISPOPMAP(0x0f, 0x01), /* 0x12: MBOX_EXECUTE_IOCB */
ISPOPMAP(0x03, 0x03), /* 0x13: MBOX_WAKE_UP */
ISPOPMAP(0x01, 0xff), /* 0x14: MBOX_STOP_FIRMWARE */
ISPOPMAP(0x4f, 0x01), /* 0x15: MBOX_ABORT */
ISPOPMAP(0x07, 0x01), /* 0x16: MBOX_ABORT_DEVICE */
ISPOPMAP(0x07, 0x01), /* 0x17: MBOX_ABORT_TARGET */
ISPOPMAP(0x03, 0x03), /* 0x18: MBOX_BUS_RESET */
ISPOPMAP(0x07, 0x05), /* 0x19: MBOX_STOP_QUEUE */
ISPOPMAP(0x07, 0x05), /* 0x1a: MBOX_START_QUEUE */
ISPOPMAP(0x07, 0x05), /* 0x1b: MBOX_SINGLE_STEP_QUEUE */
ISPOPMAP(0x07, 0x05), /* 0x1c: MBOX_ABORT_QUEUE */
ISPOPMAP(0x07, 0x03), /* 0x1d: MBOX_GET_DEV_QUEUE_STATUS */
ISPOPMAP(0x00, 0x00), /* 0x1e: */
ISPOPMAP(0x01, 0x07), /* 0x1f: MBOX_GET_FIRMWARE_STATUS */
ISPOPMAP(0x01, 0x4f), /* 0x20: MBOX_GET_LOOP_ID */
ISPOPMAP(0x00, 0x00), /* 0x21: */
ISPOPMAP(0x01, 0x07), /* 0x22: MBOX_GET_RETRY_COUNT */
ISPOPMAP(0x00, 0x00), /* 0x23: */
ISPOPMAP(0x00, 0x00), /* 0x24: */
ISPOPMAP(0x00, 0x00), /* 0x25: */
ISPOPMAP(0x00, 0x00), /* 0x26: */
ISPOPMAP(0x00, 0x00), /* 0x27: */
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
ISPOPMAP(0x01, 0x03), /* 0x28: MBOX_GET_FIRMWARE_OPTIONS */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x03, 0x07), /* 0x29: MBOX_GET_PORT_QUEUE_PARAMS */
ISPOPMAP(0x00, 0x00), /* 0x2a: */
ISPOPMAP(0x00, 0x00), /* 0x2b: */
ISPOPMAP(0x00, 0x00), /* 0x2c: */
ISPOPMAP(0x00, 0x00), /* 0x2d: */
ISPOPMAP(0x00, 0x00), /* 0x2e: */
ISPOPMAP(0x00, 0x00), /* 0x2f: */
ISPOPMAP(0x00, 0x00), /* 0x30: */
ISPOPMAP(0x00, 0x00), /* 0x31: */
ISPOPMAP(0x07, 0x07), /* 0x32: MBOX_SET_RETRY_COUNT */
ISPOPMAP(0x00, 0x00), /* 0x33: */
ISPOPMAP(0x00, 0x00), /* 0x34: */
ISPOPMAP(0x00, 0x00), /* 0x35: */
ISPOPMAP(0x00, 0x00), /* 0x36: */
ISPOPMAP(0x00, 0x00), /* 0x37: */
ISPOPMAP(0x0f, 0x01), /* 0x38: MBOX_SET_FIRMWARE_OPTIONS */
ISPOPMAP(0x0f, 0x07), /* 0x39: MBOX_SET_PORT_QUEUE_PARAMS */
ISPOPMAP(0x00, 0x00), /* 0x3a: */
ISPOPMAP(0x00, 0x00), /* 0x3b: */
ISPOPMAP(0x00, 0x00), /* 0x3c: */
ISPOPMAP(0x00, 0x00), /* 0x3d: */
ISPOPMAP(0x00, 0x00), /* 0x3e: */
ISPOPMAP(0x00, 0x00), /* 0x3f: */
ISPOPMAP(0x03, 0x01), /* 0x40: MBOX_LOOP_PORT_BYPASS */
ISPOPMAP(0x03, 0x01), /* 0x41: MBOX_LOOP_PORT_ENABLE */
ISPOPMAP(0x03, 0x07), /* 0x42: MBOX_GET_RESOURCE_COUNTS */
ISPOPMAP(0x01, 0x01), /* 0x43: MBOX_REQUEST_NON_PARTICIPATING_MODE */
ISPOPMAP(0x00, 0x00), /* 0x44: */
ISPOPMAP(0x00, 0x00), /* 0x45: */
ISPOPMAP(0x00, 0x00), /* 0x46: */
ISPOPMAP(0xcf, 0x03), /* 0x47: GET PORT_DATABASE ENHANCED */
ISPOPMAP(0x00, 0x00), /* 0x48: */
ISPOPMAP(0x00, 0x00), /* 0x49: */
ISPOPMAP(0x00, 0x00), /* 0x4a: */
ISPOPMAP(0x00, 0x00), /* 0x4b: */
ISPOPMAP(0x00, 0x00), /* 0x4c: */
ISPOPMAP(0x00, 0x00), /* 0x4d: */
ISPOPMAP(0x00, 0x00), /* 0x4e: */
ISPOPMAP(0x00, 0x00), /* 0x4f: */
ISPOPMAP(0x00, 0x00), /* 0x50: */
ISPOPMAP(0x00, 0x00), /* 0x51: */
ISPOPMAP(0x00, 0x00), /* 0x52: */
ISPOPMAP(0x00, 0x00), /* 0x53: */
ISPOPMAP(0xcf, 0x01), /* 0x54: EXECUTE IOCB A64 */
ISPOPMAP(0x00, 0x00), /* 0x55: */
ISPOPMAP(0x00, 0x00), /* 0x56: */
ISPOPMAP(0x00, 0x00), /* 0x57: */
ISPOPMAP(0x00, 0x00), /* 0x58: */
ISPOPMAP(0x00, 0x00), /* 0x59: */
ISPOPMAP(0x00, 0x00), /* 0x5a: */
If we get a DATA UNDERRUN error from QLogic FC cards, but the RQCS_RU bit is not set in the scsi completion status, or if the residual is clearly nonsense, then this was a command that suffered the loss of one or more FC frames in the middle of the exchange. Set HBA_BOTCH and hope it will get retried. It's the only thing we can do. MFC after: 1 day
2002-05-01 21:58:36 +00:00
ISPOPMAP(0x03, 0x01), /* 0x5b: MBOX_DRIVER_HEARTBEAT */
ISPOPMAP(0xcf, 0x01), /* 0x5c: MBOX_FW_HEARTBEAT */
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
ISPOPMAP(0x07, 0x03), /* 0x5d: MBOX_GET_SET_DATA_RATE */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x00, 0x00), /* 0x5e: */
ISPOPMAP(0x00, 0x00), /* 0x5f: */
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
ISPOPMAP(0xcd, 0x31), /* 0x60: MBOX_INIT_FIRMWARE */
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISPOPMAP(0x00, 0x00), /* 0x61: */
ISPOPMAP(0x01, 0x01), /* 0x62: MBOX_INIT_LIP */
ISPOPMAP(0xcd, 0x03), /* 0x63: MBOX_GET_FC_AL_POSITION_MAP */
ISPOPMAP(0xcf, 0x01), /* 0x64: MBOX_GET_PORT_DB */
ISPOPMAP(0x07, 0x01), /* 0x65: MBOX_CLEAR_ACA */
ISPOPMAP(0x07, 0x01), /* 0x66: MBOX_TARGET_RESET */
ISPOPMAP(0x07, 0x01), /* 0x67: MBOX_CLEAR_TASK_SET */
ISPOPMAP(0x07, 0x01), /* 0x68: MBOX_ABORT_TASK_SET */
ISPOPMAP(0x01, 0x07), /* 0x69: MBOX_GET_FW_STATE */
ISPOPMAP(0x03, 0xcf), /* 0x6a: MBOX_GET_PORT_NAME */
ISPOPMAP(0xcf, 0x01), /* 0x6b: MBOX_GET_LINK_STATUS */
ISPOPMAP(0x0f, 0x01), /* 0x6c: MBOX_INIT_LIP_RESET */
ISPOPMAP(0x00, 0x00), /* 0x6d: */
ISPOPMAP(0xcf, 0x03), /* 0x6e: MBOX_SEND_SNS */
ISPOPMAP(0x0f, 0x07), /* 0x6f: MBOX_FABRIC_LOGIN */
ISPOPMAP(0x03, 0x01), /* 0x70: MBOX_SEND_CHANGE_REQUEST */
ISPOPMAP(0x03, 0x03), /* 0x71: MBOX_FABRIC_LOGOUT */
ISPOPMAP(0x0f, 0x0f), /* 0x72: MBOX_INIT_LIP_LOGIN */
ISPOPMAP(0x00, 0x00), /* 0x73: */
ISPOPMAP(0x07, 0x01), /* 0x74: LOGIN LOOP PORT */
ISPOPMAP(0xcf, 0x03), /* 0x75: GET PORT/NODE NAME LIST */
ISPOPMAP(0x4f, 0x01), /* 0x76: SET VENDOR ID */
ISPOPMAP(0xcd, 0x01), /* 0x77: INITIALIZE IP MAILBOX */
ISPOPMAP(0x00, 0x00), /* 0x78: */
ISPOPMAP(0x00, 0x00), /* 0x79: */
ISPOPMAP(0x00, 0x00), /* 0x7a: */
ISPOPMAP(0x00, 0x00), /* 0x7b: */
ISPOPMAP(0x4f, 0x03), /* 0x7c: Get ID List */
ISPOPMAP(0xcf, 0x01), /* 0x7d: SEND LFA */
Add an ioctl framework for doing FC task management functions from a user space tool- useful for doing FC target mode certification.
2005-10-29 02:46:59 +00:00
ISPOPMAP(0x0f, 0x01) /* 0x7e: LUN RESET */
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
};
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
/*
* Footnotes
*
* (1): this sets bits 21..16 in mailbox register #8, which we nominally
* do not access at this time in the core driver. The caller is
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
* responsible for setting this register first (Gross!). The assumption
* is that we won't overflow.
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
#ifndef ISP_STRIPPED
static char *fc_mbcmd_names[] = {
"NO-OP",
"LOAD RAM",
"EXEC FIRMWARE",
"DUMP RAM",
"WRITE RAM WORD",
"READ RAM WORD",
"MAILBOX REG TEST",
"VERIFY CHECKSUM",
"ABOUT FIRMWARE",
"LOAD RAM",
"DUMP RAM",
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
"WRITE RAM WORD EXTENDED",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
NULL,
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
"READ RAM WORD EXTENDED",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"CHECK FIRMWARE",
NULL,
"INIT REQUEST QUEUE",
"INIT RESULT QUEUE",
"EXECUTE IOCB",
"WAKE UP",
"STOP FIRMWARE",
"ABORT",
"ABORT DEVICE",
"ABORT TARGET",
"BUS RESET",
"STOP QUEUE",
"START QUEUE",
"SINGLE STEP QUEUE",
"ABORT QUEUE",
"GET DEV QUEUE STATUS",
NULL,
"GET FIRMWARE STATUS",
"GET LOOP ID",
NULL,
"GET RETRY COUNT",
NULL,
NULL,
NULL,
NULL,
NULL,
"GET FIRMWARE OPTIONS",
"GET PORT QUEUE PARAMS",
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
"SET RETRY COUNT",
NULL,
NULL,
NULL,
NULL,
NULL,
"SET FIRMWARE OPTIONS",
"SET PORT QUEUE PARAMS",
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
"LOOP PORT BYPASS",
"LOOP PORT ENABLE",
"GET RESOURCE COUNTS",
"REQUEST NON PARTICIPATING MODE",
NULL,
NULL,
NULL,
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
"GET PORT DATABASE ENHANCED",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
"EXECUTE IOCB A64",
NULL,
NULL,
NULL,
NULL,
NULL,
NULL,
Pick up some compilation warning fixes from NetBSD. If we don't have ISP_FW_CRASH_DUMP defined, we have to do a isp_reinit in the core code- not the platform code- so fix the ISP_CONN_FATAL case.
2003-02-16 01:32:52 +00:00
"DRIVER HEARTBEAT",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
NULL,
Implement a call to get the actual link data rate (if 23XX) so we can set whether it's a 2Gps or 1Gps link. MFC after: 1 week
2001-10-01 03:45:54 +00:00
"GET/SET DATA RATE",
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
NULL,
NULL,
"INIT FIRMWARE",
NULL,
"INIT LIP",
"GET FC-AL POSITION MAP",
"GET PORT DATABASE",
"CLEAR ACA",
"TARGET RESET",
"CLEAR TASK SET",
"ABORT TASK SET",
"GET FW STATE",
"GET PORT NAME",
"GET LINK STATUS",
"INIT LIP RESET",
NULL,
"SEND SNS",
"FABRIC LOGIN",
"SEND CHANGE REQUEST",
"FABRIC LOGOUT",
"INIT LIP LOGIN",
NULL,
"LOGIN LOOP PORT",
"GET PORT/NODE NAME LIST",
"SET VENDOR ID",
"INITIALIZE IP MAILBOX",
NULL,
NULL,
NULL,
NULL,
"Get ID List",
"SEND LFA",
"Lun RESET"
};
#endif
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_mboxcmd_qnw(ispsoftc_t *isp, mbreg_t *mbp, int nodelay)
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
{
Remove unused variables Add /* FALLTHROUGH */ Found by: FlexeLint
2003-05-31 19:49:49 +00:00
unsigned int ibits, obits, box, opcode;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
const uint32_t *mcp;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
Restore parentheses removed inappropriately in last commit.
2003-06-01 19:01:01 +00:00
if (IS_FC(isp)) {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
mcp = mbpfc;
Restore parentheses removed inappropriately in last commit.
2003-06-01 19:01:01 +00:00
} else {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
mcp = mbpscsi;
Restore parentheses removed inappropriately in last commit.
2003-06-01 19:01:01 +00:00
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
opcode = mbp->param[0];
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
ibits = HIWRD(mcp[opcode]) & NMBOX_BMASK(isp);
obits = LOWRD(mcp[opcode]) & NMBOX_BMASK(isp);
ibits |= mbp->ibits;
obits |= mbp->obits;
for (box = 0; box < MAX_MAILBOX(isp); box++) {
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (ibits & (1 << box)) {
ISP_WRITE(isp, MBOX_OFF(box), mbp->param[box]);
}
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
if (nodelay == 0) {
isp->isp_mboxtmp[box] = mbp->param[box] = 0;
}
}
if (nodelay == 0) {
isp->isp_lastmbxcmd = opcode;
isp->isp_obits = obits;
isp->isp_mboxbsy = 1;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
}
ISP_WRITE(isp, HCCR, HCCR_CMD_SET_HOST_INT);
Fix a problem where a local loop disk logs out- and we get a PORT LOGGED OUT status. We are, apparently, required to force the f/w to log back in if we want to try and talk to that disk again. This means either issuing a LOGIN LOCAL LOOP PORT mailbox command, or by issuing a LIP. I've elected to issue a LIP because this has a better chance of waking up the disk which clearly just crashed and burned. These should not occur at all. If they do, they should be darned rare. MFC after: 1 week
2002-02-21 01:56:08 +00:00
/*
* Oddly enough, if we're not delaying for an answer,
* delay a bit to give the f/w a chance to pick up the
* command.
*/
if (nodelay) {
USEC_DELAY(1000);
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_mboxcmd(ispsoftc_t *isp, mbreg_t *mbp, int logmask)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
char *cname, *xname, tname[16], mname[16];
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
unsigned int lim, ibits, obits, box, opcode;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
const uint32_t *mcp;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if (IS_FC(isp)) {
mcp = mbpfc;
lim = (sizeof (mbpfc) / sizeof (mbpfc[0]));
} else {
mcp = mbpscsi;
lim = (sizeof (mbpscsi) / sizeof (mbpscsi[0]));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if ((opcode = mbp->param[0]) >= lim) {
mbp->param[0] = MBOX_INVALID_COMMAND;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "Unknown Command 0x%x", opcode);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
return;
Remove the Target mode functions until they're in better shape. Implement some suggested compilation cleanups from Eklund. Wire down a hard loop id if we are not on a platform that has the ability to get to a PCI BIOS (it still will float to the ID it gets after a LIP but at least we can try). Clarify that the expanded lun is based upon SCCLUN defines (in f/w).
1998-12-05 01:33:57 +00:00
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
ibits = HIWRD(mcp[opcode]) & NMBOX_BMASK(isp);
obits = LOWRD(mcp[opcode]) & NMBOX_BMASK(isp);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
/*
* Pick up any additional bits that the caller might have set.
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
ibits |= mbp->ibits;
obits |= mbp->obits;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if (ibits == 0 && obits == 0) {
mbp->param[0] = MBOX_COMMAND_PARAM_ERROR;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGERR, "no parameters for 0x%x", opcode);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
return;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
/*
* Get exclusive usage of mailbox registers.
*/
MBOX_ACQUIRE(isp);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (box = 0; box < MAX_MAILBOX(isp); box++) {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if (ibits & (1 << box)) {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
isp_prt(isp, ISP_LOGDEBUG1, "IN mbox %d = 0x%x", box,
mbp->param[box]);
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
ISP_WRITE(isp, MBOX_OFF(box), mbp->param[box]);
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
}
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
isp->isp_mboxtmp[box] = mbp->param[box] = 0;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
When resetting the Qlogic 2X00 units, reset the FPM (Fibre Protocol Module) and FBM (Fibre Buffer Modules). Also remember to clear the semaphore registers. Tell the RISC processor to not halt on FPM parity errors. Throw out the ISP_CFG_NOINIT silliness and instead go to the use of adapter 'roles' to see whether one completes initialization or not (mostly for Fibre Channel). The ultimate intent, btw, of all of this is to have a warm standby adapter for failover reasons. Because we do roles now, setting of Target Capable Class 3 service parameters in the ICB for the 2x00 cards reflects from role. Also, in isp_start, if we're not supporting an initiator role, we bounce outgoing commands with a Selection Timeout error. Also clean out the TOGGLE_TMODE goop for FC- there is no toggling of target mode like there is for parallel SCSI cards. Do more cleanup with respect to using target ids 0..125 in F-port topologies. Also keep track of things which *were* fabric devices so that when you rescan the fabric you can notify the outer layers when fabric devices go away. Only force a LOGOUT for fabric devices if they're still logged in (i.e., you cat their Port Database entry. Clean up the Get All Next scanning. Finally, use a new tag in the softc to store the opcode for the last mailbox command used so we can report which opcode timed out.
2001-01-15 18:33:08 +00:00
isp->isp_lastmbxcmd = opcode;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
* We assume that we can't overwrite a previous command.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
isp->isp_obits = obits;
isp->isp_mboxbsy = 1;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Set Host Interrupt condition so that RISC will pick up mailbox regs.
*/
ISP_WRITE(isp, HCCR, HCCR_CMD_SET_HOST_INT);
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
/*
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
* While we haven't finished the command, spin our wheels here.
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
*/
Change delay loop in new isp_mboxcmd to the use of the new MBOX_WAIT_COMPLETE macro. Change notification of completion of a mailbox command in isp_intr to MBOX_NOTIFY_COMPLETE macro.
2000-07-04 01:02:38 +00:00
MBOX_WAIT_COMPLETE(isp);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
/*
* Did the command time out?
*/
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (isp->isp_mboxbsy) {
isp->isp_mboxbsy = 0;
MBOX_RELEASE(isp);
return;
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
* Copy back output registers.
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
*/
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (box = 0; box < MAX_MAILBOX(isp); box++) {
Fix usage of DELAY (SYS_DELAY is the platform independent local define). Fix stupidity wrt checking whether we've gone to LOOP_PDB_RCVD loopstate- it's okay to be greater than this state. D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC macros. Completely redo mailbox command routine (in preparation to make this possibly wait rather than poll for completion). Make a major attempt to solve the 'lost interrupt' problem 1. Problem The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate regular SCSI command placed on the request queue would never complete and the watchdog routine in the driver would eventually wakeup and catch it. This would typically only happen on Alphas, although a couple folks with 700MHz Intel platforms have also seen this. For a long time I thought it was a foulup with f/w negotiations of SYNC and/or WIDE as it always seemed to happen right after the platform it was running on had done a SET TARGET PARAMETERS mailbox command to (re)enable sync && wide (after initially forcing ASYNC/NARROW at startup). However, occasionally, the same thing would also occur for the Fibre Channel cards as well (which, ahem, have no SET TARGET PARAMETERS for transfer mode). After finally putting in a better set of watchdog routines for the platforms for this driver, it seemed to be the case that the command in question (usually a READ CAPACITY) just had up and died- the watchdog routine would catch it after ~10 seconds. For some platforms (NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which would always fail- indicating that the f/w denied knowledge of this command, i.e., the f/w thought it was a done command). In any case, retrying the command worked. But this whole problem needed to be really fixed. 2. A False Step That Went in The Right Direction The mailbox code was completely rewritten to no longer try and grab the mailbox semaphore register and to try and 'by hand' complete async fast posting completions. It was also rewritten to now have separate in && out bitpatterns for registers to load to start and retrieve to complete. This means that isp_intr now handles mailbox completions. This substantially simplifies the mailbox handling code, and carries things 90% toward getting this to be a non-polled routine for this driver. This did not solve the problem, though. 3. Register Debouncing I saw some comments in some errata sheets and some notes in a Qlogic produced Linux driver (for the Qlogic 2100) that seemed to indicate that debouncing of reads of the mailbox registers might be needed, so I added this. This did not affect the problem. In fact, it made the problem worse for non-2100 cards. 5. Interrupt masking/unmasking The driver *used* to do a substantial amount of masking/unmasking of the interrupt control register. This was done to make sure that the core common code could just assume it would never get pre-empted. This apparently substantially contributed to the lost interrupt problem. The rewrite of the ICR (Interrupt Control Register), which is a separate register from the ISR (Interrupt Status Register) should not have caused any change to interrupt assertions pending. The manual does not state that it will, and the register layout seems to imply that the ICR is just an active route gate. We only enable PCI Interrupts and RISC Interrupts- this should mean that when the f/w asserts a RISC interrupt and (and the ICR allows RISC Interrupts) and we have PCI Interrupts enabled, we should get a PCI interrupt. Apparently this is a latch- not a signal route. Removing this got rid of *most* but not all, lost interrupts. 5. Watchdog Smartening I made sure that the watchdog routine would catch cases where the Qlogic's ISR showed an interrupt assertion. The watchdog routine now calls the interrupt service routine if it sees this. Some additional internal state flags were added so that the watchdog routine could then know whether the command it was in the middle of burying (because we had time it out) was in fact completed by the interrupt service routine. 6. Occasional Constipation Of Commands.. In running some very strenous high IOPs tests (generating about 11000 interrupts/second across one Qlogic 1040, one Qlogic 1080 and one Qlogic 2200 on an Alpha PC164), I found that I would get occasional but regular 'watchdog timeouts' on both the 1080 and the 2100 cards. This is under FreeBSD, and the watchdog timeout routine just marks the command in error and retries it. Invariably, right after this 'watchdog timeout' error, I'd get a command completion for the command that I had thought timed out. That is, I'd get a command completion, but the handle returned by the firmware mapped to no current command. The frequency of this problem is low under such a load- it would usually take an 30 minutes per 'lost' interrupt. I doubled the timeout for commands to see if it just was an edge case of waiting too short a period. This has no effect. I gathered and printed out microtimes for the watchdog completed command and the completion that couldn't find a command- it was always the case that the order of occurrence was "timeout, completion" separated by a time on the order of 100 to 150 ms. This caused me to consider 'firmware constipation' as to be a possible culprit. That is, resubmission of a command to the device that had suffered a watchdog timeout seemed to cause the presumed dead command to show back up. I added code in the watchdog routine that, when first entered for the command, marks the command with a flag, reissues a local timeout call for one second later, but also then issues a MARKER Request Queue entry to the Qlogic f/w. A MARKER entry is used typically after a Bus Reset to cause the f/w to get synchronized with respect to either a Bus, a Nexus or a Target. Since I've added this code, I always now see the occasional watchdog timeout, but the command that was about to be terminated always now seems to be completed after the MARKER entry is issued (and before the timeout extension fires, which would come back and *really* terminate the command).
2000-06-27 19:44:31 +00:00
if (obits & (1 << box)) {
mbp->param[box] = isp->isp_mboxtmp[box];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
isp_prt(isp, ISP_LOGDEBUG1, "OUT mbox %d = 0x%x", box,
mbp->param[box]);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
MBOX_RELEASE(isp);
if (logmask == 0 || opcode == MBOX_EXEC_FIRMWARE) {
return;
}
#ifdef ISP_STRIPPED
cname = NULL;
#else
cname = (IS_FC(isp))? fc_mbcmd_names[opcode] : scsi_mbcmd_names[opcode];
#endif
if (cname == NULL) {
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
cname = tname;
SNPRINTF(tname, sizeof tname, "opcode %x", opcode);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
/*
* Just to be chatty here...
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = NULL;
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
switch (mbp->param[0]) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
case MBOX_COMMAND_COMPLETE:
break;
case MBOX_INVALID_COMMAND:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_COMMAND_COMPLETE)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "INVALID COMMAND";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case MBOX_HOST_INTERFACE_ERROR:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_HOST_INTERFACE_ERROR)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "HOST INTERFACE ERROR";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case MBOX_TEST_FAILED:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_TEST_FAILED)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "TEST FAILED";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case MBOX_COMMAND_ERROR:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_COMMAND_ERROR)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "COMMAND ERROR";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
case MBOX_COMMAND_PARAM_ERROR:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_COMMAND_PARAM_ERROR)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "COMMAND PARAMETER ERROR";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
case MBOX_LOOP_ID_USED:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_LOOP_ID_USED)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "LOOP ID ALREADY IN USE";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
break;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
case MBOX_PORT_ID_USED:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_PORT_ID_USED)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "PORT ID ALREADY IN USE";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
break;
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
case MBOX_ALL_IDS_USED:
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (logmask & MBLOGMASK(MBOX_ALL_IDS_USED)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
xname = "ALL LOOP IDS IN USE";
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
break;
case 0: /* special case */
xname = "TIMEOUT";
Slightly cleaner fabric support (whiter whites! redder reds!).. No, seriously- only attempt to logout a previously logged in fabric device. Fix a longstanding bug for aborting overtime commands- handle halves have always been reversed. Clean up some error messages to indicate channel number. Approved:jkh
2000-02-29 05:52:14 +00:00
break;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
default:
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
SNPRINTF(mname, sizeof mname, "error 0x%x", mbp->param[0]);
xname = mname;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
break;
}
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
if (xname) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGALL, "Mailbox Command '%s' failed (%s)",
cname, xname);
Add some missing braces. Add MEMORY_BARRIER for the few scratch dma ops that were missing them plus add a couple of hi 32 bit dma ops (we could probably allow 64 bit scratch and request/response queue dma now).
2006-07-14 05:14:48 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_fw_state(ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_FC(isp)) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
mbreg_t mbs;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
fcparam *fcp = isp->isp_param;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbs.param[0] = MBOX_GET_FW_STATE;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
fcp->isp_fwstate = mbs.param[1];
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_update(ispsoftc_t *isp)
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
{
various fixes
2000-08-27 23:38:44 +00:00
int bus, upmask;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
various fixes
2000-08-27 23:38:44 +00:00
for (bus = 0, upmask = isp->isp_update; upmask != 0; bus++) {
if (upmask & (1 << bus)) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp_update_bus(isp, bus);
}
various fixes
2000-08-27 23:38:44 +00:00
upmask &= ~(1 << bus);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_update_bus(ispsoftc_t *isp, int bus)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
int tgt;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
mbreg_t mbs;
sdparam *sdp;
various fixes
2000-08-27 23:38:44 +00:00
isp->isp_update &= ~(1 << bus);
More code cleanup. Go back to using FULL_LOGIN Fibre Chan if f/w is less than 1.17.0 level. Change where we do the loop database init. Add in the CMD_RQLATER return. Add some register debounce.
1999-08-16 19:59:55 +00:00
if (IS_FC(isp)) {
various fixes
2000-08-27 23:38:44 +00:00
/*
* There are no 'per-bus' settings for Fibre Channel.
*/
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
return;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
sdp = isp->isp_param;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
sdp += bus;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t flags, period, offset;
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
int get;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
if (sdp->isp_devparam[tgt].dev_enable == 0) {
various fixes
2000-08-27 23:38:44 +00:00
sdp->isp_devparam[tgt].dev_update = 0;
sdp->isp_devparam[tgt].dev_refresh = 0;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"skipping target %d bus %d update", tgt, bus);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
continue;
}
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
/*
* If the goal is to update the status of the device,
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
* take what's in goal_flags and try and set the device
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
* toward that. Otherwise, if we're just refreshing the
* current device state, get the current parameters.
*/
various fixes
2000-08-27 23:38:44 +00:00
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
various fixes
2000-08-27 23:38:44 +00:00
/*
* Refresh overrides set
*/
if (sdp->isp_devparam[tgt].dev_refresh) {
mbs.param[0] = MBOX_GET_TARGET_PARAMS;
sdp->isp_devparam[tgt].dev_refresh = 0;
get = 1;
} else if (sdp->isp_devparam[tgt].dev_update) {
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
mbs.param[0] = MBOX_SET_TARGET_PARAMS;
various fixes
2000-08-27 23:38:44 +00:00
/*
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
* Make sure goal_flags has "Renegotiate on Error"
various fixes
2000-08-27 23:38:44 +00:00
* on and "Freeze Queue on Error" off.
*/
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].goal_flags |= DPARM_RENEG;
sdp->isp_devparam[tgt].goal_flags &= ~DPARM_QFRZ;
various fixes
2000-08-27 23:38:44 +00:00
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
mbs.param[2] = sdp->isp_devparam[tgt].goal_flags;
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/*
various fixes
2000-08-27 23:38:44 +00:00
* Insist that PARITY must be enabled
* if SYNC or WIDE is enabled.
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
*/
various fixes
2000-08-27 23:38:44 +00:00
if ((mbs.param[2] & (DPARM_SYNC|DPARM_WIDE)) != 0) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
mbs.param[2] |= DPARM_PARITY;
}
various fixes
2000-08-27 23:38:44 +00:00
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
if (mbs.param[2] & DPARM_SYNC) {
various fixes
2000-08-27 23:38:44 +00:00
mbs.param[3] =
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
(sdp->isp_devparam[tgt].goal_offset << 8) |
(sdp->isp_devparam[tgt].goal_period);
various fixes
2000-08-27 23:38:44 +00:00
}
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
/*
* A command completion later that has
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
* RQSTF_NEGOTIATION set can cause
various fixes
2000-08-27 23:38:44 +00:00
* the dev_refresh/announce cycle also.
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
*
* Note: It is really important to update our current
* flags with at least the state of TAG capabilities-
* otherwise we might try and send a tagged command
* when we have it all turned off. So change it here
* to say that current already matches goal.
*/
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].actv_flags &= ~DPARM_TQING;
sdp->isp_devparam[tgt].actv_flags |=
(sdp->isp_devparam[tgt].goal_flags & DPARM_TQING);
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
"bus %d set tgt %d flags 0x%x off 0x%x period 0x%x",
bus, tgt, mbs.param[2], mbs.param[3] >> 8,
mbs.param[3] & 0xff);
various fixes
2000-08-27 23:38:44 +00:00
sdp->isp_devparam[tgt].dev_update = 0;
sdp->isp_devparam[tgt].dev_refresh = 1;
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
get = 0;
} else {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
continue;
}
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
mbs.param[1] = (bus << 15) | (tgt << 8);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_mboxcmd(isp, &mbs, MBLOGALL);
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
if (get == 0) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
isp->isp_sendmarker |= (1 << bus);
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 07:29:00 +00:00
continue;
}
flags = mbs.param[2];
period = mbs.param[3] & 0xff;
offset = mbs.param[3] >> 8;
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].actv_flags = flags;
sdp->isp_devparam[tgt].actv_period = period;
sdp->isp_devparam[tgt].actv_offset = offset;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
get = (bus << 16) | tgt;
(void) isp_async(isp, ISPASYNC_NEW_TGT_PARAMS, &get);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
if (sdp->isp_devparam[tgt].dev_update ||
sdp->isp_devparam[tgt].dev_refresh) {
isp->isp_update |= (1 << bus);
break;
}
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
#ifndef DEFAULT_FRAMESIZE
#define DEFAULT_FRAMESIZE(isp) ICB_DFLT_FRMLEN
#endif
#ifndef DEFAULT_EXEC_THROTTLE
#define DEFAULT_EXEC_THROTTLE(isp) ISP_EXEC_THROTTLE
#endif
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_setdfltparm(ispsoftc_t *isp, int channel)
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
{
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
int tgt;
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
sdparam *sdp;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_FC(isp)) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
fcparam *fcp = (fcparam *) isp->isp_param;
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
int nvfail;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
fcp += channel;
if (fcp->isp_gotdparms) {
return;
}
fcp->isp_gotdparms = 1;
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
fcp->isp_maxfrmlen = DEFAULT_FRAMESIZE(isp);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
fcp->isp_maxalloc = ICB_DFLT_ALLOC;
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
fcp->isp_execthrottle = DEFAULT_EXEC_THROTTLE(isp);
Roll revision levels. Add support for the Qlogic 2200 (warn about not having SCSI_ISP_SCCLUN config defined if we don't have f/w for the 2200- it's resident firmware uses SCCLUN (65535 luns)). Change the way the default LoopID is gathered (it's now a platform specific define so that some attempt at a synthetic WWN can be made in case NVRAM isn't readable). Change initialization of options a bit- don't use ADISC. Set FullDuplex mode if config options tells us to do so. Do not use FULL_LOGIN after LIP- it's the right thing to do but it causes too much loop disruption (Loop Resets). Sanity check some default values. Redo construction of port and node WWNs based upon what we have- if we have 2 in the top nibble, we can have distinct port and node WWNs. Clean up some SCCLUN related code that we obviously had never compiled (:-(). Audit commands coming int ispscsicmd and don't throw commands at Fibre devices that do not have Class 3 service parameters TARGET ROLE defined. Clean up f/w initialization a bit. Add Fabric support (or at least the first blush of it). Whew - way too much to describe here. Basically, after a LIP, hang out until we see a Loop Up or a Port DataBase Change async event, then see if we're on a Fabric (GET_PORT_NAME of FL_PORT_ID). If we are, try and scan the fabric controller for fabric devices using the GetAllNext SNS subcommand. As we find devices, announce them to the outer layer. Try and do some guard code for broken (Brocade) SNS servers (that get stuck in loops- gotta maybe do this a different way using the GP_ID3 cmd instead). Then do a scan of the lower (local loop) ids using a GET_PORT_NAME to see if the f/w has logged into anything at that loop id. If so, then do a GET_PORT_DATABASE command. Do this scan into a local database. At this point we can say the loop is 'Ready'. After this, we merge our local loop port database with our stored port database- in a as yet to be really fully exercised fashion we try and follow the logic of something having moved around. The first time we see something at a Loop ID, we fix it, for the purpose of this system instance, at that Loop ID. If things shift around so it ends up somewhere else, we still keep it at this Loop ID (our 'Target') but use the new (moved) Loop ID when we actually throw commands at it. Check for insane cases of different Loop IDs both claiming to have the same WWN- if that happens, invalidate both. Notify the outer layer of devices that have arrived and devices that have gone away. *Finally*, when this is done, search the softc's database of Fabric devices and perform logout/login actions. The Qlogic f/w maintains logout/login for all local loop devices. We have to maintain logout/login for fabric devices- total PITA. Expect to see this area undergo more change over time.
1999-07-02 23:06:38 +00:00
fcp->isp_retry_delay = ICB_DFLT_RDELAY;
fcp->isp_retry_count = ICB_DFLT_RCOUNT;
/* Platform specific.... */
fcp->isp_loopid = DEFAULT_LOOPID(isp);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
fcp->isp_nodewwn = DEFAULT_NODEWWN(isp);
fcp->isp_portwwn = DEFAULT_PORTWWN(isp);
various fixes
2000-08-27 23:38:44 +00:00
fcp->isp_fwoptions = 0;
fcp->isp_fwoptions |= ICBOPT_FAIRNESS;
fcp->isp_fwoptions |= ICBOPT_PDBCHANGE_AE;
fcp->isp_fwoptions |= ICBOPT_HARD_ADDRESS;
fcp->isp_fwoptions |= ICBOPT_FAST_POST;
if (isp->isp_confopts & ISP_CFG_FULL_DUPLEX)
fcp->isp_fwoptions |= ICBOPT_FULL_DUPLEX;
/*
* Make sure this is turned off now until we get
* extended options from NVRAM
*/
fcp->isp_fwoptions &= ~ICBOPT_EXTENDED;
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
* Now try and read NVRAM unless told to not do so.
* This will set fcparam's isp_nodewwn && isp_portwwn.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*/
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
if ((isp->isp_confopts & ISP_CFG_NONVRAM) == 0) {
nvfail = isp_read_nvram(isp);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (nvfail) {
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
isp->isp_confopts |= ISP_CFG_NONVRAM;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
}
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
} else {
nvfail = 1;
}
/*
* Set node && port to override platform set defaults
* unless the nvram read failed (or none was done),
* or the platform code wants to use what had been
* set in the defaults.
*/
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
if (nvfail) {
isp->isp_confopts |= ISP_CFG_OWNWWPN|ISP_CFG_OWNWWNN;
}
if (isp->isp_confopts & ISP_CFG_OWNWWNN) {
isp_prt(isp, ISP_LOGCONFIG, "Using Node WWN 0x%08x%08x",
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(uint32_t) (DEFAULT_NODEWWN(isp) >> 32),
(uint32_t) (DEFAULT_NODEWWN(isp) & 0xffffffff));
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
ISP_NODEWWN(isp) = DEFAULT_NODEWWN(isp);
} else {
/*
* We always start out with values derived
* from NVRAM or our platform default.
*/
ISP_NODEWWN(isp) = fcp->isp_nodewwn;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (fcp->isp_nodewwn == 0) {
isp_prt(isp, ISP_LOGCONFIG,
"bad WWNN- using default\n");
ISP_NODEWWN(isp) = DEFAULT_NODEWWN(isp);
}
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
}
if (isp->isp_confopts & ISP_CFG_OWNWWPN) {
isp_prt(isp, ISP_LOGCONFIG, "Using Port WWN 0x%08x%08x",
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(uint32_t) (DEFAULT_PORTWWN(isp) >> 32),
(uint32_t) (DEFAULT_PORTWWN(isp) & 0xffffffff));
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
ISP_PORTWWN(isp) = DEFAULT_PORTWWN(isp);
} else {
/*
* We always start out with values derived
* from NVRAM or our platform default.
*/
ISP_PORTWWN(isp) = fcp->isp_portwwn;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
if (fcp->isp_portwwn == 0) {
isp_prt(isp, ISP_LOGCONFIG,
"bad WWPN- using default\n");
ISP_PORTWWN(isp) = DEFAULT_PORTWWN(isp);
}
Add a isp_register_fc4_type function so that we work with McData switches that require us to register our FC4 types of interest. Allow ourselves, in F-port topologies, to start logging in fabric devices in the target 0..125 range. Change ISPASYNC_PDB_CHANGED (misnamed) to ISPASYNC_LOGGED_INOUT. Fix (*SMACK*) again some default WWN stuff. This is *really* hard to get right across all the range of platforms.
2001-01-09 02:46:23 +00:00
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
return;
}
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
sdp = (sdparam *) isp->isp_param;
sdp += channel;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/*
* Been there, done that, got the T-shirt...
*/
if (sdp->isp_gotdparms) {
return;
}
sdp->isp_gotdparms = 1;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
/*
* Establish some default parameters.
*/
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
sdp->isp_cmd_dma_burst_enable = 0;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
sdp->isp_data_dma_burst_enabl = 1;
sdp->isp_fifo_threshold = 0;
sdp->isp_initiator_id = DEFAULT_IID(isp);
if (isp->isp_type >= ISP_HA_SCSI_1040) {
sdp->isp_async_data_setup = 9;
} else {
sdp->isp_async_data_setup = 6;
}
sdp->isp_selection_timeout = 250;
sdp->isp_max_queue_depth = MAXISPREQUEST(isp);
sdp->isp_tag_aging = 8;
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
sdp->isp_bus_reset_delay = 5;
/*
* Don't retry selection, busy or queue full automatically- reflect
* these back to us.
*/
sdp->isp_retry_count = 0;
sdp->isp_retry_delay = 0;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
sdp->isp_devparam[tgt].exc_throttle = ISP_EXEC_THROTTLE;
sdp->isp_devparam[tgt].dev_enable = 1;
}
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
/*
* If we've not been told to avoid reading NVRAM, try and read it.
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
* If we're successful reading it, we can then return because NVRAM
* will tell us what the desired settings are. Otherwise, we establish
* some reasonable 'fake' nvram and goal defaults.
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
*/
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if ((isp->isp_confopts & ISP_CFG_NONVRAM) == 0) {
if (isp_read_nvram(isp) == 0) {
return;
}
}
/*
* Now try and see whether we have specific values for them.
*/
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
if ((isp->isp_confopts & ISP_CFG_NONVRAM) == 0) {
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
mbreg_t mbs;
MEMZERO(&mbs, sizeof (mbs));
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
mbs.param[0] = MBOX_GET_ACT_NEG_STATE;
After loading f/w, for FC cards print out Firmware Attributes. Redo establishment of default SCSI parameters whether or not we've been compiled for target mode. Unfortunately, the Qlogic f/w is confused so that if we set all targets to be 'safe' (i.e., narrow/async), it will also then report narrow, async if we're contacted in target mode from that target (acting in initiator role). D'oh! Fix ISPCTL_TOGGLE_TMODE to correctly enable the right channel for dual channel cards. Add some more opcodes. Fix a stupid NULL pointer bug.
2001-04-04 21:42:59 +00:00
isp_mboxcmd(isp, &mbs, MBLOGNONE);
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
sdp->isp_req_ack_active_neg = 1;
sdp->isp_data_line_active_neg = 1;
} else {
sdp->isp_req_ack_active_neg =
(mbs.param[1+channel] >> 4) & 0x1;
sdp->isp_data_line_active_neg =
(mbs.param[1+channel] >> 5) & 0x1;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
isp_prt(isp, ISP_LOGDEBUG0, sc0, sc3,
0, sdp->isp_fifo_threshold, sdp->isp_initiator_id,
sdp->isp_bus_reset_delay, sdp->isp_retry_count,
sdp->isp_retry_delay, sdp->isp_async_data_setup);
isp_prt(isp, ISP_LOGDEBUG0, sc1, sc3,
sdp->isp_req_ack_active_neg, sdp->isp_data_line_active_neg,
sdp->isp_data_dma_burst_enabl, sdp->isp_cmd_dma_burst_enable,
sdp->isp_selection_timeout, sdp->isp_max_queue_depth);
Clean up some f/w revision checking wrt enabling fast posting. Make sure we set defaults sanely for dual-bus adapters.
1999-12-20 01:34:01 +00:00
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
/*
* The trick here is to establish a default for the default (honk!)
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
* state (goal_flags). Then try and get the current status from
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
* the card to fill in the current state. We don't, in fact, set
* the default to the SAFE default state- that's not the goal state.
*/
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t off, per;
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].actv_offset = 0;
sdp->isp_devparam[tgt].actv_period = 0;
sdp->isp_devparam[tgt].actv_flags = 0;
sdp->isp_devparam[tgt].goal_flags =
sdp->isp_devparam[tgt].nvrm_flags = DPARM_DEFAULT;
clean up for SBus Ultra (yes, we do not do that here yet)
2000-01-15 01:52:01 +00:00
/*
* We default to Wide/Fast for versions less than a 1040
* (unless it's SBus).
*/
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
if (IS_ULTRA3(isp)) {
off = ISP_80M_SYNCPARMS >> 8;
per = ISP_80M_SYNCPARMS & 0xff;
} else if (IS_ULTRA2(isp)) {
off = ISP_40M_SYNCPARMS >> 8;
per = ISP_40M_SYNCPARMS & 0xff;
} else if (IS_1240(isp)) {
off = ISP_20M_SYNCPARMS >> 8;
per = ISP_20M_SYNCPARMS & 0xff;
} else if ((isp->isp_bustype == ISP_BT_SBUS &&
clean up for SBus Ultra (yes, we do not do that here yet)
2000-01-15 01:52:01 +00:00
isp->isp_type < ISP_HA_SCSI_1020A) ||
(isp->isp_bustype == ISP_BT_PCI &&
isp->isp_type < ISP_HA_SCSI_1040) ||
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
(isp->isp_clock && isp->isp_clock < 60) ||
(sdp->isp_ultramode == 0)) {
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
off = ISP_10M_SYNCPARMS >> 8;
per = ISP_10M_SYNCPARMS & 0xff;
various fixes
2000-08-27 23:38:44 +00:00
} else {
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
off = ISP_20M_SYNCPARMS_1040 >> 8;
per = ISP_20M_SYNCPARMS_1040 & 0xff;
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
}
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
sdp->isp_devparam[tgt].goal_offset =
sdp->isp_devparam[tgt].nvrm_offset = off;
sdp->isp_devparam[tgt].goal_period =
Oops- don't set 'goal' twice when you mean to set 'nvrm' as well. This breaks bogus NVRAM boards. MFC after: 1 day
2001-08-02 00:34:56 +00:00
sdp->isp_devparam[tgt].nvrm_period = per;
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
isp_prt(isp, ISP_LOGDEBUG0, sc2, sc3,
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
channel, tgt, sdp->isp_devparam[tgt].nvrm_flags,
sdp->isp_devparam[tgt].nvrm_offset,
sdp->isp_devparam[tgt].nvrm_period);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
Add in 1080 LVD support and some basis also for the 1240. The port database printout is now enabled.
1999-03-25 22:52:45 +00:00
/*
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
* Re-initialize the ISP and complete all orphaned commands
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
* with a 'botched' notice. The reset/init routines should
* not disturb an already active list of commands.
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*
* Locks held prior to coming here.
*/
void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_reinit(ispsoftc_t *isp)
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
{
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
XS_T *xs;
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
int i;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Revert previous commit. Violates Maintainer (O'Brien knows how to reach me directly), but more importantly, breaks compiles on non-FreeBSD platforms.
2003-08-25 17:58:23 +00:00
if (IS_FC(isp)) {
isp_mark_getpdb_all(isp);
}
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
isp_reset(isp);
Minor stuff: Remove ISP2100_FABRIC defines- we always handle fabric now. Insert isp_getmap helper function (for getting Loop Position map). Make sure we (for our own benefit) mark req_state_flags with RQSF_GOT_SENSE for Fibre Channel if we got sense data- the !*$)!*$)~*$)*$ Qlogic f/w doesn't do so. Add ISPCTL_SCAN_FABRIC, ISPCTL_SCAN_LOOP, ISPCTL_SEND_LIP, and ISPCTL_GET_POSMAP isp_control functions. Correctly send async notifications upstream for changes in the name server, changes in the port database, and f/w crashes. Correctly set topology when we get a ASYNC_PTPMODE event. Major stuff: Quite massively redo how we handle Loop events- we've now added several intermediate states between LOOP_PDB_RCVD and LOOP_READY. This allows us a lot finer control about how we scan fabric, whether we go further than scanning fabric, how we look at the local loop, and whether we merge entries at the level or not. This is the next to last step for moving managing loop state out of the core module entirely (whereupon loop && fabric events will simply freeze the command queue and a thread will run to figure out what's changed and *it* will re-enable the queu). This fine amount of control also gets us closer to having an external policy engine decide which fabric devices we really want to log into.
2001-02-11 03:44:43 +00:00
if (isp->isp_state != ISP_RESETSTATE) {
isp_prt(isp, ISP_LOGERR, "isp_reinit cannot reset card");
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
} else if (isp->isp_role != ISP_ROLE_NONE) {
isp_init(isp);
if (isp->isp_state == ISP_INITSTATE) {
isp->isp_state = ISP_RUNSTATE;
}
if (isp->isp_state != ISP_RUNSTATE) {
isp_prt(isp, ISP_LOGERR,
"isp_reinit cannot restart card");
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
isp->isp_nactive = 0;
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
for (i = 0; i < isp->isp_maxcmds; i++) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t handle;
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
xs = isp->isp_xflist[i];
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
if (xs == NULL) {
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
continue;
ISP_DMASETUP now returns a value to be possibly punted to outer layers. Turn request queue overflow messages into debug messages. Ensure on isp_restarts that we nullify the xflist array.
1998-09-17 21:03:45 +00:00
}
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
handle = isp_index_handle(i);
Remove some target mode stuff. It will get re-introduced in a different file later. Do some pencil-sharpening types of minor changes. Change how active commands are remembered (using new inline functions to get handles, etc..). Now do a GET FIRMWARE STATUS after firing up the f/w as outgoing mailbox 2 will tell you the f/w's notion of the max commands that can be supported. Attempt to retrieve loop topology. Add in the appropriate SWIZZLE/UNSWIZZLE macros calls (this is a no-op on Little Endian machines but is needed for sparc (on other platforms)). Move the temp port database we use to find out where things have moved to after a LIP to the softc and off the kernel stack. Follow Qlogic's hint and don't bother setting a tag for commands that don't have this enabled (presumably the f/w will do it's own selection then). Use an INT_PENDING macro to check for an interrupt. The call to ISP_DMAFREE now just takes the handle- not the 'handle-1' which was a layering violation. Use CFGPRINTF in a couple of places to make things less chatty if not booting verbose, or CAMDEBUG compiles, etc..
1999-10-17 18:58:22 +00:00
isp_destroy_handle(isp, handle);
if (XS_XFRLEN(xs)) {
ISP_DMAFREE(isp, xs, handle);
XS_RESID(xs) = XS_XFRLEN(xs);
} else {
XS_RESID(xs) = 0;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
XS_SETERR(xs, HBA_BUSRESET);
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_done(xs);
Add support for the Qlogic ISP SCSI && FC/AL Adapters
1998-04-22 17:54:58 +00:00
}
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
/*
* NVRAM Routines
*/
static int
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_read_nvram(ispsoftc_t *isp)
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
{
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
int i, amt, retval;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t csum, minversion;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
union {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t _x[ISP2100_NVRAM_SIZE];
uint16_t _s[ISP2100_NVRAM_SIZE>>1];
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
} _n;
#define nvram_data _n._x
#define nvram_words _n._s
A wad of changes- prepping for 1080/1240 support (which caused a massive thwank in register layout goop). A different mboxcmd approach. Some PDB change infrastructure. Some better management of loopdown/loopup events (keep them distinct from resource starvation for simq freeze/unfreeze actions).
1999-03-17 05:04:39 +00:00
if (IS_FC(isp)) {
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
amt = ISP2100_NVRAM_SIZE;
minversion = 1;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
} else if (IS_ULTRA2(isp)) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
amt = ISP1080_NVRAM_SIZE;
minversion = 0;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
} else {
amt = ISP_NVRAM_SIZE;
minversion = 2;
}
/*
* Just read the first two words first to see if we have a valid
* NVRAM to continue reading the rest with.
*/
for (i = 0; i < 2; i++) {
isp_rdnvram_word(isp, i, &nvram_words[i]);
}
if (nvram_data[0] != 'I' || nvram_data[1] != 'S' ||
nvram_data[2] != 'P') {
if (isp->isp_bustype != ISP_BT_SBUS) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "invalid NVRAM header");
isp_prt(isp, ISP_LOGDEBUG0, "%x %x %x",
nvram_data[0], nvram_data[1], nvram_data[2]);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
retval = -1;
goto out;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
for (i = 2; i < amt>>1; i++) {
isp_rdnvram_word(isp, i, &nvram_words[i]);
}
for (csum = 0, i = 0; i < amt; i++) {
csum += nvram_data[i];
}
if (csum != 0) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "invalid NVRAM checksum");
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
retval = -1;
goto out;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
if (ISP_NVRAM_VERSION(nvram_data) < minversion) {
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
isp_prt(isp, ISP_LOGWARN, "version %d NVRAM not understood",
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
ISP_NVRAM_VERSION(nvram_data));
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
retval = -1;
goto out;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
if (IS_ULTRA3(isp)) {
isp_parse_nvram_12160(isp, 0, nvram_data);
This should enable 10160 support. As best as I can tell, the same f/w as 12160 is used, and otherwise, this is just a single channel variant of the 10160. MFC after: 0 days
2002-10-11 17:28:01 +00:00
if (IS_12160(isp))
isp_parse_nvram_12160(isp, 1, nvram_data);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
} else if (IS_1080(isp)) {
isp_parse_nvram_1080(isp, 0, nvram_data);
} else if (IS_1280(isp) || IS_1240(isp)) {
isp_parse_nvram_1080(isp, 0, nvram_data);
isp_parse_nvram_1080(isp, 1, nvram_data);
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
} else if (IS_SCSI(isp)) {
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
isp_parse_nvram_1020(isp, nvram_data);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
} else {
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
isp_parse_nvram_2100(isp, nvram_data);
Roll internal release tag. Print out if we're in a 64 bit PCI slot. Use fast memory timing NVRAM parameter. Clean up and fix establishment of default target parameters. Don't use NVRAM if are flagged as not to do so (I had a busted NVRAM setup which I couldn't edit that enabled SYNC mode but disabled disconnect/reconnect and wide!!). Fix delays after resets. BUS resets not done in isp_init anymore- relegated to OS specific outer layers. Fix a buglet where you can get in a loop for a NULL xs in the completion list in isp_intr. Add in some defines that can disable fast posting. Add in code for Loop Up/Loop Down events that call into the outer layers as to what to do.
1999-02-09 01:07:06 +00:00
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
retval = 0;
out:
return (retval);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
#undef nvram_data
#undef nvram_words
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_rdnvram_word(ispsoftc_t *isp, int wo, uint16_t *rp)
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
{
int i, cbits;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
uint16_t bit, rqst, junk;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT|BIU_NVRAM_CLOCK);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
if (IS_FC(isp)) {
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
wo &= ((ISP2100_NVRAM_SIZE >> 1) - 1);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (IS_2312(isp) && isp->isp_port) {
wo += 128;
}
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
rqst = (ISP_NVRAM_READ << 8) | wo;
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
cbits = 10;
Add Dual LVD bus (1280) support
1999-12-16 05:42:02 +00:00
} else if (IS_ULTRA2(isp)) {
Some massive thwunking in initialization to handle dual bus adapters. More massive thwunking to include an XS_CHANNEL value. Some changes of how parameters are reported to outer layers (including bus, e.g.). Yet more stirring around in isp_mboxcmd to try and get it right. Decode of 1080/1240 NVRAM.
1999-05-11 05:06:55 +00:00
wo &= ((ISP1080_NVRAM_SIZE >> 1) - 1);
rqst = (ISP_NVRAM_READ << 8) | wo;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
cbits = 10;
} else {
wo &= ((ISP_NVRAM_SIZE >> 1) - 1);
rqst = (ISP_NVRAM_READ << 6) | wo;
cbits = 8;
}
/*
* Clock the word select request out...
*/
for (i = cbits; i >= 0; i--) {
if ((rqst >> i) & 1) {
bit = BIU_NVRAM_SELECT | BIU_NVRAM_DATAOUT;
} else {
bit = BIU_NVRAM_SELECT;
}
ISP_WRITE(isp, BIU_NVRAM, bit);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
junk = ISP_READ(isp, BIU_NVRAM); /* force PCI flush */
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
ISP_WRITE(isp, BIU_NVRAM, bit | BIU_NVRAM_CLOCK);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
junk = ISP_READ(isp, BIU_NVRAM); /* force PCI flush */
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
ISP_WRITE(isp, BIU_NVRAM, bit);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
junk = ISP_READ(isp, BIU_NVRAM); /* force PCI flush */
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
/*
* Now read the result back in (bits come back in MSB format).
*/
*rp = 0;
for (i = 0; i < 16; i++) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t rv;
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
*rp <<= 1;
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT|BIU_NVRAM_CLOCK);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
rv = ISP_READ(isp, BIU_NVRAM);
if (rv & BIU_NVRAM_DATAIN) {
*rp |= 1;
}
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
junk = ISP_READ(isp, BIU_NVRAM); /* force PCI flush */
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
ISP_WRITE(isp, BIU_NVRAM, 0);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
USEC_DELAY(10);
junk = ISP_READ(isp, BIU_NVRAM); /* force PCI flush */
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
ISP_SWIZZLE_NVRAM_WORD(isp, rp);
Update QLogic ISP support for CAM. Add preliminary target mode support. Submitted by: Matthew Jacob <mjacob@feral.com>
1998-09-15 08:42:56 +00:00
}
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_nvram_1020(ispsoftc_t *isp, uint8_t *nvram_data)
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
{
sdparam *sdp = (sdparam *) isp->isp_param;
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
int tgt;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_fifo_threshold =
ISP_NVRAM_FIFO_THRESHOLD(nvram_data) |
(ISP_NVRAM_FIFO_THRESHOLD_128(nvram_data) << 2);
Remove the 'bogus registrant' hack for fabric searches. It really turns out that there's something of a hole in our new fabric name server stuff. We ask the name server for entities that have registered as a specific type. That type is FC-SCSI. If the entity hasn't performed a REGISTER FC4 TYPES, the fabric nameserver won't return it. This brings this driver to a bit of a fork in the road as to what the right thing to do is. For servicing the needs of accessing FC-SCSI devices, this method is fine, and to be preferred. It is extremely unlikely we're interested in fabric devices that *don't* register correctly. If I ever get around to adding an FC-IP stack, then asking for devices that have registers as FC-IP types is also the right thing to do. So- asking the fabric nameserver for a specific type is fine, *as long as you are only interested in specific types*. If, on the other hand, you want to create (as for management tool support) a picture of everything on the fabric, this is *not* so fine. There are a large class of FC-SCSI *initiators* who *don't* correctly register, so we never will *see* them. Is this a problem? Yes, but only a little one. If we want to do such management tool support, we should probably run a *different* fabric nameserver query algorithm. Better yet, we should talk to the management nameserver in Brocade switches instead of the standard FC-GS-2 fabric nameserver (which can be unwieldy). Other changes: if we've overrrides marked, don't set some default values from reading NVRAM. This allows us to override things like EXEC throttle without having to ignore NVRAM entirely. MFC after: 1 week
2002-07-08 17:33:37 +00:00
if ((isp->isp_confopts & ISP_CFG_OWNLOOPID) == 0)
sdp->isp_initiator_id =
ISP_NVRAM_INITIATOR_ID(nvram_data);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_bus_reset_delay =
ISP_NVRAM_BUS_RESET_DELAY(nvram_data);
sdp->isp_retry_count =
ISP_NVRAM_BUS_RETRY_COUNT(nvram_data);
sdp->isp_retry_delay =
ISP_NVRAM_BUS_RETRY_DELAY(nvram_data);
sdp->isp_async_data_setup =
ISP_NVRAM_ASYNC_DATA_SETUP_TIME(nvram_data);
if (isp->isp_type >= ISP_HA_SCSI_1040) {
if (sdp->isp_async_data_setup < 9) {
sdp->isp_async_data_setup = 9;
}
} else {
if (sdp->isp_async_data_setup != 6) {
sdp->isp_async_data_setup = 6;
}
}
sdp->isp_req_ack_active_neg =
ISP_NVRAM_REQ_ACK_ACTIVE_NEGATION(nvram_data);
sdp->isp_data_line_active_neg =
ISP_NVRAM_DATA_LINE_ACTIVE_NEGATION(nvram_data);
sdp->isp_data_dma_burst_enabl =
ISP_NVRAM_DATA_DMA_BURST_ENABLE(nvram_data);
sdp->isp_cmd_dma_burst_enable =
ISP_NVRAM_CMD_DMA_BURST_ENABLE(nvram_data);
sdp->isp_tag_aging =
ISP_NVRAM_TAG_AGE_LIMIT(nvram_data);
sdp->isp_selection_timeout =
ISP_NVRAM_SELECTION_TIMEOUT(nvram_data);
sdp->isp_max_queue_depth =
ISP_NVRAM_MAX_QUEUE_DEPTH(nvram_data);
various fixes
2000-08-27 23:38:44 +00:00
sdp->isp_fast_mttr = ISP_NVRAM_FAST_MTTR_ENABLE(nvram_data);
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
isp_prt(isp, ISP_LOGDEBUG0, sc0, sc4,
0, sdp->isp_fifo_threshold, sdp->isp_initiator_id,
sdp->isp_bus_reset_delay, sdp->isp_retry_count,
sdp->isp_retry_delay, sdp->isp_async_data_setup);
isp_prt(isp, ISP_LOGDEBUG0, sc1, sc4,
sdp->isp_req_ack_active_neg, sdp->isp_data_line_active_neg,
sdp->isp_data_dma_burst_enabl, sdp->isp_cmd_dma_burst_enable,
sdp->isp_selection_timeout, sdp->isp_max_queue_depth);
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
sdp->isp_devparam[tgt].dev_enable =
ISP_NVRAM_TGT_DEVICE_ENABLE(nvram_data, tgt);
sdp->isp_devparam[tgt].exc_throttle =
ISP_NVRAM_TGT_EXEC_THROTTLE(nvram_data, tgt);
sdp->isp_devparam[tgt].nvrm_offset =
ISP_NVRAM_TGT_SYNC_OFFSET(nvram_data, tgt);
sdp->isp_devparam[tgt].nvrm_period =
ISP_NVRAM_TGT_SYNC_PERIOD(nvram_data, tgt);
Redo how we manage SCSI device settings- have a 3rd flags (nvram) that records either what's in NVRAM or what the safe defaults would be if we lack NVRAM. Then we rename cur_XXXX to actv_XXXX (these are the currently active settings) and the dev_XXX settings to goal_XXXX (these are the settings which we want cur_XXXX to converge to).
2001-07-30 00:59:06 +00:00
/*
* We probably shouldn't lie about this, but it
* it makes it much safer if we limit NVRAM values
* to sanity.
*/
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
if (isp->isp_type < ISP_HA_SCSI_1040) {
/*
* If we're not ultra, we can't possibly
* be a shorter period than this.
*/
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
if (sdp->isp_devparam[tgt].nvrm_period < 0x19) {
sdp->isp_devparam[tgt].nvrm_period = 0x19;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
if (sdp->isp_devparam[tgt].nvrm_offset > 0xc) {
sdp->isp_devparam[tgt].nvrm_offset = 0x0c;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
} else {
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
if (sdp->isp_devparam[tgt].nvrm_offset > 0x8) {
sdp->isp_devparam[tgt].nvrm_offset = 0x8;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
}
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
sdp->isp_devparam[tgt].nvrm_flags = 0;
if (ISP_NVRAM_TGT_RENEG(nvram_data, tgt))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_RENEG;
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_ARQ;
if (ISP_NVRAM_TGT_TQING(nvram_data, tgt))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_TQING;
if (ISP_NVRAM_TGT_SYNC(nvram_data, tgt))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_SYNC;
if (ISP_NVRAM_TGT_WIDE(nvram_data, tgt))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_WIDE;
if (ISP_NVRAM_TGT_PARITY(nvram_data, tgt))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_PARITY;
if (ISP_NVRAM_TGT_DISC(nvram_data, tgt))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_DISC;
sdp->isp_devparam[tgt].actv_flags = 0; /* we don't know */
isp_prt(isp, ISP_LOGDEBUG0, sc2, sc4,
0, tgt, sdp->isp_devparam[tgt].nvrm_flags,
sdp->isp_devparam[tgt].nvrm_offset,
sdp->isp_devparam[tgt].nvrm_period);
sdp->isp_devparam[tgt].goal_offset =
sdp->isp_devparam[tgt].nvrm_offset;
sdp->isp_devparam[tgt].goal_period =
sdp->isp_devparam[tgt].nvrm_period;
sdp->isp_devparam[tgt].goal_flags =
sdp->isp_devparam[tgt].nvrm_flags;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_nvram_1080(ispsoftc_t *isp, int bus, uint8_t *nvram_data)
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
{
sdparam *sdp = (sdparam *) isp->isp_param;
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
int tgt;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp += bus;
More 2300 support prep- the Request/Response in/out pointers are part of the PCI block for the 2300- not software convention usage of the mailbox registers- so we macrosize in/out pointer usage. Only report that a LIP destroyed commands if it actually destroyed commands. Get the chan/tgt/lun order correct. Fix a longstanding stupid bug that caused us to try and issue a command with a tag on Channel B because we were checking the tagged capability for the target against Channel A. A firmware crash is now vectored out to platform specific code as an async event. Some minor formatting tweaks.
2001-07-04 18:42:41 +00:00
sdp->isp_fifo_threshold =
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
ISP1080_NVRAM_FIFO_THRESHOLD(nvram_data);
Remove the 'bogus registrant' hack for fabric searches. It really turns out that there's something of a hole in our new fabric name server stuff. We ask the name server for entities that have registered as a specific type. That type is FC-SCSI. If the entity hasn't performed a REGISTER FC4 TYPES, the fabric nameserver won't return it. This brings this driver to a bit of a fork in the road as to what the right thing to do is. For servicing the needs of accessing FC-SCSI devices, this method is fine, and to be preferred. It is extremely unlikely we're interested in fabric devices that *don't* register correctly. If I ever get around to adding an FC-IP stack, then asking for devices that have registers as FC-IP types is also the right thing to do. So- asking the fabric nameserver for a specific type is fine, *as long as you are only interested in specific types*. If, on the other hand, you want to create (as for management tool support) a picture of everything on the fabric, this is *not* so fine. There are a large class of FC-SCSI *initiators* who *don't* correctly register, so we never will *see* them. Is this a problem? Yes, but only a little one. If we want to do such management tool support, we should probably run a *different* fabric nameserver query algorithm. Better yet, we should talk to the management nameserver in Brocade switches instead of the standard FC-GS-2 fabric nameserver (which can be unwieldy). Other changes: if we've overrrides marked, don't set some default values from reading NVRAM. This allows us to override things like EXEC throttle without having to ignore NVRAM entirely. MFC after: 1 week
2002-07-08 17:33:37 +00:00
if ((isp->isp_confopts & ISP_CFG_OWNLOOPID) == 0)
sdp->isp_initiator_id =
ISP1080_NVRAM_INITIATOR_ID(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_bus_reset_delay =
ISP1080_NVRAM_BUS_RESET_DELAY(nvram_data, bus);
sdp->isp_retry_count =
ISP1080_NVRAM_BUS_RETRY_COUNT(nvram_data, bus);
sdp->isp_retry_delay =
ISP1080_NVRAM_BUS_RETRY_DELAY(nvram_data, bus);
sdp->isp_async_data_setup =
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
ISP1080_NVRAM_ASYNC_DATA_SETUP_TIME(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_req_ack_active_neg =
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
ISP1080_NVRAM_REQ_ACK_ACTIVE_NEGATION(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_data_line_active_neg =
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
ISP1080_NVRAM_DATA_LINE_ACTIVE_NEGATION(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_data_dma_burst_enabl =
ISP1080_NVRAM_BURST_ENABLE(nvram_data);
sdp->isp_cmd_dma_burst_enable =
ISP1080_NVRAM_BURST_ENABLE(nvram_data);
sdp->isp_selection_timeout =
ISP1080_NVRAM_SELECTION_TIMEOUT(nvram_data, bus);
sdp->isp_max_queue_depth =
ISP1080_NVRAM_MAX_QUEUE_DEPTH(nvram_data, bus);
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
isp_prt(isp, ISP_LOGDEBUG0, sc0, sc4,
bus, sdp->isp_fifo_threshold, sdp->isp_initiator_id,
sdp->isp_bus_reset_delay, sdp->isp_retry_count,
sdp->isp_retry_delay, sdp->isp_async_data_setup);
isp_prt(isp, ISP_LOGDEBUG0, sc1, sc4,
sdp->isp_req_ack_active_neg, sdp->isp_data_line_active_neg,
sdp->isp_data_dma_burst_enabl, sdp->isp_cmd_dma_burst_enable,
sdp->isp_selection_timeout, sdp->isp_max_queue_depth);
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
sdp->isp_devparam[tgt].dev_enable =
ISP1080_NVRAM_TGT_DEVICE_ENABLE(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].exc_throttle =
ISP1080_NVRAM_TGT_EXEC_THROTTLE(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].nvrm_offset =
ISP1080_NVRAM_TGT_SYNC_OFFSET(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].nvrm_period =
ISP1080_NVRAM_TGT_SYNC_PERIOD(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].nvrm_flags = 0;
if (ISP1080_NVRAM_TGT_RENEG(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_RENEG;
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_ARQ;
if (ISP1080_NVRAM_TGT_TQING(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_TQING;
if (ISP1080_NVRAM_TGT_SYNC(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_SYNC;
if (ISP1080_NVRAM_TGT_WIDE(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_WIDE;
if (ISP1080_NVRAM_TGT_PARITY(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_PARITY;
if (ISP1080_NVRAM_TGT_DISC(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_DISC;
sdp->isp_devparam[tgt].actv_flags = 0;
isp_prt(isp, ISP_LOGDEBUG0, sc2, sc4,
bus, tgt, sdp->isp_devparam[tgt].nvrm_flags,
sdp->isp_devparam[tgt].nvrm_offset,
sdp->isp_devparam[tgt].nvrm_period);
sdp->isp_devparam[tgt].goal_offset =
sdp->isp_devparam[tgt].nvrm_offset;
sdp->isp_devparam[tgt].goal_period =
sdp->isp_devparam[tgt].nvrm_period;
sdp->isp_devparam[tgt].goal_flags =
sdp->isp_devparam[tgt].nvrm_flags;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_nvram_12160(ispsoftc_t *isp, int bus, uint8_t *nvram_data)
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
{
sdparam *sdp = (sdparam *) isp->isp_param;
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
int tgt;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp += bus;
sdp->isp_fifo_threshold =
ISP12160_NVRAM_FIFO_THRESHOLD(nvram_data);
Remove the 'bogus registrant' hack for fabric searches. It really turns out that there's something of a hole in our new fabric name server stuff. We ask the name server for entities that have registered as a specific type. That type is FC-SCSI. If the entity hasn't performed a REGISTER FC4 TYPES, the fabric nameserver won't return it. This brings this driver to a bit of a fork in the road as to what the right thing to do is. For servicing the needs of accessing FC-SCSI devices, this method is fine, and to be preferred. It is extremely unlikely we're interested in fabric devices that *don't* register correctly. If I ever get around to adding an FC-IP stack, then asking for devices that have registers as FC-IP types is also the right thing to do. So- asking the fabric nameserver for a specific type is fine, *as long as you are only interested in specific types*. If, on the other hand, you want to create (as for management tool support) a picture of everything on the fabric, this is *not* so fine. There are a large class of FC-SCSI *initiators* who *don't* correctly register, so we never will *see* them. Is this a problem? Yes, but only a little one. If we want to do such management tool support, we should probably run a *different* fabric nameserver query algorithm. Better yet, we should talk to the management nameserver in Brocade switches instead of the standard FC-GS-2 fabric nameserver (which can be unwieldy). Other changes: if we've overrrides marked, don't set some default values from reading NVRAM. This allows us to override things like EXEC throttle without having to ignore NVRAM entirely. MFC after: 1 week
2002-07-08 17:33:37 +00:00
if ((isp->isp_confopts & ISP_CFG_OWNLOOPID) == 0)
sdp->isp_initiator_id =
ISP12160_NVRAM_INITIATOR_ID(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_bus_reset_delay =
ISP12160_NVRAM_BUS_RESET_DELAY(nvram_data, bus);
sdp->isp_retry_count =
ISP12160_NVRAM_BUS_RETRY_COUNT(nvram_data, bus);
sdp->isp_retry_delay =
ISP12160_NVRAM_BUS_RETRY_DELAY(nvram_data, bus);
sdp->isp_async_data_setup =
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
ISP12160_NVRAM_ASYNC_DATA_SETUP_TIME(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_req_ack_active_neg =
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
ISP12160_NVRAM_REQ_ACK_ACTIVE_NEGATION(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_data_line_active_neg =
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
ISP12160_NVRAM_DATA_LINE_ACTIVE_NEGATION(nvram_data, bus);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
sdp->isp_data_dma_burst_enabl =
ISP12160_NVRAM_BURST_ENABLE(nvram_data);
sdp->isp_cmd_dma_burst_enable =
ISP12160_NVRAM_BURST_ENABLE(nvram_data);
sdp->isp_selection_timeout =
ISP12160_NVRAM_SELECTION_TIMEOUT(nvram_data, bus);
sdp->isp_max_queue_depth =
ISP12160_NVRAM_MAX_QUEUE_DEPTH(nvram_data, bus);
Clean up some ways in which we set defaults for SCSI cards that do not have valid NVRAM. In particular, we were leaving a retry count set (to retry selection timeouts) when thats not really what we want. Do some constant string additions so that LOGDEBUG0 info is useful across all cards. MFC after: 2 weeks
2001-08-20 17:28:32 +00:00
isp_prt(isp, ISP_LOGDEBUG0, sc0, sc4,
bus, sdp->isp_fifo_threshold, sdp->isp_initiator_id,
sdp->isp_bus_reset_delay, sdp->isp_retry_count,
sdp->isp_retry_delay, sdp->isp_async_data_setup);
isp_prt(isp, ISP_LOGDEBUG0, sc1, sc4,
sdp->isp_req_ack_active_neg, sdp->isp_data_line_active_neg,
sdp->isp_data_dma_burst_enabl, sdp->isp_cmd_dma_burst_enable,
sdp->isp_selection_timeout, sdp->isp_max_queue_depth);
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
sdp->isp_devparam[tgt].dev_enable =
ISP12160_NVRAM_TGT_DEVICE_ENABLE(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].exc_throttle =
ISP12160_NVRAM_TGT_EXEC_THROTTLE(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].nvrm_offset =
ISP12160_NVRAM_TGT_SYNC_OFFSET(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].nvrm_period =
ISP12160_NVRAM_TGT_SYNC_PERIOD(nvram_data, tgt, bus);
sdp->isp_devparam[tgt].nvrm_flags = 0;
if (ISP12160_NVRAM_TGT_RENEG(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_RENEG;
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_ARQ;
if (ISP12160_NVRAM_TGT_TQING(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_TQING;
if (ISP12160_NVRAM_TGT_SYNC(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_SYNC;
if (ISP12160_NVRAM_TGT_WIDE(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_WIDE;
if (ISP12160_NVRAM_TGT_PARITY(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_PARITY;
if (ISP12160_NVRAM_TGT_DISC(nvram_data, tgt, bus))
sdp->isp_devparam[tgt].nvrm_flags |= DPARM_DISC;
sdp->isp_devparam[tgt].actv_flags = 0;
isp_prt(isp, ISP_LOGDEBUG0, sc2, sc4,
bus, tgt, sdp->isp_devparam[tgt].nvrm_flags,
sdp->isp_devparam[tgt].nvrm_offset,
sdp->isp_devparam[tgt].nvrm_period);
sdp->isp_devparam[tgt].goal_offset =
sdp->isp_devparam[tgt].nvrm_offset;
sdp->isp_devparam[tgt].goal_period =
sdp->isp_devparam[tgt].nvrm_period;
sdp->isp_devparam[tgt].goal_flags =
sdp->isp_devparam[tgt].nvrm_flags;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_parse_nvram_2100(ispsoftc_t *isp, uint8_t *nvram_data)
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
{
fcparam *fcp = (fcparam *) isp->isp_param;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint64_t wwn;
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
/*
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
* There is NVRAM storage for both Port and Node entities-
* but the Node entity appears to be unused on all the cards
* I can find. However, we should account for this being set
* at some point in the future.
*
* Qlogic WWNs have an NAA of 2, but usually nothing shows up in
* bits 48..60. In the case of the 2202, it appears that they do
* use bit 48 to distinguish between the two instances on the card.
* The 2204, which I've never seen, *probably* extends this method.
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
*/
Major whacking for core version 2.0. A major motivator for 2.0 and these changes is that there's now a Solaris port of this driver, so some things in the core version had to change (not much, but some). In order, from the top.....: A lot of error strings are gathered in one place at the head of the file. This caused me to rewrite them to look consistent (with respect to things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'. The major mailbox function, isp_mboxcmd, now takes a third argument, which is a mask that selectively says whether mailbox command failures will be logged. This will substantially reduce a lot of spurious noise from the driver. At the first run through isp_reset we used to try and get the current running firmware's revision by issuing a mailbox command. This would invariably fail on alpha's with anything but a Qlogic 1040 since SRM doesn't *start* the f/w on these cards. Instead, we now see whether we're sitting ROM state before trying to get a running BIOS loaded f/w version. All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to isp_prt. There are seperate print levels that can be independently set (see ispvar.h), which include debugging, etc. All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry indices, so we can make a 1024 entry index for them instead of the 256 entries we've had until now. A major change it to fix isp_fclink_test to actually only wait the delay of time specified in the microsecond argument being passed. The problem has always been that a call to isp_mboxcmd to get he current firmware state takes an unknown (sometimes long) amount of time- this is if the firmware is busy doing PLOGIs while we ask it what's up. So, up until now, the usdelay argument has been a joke. The net effect has been that if you boot without being plugged into a good loop or into a switch, you hang. Massively annonying, and hard to fix because the actual time delta was impossible to know from just guessing. Now, using the new GET_NANOTIME macros, a precise and measured amount of USEC_DELAY calls are done so that only the specified usecdelay is allowed to pass. This means that if the initial startup of the firmware if followed by a call from isp_freebsd.c:isp_attach to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000, no more than two seconds will actually elapse before we leave concluding that the cable is unhooked. Jeez. About time.... Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE macro to a call to the platform supplied isp_done (sane naming). Limit our size of request queue completions we'll look at at interrupt time. Since we've increased the size of the Request Queue (and the size of the Response Queue proportionally), let's not create an interrupt stack overflow by having to keep a max completion list (forw links are not an option because this is common code with some platforms that don't have link space in their XS_T structures). A limit of 32 is not unreasonable- I doubt there'd be even this many request queue completions at a time- remember, most boards now use fast posting for normal command completion instead of filling out response queue entries. In the isp_mboxcmd cleanup, also create an array of command names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8". Remove the isp_lostcmd function- it's been deprecated for a while. Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus register dump fucntion. Various other cleanups.
2000-08-01 06:51:05 +00:00
wwn = ISP2100_NVRAM_PORT_NAME(nvram_data);
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
if (wwn) {
isp_prt(isp, ISP_LOGCONFIG, "NVRAM Port WWN 0x%08x%08x",
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(uint32_t) (wwn >> 32), (uint32_t) (wwn & 0xffffffff));
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
if ((wwn >> 60) == 0) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
wwn |= (((uint64_t) 2)<< 60);
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
}
}
fcp->isp_portwwn = wwn;
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (IS_2200(isp) || IS_23XX(isp)) {
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
wwn = ISP2100_NVRAM_NODE_NAME(nvram_data);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if (wwn) {
isp_prt(isp, ISP_LOGCONFIG, "NVRAM Node WWN 0x%08x%08x",
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
(uint32_t) (wwn >> 32),
(uint32_t) (wwn & 0xffffffff));
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
if ((wwn >> 60) == 0) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
wwn |= (((uint64_t) 2)<< 60);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
}
Fix some breakage about how we build WWNs. Do some other fabric related changes: consider a new PDB entry different if Class 3 service parameter roles change (!!!). Do some checking as we're getting a port database that traps whether things change while we're doing so. Handle N-port and F-ports correctly. Fix the fabric login loop to retain a login/binding if things haven't changed (I mean, why logout a device only to log it back in). No longer accept, after fabric logins, garbage if we can't get a PDB entry that matches the device we've just logged into- if it doesn't, log it out as it is very unlikely to still be what we thought it was. Get rid of some of the debounce loops because we could get stuck there.
2000-05-09 01:14:43 +00:00
}
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
} else {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
wwn &= ~((uint64_t) 0xfff << 48);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
fcp->isp_nodewwn = wwn;
/*
* Make sure we have both Node and Port as non-zero values.
*/
if (fcp->isp_nodewwn != 0 && fcp->isp_portwwn == 0) {
fcp->isp_portwwn = fcp->isp_nodewwn;
} else if (fcp->isp_nodewwn == 0 && fcp->isp_portwwn != 0) {
fcp->isp_nodewwn = fcp->isp_portwwn;
}
/*
* Make the Node and Port values sane if they're NAA == 2.
* This means to clear bits 48..56 for the Node WWN and
* make sure that there's some non-zero value in 48..56
* for the Port WWN.
*/
if (fcp->isp_nodewwn && fcp->isp_portwwn) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if ((fcp->isp_nodewwn & (((uint64_t) 0xfff) << 48)) != 0 &&
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
(fcp->isp_nodewwn >> 60) == 2) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
fcp->isp_nodewwn &= ~((uint64_t) 0xfff << 48);
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if ((fcp->isp_portwwn & (((uint64_t) 0xfff) << 48)) == 0 &&
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
(fcp->isp_portwwn >> 60) == 2) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
fcp->isp_portwwn |= ((uint64_t) 1 << 56);
Redo how default Node and Port WWNs are determined (again!). This is so we don't stomp on the differences between ports for a Qlogic 2202.
2000-10-12 23:49:09 +00:00
}
}
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
fcp->isp_maxalloc =
ISP2100_NVRAM_MAXIOCBALLOCATION(nvram_data);
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if ((isp->isp_confopts & ISP_CFG_OWNFSZ) == 0)
fcp->isp_maxfrmlen =
ISP2100_NVRAM_MAXFRAMELENGTH(nvram_data);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
fcp->isp_retry_delay =
ISP2100_NVRAM_RETRY_DELAY(nvram_data);
fcp->isp_retry_count =
ISP2100_NVRAM_RETRY_COUNT(nvram_data);
If the HBA is already 'touched', still set maxluns. Othewise for CAM_QUIRK_HILUN devices we loop thru 32bits of lun. Oops. Switch to using USEC_DELAY rather than USEC_SLEEP at isp_reset time. Try to paper around a defect in clients that don't correctly registers themeselves with the fabric nameserver. Minor updates for Mirapoint support- they still use code that is not HANDLE_LOOPSTATE_IN_OUTER_LAYERS, and, surprise surprise, this old stuff had some bugs in it. Clean up some target mode stuff. MFC after: 1 week
2002-06-16 05:18:22 +00:00
if ((isp->isp_confopts & ISP_CFG_OWNLOOPID) == 0)
fcp->isp_loopid =
ISP2100_NVRAM_HARDLOOPID(nvram_data);
if ((isp->isp_confopts & ISP_CFG_OWNEXCTHROTTLE) == 0)
fcp->isp_execthrottle =
ISP2100_NVRAM_EXECUTION_THROTTLE(nvram_data);
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
fcp->isp_fwoptions = ISP2100_NVRAM_OPTIONS(nvram_data);
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp_prt(isp, ISP_LOGDEBUG0,
"NVRAM 0x%08x%08x 0x%08x%08x maxalloc %d maxframelen %d",
(uint32_t) (fcp->isp_nodewwn >> 32), (uint32_t) fcp->isp_nodewwn,
(uint32_t) (fcp->isp_portwwn >> 32), (uint32_t) fcp->isp_portwwn,
ISP2100_NVRAM_MAXIOCBALLOCATION(nvram_data),
ISP2100_NVRAM_MAXFRAMELENGTH(nvram_data));
isp_prt(isp, ISP_LOGDEBUG0,
"execthrottle %d fwoptions 0x%x hardloop %d tov %d",
ISP2100_NVRAM_EXECUTION_THROTTLE(nvram_data),
ISP2100_NVRAM_OPTIONS(nvram_data),
ISP2100_NVRAM_HARDLOOPID(nvram_data),
ISP2100_NVRAM_TOV(nvram_data));
fcp->isp_xfwoptions = ISP2100_XFW_OPTIONS(nvram_data);
fcp->isp_zfwoptions = ISP2100_ZFW_OPTIONS(nvram_data);
isp_prt(isp, ISP_LOGDEBUG0,
"xfwoptions 0x%x zfw options 0x%x",
ISP2100_XFW_OPTIONS(nvram_data), ISP2100_ZFW_OPTIONS(nvram_data));
Restructure nvram reading routine to split out to separate functions for 1020/1X80/12160/2X00- for readability. Add in 12160 (Ultra3) support- but not with PPR just yet. Fix and clarify fetching of return parameter for getting firmware rev which for the 2200 contains the connection topology (Private Loop (NL-port), N-port, FL-port, F-port). Synthesize the connection topology for the 2100 which can only be Private Loop or FL-port. Handle a couple of new async mailbox commands which signify connection in Point-to-Point mode (N-port or F-port) or indicate various toe stubbing getting to same. Approved: jkh@freebsd.org
2000-02-11 19:31:32 +00:00
}
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
#ifdef ISP_FW_CRASH_DUMP
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
static void isp2200_fw_dump(ispsoftc_t *);
static void isp2300_fw_dump(ispsoftc_t *);
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp2200_fw_dump(ispsoftc_t *isp)
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
{
More for f/w crash dumps (bug fixing and adding ioctl entry points and hints to enable for specific units) MFC after: 1 week
2002-02-18 00:00:34 +00:00
int i, j;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
mbreg_t mbs;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *ptr;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
ptr = FCPARAM(isp)->isp_dump_data;
if (ptr == NULL) {
isp_prt(isp, ISP_LOGERR,
"No place to dump RISC registers and SRAM");
return;
}
if (*ptr++) {
isp_prt(isp, ISP_LOGERR,
"dump area for RISC registers and SRAM already used");
return;
}
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
for (i = 0; i < 100; i++) {
USEC_DELAY(100);
if (ISP_READ(isp, HCCR) & HCCR_PAUSE) {
break;
}
}
if (ISP_READ(isp, HCCR) & HCCR_PAUSE) {
/*
* PBIU Registers
*/
for (i = 0; i < 8; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + (i << 1));
}
/*
* Mailbox Registers
*/
for (i = 0; i < 8; i++) {
*ptr++ = ISP_READ(isp, MBOX_BLOCK + (i << 1));
}
/*
* DMA Registers
*/
for (i = 0; i < 48; i++) {
*ptr++ = ISP_READ(isp, DMA_BLOCK + 0x20 + (i << 1));
}
/*
* RISC H/W Registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0);
for (i = 0; i < 16; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0xA0 + (i << 1));
}
/*
* RISC GP Registers
*/
for (j = 0; j < 8; j++) {
ISP_WRITE(isp, BIU_BLOCK + 0xA4, 0x2000 + (j << 8));
for (i = 0; i < 16; i++) {
*ptr++ =
ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
}
/*
* Frame Buffer Hardware Registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x10);
for (i = 0; i < 16; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
/*
* Fibre Protocol Module 0 Hardware Registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x20);
for (i = 0; i < 64; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
/*
* Fibre Protocol Module 1 Hardware Registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x30);
for (i = 0; i < 64; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
} else {
isp_prt(isp, ISP_LOGERR, "RISC Would Not Pause");
return;
}
isp_prt(isp, ISP_LOGALL,
"isp_fw_dump: RISC registers dumped successfully");
ISP_WRITE(isp, BIU2100_CSR, BIU2100_SOFT_RESET);
for (i = 0; i < 100; i++) {
USEC_DELAY(100);
if (ISP_READ(isp, OUTMAILBOX0) == 0) {
break;
}
}
if (ISP_READ(isp, OUTMAILBOX0) != 0) {
isp_prt(isp, ISP_LOGERR, "Board Would Not Reset");
return;
}
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
for (i = 0; i < 100; i++) {
USEC_DELAY(100);
if (ISP_READ(isp, HCCR) & HCCR_PAUSE) {
break;
}
}
if ((ISP_READ(isp, HCCR) & HCCR_PAUSE) == 0) {
isp_prt(isp, ISP_LOGERR, "RISC Would Not Pause After Reset");
return;
}
ISP_WRITE(isp, RISC_EMB, 0xf2);
ISP_WRITE(isp, HCCR, HCCR_CMD_RELEASE);
for (i = 0; i < 100; i++) {
USEC_DELAY(100);
if ((ISP_READ(isp, HCCR) & HCCR_PAUSE) == 0) {
break;
}
}
ENABLE_INTS(isp);
mbs.param[0] = MBOX_READ_RAM_WORD;
mbs.param[1] = 0x1000;
isp->isp_mbxworkp = (void *) ptr;
isp->isp_mbxwrk0 = 0xefff; /* continuation count */
isp->isp_mbxwrk1 = 0x1001; /* next SRAM address */
isp_control(isp, ISPCTL_RUN_MBOXCMD, &mbs);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGWARN,
"RAM DUMP FAILED @ WORD %x", isp->isp_mbxwrk1);
return;
}
ptr = isp->isp_mbxworkp; /* finish fetch of final word */
*ptr++ = isp->isp_mboxtmp[2];
Revert previous commit. Violates Maintainer (O'Brien knows how to reach me directly), but more importantly, breaks compiles on non-FreeBSD platforms.
2003-08-25 17:58:23 +00:00
isp_prt(isp, ISP_LOGALL, "isp_fw_dump: SRAM dumped successfully");
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
FCPARAM(isp)->isp_dump_data[0] = isp->isp_type; /* now used */
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
(void) isp_async(isp, ISPASYNC_FW_DUMPED, 0);
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
}
static void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp2300_fw_dump(ispsoftc_t *isp)
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
{
More for f/w crash dumps (bug fixing and adding ioctl entry points and hints to enable for specific units) MFC after: 1 week
2002-02-18 00:00:34 +00:00
int i, j;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
mbreg_t mbs;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t *ptr;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
MEMZERO(&mbs, sizeof (mbs));
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
ptr = FCPARAM(isp)->isp_dump_data;
if (ptr == NULL) {
isp_prt(isp, ISP_LOGERR,
"No place to dump RISC registers and SRAM");
return;
}
if (*ptr++) {
isp_prt(isp, ISP_LOGERR,
"dump area for RISC registers and SRAM already used");
return;
}
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
for (i = 0; i < 100; i++) {
USEC_DELAY(100);
if (ISP_READ(isp, HCCR) & HCCR_PAUSE) {
break;
}
}
if (ISP_READ(isp, HCCR) & HCCR_PAUSE) {
/*
* PBIU registers
*/
for (i = 0; i < 8; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + (i << 1));
}
/*
* ReqQ-RspQ-Risc2Host Status registers
*/
for (i = 0; i < 8; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x10 + (i << 1));
}
/*
* Mailbox Registers
*/
for (i = 0; i < 32; i++) {
*ptr++ =
ISP_READ(isp, PCI_MBOX_REGS2300_OFF + (i << 1));
}
/*
* Auto Request Response DMA registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x40);
for (i = 0; i < 32; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
/*
* DMA registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x50);
for (i = 0; i < 48; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
/*
* RISC hardware registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0);
for (i = 0; i < 16; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0xA0 + (i << 1));
}
/*
* RISC GP? registers
*/
for (j = 0; j < 8; j++) {
ISP_WRITE(isp, BIU_BLOCK + 0xA4, 0x2000 + (j << 9));
for (i = 0; i < 16; i++) {
*ptr++ =
ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
}
/*
* frame buffer hardware registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x10);
for (i = 0; i < 64; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
/*
* FPM B0 hardware registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x20);
for (i = 0; i < 64; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
/*
* FPM B1 hardware registers
*/
ISP_WRITE(isp, BIU2100_CSR, 0x30);
for (i = 0; i < 64; i++) {
*ptr++ = ISP_READ(isp, BIU_BLOCK + 0x80 + (i << 1));
}
} else {
isp_prt(isp, ISP_LOGERR, "RISC Would Not Pause");
return;
}
isp_prt(isp, ISP_LOGALL,
"isp_fw_dump: RISC registers dumped successfully");
ISP_WRITE(isp, BIU2100_CSR, BIU2100_SOFT_RESET);
for (i = 0; i < 100; i++) {
USEC_DELAY(100);
if (ISP_READ(isp, OUTMAILBOX0) == 0) {
break;
}
}
if (ISP_READ(isp, OUTMAILBOX0) != 0) {
isp_prt(isp, ISP_LOGERR, "Board Would Not Reset");
return;
}
ENABLE_INTS(isp);
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
mbs.param[0] = MBOX_READ_RAM_WORD;
mbs.param[1] = 0x800;
isp->isp_mbxworkp = (void *) ptr;
isp->isp_mbxwrk0 = 0xf7ff; /* continuation count */
isp->isp_mbxwrk1 = 0x801; /* next SRAM address */
isp_control(isp, ISPCTL_RUN_MBOXCMD, &mbs);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGWARN,
"RAM DUMP FAILED @ WORD %x", isp->isp_mbxwrk1);
return;
}
ptr = isp->isp_mbxworkp; /* finish fetch of final word */
*ptr++ = isp->isp_mboxtmp[2];
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
MEMZERO(&mbs, sizeof (mbs));
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
mbs.param[0] = MBOX_READ_RAM_WORD_EXTENDED;
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
mbs.param[8] = 1;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
isp->isp_mbxworkp = (void *) ptr;
isp->isp_mbxwrk0 = 0xffff; /* continuation count */
isp->isp_mbxwrk1 = 0x1; /* next SRAM address */
Do various fixes to support firmware loading for the 2322 (and by extension, the 2422). One peculiar thing I've found with the 2322 is that if you don't force it to do Hard LoopID acquisition, the firmware crashes. This took a while to figure out. While we're at it, fix various bugs having to do with NVRAM reading and option setting with respect to pieces of NVRAM.
2006-07-03 08:24:09 +00:00
isp->isp_mbxwrk8 = 0x1;
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
isp_control(isp, ISPCTL_RUN_MBOXCMD, &mbs);
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
isp_prt(isp, ISP_LOGWARN,
"RAM DUMP FAILED @ WORD %x", 0x10000 + isp->isp_mbxwrk1);
return;
}
ptr = isp->isp_mbxworkp; /* finish final word */
*ptr++ = mbs.param[2];
Revert previous commit. Violates Maintainer (O'Brien knows how to reach me directly), but more importantly, breaks compiles on non-FreeBSD platforms.
2003-08-25 17:58:23 +00:00
isp_prt(isp, ISP_LOGALL, "isp_fw_dump: SRAM dumped successfully");
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
FCPARAM(isp)->isp_dump_data[0] = isp->isp_type; /* now used */
If we're using ancient (pre 1.17.0) 2100 f/w (for the cards that cannot load f/w images > 0x7fff words), set ISP_FW_ATTR_SCCLUN. We explicitly don't believe we can find attributes if f/w is < 1.17.0, so we have to set SCCLUN for the 1.15.37 f/w we're using manually- otherwise every target will replicate itself across all 16 supported luns for non-SCCLUN f/w. Correctly set things up for 23XX and either fast posting or ZIO. The 23XX, it turns out, does not support RIO. If you put a non-zero value in xfwoptions, this will disable fast posting. If you put ICBXOPT_ZIO in xfwoptions, then the 23XX will do interrupt delays but post to the response queue- apparently QLogic *now* believes that reading multiple handles from registers is less of a win than writing (and delaying) multiple 64 byte responses to the response queue. At the end of taking a a good f/w crash dump, send the ISPASYNC_FW_DUMPED event to the outer layers (who can then do things like wake a user daemon to *fetch* the crash image, etc.).
2002-08-17 17:29:15 +00:00
(void) isp_async(isp, ISPASYNC_FW_DUMPED, 0);
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
}
void
Some more gratuitous format and name changes. Pull in some target mode changes from a private branch. Pull in some more RELENG_4 compilation changes. A lot of lines changed, but not much content change yet.
2006-04-21 18:30:01 +00:00
isp_fw_dump(ispsoftc_t *isp)
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
{
if (IS_2200(isp))
isp2200_fw_dump(isp);
More for f/w crash dumps (bug fixing and adding ioctl entry points and hints to enable for specific units) MFC after: 1 week
2002-02-18 00:00:34 +00:00
else if (IS_23XX(isp))
Support for f/w crash dumps (2200 && 23XX). If you want QLogic to look at a potential f/w problem for FC cards, you really have to provide them info in the format they expect. This involves dumping a lot of hardware registers (> 300 16 bit registers) and a lot of SRAM (> 128KB minimum). Thus all of this code is #ifdef protected which will become an option so that the memory allocation of where to dump the crash image is pretty expensive. It's worth it if you have a reproducible problem because they have some tools that can tell them, given the f/w version, the precise state of everything. MFC after: 1 week
2002-02-17 06:38:22 +00:00
isp2300_fw_dump(isp);
}
#endif
Reference in New Issue Copy Permalink