1999-08-28 01:08:13 +00:00
|
|
|
/* $FreeBSD$ */
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
1998-09-15 08:42:56 +00:00
|
|
|
* Machine and OS Independent (well, as best as possible)
|
1998-04-22 17:54:58 +00:00
|
|
|
* code for the Qlogic ISP SCSI adapters.
|
|
|
|
|
*
|
2001-03-02 06:28:55 +00:00
|
|
|
* Copyright (c) 1997, 1998, 1999, 2000, 2001 by Matthew Jacob
|
2000-09-21 20:16:04 +00:00
|
|
|
* Feral Software
|
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.
|
2000-09-21 20:16:04 +00:00
|
|
|
* 2. The name of the author may not be used to endorse or promote products
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* 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__
|
|
|
|
|
#include <dev/isp/isp_freebsd.h>
|
|
|
|
|
#endif
|
1999-03-17 05:04:39 +00:00
|
|
|
#ifdef __OpenBSD__
|
|
|
|
|
#include <dev/ic/isp_openbsd.h>
|
|
|
|
|
#endif
|
1998-04-22 17:54:58 +00:00
|
|
|
#ifdef __linux__
|
1998-12-28 19:22:27 +00:00
|
|
|
#include "isp_linux.h"
|
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
|
1998-04-22 17:54:58 +00:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* General defines
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#define MBOX_DELAY_COUNT 1000000 / 100
|
|
|
|
|
|
|
|
|
|
/*
|
1998-09-15 08:42:56 +00:00
|
|
|
* Local static data
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
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)";
|
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";
|
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)";
|
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";
|
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)";
|
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)";
|
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";
|
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)";
|
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)";
|
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";
|
2000-12-29 19:12:44 +00:00
|
|
|
static const char notresp[] =
|
2000-12-02 18:08:35 +00:00
|
|
|
"Not RESPONSE in RESPONSE Queue (type 0x%x) @ idx %d (next %d) nlooked %d";
|
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";
|
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";
|
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";
|
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";
|
2000-12-29 19:12:44 +00:00
|
|
|
static const char topology[] =
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus
register dump fucntion.
Various other cleanups.
2000-08-01 06:51:05 +00:00
|
|
|
"Loop ID %d, AL_PA 0x%x, Port ID 0x%x, Loop State 0x%x, Topology '%s'";
|
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";
|
2000-12-29 19:12:44 +00:00
|
|
|
static const char finmsg[] =
|
2001-02-11 03:44:43 +00:00
|
|
|
"(%d.%d.%d): FIN dl%d resid %d STS 0x%x SKEY %c XS_ERR=0x%x";
|
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";
|
2002-05-01 21:58:36 +00:00
|
|
|
static const char bun[] =
|
|
|
|
|
"bad underrun for %d.%d (count %d, resid %d, status %s)";
|
2001-08-20 17:28:32 +00:00
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
|
|
|
|
* Local function prototypes.
|
|
|
|
|
*/
|
2002-01-03 20:43:22 +00:00
|
|
|
static int isp_parse_async(struct ispsoftc *, u_int16_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 int isp_handle_other_response(struct ispsoftc *, int, isphdr_t *,
|
|
|
|
|
u_int16_t *);
|
|
|
|
|
static void
|
|
|
|
|
isp_parse_status(struct ispsoftc *, ispstatusreq_t *, XS_T *);
|
|
|
|
|
static void isp_fastpost_complete(struct ispsoftc *, u_int16_t);
|
2002-02-04 21:04:25 +00:00
|
|
|
static int isp_mbox_continue(struct ispsoftc *);
|
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 isp_scsi_init(struct ispsoftc *);
|
|
|
|
|
static void isp_scsi_channel_init(struct ispsoftc *, int);
|
|
|
|
|
static void isp_fibre_init(struct ispsoftc *);
|
|
|
|
|
static void isp_mark_getpdb_all(struct ispsoftc *);
|
|
|
|
|
static int isp_getmap(struct ispsoftc *, fcpos_map_t *);
|
|
|
|
|
static int isp_getpdb(struct ispsoftc *, int, isp_pdb_t *);
|
|
|
|
|
static u_int64_t isp_get_portname(struct ispsoftc *, int, int);
|
|
|
|
|
static int isp_fclink_test(struct ispsoftc *, int);
|
|
|
|
|
static char *isp2100_fw_statename(int);
|
|
|
|
|
static int isp_pdb_sync(struct ispsoftc *);
|
|
|
|
|
static int isp_scan_loop(struct ispsoftc *);
|
2002-04-04 23:46:01 +00:00
|
|
|
static int isp_fabric_mbox_cmd(struct ispsoftc *, mbreg_t *);
|
|
|
|
|
static int isp_scan_fabric(struct ispsoftc *, int);
|
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 isp_register_fc4_type(struct ispsoftc *);
|
|
|
|
|
static void isp_fw_state(struct ispsoftc *);
|
2002-02-21 01:56:08 +00:00
|
|
|
static void isp_mboxcmd_qnw(struct ispsoftc *, mbreg_t *, int);
|
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 isp_mboxcmd(struct ispsoftc *, mbreg_t *, int);
|
|
|
|
|
|
|
|
|
|
static void isp_update(struct ispsoftc *);
|
|
|
|
|
static void isp_update_bus(struct ispsoftc *, int);
|
|
|
|
|
static void isp_setdfltparm(struct ispsoftc *, int);
|
|
|
|
|
static int isp_read_nvram(struct ispsoftc *);
|
|
|
|
|
static void isp_rdnvram_word(struct ispsoftc *, int, u_int16_t *);
|
|
|
|
|
static void isp_parse_nvram_1020(struct ispsoftc *, u_int8_t *);
|
|
|
|
|
static void isp_parse_nvram_1080(struct ispsoftc *, int, u_int8_t *);
|
|
|
|
|
static void isp_parse_nvram_12160(struct ispsoftc *, int, u_int8_t *);
|
|
|
|
|
static void isp_parse_nvram_2100(struct ispsoftc *, u_int8_t *);
|
2000-02-11 19:31:32 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
|
|
|
|
* Reset Hardware.
|
1998-09-15 08:42:56 +00:00
|
|
|
*
|
1999-10-30 19:32:44 +00:00
|
|
|
* Hit the chip over the head, download new f/w if available and set it running.
|
1998-09-15 08:42:56 +00:00
|
|
|
*
|
|
|
|
|
* Locking done elsewhere.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
2001-08-31 21:39:04 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_reset(struct ispsoftc *isp)
|
1998-04-22 17:54:58 +00:00
|
|
|
{
|
|
|
|
|
mbreg_t mbs;
|
2001-08-31 21:39:04 +00:00
|
|
|
u_int16_t code_org;
|
2002-06-16 05:18:22 +00:00
|
|
|
int loops, i, dodnld = 1;
|
2001-08-31 21:39:04 +00:00
|
|
|
char *btype = "????";
|
1998-04-22 17:54:58 +00:00
|
|
|
|
|
|
|
|
isp->isp_state = ISP_NILSTATE;
|
|
|
|
|
|
|
|
|
|
/*
|
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
|
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
|
|
|
*
|
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.
|
1999-02-09 01:07:06 +00:00
|
|
|
*/
|
|
|
|
|
|
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.
|
|
|
|
|
*/
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp->isp_touched == 0) {
|
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 "
|
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.
|
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);
|
|
|
|
|
mbs.param[0] = MBOX_ABOUT_FIRMWARE;
|
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];
|
|
|
|
|
}
|
1999-01-10 02:55:10 +00:00
|
|
|
}
|
1999-11-21 03:18:22 +00:00
|
|
|
isp->isp_touched = 1;
|
1999-01-10 02:55:10 +00:00
|
|
|
}
|
|
|
|
|
|
1999-04-04 02:28:29 +00:00
|
|
|
DISABLE_INTS(isp);
|
2001-09-03 03:09:48 +00:00
|
|
|
|
2001-08-31 21:39:04 +00:00
|
|
|
/*
|
|
|
|
|
* Set up default request/response queue in-pointer/out-pointer
|
|
|
|
|
* register indices.
|
|
|
|
|
*/
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
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;
|
|
|
|
|
} else {
|
|
|
|
|
isp->isp_rqstinrp = INMAILBOX4;
|
|
|
|
|
isp->isp_rqstoutrp = OUTMAILBOX4;
|
|
|
|
|
isp->isp_respinrp = OUTMAILBOX5;
|
|
|
|
|
isp->isp_respoutrp = INMAILBOX5;
|
|
|
|
|
}
|
1999-04-04 02:28:29 +00:00
|
|
|
|
1999-01-30 07:29:00 +00:00
|
|
|
/*
|
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).
|
1999-01-30 07:29:00 +00:00
|
|
|
*/
|
|
|
|
|
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
|
|
|
|
|
|
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:
|
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:
|
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:
|
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;
|
2002-02-04 21:04:25 +00:00
|
|
|
case ISP_HA_FC_2312:
|
|
|
|
|
btype = "2312";
|
|
|
|
|
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;
|
|
|
|
|
}
|
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);
|
1999-12-16 05:42:02 +00:00
|
|
|
} else if (IS_1240(isp)) {
|
|
|
|
|
sdparam *sdp = isp->isp_param;
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1240";
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 60;
|
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)) {
|
2000-12-29 19:12:44 +00:00
|
|
|
static const char m[] = "bus %d is in %s Mode";
|
1999-03-25 22:52:45 +00:00
|
|
|
u_int16_t l;
|
1999-03-17 05:04:39 +00:00
|
|
|
sdparam *sdp = isp->isp_param;
|
1999-12-16 05:42:02 +00:00
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 100;
|
1999-12-16 05:42:02 +00:00
|
|
|
|
2000-02-11 19:31:32 +00:00
|
|
|
if (IS_1280(isp))
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1280";
|
2000-02-11 19:31:32 +00:00
|
|
|
else if (IS_1080(isp))
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1080";
|
2002-10-11 17:28:01 +00:00
|
|
|
else if (IS_10160(isp))
|
|
|
|
|
btype = "10160";
|
2000-02-11 19:31:32 +00:00
|
|
|
else if (IS_12160(isp))
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "12160";
|
2000-02-11 19:31:32 +00:00
|
|
|
else
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "<UNKLVD>";
|
2000-02-11 19:31:32 +00:00
|
|
|
|
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");
|
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");
|
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");
|
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);
|
1999-03-25 22:52:45 +00:00
|
|
|
break;
|
|
|
|
|
}
|
1999-12-16 05:42:02 +00:00
|
|
|
|
2000-02-11 19:31:32 +00:00
|
|
|
if (IS_DUALBUS(isp)) {
|
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");
|
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");
|
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");
|
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);
|
1999-12-16 05:42:02 +00:00
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
} else {
|
1998-09-15 08:42:56 +00:00
|
|
|
sdparam *sdp = isp->isp_param;
|
1999-01-10 02:55:10 +00:00
|
|
|
i = ISP_READ(isp, BIU_CONF0) & BIU_CONF0_HW_MASK;
|
|
|
|
|
switch (i) {
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
/* FALLTHROUGH */
|
|
|
|
|
case 1:
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1020";
|
1998-04-22 17:54:58 +00:00
|
|
|
isp->isp_type = ISP_HA_SCSI_1020;
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 40;
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 2:
|
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.
|
|
|
|
|
*/
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1020A";
|
1998-09-15 08:42:56 +00:00
|
|
|
isp->isp_type = ISP_HA_SCSI_1020A;
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 40;
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 3:
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1040";
|
1998-09-15 08:42:56 +00:00
|
|
|
isp->isp_type = ISP_HA_SCSI_1040;
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 60;
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
case 4:
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1040A";
|
1998-04-22 17:54:58 +00:00
|
|
|
isp->isp_type = ISP_HA_SCSI_1040A;
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 60;
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 5:
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1040B";
|
1998-04-22 17:54:58 +00:00
|
|
|
isp->isp_type = ISP_HA_SCSI_1040B;
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 60;
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
2001-07-04 18:42:41 +00:00
|
|
|
case 6:
|
2001-08-31 21:39:04 +00:00
|
|
|
btype = "1040C";
|
1999-04-04 02:28:29 +00:00
|
|
|
isp->isp_type = ISP_HA_SCSI_1040C;
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 60;
|
2001-07-04 18:42:41 +00:00
|
|
|
break;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
1999-01-30 07:29:00 +00:00
|
|
|
* Now, while we're at it, gather info about ultra
|
|
|
|
|
* and/or differential mode.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
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");
|
1999-01-30 07:29:00 +00:00
|
|
|
sdp->isp_diffmode = 1;
|
1999-01-10 02:55:10 +00:00
|
|
|
} else {
|
1999-01-30 07:29:00 +00:00
|
|
|
sdp->isp_diffmode = 0;
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
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");
|
1999-01-30 07:29:00 +00:00
|
|
|
sdp->isp_ultramode = 1;
|
1999-03-17 05:04:39 +00:00
|
|
|
/*
|
|
|
|
|
* If we're in Ultra Mode, we have to be 60Mhz clock-
|
|
|
|
|
* even for the SBus version.
|
|
|
|
|
*/
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_clock = 60;
|
1999-03-25 22:52:45 +00:00
|
|
|
} else {
|
1999-01-30 07:29:00 +00:00
|
|
|
sdp->isp_ultramode = 0;
|
1999-03-17 05:04:39 +00:00
|
|
|
/*
|
|
|
|
|
* Clock is known. Gronk.
|
|
|
|
|
*/
|
1999-01-30 07:29:00 +00:00
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
|
|
|
|
* Machine dependent clock (if set) overrides
|
|
|
|
|
* our generic determinations.
|
|
|
|
|
*/
|
|
|
|
|
if (isp->isp_mdvec->dv_clock) {
|
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;
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
2000-12-02 18:08:35 +00:00
|
|
|
/*
|
|
|
|
|
* Clear instrumentation
|
|
|
|
|
*/
|
|
|
|
|
isp->isp_intcnt = isp->isp_intbogus = 0;
|
|
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
|
|
|
|
* Do MD specific pre initialization
|
|
|
|
|
*/
|
|
|
|
|
ISP_RESET0(isp);
|
1998-09-15 08:42:56 +00:00
|
|
|
|
1999-01-30 07:29:00 +00:00
|
|
|
again:
|
1999-01-10 02:55:10 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
|
|
|
|
* Hit the chip over the head with hammer,
|
|
|
|
|
* and give the ISP a chance to recover.
|
|
|
|
|
*/
|
|
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
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);
|
1999-03-17 05:04:39 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
|
|
|
|
* Clear data && control DMA engines.
|
|
|
|
|
*/
|
|
|
|
|
ISP_WRITE(isp, CDMA_CONTROL,
|
1999-03-25 22:52:45 +00:00
|
|
|
DMA_CNTRL_CLEAR_CHAN | DMA_CNTRL_RESET_INT);
|
1998-04-22 17:54:58 +00:00
|
|
|
ISP_WRITE(isp, DDMA_CONTROL,
|
1999-03-25 22:52:45 +00:00
|
|
|
DMA_CNTRL_CLEAR_CHAN | DMA_CNTRL_RESET_INT);
|
1999-03-17 05:04:39 +00:00
|
|
|
|
|
|
|
|
|
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);
|
1999-01-30 07:29:00 +00:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Clear data && control DMA engines.
|
|
|
|
|
*/
|
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 (;;) {
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
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);
|
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");
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
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.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
1999-01-30 07:29:00 +00:00
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1998-04-22 17:54:58 +00:00
|
|
|
ISP_WRITE(isp, BIU_CONF1, 0);
|
|
|
|
|
} else {
|
|
|
|
|
ISP_WRITE(isp, BIU2100_CSR, 0);
|
|
|
|
|
}
|
|
|
|
|
|
1999-01-30 07:29:00 +00:00
|
|
|
/*
|
|
|
|
|
* Reset RISC Processor
|
|
|
|
|
*/
|
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);
|
2001-01-15 18:33:08 +00:00
|
|
|
/* Clear semaphore register (just to be sure) */
|
|
|
|
|
ISP_WRITE(isp, BIU_SEMA, 0);
|
1998-04-22 17:54:58 +00:00
|
|
|
|
1999-01-10 02:55:10 +00:00
|
|
|
/*
|
1999-01-30 07:29:00 +00:00
|
|
|
* Establish some initial burst rate stuff.
|
1999-01-10 02:55:10 +00:00
|
|
|
* (only for the 1XX0 boards). This really should
|
|
|
|
|
* be done later after fetching from NVRAM.
|
|
|
|
|
*/
|
1999-03-17 05:04:39 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1999-01-30 07:29:00 +00:00
|
|
|
u_int16_t tmp = isp->isp_mdvec->dv_conf1;
|
1999-01-10 02:55:10 +00:00
|
|
|
/*
|
|
|
|
|
* Busted FIFO. Turn off all but burst enables.
|
|
|
|
|
*/
|
|
|
|
|
if (isp->isp_type == ISP_HA_SCSI_1040A) {
|
1999-01-30 07:29:00 +00:00
|
|
|
tmp &= BIU_BURST_ENABLE;
|
1999-01-10 02:55:10 +00:00
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
ISP_SETBITS(isp, BIU_CONF1, tmp);
|
|
|
|
|
if (tmp & BIU_BURST_ENABLE) {
|
1998-04-22 17:54:58 +00:00
|
|
|
ISP_SETBITS(isp, CDMA_CONF, DMA_ENABLE_BURST);
|
|
|
|
|
ISP_SETBITS(isp, DDMA_CONF, DMA_ENABLE_BURST);
|
|
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
#ifdef PTI_CARDS
|
|
|
|
|
if (((sdparam *) isp->isp_param)->isp_ultramode) {
|
1999-03-25 22:52:45 +00:00
|
|
|
while (ISP_READ(isp, RISC_MTR) != 0x1313) {
|
1999-01-30 07:29:00 +00:00
|
|
|
ISP_WRITE(isp, RISC_MTR, 0x1313);
|
|
|
|
|
ISP_WRITE(isp, HCCR, HCCR_CMD_STEP);
|
|
|
|
|
}
|
1999-03-25 22:52:45 +00:00
|
|
|
} else {
|
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);
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_2200(isp) || IS_23XX(isp)) {
|
2001-01-15 18:33:08 +00:00
|
|
|
ISP_WRITE(isp, HCCR, HCCR_2X00_DISABLE_PARITY_PAUSE);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
|
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);
|
|
|
|
|
|
1999-01-30 07:29:00 +00:00
|
|
|
/*
|
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).
|
1999-01-30 07:29:00 +00:00
|
|
|
*/
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
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;
|
|
|
|
|
}
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
|
|
|
|
* Do some sanity checking.
|
|
|
|
|
*/
|
|
|
|
|
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);
|
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)) {
|
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);
|
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]);
|
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.
|
|
|
|
|
*/
|
1999-11-01 04:39:52 +00:00
|
|
|
if ((isp->isp_mdvec->dv_ispfw == NULL) ||
|
1998-04-22 17:54:58 +00:00
|
|
|
(isp->isp_confopts & ISP_CFG_NORELOAD)) {
|
|
|
|
|
dodnld = 0;
|
|
|
|
|
}
|
|
|
|
|
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp))
|
2001-08-31 21:39:04 +00:00
|
|
|
code_org = ISP_CODE_ORG_2300;
|
|
|
|
|
else
|
|
|
|
|
code_org = ISP_CODE_ORG;
|
|
|
|
|
|
1999-11-01 04:39:52 +00:00
|
|
|
if (dodnld) {
|
2002-02-04 21:04:25 +00:00
|
|
|
isp->isp_mbxworkp = (void *) &isp->isp_mdvec->dv_ispfw[1];
|
|
|
|
|
isp->isp_mbxwrk0 = isp->isp_mdvec->dv_ispfw[3] - 1;
|
|
|
|
|
isp->isp_mbxwrk1 = code_org + 1;
|
|
|
|
|
mbs.param[0] = MBOX_WRITE_RAM_WORD;
|
|
|
|
|
mbs.param[1] = code_org;
|
|
|
|
|
mbs.param[2] = isp->isp_mdvec->dv_ispfw[0];
|
|
|
|
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
/*
|
|
|
|
|
* Verify that it downloaded correctly.
|
|
|
|
|
*/
|
|
|
|
|
mbs.param[0] = MBOX_VERIFY_CHECKSUM;
|
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);
|
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");
|
1999-01-30 07:29:00 +00:00
|
|
|
return;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
2000-08-27 23:38:44 +00:00
|
|
|
isp->isp_loaded_fw = 1;
|
1998-04-22 17:54:58 +00:00
|
|
|
} else {
|
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");
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
2001-08-31 21:39:04 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
mbs.param[0] = MBOX_EXEC_FIRMWARE;
|
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);
|
2002-06-16 05:18:22 +00:00
|
|
|
/*
|
|
|
|
|
* Give it a chance to start.
|
|
|
|
|
*/
|
|
|
|
|
USEC_DELAY(500);
|
1998-04-22 17:54:58 +00:00
|
|
|
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
1998-09-15 08:42:56 +00:00
|
|
|
* Set CLOCK RATE, but only if asked to.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
1999-05-11 05:06:55 +00:00
|
|
|
if (isp->isp_clock) {
|
1998-04-22 17:54:58 +00:00
|
|
|
mbs.param[0] = MBOX_SET_CLOCK_RATE;
|
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 */
|
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
|
|
|
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
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) {
|
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];
|
|
|
|
|
}
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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,
|
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",
|
2001-09-03 03:09:48 +00:00
|
|
|
isp->isp_fwrev[0], isp->isp_fwrev[1], isp->isp_fwrev[2]);
|
|
|
|
|
|
1999-03-25 22:52:45 +00:00
|
|
|
if (IS_FC(isp)) {
|
2001-09-03 03:09:48 +00:00
|
|
|
/*
|
|
|
|
|
* We do not believe firmware attributes for 2100 code less
|
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.
|
2001-09-03 03:09:48 +00:00
|
|
|
*/
|
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
|
2001-09-03 03:09:48 +00:00
|
|
|
FCPARAM(isp)->isp_fwattr = 0;
|
2002-08-17 17:29:15 +00:00
|
|
|
#endif
|
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]);
|
|
|
|
|
}
|
1999-02-09 01:07:06 +00:00
|
|
|
if (ISP_READ(isp, BIU2100_CSR) & BIU2100_PCI64) {
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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,
|
|
|
|
|
"Installed in 64-Bit PCI slot");
|
1999-02-09 01:07:06 +00:00
|
|
|
}
|
|
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
|
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",
|
1999-04-04 02:28:29 +00:00
|
|
|
isp->isp_romfw_rev[0], isp->isp_romfw_rev[1],
|
|
|
|
|
isp->isp_romfw_rev[2]);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
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);
|
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]);
|
1998-04-22 17:54:58 +00:00
|
|
|
isp_fw_state(isp);
|
1999-10-17 18:58:22 +00:00
|
|
|
|
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");
|
1999-05-11 05:06:55 +00:00
|
|
|
return;
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
isp->isp_state = ISP_RESETSTATE;
|
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
|
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.
|
2000-06-18 04:56:17 +00:00
|
|
|
*
|
|
|
|
|
* Expanded lun firmware gives you 32 luns for SCSI cards and
|
2002-04-16 19:55:35 +00:00
|
|
|
* 16384 luns for Fibre Channel cards.
|
2000-06-18 04:56:17 +00:00
|
|
|
*
|
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.
|
|
|
|
|
*
|
2000-06-18 04:56:17 +00:00
|
|
|
* Because the lun is in a a different position in the Request Queue
|
|
|
|
|
* 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.
|
|
|
|
|
*
|
|
|
|
|
* Note that we only do this once (the first time thru isp_reset)
|
|
|
|
|
* because we may be called again after firmware has been loaded once
|
|
|
|
|
* and released.
|
|
|
|
|
*/
|
2002-06-16 05:18:22 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
|
|
|
|
if (dodnld) {
|
|
|
|
|
if (IS_ULTRA2(isp) || IS_ULTRA3(isp)) {
|
2000-06-18 04:56:17 +00:00
|
|
|
isp->isp_maxluns = 32;
|
|
|
|
|
} else {
|
2001-09-03 03:09:48 +00:00
|
|
|
isp->isp_maxluns = 8;
|
2000-06-18 04:56:17 +00:00
|
|
|
}
|
|
|
|
|
} else {
|
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;
|
2000-06-18 04:56:17 +00:00
|
|
|
}
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
1999-01-30 07:29:00 +00:00
|
|
|
* Initialize Parameters of Hardware to a known state.
|
1998-09-15 08:42:56 +00:00
|
|
|
*
|
|
|
|
|
* Locks are held before coming here.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
1998-09-15 08:42:56 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_init(struct ispsoftc *isp)
|
1998-04-22 17:54:58 +00:00
|
|
|
{
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Must do this first to get defaults established.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
1999-05-11 05:06:55 +00:00
|
|
|
isp_setdfltparm(isp, 0);
|
1999-12-16 05:42:02 +00:00
|
|
|
if (IS_DUALBUS(isp)) {
|
1999-05-11 05:06:55 +00:00
|
|
|
isp_setdfltparm(isp, 1);
|
1999-03-17 05:04:39 +00:00
|
|
|
}
|
2001-01-15 18:33:08 +00:00
|
|
|
if (IS_FC(isp)) {
|
|
|
|
|
isp_fibre_init(isp);
|
|
|
|
|
} else {
|
|
|
|
|
isp_scsi_init(isp);
|
1999-05-11 05:06:55 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_scsi_init(struct ispsoftc *isp)
|
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;
|
1999-12-16 05:42:02 +00:00
|
|
|
if (IS_DUALBUS(isp)) {
|
1999-05-11 05:06:55 +00:00
|
|
|
sdp_chan1++;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1999-05-11 05:06:55 +00:00
|
|
|
|
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;
|
|
|
|
|
}
|
|
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
/* First do overall per-card settings. */
|
1998-04-22 17:54:58 +00:00
|
|
|
|
1999-02-09 01:07:06 +00:00
|
|
|
/*
|
|
|
|
|
* If we have fast memory timing enabled, turn it on.
|
|
|
|
|
*/
|
2000-08-27 23:38:44 +00:00
|
|
|
if (sdp_chan0->isp_fast_mttr) {
|
1999-02-09 01:07:06 +00:00
|
|
|
ISP_WRITE(isp, RISC_MTR, 0x1313);
|
|
|
|
|
}
|
|
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Set Retry Delay and Count.
|
|
|
|
|
* You set both channels at the same time.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Set ASYNC DATA SETUP time. This is very important.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Set ACTIVE Negation State.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
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);
|
|
|
|
|
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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);
|
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)",
|
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
|
|
|
|
|
*/
|
|
|
|
|
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);
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Set selection timeout.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
/* now do per-channel settings */
|
|
|
|
|
isp_scsi_channel_init(isp, 0);
|
1999-12-16 05:42:02 +00:00
|
|
|
if (IS_DUALBUS(isp))
|
1999-05-11 05:06:55 +00:00
|
|
|
isp_scsi_channel_init(isp, 1);
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Now enable request/response queues
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
|
|
|
|
|
2002-09-23 04:59:42 +00:00
|
|
|
if (IS_ULTRA2(isp) || IS_1240(isp)) {
|
|
|
|
|
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];
|
1999-05-11 05:06:55 +00:00
|
|
|
|
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 {
|
|
|
|
|
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];
|
|
|
|
|
|
|
|
|
|
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];
|
1999-05-11 05:06:55 +00:00
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
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)
|
|
|
|
|
*
|
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.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
|
|
|
|
|
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;
|
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
|
|
|
|
|
#ifndef ISP_NO_FASTPOST
|
2000-02-11 19:31:32 +00:00
|
|
|
if (IS_ULTRA2(isp) || IS_1240(isp))
|
|
|
|
|
mbs.param[1] |= FW_FEATURE_FAST_POST;
|
2002-02-04 21:04:25 +00:00
|
|
|
#endif
|
|
|
|
|
#endif
|
1999-12-20 01:34:01 +00:00
|
|
|
if (mbs.param[1] != 0) {
|
|
|
|
|
u_int16_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);
|
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
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_scsi_channel_init(struct ispsoftc *isp, int channel)
|
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.
|
|
|
|
|
*/
|
|
|
|
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
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);
|
2000-09-21 17:06:45 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
2001-07-30 00:59:06 +00:00
|
|
|
* Set current per-target parameters to an initial safe minimum.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
|
|
|
|
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
|
2000-06-18 04:56:17 +00:00
|
|
|
int lun;
|
1999-02-09 01:07:06 +00:00
|
|
|
u_int16_t sdf;
|
1998-09-15 08:42:56 +00:00
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
if (sdp->isp_devparam[tgt].dev_enable == 0) {
|
1998-04-22 17:54:58 +00:00
|
|
|
continue;
|
1999-05-11 05:06:55 +00:00
|
|
|
}
|
2001-04-04 21:42:59 +00:00
|
|
|
#ifndef ISP_TARGET_MODE
|
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;
|
2001-04-04 21:42:59 +00:00
|
|
|
}
|
|
|
|
|
#else
|
1999-05-11 05:06:55 +00:00
|
|
|
/*
|
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.
|
1999-05-11 05:06:55 +00:00
|
|
|
*/
|
2001-07-30 00:59:06 +00:00
|
|
|
sdp->isp_devparam[tgt].goal_flags = sdf = DPARM_DEFAULT;
|
2001-04-04 21:42:59 +00:00
|
|
|
#endif
|
1999-02-09 01:07:06 +00:00
|
|
|
mbs.param[0] = MBOX_SET_TARGET_PARAMS;
|
2001-04-04 21:42:59 +00:00
|
|
|
mbs.param[1] = (channel << 15) | (tgt << 8);
|
1999-02-09 01:07:06 +00:00
|
|
|
mbs.param[2] = sdf;
|
2000-08-27 23:38:44 +00:00
|
|
|
if ((sdf & DPARM_SYNC) == 0) {
|
|
|
|
|
mbs.param[3] = 0;
|
|
|
|
|
} else {
|
|
|
|
|
mbs.param[3] =
|
2001-07-30 00:59:06 +00:00
|
|
|
(sdp->isp_devparam[tgt].goal_offset << 8) |
|
|
|
|
|
(sdp->isp_devparam[tgt].goal_period);
|
2000-08-27 23:38:44 +00:00
|
|
|
}
|
2001-04-04 21:42:59 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG0,
|
2001-07-30 00:59:06 +00:00
|
|
|
"Initial Settings bus%d tgt%d flags 0x%x off 0x%x per 0x%x",
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
1999-02-09 01:07:06 +00:00
|
|
|
sdf = DPARM_SAFE_DFLT;
|
1998-04-22 17:54:58 +00:00
|
|
|
mbs.param[0] = MBOX_SET_TARGET_PARAMS;
|
1999-05-11 05:06:55 +00:00
|
|
|
mbs.param[1] = (tgt << 8) | (channel << 15);
|
1999-02-09 01:07:06 +00:00
|
|
|
mbs.param[2] = sdf;
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
1998-09-15 08:42:56 +00:00
|
|
|
continue;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
}
|
1999-12-20 01:34:01 +00:00
|
|
|
|
1999-03-26 00:33:13 +00:00
|
|
|
/*
|
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.
|
|
|
|
|
*
|
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.
|
|
|
|
|
*/
|
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++) {
|
1998-04-22 17:54:58 +00:00
|
|
|
mbs.param[0] = MBOX_SET_DEV_QUEUE_PARAMS;
|
1999-05-11 05:06:55 +00:00
|
|
|
mbs.param[1] = (channel << 15) | (tgt << 8) | lun;
|
1998-04-22 17:54:58 +00:00
|
|
|
mbs.param[2] = sdp->isp_max_queue_depth;
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
}
|
1999-01-30 07:29:00 +00:00
|
|
|
}
|
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;
|
|
|
|
|
}
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
|
|
|
|
* Fibre Channel specific initialization.
|
|
|
|
|
*
|
|
|
|
|
* Locks are held before coming here.
|
|
|
|
|
*/
|
1998-04-22 17:54:58 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_fibre_init(struct ispsoftc *isp)
|
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;
|
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;
|
2000-10-12 23:49:09 +00:00
|
|
|
u_int64_t nwwn, pwwn;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
|
|
|
|
fcp = isp->isp_param;
|
|
|
|
|
|
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;
|
|
|
|
|
}
|
|
|
|
|
|
2002-02-21 01:56:08 +00:00
|
|
|
loopid = fcp->isp_loopid;
|
1999-01-10 02:55:10 +00:00
|
|
|
MEMZERO(icbp, sizeof (*icbp));
|
1998-09-15 08:42:56 +00:00
|
|
|
icbp->icb_version = ICB_VERSION1;
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
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;
|
|
|
|
|
}
|
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
|
|
|
/*
|
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
|
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
|
|
|
*/
|
2002-08-17 17:29:15 +00:00
|
|
|
if (!ISP_FW_NEWER_THAN(isp, 1, 17, 0)) {
|
1999-08-16 19:59:55 +00:00
|
|
|
fcp->isp_fwoptions |= ICBOPT_FULL_LOGIN;
|
|
|
|
|
}
|
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
|
|
|
|
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;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Propagate all of this into the ICB structure.
|
|
|
|
|
*/
|
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);
|
1999-04-04 02:28:29 +00:00
|
|
|
icbp->icb_maxfrmlen = ICB_DFLT_FRMLEN;
|
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);
|
1999-04-04 02:28:29 +00:00
|
|
|
icbp->icb_maxalloc = 16;
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
icbp->icb_execthrottle = fcp->isp_execthrottle;
|
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;
|
1999-04-04 02:28:29 +00:00
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
icbp->icb_retry_delay = fcp->isp_retry_delay;
|
|
|
|
|
icbp->icb_retry_count = fcp->isp_retry_count;
|
1998-12-05 01:46:40 +00:00
|
|
|
icbp->icb_hardaddr = loopid;
|
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.
|
2001-08-31 21:39:04 +00:00
|
|
|
*
|
|
|
|
|
* NB: for the 2300, ICBOPT_EXTENDED is required.
|
2000-08-27 23:38:44 +00:00
|
|
|
*/
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_2200(isp) || IS_23XX(isp)) {
|
2000-02-11 19:31:32 +00:00
|
|
|
icbp->icb_fwoptions |= ICBOPT_EXTENDED;
|
|
|
|
|
/*
|
|
|
|
|
* Prefer or force Point-To-Point instead Loop?
|
|
|
|
|
*/
|
2000-12-02 18:08:35 +00:00
|
|
|
switch(isp->isp_confopts & ISP_CFG_PORT_PREF) {
|
|
|
|
|
case ISP_CFG_NPORT:
|
2001-10-01 03:45:54 +00:00
|
|
|
icbp->icb_xfwoptions |= ICBXOPT_PTP_2_LOOP;
|
2000-12-02 18:08:35 +00:00
|
|
|
break;
|
|
|
|
|
case ISP_CFG_NPORT_ONLY:
|
2001-10-01 03:45:54 +00:00
|
|
|
icbp->icb_xfwoptions |= ICBXOPT_PTP_ONLY;
|
2000-12-02 18:08:35 +00:00
|
|
|
break;
|
|
|
|
|
case ISP_CFG_LPORT_ONLY:
|
2001-10-01 03:45:54 +00:00
|
|
|
icbp->icb_xfwoptions |= ICBXOPT_LOOP_ONLY;
|
2000-12-02 18:08:35 +00:00
|
|
|
break;
|
|
|
|
|
default:
|
2001-10-01 03:45:54 +00:00
|
|
|
icbp->icb_xfwoptions |= ICBXOPT_LOOP_2_PTP;
|
2000-12-02 18:08:35 +00:00
|
|
|
break;
|
|
|
|
|
}
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
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).
|
|
|
|
|
*
|
|
|
|
|
* If we set ZIO, it will disable fast posting,
|
|
|
|
|
* so we don't need to clear it in fwoptions.
|
|
|
|
|
*/
|
|
|
|
|
icbp->icb_xfwoptions |= ICBXOPT_ZIO;
|
|
|
|
|
|
2001-10-01 03:45:54 +00:00
|
|
|
if (isp->isp_confopts & ISP_CFG_ONEGB) {
|
2001-10-06 20:41:18 +00:00
|
|
|
icbp->icb_zfwoptions |= ICBZOPT_RATE_ONEGB;
|
2001-10-01 03:45:54 +00:00
|
|
|
} else if (isp->isp_confopts & ISP_CFG_TWOGB) {
|
2001-10-06 20:41:18 +00:00
|
|
|
icbp->icb_zfwoptions |= ICBZOPT_RATE_TWOGB;
|
2001-10-01 03:45:54 +00:00
|
|
|
} else {
|
2001-10-06 20:41:18 +00:00
|
|
|
icbp->icb_zfwoptions |= ICBZOPT_RATE_AUTO;
|
2001-10-01 03:45:54 +00:00
|
|
|
}
|
2001-08-31 21:39:04 +00:00
|
|
|
}
|
2002-02-17 06:38:22 +00:00
|
|
|
}
|
|
|
|
|
|
2002-02-04 21:04:25 +00:00
|
|
|
#ifndef ISP_NO_RIO_FC
|
2002-03-07 17:32:45 +00:00
|
|
|
/*
|
|
|
|
|
* RIO seems to be enabled in 2100s for fw >= 1.17.0.
|
|
|
|
|
*
|
|
|
|
|
* I've had some questionable problems with RIO on 2200.
|
|
|
|
|
* More specifically, on a 2204 I had problems with RIO
|
|
|
|
|
* on a Linux system where I was dropping commands right
|
|
|
|
|
* and left. It's not clear to me what the actual problem
|
2002-08-17 17:29:15 +00:00
|
|
|
* was.
|
2002-03-07 17:32:45 +00:00
|
|
|
*
|
2002-08-17 17:29:15 +00:00
|
|
|
* 23XX Cards do not support RIO. Instead they support ZIO.
|
2002-03-07 17:32:45 +00:00
|
|
|
*/
|
2002-08-17 17:29:15 +00:00
|
|
|
#if 0
|
|
|
|
|
if (!IS_23XX(isp) && ISP_FW_NEWER_THAN(isp, 1, 17, 0)) {
|
2002-02-17 06:38:22 +00:00
|
|
|
icbp->icb_xfwoptions |= ICBXOPT_RIO_16BIT;
|
|
|
|
|
icbp->icb_racctimer = 4;
|
|
|
|
|
icbp->icb_idelaytimer = 8;
|
2001-08-31 21:39:04 +00:00
|
|
|
}
|
2002-08-17 17:29:15 +00:00
|
|
|
#endif
|
2002-02-17 06:38:22 +00:00
|
|
|
#endif
|
2001-08-31 21:39:04 +00:00
|
|
|
|
2002-08-17 17:29:15 +00:00
|
|
|
/*
|
|
|
|
|
* For 22XX > 2.1.26 && 23XX, set someoptions.
|
|
|
|
|
* XXX: Probably okay for newer 2100 f/w too.
|
|
|
|
|
*/
|
|
|
|
|
if (ISP_FW_NEWER_THAN(isp, 2, 26, 0)) {
|
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)
|
|
|
|
|
*/
|
|
|
|
|
mbs.param[0] = MBOX_SET_FIRMWARE_OPTIONS;
|
|
|
|
|
mbs.param[1] = 0xb;
|
|
|
|
|
mbs.param[2] = 0;
|
|
|
|
|
mbs.param[3] = 0;
|
|
|
|
|
isp_mboxcmd(isp, &mbs, MBLOGALL);
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
2001-08-31 21:39:04 +00:00
|
|
|
icbp->icb_logintime = 30; /* 30 second login timeout */
|
|
|
|
|
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
2001-08-31 21:39:04 +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);
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
|
2000-10-12 23:49:09 +00:00
|
|
|
nwwn = ISP_NODEWWN(isp);
|
|
|
|
|
pwwn = ISP_PORTWWN(isp);
|
|
|
|
|
if (nwwn && pwwn) {
|
2001-02-23 05:35:50 +00:00
|
|
|
icbp->icb_fwoptions |= ICBOPT_BOTH_WWNS;
|
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",
|
|
|
|
|
((u_int32_t) (nwwn >> 32)),
|
|
|
|
|
((u_int32_t) (nwwn & 0xffffffff)),
|
|
|
|
|
((u_int32_t) (pwwn >> 32)),
|
|
|
|
|
((u_int32_t) (pwwn & 0xffffffff)));
|
1999-04-04 02:28:29 +00:00
|
|
|
} else {
|
2000-10-12 23:49:09 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG1, "Not using any WWNs");
|
2001-02-23 05:35:50 +00:00
|
|
|
icbp->icb_fwoptions &= ~(ICBOPT_BOTH_WWNS|ICBOPT_FULL_LOGIN);
|
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);
|
|
|
|
|
icbp->icb_rsltqlen = RESULT_QUEUE_LEN(isp);
|
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);
|
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);
|
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);
|
2001-01-15 18:33:08 +00:00
|
|
|
|
1999-08-16 19:59:55 +00:00
|
|
|
/*
|
2001-01-15 18:33:08 +00:00
|
|
|
* Init the firmware
|
1999-08-16 19:59:55 +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
|
|
|
mbs.param[0] = MBOX_INIT_FIRMWARE;
|
|
|
|
|
mbs.param[1] = 0;
|
2001-08-31 21:39:04 +00:00
|
|
|
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
|
|
|
|
|
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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[4] = 0;
|
|
|
|
|
mbs.param[5] = 0;
|
2001-08-31 21:39:04 +00:00
|
|
|
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
|
|
|
|
|
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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);
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
isp->isp_reqidx = isp->isp_reqodx = 0;
|
1998-04-22 17:54:58 +00:00
|
|
|
isp->isp_residx = 0;
|
1999-03-17 05:04:39 +00:00
|
|
|
isp->isp_sendmarker = 1;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
|
|
|
|
/*
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
2001-02-11 03:44:43 +00:00
|
|
|
static int
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_getmap(struct ispsoftc *isp, fcpos_map_t *map)
|
2001-02-11 03:44:43 +00:00
|
|
|
{
|
|
|
|
|
fcparam *fcp = (fcparam *) isp->isp_param;
|
|
|
|
|
mbreg_t mbs;
|
|
|
|
|
|
|
|
|
|
mbs.param[0] = MBOX_GET_FC_AL_POSITION_MAP;
|
|
|
|
|
mbs.param[1] = 0;
|
2001-08-31 21:39:04 +00:00
|
|
|
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
|
|
|
|
|
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
|
2001-02-11 03:44:43 +00:00
|
|
|
/*
|
|
|
|
|
* Unneeded. For the 2100, except for initializing f/w, registers
|
|
|
|
|
* 4/5 have to not be written to.
|
|
|
|
|
* mbs.param[4] = 0;
|
|
|
|
|
* mbs.param[5] = 0;
|
|
|
|
|
*
|
|
|
|
|
*/
|
|
|
|
|
mbs.param[6] = 0;
|
|
|
|
|
mbs.param[7] = 0;
|
2002-02-04 21:04:25 +00:00
|
|
|
FC_SCRATCH_ACQUIRE(isp);
|
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;
|
2002-02-04 21:04:25 +00:00
|
|
|
FC_SCRATCH_RELEASE(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
return (0);
|
|
|
|
|
}
|
2002-02-04 21:04:25 +00:00
|
|
|
FC_SCRATCH_RELEASE(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
return (-1);
|
|
|
|
|
}
|
|
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_mark_getpdb_all(struct ispsoftc *isp)
|
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++) {
|
2001-01-15 18:33:08 +00:00
|
|
|
fcp->portdb[i].valid = fcp->portdb[i].fabric_dev = 0;
|
1999-03-17 05:04:39 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_getpdb(struct ispsoftc *isp, int id, isp_pdb_t *pdbp)
|
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
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
mbs.param[0] = MBOX_GET_PORT_DB;
|
|
|
|
|
mbs.param[1] = id << 8;
|
2001-08-31 21:39:04 +00:00
|
|
|
mbs.param[2] = DMA_WD1(fcp->isp_scdma);
|
|
|
|
|
mbs.param[3] = DMA_WD0(fcp->isp_scdma);
|
1999-03-25 22:52:45 +00:00
|
|
|
/*
|
|
|
|
|
* Unneeded. For the 2100, except for initializing f/w, registers
|
|
|
|
|
* 4/5 have to not be written to.
|
|
|
|
|
* mbs.param[4] = 0;
|
|
|
|
|
* mbs.param[5] = 0;
|
|
|
|
|
*
|
|
|
|
|
*/
|
2001-08-31 21:39:04 +00:00
|
|
|
mbs.param[6] = DMA_WD3(fcp->isp_scdma);
|
|
|
|
|
mbs.param[7] = DMA_WD2(fcp->isp_scdma);
|
2002-02-04 21:04:25 +00:00
|
|
|
FC_SCRATCH_ACQUIRE(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
|
|
|
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);
|
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);
|
1999-03-17 05:04:39 +00:00
|
|
|
}
|
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);
|
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
|
|
|
static u_int64_t
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_get_portname(struct ispsoftc *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
|
|
|
{
|
|
|
|
|
u_int64_t wwn = 0;
|
|
|
|
|
mbreg_t mbs;
|
|
|
|
|
|
|
|
|
|
mbs.param[0] = MBOX_GET_PORT_NAME;
|
|
|
|
|
mbs.param[1] = loopid << 8;
|
|
|
|
|
if (nodename)
|
|
|
|
|
mbs.param[1] |= 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 & ~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 =
|
|
|
|
|
(((u_int64_t)(mbs.param[2] & 0xff)) << 56) |
|
|
|
|
|
(((u_int64_t)(mbs.param[2] >> 8)) << 48) |
|
|
|
|
|
(((u_int64_t)(mbs.param[3] & 0xff)) << 40) |
|
|
|
|
|
(((u_int64_t)(mbs.param[3] >> 8)) << 32) |
|
|
|
|
|
(((u_int64_t)(mbs.param[6] & 0xff)) << 24) |
|
|
|
|
|
(((u_int64_t)(mbs.param[6] >> 8)) << 16) |
|
|
|
|
|
(((u_int64_t)(mbs.param[7] & 0xff)) << 8) |
|
|
|
|
|
(((u_int64_t)(mbs.param[7] >> 8)));
|
|
|
|
|
}
|
|
|
|
|
return (wwn);
|
|
|
|
|
}
|
|
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
/*
|
|
|
|
|
* Make sure we have good FC link and know our Loop ID.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
static int
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_fclink_test(struct ispsoftc *isp, int usdelay)
|
1999-03-17 05:04:39 +00:00
|
|
|
{
|
1999-10-17 18:58:22 +00:00
|
|
|
static char *toponames[] = {
|
|
|
|
|
"Private Loop",
|
|
|
|
|
"FL Port",
|
|
|
|
|
"N-Port to N-Port",
|
2000-05-09 01:14:43 +00:00
|
|
|
"F Port",
|
|
|
|
|
"F Port (no FLOGI_ACC response)"
|
1999-10-17 18:58:22 +00:00
|
|
|
};
|
1999-03-17 05:04:39 +00:00
|
|
|
mbreg_t mbs;
|
2001-02-11 03:44:43 +00:00
|
|
|
int count, check_for_fabric;
|
1999-03-17 05:04:39 +00:00
|
|
|
u_int8_t lwfs;
|
|
|
|
|
fcparam *fcp;
|
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;
|
2001-02-11 03:44:43 +00:00
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
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
|
|
|
/*
|
|
|
|
|
* XXX: Here is where we would start a 'loop dead' timeout
|
|
|
|
|
*/
|
|
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
/*
|
|
|
|
|
* Wait up to N microseconds for F/W to go to a ready state.
|
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) {
|
|
|
|
|
u_int64_t enano;
|
|
|
|
|
u_int32_t wrk;
|
|
|
|
|
NANOTIME_T hra, hrb;
|
|
|
|
|
|
|
|
|
|
GET_NANOTIME(&hra);
|
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),
|
1998-12-05 01:33:57 +00:00
|
|
|
isp2100_fw_statename((int)fcp->isp_fwstate));
|
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);
|
|
|
|
|
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG1,
|
|
|
|
|
"usec%d: 0x%lx->0x%lx enano 0x%x%08x",
|
2000-12-29 19:12:44 +00:00
|
|
|
count, (long) GET_NANOSEC(&hra), (long) GET_NANOSEC(&hrb),
|
2001-03-14 04:11:56 +00:00
|
|
|
(u_int32_t)(enano >> 32), (u_int32_t)(enano & 0xffffffff));
|
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.
|
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
|
|
|
|
|
* invoking u_int64_t math support functions for some
|
|
|
|
|
* platforms where linkage is a problem.
|
|
|
|
|
*/
|
|
|
|
|
if (enano < (1000 * 1000)) {
|
|
|
|
|
count += 1000;
|
|
|
|
|
enano = (1000 * 1000) - enano;
|
|
|
|
|
while (enano > (u_int64_t) 4000000000U) {
|
2000-12-02 18:08:35 +00:00
|
|
|
USEC_SLEEP(isp, 4000000);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove isp_dumpregs- the ISP_DUMPREGS goes to the specific bus
register dump fucntion.
Various other cleanups.
2000-08-01 06:51:05 +00:00
|
|
|
enano -= (u_int64_t) 4000000000U;
|
|
|
|
|
}
|
|
|
|
|
wrk = enano;
|
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 {
|
|
|
|
|
while (enano > (u_int64_t) 4000000000U) {
|
|
|
|
|
count += 4000000;
|
|
|
|
|
enano -= (u_int64_t) 4000000000U;
|
|
|
|
|
}
|
|
|
|
|
wrk = enano;
|
|
|
|
|
count += (wrk / 1000);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
1999-03-17 05:04:39 +00:00
|
|
|
* If we haven't gone to 'ready' state, return.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
1999-03-17 05:04:39 +00:00
|
|
|
if (fcp->isp_fwstate != FW_READY) {
|
|
|
|
|
return (-1);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1999-03-25 22:52:45 +00:00
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
/*
|
|
|
|
|
* Get our Loop ID (if possible). We really need to have it.
|
|
|
|
|
*/
|
|
|
|
|
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);
|
1999-03-17 05:04:39 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
return (-1);
|
|
|
|
|
}
|
|
|
|
|
fcp->isp_loopid = mbs.param[1];
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_2200(isp) || IS_23XX(isp)) {
|
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;
|
2000-02-11 19:31:32 +00:00
|
|
|
} else {
|
2000-04-21 02:04:34 +00:00
|
|
|
fcp->isp_topo = TOPO_NL_PORT;
|
1999-10-17 18:58:22 +00:00
|
|
|
}
|
2001-02-11 03:44:43 +00:00
|
|
|
fcp->isp_portid = fcp->isp_alpa = mbs.param[2] & 0xff;
|
2000-05-09 01:14:43 +00:00
|
|
|
|
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
|
|
|
|
|
* (in case we're connected to an server doing aliasing).
|
|
|
|
|
*/
|
1999-08-16 19:59:55 +00:00
|
|
|
fcp->isp_onfabric = 0;
|
2001-02-11 03:44:43 +00:00
|
|
|
|
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) {
|
2001-02-11 03:44:43 +00:00
|
|
|
check_for_fabric = 1;
|
2002-08-17 17:29:15 +00:00
|
|
|
} else
|
2001-02-11 03:44:43 +00:00
|
|
|
check_for_fabric = 0;
|
|
|
|
|
|
|
|
|
|
if (check_for_fabric && isp_getpdb(isp, FL_PORT_ID, &pdb) == 0) {
|
2001-01-15 18:33:08 +00:00
|
|
|
int loopid = FL_PORT_ID;
|
2000-05-09 01:14:43 +00:00
|
|
|
if (IS_2100(isp)) {
|
2000-04-21 02:04:34 +00:00
|
|
|
fcp->isp_topo = TOPO_FL_PORT;
|
2000-05-09 01:14:43 +00:00
|
|
|
}
|
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
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
2001-01-15 18:33:08 +00:00
|
|
|
lp = &fcp->portdb[loopid];
|
2000-05-09 01:14:43 +00:00
|
|
|
lp->node_wwn =
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[7]));
|
|
|
|
|
lp->port_wwn =
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[7]));
|
|
|
|
|
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;
|
2001-02-11 03:44:43 +00:00
|
|
|
fcp->isp_onfabric = 1;
|
|
|
|
|
(void) isp_async(isp, ISPASYNC_PROMENADE, &loopid);
|
2001-01-09 02:46:23 +00:00
|
|
|
isp_register_fc4_type(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
} else {
|
|
|
|
|
not_on_fabric:
|
2000-05-09 01:14:43 +00:00
|
|
|
fcp->isp_onfabric = 0;
|
|
|
|
|
fcp->portdb[FL_PORT_ID].valid = 0;
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-01 03:45:54 +00:00
|
|
|
fcp->isp_gbspeed = 1;
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
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;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2002-02-04 21:04:25 +00:00
|
|
|
isp_prt(isp, ISP_LOGCONFIG, topology, fcp->isp_loopid, fcp->isp_alpa,
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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]);
|
2000-05-09 01:14:43 +00:00
|
|
|
|
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);
|
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
|
|
|
}
|
|
|
|
|
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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 *
|
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
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_pdb_sync(struct ispsoftc *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
|
|
|
{
|
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;
|
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
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
if (fcp->isp_loopstate != LOOP_PDB_RCVD &&
|
2001-07-04 18:42:41 +00:00
|
|
|
fcp->isp_loopstate != LOOP_FSCAN_DONE &&
|
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
|
|
|
/*
|
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
|
|
|
*/
|
1999-08-16 19:59:55 +00:00
|
|
|
|
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);
|
2000-05-09 01:14:43 +00:00
|
|
|
}
|
|
|
|
|
|
2001-01-15 18:33:08 +00:00
|
|
|
/*
|
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.
|
2001-01-15 18:33:08 +00:00
|
|
|
*/
|
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;
|
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
|
|
|
/*
|
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
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
for (lp = &fcp->portdb[base]; lp < &fcp->portdb[lim]; lp++) {
|
|
|
|
|
u_int32_t portid;
|
|
|
|
|
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;
|
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;
|
|
|
|
|
}
|
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
|
|
|
/*
|
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
|
|
|
*/
|
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;
|
|
|
|
|
}
|
1999-08-16 19:59:55 +00:00
|
|
|
|
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
|
|
|
/*
|
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
|
|
|
*/
|
2002-04-16 19:55:35 +00:00
|
|
|
if (lp->loggedin && lp->force_logout == 0 &&
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_getpdb(isp, lp->loopid, &pdb) == 0) {
|
|
|
|
|
int nrole;
|
|
|
|
|
u_int64_t nwwnn, nwwpn;
|
2001-07-04 18:42:41 +00:00
|
|
|
nwwnn =
|
2001-02-11 03:44:43 +00:00
|
|
|
(((u_int64_t)pdb.pdb_nodename[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[7]));
|
2001-07-04 18:42:41 +00:00
|
|
|
nwwpn =
|
2001-02-11 03:44:43 +00:00
|
|
|
(((u_int64_t)pdb.pdb_portname[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[7]));
|
|
|
|
|
nrole = (pdb.pdb_prli_svc3 & SVC3_ROLE_MASK) >>
|
|
|
|
|
SVC3_ROLE_SHIFT;
|
|
|
|
|
if (pdb.pdb_loopid == lp->loopid && lp->portid ==
|
|
|
|
|
(u_int32_t) BITS2WORD(pdb.pdb_portid_bits) &&
|
|
|
|
|
nwwnn == lp->node_wwn && nwwpn == lp->port_wwn &&
|
2001-06-05 17:11:06 +00:00
|
|
|
lp->roles == nrole && lp->force_logout == 0) {
|
2001-02-11 03:44:43 +00:00
|
|
|
lp->loggedin = lp->valid = 1;
|
2002-02-04 21:04:25 +00:00
|
|
|
isp_prt(isp, ISP_LOGCONFIG, lretained,
|
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;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
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
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
if (lp->loggedin) {
|
2002-04-16 19:55:35 +00:00
|
|
|
if (lp->force_logout ||
|
|
|
|
|
isp_getpdb(isp, lp->loopid, &pdb) == 0) {
|
2001-02-11 03:44:43 +00:00
|
|
|
mbs.param[0] = MBOX_FABRIC_LOGOUT;
|
|
|
|
|
mbs.param[1] = lp->loopid << 8;
|
|
|
|
|
mbs.param[2] = 0;
|
|
|
|
|
mbs.param[3] = 0;
|
|
|
|
|
isp_mboxcmd(isp, &mbs, MBLOGNONE);
|
|
|
|
|
isp_prt(isp, ISP_LOGINFO, plogout,
|
|
|
|
|
(int) (lp - fcp->portdb), lp->loopid,
|
|
|
|
|
lp->portid);
|
|
|
|
|
}
|
2002-04-16 19:55:35 +00:00
|
|
|
lp->force_logout = lp->loggedin = 0;
|
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
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
loopid = lp - fcp->portdb;
|
|
|
|
|
lp->loopid = FL_PORT_ID;
|
|
|
|
|
do {
|
|
|
|
|
mbs.param[0] = MBOX_FABRIC_LOGIN;
|
|
|
|
|
mbs.param[1] = loopid << 8;
|
|
|
|
|
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
|
|
|
/*
|
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
|
|
|
*/
|
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];
|
|
|
|
|
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
|
|
|
}
|
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
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
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;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
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
|
|
|
}
|
|
|
|
|
|
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
|
|
|
|
2001-02-11 03:44:43 +00:00
|
|
|
if (lp->portid != (u_int32_t) BITS2WORD(pdb.pdb_portid_bits)) {
|
|
|
|
|
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 =
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[7]));
|
|
|
|
|
lp->port_wwn =
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[6]) << 8) |
|
|
|
|
|
(((u_int64_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
|
|
|
/*
|
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
|
|
|
*/
|
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);
|
|
|
|
|
mbs.param[0] = MBOX_FABRIC_LOGOUT;
|
|
|
|
|
mbs.param[1] = lp->loopid << 8;
|
|
|
|
|
mbs.param[2] = 0;
|
|
|
|
|
mbs.param[3] = 0;
|
|
|
|
|
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
|
|
|
}
|
|
|
|
|
/*
|
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
|
|
|
*/
|
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
|
|
|
|
2001-02-11 03:44:43 +00:00
|
|
|
static int
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_scan_loop(struct ispsoftc *isp)
|
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:
|
|
|
|
|
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
|
|
|
}
|
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
|
|
|
|
2001-01-15 18:33:08 +00:00
|
|
|
/*
|
2001-02-11 03:44:43 +00:00
|
|
|
* make sure the temp port database is clean...
|
2001-01-15 18:33:08 +00:00
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
MEMZERO((void *)fcp->tport, sizeof (fcp->tport));
|
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
|
|
|
/*
|
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
|
|
|
*/
|
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
|
|
|
|
2001-02-11 03:44:43 +00:00
|
|
|
/*
|
|
|
|
|
* Don't even try for ourselves...
|
|
|
|
|
*/
|
|
|
|
|
if (loopid == fcp->isp_loopid)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
lp->node_wwn = isp_get_portname(isp, loopid, 1);
|
|
|
|
|
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP)
|
|
|
|
|
return (-1);
|
|
|
|
|
if (lp->node_wwn == 0)
|
|
|
|
|
continue;
|
|
|
|
|
lp->port_wwn = isp_get_portname(isp, loopid, 0);
|
|
|
|
|
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP)
|
|
|
|
|
return (-1);
|
|
|
|
|
if (lp->port_wwn == 0) {
|
|
|
|
|
lp->node_wwn = 0;
|
2001-01-09 02:46:23 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
|
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
|
|
|
/*
|
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
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
if (isp_getpdb(isp, loopid, &pdb) != 0) {
|
|
|
|
|
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP)
|
|
|
|
|
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
|
|
|
continue;
|
2001-01-09 02:46:23 +00:00
|
|
|
}
|
2001-02-11 03:44:43 +00:00
|
|
|
if (fcp->isp_loopstate < LOOP_SCANNING_LOOP) {
|
|
|
|
|
return (-1);
|
|
|
|
|
}
|
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
|
|
|
/*
|
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
|
|
|
*/
|
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 =
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_nodename[7]));
|
|
|
|
|
lp->port_wwn =
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[0]) << 56) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[1]) << 48) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[2]) << 40) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[3]) << 32) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[4]) << 24) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[5]) << 16) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[6]) << 8) |
|
|
|
|
|
(((u_int64_t)pdb.pdb_portname[7]));
|
|
|
|
|
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;
|
2001-01-09 02:46:23 +00:00
|
|
|
}
|
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
|
|
|
|
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;
|
2000-05-09 01:14:43 +00:00
|
|
|
|
|
|
|
|
/*
|
2001-02-11 03:44:43 +00:00
|
|
|
* If we don't have a non-zero Port WWN, we're not here.
|
2000-05-09 01:14:43 +00:00
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
if (fcp->tport[loopid].port_wwn == 0) {
|
|
|
|
|
continue;
|
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
|
|
|
/*
|
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
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
if (loopid == fcp->isp_iid) {
|
|
|
|
|
continue;
|
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
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
*/
|
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);
|
2000-02-29 05:52:14 +00:00
|
|
|
}
|
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;
|
2000-02-29 05:52:14 +00:00
|
|
|
}
|
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;
|
1999-08-16 19:59:55 +00:00
|
|
|
}
|
2001-02-11 03:44:43 +00:00
|
|
|
break;
|
|
|
|
|
}
|
2000-05-09 01:14:43 +00:00
|
|
|
|
|
|
|
|
/*
|
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.
|
2000-05-09 01:14:43 +00:00
|
|
|
*/
|
2001-02-11 03:44:43 +00:00
|
|
|
if (i < hival) {
|
2001-06-05 17:11:06 +00:00
|
|
|
isp_prt(isp, ISP_LOGINFO, retained,
|
|
|
|
|
fcp->portdb[i].loopid, i, fcp->portdb[i].portid);
|
1999-11-21 03:18:22 +00:00
|
|
|
continue;
|
2000-05-09 01:14:43 +00:00
|
|
|
}
|
1999-11-21 03:18:22 +00:00
|
|
|
|
2000-05-09 01:14:43 +00:00
|
|
|
/*
|
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).
|
2000-05-09 01:14:43 +00:00
|
|
|
*/
|
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;
|
2000-05-09 01:14:43 +00:00
|
|
|
}
|
|
|
|
|
|
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;
|
2000-05-09 01:14:43 +00:00
|
|
|
|
1999-11-21 03:18:22 +00:00
|
|
|
/*
|
2001-02-11 03:44:43 +00:00
|
|
|
* Tell the outside world we've arrived.
|
1999-11-21 03:18:22 +00:00
|
|
|
*/
|
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
|
|
|
}
|
2001-02-11 03:44:43 +00:00
|
|
|
|
2000-01-03 23:52:41 +00:00
|
|
|
/*
|
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.
|
2000-01-03 23:52:41 +00:00
|
|
|
*/
|
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.
|
2001-07-04 18:42:41 +00:00
|
|
|
*
|
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);
|
|
|
|
|
}
|
|
|
|
|
|
2002-04-04 23:46:01 +00:00
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
isp_fabric_mbox_cmd(struct ispsoftc *isp, mbreg_t *mbp)
|
|
|
|
|
{
|
|
|
|
|
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
|
2002-04-04 23:46:01 +00:00
|
|
|
isp_scan_fabric(struct ispsoftc *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;
|
2001-10-18 17:26:52 +00:00
|
|
|
u_int32_t portid, first_portid, last_portid;
|
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
|
|
|
|
2001-02-11 03:44:43 +00:00
|
|
|
if (fcp->isp_onfabric == 0) {
|
|
|
|
|
fcp->isp_loopstate = LOOP_FSCAN_DONE;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
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
|
|
|
|
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.
|
|
|
|
|
*/
|
2001-10-18 17:26:52 +00:00
|
|
|
last_port_same = 0;
|
|
|
|
|
last_portid = 0xffffffff; /* not a port */
|
2001-01-15 18:33:08 +00:00
|
|
|
first_portid = portid = fcp->isp_portid;
|
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
|
|
|
|
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;
|
2002-04-04 23:46:01 +00:00
|
|
|
sns_ga_nxt_rsp_t *rs0, *rs1;
|
|
|
|
|
struct lportdb lcl;
|
|
|
|
|
u_int8_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;
|
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;
|
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);
|
2002-04-04 23:46:01 +00:00
|
|
|
MEMORYBARRIER(isp, SYNC_SFORDEV, 0, SNS_GA_NXT_REQ_SIZE);
|
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;
|
2002-04-04 23:46:01 +00:00
|
|
|
mbs.param[1] = SNS_GA_NXT_REQ_SIZE >> 1;
|
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);
|
2002-04-04 23:46:01 +00:00
|
|
|
if (isp_fabric_mbox_cmd(isp, &mbs)) {
|
|
|
|
|
if (fcp->isp_loopstate >= LOOP_SCANNING_FABRIC) {
|
2001-02-11 03:44:43 +00:00
|
|
|
fcp->isp_loopstate = LOOP_PDB_RCVD;
|
|
|
|
|
}
|
2002-04-04 23:46:01 +00:00
|
|
|
FC_SCRATCH_RELEASE(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
return (-1);
|
|
|
|
|
}
|
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;
|
|
|
|
|
rs0 = (sns_ga_nxt_rsp_t *) ((u_int8_t *)fcp->isp_scratch+0x100);
|
|
|
|
|
isp_get_ga_nxt_response(isp, rs0, rs1);
|
|
|
|
|
if (rs1->snscb_cthdr.ct_response != FS_ACC) {
|
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",
|
2002-04-04 23:46:01 +00:00
|
|
|
rs1->snscb_cthdr.ct_reason,
|
|
|
|
|
rs1->snscb_cthdr.ct_explanation, portid);
|
2002-02-04 21:04:25 +00:00
|
|
|
FC_SCRATCH_RELEASE(isp);
|
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
|
|
|
}
|
2002-04-04 23:46:01 +00:00
|
|
|
portid =
|
|
|
|
|
(((u_int32_t) rs1->snscb_port_id[0]) << 16) |
|
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
|
|
|
(((u_int32_t) rs1->snscb_port_id[1]) << 8) |
|
|
|
|
|
(((u_int32_t) rs1->snscb_port_id[2]));
|
2002-04-04 23:46:01 +00:00
|
|
|
|
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.
|
|
|
|
|
*/
|
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 =
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[0]) << 56) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[1]) << 48) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[2]) << 40) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[3]) << 32) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[4]) << 24) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[5]) << 16) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[6]) << 8) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_nodename[7]));
|
|
|
|
|
lcl.port_wwn =
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[0]) << 56) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[1]) << 48) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[2]) << 40) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[3]) << 32) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[4]) << 24) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[5]) << 16) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[6]) << 8) |
|
|
|
|
|
(((u_int64_t)rs1->snscb_portname[7]));
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* 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);
|
|
|
|
|
}
|
2001-01-15 18:33:08 +00:00
|
|
|
if (first_portid == portid) {
|
2001-02-11 03:44:43 +00:00
|
|
|
fcp->isp_loopstate = LOOP_FSCAN_DONE;
|
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);
|
|
|
|
|
}
|
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 {
|
2002-04-04 23:46:01 +00:00
|
|
|
last_port_same = 0 ;
|
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
|
|
|
}
|
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)
|
2001-02-11 03:44:43 +00:00
|
|
|
|
2002-04-04 23:46:01 +00:00
|
|
|
static int
|
|
|
|
|
isp_scan_fabric(struct ispsoftc *isp, int ftype)
|
|
|
|
|
{
|
|
|
|
|
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);
|
2001-10-18 17:26:52 +00:00
|
|
|
}
|
2001-02-11 03:44:43 +00:00
|
|
|
|
2002-04-04 23:46:01 +00:00
|
|
|
FC_SCRATCH_ACQUIRE(isp);
|
|
|
|
|
fcp->isp_loopstate = LOOP_SCANNING_FABRIC;
|
|
|
|
|
|
2002-04-16 19:55:35 +00:00
|
|
|
rq = (sns_gid_ft_req_t *)fcp->tport;
|
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);
|
|
|
|
|
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
|
|
|
/*
|
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
|
|
|
*/
|
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);
|
2002-04-16 19:55:35 +00:00
|
|
|
rs1 = (sns_gid_ft_rsp_t *) fcp->tport;
|
2002-04-04 23:46:01 +00:00
|
|
|
rs0 = (sns_gid_ft_rsp_t *) ((u_int8_t *)fcp->isp_scratch+IGPOFF);
|
|
|
|
|
isp_get_gid_ft_response(isp, rs0, rs1, NGENT);
|
|
|
|
|
if (rs1->snscb_cthdr.ct_response != FS_ACC) {
|
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",
|
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;
|
2002-04-16 19:55:35 +00:00
|
|
|
#if 0
|
|
|
|
|
sns_gff_id_rsp_t *fs0, ffsbuf, *fs1 = &ffsbuf;
|
|
|
|
|
#endif
|
2002-04-04 23:46:01 +00:00
|
|
|
|
|
|
|
|
i++;
|
|
|
|
|
MEMZERO(&lcl, sizeof (lcl));
|
|
|
|
|
lcl.fc4_type = ftype;
|
|
|
|
|
lcl.portid =
|
|
|
|
|
(((u_int32_t) rs1->snscb_ports[i].portid[0]) << 16) |
|
|
|
|
|
(((u_int32_t) rs1->snscb_ports[i].portid[1]) << 8) |
|
|
|
|
|
(((u_int32_t) rs1->snscb_ports[i].portid[2]));
|
|
|
|
|
|
|
|
|
|
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);
|
|
|
|
|
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);
|
|
|
|
|
gs0 = (sns_gxn_id_rsp_t *) ((u_int8_t *)fcp->isp_scratch+GXOFF);
|
|
|
|
|
isp_get_gxn_id_response(isp, gs0, gs1);
|
|
|
|
|
if (gs1->snscb_cthdr.ct_response != FS_ACC) {
|
|
|
|
|
isp_prt(isp, ISP_LOGWARN, swrej, "GPN_ID",
|
2002-04-16 19:55:35 +00:00
|
|
|
gs1->snscb_cthdr.ct_reason,
|
|
|
|
|
gs1->snscb_cthdr.ct_explanation, lcl.portid);
|
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 =
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[0]) << 56) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[1]) << 48) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[2]) << 40) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[3]) << 32) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[4]) << 24) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[5]) << 16) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[6]) << 8) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[7]));
|
|
|
|
|
|
|
|
|
|
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);
|
|
|
|
|
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);
|
|
|
|
|
gs0 = (sns_gxn_id_rsp_t *) ((u_int8_t *)fcp->isp_scratch+GXOFF);
|
|
|
|
|
isp_get_gxn_id_response(isp, gs0, gs1);
|
|
|
|
|
if (gs1->snscb_cthdr.ct_response != FS_ACC) {
|
|
|
|
|
isp_prt(isp, ISP_LOGWARN, swrej, "GNN_ID",
|
2002-04-16 19:55:35 +00:00
|
|
|
gs1->snscb_cthdr.ct_reason,
|
|
|
|
|
gs1->snscb_cthdr.ct_explanation, lcl.portid);
|
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 =
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[0]) << 56) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[1]) << 48) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[2]) << 40) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[3]) << 32) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[4]) << 24) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[5]) << 16) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[6]) << 8) |
|
|
|
|
|
(((u_int64_t)gs1->snscb_wwn[7]));
|
|
|
|
|
|
|
|
|
|
/*
|
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);
|
|
|
|
|
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
|
2002-04-04 23:46:01 +00:00
|
|
|
*/
|
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);
|
|
|
|
|
fs0 = (sns_gff_id_rsp_t *) ((u_int8_t *)fcp->isp_scratch+GXOFF);
|
|
|
|
|
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
|
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);
|
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);
|
|
|
|
|
}
|
2002-04-04 23:46:01 +00:00
|
|
|
#endif
|
2001-01-09 02:46:23 +00:00
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
isp_register_fc4_type(struct ispsoftc *isp)
|
|
|
|
|
{
|
|
|
|
|
fcparam *fcp = isp->isp_param;
|
2002-04-04 23:46:01 +00:00
|
|
|
u_int8_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;
|
2001-01-09 02:46:23 +00:00
|
|
|
mbreg_t mbs;
|
|
|
|
|
|
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;
|
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);
|
2001-01-09 02:46:23 +00:00
|
|
|
reqp->snscb_sblen = 22;
|
2002-04-04 23:46:01 +00:00
|
|
|
reqp->snscb_data[0] = SNS_RFT_ID;
|
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;
|
2002-04-04 23:46:01 +00:00
|
|
|
reqp->snscb_data[6] = (1 << FC4_SCSI);
|
2001-01-09 02:46:23 +00:00
|
|
|
#if 0
|
2002-04-04 23:46:01 +00:00
|
|
|
reqp->snscb_data[6] |= (1 << FC4_IP); /* ISO/IEC 8802-2 LLC/SNAP */
|
2001-01-09 02:46:23 +00:00
|
|
|
#endif
|
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);
|
2001-01-09 02:46:23 +00:00
|
|
|
mbs.param[0] = MBOX_SEND_SNS;
|
2002-04-04 23:46:01 +00:00
|
|
|
mbs.param[1] = SNS_RFT_ID_REQ_SIZE >> 1;
|
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);
|
2001-01-09 02:46:23 +00:00
|
|
|
isp_mboxcmd(isp, &mbs, MBLOGALL);
|
2002-02-04 21:04:25 +00:00
|
|
|
FC_SCRATCH_RELEASE(isp);
|
2001-01-09 02:46:23 +00:00
|
|
|
if (mbs.param[0] == MBOX_COMMAND_COMPLETE) {
|
2001-01-15 18:33:08 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG0, "Register FC4 types succeeded");
|
2001-01-09 02:46:23 +00:00
|
|
|
}
|
|
|
|
|
}
|
2001-02-11 03:44:43 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
1998-09-15 08:42:56 +00:00
|
|
|
* Start a command. Locking is assumed done in the caller.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
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
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_start(XS_T *xs)
|
1998-04-22 17:54:58 +00:00
|
|
|
{
|
|
|
|
|
struct ispsoftc *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
|
|
|
u_int16_t nxti, optr, handle;
|
|
|
|
|
u_int8_t local[QENTRY_LEN];
|
|
|
|
|
ispreq_t *reqp, *qep;
|
1999-10-17 18:58:22 +00:00
|
|
|
int target, i;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
|
|
|
|
XS_INITERR(xs);
|
|
|
|
|
isp = XS_ISP(xs);
|
|
|
|
|
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
return (CMD_COMPLETE);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
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).
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
1999-03-17 05:04:39 +00:00
|
|
|
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
return (CMD_COMPLETE);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
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);
|
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;
|
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);
|
|
|
|
|
}
|
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.
|
|
|
|
|
*/
|
2001-03-02 06:28:55 +00:00
|
|
|
}
|
|
|
|
|
#else
|
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);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2000-04-21 02:04:34 +00:00
|
|
|
/*
|
|
|
|
|
* If we're not on a Fabric, we can't have a target
|
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.
|
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);
|
|
|
|
|
}
|
2001-01-09 02:46:23 +00:00
|
|
|
if (fcp->isp_topo != TOPO_F_PORT &&
|
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
|
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);
|
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
|
|
|
}
|
|
|
|
|
|
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) {
|
2002-04-04 23:46:01 +00:00
|
|
|
if (isp_scan_fabric(isp, FC4_SCSI)) {
|
2001-02-11 03:44:43 +00:00
|
|
|
return (CMD_RQLATER);
|
|
|
|
|
}
|
|
|
|
|
if (fcp->isp_fwstate != FW_READY ||
|
2002-06-16 05:18:22 +00:00
|
|
|
fcp->isp_loopstate < LOOP_FSCAN_DONE) {
|
2001-02-11 03:44:43 +00:00
|
|
|
return (CMD_RQLATER);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
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
|
1999-11-21 03:18:22 +00:00
|
|
|
* we have to go off and (re)synchronize our port
|
|
|
|
|
* database.
|
1999-03-17 05:04:39 +00:00
|
|
|
*/
|
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);
|
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
|
|
|
|
|
*/
|
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);
|
|
|
|
|
}
|
|
|
|
|
/*
|
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;
|
1999-03-17 05:04:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Next check to see if any HBA or Device
|
1998-09-15 08:42:56 +00:00
|
|
|
* parameters need to be updated.
|
|
|
|
|
*/
|
1999-05-11 05:06:55 +00:00
|
|
|
if (isp->isp_update != 0) {
|
1998-09-15 08:42:56 +00:00
|
|
|
isp_update(isp);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
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
|
|
|
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");
|
1998-04-22 17:54:58 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
return (CMD_EAGAIN);
|
|
|
|
|
}
|
|
|
|
|
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
if (isp->isp_sendmarker) {
|
1999-12-16 05:42:02 +00:00
|
|
|
u_int8_t n = (IS_DUALBUS(isp)? 2: 1);
|
1998-09-17 21:03:45 +00:00
|
|
|
/*
|
1999-05-11 05:06:55 +00:00
|
|
|
* Check ports to send markers for...
|
1998-09-17 21:03:45 +00:00
|
|
|
*/
|
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);
|
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;
|
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);
|
|
|
|
|
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+");
|
1999-05-11 05:06:55 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
return (CMD_EAGAIN);
|
|
|
|
|
}
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
reqp->req_header.rqs_entry_count = 1;
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_FC(isp)) {
|
1998-04-22 17:54:58 +00:00
|
|
|
reqp->req_header.rqs_entry_type = RQSTYPE_T2RQS;
|
|
|
|
|
} else {
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
2001-10-23 23:05:20 +00:00
|
|
|
/* reqp->req_header.rqs_flags = 0; */
|
|
|
|
|
/* reqp->req_header.rqs_seqno = 0; */
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_FC(isp)) {
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
|
|
|
|
* See comment in isp_intr
|
|
|
|
|
*/
|
2001-10-23 23:05:20 +00:00
|
|
|
/* XS_RESID(xs) = 0; */
|
1999-03-25 22:52:45 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
/*
|
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).
|
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);
|
1999-10-28 02:48:42 +00:00
|
|
|
} else {
|
2000-08-27 23:38:44 +00:00
|
|
|
/*
|
|
|
|
|
* If we don't know what tag to use, use HEAD OF QUEUE
|
2001-08-31 21:39:04 +00:00
|
|
|
* for Request Sense or Simple.
|
2000-08-27 23:38:44 +00:00
|
|
|
*/
|
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;
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
sdparam *sdp = (sdparam *)isp->isp_param;
|
2001-07-04 18:42:41 +00:00
|
|
|
sdp += XS_CHANNEL(xs);
|
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);
|
1998-04-22 17:54:58 +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
|
|
|
reqp->req_target = target | (XS_CHANNEL(xs) << 7);
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1998-12-05 01:33:57 +00:00
|
|
|
reqp->req_lun_trn = XS_LUN(xs);
|
1998-04-22 17:54:58 +00:00
|
|
|
reqp->req_cdblen = XS_CDBLEN(xs);
|
1998-12-05 01:33:57 +00:00
|
|
|
} else {
|
2001-09-03 03:09:48 +00:00
|
|
|
if (FCPARAM(isp)->isp_fwattr & ISP_FW_ATTR_SCCLUN)
|
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_scclun = XS_LUN(xs);
|
2000-06-18 04:56:17 +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_lun_trn = XS_LUN(xs);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1999-01-10 11:15:23 +00:00
|
|
|
MEMCPY(reqp->req_cdb, XS_CDBP(xs), XS_CDBLEN(xs));
|
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)) {
|
1998-04-22 17:54:58 +00:00
|
|
|
reqp->req_time = 1;
|
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
|
|
|
|
2001-03-02 06:28:55 +00:00
|
|
|
if (isp_save_xs(isp, xs, &handle)) {
|
2001-10-23 23:05:20 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG0, "out of xflist pointers");
|
1999-10-17 18:58:22 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
return (CMD_EAGAIN);
|
|
|
|
|
}
|
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
|
|
|
|
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);
|
1998-09-17 21:03:45 +00:00
|
|
|
if (i != CMD_QUEUED) {
|
2001-03-02 06:28:55 +00:00
|
|
|
isp_destroy_handle(isp, handle);
|
1998-09-17 21:03:45 +00:00
|
|
|
/*
|
|
|
|
|
* dmasetup sets actual error in packet, and
|
|
|
|
|
* return what we were given to return.
|
|
|
|
|
*/
|
|
|
|
|
return (i);
|
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,
|
2001-01-15 18:33:08 +00:00
|
|
|
"START cmd for %d.%d.%d cmd 0x%x datalen %ld",
|
2002-06-16 05:18:22 +00:00
|
|
|
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs), XS_CDBP(xs)[0],
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
isp->isp_nactive++;
|
|
|
|
|
return (CMD_QUEUED);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
1998-09-15 08:42:56 +00:00
|
|
|
* isp control
|
|
|
|
|
* Locks (ints blocked) assumed held.
|
1998-04-22 17:54:58 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
int
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_control(struct ispsoftc *isp, ispctl_t ctl, void *arg)
|
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;
|
1998-09-15 08:42:56 +00:00
|
|
|
mbreg_t mbs;
|
1999-10-17 18:58:22 +00:00
|
|
|
int bus, tgt;
|
2001-03-02 06:28:55 +00:00
|
|
|
u_int16_t handle;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ISPCTL_RESET_BUS:
|
1999-03-17 05:04:39 +00:00
|
|
|
/*
|
|
|
|
|
* Issue a bus reset.
|
|
|
|
|
*/
|
1998-09-15 08:42:56 +00:00
|
|
|
mbs.param[0] = MBOX_BUS_RESET;
|
2000-02-29 05:52:14 +00:00
|
|
|
mbs.param[2] = 0;
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1999-02-09 01:07:06 +00:00
|
|
|
mbs.param[1] =
|
|
|
|
|
((sdparam *) isp->isp_param)->isp_bus_reset_delay;
|
1999-03-17 05:04:39 +00:00
|
|
|
if (mbs.param[1] < 2)
|
|
|
|
|
mbs.param[1] = 2;
|
1999-08-16 19:59:55 +00:00
|
|
|
bus = *((int *) arg);
|
2000-02-29 05:52:14 +00:00
|
|
|
if (IS_DUALBUS(isp))
|
|
|
|
|
mbs.param[2] = bus;
|
1999-02-09 01:07:06 +00:00
|
|
|
} else {
|
1999-08-16 19:59:55 +00:00
|
|
|
mbs.param[1] = 10;
|
|
|
|
|
bus = 0;
|
1999-02-09 01:07:06 +00:00
|
|
|
}
|
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);
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
return (0);
|
|
|
|
|
|
1999-03-25 22:52:45 +00:00
|
|
|
case ISPCTL_RESET_DEV:
|
1999-05-11 05:06:55 +00:00
|
|
|
tgt = (*((int *) arg)) & 0xffff;
|
|
|
|
|
bus = (*((int *) arg)) >> 16;
|
1998-09-15 08:42:56 +00:00
|
|
|
mbs.param[0] = MBOX_ABORT_TARGET;
|
1999-05-11 05:06:55 +00:00
|
|
|
mbs.param[1] = (tgt << 8) | (bus << 15);
|
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);
|
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);
|
2000-04-21 02:04:34 +00:00
|
|
|
isp->isp_sendmarker |= (1 << bus);
|
1998-09-15 08:42:56 +00:00
|
|
|
return (0);
|
|
|
|
|
|
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;
|
2000-07-05 06:41:36 +00:00
|
|
|
tgt = XS_TGT(xs);
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
bus = XS_CHANNEL(xs);
|
1998-09-15 08:42:56 +00:00
|
|
|
mbs.param[0] = MBOX_ABORT;
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_FC(isp)) {
|
2001-09-03 03:09:48 +00:00
|
|
|
if (FCPARAM(isp)->isp_fwattr & ISP_FW_ATTR_SCCLUN) {
|
2000-07-05 06:41:36 +00:00
|
|
|
mbs.param[1] = tgt << 8;
|
2000-06-18 04:56:17 +00:00
|
|
|
mbs.param[4] = 0;
|
|
|
|
|
mbs.param[5] = 0;
|
|
|
|
|
mbs.param[6] = XS_LUN(xs);
|
|
|
|
|
} else {
|
2000-07-05 06:41:36 +00:00
|
|
|
mbs.param[1] = tgt << 8 | XS_LUN(xs);
|
2000-06-18 04:56:17 +00:00
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
} else {
|
|
|
|
|
mbs.param[1] =
|
|
|
|
|
(bus << 15) | (XS_TGT(xs) << 8) | XS_LUN(xs);
|
|
|
|
|
}
|
2001-03-02 06:28:55 +00:00
|
|
|
mbs.param[3] = 0;
|
|
|
|
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
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;
|
1998-09-15 08:42:56 +00:00
|
|
|
|
|
|
|
|
case ISPCTL_UPDATE_PARAMS:
|
2001-02-11 03:44:43 +00:00
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
isp_update(isp);
|
1999-03-25 22:52:45 +00:00
|
|
|
return (0);
|
1999-03-17 05:04:39 +00:00
|
|
|
|
|
|
|
|
case ISPCTL_FCLINK_TEST:
|
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)) {
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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 usdelay = (arg)? *((int *) arg) : 250000;
|
|
|
|
|
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;
|
1999-11-21 03:18:22 +00:00
|
|
|
|
2001-02-11 03:44:43 +00:00
|
|
|
case ISPCTL_SCAN_FABRIC:
|
|
|
|
|
|
|
|
|
|
if (IS_FC(isp)) {
|
2002-04-04 23:46:01 +00:00
|
|
|
int ftype = (arg)? *((int *) arg) : FC4_SCSI;
|
|
|
|
|
return (isp_scan_fabric(isp, ftype));
|
2001-02-11 03:44:43 +00:00
|
|
|
}
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ISPCTL_SCAN_LOOP:
|
|
|
|
|
|
|
|
|
|
if (IS_FC(isp)) {
|
|
|
|
|
return (isp_scan_loop(isp));
|
|
|
|
|
}
|
|
|
|
|
break;
|
|
|
|
|
|
2000-01-03 23:52:41 +00:00
|
|
|
case ISPCTL_PDB_SYNC:
|
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)) {
|
2001-02-11 03:44:43 +00:00
|
|
|
mbs.param[0] = MBOX_INIT_LIP;
|
|
|
|
|
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;
|
2001-02-11 03:44:43 +00:00
|
|
|
|
2001-02-23 05:35:50 +00:00
|
|
|
case ISPCTL_RUN_MBOXCMD:
|
|
|
|
|
|
|
|
|
|
isp_mboxcmd(isp, arg, MBLOGALL);
|
|
|
|
|
return(0);
|
|
|
|
|
|
2000-01-03 23:52:41 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
|
|
|
|
case ISPCTL_TOGGLE_TMODE:
|
2000-08-27 23:38:44 +00:00
|
|
|
{
|
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.
|
|
|
|
|
*/
|
2000-01-03 23:52:41 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
2001-04-04 21:42:59 +00:00
|
|
|
int param = *(int *)arg;
|
2000-01-03 23:52:41 +00:00
|
|
|
mbs.param[0] = MBOX_ENABLE_TARGET_MODE;
|
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);
|
2000-01-03 23:52:41 +00:00
|
|
|
if (mbs.param[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return (0);
|
2000-08-27 23:38:44 +00:00
|
|
|
}
|
1999-11-21 03:18:22 +00:00
|
|
|
#endif
|
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.
|
|
|
|
|
*/
|
2002-02-04 21:04:25 +00:00
|
|
|
#ifndef MAX_REQUESTQ_COMPLETIONS
|
2002-01-03 20:43:22 +00:00
|
|
|
#define MAX_REQUESTQ_COMPLETIONS 64
|
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
|
|
|
|
2001-08-31 21:39:04 +00:00
|
|
|
void
|
|
|
|
|
isp_intr(struct ispsoftc *isp, u_int16_t isr, u_int16_t sema, u_int16_t mbox)
|
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;
|
2001-08-31 21:39:04 +00:00
|
|
|
u_int16_t iptr, optr, junk;
|
1999-02-09 01:07:06 +00:00
|
|
|
int i, nlooked = 0, ndone = 0;
|
1998-09-15 08:42:56 +00:00
|
|
|
|
2002-02-04 21:04:25 +00:00
|
|
|
again:
|
2001-08-31 21:39:04 +00:00
|
|
|
/*
|
|
|
|
|
* Is this a mailbox related interrupt?
|
|
|
|
|
* The mailbox semaphore will be nonzero if so.
|
|
|
|
|
*/
|
1999-05-11 05:06:55 +00:00
|
|
|
if (sema) {
|
1999-03-17 05:04:39 +00:00
|
|
|
if (mbox & 0x4000) {
|
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;
|
2000-12-29 19:12:44 +00:00
|
|
|
for (i = 1; i < MAX_MAILBOX; 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));
|
|
|
|
|
}
|
2002-02-04 21:04:25 +00:00
|
|
|
if (isp->isp_mbxwrk0) {
|
|
|
|
|
if (isp_mbox_continue(isp) == 0) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
}
|
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
|
|
|
}
|
2002-01-03 20:43:22 +00:00
|
|
|
} else if (isp_parse_async(isp, mbox) < 0) {
|
|
|
|
|
return;
|
1999-01-30 07:29:00 +00:00
|
|
|
}
|
2002-08-17 17:29:15 +00:00
|
|
|
if ((IS_FC(isp) && mbox != ASYNC_RIO_RESP) ||
|
|
|
|
|
isp->isp_state != ISP_RUNSTATE) {
|
2000-06-18 04:56:17 +00:00
|
|
|
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
|
2001-08-31 21:39:04 +00:00
|
|
|
ISP_WRITE(isp, BIU_SEMA, 0);
|
|
|
|
|
return;
|
2000-06-18 04:56:17 +00:00
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
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,
|
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);
|
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
|
|
|
}
|
|
|
|
|
|
1999-04-04 02:28:29 +00:00
|
|
|
/*
|
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.
|
1999-04-04 02:28:29 +00:00
|
|
|
*/
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
2001-08-31 21:39:04 +00:00
|
|
|
optr = ISP_READ(isp, isp->isp_respoutrp);
|
2002-02-04 21:04:25 +00:00
|
|
|
/*
|
|
|
|
|
* Debug: to be taken out eventually
|
|
|
|
|
*/
|
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
|
|
|
|
2001-08-31 21:39:04 +00:00
|
|
|
/*
|
|
|
|
|
* You *must* read the Response Queue In Pointer
|
|
|
|
|
* prior to clearing the RISC interrupt.
|
2002-01-03 20:43:22 +00:00
|
|
|
*
|
|
|
|
|
* Debounce the 2300 if revision less than 2.
|
2001-08-31 21:39:04 +00:00
|
|
|
*/
|
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 {
|
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,
|
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 {
|
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
|
|
|
}
|
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
|
|
|
|
|
|
|
|
|
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.
|
|
|
|
|
*/
|
2002-01-03 20:43:22 +00:00
|
|
|
if (IS_23XX(isp)) {
|
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) {
|
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,
|
2001-08-31 21:39:04 +00:00
|
|
|
"bogus intr- isr %x (%x) iptr %x optr %x",
|
|
|
|
|
isr, junk, iptr, optr);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
2002-03-21 21:10:16 +00:00
|
|
|
isp->isp_resodx = iptr;
|
2001-08-31 21:39:04 +00:00
|
|
|
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
|
|
|
|
|
ISP_WRITE(isp, BIU_SEMA, 0);
|
1998-04-22 17:54:58 +00:00
|
|
|
|
2002-03-21 21:10:16 +00:00
|
|
|
if (isp->isp_rspbsy) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
isp->isp_rspbsy = 1;
|
|
|
|
|
|
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;
|
1999-11-21 03:18:22 +00:00
|
|
|
u_int16_t oop;
|
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);
|
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));
|
1999-02-09 01:07:06 +00:00
|
|
|
nlooked++;
|
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.
|
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);
|
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;
|
2002-03-21 21:10:16 +00:00
|
|
|
MEMZERO(hp, QENTRY_LEN); /* PERF */
|
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 {
|
2002-03-07 17:32:45 +00:00
|
|
|
/*
|
|
|
|
|
* Somebody reachable via isp_handle_other_response
|
|
|
|
|
* may have updated the response queue pointers for
|
2002-03-21 21:10:16 +00:00
|
|
|
* us, so we reload our goal index.
|
2002-03-07 17:32:45 +00:00
|
|
|
*/
|
2002-03-21 21:10:16 +00:00
|
|
|
if (isp_handle_other_response(isp, type, hp, &optr)) {
|
|
|
|
|
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 */
|
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);
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
|
|
|
|
* It really has to be a bounced request just copied
|
1999-04-14 17:37:36 +00:00
|
|
|
* from the request queue to the response queue. If
|
|
|
|
|
* not, something bad has happened.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
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,
|
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 */
|
1998-04-22 17:54:58 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
buddaboom = 1;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (sp->req_header.rqs_flags & 0xf) {
|
1999-04-14 17:37:36 +00:00
|
|
|
#define _RQS_OFLAGS \
|
|
|
|
|
~(RQSFLAG_CONTINUATION|RQSFLAG_FULL|RQSFLAG_BADHEADER|RQSFLAG_BADPACKET)
|
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");
|
2001-08-31 21:39:04 +00:00
|
|
|
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
|
1998-04-22 17:54:58 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
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");
|
1999-04-14 17:37:36 +00:00
|
|
|
/*
|
|
|
|
|
* We'll synthesize a QUEUE FULL message below.
|
|
|
|
|
*/
|
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");
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
buddaboom++;
|
|
|
|
|
}
|
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);
|
1999-04-14 17:37:36 +00:00
|
|
|
buddaboom++;
|
|
|
|
|
}
|
|
|
|
|
#undef _RQS_OFLAGS
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
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,
|
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);
|
2001-08-31 21:39:04 +00:00
|
|
|
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
|
1998-04-22 17:54:58 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
xs = isp_find_xs(isp, sp->req_handle);
|
1998-04-22 17:54:58 +00:00
|
|
|
if (xs == NULL) {
|
2002-04-16 19:55:35 +00:00
|
|
|
u_int8_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 */
|
2002-04-16 19:55:35 +00:00
|
|
|
/*
|
|
|
|
|
* Only whine if this isn't the expected fallout of
|
|
|
|
|
* aborting the command.
|
|
|
|
|
*/
|
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) {
|
2002-04-16 19:55:35 +00:00
|
|
|
isp_prt(isp, ISP_LOGERR,
|
2002-06-16 05:18:22 +00:00
|
|
|
"cannot find handle 0x%x (status 0x%x)",
|
|
|
|
|
sp->req_handle, ts);
|
2002-04-16 19:55:35 +00:00
|
|
|
}
|
2001-08-31 21:39:04 +00:00
|
|
|
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
|
1998-04-22 17:54:58 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
isp_destroy_handle(isp, sp->req_handle);
|
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);
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_sendmarker |= (1 << XS_CHANNEL(xs));
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
if (buddaboom) {
|
|
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
}
|
2000-08-27 23:38:44 +00:00
|
|
|
|
|
|
|
|
if (IS_FC(isp) && (sp->req_scsi_status & RQCS_SV)) {
|
1999-01-30 07:29:00 +00:00
|
|
|
/*
|
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.
|
1999-01-30 07:29:00 +00:00
|
|
|
*/
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
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);
|
2001-02-11 03:44:43 +00:00
|
|
|
/* solely for the benefit of debug */
|
|
|
|
|
sp->req_state_flags |= RQSF_GOT_SENSE;
|
2000-08-27 23:38:44 +00:00
|
|
|
}
|
|
|
|
|
}
|
2001-01-15 18:33:08 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG2,
|
|
|
|
|
"asked for %ld got resid %ld", (long) XS_XFRLEN(xs),
|
|
|
|
|
(long) sp->req_resid);
|
2000-08-27 23:38:44 +00:00
|
|
|
break;
|
|
|
|
|
case RQSTYPE_REQUEST:
|
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;
|
1999-05-11 05:06:55 +00:00
|
|
|
XS_SETERR(xs, HBA_NOERROR);
|
|
|
|
|
} else if (XS_NOERR(xs)) {
|
2001-10-23 23:05:20 +00:00
|
|
|
/*
|
|
|
|
|
* ????
|
|
|
|
|
*/
|
|
|
|
|
isp_prt(isp, ISP_LOGDEBUG0,
|
|
|
|
|
"Request Queue Entry bounced back");
|
1999-05-11 05:06:55 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
|
|
|
|
}
|
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,
|
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);
|
1999-01-10 02:55:10 +00:00
|
|
|
if (XS_NOERR(xs)) {
|
1998-04-22 17:54:58 +00:00
|
|
|
XS_SETERR(xs, HBA_BOTCH);
|
1999-01-10 02:55:10 +00:00
|
|
|
}
|
2000-08-27 23:38:44 +00:00
|
|
|
break;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
2000-08-27 23:38:44 +00:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Free any dma resources. As a side effect, this may
|
|
|
|
|
* also do any cache flushing necessary for data coherence. */
|
1998-04-22 17:54:58 +00:00
|
|
|
if (XS_XFRLEN(xs)) {
|
1999-10-17 18:58:22 +00:00
|
|
|
ISP_DMAFREE(isp, xs, sp->req_handle);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
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))) ||
|
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;
|
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
|
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 = '.';
|
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));
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
|
1998-12-05 01:33:57 +00:00
|
|
|
if (isp->isp_nactive > 0)
|
|
|
|
|
isp->isp_nactive--;
|
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;
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
1999-02-09 01:07:06 +00:00
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
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.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
1999-03-17 05:04:39 +00:00
|
|
|
if (nlooked) {
|
2001-08-31 21:39:04 +00:00
|
|
|
WRITE_RESPONSE_QUEUE_OUT_POINTER(isp, optr);
|
|
|
|
|
/*
|
2002-03-07 17:32:45 +00:00
|
|
|
* While we're at it, read the requst queue out pointer.
|
2001-08-31 21:39:04 +00:00
|
|
|
*/
|
2001-07-04 18:42:41 +00:00
|
|
|
isp->isp_reqodx = READ_REQUEST_QUEUE_OUT_POINTER(isp);
|
2002-01-03 20:43:22 +00:00
|
|
|
if (isp->isp_rscchiwater < ndone)
|
|
|
|
|
isp->isp_rscchiwater = ndone;
|
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
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
isp->isp_residx = optr;
|
2002-03-21 21:10:16 +00:00
|
|
|
isp->isp_rspbsy = 0;
|
1998-09-15 08:42:56 +00:00
|
|
|
for (i = 0; i < ndone; i++) {
|
|
|
|
|
xs = complist[i];
|
|
|
|
|
if (xs) {
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Support routines.
|
|
|
|
|
*/
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
static int
|
2002-01-03 20:43:22 +00:00
|
|
|
isp_parse_async(struct ispsoftc *isp, u_int16_t mbox)
|
1998-09-15 08:42:56 +00:00
|
|
|
{
|
2002-06-16 05:18:22 +00:00
|
|
|
int rval = 0;
|
1999-11-21 03:18:22 +00:00
|
|
|
int bus;
|
1999-01-30 07:29:00 +00:00
|
|
|
|
1999-12-16 05:42:02 +00:00
|
|
|
if (IS_DUALBUS(isp)) {
|
1999-11-21 03:18:22 +00:00
|
|
|
bus = ISP_READ(isp, OUTMAILBOX6);
|
|
|
|
|
} else {
|
|
|
|
|
bus = 0;
|
|
|
|
|
}
|
2002-01-03 20:43:22 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG2, "Async Mbox 0x%x", mbox);
|
1999-11-21 03:18:22 +00:00
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
switch (mbox) {
|
|
|
|
|
case ASYNC_BUS_RESET:
|
2000-04-21 02:04:34 +00:00
|
|
|
isp->isp_sendmarker |= (1 << bus);
|
1998-12-05 01:33:57 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
1998-12-05 01:33:57 +00:00
|
|
|
#endif
|
1999-11-21 03:18:22 +00:00
|
|
|
isp_async(isp, ISPASYNC_BUS_RESET, &bus);
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
case ASYNC_SYSTEM_ERROR:
|
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
|
2001-07-04 18:42:41 +00:00
|
|
|
isp_async(isp, ISPASYNC_FW_CRASH, NULL);
|
2002-06-16 05:18:22 +00:00
|
|
|
isp_reinit(isp);
|
|
|
|
|
isp_async(isp, ISPASYNC_FW_RESTARTED, NULL);
|
2002-08-17 17:29:15 +00:00
|
|
|
#endif
|
2002-06-16 05:18:22 +00:00
|
|
|
rval = -1;
|
|
|
|
|
break;
|
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");
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_QWAKEUP:
|
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.
|
|
|
|
|
*/
|
2001-07-04 18:42:41 +00:00
|
|
|
mbox = READ_REQUEST_QUEUE_OUT_POINTER(isp);
|
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,
|
2001-03-04 18:41:23 +00:00
|
|
|
"timeout initiated SCSI bus reset of bus %d", bus);
|
2000-04-21 02:04:34 +00:00
|
|
|
isp->isp_sendmarker |= (1 << bus);
|
1998-12-05 01:33:57 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
1998-12-05 01:33:57 +00:00
|
|
|
#endif
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
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);
|
2000-04-21 02:04:34 +00:00
|
|
|
isp->isp_sendmarker |= (1 << bus);
|
1998-12-05 01:33:57 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
1998-12-05 01:33:57 +00:00
|
|
|
#endif
|
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");
|
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");
|
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");
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_BUS_TRANSIT:
|
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");
|
2000-08-27 23:38:44 +00:00
|
|
|
SDPARAM(isp)->isp_diffmode = 0;
|
|
|
|
|
SDPARAM(isp)->isp_ultramode = 0;
|
|
|
|
|
SDPARAM(isp)->isp_lvdmode = 1;
|
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");
|
2000-08-27 23:38:44 +00:00
|
|
|
SDPARAM(isp)->isp_diffmode = 1;
|
|
|
|
|
SDPARAM(isp)->isp_ultramode = 0;
|
|
|
|
|
SDPARAM(isp)->isp_lvdmode = 0;
|
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");
|
2000-08-27 23:38:44 +00:00
|
|
|
SDPARAM(isp)->isp_diffmode = 0;
|
|
|
|
|
SDPARAM(isp)->isp_ultramode = 1;
|
|
|
|
|
SDPARAM(isp)->isp_lvdmode = 0;
|
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);
|
1999-03-25 22:52:45 +00:00
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
/*
|
|
|
|
|
* XXX: Set up to renegotiate again!
|
|
|
|
|
*/
|
1999-05-11 05:06:55 +00:00
|
|
|
/* Can only be for a 1080... */
|
2000-04-21 02:04:34 +00:00
|
|
|
isp->isp_sendmarker |= (1 << bus);
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
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:
|
1998-09-15 08:42:56 +00:00
|
|
|
case ASYNC_CMD_CMPLT:
|
2002-01-03 20:43:22 +00:00
|
|
|
bus = 0x102; /* outgoing mailbox regs 1 */
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_RIO_RESP:
|
2002-08-17 17:29:15 +00:00
|
|
|
return (rval);
|
1998-09-15 08:42:56 +00:00
|
|
|
|
|
|
|
|
case ASYNC_CTIO_DONE:
|
2002-06-16 05:18:22 +00:00
|
|
|
{
|
2000-08-27 23:38:44 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
int handle =
|
|
|
|
|
(ISP_READ(isp, OUTMAILBOX2) << 16) |
|
|
|
|
|
(ISP_READ(isp, OUTMAILBOX1));
|
|
|
|
|
if (isp_target_async(isp, handle, mbox))
|
|
|
|
|
rval = -1;
|
2000-08-27 23:38:44 +00:00
|
|
|
#else
|
|
|
|
|
isp_prt(isp, ISP_LOGINFO, "Fast Posting CTIO done");
|
|
|
|
|
#endif
|
2002-02-04 21:04:25 +00:00
|
|
|
isp->isp_fphccmplt++; /* count it as a fast posting intr */
|
2002-06-16 05:18:22 +00:00
|
|
|
break;
|
|
|
|
|
}
|
2001-08-16 17:25:08 +00:00
|
|
|
case ASYNC_LIP_F8:
|
1998-09-15 08:42:56 +00:00
|
|
|
case ASYNC_LIP_OCCURRED:
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_lipseq =
|
1999-10-17 18:58:22 +00:00
|
|
|
ISP_READ(isp, OUTMAILBOX1);
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
|
|
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
|
1999-03-17 05:04:39 +00:00
|
|
|
isp->isp_sendmarker = 1;
|
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);
|
2000-01-03 23:52:41 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
2000-01-03 23:52:41 +00:00
|
|
|
#endif
|
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);
|
|
|
|
|
}
|
|
|
|
|
}
|
2001-07-04 18:42:41 +00:00
|
|
|
if (j) {
|
|
|
|
|
isp_prt(isp, ISP_LOGERR,
|
|
|
|
|
"LIP destroyed %d active commands", j);
|
|
|
|
|
}
|
2001-06-14 17:13:24 +00:00
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_LOOP_UP:
|
1999-03-17 05:04:39 +00:00
|
|
|
isp->isp_sendmarker = 1;
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
|
|
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
|
1999-04-04 02:28:29 +00:00
|
|
|
isp_mark_getpdb_all(isp);
|
1999-02-09 01:07:06 +00:00
|
|
|
isp_async(isp, ISPASYNC_LOOP_UP, NULL);
|
2000-01-03 23:52:41 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
2000-01-03 23:52:41 +00:00
|
|
|
#endif
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_LOOP_DOWN:
|
1999-04-04 02:28:29 +00:00
|
|
|
isp->isp_sendmarker = 1;
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
|
|
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_NIL;
|
1999-04-04 02:28:29 +00:00
|
|
|
isp_mark_getpdb_all(isp);
|
1999-02-09 01:07:06 +00:00
|
|
|
isp_async(isp, ISPASYNC_LOOP_DOWN, NULL);
|
2000-01-03 23:52:41 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
2000-01-03 23:52:41 +00:00
|
|
|
#endif
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_LOOP_RESET:
|
2000-04-21 02:04:34 +00:00
|
|
|
isp->isp_sendmarker = 1;
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
|
|
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_NIL;
|
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);
|
1998-12-05 01:33:57 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
1998-12-05 01:33:57 +00:00
|
|
|
#endif
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_PDB_CHANGED:
|
1999-03-17 05:04:39 +00:00
|
|
|
isp->isp_sendmarker = 1;
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_PDB_RCVD;
|
1999-03-17 05:04:39 +00:00
|
|
|
isp_mark_getpdb_all(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_PDB);
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_CHANGE_NOTIFY:
|
1999-08-16 19:59:55 +00:00
|
|
|
/*
|
|
|
|
|
* Not correct, but it will force us to rescan the loop.
|
|
|
|
|
*/
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_PDB_RCVD;
|
2001-01-09 02:46:23 +00:00
|
|
|
isp_mark_getpdb_all(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_SNS);
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
2000-02-11 19:31:32 +00:00
|
|
|
case ASYNC_PTPMODE:
|
2000-08-27 23:38:44 +00:00
|
|
|
if (FCPARAM(isp)->isp_onfabric)
|
|
|
|
|
FCPARAM(isp)->isp_topo = TOPO_F_PORT;
|
2001-02-11 03:44:43 +00:00
|
|
|
else
|
|
|
|
|
FCPARAM(isp)->isp_topo = TOPO_N_PORT;
|
2000-05-09 01:14:43 +00:00
|
|
|
isp_mark_getpdb_all(isp);
|
|
|
|
|
isp->isp_sendmarker = 1;
|
2000-08-27 23:38:44 +00:00
|
|
|
FCPARAM(isp)->isp_fwstate = FW_CONFIG_WAIT;
|
|
|
|
|
FCPARAM(isp)->isp_loopstate = LOOP_LIP_RCVD;
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_async(isp, ISPASYNC_CHANGE_NOTIFY, ISPASYNC_CHANGE_OTHER);
|
2000-05-09 01:14:43 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-06-16 05:18:22 +00:00
|
|
|
if (isp_target_async(isp, bus, mbox))
|
|
|
|
|
rval = -1;
|
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");
|
2000-02-11 19:31:32 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case ASYNC_CONNMODE:
|
|
|
|
|
mbox = ISP_READ(isp, OUTMAILBOX1);
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_mark_getpdb_all(isp);
|
2000-02-11 19:31:32 +00:00
|
|
|
switch (mbox) {
|
|
|
|
|
case ISP_CONN_LOOP:
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_prt(isp, ISP_LOGINFO,
|
|
|
|
|
"Point-to-Point -> Loop mode");
|
2000-02-11 19:31:32 +00:00
|
|
|
break;
|
|
|
|
|
case ISP_CONN_PTP:
|
2001-02-11 03:44:43 +00:00
|
|
|
isp_prt(isp, ISP_LOGINFO,
|
|
|
|
|
"Loop -> Point-to-Point mode");
|
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,
|
2001-02-11 03:44:43 +00:00
|
|
|
"Point-to-Point -> Loop mode (BAD LIP)");
|
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");
|
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);
|
2002-06-16 05:18:22 +00:00
|
|
|
isp_async(isp, ISPASYNC_FW_RESTARTED, NULL);
|
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");
|
2001-02-11 03:44:43 +00:00
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
isp_prt(isp, ISP_LOGWARN,
|
|
|
|
|
"Unknown connection mode (0x%x)", mbox);
|
|
|
|
|
break;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
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;
|
2000-02-11 19:31:32 +00:00
|
|
|
break;
|
|
|
|
|
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
}
|
2002-01-03 20:43:22 +00:00
|
|
|
|
|
|
|
|
if (bus & 0x100) {
|
|
|
|
|
int i, nh;
|
|
|
|
|
u_int16_t handles[5];
|
|
|
|
|
|
|
|
|
|
for (nh = 0, i = 1; i < MAX_MAILBOX; i++) {
|
|
|
|
|
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++;
|
|
|
|
|
}
|
2002-06-16 05:18:22 +00:00
|
|
|
return (rval);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
|
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.
|
|
|
|
|
*/
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
static int
|
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_handle_other_response(struct ispsoftc *isp, int type,
|
|
|
|
|
isphdr_t *hp, u_int16_t *optrp)
|
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) {
|
2000-08-27 23:38:44 +00:00
|
|
|
case RQSTYPE_STATUS_CONT:
|
|
|
|
|
isp_prt(isp, ISP_LOGINFO, "Ignored Continuation Response");
|
2002-03-21 21:10:16 +00:00
|
|
|
return (1);
|
1999-10-17 18:58:22 +00:00
|
|
|
case RQSTYPE_ATIO:
|
2000-01-03 23:52:41 +00:00
|
|
|
case RQSTYPE_CTIO:
|
1998-09-15 08:42:56 +00:00
|
|
|
case RQSTYPE_ENABLE_LUN:
|
1998-12-05 01:33:57 +00:00
|
|
|
case RQSTYPE_MODIFY_LUN:
|
1999-10-17 18:58:22 +00:00
|
|
|
case RQSTYPE_NOTIFY:
|
|
|
|
|
case RQSTYPE_NOTIFY_ACK:
|
|
|
|
|
case RQSTYPE_CTIO1:
|
1998-09-15 08:42:56 +00:00
|
|
|
case RQSTYPE_ATIO2:
|
|
|
|
|
case RQSTYPE_CTIO2:
|
1999-10-17 18:58:22 +00:00
|
|
|
case RQSTYPE_CTIO3:
|
2002-02-04 21:04:25 +00:00
|
|
|
isp->isp_rsltccmplt++; /* count as a response completion */
|
1999-10-17 18:58:22 +00:00
|
|
|
#ifdef ISP_TARGET_MODE
|
2002-03-21 21:10:16 +00:00
|
|
|
if (isp_target_notify(isp, (ispstatusreq_t *) hp, optrp)) {
|
|
|
|
|
return (1);
|
|
|
|
|
}
|
1999-10-17 18:58:22 +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
|
|
|
optrp = optrp;
|
1999-10-17 18:58:22 +00:00
|
|
|
/* FALLTHROUGH */
|
1998-09-15 08:42:56 +00:00
|
|
|
#endif
|
1999-10-17 18:58:22 +00:00
|
|
|
case RQSTYPE_REQUEST:
|
1998-09-15 08:42:56 +00:00
|
|
|
default:
|
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_async(isp, ISPASYNC_UNHANDLED_RESPONSE, hp)) {
|
2002-03-21 21:10:16 +00:00
|
|
|
return (1);
|
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));
|
2002-03-21 21:10:16 +00:00
|
|
|
return (0);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
}
|
1998-12-05 01:33:57 +00:00
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_parse_status(struct ispsoftc *isp, ispstatusreq_t *sp, XS_T *xs)
|
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) {
|
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);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
|
|
|
|
|
case RQCS_INCOMPLETE:
|
|
|
|
|
if ((sp->req_state_flags & RQSF_GOT_TARGET) == 0) {
|
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",
|
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);
|
|
|
|
|
}
|
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),
|
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));
|
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];
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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
|
|
|
}
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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) {
|
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);
|
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
|
|
|
}
|
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));
|
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);
|
|
|
|
|
}
|
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));
|
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);
|
|
|
|
|
}
|
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));
|
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);
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
return;
|
|
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
case RQCS_DATA_OVERRUN:
|
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);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
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));
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
case RQCS_DATA_UNDERRUN:
|
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;
|
|
|
|
|
}
|
|
|
|
|
}
|
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);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
2002-05-01 21:58:36 +00:00
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
|
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),
|
2000-02-29 05:52:14 +00:00
|
|
|
XS_TGT(xs), XS_LUN(xs));
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
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,
|
2000-02-29 05:52:14 +00:00
|
|
|
XS_LUN(xs), XS_TGT(xs), XS_CHANNEL(xs));
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case RQCS_XACT_ERR3:
|
2001-07-04 18:42:41 +00:00
|
|
|
isp_prt(isp, ISP_LOGERR, xact3,
|
|
|
|
|
XS_CHANNEL(xs), XS_TGT(xs), XS_LUN(xs));
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case RQCS_QUEUE_FULL:
|
2001-10-23 23:05:20 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG0,
|
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));
|
2001-10-23 23:05:20 +00:00
|
|
|
|
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.
|
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)) {
|
1999-01-10 02:55:10 +00:00
|
|
|
XS_SETERR(xs, HBA_NOERROR);
|
|
|
|
|
return;
|
|
|
|
|
}
|
2001-10-23 23:05:20 +00:00
|
|
|
|
|
|
|
|
*
|
|
|
|
|
*
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
*XS_STSP(xs) = SCSI_QFULL;
|
|
|
|
|
XS_SETERR(xs, HBA_NOERROR);
|
|
|
|
|
return;
|
1998-09-15 08:42:56 +00:00
|
|
|
|
|
|
|
|
case RQCS_PHASE_SKIPPED:
|
2001-08-16 17:39:45 +00:00
|
|
|
isp_prt(isp, ISP_LOGERR, pskip, XS_CHANNEL(xs),
|
2001-08-16 17:25:08 +00:00
|
|
|
XS_TGT(xs), XS_LUN(xs));
|
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);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
|
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));
|
1999-05-11 05:06:55 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1998-09-15 08:42:56 +00:00
|
|
|
sdparam *sdp = isp->isp_param;
|
1999-05-11 05:06:55 +00:00
|
|
|
sdp += XS_CHANNEL(xs);
|
2001-07-30 00:59:06 +00:00
|
|
|
sdp->isp_devparam[XS_TGT(xs)].goal_flags &= ~DPARM_WIDE;
|
1999-01-30 07:29:00 +00:00
|
|
|
sdp->isp_devparam[XS_TGT(xs)].dev_update = 1;
|
2000-02-29 05:52:14 +00:00
|
|
|
isp->isp_update |= (1 << XS_CHANNEL(xs));
|
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);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
|
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));
|
1999-05-11 05:06:55 +00:00
|
|
|
if (IS_SCSI(isp)) {
|
1998-09-15 08:42:56 +00:00
|
|
|
sdparam *sdp = isp->isp_param;
|
1999-05-11 05:06:55 +00:00
|
|
|
sdp += XS_CHANNEL(xs);
|
2001-07-30 00:59:06 +00:00
|
|
|
sdp->isp_devparam[XS_TGT(xs)].goal_flags &= ~DPARM_SYNC;
|
1999-01-30 07:29:00 +00:00
|
|
|
sdp->isp_devparam[XS_TGT(xs)].dev_update = 1;
|
2000-02-29 05:52:14 +00:00
|
|
|
isp->isp_update |= (1 << XS_CHANNEL(xs));
|
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));
|
1998-09-15 08:42:56 +00:00
|
|
|
break;
|
|
|
|
|
|
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:
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
|
|
|
|
* It was there (maybe)- treat as a selection timeout.
|
|
|
|
|
*/
|
2002-03-21 21:10:16 +00:00
|
|
|
if ((sp->req_completion_status & 0xff) == RQCS_PORT_UNAVAILABLE)
|
|
|
|
|
isp_prt(isp, ISP_LOGINFO,
|
2002-05-01 21:58:36 +00:00
|
|
|
"port unavailable for target %d", XS_TGT(xs));
|
2002-03-21 21:10:16 +00:00
|
|
|
else
|
|
|
|
|
isp_prt(isp, ISP_LOGINFO,
|
|
|
|
|
"port logout for target %d", XS_TGT(xs));
|
2002-02-21 01:56:08 +00:00
|
|
|
/*
|
|
|
|
|
* If we're on a local loop, force a LIP (which is overkill)
|
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.
|
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;
|
|
|
|
|
mbs.param[0] = MBOX_INIT_LIP;
|
|
|
|
|
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);
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
return;
|
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);
|
|
|
|
|
}
|
2000-02-29 05:52:14 +00:00
|
|
|
return;
|
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);
|
|
|
|
|
}
|
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);
|
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);
|
|
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_fastpost_complete(struct ispsoftc *isp, u_int16_t fph)
|
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;
|
1999-03-17 05:04:39 +00:00
|
|
|
|
2001-03-02 06:28:55 +00:00
|
|
|
if (fph == 0) {
|
1999-03-17 05:04:39 +00:00
|
|
|
return;
|
1999-10-17 18:58:22 +00:00
|
|
|
}
|
|
|
|
|
xs = isp_find_xs(isp, fph);
|
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);
|
1999-03-17 05:04:39 +00:00
|
|
|
return;
|
|
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
isp_destroy_handle(isp, fph);
|
|
|
|
|
|
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);
|
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;
|
1999-03-17 05:04:39 +00:00
|
|
|
if (XS_XFRLEN(xs)) {
|
1999-10-17 18:58:22 +00:00
|
|
|
ISP_DMAFREE(isp, xs, fph);
|
1999-03-17 05:04:39 +00:00
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
if (isp->isp_nactive)
|
|
|
|
|
isp->isp_nactive--;
|
2002-01-03 20:43:22 +00:00
|
|
|
isp->isp_fphccmplt++;
|
2000-08-27 23:38:44 +00:00
|
|
|
isp_done(xs);
|
1999-03-17 05:04:39 +00:00
|
|
|
}
|
|
|
|
|
|
2002-02-04 21:04:25 +00:00
|
|
|
static int
|
|
|
|
|
isp_mbox_continue(struct ispsoftc *isp)
|
|
|
|
|
{
|
|
|
|
|
mbreg_t mbs;
|
|
|
|
|
u_int16_t *ptr;
|
|
|
|
|
|
2002-02-17 06:38:22 +00:00
|
|
|
switch (isp->isp_lastmbxcmd) {
|
|
|
|
|
case MBOX_WRITE_RAM_WORD:
|
|
|
|
|
case MBOX_READ_RAM_WORD:
|
|
|
|
|
case MBOX_READ_RAM_WORD_EXTENDED:
|
|
|
|
|
break;
|
|
|
|
|
default:
|
2002-02-04 21:04:25 +00:00
|
|
|
return (1);
|
|
|
|
|
}
|
|
|
|
|
if (isp->isp_mboxtmp[0] != MBOX_COMMAND_COMPLETE) {
|
|
|
|
|
isp->isp_mbxwrk0 = 0;
|
|
|
|
|
return (-1);
|
|
|
|
|
}
|
|
|
|
|
|
2002-02-17 06:38:22 +00:00
|
|
|
|
2002-02-04 21:04:25 +00:00
|
|
|
/*
|
|
|
|
|
* Clear the previous interrupt.
|
|
|
|
|
*/
|
|
|
|
|
ISP_WRITE(isp, HCCR, HCCR_CMD_CLEAR_RISC_INT);
|
|
|
|
|
ISP_WRITE(isp, BIU_SEMA, 0);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Continue with next word.
|
|
|
|
|
*/
|
|
|
|
|
ptr = isp->isp_mbxworkp;
|
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;
|
|
|
|
|
case MBOX_READ_RAM_WORD:
|
|
|
|
|
case MBOX_READ_RAM_WORD_EXTENDED:
|
|
|
|
|
*ptr++ = isp->isp_mboxtmp[2];
|
|
|
|
|
mbs.param[1] = isp->isp_mbxwrk1++;
|
|
|
|
|
break;
|
|
|
|
|
}
|
2002-02-04 21:04:25 +00:00
|
|
|
isp->isp_mbxworkp = ptr;
|
2002-02-17 06:38:22 +00:00
|
|
|
mbs.param[0] = isp->isp_lastmbxcmd;
|
2002-02-04 21:04:25 +00:00
|
|
|
isp->isp_mbxwrk0 -= 1;
|
2002-02-21 01:56:08 +00:00
|
|
|
isp_mboxcmd_qnw(isp, &mbs, 0);
|
2002-02-04 21:04:25 +00:00
|
|
|
return (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
|
|
|
#define HIBYT(x) ((x) >> 0x8)
|
|
|
|
|
#define LOBYT(x) ((x) & 0xff)
|
|
|
|
|
#define ISPOPMAP(a, b) (((a) << 8) | (b))
|
|
|
|
|
static u_int16_t mbpscsi[] = {
|
|
|
|
|
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 */
|
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 */
|
2002-09-23 04:59:42 +00:00
|
|
|
ISPOPMAP(0xdf, 0xff), /* 0x52: INITIALIZE REQUEST QUEUE A64 */
|
|
|
|
|
ISPOPMAP(0xef, 0xff), /* 0x53: INITIALIZE RESPONSE QUEUE A64 */
|
2001-10-23 23:05:20 +00:00
|
|
|
ISPOPMAP(0xcf, 0x01), /* 0x54: EXECUTE IOCB A64 */
|
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 */
|
|
|
|
|
};
|
|
|
|
|
|
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",
|
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
|
|
|
|
|
|
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
|
|
|
static u_int16_t mbpfc[] = {
|
|
|
|
|
ISPOPMAP(0x01, 0x01), /* 0x00: MBOX_NO_OP */
|
|
|
|
|
ISPOPMAP(0x1f, 0x01), /* 0x01: MBOX_LOAD_RAM */
|
|
|
|
|
ISPOPMAP(0x03, 0x01), /* 0x02: MBOX_EXEC_FIRMWARE */
|
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 */
|
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: */
|
|
|
|
|
ISPOPMAP(0x00, 0x00), /* 0x0d: */
|
|
|
|
|
ISPOPMAP(0x01, 0x05), /* 0x0e: MBOX_CHECK_FIRMWARE */
|
2002-02-17 06:38:22 +00:00
|
|
|
ISPOPMAP(0x03, 0x07), /* 0x0f: MBOX_READ_RAM_WORD_EXTENDED(1) */
|
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: */
|
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: */
|
2002-05-01 21:58:36 +00:00
|
|
|
ISPOPMAP(0x03, 0x01), /* 0x5b: MBOX_DRIVER_HEARTBEAT */
|
|
|
|
|
ISPOPMAP(0xcf, 0x01), /* 0x5c: MBOX_FW_HEARTBEAT */
|
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: */
|
|
|
|
|
ISPOPMAP(0xfd, 0x31), /* 0x60: MBOX_INIT_FIRMWARE */
|
|
|
|
|
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 */
|
|
|
|
|
ISPOPMAP(0x07, 0x01) /* 0x7e: Lun RESET */
|
1998-04-22 17:54:58 +00:00
|
|
|
};
|
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
|
|
|
|
|
* responsible for setting this register first (Gross!).
|
|
|
|
|
*/
|
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",
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
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,
|
|
|
|
|
"GET PORT DATABASE,, ENHANCED",
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
"EXECUTE IOCB A64",
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
|
|
|
|
NULL,
|
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
|
|
|
|
|
|
2002-02-04 21:04:25 +00:00
|
|
|
static void
|
2002-02-21 01:56:08 +00:00
|
|
|
isp_mboxcmd_qnw(struct ispsoftc *isp, mbreg_t *mbp, int nodelay)
|
2002-02-04 21:04:25 +00:00
|
|
|
{
|
|
|
|
|
unsigned int lim, ibits, obits, box, opcode;
|
|
|
|
|
u_int16_t *mcp;
|
|
|
|
|
|
|
|
|
|
if (IS_FC(isp)) {
|
|
|
|
|
mcp = mbpfc;
|
|
|
|
|
lim = (sizeof (mbpfc) / sizeof (mbpfc[0]));
|
|
|
|
|
} else {
|
|
|
|
|
mcp = mbpscsi;
|
|
|
|
|
lim = (sizeof (mbpscsi) / sizeof (mbpscsi[0]));
|
|
|
|
|
}
|
|
|
|
|
opcode = mbp->param[0];
|
|
|
|
|
ibits = HIBYT(mcp[opcode]) & NMBOX_BMASK(isp);
|
|
|
|
|
obits = LOBYT(mcp[opcode]) & NMBOX_BMASK(isp);
|
|
|
|
|
for (box = 0; box < MAX_MAILBOX; box++) {
|
|
|
|
|
if (ibits & (1 << box)) {
|
|
|
|
|
ISP_WRITE(isp, MBOX_OFF(box), mbp->param[box]);
|
|
|
|
|
}
|
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;
|
2002-02-04 21:04:25 +00:00
|
|
|
}
|
|
|
|
|
ISP_WRITE(isp, HCCR, HCCR_CMD_SET_HOST_INT);
|
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);
|
|
|
|
|
}
|
2002-02-04 21:04:25 +00:00
|
|
|
}
|
|
|
|
|
|
1998-04-22 17:54:58 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_mboxcmd(struct ispsoftc *isp, mbreg_t *mbp, int logmask)
|
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;
|
|
|
|
|
u_int16_t *mcp;
|
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]));
|
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;
|
1998-12-05 01:33:57 +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
|
|
|
ibits = HIBYT(mcp[opcode]) & NMBOX_BMASK(isp);
|
|
|
|
|
obits = LOBYT(mcp[opcode]) & NMBOX_BMASK(isp);
|
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;
|
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);
|
|
|
|
|
|
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
|
|
|
for (box = 0; box < MAX_MAILBOX; box++) {
|
|
|
|
|
if (ibits & (1 << box)) {
|
|
|
|
|
ISP_WRITE(isp, MBOX_OFF(box), mbp->param[box]);
|
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;
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
2001-01-15 18:33:08 +00:00
|
|
|
isp->isp_lastmbxcmd = opcode;
|
|
|
|
|
|
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.
|
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;
|
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);
|
|
|
|
|
|
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.
|
1999-03-17 05:04:39 +00:00
|
|
|
*/
|
2000-07-04 01:02:38 +00:00
|
|
|
MBOX_WAIT_COMPLETE(isp);
|
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
|
|
|
if (isp->isp_mboxbsy) {
|
|
|
|
|
/*
|
|
|
|
|
* Command timed out.
|
|
|
|
|
*/
|
|
|
|
|
isp->isp_mboxbsy = 0;
|
|
|
|
|
MBOX_RELEASE(isp);
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
|
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.
|
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
|
|
|
for (box = 0; box < MAX_MAILBOX; box++) {
|
|
|
|
|
if (obits & (1 << box)) {
|
|
|
|
|
mbp->param[box] = isp->isp_mboxtmp[box];
|
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) {
|
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
|
|
|
}
|
|
|
|
|
|
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;
|
1999-03-25 22:52:45 +00:00
|
|
|
switch (mbp->param[0]) {
|
1998-04-22 17:54:58 +00:00
|
|
|
case MBOX_COMMAND_COMPLETE:
|
|
|
|
|
break;
|
|
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_COMMAND_COMPLETE))
|
|
|
|
|
xname = "INVALID COMMAND";
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_HOST_INTERFACE_ERROR))
|
|
|
|
|
xname = "HOST INTERFACE ERROR";
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_TEST_FAILED))
|
|
|
|
|
xname = "TEST FAILED";
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_COMMAND_ERROR))
|
|
|
|
|
xname = "COMMAND ERROR";
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
|
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_COMMAND_PARAM_ERROR))
|
|
|
|
|
xname = "COMMAND PARAMETER ERROR";
|
1998-04-22 17:54:58 +00:00
|
|
|
break;
|
2000-02-29 05:52:14 +00:00
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_LOOP_ID_USED))
|
|
|
|
|
xname = "LOOP ID ALREADY IN USE";
|
|
|
|
|
break;
|
2000-02-29 05:52:14 +00:00
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_PORT_ID_USED))
|
|
|
|
|
xname = "PORT ID ALREADY IN USE";
|
|
|
|
|
break;
|
2000-02-29 05:52:14 +00:00
|
|
|
case 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
|
|
|
if (logmask & MBLOGMASK(MBOX_ALL_IDS_USED))
|
|
|
|
|
xname = "ALL LOOP IDS IN USE";
|
|
|
|
|
break;
|
|
|
|
|
case 0: /* special case */
|
|
|
|
|
xname = "TIMEOUT";
|
2000-02-29 05:52:14 +00:00
|
|
|
break;
|
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;
|
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 (xname)
|
|
|
|
|
isp_prt(isp, ISP_LOGALL, "Mailbox Command '%s' failed (%s)",
|
|
|
|
|
cname, xname);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_fw_state(struct ispsoftc *isp)
|
1998-04-22 17:54:58 +00:00
|
|
|
{
|
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;
|
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
|
|
|
|
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);
|
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];
|
2001-02-11 03:44:43 +00:00
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_update(struct ispsoftc *isp)
|
1999-05-11 05:06:55 +00:00
|
|
|
{
|
2000-08-27 23:38:44 +00:00
|
|
|
int bus, upmask;
|
1999-05-11 05:06:55 +00:00
|
|
|
|
2000-08-27 23:38:44 +00:00
|
|
|
for (bus = 0, upmask = isp->isp_update; upmask != 0; bus++) {
|
|
|
|
|
if (upmask & (1 << bus)) {
|
1999-05-11 05:06:55 +00:00
|
|
|
isp_update_bus(isp, bus);
|
|
|
|
|
}
|
2000-08-27 23:38:44 +00:00
|
|
|
upmask &= ~(1 << bus);
|
1999-05-11 05:06:55 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_update_bus(struct ispsoftc *isp, int bus)
|
1998-04-22 17:54:58 +00:00
|
|
|
{
|
1998-09-15 08:42:56 +00:00
|
|
|
int tgt;
|
1998-04-22 17:54:58 +00:00
|
|
|
mbreg_t mbs;
|
|
|
|
|
sdparam *sdp;
|
|
|
|
|
|
2000-08-27 23:38:44 +00:00
|
|
|
isp->isp_update &= ~(1 << bus);
|
1999-08-16 19:59:55 +00:00
|
|
|
if (IS_FC(isp)) {
|
2000-08-27 23:38:44 +00:00
|
|
|
/*
|
|
|
|
|
* There are no 'per-bus' settings for Fibre Channel.
|
|
|
|
|
*/
|
1998-04-22 17:54:58 +00:00
|
|
|
return;
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
sdp = isp->isp_param;
|
1999-05-11 05:06:55 +00:00
|
|
|
sdp += bus;
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
for (tgt = 0; tgt < MAX_TARGETS; tgt++) {
|
1999-05-11 05:06:55 +00:00
|
|
|
u_int16_t flags, period, offset;
|
1999-01-30 07:29:00 +00:00
|
|
|
int get;
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
if (sdp->isp_devparam[tgt].dev_enable == 0) {
|
2000-08-27 23:38:44 +00:00
|
|
|
sdp->isp_devparam[tgt].dev_update = 0;
|
|
|
|
|
sdp->isp_devparam[tgt].dev_refresh = 0;
|
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);
|
1998-09-15 08:42:56 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
1999-03-25 22:52:45 +00:00
|
|
|
/*
|
|
|
|
|
* If the goal is to update the status of the device,
|
2001-07-30 00:59:06 +00:00
|
|
|
* take what's in goal_flags and try and set the device
|
1999-03-25 22:52:45 +00:00
|
|
|
* toward that. Otherwise, if we're just refreshing the
|
|
|
|
|
* current device state, get the current parameters.
|
|
|
|
|
*/
|
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) {
|
1999-01-30 07:29:00 +00:00
|
|
|
mbs.param[0] = MBOX_SET_TARGET_PARAMS;
|
2000-08-27 23:38:44 +00:00
|
|
|
/*
|
2001-07-30 00:59:06 +00:00
|
|
|
* Make sure goal_flags has "Renegotiate on Error"
|
2000-08-27 23:38:44 +00:00
|
|
|
* on and "Freeze Queue on Error" off.
|
|
|
|
|
*/
|
2001-07-30 00:59:06 +00:00
|
|
|
sdp->isp_devparam[tgt].goal_flags |= DPARM_RENEG;
|
|
|
|
|
sdp->isp_devparam[tgt].goal_flags &= ~DPARM_QFRZ;
|
2000-08-27 23:38:44 +00:00
|
|
|
|
2001-07-30 00:59:06 +00:00
|
|
|
mbs.param[2] = sdp->isp_devparam[tgt].goal_flags;
|
2001-07-04 18:42:41 +00:00
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
/*
|
2000-08-27 23:38:44 +00:00
|
|
|
* Insist that PARITY must be enabled
|
|
|
|
|
* if SYNC or WIDE is enabled.
|
1999-05-11 05:06:55 +00:00
|
|
|
*/
|
2000-08-27 23:38:44 +00:00
|
|
|
if ((mbs.param[2] & (DPARM_SYNC|DPARM_WIDE)) != 0) {
|
1999-05-11 05:06:55 +00:00
|
|
|
mbs.param[2] |= DPARM_PARITY;
|
|
|
|
|
}
|
2000-08-27 23:38:44 +00:00
|
|
|
|
|
|
|
|
if ((mbs.param[2] & DPARM_SYNC) == 0) {
|
|
|
|
|
mbs.param[3] = 0;
|
|
|
|
|
} else {
|
|
|
|
|
mbs.param[3] =
|
2001-07-30 00:59:06 +00:00
|
|
|
(sdp->isp_devparam[tgt].goal_offset << 8) |
|
|
|
|
|
(sdp->isp_devparam[tgt].goal_period);
|
2000-08-27 23:38:44 +00:00
|
|
|
}
|
1999-03-25 22:52:45 +00:00
|
|
|
/*
|
|
|
|
|
* A command completion later that has
|
2001-07-30 00:59:06 +00:00
|
|
|
* RQSTF_NEGOTIATION set can cause
|
2000-08-27 23:38:44 +00:00
|
|
|
* the dev_refresh/announce cycle also.
|
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.
|
|
|
|
|
*/
|
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);
|
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);
|
2000-08-27 23:38:44 +00:00
|
|
|
sdp->isp_devparam[tgt].dev_update = 0;
|
|
|
|
|
sdp->isp_devparam[tgt].dev_refresh = 1;
|
1999-01-30 07:29:00 +00:00
|
|
|
get = 0;
|
|
|
|
|
} else {
|
1998-09-15 08:42:56 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
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);
|
1999-01-30 07:29:00 +00:00
|
|
|
if (get == 0) {
|
1999-05-11 05:06:55 +00:00
|
|
|
isp->isp_sendmarker |= (1 << bus);
|
1999-01-30 07:29:00 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
flags = mbs.param[2];
|
|
|
|
|
period = mbs.param[3] & 0xff;
|
|
|
|
|
offset = mbs.param[3] >> 8;
|
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;
|
1999-05-11 05:06:55 +00:00
|
|
|
get = (bus << 16) | tgt;
|
|
|
|
|
(void) isp_async(isp, ISPASYNC_NEW_TGT_PARAMS, &get);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
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;
|
|
|
|
|
}
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
|
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
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_setdfltparm(struct ispsoftc *isp, int channel)
|
1998-09-15 08:42:56 +00:00
|
|
|
{
|
1999-03-25 22:52:45 +00:00
|
|
|
int tgt;
|
1998-09-15 08:42:56 +00:00
|
|
|
mbreg_t mbs;
|
1999-12-20 01:34:01 +00:00
|
|
|
sdparam *sdp;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
if (IS_FC(isp)) {
|
1998-09-15 08:42:56 +00:00
|
|
|
fcparam *fcp = (fcparam *) isp->isp_param;
|
2001-01-09 02:46:23 +00:00
|
|
|
int nvfail;
|
|
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
fcp += channel;
|
|
|
|
|
if (fcp->isp_gotdparms) {
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
fcp->isp_gotdparms = 1;
|
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;
|
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);
|
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;
|
|
|
|
|
#ifndef ISP_NO_FASTPOST_FC
|
|
|
|
|
fcp->isp_fwoptions |= ICBOPT_FAST_POST;
|
|
|
|
|
#endif
|
|
|
|
|
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
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
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.
|
1998-09-15 08:42:56 +00:00
|
|
|
*/
|
2001-01-09 02:46:23 +00:00
|
|
|
if ((isp->isp_confopts & ISP_CFG_NONVRAM) == 0) {
|
|
|
|
|
nvfail = isp_read_nvram(isp);
|
|
|
|
|
if (nvfail)
|
|
|
|
|
isp->isp_confopts |= ISP_CFG_NONVRAM;
|
|
|
|
|
} 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.
|
|
|
|
|
*/
|
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",
|
2001-01-09 02:46:23 +00:00
|
|
|
(u_int32_t) (DEFAULT_NODEWWN(isp) >> 32),
|
2002-02-17 06:38:22 +00:00
|
|
|
(u_int32_t) (DEFAULT_NODEWWN(isp) & 0xffffffff));
|
|
|
|
|
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;
|
|
|
|
|
}
|
|
|
|
|
if (isp->isp_confopts & ISP_CFG_OWNWWPN) {
|
|
|
|
|
isp_prt(isp, ISP_LOGCONFIG, "Using Port WWN 0x%08x%08x",
|
2001-01-09 02:46:23 +00:00
|
|
|
(u_int32_t) (DEFAULT_PORTWWN(isp) >> 32),
|
|
|
|
|
(u_int32_t) (DEFAULT_PORTWWN(isp) & 0xffffffff));
|
|
|
|
|
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;
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
|
1999-12-20 01:34:01 +00:00
|
|
|
sdp = (sdparam *) isp->isp_param;
|
|
|
|
|
sdp += channel;
|
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;
|
|
|
|
|
|
2001-04-04 21:42:59 +00:00
|
|
|
/*
|
|
|
|
|
* Establish some default parameters.
|
|
|
|
|
*/
|
2001-08-20 17:28:32 +00:00
|
|
|
sdp->isp_cmd_dma_burst_enable = 0;
|
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;
|
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;
|
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;
|
|
|
|
|
}
|
|
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
/*
|
|
|
|
|
* If we've not been told to avoid reading NVRAM, try and read it.
|
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.
|
1999-05-11 05:06:55 +00:00
|
|
|
*/
|
2001-07-30 00:59:06 +00:00
|
|
|
|
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.
|
|
|
|
|
*/
|
1999-12-20 01:34:01 +00:00
|
|
|
if ((isp->isp_confopts & ISP_CFG_NONVRAM) == 0) {
|
|
|
|
|
mbs.param[0] = MBOX_GET_ACT_NEG_STATE;
|
2001-04-04 21:42:59 +00:00
|
|
|
isp_mboxcmd(isp, &mbs, MBLOGNONE);
|
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;
|
1999-05-11 05:06:55 +00:00
|
|
|
}
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
1999-02-09 01:07:06 +00:00
|
|
|
|
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);
|
1999-12-20 01:34:01 +00:00
|
|
|
|
1999-03-25 22:52:45 +00:00
|
|
|
/*
|
|
|
|
|
* The trick here is to establish a default for the default (honk!)
|
2001-07-30 00:59:06 +00:00
|
|
|
* state (goal_flags). Then try and get the current status from
|
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++) {
|
2001-07-30 00:59:06 +00:00
|
|
|
u_int8_t off, per;
|
|
|
|
|
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;
|
|
|
|
|
|
2000-01-15 01:52:01 +00:00
|
|
|
/*
|
|
|
|
|
* We default to Wide/Fast for versions less than a 1040
|
|
|
|
|
* (unless it's SBus).
|
|
|
|
|
*/
|
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 &&
|
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) ||
|
2000-05-09 01:14:43 +00:00
|
|
|
(isp->isp_clock && isp->isp_clock < 60) ||
|
|
|
|
|
(sdp->isp_ultramode == 0)) {
|
2001-07-30 00:59:06 +00:00
|
|
|
off = ISP_10M_SYNCPARMS >> 8;
|
|
|
|
|
per = ISP_10M_SYNCPARMS & 0xff;
|
2000-08-27 23:38:44 +00:00
|
|
|
} else {
|
2001-07-30 00:59:06 +00:00
|
|
|
off = ISP_20M_SYNCPARMS_1040 >> 8;
|
|
|
|
|
per = ISP_20M_SYNCPARMS_1040 & 0xff;
|
1999-02-09 01:07:06 +00:00
|
|
|
}
|
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 =
|
2001-08-02 00:34:56 +00:00
|
|
|
sdp->isp_devparam[tgt].nvrm_period = per;
|
1999-02-09 01:07:06 +00:00
|
|
|
|
2001-08-20 17:28:32 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG0, sc2, sc3,
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
1999-03-25 22:52:45 +00:00
|
|
|
/*
|
1998-09-15 08:42:56 +00:00
|
|
|
* Re-initialize the ISP and complete all orphaned commands
|
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.
|
1998-09-15 08:42:56 +00:00
|
|
|
*
|
|
|
|
|
* Locks held prior to coming here.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_reinit(struct ispsoftc *isp)
|
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;
|
2001-03-04 18:41:23 +00:00
|
|
|
u_int16_t handle;
|
1998-04-22 17:54:58 +00:00
|
|
|
|
|
|
|
|
isp_reset(isp);
|
2001-02-11 03:44:43 +00:00
|
|
|
if (isp->isp_state != ISP_RESETSTATE) {
|
|
|
|
|
isp_prt(isp, ISP_LOGERR, "isp_reinit cannot reset card");
|
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");
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
1999-10-17 18:58:22 +00:00
|
|
|
isp->isp_nactive = 0;
|
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
|
|
|
for (handle = 1; (int) handle <= isp->isp_maxcmds; handle++) {
|
1999-10-17 18:58:22 +00:00
|
|
|
xs = isp_find_xs(isp, handle);
|
|
|
|
|
if (xs == NULL) {
|
1998-04-22 17:54:58 +00:00
|
|
|
continue;
|
1998-09-17 21:03:45 +00:00
|
|
|
}
|
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;
|
|
|
|
|
}
|
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);
|
1998-04-22 17:54:58 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
1998-09-15 08:42:56 +00:00
|
|
|
/*
|
|
|
|
|
* NVRAM Routines
|
|
|
|
|
*/
|
|
|
|
|
static int
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_read_nvram(struct ispsoftc *isp)
|
1998-09-15 08:42:56 +00:00
|
|
|
{
|
|
|
|
|
int i, amt;
|
|
|
|
|
u_int8_t csum, minversion;
|
|
|
|
|
union {
|
|
|
|
|
u_int8_t _x[ISP2100_NVRAM_SIZE];
|
|
|
|
|
u_int16_t _s[ISP2100_NVRAM_SIZE>>1];
|
|
|
|
|
} _n;
|
|
|
|
|
#define nvram_data _n._x
|
|
|
|
|
#define nvram_words _n._s
|
|
|
|
|
|
1999-03-17 05:04:39 +00:00
|
|
|
if (IS_FC(isp)) {
|
1998-09-15 08:42:56 +00:00
|
|
|
amt = ISP2100_NVRAM_SIZE;
|
|
|
|
|
minversion = 1;
|
1999-12-16 05:42:02 +00:00
|
|
|
} else if (IS_ULTRA2(isp)) {
|
1999-05-11 05:06:55 +00:00
|
|
|
amt = ISP1080_NVRAM_SIZE;
|
|
|
|
|
minversion = 0;
|
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]);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
return (-1);
|
|
|
|
|
}
|
|
|
|
|
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");
|
1998-09-15 08:42:56 +00:00
|
|
|
return (-1);
|
|
|
|
|
}
|
|
|
|
|
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",
|
1998-09-15 08:42:56 +00:00
|
|
|
ISP_NVRAM_VERSION(nvram_data));
|
|
|
|
|
return (-1);
|
|
|
|
|
}
|
|
|
|
|
|
2000-02-11 19:31:32 +00:00
|
|
|
if (IS_ULTRA3(isp)) {
|
|
|
|
|
isp_parse_nvram_12160(isp, 0, nvram_data);
|
2002-10-11 17:28:01 +00:00
|
|
|
if (IS_12160(isp))
|
|
|
|
|
isp_parse_nvram_12160(isp, 1, nvram_data);
|
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);
|
1999-05-11 05:06:55 +00:00
|
|
|
} else if (IS_SCSI(isp)) {
|
2000-02-11 19:31:32 +00:00
|
|
|
isp_parse_nvram_1020(isp, nvram_data);
|
1998-09-15 08:42:56 +00:00
|
|
|
} else {
|
2000-02-11 19:31:32 +00:00
|
|
|
isp_parse_nvram_2100(isp, nvram_data);
|
1999-02-09 01:07:06 +00:00
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
return (0);
|
2000-02-11 19:31:32 +00:00
|
|
|
#undef nvram_data
|
|
|
|
|
#undef nvram_words
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_rdnvram_word(struct ispsoftc *isp, int wo, u_int16_t *rp)
|
1998-09-15 08:42:56 +00:00
|
|
|
{
|
|
|
|
|
int i, cbits;
|
|
|
|
|
u_int16_t bit, rqst;
|
|
|
|
|
|
|
|
|
|
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT|BIU_NVRAM_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
|
|
|
USEC_DELAY(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
|
1999-05-11 05:06:55 +00:00
|
|
|
if (IS_FC(isp)) {
|
1998-09-15 08:42:56 +00:00
|
|
|
wo &= ((ISP2100_NVRAM_SIZE >> 1) - 1);
|
2002-02-04 21:04:25 +00:00
|
|
|
if (IS_2312(isp) && isp->isp_port) {
|
|
|
|
|
wo += 128;
|
|
|
|
|
}
|
1998-09-15 08:42:56 +00:00
|
|
|
rqst = (ISP_NVRAM_READ << 8) | wo;
|
1999-05-11 05:06:55 +00:00
|
|
|
cbits = 10;
|
1999-12-16 05:42:02 +00:00
|
|
|
} else if (IS_ULTRA2(isp)) {
|
1999-05-11 05:06:55 +00:00
|
|
|
wo &= ((ISP1080_NVRAM_SIZE >> 1) - 1);
|
|
|
|
|
rqst = (ISP_NVRAM_READ << 8) | wo;
|
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);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
ISP_WRITE(isp, BIU_NVRAM, bit | BIU_NVRAM_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
|
|
|
USEC_DELAY(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
ISP_WRITE(isp, BIU_NVRAM, bit);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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(2);
|
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++) {
|
|
|
|
|
u_int16_t rv;
|
|
|
|
|
*rp <<= 1;
|
|
|
|
|
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT|BIU_NVRAM_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
|
|
|
USEC_DELAY(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
rv = ISP_READ(isp, BIU_NVRAM);
|
|
|
|
|
if (rv & BIU_NVRAM_DATAIN) {
|
|
|
|
|
*rp |= 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
|
|
|
USEC_DELAY(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
ISP_WRITE(isp, BIU_NVRAM, BIU_NVRAM_SELECT);
|
Major whacking for core version 2.0. A major motivator for 2.0 and these
changes is that there's now a Solaris port of this driver, so some things
in the core version had to change (not much, but some).
In order, from the top.....:
A lot of error strings are gathered in one place at the head of the file.
This caused me to rewrite them to look consistent (with respect to
things like 'Port 0x%' and 'Target %d' and 'Loop ID 0x%x'.
The major mailbox function, isp_mboxcmd, now takes a third argument,
which is a mask that selectively says whether mailbox command failures
will be logged. This will substantially reduce a lot of spurious noise
from the driver.
At the first run through isp_reset we used to try and get the current
running firmware's revision by issuing a mailbox command. This would
invariably fail on alpha's with anything but a Qlogic 1040 since SRM
doesn't *start* the f/w on these cards. Instead, we now see whether we're
sitting ROM state before trying to get a running BIOS loaded f/w version.
All CFGPRINTF/PRINTF/IDPRINTF macros have been replaced with calls to
isp_prt. There are seperate print levels that can be independently
set (see ispvar.h), which include debugging, etc.
All SYS_DELAY macros are now USEC_DELAY macros. RQUEST_QUEUE_LEN and
RESULT_QUEUE_LEN now take ispsoftc as a parameter- the Fibre Channel
cards and the Ultra2/Ultra3 cards can have 16 bit request queue entry
indices, so we can make a 1024 entry index for them instead of the
256 entries we've had until now.
A major change it to fix isp_fclink_test to actually only wait the
delay of time specified in the microsecond argument being passed.
The problem has always been that a call to isp_mboxcmd to get he
current firmware state takes an unknown (sometimes long) amount of
time- this is if the firmware is busy doing PLOGIs while we ask
it what's up. So, up until now, the usdelay argument has been
a joke. The net effect has been that if you boot without being plugged
into a good loop or into a switch, you hang. Massively annonying, and
hard to fix because the actual time delta was impossible to know
from just guessing. Now, using the new GET_NANOTIME macros, a precise
and measured amount of USEC_DELAY calls are done so that only the
specified usecdelay is allowed to pass. This means that if the initial
startup of the firmware if followed by a call from isp_freebsd.c:isp_attach
to isp_control(isp, ISP_FCLINK_TEST, &tdelay) where tdelay is 2 * 1000000,
no more than two seconds will actually elapse before we leave concluding
that the cable is unhooked. Jeez. About time....
Change the ispscsicmd entry point to isp_start, and the XS_CMD_DONE
macro to a call to the platform supplied isp_done (sane naming).
Limit our size of request queue completions we'll look at at interrupt
time. Since we've increased the size of the Request Queue (and the
size of the Response Queue proportionally), let's not create an
interrupt stack overflow by having to keep a max completion list
(forw links are not an option because this is common code with
some platforms that don't have link space in their XS_T structures).
A limit of 32 is not unreasonable- I doubt there'd be even this many
request queue completions at a time- remember, most boards now use
fast posting for normal command completion instead of filling out
response queue entries.
In the isp_mboxcmd cleanup, also create an array of command
names so that "ABOUT FIRMWARE" can be printed instead of "CMD #8".
Remove the isp_lostcmd function- it's been deprecated for a while.
Remove 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(2);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
|
|
|
|
ISP_WRITE(isp, BIU_NVRAM, 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
|
|
|
USEC_DELAY(2);
|
|
|
|
|
ISP_SWIZZLE_NVRAM_WORD(isp, rp);
|
1998-09-15 08:42:56 +00:00
|
|
|
}
|
2000-02-11 19:31:32 +00:00
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_parse_nvram_1020(struct ispsoftc *isp, u_int8_t *nvram_data)
|
2000-02-11 19:31:32 +00:00
|
|
|
{
|
|
|
|
|
sdparam *sdp = (sdparam *) isp->isp_param;
|
2001-08-20 17:28:32 +00:00
|
|
|
int tgt;
|
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);
|
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);
|
|
|
|
|
|
2000-08-27 23:38:44 +00:00
|
|
|
sdp->isp_fast_mttr = ISP_NVRAM_FAST_MTTR_ENABLE(nvram_data);
|
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);
|
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.
|
|
|
|
|
*/
|
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.
|
|
|
|
|
*/
|
2001-08-20 17:28:32 +00:00
|
|
|
if (sdp->isp_devparam[tgt].nvrm_period < 0x19) {
|
|
|
|
|
sdp->isp_devparam[tgt].nvrm_period = 0x19;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
2001-08-20 17:28:32 +00:00
|
|
|
if (sdp->isp_devparam[tgt].nvrm_offset > 0xc) {
|
|
|
|
|
sdp->isp_devparam[tgt].nvrm_offset = 0x0c;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
|
|
|
|
} else {
|
2001-08-20 17:28:32 +00:00
|
|
|
if (sdp->isp_devparam[tgt].nvrm_offset > 0x8) {
|
|
|
|
|
sdp->isp_devparam[tgt].nvrm_offset = 0x8;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
|
|
|
|
}
|
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;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_parse_nvram_1080(struct ispsoftc *isp, int bus, u_int8_t *nvram_data)
|
2000-02-11 19:31:32 +00:00
|
|
|
{
|
|
|
|
|
sdparam *sdp = (sdparam *) isp->isp_param;
|
2001-08-20 17:28:32 +00:00
|
|
|
int tgt;
|
|
|
|
|
|
2000-02-11 19:31:32 +00:00
|
|
|
sdp += bus;
|
|
|
|
|
|
2001-07-04 18:42:41 +00:00
|
|
|
sdp->isp_fifo_threshold =
|
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);
|
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 =
|
2001-08-20 17:28:32 +00:00
|
|
|
ISP1080_NVRAM_ASYNC_DATA_SETUP_TIME(nvram_data, bus);
|
2000-02-11 19:31:32 +00:00
|
|
|
|
|
|
|
|
sdp->isp_req_ack_active_neg =
|
2001-08-20 17:28:32 +00:00
|
|
|
ISP1080_NVRAM_REQ_ACK_ACTIVE_NEGATION(nvram_data, bus);
|
2000-02-11 19:31:32 +00:00
|
|
|
|
|
|
|
|
sdp->isp_data_line_active_neg =
|
2001-08-20 17:28:32 +00:00
|
|
|
ISP1080_NVRAM_DATA_LINE_ACTIVE_NEGATION(nvram_data, bus);
|
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);
|
|
|
|
|
|
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;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_parse_nvram_12160(struct ispsoftc *isp, int bus, u_int8_t *nvram_data)
|
2000-02-11 19:31:32 +00:00
|
|
|
{
|
|
|
|
|
sdparam *sdp = (sdparam *) isp->isp_param;
|
2001-08-20 17:28:32 +00:00
|
|
|
int tgt;
|
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);
|
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 =
|
2001-08-20 17:28:32 +00:00
|
|
|
ISP12160_NVRAM_ASYNC_DATA_SETUP_TIME(nvram_data, bus);
|
2000-02-11 19:31:32 +00:00
|
|
|
|
|
|
|
|
sdp->isp_req_ack_active_neg =
|
2001-08-20 17:28:32 +00:00
|
|
|
ISP12160_NVRAM_REQ_ACK_ACTIVE_NEGATION(nvram_data, bus);
|
2000-02-11 19:31:32 +00:00
|
|
|
|
|
|
|
|
sdp->isp_data_line_active_neg =
|
2001-08-20 17:28:32 +00:00
|
|
|
ISP12160_NVRAM_DATA_LINE_ACTIVE_NEGATION(nvram_data, bus);
|
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);
|
|
|
|
|
|
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;
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2001-03-14 04:11:56 +00:00
|
|
|
isp_parse_nvram_2100(struct ispsoftc *isp, u_int8_t *nvram_data)
|
2000-02-11 19:31:32 +00:00
|
|
|
{
|
|
|
|
|
fcparam *fcp = (fcparam *) 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
|
|
|
u_int64_t wwn;
|
2000-02-11 19:31:32 +00:00
|
|
|
|
|
|
|
|
/*
|
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.
|
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);
|
2000-10-12 23:49:09 +00:00
|
|
|
if (wwn) {
|
|
|
|
|
isp_prt(isp, ISP_LOGCONFIG, "NVRAM Port WWN 0x%08x%08x",
|
|
|
|
|
(u_int32_t) (wwn >> 32), (u_int32_t) (wwn & 0xffffffff));
|
|
|
|
|
if ((wwn >> 60) == 0) {
|
2001-07-04 18:42:41 +00:00
|
|
|
wwn |= (((u_int64_t) 2)<< 60);
|
2000-10-12 23:49:09 +00:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
fcp->isp_portwwn = wwn;
|
2002-02-04 21:04:25 +00:00
|
|
|
if (IS_2200(isp) || IS_23XX(isp)) {
|
|
|
|
|
wwn = ISP2200_NVRAM_NODE_NAME(nvram_data);
|
|
|
|
|
if (wwn) {
|
|
|
|
|
isp_prt(isp, ISP_LOGCONFIG, "NVRAM Node WWN 0x%08x%08x",
|
|
|
|
|
(u_int32_t) (wwn >> 32),
|
|
|
|
|
(u_int32_t) (wwn & 0xffffffff));
|
|
|
|
|
if ((wwn >> 60) == 0) {
|
|
|
|
|
wwn |= (((u_int64_t) 2)<< 60);
|
|
|
|
|
}
|
2000-05-09 01:14:43 +00:00
|
|
|
}
|
2002-02-04 21:04:25 +00:00
|
|
|
} else {
|
|
|
|
|
wwn &= ~((u_int64_t) 0xfff << 48);
|
2000-02-11 19:31:32 +00:00
|
|
|
}
|
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) {
|
|
|
|
|
if ((fcp->isp_nodewwn & (((u_int64_t) 0xfff) << 48)) != 0 &&
|
|
|
|
|
(fcp->isp_nodewwn >> 60) == 2) {
|
|
|
|
|
fcp->isp_nodewwn &= ~((u_int64_t) 0xfff << 48);
|
|
|
|
|
}
|
|
|
|
|
if ((fcp->isp_portwwn & (((u_int64_t) 0xfff) << 48)) == 0 &&
|
|
|
|
|
(fcp->isp_portwwn >> 60) == 2) {
|
|
|
|
|
fcp->isp_portwwn |= ((u_int64_t) 1 << 56);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2002-06-16 05:18:22 +00:00
|
|
|
isp_prt(isp, ISP_LOGDEBUG0,
|
|
|
|
|
"NVRAM: maxfrmlen %d execthrottle %d fwoptions 0x%x loopid %x",
|
|
|
|
|
ISP2100_NVRAM_MAXFRAMELENGTH(nvram_data),
|
|
|
|
|
ISP2100_NVRAM_EXECUTION_THROTTLE(nvram_data),
|
|
|
|
|
ISP2100_NVRAM_OPTIONS(nvram_data),
|
|
|
|
|
ISP2100_NVRAM_HARDLOOPID(nvram_data));
|
|
|
|
|
|
2000-02-11 19:31:32 +00:00
|
|
|
fcp->isp_maxalloc =
|
|
|
|
|
ISP2100_NVRAM_MAXIOCBALLOCATION(nvram_data);
|
2002-06-16 05:18:22 +00:00
|
|
|
if ((isp->isp_confopts & ISP_CFG_OWNFSZ) == 0)
|
|
|
|
|
fcp->isp_maxfrmlen =
|
|
|
|
|
ISP2100_NVRAM_MAXFRAMELENGTH(nvram_data);
|
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);
|
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);
|
2000-02-11 19:31:32 +00:00
|
|
|
fcp->isp_fwoptions = ISP2100_NVRAM_OPTIONS(nvram_data);
|
|
|
|
|
}
|
2002-02-17 06:38:22 +00:00
|
|
|
|
|
|
|
|
#ifdef ISP_FW_CRASH_DUMP
|
|
|
|
|
static void isp2200_fw_dump(struct ispsoftc *);
|
|
|
|
|
static void isp2300_fw_dump(struct ispsoftc *);
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
isp2200_fw_dump(struct ispsoftc *isp)
|
|
|
|
|
{
|
2002-02-18 00:00:34 +00:00
|
|
|
int i, j;
|
2002-02-17 06:38:22 +00:00
|
|
|
mbreg_t mbs;
|
|
|
|
|
u_int16_t *ptr;
|
|
|
|
|
|
|
|
|
|
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];
|
|
|
|
|
isp_prt(isp, ISP_LOGALL, "isp_fw_dump: SRAM dumped succesfully");
|
|
|
|
|
FCPARAM(isp)->isp_dump_data[0] = isp->isp_type; /* now used */
|
2002-08-17 17:29:15 +00:00
|
|
|
(void) isp_async(isp, ISPASYNC_FW_DUMPED, 0);
|
2002-02-17 06:38:22 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
isp2300_fw_dump(struct ispsoftc *isp)
|
|
|
|
|
{
|
2002-02-18 00:00:34 +00:00
|
|
|
int i, j;
|
2002-02-17 06:38:22 +00:00
|
|
|
mbreg_t mbs;
|
|
|
|
|
u_int16_t *ptr;
|
|
|
|
|
|
|
|
|
|
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);
|
|
|
|
|
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];
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* We don't have access to mailbox registers 8.. onward
|
|
|
|
|
* in our 'common' device model- so we have to set it
|
|
|
|
|
* here and hope it stays the same!
|
|
|
|
|
*/
|
|
|
|
|
ISP_WRITE(isp, PCI_MBOX_REGS2300_OFF + (8 << 1), 0x1);
|
|
|
|
|
|
|
|
|
|
mbs.param[0] = MBOX_READ_RAM_WORD_EXTENDED;
|
|
|
|
|
mbs.param[1] = 0;
|
|
|
|
|
isp->isp_mbxworkp = (void *) ptr;
|
|
|
|
|
isp->isp_mbxwrk0 = 0xffff; /* continuation count */
|
|
|
|
|
isp->isp_mbxwrk1 = 0x1; /* 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", 0x10000 + isp->isp_mbxwrk1);
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
ptr = isp->isp_mbxworkp; /* finish final word */
|
|
|
|
|
*ptr++ = mbs.param[2];
|
|
|
|
|
isp_prt(isp, ISP_LOGALL, "isp_fw_dump: SRAM dumped succesfully");
|
|
|
|
|
FCPARAM(isp)->isp_dump_data[0] = isp->isp_type; /* now used */
|
2002-08-17 17:29:15 +00:00
|
|
|
(void) isp_async(isp, ISPASYNC_FW_DUMPED, 0);
|
2002-02-17 06:38:22 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
|
isp_fw_dump(struct ispsoftc *isp)
|
|
|
|
|
{
|
|
|
|
|
if (IS_2200(isp))
|
|
|
|
|
isp2200_fw_dump(isp);
|
2002-02-18 00:00:34 +00:00
|
|
|
else if (IS_23XX(isp))
|
2002-02-17 06:38:22 +00:00
|
|
|
isp2300_fw_dump(isp);
|
|
|
|
|
}
|
|
|
|
|
#endif
|
