d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
b4110d4604
freebsd-skq/sys/dev/isp/isp_library.c

1726 lines
60 KiB
C
Raw Normal View History

Start each of the license/copyright comments with /*-, minor shuffle of lines
2005-01-06 01:43:34 +00:00
/*-
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
* Copyright (c) 1999-2006 by Matthew Jacob
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice immediately at the beginning of the file, without modification,
* this list of conditions, and the following disclaimer.
some copyright cleanups
2000-09-21 20:16:04 +00:00
* 2. The name of the author may not be used to endorse or promote products
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
/*
* Qlogic Host Adapter Internal Library Functions
*/
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
#ifdef __NetBSD__
#include <dev/ic/isp_netbsd.h>
#endif
#ifdef __FreeBSD__
Some rearrangement of headers to minimize diffs with outside of FreeBSD repository and to clean up the license header so as to not pollute the license with file function. Zero all mailbox structures prior to use (just in case). Change the outgoing mailbox count for INIT_FIRMWARE to be correct.
2006-07-16 20:11:50 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
#include <dev/isp/isp_freebsd.h>
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
#endif
#ifdef __OpenBSD__
#include <dev/ic/isp_openbsd.h>
#endif
#ifdef __linux__
#include "isp_linux.h"
#endif
#ifdef __svr4__
#include "isp_solaris.h"
#endif
'Support' for ISP SBus cards. This code does not imply that SBus cards work yet. They hang for me. But I can't netboot the latest snapshot on my ultra1e, and things hang at bus_setup_intr time. Since I'm offline for a while, I thought I'd toss this in in case somebody else who has a bit better luck wants to fart around with it. Please try and wait until I get back to check things in.
2002-07-11 03:25:04 +00:00
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
int
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_save_xs(ispsoftc_t *isp, XS_T *xs, uint16_t *handlep)
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
{
int i, j;
for (j = isp->isp_lasthdls, i = 0; i < (int) isp->isp_maxcmds; i++) {
if (isp->isp_xflist[j] == NULL) {
break;
}
if (++j == isp->isp_maxcmds) {
j = 0;
}
}
if (i == isp->isp_maxcmds) {
return (-1);
}
isp->isp_xflist[j] = xs;
*handlep = j+1;
if (++j == isp->isp_maxcmds)
j = 0;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp->isp_lasthdls = (uint16_t)j;
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
return (0);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
XS_T *
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_find_xs(ispsoftc_t *isp, uint16_t handle)
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if (handle < 1 || handle > (uint16_t) isp->isp_maxcmds) {
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
return (NULL);
} else {
return (isp->isp_xflist[handle - 1]);
}
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t
isp_find_handle(ispsoftc_t *isp, XS_T *xs)
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
{
int i;
if (xs != NULL) {
for (i = 0; i < isp->isp_maxcmds; i++) {
if (isp->isp_xflist[i] == xs) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
return ((uint16_t) i+1);
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
}
}
}
return (0);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
int
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_handle_index(uint16_t handle)
Add an isp_handle_index function- this is prepatory to loading more into the handle (i.e., generation number), so we will now need a function that will take a handle and return a flat index [ 0 .. maxhandles-1 ] for auxillary routines that need an index to get at buddy store values (like dma maps or xflist pointers).
2000-06-27 19:14:14 +00:00
{
return (handle-1);
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
isp_index_handle(int index)
{
return (index+1);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_destroy_handle(ispsoftc_t *isp, uint16_t handle)
Add a file where inline functions for the Qlogic isp cards can go. This reduces duplication in all the platform specific header files.
1999-10-17 18:15:05 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if (handle > 0 && handle <= (uint16_t) isp->isp_maxcmds) {
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
isp->isp_xflist[handle - 1] = NULL;
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
int
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_getrqentry(ispsoftc_t *isp, uint16_t *iptrp,
uint16_t *optrp, void **resultp)
Add a isp_getrqentry inline function (that is, find the next available request queue entry if any are left).
1999-11-21 03:14:04 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
volatile uint16_t iptr, optr;
Add a isp_getrqentry inline function (that is, find the next available request queue entry if any are left).
1999-11-21 03:14:04 +00:00
Macroize request/response in/out queue pointer access.
2001-07-04 18:52:23 +00:00
optr = isp->isp_reqodx = READ_REQUEST_QUEUE_OUT_POINTER(isp);
Add a isp_getrqentry inline function (that is, find the next available request queue entry if any are left).
1999-11-21 03:14:04 +00:00
iptr = isp->isp_reqidx;
*resultp = ISP_QUEUE_ENTRY(isp->isp_rquest, iptr);
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
iptr = ISP_NXT_QENTRY(iptr, RQUEST_QUEUE_LEN(isp));
Add a isp_getrqentry inline function (that is, find the next available request queue entry if any are left).
1999-11-21 03:14:04 +00:00
if (iptr == optr) {
return (1);
}
Add an isp_handle_index function- this is prepatory to loading more into the handle (i.e., generation number), so we will now need a function that will take a handle and return a flat index [ 0 .. maxhandles-1 ] for auxillary routines that need an index to get at buddy store values (like dma maps or xflist pointers).
2000-06-27 19:14:14 +00:00
if (optrp)
*optrp = optr;
if (iptrp)
*iptrp = iptr;
Add a isp_getrqentry inline function (that is, find the next available request queue entry if any are left).
1999-11-21 03:14:04 +00:00
return (0);
}
add isp_print_qentry inline function
2000-01-03 22:14:24 +00:00
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
#define TBA (4 * (((QENTRY_LEN >> 2) * 3) + 1) + 1)
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_print_qentry(ispsoftc_t *isp, char *msg, int idx, void *arg)
add isp_print_qentry inline function
2000-01-03 22:14:24 +00:00
{
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
char buf[TBA];
add isp_print_qentry inline function
2000-01-03 22:14:24 +00:00
int amt, i, j;
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t *ptr = arg;
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
Fix isp_print_qentry to print all four lines- it's been broken for months.
2001-03-02 04:48:41 +00:00
isp_prt(isp, ISP_LOGALL, "%s index %d=>", msg, idx);
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
for (buf[0] = 0, amt = i = 0; i < 4; i++) {
buf[0] = 0;
Fix isp_print_qentry to print all four lines- it's been broken for months.
2001-03-02 04:48:41 +00:00
SNPRINTF(buf, TBA, " ");
add isp_print_qentry inline function
2000-01-03 22:14:24 +00:00
for (j = 0; j < (QENTRY_LEN >> 2); j++) {
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
SNPRINTF(buf, TBA, "%s %02x", buf, ptr[amt++] & 0xff);
}
Fix isp_print_qentry to print all four lines- it's been broken for months.
2001-03-02 04:48:41 +00:00
isp_prt(isp, ISP_LOGALL, buf);
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_print_bytes(ispsoftc_t *isp, char *msg, int amt, void *arg)
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
{
char buf[128];
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t *ptr = arg;
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
int off;
various fixes
2000-08-27 23:38:44 +00:00
if (msg)
isp_prt(isp, ISP_LOGALL, "%s:", msg);
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
off = 0;
various fixes
2000-08-27 23:38:44 +00:00
buf[0] = 0;
Remove isp_prtstst (now in case statement in isp.c). Remove isp2100_fw_statename as an INLINE (now a function in isp.c). Remove isp2100_pdb_statename (unused). Redo all ISP_SCSI_XFER_T as XS_T types. Change all RQUEST_QUEUE_LEN/RESULT_QUEUE_LEN macros to take a parameter. Add isp_print_bytes function.
2000-08-01 06:26:04 +00:00
while (off < amt) {
int j, to;
to = off;
for (j = 0; j < 16; j++) {
SNPRINTF(buf, 128, "%s %02x", buf, ptr[off++] & 0xff);
if (off == amt)
break;
add isp_print_qentry inline function
2000-01-03 22:14:24 +00:00
}
various fixes
2000-08-27 23:38:44 +00:00
isp_prt(isp, ISP_LOGALL, "0x%08x:%s", to, buf);
buf[0] = 0;
add isp_print_qentry inline function
2000-01-03 22:14:24 +00:00
}
}
Add isp_fc_runstate function- this function's purpose is to, in stages, and depending on role, make sure link is up, scan the fabric (if we're connected to a fabric), scan the local loop (if appropriate), merge the results into the local port database then, check once again to make sure we have f/w at FW_READY state and the the loopstate is LOOP_READY.
2001-02-11 03:52:04 +00:00
/*
* Do the common path to try and ensure that link is up, we've scanned
* the fabric (if we're on a fabric), and that we've synchronized this
* all with our own database and done the appropriate logins.
*
* We repeatedly check for firmware state and loop state after each
* action because things may have changed while we were doing this.
* Any failure or change of state causes us to return a nonzero value.
*
* We honor HBA roles in that if we're not in Initiator mode, we don't
* attempt to sync up the database (that's for somebody else to do,
* if ever).
*
* We assume we enter here with any locks held.
*/
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
int
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_fc_runstate(ispsoftc_t *isp, int tval)
Add isp_fc_runstate function- this function's purpose is to, in stages, and depending on role, make sure link is up, scan the fabric (if we're connected to a fabric), scan the local loop (if appropriate), merge the results into the local port database then, check once again to make sure we have f/w at FW_READY state and the the loopstate is LOOP_READY.
2001-02-11 03:52:04 +00:00
{
fcparam *fcp;
int *tptr;
Add ISP_FC_GETHINFO ioctl. MFC after: 1 week
2002-06-16 04:59:30 +00:00
if (IS_SCSI(isp))
Add isp_fc_runstate function- this function's purpose is to, in stages, and depending on role, make sure link is up, scan the fabric (if we're connected to a fabric), scan the local loop (if appropriate), merge the results into the local port database then, check once again to make sure we have f/w at FW_READY state and the the loopstate is LOOP_READY.
2001-02-11 03:52:04 +00:00
return (0);
Redo some code based upon issues found by Coverity.
2006-04-21 18:46:35 +00:00
tptr = &tval;
Add isp_fc_runstate function- this function's purpose is to, in stages, and depending on role, make sure link is up, scan the fabric (if we're connected to a fabric), scan the local loop (if appropriate), merge the results into the local port database then, check once again to make sure we have f/w at FW_READY state and the the loopstate is LOOP_READY.
2001-02-11 03:52:04 +00:00
if (isp_control(isp, ISPCTL_FCLINK_TEST, tptr) != 0) {
return (-1);
}
fcp = FCPARAM(isp);
if (fcp->isp_fwstate != FW_READY || fcp->isp_loopstate < LOOP_PDB_RCVD)
return (-1);
if (isp_control(isp, ISPCTL_SCAN_FABRIC, NULL) != 0) {
return (-1);
}
if (isp_control(isp, ISPCTL_SCAN_LOOP, NULL) != 0) {
return (-1);
}
if ((isp->isp_role & ISP_ROLE_INITIATOR) == 0) {
return (0);
}
if (isp_control(isp, ISPCTL_PDB_SYNC, NULL) != 0) {
return (-1);
}
if (fcp->isp_fwstate != FW_READY || fcp->isp_loopstate != LOOP_READY) {
return (-1);
}
return (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
/*
* Functions to move stuff to a form that the QLogic RISC engine understands
* and functions to move stuff back to a form the processor understands.
*
* Each platform is required to provide the 8, 16 and 32 bit
* swizzle and unswizzle macros (ISP_IOX{PUT|GET}_{8,16,32})
*
* The assumption is that swizzling and unswizzling is mostly done 'in place'
* (with a few exceptions for efficiency).
*/
#define ISP_IS_SBUS(isp) \
(ISP_SBUS_SUPPORTED && (isp)->isp_bustype == ISP_BT_SBUS)
/*
* Swizzle/Copy Functions
*/
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_copy_out_hdr(ispsoftc_t *isp, isphdr_t *hpsrc, isphdr_t *hpdst)
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_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, hpsrc->rqs_entry_type,
&hpdst->rqs_entry_count);
ISP_IOXPUT_8(isp, hpsrc->rqs_entry_count,
&hpdst->rqs_entry_type);
ISP_IOXPUT_8(isp, hpsrc->rqs_seqno,
&hpdst->rqs_flags);
ISP_IOXPUT_8(isp, hpsrc->rqs_flags,
&hpdst->rqs_seqno);
} else {
ISP_IOXPUT_8(isp, hpsrc->rqs_entry_type,
&hpdst->rqs_entry_type);
ISP_IOXPUT_8(isp, hpsrc->rqs_entry_count,
&hpdst->rqs_entry_count);
ISP_IOXPUT_8(isp, hpsrc->rqs_seqno,
&hpdst->rqs_seqno);
ISP_IOXPUT_8(isp, hpsrc->rqs_flags,
&hpdst->rqs_flags);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_copy_in_hdr(ispsoftc_t *isp, isphdr_t *hpsrc, isphdr_t *hpdst)
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_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &hpsrc->rqs_entry_type,
hpdst->rqs_entry_count);
ISP_IOXGET_8(isp, &hpsrc->rqs_entry_count,
hpdst->rqs_entry_type);
ISP_IOXGET_8(isp, &hpsrc->rqs_seqno,
hpdst->rqs_flags);
ISP_IOXGET_8(isp, &hpsrc->rqs_flags,
hpdst->rqs_seqno);
} else {
ISP_IOXGET_8(isp, &hpsrc->rqs_entry_type,
hpdst->rqs_entry_type);
ISP_IOXGET_8(isp, &hpsrc->rqs_entry_count,
hpdst->rqs_entry_count);
ISP_IOXGET_8(isp, &hpsrc->rqs_seqno,
hpdst->rqs_seqno);
ISP_IOXGET_8(isp, &hpsrc->rqs_flags,
hpdst->rqs_flags);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
int
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_response_type(ispsoftc_t *isp, isphdr_t *hp)
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
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint8_t type;
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_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &hp->rqs_entry_count, type);
} else {
ISP_IOXGET_8(isp, &hp->rqs_entry_type, type);
}
return ((int)type);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_request(ispsoftc_t *isp, ispreq_t *rqsrc, ispreq_t *rqdst)
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
{
int i;
isp_copy_out_hdr(isp, &rqsrc->req_header, &rqdst->req_header);
ISP_IOXPUT_32(isp, rqsrc->req_handle, &rqdst->req_handle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, rqsrc->req_lun_trn, &rqdst->req_target);
ISP_IOXPUT_8(isp, rqsrc->req_target, &rqdst->req_lun_trn);
} else {
ISP_IOXPUT_8(isp, rqsrc->req_lun_trn, &rqdst->req_lun_trn);
ISP_IOXPUT_8(isp, rqsrc->req_target, &rqdst->req_target);
}
ISP_IOXPUT_16(isp, rqsrc->req_cdblen, &rqdst->req_cdblen);
ISP_IOXPUT_16(isp, rqsrc->req_flags, &rqdst->req_flags);
ISP_IOXPUT_16(isp, rqsrc->req_time, &rqdst->req_time);
ISP_IOXPUT_16(isp, rqsrc->req_seg_count, &rqdst->req_seg_count);
for (i = 0; i < 12; i++) {
ISP_IOXPUT_8(isp, rqsrc->req_cdb[i], &rqdst->req_cdb[i]);
}
for (i = 0; i < ISP_RQDSEG; i++) {
ISP_IOXPUT_32(isp, rqsrc->req_dataseg[i].ds_base,
&rqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, rqsrc->req_dataseg[i].ds_count,
&rqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_request_t2(ispsoftc_t *isp, ispreqt2_t *tqsrc, ispreqt2_t *tqdst)
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
{
int i;
isp_copy_out_hdr(isp, &tqsrc->req_header, &tqdst->req_header);
ISP_IOXPUT_32(isp, tqsrc->req_handle, &tqdst->req_handle);
ISP_IOXPUT_8(isp, tqsrc->req_lun_trn, &tqdst->req_lun_trn);
ISP_IOXPUT_8(isp, tqsrc->req_target, &tqdst->req_target);
ISP_IOXPUT_16(isp, tqsrc->req_scclun, &tqdst->req_scclun);
ISP_IOXPUT_16(isp, tqsrc->req_flags, &tqdst->req_flags);
ISP_IOXPUT_16(isp, tqsrc->_res2, &tqdst->_res2);
ISP_IOXPUT_16(isp, tqsrc->req_time, &tqdst->req_time);
ISP_IOXPUT_16(isp, tqsrc->req_seg_count, &tqdst->req_seg_count);
for (i = 0; i < 16; i++) {
ISP_IOXPUT_8(isp, tqsrc->req_cdb[i], &tqdst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, tqsrc->req_totalcnt, &tqdst->req_totalcnt);
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_base,
&tqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_count,
&tqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_request_t2e(ispsoftc_t *isp, ispreqt2e_t *tqsrc, ispreqt2e_t *tqdst)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
{
int i;
isp_copy_out_hdr(isp, &tqsrc->req_header, &tqdst->req_header);
ISP_IOXPUT_32(isp, tqsrc->req_handle, &tqdst->req_handle);
ISP_IOXPUT_16(isp, tqsrc->req_target, &tqdst->req_target);
ISP_IOXPUT_16(isp, tqsrc->req_scclun, &tqdst->req_scclun);
ISP_IOXPUT_16(isp, tqsrc->req_flags, &tqdst->req_flags);
ISP_IOXPUT_16(isp, tqsrc->_res2, &tqdst->_res2);
ISP_IOXPUT_16(isp, tqsrc->req_time, &tqdst->req_time);
ISP_IOXPUT_16(isp, tqsrc->req_seg_count, &tqdst->req_seg_count);
for (i = 0; i < 16; i++) {
ISP_IOXPUT_8(isp, tqsrc->req_cdb[i], &tqdst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, tqsrc->req_totalcnt, &tqdst->req_totalcnt);
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_base,
&tqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_count,
&tqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_request_t3(ispsoftc_t *isp, ispreqt3_t *tqsrc, ispreqt3_t *tqdst)
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
{
int i;
isp_copy_out_hdr(isp, &tqsrc->req_header, &tqdst->req_header);
ISP_IOXPUT_32(isp, tqsrc->req_handle, &tqdst->req_handle);
ISP_IOXPUT_8(isp, tqsrc->req_lun_trn, &tqdst->req_lun_trn);
ISP_IOXPUT_8(isp, tqsrc->req_target, &tqdst->req_target);
ISP_IOXPUT_16(isp, tqsrc->req_scclun, &tqdst->req_scclun);
ISP_IOXPUT_16(isp, tqsrc->req_flags, &tqdst->req_flags);
ISP_IOXPUT_16(isp, tqsrc->_res2, &tqdst->_res2);
ISP_IOXPUT_16(isp, tqsrc->req_time, &tqdst->req_time);
ISP_IOXPUT_16(isp, tqsrc->req_seg_count, &tqdst->req_seg_count);
for (i = 0; i < 16; i++) {
ISP_IOXPUT_8(isp, tqsrc->req_cdb[i], &tqdst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, tqsrc->req_totalcnt, &tqdst->req_totalcnt);
Fix buglet in A64 CTIO3 structure copy.
2002-09-23 05:00:37 +00:00
for (i = 0; i < ISP_RQDSEG_T3; i++) {
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_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_base,
&tqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_basehi,
&tqdst->req_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_count,
&tqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_request_t3e(ispsoftc_t *isp, ispreqt3e_t *tqsrc, ispreqt3e_t *tqdst)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
{
int i;
isp_copy_out_hdr(isp, &tqsrc->req_header, &tqdst->req_header);
ISP_IOXPUT_32(isp, tqsrc->req_handle, &tqdst->req_handle);
ISP_IOXPUT_16(isp, tqsrc->req_target, &tqdst->req_target);
ISP_IOXPUT_16(isp, tqsrc->req_scclun, &tqdst->req_scclun);
ISP_IOXPUT_16(isp, tqsrc->req_flags, &tqdst->req_flags);
ISP_IOXPUT_16(isp, tqsrc->_res2, &tqdst->_res2);
ISP_IOXPUT_16(isp, tqsrc->req_time, &tqdst->req_time);
ISP_IOXPUT_16(isp, tqsrc->req_seg_count, &tqdst->req_seg_count);
for (i = 0; i < 16; i++) {
ISP_IOXPUT_8(isp, tqsrc->req_cdb[i], &tqdst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, tqsrc->req_totalcnt, &tqdst->req_totalcnt);
for (i = 0; i < ISP_RQDSEG_T3; i++) {
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_base,
&tqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_basehi,
&tqdst->req_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, tqsrc->req_dataseg[i].ds_count,
&tqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_extended_request(ispsoftc_t *isp, ispextreq_t *xqsrc,
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
ispextreq_t *xqdst)
{
int i;
isp_copy_out_hdr(isp, &xqsrc->req_header, &xqdst->req_header);
ISP_IOXPUT_32(isp, xqsrc->req_handle, &xqdst->req_handle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, xqsrc->req_lun_trn, &xqdst->req_target);
ISP_IOXPUT_8(isp, xqsrc->req_target, &xqdst->req_lun_trn);
} else {
ISP_IOXPUT_8(isp, xqsrc->req_lun_trn, &xqdst->req_lun_trn);
ISP_IOXPUT_8(isp, xqsrc->req_target, &xqdst->req_target);
}
ISP_IOXPUT_16(isp, xqsrc->req_cdblen, &xqdst->req_cdblen);
ISP_IOXPUT_16(isp, xqsrc->req_flags, &xqdst->req_flags);
ISP_IOXPUT_16(isp, xqsrc->req_time, &xqdst->req_time);
ISP_IOXPUT_16(isp, xqsrc->req_seg_count, &xqdst->req_seg_count);
for (i = 0; i < 44; i++) {
ISP_IOXPUT_8(isp, xqsrc->req_cdb[i], &xqdst->req_cdb[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_cont_req(ispsoftc_t *isp, ispcontreq_t *cqsrc, ispcontreq_t *cqdst)
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
{
int i;
isp_copy_out_hdr(isp, &cqsrc->req_header, &cqdst->req_header);
for (i = 0; i < ISP_CDSEG; i++) {
ISP_IOXPUT_32(isp, cqsrc->req_dataseg[i].ds_base,
&cqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, cqsrc->req_dataseg[i].ds_count,
&cqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_cont64_req(ispsoftc_t *isp, ispcontreq64_t *cqsrc,
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
ispcontreq64_t *cqdst)
{
int i;
isp_copy_out_hdr(isp, &cqsrc->req_header, &cqdst->req_header);
for (i = 0; i < ISP_CDSEG64; i++) {
ISP_IOXPUT_32(isp, cqsrc->req_dataseg[i].ds_base,
&cqdst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, cqsrc->req_dataseg[i].ds_basehi,
&cqdst->req_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, cqsrc->req_dataseg[i].ds_count,
&cqdst->req_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_response(ispsoftc_t *isp, ispstatusreq_t *spsrc,
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 *spdst)
{
int i;
isp_copy_in_hdr(isp, &spsrc->req_header, &spdst->req_header);
ISP_IOXGET_32(isp, &spsrc->req_handle, spdst->req_handle);
ISP_IOXGET_16(isp, &spsrc->req_scsi_status, spdst->req_scsi_status);
ISP_IOXGET_16(isp, &spsrc->req_completion_status,
spdst->req_completion_status);
ISP_IOXGET_16(isp, &spsrc->req_state_flags, spdst->req_state_flags);
ISP_IOXGET_16(isp, &spsrc->req_status_flags, spdst->req_status_flags);
ISP_IOXGET_16(isp, &spsrc->req_time, spdst->req_time);
ISP_IOXGET_16(isp, &spsrc->req_sense_len, spdst->req_sense_len);
ISP_IOXGET_32(isp, &spsrc->req_resid, spdst->req_resid);
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &spsrc->req_response[i],
spdst->req_response[i]);
}
for (i = 0; i < 32; i++) {
ISP_IOXGET_8(isp, &spsrc->req_sense_data[i],
spdst->req_sense_data[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_response_x(ispsoftc_t *isp, ispstatus_cont_t *cpsrc,
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
ispstatus_cont_t *cpdst)
{
int i;
isp_copy_in_hdr(isp, &cpsrc->req_header, &cpdst->req_header);
for (i = 0; i < 60; i++) {
ISP_IOXGET_8(isp, &cpsrc->req_sense_data[i],
cpdst->req_sense_data[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_rio2(ispsoftc_t *isp, isp_rio2_t *r2src, isp_rio2_t *r2dst)
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
{
int i;
isp_copy_in_hdr(isp, &r2src->req_header, &r2dst->req_header);
if (r2dst->req_header.rqs_seqno > 30)
r2dst->req_header.rqs_seqno = 30;
for (i = 0; i < r2dst->req_header.rqs_seqno; i++) {
ISP_IOXGET_16(isp, &r2src->req_handles[i],
r2dst->req_handles[i]);
}
while (i < 30) {
r2dst->req_handles[i++] = 0;
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_icb(ispsoftc_t *isp, isp_icb_t *Is, isp_icb_t *Id)
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
{
int i;
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, Is->icb_version, &Id->_reserved0);
ISP_IOXPUT_8(isp, Is->_reserved0, &Id->icb_version);
} else {
ISP_IOXPUT_8(isp, Is->icb_version, &Id->icb_version);
ISP_IOXPUT_8(isp, Is->_reserved0, &Id->_reserved0);
}
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_IOXPUT_16(isp, Is->icb_fwoptions, &Id->icb_fwoptions);
ISP_IOXPUT_16(isp, Is->icb_maxfrmlen, &Id->icb_maxfrmlen);
ISP_IOXPUT_16(isp, Is->icb_maxalloc, &Id->icb_maxalloc);
ISP_IOXPUT_16(isp, Is->icb_execthrottle, &Id->icb_execthrottle);
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, Is->icb_retry_count, &Id->icb_retry_delay);
ISP_IOXPUT_8(isp, Is->icb_retry_delay, &Id->icb_retry_count);
} else {
ISP_IOXPUT_8(isp, Is->icb_retry_count, &Id->icb_retry_count);
ISP_IOXPUT_8(isp, Is->icb_retry_delay, &Id->icb_retry_delay);
}
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
for (i = 0; i < 8; i++) {
ISP_IOXPUT_8(isp, Is->icb_portname[i], &Id->icb_portname[i]);
}
ISP_IOXPUT_16(isp, Is->icb_hardaddr, &Id->icb_hardaddr);
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, Is->icb_iqdevtype, &Id->icb_logintime);
ISP_IOXPUT_8(isp, Is->icb_logintime, &Id->icb_iqdevtype);
} else {
ISP_IOXPUT_8(isp, Is->icb_iqdevtype, &Id->icb_iqdevtype);
ISP_IOXPUT_8(isp, Is->icb_logintime, &Id->icb_logintime);
}
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
for (i = 0; i < 8; i++) {
ISP_IOXPUT_8(isp, Is->icb_nodename[i], &Id->icb_nodename[i]);
}
ISP_IOXPUT_16(isp, Is->icb_rqstout, &Id->icb_rqstout);
ISP_IOXPUT_16(isp, Is->icb_rspnsin, &Id->icb_rspnsin);
ISP_IOXPUT_16(isp, Is->icb_rqstqlen, &Id->icb_rqstqlen);
ISP_IOXPUT_16(isp, Is->icb_rsltqlen, &Id->icb_rsltqlen);
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, Is->icb_rqstaddr[i], &Id->icb_rqstaddr[i]);
}
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, Is->icb_respaddr[i], &Id->icb_respaddr[i]);
}
ISP_IOXPUT_16(isp, Is->icb_lunenables, &Id->icb_lunenables);
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, Is->icb_ccnt, &Id->icb_icnt);
ISP_IOXPUT_8(isp, Is->icb_icnt, &Id->icb_ccnt);
} else {
ISP_IOXPUT_8(isp, Is->icb_ccnt, &Id->icb_ccnt);
ISP_IOXPUT_8(isp, Is->icb_icnt, &Id->icb_icnt);
}
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_IOXPUT_16(isp, Is->icb_lunetimeout, &Id->icb_lunetimeout);
ISP_IOXPUT_16(isp, Is->icb_xfwoptions, &Id->icb_xfwoptions);
Fix some incorrectly swapped fields in an ICB. Access a PCI register with correct width. Obtained from: Dmitry Valeryevich Trikoz
2005-05-11 03:00:50 +00:00
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, Is->icb_racctimer, &Id->icb_idelaytimer);
ISP_IOXPUT_8(isp, Is->icb_idelaytimer, &Id->icb_racctimer);
} else {
ISP_IOXPUT_8(isp, Is->icb_racctimer, &Id->icb_racctimer);
ISP_IOXPUT_8(isp, Is->icb_idelaytimer, &Id->icb_idelaytimer);
}
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_IOXPUT_16(isp, Is->icb_zfwoptions, &Id->icb_zfwoptions);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_pdb(ispsoftc_t *isp, isp_pdb_t *src, isp_pdb_t *dst)
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
{
int i;
ISP_IOXGET_16(isp, &src->pdb_options, dst->pdb_options);
ISP_IOXGET_8(isp, &src->pdb_mstate, dst->pdb_mstate);
ISP_IOXGET_8(isp, &src->pdb_sstate, dst->pdb_sstate);
for (i = 0; i < 4; i++) {
ISP_IOXGET_8(isp, &src->pdb_hardaddr_bits[i],
dst->pdb_hardaddr_bits[i]);
}
for (i = 0; i < 4; i++) {
ISP_IOXGET_8(isp, &src->pdb_portid_bits[i],
dst->pdb_portid_bits[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->pdb_nodename[i], dst->pdb_nodename[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->pdb_portname[i], dst->pdb_portname[i]);
}
ISP_IOXGET_16(isp, &src->pdb_execthrottle, dst->pdb_execthrottle);
ISP_IOXGET_16(isp, &src->pdb_exec_count, dst->pdb_exec_count);
ISP_IOXGET_8(isp, &src->pdb_retry_count, dst->pdb_retry_count);
ISP_IOXGET_8(isp, &src->pdb_retry_delay, dst->pdb_retry_delay);
ISP_IOXGET_16(isp, &src->pdb_resalloc, dst->pdb_resalloc);
ISP_IOXGET_16(isp, &src->pdb_curalloc, dst->pdb_curalloc);
ISP_IOXGET_16(isp, &src->pdb_qhead, dst->pdb_qhead);
ISP_IOXGET_16(isp, &src->pdb_qtail, dst->pdb_qtail);
ISP_IOXGET_16(isp, &src->pdb_tl_next, dst->pdb_tl_next);
ISP_IOXGET_16(isp, &src->pdb_tl_last, dst->pdb_tl_last);
ISP_IOXGET_16(isp, &src->pdb_features, dst->pdb_features);
ISP_IOXGET_16(isp, &src->pdb_pconcurrnt, dst->pdb_pconcurrnt);
ISP_IOXGET_16(isp, &src->pdb_roi, dst->pdb_roi);
ISP_IOXGET_8(isp, &src->pdb_target, dst->pdb_target);
ISP_IOXGET_8(isp, &src->pdb_initiator, dst->pdb_initiator);
ISP_IOXGET_16(isp, &src->pdb_rdsiz, dst->pdb_rdsiz);
ISP_IOXGET_16(isp, &src->pdb_ncseq, dst->pdb_ncseq);
ISP_IOXGET_16(isp, &src->pdb_noseq, dst->pdb_noseq);
ISP_IOXGET_16(isp, &src->pdb_labrtflg, dst->pdb_labrtflg);
ISP_IOXGET_16(isp, &src->pdb_lstopflg, dst->pdb_lstopflg);
ISP_IOXGET_16(isp, &src->pdb_sqhead, dst->pdb_sqhead);
ISP_IOXGET_16(isp, &src->pdb_sqtail, dst->pdb_sqtail);
ISP_IOXGET_16(isp, &src->pdb_ptimer, dst->pdb_ptimer);
ISP_IOXGET_16(isp, &src->pdb_nxt_seqid, dst->pdb_nxt_seqid);
ISP_IOXGET_16(isp, &src->pdb_fcount, dst->pdb_fcount);
ISP_IOXGET_16(isp, &src->pdb_prli_len, dst->pdb_prli_len);
ISP_IOXGET_16(isp, &src->pdb_prli_svc0, dst->pdb_prli_svc0);
ISP_IOXGET_16(isp, &src->pdb_prli_svc3, dst->pdb_prli_svc3);
ISP_IOXGET_16(isp, &src->pdb_loopid, dst->pdb_loopid);
ISP_IOXGET_16(isp, &src->pdb_il_ptr, dst->pdb_il_ptr);
ISP_IOXGET_16(isp, &src->pdb_sl_ptr, dst->pdb_sl_ptr);
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
/*
* CT_HDR canonicalization- only needed for SNS responses
*/
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_ct_hdr(ispsoftc_t *isp, ct_hdr_t *src, ct_hdr_t *dst)
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
{
ISP_IOXGET_8(isp, &src->ct_revision, dst->ct_revision);
ISP_IOXGET_8(isp, &src->ct_portid[0], dst->ct_portid[0]);
ISP_IOXGET_8(isp, &src->ct_portid[1], dst->ct_portid[1]);
ISP_IOXGET_8(isp, &src->ct_portid[2], dst->ct_portid[2]);
ISP_IOXGET_8(isp, &src->ct_fcs_type, dst->ct_fcs_type);
ISP_IOXGET_8(isp, &src->ct_fcs_subtype, dst->ct_fcs_subtype);
ISP_IOXGET_8(isp, &src->ct_options, dst->ct_options);
ISP_IOXGET_8(isp, &src->ct_res0, dst->ct_res0);
ISP_IOXGET_16(isp, &src->ct_response, dst->ct_response);
dst->ct_response = (dst->ct_response << 8) | (dst->ct_response >> 8);
ISP_IOXGET_16(isp, &src->ct_resid, dst->ct_resid);
dst->ct_resid = (dst->ct_resid << 8) | (dst->ct_resid >> 8);
ISP_IOXGET_8(isp, &src->ct_res1, dst->ct_res1);
ISP_IOXGET_8(isp, &src->ct_reason, dst->ct_reason);
ISP_IOXGET_8(isp, &src->ct_explanation, dst->ct_explanation);
ISP_IOXGET_8(isp, &src->ct_vunique, dst->ct_vunique);
}
/*
* Generic SNS request - not particularly useful since the per-command data
* isn't always 16 bit words.
*/
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_sns_request(ispsoftc_t *isp, sns_screq_t *src, sns_screq_t *dst)
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
{
int i, nw = (int) src->snscb_sblen;
ISP_IOXPUT_16(isp, src->snscb_rblen, &dst->snscb_rblen);
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, src->snscb_addr[i], &dst->snscb_addr[i]);
}
ISP_IOXPUT_16(isp, src->snscb_sblen, &dst->snscb_sblen);
for (i = 0; i < nw; i++) {
ISP_IOXPUT_16(isp, src->snscb_data[i], &dst->snscb_data[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_gid_ft_request(ispsoftc_t *isp, sns_gid_ft_req_t *src,
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
sns_gid_ft_req_t *dst)
{
ISP_IOXPUT_16(isp, src->snscb_rblen, &dst->snscb_rblen);
ISP_IOXPUT_16(isp, src->snscb_res0, &dst->snscb_res0);
ISP_IOXPUT_16(isp, src->snscb_addr[0], &dst->snscb_addr[0]);
ISP_IOXPUT_16(isp, src->snscb_addr[1], &dst->snscb_addr[1]);
ISP_IOXPUT_16(isp, src->snscb_addr[2], &dst->snscb_addr[2]);
ISP_IOXPUT_16(isp, src->snscb_addr[3], &dst->snscb_addr[3]);
ISP_IOXPUT_16(isp, src->snscb_sblen, &dst->snscb_sblen);
ISP_IOXPUT_16(isp, src->snscb_res1, &dst->snscb_res1);
ISP_IOXPUT_16(isp, src->snscb_cmd, &dst->snscb_cmd);
ISP_IOXPUT_16(isp, src->snscb_mword_div_2, &dst->snscb_mword_div_2);
ISP_IOXPUT_32(isp, src->snscb_res3, &dst->snscb_res3);
Send 32 bytes out for fc4_types... Interestingly enough the Solaris/Sparc version worked fine, but Linux/Sparc && FreeBSD/Sparc choked. MFC after: 1 week
2002-04-05 01:40:05 +00:00
ISP_IOXPUT_32(isp, src->snscb_fc4_type, &dst->snscb_fc4_type);
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_gxn_id_request(ispsoftc_t *isp, sns_gxn_id_req_t *src,
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
sns_gxn_id_req_t *dst)
{
ISP_IOXPUT_16(isp, src->snscb_rblen, &dst->snscb_rblen);
ISP_IOXPUT_16(isp, src->snscb_res0, &dst->snscb_res0);
ISP_IOXPUT_16(isp, src->snscb_addr[0], &dst->snscb_addr[0]);
ISP_IOXPUT_16(isp, src->snscb_addr[1], &dst->snscb_addr[1]);
ISP_IOXPUT_16(isp, src->snscb_addr[2], &dst->snscb_addr[2]);
ISP_IOXPUT_16(isp, src->snscb_addr[3], &dst->snscb_addr[3]);
ISP_IOXPUT_16(isp, src->snscb_sblen, &dst->snscb_sblen);
ISP_IOXPUT_16(isp, src->snscb_res1, &dst->snscb_res1);
ISP_IOXPUT_16(isp, src->snscb_cmd, &dst->snscb_cmd);
ISP_IOXPUT_16(isp, src->snscb_res2, &dst->snscb_res2);
ISP_IOXPUT_32(isp, src->snscb_res3, &dst->snscb_res3);
ISP_IOXPUT_32(isp, src->snscb_portid, &dst->snscb_portid);
}
/*
* Generic SNS response - not particularly useful since the per-command data
* isn't always 16 bit words.
*/
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_sns_response(ispsoftc_t *isp, sns_scrsp_t *src,
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_scrsp_t *dst, int nwords)
{
int i;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_get_ct_hdr(isp, &src->snscb_cthdr, &dst->snscb_cthdr);
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_IOXGET_8(isp, &src->snscb_port_type, dst->snscb_port_type);
for (i = 0; i < 3; i++) {
ISP_IOXGET_8(isp, &src->snscb_port_id[i],
dst->snscb_port_id[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->snscb_portname[i],
dst->snscb_portname[i]);
}
for (i = 0; i < nwords; i++) {
ISP_IOXGET_16(isp, &src->snscb_data[i], dst->snscb_data[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_gid_ft_response(ispsoftc_t *isp, sns_gid_ft_rsp_t *src,
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
sns_gid_ft_rsp_t *dst, int nwords)
Explicitly decode GetAllNext SNS Response back *as* a GetAllNext response. Otherwise, we won't unswizzle it correctly. This was found on linux/PPC. This mandated creating another inline: isp_get_gan_response.
2001-12-11 21:58:04 +00:00
{
int i;
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
isp_get_ct_hdr(isp, &src->snscb_cthdr, &dst->snscb_cthdr);
for (i = 0; i < nwords; i++) {
int j;
ISP_IOXGET_8(isp,
&src->snscb_ports[i].control,
dst->snscb_ports[i].control);
for (j = 0; j < 3; j++) {
ISP_IOXGET_8(isp,
&src->snscb_ports[i].portid[j],
dst->snscb_ports[i].portid[j]);
}
if (dst->snscb_ports[i].control & 0x80) {
break;
}
Explicitly decode GetAllNext SNS Response back *as* a GetAllNext response. Otherwise, we won't unswizzle it correctly. This was found on linux/PPC. This mandated creating another inline: isp_get_gan_response.
2001-12-11 21:58:04 +00:00
}
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_gxn_id_response(ispsoftc_t *isp, sns_gxn_id_rsp_t *src,
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
sns_gxn_id_rsp_t *dst)
{
int i;
isp_get_ct_hdr(isp, &src->snscb_cthdr, &dst->snscb_cthdr);
for (i = 0; i < 8; i++)
ISP_IOXGET_8(isp, &src->snscb_wwn[i], dst->snscb_wwn[i]);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_gff_id_response(ispsoftc_t *isp, sns_gff_id_rsp_t *src,
Scale back # of luns supported for SCC to 16384- oops- top 3 bits are a lun address modifier of sorts. Only an HP XP-512 seems to have cared. Fix a few misplaced pointers for the new fabric goop, which has been demonstrated to work on newer Brocades and McData switches now. Put in commented out code which would run GFF_ID if the QLogic f/w allowed it. Don't whine about not being able to find a handle for a command if it was a command aborted (by us).
2002-04-16 19:55:35 +00:00
sns_gff_id_rsp_t *dst)
{
int i;
isp_get_ct_hdr(isp, &src->snscb_cthdr, &dst->snscb_cthdr);
for (i = 0; i < 32; i++) {
ISP_IOXGET_32(isp, &src->snscb_fc4_features[i],
dst->snscb_fc4_features[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_ga_nxt_response(ispsoftc_t *isp, sns_ga_nxt_rsp_t *src,
Fix bus dma segment count to be based off of MAXPHYS, not BUS_SPACE_MAXSIZE. Grumble. I've seen better documented architectures out of Redmond. Redo fabric evaluation to not use GET ALL NEXT (GA_NXT). Switches seem to be trying to wriggle out of supporting this well. Instead, use GID_FT to get a list of Port IDs and then use GPN_ID/GNN_ID to find the port and node wwn. This should make working on fabrics a bit cleaner and more stable. This also caused some cleanup of SNS subcommand canonicalization so that we can actually check for FS_ACC and FS_RJT, and if we get an FS_RJT, print out the reason and explanation codes. We'll keep the old GA_NXT method around if people want to uncomment a controlling definition in ispvar.h. This also had us clean up ISPASYNC_FABRICDEV to use a local lportdb argument and to have the caller explicitly say that a device is at the end of the fabric list. MFC after: 1 week
2002-04-04 23:46:01 +00:00
sns_ga_nxt_rsp_t *dst)
{
int i;
isp_get_ct_hdr(isp, &src->snscb_cthdr, &dst->snscb_cthdr);
Explicitly decode GetAllNext SNS Response back *as* a GetAllNext response. Otherwise, we won't unswizzle it correctly. This was found on linux/PPC. This mandated creating another inline: isp_get_gan_response.
2001-12-11 21:58:04 +00:00
ISP_IOXGET_8(isp, &src->snscb_port_type, dst->snscb_port_type);
for (i = 0; i < 3; i++) {
ISP_IOXGET_8(isp, &src->snscb_port_id[i],
dst->snscb_port_id[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->snscb_portname[i],
dst->snscb_portname[i]);
}
ISP_IOXGET_8(isp, &src->snscb_pnlen, dst->snscb_pnlen);
for (i = 0; i < 255; i++) {
ISP_IOXGET_8(isp, &src->snscb_pname[i], dst->snscb_pname[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->snscb_nodename[i],
dst->snscb_nodename[i]);
}
ISP_IOXGET_8(isp, &src->snscb_nnlen, dst->snscb_nnlen);
for (i = 0; i < 255; i++) {
ISP_IOXGET_8(isp, &src->snscb_nname[i], dst->snscb_nname[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->snscb_ipassoc[i],
dst->snscb_ipassoc[i]);
}
for (i = 0; i < 16; i++) {
ISP_IOXGET_8(isp, &src->snscb_ipaddr[i], dst->snscb_ipaddr[i]);
}
for (i = 0; i < 4; i++) {
ISP_IOXGET_8(isp, &src->snscb_svc_class[i],
dst->snscb_svc_class[i]);
}
for (i = 0; i < 32; i++) {
ISP_IOXGET_8(isp, &src->snscb_fc4_types[i],
dst->snscb_fc4_types[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->snscb_fpname[i], dst->snscb_fpname[i]);
}
ISP_IOXGET_8(isp, &src->snscb_reserved, dst->snscb_reserved);
for (i = 0; i < 3; i++) {
ISP_IOXGET_8(isp, &src->snscb_hardaddr[i],
dst->snscb_hardaddr[i]);
}
}
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
#ifdef ISP_TARGET_MODE
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
int
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_save_xs_tgt(ispsoftc_t *isp, void *xs, uint16_t *handlep)
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
{
int i;
for (i = 0; i < (int) isp->isp_maxcmds; i++) {
if (isp->isp_tgtlist[i] == NULL) {
break;
}
}
if (i == isp->isp_maxcmds) {
return (-1);
}
isp->isp_tgtlist[i] = xs;
*handlep = i+1;
return (0);
}
void *
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_find_xs_tgt(ispsoftc_t *isp, uint16_t handle)
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if (handle < 1 || handle > (uint16_t) isp->isp_maxcmds) {
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
return (NULL);
} else {
return (isp->isp_tgtlist[handle - 1]);
}
}
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
uint16_t
isp_find_tgt_handle(ispsoftc_t *isp, void *xs)
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
{
int i;
if (xs != NULL) {
for (i = 0; i < isp->isp_maxcmds; i++) {
if (isp->isp_tgtlist[i] == xs) {
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
return ((uint16_t) i+1);
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
}
}
}
return (0);
}
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_destroy_tgt_handle(ispsoftc_t *isp, uint16_t handle)
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
{
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
if (handle > 0 && handle <= (uint16_t) isp->isp_maxcmds) {
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
isp->isp_tgtlist[handle - 1] = NULL;
}
}
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_atio(ispsoftc_t *isp, at_entry_t *atsrc, at_entry_t *atdst)
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
{
int i;
isp_copy_out_hdr(isp, &atsrc->at_header, &atdst->at_header);
ISP_IOXPUT_16(isp, atsrc->at_reserved, &atdst->at_reserved);
ISP_IOXPUT_16(isp, atsrc->at_handle, &atdst->at_handle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, atsrc->at_lun, &atdst->at_iid);
ISP_IOXPUT_8(isp, atsrc->at_iid, &atdst->at_lun);
ISP_IOXPUT_8(isp, atsrc->at_cdblen, &atdst->at_tgt);
ISP_IOXPUT_8(isp, atsrc->at_tgt, &atdst->at_cdblen);
ISP_IOXPUT_8(isp, atsrc->at_status, &atdst->at_scsi_status);
ISP_IOXPUT_8(isp, atsrc->at_scsi_status, &atdst->at_status);
ISP_IOXPUT_8(isp, atsrc->at_tag_val, &atdst->at_tag_type);
ISP_IOXPUT_8(isp, atsrc->at_tag_type, &atdst->at_tag_val);
} else {
ISP_IOXPUT_8(isp, atsrc->at_lun, &atdst->at_lun);
ISP_IOXPUT_8(isp, atsrc->at_iid, &atdst->at_iid);
ISP_IOXPUT_8(isp, atsrc->at_cdblen, &atdst->at_cdblen);
ISP_IOXPUT_8(isp, atsrc->at_tgt, &atdst->at_tgt);
ISP_IOXPUT_8(isp, atsrc->at_status, &atdst->at_status);
ISP_IOXPUT_8(isp, atsrc->at_scsi_status,
&atdst->at_scsi_status);
ISP_IOXPUT_8(isp, atsrc->at_tag_val, &atdst->at_tag_val);
ISP_IOXPUT_8(isp, atsrc->at_tag_type, &atdst->at_tag_type);
}
ISP_IOXPUT_32(isp, atsrc->at_flags, &atdst->at_flags);
for (i = 0; i < ATIO_CDBLEN; i++) {
ISP_IOXPUT_8(isp, atsrc->at_cdb[i], &atdst->at_cdb[i]);
}
for (i = 0; i < QLTM_SENSELEN; i++) {
ISP_IOXPUT_8(isp, atsrc->at_sense[i], &atdst->at_sense[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_atio(ispsoftc_t *isp, at_entry_t *atsrc, at_entry_t *atdst)
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
{
int i;
isp_copy_in_hdr(isp, &atsrc->at_header, &atdst->at_header);
ISP_IOXGET_16(isp, &atsrc->at_reserved, atdst->at_reserved);
ISP_IOXGET_16(isp, &atsrc->at_handle, atdst->at_handle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &atsrc->at_lun, atdst->at_iid);
ISP_IOXGET_8(isp, &atsrc->at_iid, atdst->at_lun);
ISP_IOXGET_8(isp, &atsrc->at_cdblen, atdst->at_tgt);
ISP_IOXGET_8(isp, &atsrc->at_tgt, atdst->at_cdblen);
ISP_IOXGET_8(isp, &atsrc->at_status, atdst->at_scsi_status);
ISP_IOXGET_8(isp, &atsrc->at_scsi_status, atdst->at_status);
ISP_IOXGET_8(isp, &atsrc->at_tag_val, atdst->at_tag_type);
ISP_IOXGET_8(isp, &atsrc->at_tag_type, atdst->at_tag_val);
} else {
ISP_IOXGET_8(isp, &atsrc->at_lun, atdst->at_lun);
ISP_IOXGET_8(isp, &atsrc->at_iid, atdst->at_iid);
ISP_IOXGET_8(isp, &atsrc->at_cdblen, atdst->at_cdblen);
ISP_IOXGET_8(isp, &atsrc->at_tgt, atdst->at_tgt);
ISP_IOXGET_8(isp, &atsrc->at_status, atdst->at_status);
ISP_IOXGET_8(isp, &atsrc->at_scsi_status,
atdst->at_scsi_status);
ISP_IOXGET_8(isp, &atsrc->at_tag_val, atdst->at_tag_val);
ISP_IOXGET_8(isp, &atsrc->at_tag_type, atdst->at_tag_type);
}
ISP_IOXGET_32(isp, &atsrc->at_flags, atdst->at_flags);
for (i = 0; i < ATIO_CDBLEN; i++) {
ISP_IOXGET_8(isp, &atsrc->at_cdb[i], atdst->at_cdb[i]);
}
for (i = 0; i < QLTM_SENSELEN; i++) {
ISP_IOXGET_8(isp, &atsrc->at_sense[i], atdst->at_sense[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_atio2(ispsoftc_t *isp, at2_entry_t *atsrc, at2_entry_t *atdst)
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
{
int i;
isp_copy_out_hdr(isp, &atsrc->at_header, &atdst->at_header);
ISP_IOXPUT_32(isp, atsrc->at_reserved, &atdst->at_reserved);
ISP_IOXPUT_8(isp, atsrc->at_lun, &atdst->at_lun);
ISP_IOXPUT_8(isp, atsrc->at_iid, &atdst->at_iid);
ISP_IOXPUT_16(isp, atsrc->at_rxid, &atdst->at_rxid);
ISP_IOXPUT_16(isp, atsrc->at_flags, &atdst->at_flags);
ISP_IOXPUT_16(isp, atsrc->at_status, &atdst->at_status);
Some canonicalization function cleanups based upon some target mode structure cleanups.
2003-09-13 01:57:31 +00:00
ISP_IOXPUT_8(isp, atsrc->at_crn, &atdst->at_crn);
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_IOXPUT_8(isp, atsrc->at_taskcodes, &atdst->at_taskcodes);
ISP_IOXPUT_8(isp, atsrc->at_taskflags, &atdst->at_taskflags);
ISP_IOXPUT_8(isp, atsrc->at_execodes, &atdst->at_execodes);
for (i = 0; i < ATIO2_CDBLEN; i++) {
ISP_IOXPUT_8(isp, atsrc->at_cdb[i], &atdst->at_cdb[i]);
}
ISP_IOXPUT_32(isp, atsrc->at_datalen, &atdst->at_datalen);
ISP_IOXPUT_16(isp, atsrc->at_scclun, &atdst->at_scclun);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, atsrc->at_wwpn[i], &atdst->at_wwpn[i]);
}
for (i = 0; i < 6; i++) {
ISP_IOXPUT_16(isp, atsrc->at_reserved2[i],
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
&atdst->at_reserved2[i]);
}
ISP_IOXPUT_16(isp, atsrc->at_oxid, &atdst->at_oxid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_atio2e(ispsoftc_t *isp, at2e_entry_t *atsrc, at2e_entry_t *atdst)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
{
int i;
isp_copy_out_hdr(isp, &atsrc->at_header, &atdst->at_header);
ISP_IOXPUT_32(isp, atsrc->at_reserved, &atdst->at_reserved);
ISP_IOXPUT_16(isp, atsrc->at_iid, &atdst->at_iid);
ISP_IOXPUT_16(isp, atsrc->at_rxid, &atdst->at_rxid);
ISP_IOXPUT_16(isp, atsrc->at_flags, &atdst->at_flags);
ISP_IOXPUT_16(isp, atsrc->at_status, &atdst->at_status);
ISP_IOXPUT_8(isp, atsrc->at_crn, &atdst->at_crn);
ISP_IOXPUT_8(isp, atsrc->at_taskcodes, &atdst->at_taskcodes);
ISP_IOXPUT_8(isp, atsrc->at_taskflags, &atdst->at_taskflags);
ISP_IOXPUT_8(isp, atsrc->at_execodes, &atdst->at_execodes);
for (i = 0; i < ATIO2_CDBLEN; i++) {
ISP_IOXPUT_8(isp, atsrc->at_cdb[i], &atdst->at_cdb[i]);
}
ISP_IOXPUT_32(isp, atsrc->at_datalen, &atdst->at_datalen);
ISP_IOXPUT_16(isp, atsrc->at_scclun, &atdst->at_scclun);
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, atsrc->at_wwpn[i], &atdst->at_wwpn[i]);
}
for (i = 0; i < 6; i++) {
ISP_IOXPUT_16(isp, atsrc->at_reserved2[i],
&atdst->at_reserved2[i]);
}
ISP_IOXPUT_16(isp, atsrc->at_oxid, &atdst->at_oxid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_atio2(ispsoftc_t *isp, at2_entry_t *atsrc, at2_entry_t *atdst)
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
{
int i;
isp_copy_in_hdr(isp, &atsrc->at_header, &atdst->at_header);
ISP_IOXGET_32(isp, &atsrc->at_reserved, atdst->at_reserved);
ISP_IOXGET_8(isp, &atsrc->at_lun, atdst->at_lun);
ISP_IOXGET_8(isp, &atsrc->at_iid, atdst->at_iid);
ISP_IOXGET_16(isp, &atsrc->at_rxid, atdst->at_rxid);
ISP_IOXGET_16(isp, &atsrc->at_flags, atdst->at_flags);
ISP_IOXGET_16(isp, &atsrc->at_status, atdst->at_status);
Some canonicalization function cleanups based upon some target mode structure cleanups.
2003-09-13 01:57:31 +00:00
ISP_IOXGET_8(isp, &atsrc->at_crn, atdst->at_crn);
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_IOXGET_8(isp, &atsrc->at_taskcodes, atdst->at_taskcodes);
ISP_IOXGET_8(isp, &atsrc->at_taskflags, atdst->at_taskflags);
ISP_IOXGET_8(isp, &atsrc->at_execodes, atdst->at_execodes);
for (i = 0; i < ATIO2_CDBLEN; i++) {
ISP_IOXGET_8(isp, &atsrc->at_cdb[i], atdst->at_cdb[i]);
}
ISP_IOXGET_32(isp, &atsrc->at_datalen, atdst->at_datalen);
ISP_IOXGET_16(isp, &atsrc->at_scclun, atdst->at_scclun);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
for (i = 0; i < 4; i++) {
ISP_IOXGET_16(isp, &atsrc->at_wwpn[i], atdst->at_wwpn[i]);
}
for (i = 0; i < 6; i++) {
ISP_IOXGET_16(isp, &atsrc->at_reserved2[i],
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
atdst->at_reserved2[i]);
}
ISP_IOXGET_16(isp, &atsrc->at_oxid, atdst->at_oxid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_atio2e(ispsoftc_t *isp, at2e_entry_t *atsrc, at2e_entry_t *atdst)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
{
int i;
isp_copy_in_hdr(isp, &atsrc->at_header, &atdst->at_header);
ISP_IOXGET_32(isp, &atsrc->at_reserved, atdst->at_reserved);
ISP_IOXGET_16(isp, &atsrc->at_iid, atdst->at_iid);
ISP_IOXGET_16(isp, &atsrc->at_rxid, atdst->at_rxid);
ISP_IOXGET_16(isp, &atsrc->at_flags, atdst->at_flags);
ISP_IOXGET_16(isp, &atsrc->at_status, atdst->at_status);
ISP_IOXGET_8(isp, &atsrc->at_crn, atdst->at_crn);
ISP_IOXGET_8(isp, &atsrc->at_taskcodes, atdst->at_taskcodes);
ISP_IOXGET_8(isp, &atsrc->at_taskflags, atdst->at_taskflags);
ISP_IOXGET_8(isp, &atsrc->at_execodes, atdst->at_execodes);
for (i = 0; i < ATIO2_CDBLEN; i++) {
ISP_IOXGET_8(isp, &atsrc->at_cdb[i], atdst->at_cdb[i]);
}
ISP_IOXGET_32(isp, &atsrc->at_datalen, atdst->at_datalen);
ISP_IOXGET_16(isp, &atsrc->at_scclun, atdst->at_scclun);
for (i = 0; i < 4; i++) {
ISP_IOXGET_16(isp, &atsrc->at_wwpn[i], atdst->at_wwpn[i]);
}
for (i = 0; i < 6; i++) {
ISP_IOXGET_16(isp, &atsrc->at_reserved2[i],
atdst->at_reserved2[i]);
}
ISP_IOXGET_16(isp, &atsrc->at_oxid, atdst->at_oxid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_ctio(ispsoftc_t *isp, ct_entry_t *ctsrc, ct_entry_t *ctdst)
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
{
int i;
isp_copy_out_hdr(isp, &ctsrc->ct_header, &ctdst->ct_header);
ISP_IOXPUT_16(isp, ctsrc->ct_reserved, &ctdst->ct_reserved);
ISP_IOXPUT_16(isp, ctsrc->ct_fwhandle, &ctdst->ct_fwhandle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, ctsrc->ct_iid, &ctdst->ct_lun);
ISP_IOXPUT_8(isp, ctsrc->ct_lun, &ctdst->ct_iid);
ISP_IOXPUT_8(isp, ctsrc->ct_tgt, &ctdst->ct_reserved2);
ISP_IOXPUT_8(isp, ctsrc->ct_reserved2, &ctdst->ct_tgt);
ISP_IOXPUT_8(isp, ctsrc->ct_status, &ctdst->ct_scsi_status);
ISP_IOXPUT_8(isp, ctsrc->ct_scsi_status, &ctdst->ct_status);
ISP_IOXPUT_8(isp, ctsrc->ct_tag_type, &ctdst->ct_tag_val);
ISP_IOXPUT_8(isp, ctsrc->ct_tag_val, &ctdst->ct_tag_type);
} else {
ISP_IOXPUT_8(isp, ctsrc->ct_iid, &ctdst->ct_iid);
ISP_IOXPUT_8(isp, ctsrc->ct_lun, &ctdst->ct_lun);
ISP_IOXPUT_8(isp, ctsrc->ct_tgt, &ctdst->ct_tgt);
ISP_IOXPUT_8(isp, ctsrc->ct_reserved2, &ctdst->ct_reserved2);
ISP_IOXPUT_8(isp, ctsrc->ct_scsi_status,
&ctdst->ct_scsi_status);
ISP_IOXPUT_8(isp, ctsrc->ct_status, &ctdst->ct_status);
ISP_IOXPUT_8(isp, ctsrc->ct_tag_type, &ctdst->ct_tag_type);
ISP_IOXPUT_8(isp, ctsrc->ct_tag_val, &ctdst->ct_tag_val);
}
ISP_IOXPUT_32(isp, ctsrc->ct_flags, &ctdst->ct_flags);
ISP_IOXPUT_32(isp, ctsrc->ct_xfrlen, &ctdst->ct_xfrlen);
ISP_IOXPUT_32(isp, ctsrc->ct_resid, &ctdst->ct_resid);
ISP_IOXPUT_16(isp, ctsrc->ct_timeout, &ctdst->ct_timeout);
ISP_IOXPUT_16(isp, ctsrc->ct_seg_count, &ctdst->ct_seg_count);
for (i = 0; i < ISP_RQDSEG; i++) {
ISP_IOXPUT_32(isp, ctsrc->ct_dataseg[i].ds_base,
&ctdst->ct_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, ctsrc->ct_dataseg[i].ds_count,
&ctdst->ct_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_ctio(ispsoftc_t *isp, ct_entry_t *ctsrc, ct_entry_t *ctdst)
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
{
int i;
isp_copy_in_hdr(isp, &ctsrc->ct_header, &ctdst->ct_header);
ISP_IOXGET_16(isp, &ctsrc->ct_reserved, ctdst->ct_reserved);
ISP_IOXGET_16(isp, &ctsrc->ct_fwhandle, ctdst->ct_fwhandle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &ctsrc->ct_lun, ctdst->ct_iid);
ISP_IOXGET_8(isp, &ctsrc->ct_iid, ctdst->ct_lun);
ISP_IOXGET_8(isp, &ctsrc->ct_reserved2, ctdst->ct_tgt);
ISP_IOXGET_8(isp, &ctsrc->ct_tgt, ctdst->ct_reserved2);
ISP_IOXGET_8(isp, &ctsrc->ct_status, ctdst->ct_scsi_status);
ISP_IOXGET_8(isp, &ctsrc->ct_scsi_status, ctdst->ct_status);
ISP_IOXGET_8(isp, &ctsrc->ct_tag_val, ctdst->ct_tag_type);
ISP_IOXGET_8(isp, &ctsrc->ct_tag_type, ctdst->ct_tag_val);
} else {
ISP_IOXGET_8(isp, &ctsrc->ct_lun, ctdst->ct_lun);
ISP_IOXGET_8(isp, &ctsrc->ct_iid, ctdst->ct_iid);
ISP_IOXGET_8(isp, &ctsrc->ct_reserved2, ctdst->ct_reserved2);
ISP_IOXGET_8(isp, &ctsrc->ct_tgt, ctdst->ct_tgt);
ISP_IOXGET_8(isp, &ctsrc->ct_status, ctdst->ct_status);
ISP_IOXGET_8(isp, &ctsrc->ct_scsi_status,
ctdst->ct_scsi_status);
ISP_IOXGET_8(isp, &ctsrc->ct_tag_val, ctdst->ct_tag_val);
ISP_IOXGET_8(isp, &ctsrc->ct_tag_type, ctdst->ct_tag_type);
}
ISP_IOXGET_32(isp, &ctsrc->ct_flags, ctdst->ct_flags);
ISP_IOXGET_32(isp, &ctsrc->ct_xfrlen, ctdst->ct_xfrlen);
ISP_IOXGET_32(isp, &ctsrc->ct_resid, ctdst->ct_resid);
ISP_IOXGET_16(isp, &ctsrc->ct_timeout, ctdst->ct_timeout);
ISP_IOXGET_16(isp, &ctsrc->ct_seg_count, ctdst->ct_seg_count);
for (i = 0; i < ISP_RQDSEG; i++) {
ISP_IOXGET_32(isp,
&ctsrc->ct_dataseg[i].ds_base,
ctdst->ct_dataseg[i].ds_base);
ISP_IOXGET_32(isp,
&ctsrc->ct_dataseg[i].ds_count,
ctdst->ct_dataseg[i].ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_ctio2(ispsoftc_t *isp, ct2_entry_t *ctsrc, ct2_entry_t *ctdst)
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
{
int i;
isp_copy_out_hdr(isp, &ctsrc->ct_header, &ctdst->ct_header);
ISP_IOXPUT_16(isp, ctsrc->ct_reserved, &ctdst->ct_reserved);
ISP_IOXPUT_16(isp, ctsrc->ct_fwhandle, &ctdst->ct_fwhandle);
ISP_IOXPUT_8(isp, ctsrc->ct_lun, &ctdst->ct_lun);
ISP_IOXPUT_8(isp, ctsrc->ct_iid, &ctdst->ct_iid);
ISP_IOXPUT_16(isp, ctsrc->ct_rxid, &ctdst->ct_rxid);
ISP_IOXPUT_16(isp, ctsrc->ct_flags, &ctdst->ct_flags);
ISP_IOXPUT_16(isp, ctsrc->ct_timeout, &ctdst->ct_timeout);
ISP_IOXPUT_16(isp, ctsrc->ct_seg_count, &ctdst->ct_seg_count);
ISP_IOXPUT_32(isp, ctsrc->ct_resid, &ctdst->ct_resid);
Add missing move of relative offset for CTIO2 updates.
2002-01-11 23:48:25 +00:00
ISP_IOXPUT_32(isp, ctsrc->ct_reloff, &ctdst->ct_reloff);
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 ((ctsrc->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE0) {
ISP_IOXPUT_32(isp, ctsrc->rsp.m0._reserved,
&ctdst->rsp.m0._reserved);
ISP_IOXPUT_16(isp, ctsrc->rsp.m0._reserved2,
&ctdst->rsp.m0._reserved2);
ISP_IOXPUT_16(isp, ctsrc->rsp.m0.ct_scsi_status,
&ctdst->rsp.m0.ct_scsi_status);
ISP_IOXPUT_32(isp, ctsrc->rsp.m0.ct_xfrlen,
&ctdst->rsp.m0.ct_xfrlen);
if (ctsrc->ct_header.rqs_entry_type == RQSTYPE_CTIO2) {
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg[i].ds_base,
&ctdst->rsp.m0.ct_dataseg[i].ds_base);
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg[i].ds_count,
&ctdst->rsp.m0.ct_dataseg[i].ds_count);
}
} else if (ctsrc->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
for (i = 0; i < ISP_RQDSEG_T3; i++) {
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg64[i].ds_base,
&ctdst->rsp.m0.ct_dataseg64[i].ds_base);
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg64[i].ds_basehi,
&ctdst->rsp.m0.ct_dataseg64[i].ds_basehi);
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg64[i].ds_count,
&ctdst->rsp.m0.ct_dataseg64[i].ds_count);
}
} else if (ctsrc->ct_header.rqs_entry_type == RQSTYPE_CTIO4) {
ISP_IOXPUT_16(isp, ctsrc->rsp.m0.ct_dslist.ds_type,
&ctdst->rsp.m0.ct_dslist.ds_type);
ISP_IOXPUT_32(isp, ctsrc->rsp.m0.ct_dslist.ds_segment,
&ctdst->rsp.m0.ct_dslist.ds_segment);
ISP_IOXPUT_32(isp, ctsrc->rsp.m0.ct_dslist.ds_base,
&ctdst->rsp.m0.ct_dslist.ds_base);
}
} else if ((ctsrc->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE1) {
ISP_IOXPUT_16(isp, ctsrc->rsp.m1._reserved,
&ctdst->rsp.m1._reserved);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1._reserved2,
&ctdst->rsp.m1._reserved2);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1.ct_senselen,
&ctdst->rsp.m1.ct_senselen);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1.ct_scsi_status,
&ctdst->rsp.m1.ct_scsi_status);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1.ct_resplen,
&ctdst->rsp.m1.ct_resplen);
for (i = 0; i < MAXRESPLEN; i++) {
ISP_IOXPUT_8(isp, ctsrc->rsp.m1.ct_resp[i],
&ctdst->rsp.m1.ct_resp[i]);
}
} else {
ISP_IOXPUT_32(isp, ctsrc->rsp.m2._reserved,
&ctdst->rsp.m2._reserved);
ISP_IOXPUT_16(isp, ctsrc->rsp.m2._reserved2,
&ctdst->rsp.m2._reserved2);
ISP_IOXPUT_16(isp, ctsrc->rsp.m2._reserved3,
&ctdst->rsp.m2._reserved3);
ISP_IOXPUT_32(isp, ctsrc->rsp.m2.ct_datalen,
&ctdst->rsp.m2.ct_datalen);
ISP_IOXPUT_32(isp, ctsrc->rsp.m2.ct_fcp_rsp_iudata.ds_base,
&ctdst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXPUT_32(isp, ctsrc->rsp.m2.ct_fcp_rsp_iudata.ds_count,
&ctdst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_ctio2e(ispsoftc_t *isp, ct2e_entry_t *ctsrc, ct2e_entry_t *ctdst)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
{
int i;
isp_copy_out_hdr(isp, &ctsrc->ct_header, &ctdst->ct_header);
ISP_IOXPUT_16(isp, ctsrc->ct_reserved, &ctdst->ct_reserved);
ISP_IOXPUT_16(isp, ctsrc->ct_fwhandle, &ctdst->ct_fwhandle);
ISP_IOXPUT_16(isp, ctsrc->ct_iid, &ctdst->ct_iid);
ISP_IOXPUT_16(isp, ctsrc->ct_rxid, &ctdst->ct_rxid);
ISP_IOXPUT_16(isp, ctsrc->ct_flags, &ctdst->ct_flags);
ISP_IOXPUT_16(isp, ctsrc->ct_timeout, &ctdst->ct_timeout);
ISP_IOXPUT_16(isp, ctsrc->ct_seg_count, &ctdst->ct_seg_count);
ISP_IOXPUT_32(isp, ctsrc->ct_resid, &ctdst->ct_resid);
ISP_IOXPUT_32(isp, ctsrc->ct_reloff, &ctdst->ct_reloff);
if ((ctsrc->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE0) {
ISP_IOXPUT_32(isp, ctsrc->rsp.m0._reserved,
&ctdst->rsp.m0._reserved);
ISP_IOXPUT_16(isp, ctsrc->rsp.m0._reserved2,
&ctdst->rsp.m0._reserved2);
ISP_IOXPUT_16(isp, ctsrc->rsp.m0.ct_scsi_status,
&ctdst->rsp.m0.ct_scsi_status);
ISP_IOXPUT_32(isp, ctsrc->rsp.m0.ct_xfrlen,
&ctdst->rsp.m0.ct_xfrlen);
if (ctsrc->ct_header.rqs_entry_type == RQSTYPE_CTIO2) {
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg[i].ds_base,
&ctdst->rsp.m0.ct_dataseg[i].ds_base);
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg[i].ds_count,
&ctdst->rsp.m0.ct_dataseg[i].ds_count);
}
} else if (ctsrc->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
for (i = 0; i < ISP_RQDSEG_T3; i++) {
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg64[i].ds_base,
&ctdst->rsp.m0.ct_dataseg64[i].ds_base);
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg64[i].ds_basehi,
&ctdst->rsp.m0.ct_dataseg64[i].ds_basehi);
ISP_IOXPUT_32(isp,
ctsrc->rsp.m0.ct_dataseg64[i].ds_count,
&ctdst->rsp.m0.ct_dataseg64[i].ds_count);
}
} else if (ctsrc->ct_header.rqs_entry_type == RQSTYPE_CTIO4) {
ISP_IOXPUT_16(isp, ctsrc->rsp.m0.ct_dslist.ds_type,
&ctdst->rsp.m0.ct_dslist.ds_type);
ISP_IOXPUT_32(isp, ctsrc->rsp.m0.ct_dslist.ds_segment,
&ctdst->rsp.m0.ct_dslist.ds_segment);
ISP_IOXPUT_32(isp, ctsrc->rsp.m0.ct_dslist.ds_base,
&ctdst->rsp.m0.ct_dslist.ds_base);
}
} else if ((ctsrc->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE1) {
ISP_IOXPUT_16(isp, ctsrc->rsp.m1._reserved,
&ctdst->rsp.m1._reserved);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1._reserved2,
&ctdst->rsp.m1._reserved2);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1.ct_senselen,
&ctdst->rsp.m1.ct_senselen);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1.ct_scsi_status,
&ctdst->rsp.m1.ct_scsi_status);
ISP_IOXPUT_16(isp, ctsrc->rsp.m1.ct_resplen,
&ctdst->rsp.m1.ct_resplen);
for (i = 0; i < MAXRESPLEN; i++) {
ISP_IOXPUT_8(isp, ctsrc->rsp.m1.ct_resp[i],
&ctdst->rsp.m1.ct_resp[i]);
}
} else {
ISP_IOXPUT_32(isp, ctsrc->rsp.m2._reserved,
&ctdst->rsp.m2._reserved);
ISP_IOXPUT_16(isp, ctsrc->rsp.m2._reserved2,
&ctdst->rsp.m2._reserved2);
ISP_IOXPUT_16(isp, ctsrc->rsp.m2._reserved3,
&ctdst->rsp.m2._reserved3);
ISP_IOXPUT_32(isp, ctsrc->rsp.m2.ct_datalen,
&ctdst->rsp.m2.ct_datalen);
ISP_IOXPUT_32(isp, ctsrc->rsp.m2.ct_fcp_rsp_iudata.ds_base,
&ctdst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXPUT_32(isp, ctsrc->rsp.m2.ct_fcp_rsp_iudata.ds_count,
&ctdst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_ctio2(ispsoftc_t *isp, ct2_entry_t *ctsrc, ct2_entry_t *ctdst)
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_copy_in_hdr(isp, &ctsrc->ct_header, &ctdst->ct_header);
ISP_IOXGET_16(isp, &ctsrc->ct_reserved, ctdst->ct_reserved);
ISP_IOXGET_16(isp, &ctsrc->ct_fwhandle, ctdst->ct_fwhandle);
ISP_IOXGET_8(isp, &ctsrc->ct_lun, ctdst->ct_lun);
ISP_IOXGET_8(isp, &ctsrc->ct_iid, ctdst->ct_iid);
ISP_IOXGET_16(isp, &ctsrc->ct_rxid, ctdst->ct_rxid);
ISP_IOXGET_16(isp, &ctsrc->ct_flags, ctdst->ct_flags);
ISP_IOXGET_16(isp, &ctsrc->ct_status, ctdst->ct_status);
ISP_IOXGET_16(isp, &ctsrc->ct_timeout, ctdst->ct_timeout);
ISP_IOXGET_16(isp, &ctsrc->ct_seg_count, ctdst->ct_seg_count);
ISP_IOXGET_32(isp, &ctsrc->ct_reloff, ctdst->ct_reloff);
ISP_IOXGET_32(isp, &ctsrc->ct_resid, ctdst->ct_resid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_ctio2e(ispsoftc_t *isp, ct2e_entry_t *ctsrc, ct2e_entry_t *ctdst)
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
{
isp_copy_in_hdr(isp, &ctsrc->ct_header, &ctdst->ct_header);
ISP_IOXGET_16(isp, &ctsrc->ct_reserved, ctdst->ct_reserved);
ISP_IOXGET_16(isp, &ctsrc->ct_fwhandle, ctdst->ct_fwhandle);
ISP_IOXGET_16(isp, &ctsrc->ct_iid, ctdst->ct_iid);
ISP_IOXGET_16(isp, &ctsrc->ct_rxid, ctdst->ct_rxid);
ISP_IOXGET_16(isp, &ctsrc->ct_flags, ctdst->ct_flags);
ISP_IOXGET_16(isp, &ctsrc->ct_status, ctdst->ct_status);
ISP_IOXGET_16(isp, &ctsrc->ct_timeout, ctdst->ct_timeout);
ISP_IOXGET_16(isp, &ctsrc->ct_seg_count, ctdst->ct_seg_count);
ISP_IOXGET_32(isp, &ctsrc->ct_reloff, ctdst->ct_reloff);
ISP_IOXGET_32(isp, &ctsrc->ct_resid, ctdst->ct_resid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_enable_lun(ispsoftc_t *isp, lun_entry_t *lesrc, lun_entry_t *ledst)
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
{
int i;
isp_copy_out_hdr(isp, &lesrc->le_header, &ledst->le_header);
ISP_IOXPUT_32(isp, lesrc->le_reserved, &ledst->le_reserved);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, lesrc->le_lun, &ledst->le_rsvd);
ISP_IOXPUT_8(isp, lesrc->le_rsvd, &ledst->le_lun);
ISP_IOXPUT_8(isp, lesrc->le_ops, &ledst->le_tgt);
ISP_IOXPUT_8(isp, lesrc->le_tgt, &ledst->le_ops);
ISP_IOXPUT_8(isp, lesrc->le_status, &ledst->le_reserved2);
ISP_IOXPUT_8(isp, lesrc->le_reserved2, &ledst->le_status);
ISP_IOXPUT_8(isp, lesrc->le_cmd_count, &ledst->le_in_count);
ISP_IOXPUT_8(isp, lesrc->le_in_count, &ledst->le_cmd_count);
ISP_IOXPUT_8(isp, lesrc->le_cdb6len, &ledst->le_cdb7len);
ISP_IOXPUT_8(isp, lesrc->le_cdb7len, &ledst->le_cdb6len);
} else {
ISP_IOXPUT_8(isp, lesrc->le_lun, &ledst->le_lun);
ISP_IOXPUT_8(isp, lesrc->le_rsvd, &ledst->le_rsvd);
ISP_IOXPUT_8(isp, lesrc->le_ops, &ledst->le_ops);
ISP_IOXPUT_8(isp, lesrc->le_tgt, &ledst->le_tgt);
ISP_IOXPUT_8(isp, lesrc->le_status, &ledst->le_status);
ISP_IOXPUT_8(isp, lesrc->le_reserved2, &ledst->le_reserved2);
ISP_IOXPUT_8(isp, lesrc->le_cmd_count, &ledst->le_cmd_count);
ISP_IOXPUT_8(isp, lesrc->le_in_count, &ledst->le_in_count);
ISP_IOXPUT_8(isp, lesrc->le_cdb6len, &ledst->le_cdb6len);
ISP_IOXPUT_8(isp, lesrc->le_cdb7len, &ledst->le_cdb7len);
}
ISP_IOXPUT_32(isp, lesrc->le_flags, &ledst->le_flags);
ISP_IOXPUT_16(isp, lesrc->le_timeout, &ledst->le_timeout);
for (i = 0; i < 20; i++) {
ISP_IOXPUT_8(isp, lesrc->le_reserved3[i],
&ledst->le_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_enable_lun(ispsoftc_t *isp, lun_entry_t *lesrc, lun_entry_t *ledst)
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
{
int i;
isp_copy_in_hdr(isp, &lesrc->le_header, &ledst->le_header);
ISP_IOXGET_32(isp, &lesrc->le_reserved, ledst->le_reserved);
if (ISP_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &lesrc->le_lun, ledst->le_rsvd);
ISP_IOXGET_8(isp, &lesrc->le_rsvd, ledst->le_lun);
ISP_IOXGET_8(isp, &lesrc->le_ops, ledst->le_tgt);
ISP_IOXGET_8(isp, &lesrc->le_tgt, ledst->le_ops);
ISP_IOXGET_8(isp, &lesrc->le_status, ledst->le_reserved2);
ISP_IOXGET_8(isp, &lesrc->le_reserved2, ledst->le_status);
ISP_IOXGET_8(isp, &lesrc->le_cmd_count, ledst->le_in_count);
ISP_IOXGET_8(isp, &lesrc->le_in_count, ledst->le_cmd_count);
ISP_IOXGET_8(isp, &lesrc->le_cdb6len, ledst->le_cdb7len);
ISP_IOXGET_8(isp, &lesrc->le_cdb7len, ledst->le_cdb6len);
} else {
ISP_IOXGET_8(isp, &lesrc->le_lun, ledst->le_lun);
ISP_IOXGET_8(isp, &lesrc->le_rsvd, ledst->le_rsvd);
ISP_IOXGET_8(isp, &lesrc->le_ops, ledst->le_ops);
ISP_IOXGET_8(isp, &lesrc->le_tgt, ledst->le_tgt);
ISP_IOXGET_8(isp, &lesrc->le_status, ledst->le_status);
ISP_IOXGET_8(isp, &lesrc->le_reserved2, ledst->le_reserved2);
ISP_IOXGET_8(isp, &lesrc->le_cmd_count, ledst->le_cmd_count);
ISP_IOXGET_8(isp, &lesrc->le_in_count, ledst->le_in_count);
ISP_IOXGET_8(isp, &lesrc->le_cdb6len, ledst->le_cdb6len);
ISP_IOXGET_8(isp, &lesrc->le_cdb7len, ledst->le_cdb7len);
}
ISP_IOXGET_32(isp, &lesrc->le_flags, ledst->le_flags);
ISP_IOXGET_16(isp, &lesrc->le_timeout, ledst->le_timeout);
for (i = 0; i < 20; i++) {
ISP_IOXGET_8(isp, &lesrc->le_reserved3[i],
ledst->le_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_notify(ispsoftc_t *isp, in_entry_t *insrc, in_entry_t *indst)
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
{
int i;
isp_copy_out_hdr(isp, &insrc->in_header, &indst->in_header);
ISP_IOXPUT_32(isp, insrc->in_reserved, &indst->in_reserved);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, insrc->in_lun, &indst->in_iid);
ISP_IOXPUT_8(isp, insrc->in_iid, &indst->in_lun);
ISP_IOXPUT_8(isp, insrc->in_reserved2, &indst->in_tgt);
ISP_IOXPUT_8(isp, insrc->in_tgt, &indst->in_reserved2);
ISP_IOXPUT_8(isp, insrc->in_status, &indst->in_rsvd2);
ISP_IOXPUT_8(isp, insrc->in_rsvd2, &indst->in_status);
ISP_IOXPUT_8(isp, insrc->in_tag_val, &indst->in_tag_type);
ISP_IOXPUT_8(isp, insrc->in_tag_type, &indst->in_tag_val);
} else {
ISP_IOXPUT_8(isp, insrc->in_lun, &indst->in_lun);
ISP_IOXPUT_8(isp, insrc->in_iid, &indst->in_iid);
ISP_IOXPUT_8(isp, insrc->in_reserved2, &indst->in_reserved2);
ISP_IOXPUT_8(isp, insrc->in_tgt, &indst->in_tgt);
ISP_IOXPUT_8(isp, insrc->in_status, &indst->in_status);
ISP_IOXPUT_8(isp, insrc->in_rsvd2, &indst->in_rsvd2);
ISP_IOXPUT_8(isp, insrc->in_tag_val, &indst->in_tag_val);
ISP_IOXPUT_8(isp, insrc->in_tag_type, &indst->in_tag_type);
}
ISP_IOXPUT_32(isp, insrc->in_flags, &indst->in_flags);
ISP_IOXPUT_16(isp, insrc->in_seqid, &indst->in_seqid);
for (i = 0; i < IN_MSGLEN; i++) {
ISP_IOXPUT_8(isp, insrc->in_msg[i], &indst->in_msg[i]);
}
for (i = 0; i < IN_RSVDLEN; i++) {
ISP_IOXPUT_8(isp, insrc->in_reserved3[i],
&indst->in_reserved3[i]);
}
for (i = 0; i < QLTM_SENSELEN; i++) {
ISP_IOXPUT_8(isp, insrc->in_sense[i],
&indst->in_sense[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_notify(ispsoftc_t *isp, in_entry_t *insrc, in_entry_t *indst)
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
{
int i;
isp_copy_in_hdr(isp, &insrc->in_header, &indst->in_header);
ISP_IOXGET_32(isp, &insrc->in_reserved, indst->in_reserved);
if (ISP_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &insrc->in_lun, indst->in_iid);
ISP_IOXGET_8(isp, &insrc->in_iid, indst->in_lun);
ISP_IOXGET_8(isp, &insrc->in_reserved2, indst->in_tgt);
ISP_IOXGET_8(isp, &insrc->in_tgt, indst->in_reserved2);
ISP_IOXGET_8(isp, &insrc->in_status, indst->in_rsvd2);
ISP_IOXGET_8(isp, &insrc->in_rsvd2, indst->in_status);
ISP_IOXGET_8(isp, &insrc->in_tag_val, indst->in_tag_type);
ISP_IOXGET_8(isp, &insrc->in_tag_type, indst->in_tag_val);
} else {
ISP_IOXGET_8(isp, &insrc->in_lun, indst->in_lun);
ISP_IOXGET_8(isp, &insrc->in_iid, indst->in_iid);
ISP_IOXGET_8(isp, &insrc->in_reserved2, indst->in_reserved2);
ISP_IOXGET_8(isp, &insrc->in_tgt, indst->in_tgt);
ISP_IOXGET_8(isp, &insrc->in_status, indst->in_status);
ISP_IOXGET_8(isp, &insrc->in_rsvd2, indst->in_rsvd2);
ISP_IOXGET_8(isp, &insrc->in_tag_val, indst->in_tag_val);
ISP_IOXGET_8(isp, &insrc->in_tag_type, indst->in_tag_type);
}
ISP_IOXGET_32(isp, &insrc->in_flags, indst->in_flags);
ISP_IOXGET_16(isp, &insrc->in_seqid, indst->in_seqid);
for (i = 0; i < IN_MSGLEN; i++) {
ISP_IOXGET_8(isp, &insrc->in_msg[i], indst->in_msg[i]);
}
for (i = 0; i < IN_RSVDLEN; i++) {
ISP_IOXGET_8(isp, &insrc->in_reserved3[i],
indst->in_reserved3[i]);
}
for (i = 0; i < QLTM_SENSELEN; i++) {
ISP_IOXGET_8(isp, &insrc->in_sense[i],
indst->in_sense[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_notify_fc(ispsoftc_t *isp, in_fcentry_t *insrc,
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
in_fcentry_t *indst)
{
isp_copy_out_hdr(isp, &insrc->in_header, &indst->in_header);
ISP_IOXPUT_32(isp, insrc->in_reserved, &indst->in_reserved);
ISP_IOXPUT_8(isp, insrc->in_lun, &indst->in_lun);
ISP_IOXPUT_8(isp, insrc->in_iid, &indst->in_iid);
ISP_IOXPUT_16(isp, insrc->in_scclun, &indst->in_scclun);
ISP_IOXPUT_32(isp, insrc->in_reserved2, &indst->in_reserved2);
ISP_IOXPUT_16(isp, insrc->in_status, &indst->in_status);
ISP_IOXPUT_16(isp, insrc->in_task_flags, &indst->in_task_flags);
ISP_IOXPUT_16(isp, insrc->in_seqid, &indst->in_seqid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_notify_fc_e(ispsoftc_t *isp, in_fcentry_e_t *insrc,
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
in_fcentry_e_t *indst)
{
isp_copy_out_hdr(isp, &insrc->in_header, &indst->in_header);
ISP_IOXPUT_32(isp, insrc->in_reserved, &indst->in_reserved);
ISP_IOXPUT_16(isp, insrc->in_iid, &indst->in_iid);
ISP_IOXPUT_16(isp, insrc->in_scclun, &indst->in_scclun);
ISP_IOXPUT_32(isp, insrc->in_reserved2, &indst->in_reserved2);
ISP_IOXPUT_16(isp, insrc->in_status, &indst->in_status);
ISP_IOXPUT_16(isp, insrc->in_task_flags, &indst->in_task_flags);
ISP_IOXPUT_16(isp, insrc->in_seqid, &indst->in_seqid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_notify_fc(ispsoftc_t *isp, in_fcentry_t *insrc,
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
in_fcentry_t *indst)
{
isp_copy_in_hdr(isp, &insrc->in_header, &indst->in_header);
ISP_IOXGET_32(isp, &insrc->in_reserved, indst->in_reserved);
ISP_IOXGET_8(isp, &insrc->in_lun, indst->in_lun);
ISP_IOXGET_8(isp, &insrc->in_iid, indst->in_iid);
ISP_IOXGET_16(isp, &insrc->in_scclun, indst->in_scclun);
ISP_IOXGET_32(isp, &insrc->in_reserved2, indst->in_reserved2);
ISP_IOXGET_16(isp, &insrc->in_status, indst->in_status);
ISP_IOXGET_16(isp, &insrc->in_task_flags, indst->in_task_flags);
ISP_IOXGET_16(isp, &insrc->in_seqid, indst->in_seqid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_notify_fc_e(ispsoftc_t *isp, in_fcentry_e_t *insrc,
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
in_fcentry_e_t *indst)
{
isp_copy_in_hdr(isp, &insrc->in_header, &indst->in_header);
ISP_IOXGET_32(isp, &insrc->in_reserved, indst->in_reserved);
ISP_IOXGET_16(isp, &insrc->in_iid, indst->in_iid);
ISP_IOXGET_16(isp, &insrc->in_scclun, indst->in_scclun);
ISP_IOXGET_32(isp, &insrc->in_reserved2, indst->in_reserved2);
ISP_IOXGET_16(isp, &insrc->in_status, indst->in_status);
ISP_IOXGET_16(isp, &insrc->in_task_flags, indst->in_task_flags);
ISP_IOXGET_16(isp, &insrc->in_seqid, indst->in_seqid);
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_notify_ack(ispsoftc_t *isp, na_entry_t *nasrc, na_entry_t *nadst)
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
{
int i;
isp_copy_out_hdr(isp, &nasrc->na_header, &nadst->na_header);
ISP_IOXPUT_32(isp, nasrc->na_reserved, &nadst->na_reserved);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, nasrc->na_lun, &nadst->na_iid);
ISP_IOXPUT_8(isp, nasrc->na_iid, &nadst->na_lun);
ISP_IOXPUT_8(isp, nasrc->na_status, &nadst->na_event);
ISP_IOXPUT_8(isp, nasrc->na_event, &nadst->na_status);
} else {
ISP_IOXPUT_8(isp, nasrc->na_lun, &nadst->na_lun);
ISP_IOXPUT_8(isp, nasrc->na_iid, &nadst->na_iid);
ISP_IOXPUT_8(isp, nasrc->na_status, &nadst->na_status);
ISP_IOXPUT_8(isp, nasrc->na_event, &nadst->na_event);
}
ISP_IOXPUT_32(isp, nasrc->na_flags, &nadst->na_flags);
for (i = 0; i < NA_RSVDLEN; i++) {
ISP_IOXPUT_16(isp, nasrc->na_reserved3[i],
&nadst->na_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_notify_ack(ispsoftc_t *isp, na_entry_t *nasrc, na_entry_t *nadst)
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
{
int i;
isp_copy_in_hdr(isp, &nasrc->na_header, &nadst->na_header);
ISP_IOXGET_32(isp, &nasrc->na_reserved, nadst->na_reserved);
if (ISP_IS_SBUS(isp)) {
ISP_IOXGET_8(isp, &nasrc->na_lun, nadst->na_iid);
ISP_IOXGET_8(isp, &nasrc->na_iid, nadst->na_lun);
ISP_IOXGET_8(isp, &nasrc->na_status, nadst->na_event);
ISP_IOXGET_8(isp, &nasrc->na_event, nadst->na_status);
} else {
ISP_IOXGET_8(isp, &nasrc->na_lun, nadst->na_lun);
ISP_IOXGET_8(isp, &nasrc->na_iid, nadst->na_iid);
ISP_IOXGET_8(isp, &nasrc->na_status, nadst->na_status);
ISP_IOXGET_8(isp, &nasrc->na_event, nadst->na_event);
}
ISP_IOXGET_32(isp, &nasrc->na_flags, nadst->na_flags);
for (i = 0; i < NA_RSVDLEN; i++) {
ISP_IOXGET_16(isp, &nasrc->na_reserved3[i],
nadst->na_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_notify_ack_fc(ispsoftc_t *isp, na_fcentry_t *nasrc,
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
na_fcentry_t *nadst)
{
int i;
isp_copy_out_hdr(isp, &nasrc->na_header, &nadst->na_header);
ISP_IOXPUT_32(isp, nasrc->na_reserved, &nadst->na_reserved);
Fix na_fcentry_t to not have a lun field. Fix indentation in handly the notify structs. Fix messages in isp_got_msg_fc to print out the loop id of the sender- not the wwpn which will be synthesized later, if possible, in the outer layers. Put in debug printouts to pair a notify ack to a notify so one can see the start/close of an immediate notify event. Put in spsace for TASK MANAGEMENT response flags (which we don't do yet).
2006-08-04 20:20:00 +00:00
ISP_IOXPUT_8(isp, nasrc->na_reserved1, &nadst->na_reserved1);
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_IOXPUT_8(isp, nasrc->na_iid, &nadst->na_iid);
Fix na_fcentry_t to not have a lun field. Fix indentation in handly the notify structs. Fix messages in isp_got_msg_fc to print out the loop id of the sender- not the wwpn which will be synthesized later, if possible, in the outer layers. Put in debug printouts to pair a notify ack to a notify so one can see the start/close of an immediate notify event. Put in spsace for TASK MANAGEMENT response flags (which we don't do yet).
2006-08-04 20:20:00 +00:00
ISP_IOXPUT_16(isp, nasrc->na_response, &nadst->na_response);
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_IOXPUT_16(isp, nasrc->na_flags, &nadst->na_flags);
ISP_IOXPUT_16(isp, nasrc->na_reserved2, &nadst->na_reserved2);
ISP_IOXPUT_16(isp, nasrc->na_status, &nadst->na_status);
ISP_IOXPUT_16(isp, nasrc->na_task_flags, &nadst->na_task_flags);
ISP_IOXPUT_16(isp, nasrc->na_seqid, &nadst->na_seqid);
for (i = 0; i < NA2_RSVDLEN; i++) {
ISP_IOXPUT_16(isp, nasrc->na_reserved3[i],
&nadst->na_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_put_notify_ack_fc_e(ispsoftc_t *isp, na_fcentry_e_t *nasrc,
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
na_fcentry_e_t *nadst)
{
int i;
isp_copy_out_hdr(isp, &nasrc->na_header, &nadst->na_header);
ISP_IOXPUT_32(isp, nasrc->na_reserved, &nadst->na_reserved);
ISP_IOXPUT_16(isp, nasrc->na_iid, &nadst->na_iid);
Fix na_fcentry_t to not have a lun field. Fix indentation in handly the notify structs. Fix messages in isp_got_msg_fc to print out the loop id of the sender- not the wwpn which will be synthesized later, if possible, in the outer layers. Put in debug printouts to pair a notify ack to a notify so one can see the start/close of an immediate notify event. Put in spsace for TASK MANAGEMENT response flags (which we don't do yet).
2006-08-04 20:20:00 +00:00
ISP_IOXPUT_16(isp, nasrc->na_response, &nadst->na_response);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
ISP_IOXPUT_16(isp, nasrc->na_flags, &nadst->na_flags);
ISP_IOXPUT_16(isp, nasrc->na_reserved2, &nadst->na_reserved2);
ISP_IOXPUT_16(isp, nasrc->na_status, &nadst->na_status);
ISP_IOXPUT_16(isp, nasrc->na_task_flags, &nadst->na_task_flags);
ISP_IOXPUT_16(isp, nasrc->na_seqid, &nadst->na_seqid);
for (i = 0; i < NA2_RSVDLEN; i++) {
ISP_IOXPUT_16(isp, nasrc->na_reserved3[i],
&nadst->na_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_notify_ack_fc(ispsoftc_t *isp, na_fcentry_t *nasrc,
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
na_fcentry_t *nadst)
{
int i;
isp_copy_in_hdr(isp, &nasrc->na_header, &nadst->na_header);
ISP_IOXGET_32(isp, &nasrc->na_reserved, nadst->na_reserved);
Fix na_fcentry_t to not have a lun field. Fix indentation in handly the notify structs. Fix messages in isp_got_msg_fc to print out the loop id of the sender- not the wwpn which will be synthesized later, if possible, in the outer layers. Put in debug printouts to pair a notify ack to a notify so one can see the start/close of an immediate notify event. Put in spsace for TASK MANAGEMENT response flags (which we don't do yet).
2006-08-04 20:20:00 +00:00
ISP_IOXGET_8(isp, &nasrc->na_reserved1, nadst->na_reserved1);
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_IOXGET_8(isp, &nasrc->na_iid, nadst->na_iid);
Fix na_fcentry_t to not have a lun field. Fix indentation in handly the notify structs. Fix messages in isp_got_msg_fc to print out the loop id of the sender- not the wwpn which will be synthesized later, if possible, in the outer layers. Put in debug printouts to pair a notify ack to a notify so one can see the start/close of an immediate notify event. Put in spsace for TASK MANAGEMENT response flags (which we don't do yet).
2006-08-04 20:20:00 +00:00
ISP_IOXGET_16(isp, &nasrc->na_response, nadst->na_response);
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_IOXGET_16(isp, &nasrc->na_flags, nadst->na_flags);
ISP_IOXGET_16(isp, &nasrc->na_reserved2, nadst->na_reserved2);
ISP_IOXGET_16(isp, &nasrc->na_status, nadst->na_status);
ISP_IOXGET_16(isp, &nasrc->na_task_flags, nadst->na_task_flags);
ISP_IOXGET_16(isp, &nasrc->na_seqid, nadst->na_seqid);
for (i = 0; i < NA2_RSVDLEN; i++) {
ISP_IOXGET_16(isp, &nasrc->na_reserved3[i],
nadst->na_reserved3[i]);
}
}
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
void
a) clean up some declaration stuff (i.e., make more modern with respect to getting rid u_int for uint and so on). b) Turn back on 64 bit DAC support. Cheeze it a bit in that we have two DMA callback functions- one when we have bus_addr_t > 4 bits in width and the other which should be normal. Even Cheezier in that we turn off setting up DMA maps to be BUS_SPACE_MAXADDR if we're in ISP_TARGET_MODE. More work on this in a week or so. c) Tested under amd64 and 1MB DFLTPHYS, sparc64, i386 (PAE, but insufficient memory to really test > 4GB). LINT check under amd64. MFC after: 1 month
2006-02-15 00:31:48 +00:00
isp_get_notify_ack_fc_e(ispsoftc_t *isp, na_fcentry_e_t *nasrc,
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
na_fcentry_e_t *nadst)
{
int i;
isp_copy_in_hdr(isp, &nasrc->na_header, &nadst->na_header);
ISP_IOXGET_32(isp, &nasrc->na_reserved, nadst->na_reserved);
ISP_IOXGET_16(isp, &nasrc->na_iid, nadst->na_iid);
Fix na_fcentry_t to not have a lun field. Fix indentation in handly the notify structs. Fix messages in isp_got_msg_fc to print out the loop id of the sender- not the wwpn which will be synthesized later, if possible, in the outer layers. Put in debug printouts to pair a notify ack to a notify so one can see the start/close of an immediate notify event. Put in spsace for TASK MANAGEMENT response flags (which we don't do yet).
2006-08-04 20:20:00 +00:00
ISP_IOXGET_16(isp, &nasrc->na_response, nadst->na_response);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
ISP_IOXGET_16(isp, &nasrc->na_flags, nadst->na_flags);
ISP_IOXGET_16(isp, &nasrc->na_reserved2, nadst->na_reserved2);
ISP_IOXGET_16(isp, &nasrc->na_status, nadst->na_status);
ISP_IOXGET_16(isp, &nasrc->na_task_flags, nadst->na_task_flags);
ISP_IOXGET_16(isp, &nasrc->na_seqid, nadst->na_seqid);
for (i = 0; i < NA2_RSVDLEN; i++) {
ISP_IOXGET_16(isp, &nasrc->na_reserved3[i],
nadst->na_reserved3[i]);
}
}
Remove use of inlines and use the functions as a library. Larger code space, possibly performance hit, but more portable. Certainly less questionable use of inlining. Suggested by: des
2006-02-02 21:31:34 +00:00
#endif /* ISP_TARGET_MODE */
Reference in New Issue Copy Permalink