d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
c862b05ba7
freebsd-nq/sys/dev/isp/isp_library.c

2905 lines
105 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
/*-
Fix some stupid copyright mistakes that have been there for quite some time.
2007-03-10 02:39:54 +00:00
* Copyright (c) 1997-2007 by Matthew Jacob
* 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, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY 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 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.
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
*/
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__
Fix some stupid copyright mistakes that have been there for quite some time.
2007-03-10 02:39:54 +00:00
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD$");
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
#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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_save_xs(ispsoftc_t *isp, XS_T *xs, uint32_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
uint16_t i, 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
for (j = isp->isp_lasthdls, i = 0; i < isp->isp_maxcmds; i++) {
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
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
if (++j == 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
j = 0;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
}
isp->isp_lasthdls = (uint32_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 *
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_find_xs(ispsoftc_t *isp, uint32_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
if (handle < 1 || handle > (uint32_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]);
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
uint32_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_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
uint16_t i;
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
if (xs != NULL) {
for (i = 0; i < isp->isp_maxcmds; i++) {
if (isp->isp_xflist[i] == xs) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
return ((uint32_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);
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
uint32_t
isp_handle_index(uint32_t handle)
Store the target handles in a separate list from normal commands. Add a CTIO fast post routine to handle CTIO completions. Submitted by: mjacob
2004-05-24 07:02:25 +00:00
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
return (handle - 1);
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
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_destroy_handle(ispsoftc_t *isp, uint32_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
if (handle > 0 && handle <= (uint32_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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_getrqentry(ispsoftc_t *isp, uint32_t *iptrp,
uint32_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
volatile uint32_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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
optr = isp->isp_reqodx = ISP_READ(isp, isp->isp_rqstoutrp);
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
MFP4: a) Some constification from NetBSD (gcc 4.1.2) b) Split default param fetching/setting into scsi and fibre functions and retry the fibre fetch more than once. MFC after: 1 week
2007-03-22 23:38:32 +00:00
isp_print_bytes(ispsoftc_t *isp, const 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 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 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
if (isp->isp_role == ISP_ROLE_NONE) {
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);
}
fcp = FCPARAM(isp);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
tptr = &tval;
if (fcp->isp_fwstate < FW_READY ||
fcp->isp_loopstate < LOOP_PDB_RCVD) {
if (isp_control(isp, ISPCTL_FCLINK_TEST, tptr) != 0) {
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
isp_prt(isp, ISP_LOGSANCFG,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
"isp_fc_runstate: linktest failed");
return (-1);
}
if (fcp->isp_fwstate != FW_READY ||
fcp->isp_loopstate < LOOP_PDB_RCVD) {
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
isp_prt(isp, ISP_LOGSANCFG,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
"isp_fc_runstate: f/w not ready");
return (-1);
}
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->isp_role & ISP_ROLE_INITIATOR) == 0) {
return (0);
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
if (isp_control(isp, ISPCTL_SCAN_LOOP, NULL) != 0) {
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
isp_prt(isp, ISP_LOGSANCFG,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
"isp_fc_runstate: scan loop fails");
return (LOOP_PDB_RCVD);
}
if (isp_control(isp, ISPCTL_SCAN_FABRIC, NULL) != 0) {
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
isp_prt(isp, ISP_LOGSANCFG,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
"isp_fc_runstate: scan fabric fails");
return (LOOP_LSCAN_DONE);
}
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_PDB_SYNC, NULL) != 0) {
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
isp_prt(isp, ISP_LOGSANCFG, "isp_fc_runstate: pdb_sync fails");
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
return (LOOP_FSCAN_DONE);
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 (fcp->isp_fwstate != FW_READY || fcp->isp_loopstate != LOOP_READY) {
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
isp_prt(isp, ISP_LOGSANCFG,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
"isp_fc_runstate: f/w not ready again");
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 (-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
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
/*
* Fibre Channel Support- get the port database for the id.
*/
void
isp_dump_portdb(ispsoftc_t *isp)
{
fcparam *fcp = (fcparam *) isp->isp_param;
int i;
for (i = 0; i < MAX_FC_TARG; i++) {
char mb[4];
const char *dbs[8] = {
"NIL ",
"PROB",
"DEAD",
"CHGD",
"NEW ",
"PVLD",
"ZOMB",
"VLD "
};
const char *roles[4] = {
" UNK", " TGT", " INI", "TINI"
};
fcportdb_t *lp = &fcp->portdb[i];
if (lp->state == FC_PORTDB_STATE_NIL) {
continue;
}
if (lp->ini_map_idx) {
SNPRINTF(mb, sizeof (mb), "%3d",
((int) lp->ini_map_idx) - 1);
} else {
SNPRINTF(mb, sizeof (mb), "---");
}
Increase the timeout for some SAN commands. Only complain about FC Reponse errors if they're nonzero. Shorten some PortID printouts for local loop. Add an internal isp_xcmd_t data structure which we'll use for some CT-Passthru support as part of adding SMI-S.
2006-11-16 00:39:56 +00:00
isp_prt(isp, ISP_LOGALL, "%d: hdl 0x%x %s al%d tgt %s %s "
"0x%06x =>%s 0x%06x; WWNN 0x%08x%08x WWPN 0x%08x%08x", i,
lp->handle, dbs[lp->state], lp->autologin, mb,
Push things closer to path failover by implementing loop down and gone device timers and zombie state entries. There are tunables that can be used to select a number of parameters. loop_down_limit - how long to wait for loop to come back up before declaring all devices dead (default 300 seconds) gone_device_time- how long to wait for a device that has appeared to leave the loop or fabric to reappear (default 30 seconds) Internal tunables include (which should be externalized): quick_boot_time- how long to wait when booting for loop to come up change_is_bad- whether or not to accept devices with the same WWNN/WWPN that reappear at a different PortID as being the 'same' device. Keen students of some of the subtle issues here will ask how one can keep devices from being re-accepted at all (the answer is to set a gone_device_time to zero- that effectively would be the same thing).
2006-11-14 08:45:48 +00:00
roles[lp->roles], lp->portid,
roles[lp->new_roles], lp->new_portid,
(uint32_t) (lp->node_wwn >> 32),
(uint32_t) (lp->node_wwn),
(uint32_t) (lp->port_wwn >> 32),
(uint32_t) (lp->port_wwn));
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
void
isp_shutdown(ispsoftc_t *isp)
{
if (IS_FC(isp)) {
if (IS_24XX(isp)) {
ISP_WRITE(isp, BIU2400_ICR, 0);
ISP_WRITE(isp, BIU2400_HCCR, HCCR_2400_CMD_PAUSE);
} else {
ISP_WRITE(isp, BIU_ICR, 0);
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
ISP_WRITE(isp, BIU2100_CSR, BIU2100_FPM0_REGS);
ISP_WRITE(isp, FPM_DIAG_CONFIG, FPM_SOFT_RESET);
ISP_WRITE(isp, BIU2100_CSR, BIU2100_FB_REGS);
ISP_WRITE(isp, FBM_CMD, FBMCMD_FIFO_RESET_ALL);
ISP_WRITE(isp, BIU2100_CSR, BIU2100_RISC_REGS);
}
} else {
ISP_WRITE(isp, BIU_ICR, 0);
ISP_WRITE(isp, HCCR, HCCR_CMD_PAUSE);
}
}
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)
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
#define ASIZE(x) (sizeof (x) / sizeof (x[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
/*
* 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &rqsrc->req_header, &rqdst->req_header);
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, 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);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
for (i = 0; i < ASIZE(rqsrc->req_cdb); 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_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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_marker(ispsoftc_t *isp, isp_marker_t *src, isp_marker_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->mrk_header, &dst->mrk_header);
ISP_IOXPUT_32(isp, src->mrk_handle, &dst->mrk_handle);
if (ISP_IS_SBUS(isp)) {
ISP_IOXPUT_8(isp, src->mrk_reserved0, &dst->mrk_target);
ISP_IOXPUT_8(isp, src->mrk_target, &dst->mrk_reserved0);
} else {
ISP_IOXPUT_8(isp, src->mrk_reserved0, &dst->mrk_reserved0);
ISP_IOXPUT_8(isp, src->mrk_target, &dst->mrk_target);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->mrk_modifier, &dst->mrk_modifier);
ISP_IOXPUT_16(isp, src->mrk_flags, &dst->mrk_flags);
ISP_IOXPUT_16(isp, src->mrk_lun, &dst->mrk_lun);
for (i = 0; i < ASIZE(src->mrk_reserved1); i++) {
ISP_IOXPUT_8(isp, src->mrk_reserved1[i],
&dst->mrk_reserved1[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_marker_24xx(ispsoftc_t *isp,
isp_marker_24xx_t *src, isp_marker_24xx_t *dst)
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->mrk_header, &dst->mrk_header);
ISP_IOXPUT_32(isp, src->mrk_handle, &dst->mrk_handle);
ISP_IOXPUT_16(isp, src->mrk_nphdl, &dst->mrk_nphdl);
ISP_IOXPUT_8(isp, src->mrk_modifier, &dst->mrk_modifier);
ISP_IOXPUT_8(isp, src->mrk_reserved0, &dst->mrk_reserved0);
ISP_IOXPUT_8(isp, src->mrk_reserved1, &dst->mrk_reserved1);
ISP_IOXPUT_8(isp, src->mrk_vphdl, &dst->mrk_vphdl);
ISP_IOXPUT_8(isp, src->mrk_reserved2, &dst->mrk_reserved2);
for (i = 0; i < ASIZE(src->mrk_lun); i++) {
ISP_IOXPUT_8(isp, src->mrk_lun[i], &dst->mrk_lun[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
for (i = 0; i < ASIZE(src->mrk_reserved3); i++) {
ISP_IOXPUT_8(isp, src->mrk_reserved3[i],
&dst->mrk_reserved3[i]);
}
}
void
isp_put_request_t2(ispsoftc_t *isp, ispreqt2_t *src, ispreqt2_t *dst)
{
int i;
isp_put_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXPUT_32(isp, src->req_handle, &dst->req_handle);
ISP_IOXPUT_8(isp, src->req_lun_trn, &dst->req_lun_trn);
ISP_IOXPUT_8(isp, src->req_target, &dst->req_target);
ISP_IOXPUT_16(isp, src->req_scclun, &dst->req_scclun);
ISP_IOXPUT_16(isp, src->req_flags, &dst->req_flags);
ISP_IOXPUT_16(isp, src->req_reserved, &dst->req_reserved);
ISP_IOXPUT_16(isp, src->req_time, &dst->req_time);
ISP_IOXPUT_16(isp, src->req_seg_count, &dst->req_seg_count);
for (i = 0; i < ASIZE(src->req_cdb); i++) {
ISP_IOXPUT_8(isp, src->req_cdb[i], &dst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, src->req_totalcnt, &dst->req_totalcnt);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < ISP_RQDSEG_T2; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_base,
&dst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_count,
&dst->req_dataseg[i].ds_count);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_request_t2e(ispsoftc_t *isp, ispreqt2e_t *src, ispreqt2e_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXPUT_32(isp, src->req_handle, &dst->req_handle);
ISP_IOXPUT_16(isp, src->req_target, &dst->req_target);
ISP_IOXPUT_16(isp, src->req_scclun, &dst->req_scclun);
ISP_IOXPUT_16(isp, src->req_flags, &dst->req_flags);
ISP_IOXPUT_16(isp, src->req_reserved, &dst->req_reserved);
ISP_IOXPUT_16(isp, src->req_time, &dst->req_time);
ISP_IOXPUT_16(isp, src->req_seg_count, &dst->req_seg_count);
for (i = 0; i < ASIZE(src->req_cdb); i++) {
ISP_IOXPUT_8(isp, src->req_cdb[i], &dst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, src->req_totalcnt, &dst->req_totalcnt);
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_base,
&dst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_count,
&dst->req_dataseg[i].ds_count);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
}
void
isp_put_request_t3(ispsoftc_t *isp, ispreqt3_t *src, ispreqt3_t *dst)
{
int i;
isp_put_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXPUT_32(isp, src->req_handle, &dst->req_handle);
ISP_IOXPUT_8(isp, src->req_lun_trn, &dst->req_lun_trn);
ISP_IOXPUT_8(isp, src->req_target, &dst->req_target);
ISP_IOXPUT_16(isp, src->req_scclun, &dst->req_scclun);
ISP_IOXPUT_16(isp, src->req_flags, &dst->req_flags);
ISP_IOXPUT_16(isp, src->req_reserved, &dst->req_reserved);
ISP_IOXPUT_16(isp, src->req_time, &dst->req_time);
ISP_IOXPUT_16(isp, src->req_seg_count, &dst->req_seg_count);
for (i = 0; i < ASIZE(src->req_cdb); i++) {
ISP_IOXPUT_8(isp, src->req_cdb[i], &dst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, src->req_totalcnt, &dst->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++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_base,
&dst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_basehi,
&dst->req_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_count,
&dst->req_dataseg[i].ds_count);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_request_t3e(ispsoftc_t *isp, ispreqt3e_t *src, ispreqt3e_t *dst)
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXPUT_32(isp, src->req_handle, &dst->req_handle);
ISP_IOXPUT_16(isp, src->req_target, &dst->req_target);
ISP_IOXPUT_16(isp, src->req_scclun, &dst->req_scclun);
ISP_IOXPUT_16(isp, src->req_flags, &dst->req_flags);
ISP_IOXPUT_16(isp, src->req_reserved, &dst->req_reserved);
ISP_IOXPUT_16(isp, src->req_time, &dst->req_time);
ISP_IOXPUT_16(isp, src->req_seg_count, &dst->req_seg_count);
for (i = 0; i < ASIZE(src->req_cdb); i++) {
ISP_IOXPUT_8(isp, src->req_cdb[i], &dst->req_cdb[i]);
}
ISP_IOXPUT_32(isp, src->req_totalcnt, &dst->req_totalcnt);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < ISP_RQDSEG_T3; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_base,
&dst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_basehi,
&dst->req_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_count,
&dst->req_dataseg[i].ds_count);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_extended_request(ispsoftc_t *isp, ispextreq_t *src, ispextreq_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXPUT_32(isp, src->req_handle, &dst->req_handle);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (ISP_IS_SBUS(isp)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->req_lun_trn, &dst->req_target);
ISP_IOXPUT_8(isp, src->req_target, &dst->req_lun_trn);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->req_lun_trn, &dst->req_lun_trn);
ISP_IOXPUT_8(isp, src->req_target, &dst->req_target);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->req_cdblen, &dst->req_cdblen);
ISP_IOXPUT_16(isp, src->req_flags, &dst->req_flags);
ISP_IOXPUT_16(isp, src->req_time, &dst->req_time);
ISP_IOXPUT_16(isp, src->req_seg_count, &dst->req_seg_count);
for (i = 0; i < ASIZE(src->req_cdb); i++) {
ISP_IOXPUT_8(isp, src->req_cdb[i], &dst->req_cdb[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_request_t7(ispsoftc_t *isp, ispreqt7_t *src, ispreqt7_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
uint32_t *a, *b;
isp_put_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXPUT_32(isp, src->req_handle, &dst->req_handle);
ISP_IOXPUT_16(isp, src->req_nphdl, &dst->req_nphdl);
ISP_IOXPUT_16(isp, src->req_time, &dst->req_time);
ISP_IOXPUT_16(isp, src->req_seg_count, &dst->req_seg_count);
ISP_IOXPUT_16(isp, src->req_reserved, &dst->req_reserved);
a = (uint32_t *) src->req_lun;
b = (uint32_t *) dst->req_lun;
for (i = 0; i < (ASIZE(src->req_lun) >> 2); i++ ) {
Recover from some major omissions/problems with the 24XX port. First, we were never correctly checking for a 24XX Status Type 0 response- that cased us to fall through to evaluate status for commands as if this were a 2100/2200/2300 Status Type 0 response. This is *close*, but not quite the same. This has been reported to be apparent with some wierd lun configuration problems with some arrays. It became glaringly apparent on sparc64 where none of the correct byte swap things were done. Fixing this omission then caused a whole universe shifting debug cycle of endian issues for the 2400. The manual for 24XX f/w turns out to be wrong about the endianness of a couple of entities. The lun and cdb fields for the type 7 request are *not* unconditionally big endian- they happen to be opposite of whatever the endian of the current machine type is. Same with the sense data for the 24XX type 0 response. While we're at it investigate and resolve some NVRAM endian issues. Approved by: re (ken) MFC after: 3 days
2007-07-02 20:08:20 +00:00
*b++ = ISP_SWAP32(isp, *a++);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
}
ISP_IOXPUT_8(isp, src->req_alen_datadir, &dst->req_alen_datadir);
ISP_IOXPUT_8(isp, src->req_task_management, &dst->req_task_management);
ISP_IOXPUT_8(isp, src->req_task_attribute, &dst->req_task_attribute);
ISP_IOXPUT_8(isp, src->req_crn, &dst->req_crn);
a = (uint32_t *) src->req_cdb;
b = (uint32_t *) dst->req_cdb;
Recover from some major omissions/problems with the 24XX port. First, we were never correctly checking for a 24XX Status Type 0 response- that cased us to fall through to evaluate status for commands as if this were a 2100/2200/2300 Status Type 0 response. This is *close*, but not quite the same. This has been reported to be apparent with some wierd lun configuration problems with some arrays. It became glaringly apparent on sparc64 where none of the correct byte swap things were done. Fixing this omission then caused a whole universe shifting debug cycle of endian issues for the 2400. The manual for 24XX f/w turns out to be wrong about the endianness of a couple of entities. The lun and cdb fields for the type 7 request are *not* unconditionally big endian- they happen to be opposite of whatever the endian of the current machine type is. Same with the sense data for the 24XX type 0 response. While we're at it investigate and resolve some NVRAM endian issues. Approved by: re (ken) MFC after: 3 days
2007-07-02 20:08:20 +00:00
for (i = 0; i < (ASIZE(src->req_cdb) >> 2); i++) {
*b++ = ISP_SWAP32(isp, *a++);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
}
ISP_IOXPUT_32(isp, src->req_dl, &dst->req_dl);
ISP_IOXPUT_16(isp, src->req_tidlo, &dst->req_tidlo);
ISP_IOXPUT_8(isp, src->req_tidhi, &dst->req_tidhi);
ISP_IOXPUT_8(isp, src->req_vpidx, &dst->req_vpidx);
ISP_IOXPUT_32(isp, src->req_dataseg.ds_base,
&dst->req_dataseg.ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg.ds_basehi,
&dst->req_dataseg.ds_basehi);
ISP_IOXPUT_32(isp, src->req_dataseg.ds_count,
&dst->req_dataseg.ds_count);
}
void
isp_put_24xx_abrt(ispsoftc_t *isp, isp24xx_abrt_t *src, isp24xx_abrt_t *dst)
{
int i;
isp_put_hdr(isp, &src->abrt_header, &dst->abrt_header);
ISP_IOXPUT_32(isp, src->abrt_handle, &dst->abrt_handle);
ISP_IOXPUT_16(isp, src->abrt_nphdl, &dst->abrt_nphdl);
ISP_IOXPUT_16(isp, src->abrt_options, &dst->abrt_options);
ISP_IOXPUT_32(isp, src->abrt_cmd_handle, &dst->abrt_cmd_handle);
for (i = 0; i < ASIZE(src->abrt_reserved); i++) {
ISP_IOXPUT_8(isp, src->abrt_reserved[i],
&dst->abrt_reserved[i]);
}
ISP_IOXPUT_16(isp, src->abrt_tidlo, &dst->abrt_tidlo);
ISP_IOXPUT_8(isp, src->abrt_tidhi, &dst->abrt_tidhi);
ISP_IOXPUT_8(isp, src->abrt_vpidx, &dst->abrt_vpidx);
for (i = 0; i < ASIZE(src->abrt_reserved1); i++) {
ISP_IOXPUT_8(isp, src->abrt_reserved1[i],
&dst->abrt_reserved1[i]);
}
}
void
isp_put_cont_req(ispsoftc_t *isp, ispcontreq_t *src, ispcontreq_t *dst)
{
int i;
isp_put_hdr(isp, &src->req_header, &dst->req_header);
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 < ISP_CDSEG; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_base,
&dst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_count,
&dst->req_dataseg[i].ds_count);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_cont64_req(ispsoftc_t *isp, ispcontreq64_t *src, ispcontreq64_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->req_header, &dst->req_header);
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 < ISP_CDSEG64; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_base,
&dst->req_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_basehi,
&dst->req_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, src->req_dataseg[i].ds_count,
&dst->req_dataseg[i].ds_count);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_response(ispsoftc_t *isp, ispstatusreq_t *src, ispstatusreq_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXGET_32(isp, &src->req_handle, dst->req_handle);
ISP_IOXGET_16(isp, &src->req_scsi_status, dst->req_scsi_status);
ISP_IOXGET_16(isp, &src->req_completion_status,
dst->req_completion_status);
ISP_IOXGET_16(isp, &src->req_state_flags, dst->req_state_flags);
ISP_IOXGET_16(isp, &src->req_status_flags, dst->req_status_flags);
ISP_IOXGET_16(isp, &src->req_time, dst->req_time);
ISP_IOXGET_16(isp, &src->req_sense_len, dst->req_sense_len);
ISP_IOXGET_32(isp, &src->req_resid, dst->req_resid);
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++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->req_response[i],
dst->req_response[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
}
for (i = 0; i < 32; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->req_sense_data[i],
dst->req_sense_data[i]);
}
}
void
isp_get_24xx_response(ispsoftc_t *isp, isp24xx_statusreq_t *src,
isp24xx_statusreq_t *dst)
{
int i;
Recover from some major omissions/problems with the 24XX port. First, we were never correctly checking for a 24XX Status Type 0 response- that cased us to fall through to evaluate status for commands as if this were a 2100/2200/2300 Status Type 0 response. This is *close*, but not quite the same. This has been reported to be apparent with some wierd lun configuration problems with some arrays. It became glaringly apparent on sparc64 where none of the correct byte swap things were done. Fixing this omission then caused a whole universe shifting debug cycle of endian issues for the 2400. The manual for 24XX f/w turns out to be wrong about the endianness of a couple of entities. The lun and cdb fields for the type 7 request are *not* unconditionally big endian- they happen to be opposite of whatever the endian of the current machine type is. Same with the sense data for the 24XX type 0 response. While we're at it investigate and resolve some NVRAM endian issues. Approved by: re (ken) MFC after: 3 days
2007-07-02 20:08:20 +00:00
uint32_t *s, *d;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->req_header, &dst->req_header);
ISP_IOXGET_32(isp, &src->req_handle, dst->req_handle);
ISP_IOXGET_16(isp, &src->req_completion_status,
dst->req_completion_status);
ISP_IOXGET_16(isp, &src->req_oxid, dst->req_oxid);
ISP_IOXGET_32(isp, &src->req_resid, dst->req_resid);
ISP_IOXGET_16(isp, &src->req_reserved0, dst->req_reserved0);
ISP_IOXGET_16(isp, &src->req_state_flags, dst->req_state_flags);
ISP_IOXGET_16(isp, &src->req_reserved1, dst->req_reserved1);
ISP_IOXGET_16(isp, &src->req_scsi_status, dst->req_scsi_status);
ISP_IOXGET_32(isp, &src->req_fcp_residual, dst->req_fcp_residual);
ISP_IOXGET_32(isp, &src->req_sense_len, dst->req_sense_len);
ISP_IOXGET_32(isp, &src->req_response_len, dst->req_response_len);
Recover from some major omissions/problems with the 24XX port. First, we were never correctly checking for a 24XX Status Type 0 response- that cased us to fall through to evaluate status for commands as if this were a 2100/2200/2300 Status Type 0 response. This is *close*, but not quite the same. This has been reported to be apparent with some wierd lun configuration problems with some arrays. It became glaringly apparent on sparc64 where none of the correct byte swap things were done. Fixing this omission then caused a whole universe shifting debug cycle of endian issues for the 2400. The manual for 24XX f/w turns out to be wrong about the endianness of a couple of entities. The lun and cdb fields for the type 7 request are *not* unconditionally big endian- they happen to be opposite of whatever the endian of the current machine type is. Same with the sense data for the 24XX type 0 response. While we're at it investigate and resolve some NVRAM endian issues. Approved by: re (ken) MFC after: 3 days
2007-07-02 20:08:20 +00:00
s = (uint32_t *)src->req_rsp_sense;
d = (uint32_t *)dst->req_rsp_sense;
for (i = 0; i < (ASIZE(src->req_rsp_sense) >> 2); i++) {
d[i] = ISP_SWAP32(isp, s[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_24xx_abrt(ispsoftc_t *isp, isp24xx_abrt_t *src, isp24xx_abrt_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->abrt_header, &dst->abrt_header);
ISP_IOXGET_32(isp, &src->abrt_handle, dst->abrt_handle);
ISP_IOXGET_16(isp, &src->abrt_nphdl, dst->abrt_nphdl);
ISP_IOXGET_16(isp, &src->abrt_options, dst->abrt_options);
ISP_IOXGET_32(isp, &src->abrt_cmd_handle, dst->abrt_cmd_handle);
Recover from some major omissions/problems with the 24XX port. First, we were never correctly checking for a 24XX Status Type 0 response- that cased us to fall through to evaluate status for commands as if this were a 2100/2200/2300 Status Type 0 response. This is *close*, but not quite the same. This has been reported to be apparent with some wierd lun configuration problems with some arrays. It became glaringly apparent on sparc64 where none of the correct byte swap things were done. Fixing this omission then caused a whole universe shifting debug cycle of endian issues for the 2400. The manual for 24XX f/w turns out to be wrong about the endianness of a couple of entities. The lun and cdb fields for the type 7 request are *not* unconditionally big endian- they happen to be opposite of whatever the endian of the current machine type is. Same with the sense data for the 24XX type 0 response. While we're at it investigate and resolve some NVRAM endian issues. Approved by: re (ken) MFC after: 3 days
2007-07-02 20:08:20 +00:00
for (i = 0; i < ASIZE(src->abrt_reserved); i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->abrt_reserved[i],
dst->abrt_reserved[i]);
}
ISP_IOXGET_16(isp, &src->abrt_tidlo, dst->abrt_tidlo);
ISP_IOXGET_8(isp, &src->abrt_tidhi, dst->abrt_tidhi);
ISP_IOXGET_8(isp, &src->abrt_vpidx, dst->abrt_vpidx);
Recover from some major omissions/problems with the 24XX port. First, we were never correctly checking for a 24XX Status Type 0 response- that cased us to fall through to evaluate status for commands as if this were a 2100/2200/2300 Status Type 0 response. This is *close*, but not quite the same. This has been reported to be apparent with some wierd lun configuration problems with some arrays. It became glaringly apparent on sparc64 where none of the correct byte swap things were done. Fixing this omission then caused a whole universe shifting debug cycle of endian issues for the 2400. The manual for 24XX f/w turns out to be wrong about the endianness of a couple of entities. The lun and cdb fields for the type 7 request are *not* unconditionally big endian- they happen to be opposite of whatever the endian of the current machine type is. Same with the sense data for the 24XX type 0 response. While we're at it investigate and resolve some NVRAM endian issues. Approved by: re (ken) MFC after: 3 days
2007-07-02 20:08:20 +00:00
for (i = 0; i < ASIZE(src->abrt_reserved1); i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->abrt_reserved1[i],
dst->abrt_reserved1[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
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +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_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &r2src->req_header, &r2dst->req_header);
if (r2dst->req_header.rqs_seqno > 30) {
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
r2dst->req_header.rqs_seqno = 30;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
}
Implement REDUCED INTERRUPT OPERATION usage form FC cards- this allows the firmware to delay completion of commands so that it can attempt to batch a bunch of completions at once- either returning 16 bit handles in mailbox registers, or in a resposne queue entry that has a whole wad of 16 bit handles. Distinguish between 2300 and 2312 chipsets- if only because the revisions on the chips have different meanings. Add more instrumentation plus ISP_GET_STATS and ISP_CLR_STATS ioctls. Run up the maximum number of response queue entities we'll look at per interrupt. If we haven't set HBA role yet, always return success from isp_fc_runstate. MFC after: 2 weeks
2002-01-03 20:43:22 +00:00
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_icb(ispsoftc_t *isp, isp_icb_t *src, isp_icb_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;
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_version, &dst->icb_reserved0);
ISP_IOXPUT_8(isp, src->icb_reserved0, &dst->icb_version);
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
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_version, &dst->icb_version);
ISP_IOXPUT_8(isp, src->icb_reserved0, &dst->icb_reserved0);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_fwoptions, &dst->icb_fwoptions);
ISP_IOXPUT_16(isp, src->icb_maxfrmlen, &dst->icb_maxfrmlen);
ISP_IOXPUT_16(isp, src->icb_maxalloc, &dst->icb_maxalloc);
ISP_IOXPUT_16(isp, src->icb_execthrottle, &dst->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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_retry_count, &dst->icb_retry_delay);
ISP_IOXPUT_8(isp, src->icb_retry_delay, &dst->icb_retry_count);
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
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_retry_count, &dst->icb_retry_count);
ISP_IOXPUT_8(isp, src->icb_retry_delay, &dst->icb_retry_delay);
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
}
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++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_portname[i], &dst->icb_portname[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_hardaddr, &dst->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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_iqdevtype, &dst->icb_logintime);
ISP_IOXPUT_8(isp, src->icb_logintime, &dst->icb_iqdevtype);
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
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_iqdevtype, &dst->icb_iqdevtype);
ISP_IOXPUT_8(isp, src->icb_logintime, &dst->icb_logintime);
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
}
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++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_nodename[i], &dst->icb_nodename[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_rqstout, &dst->icb_rqstout);
ISP_IOXPUT_16(isp, src->icb_rspnsin, &dst->icb_rspnsin);
ISP_IOXPUT_16(isp, src->icb_rqstqlen, &dst->icb_rqstqlen);
ISP_IOXPUT_16(isp, src->icb_rsltqlen, &dst->icb_rsltqlen);
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 < 4; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_rqstaddr[i], &dst->icb_rqstaddr[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
}
for (i = 0; i < 4; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_respaddr[i], &dst->icb_respaddr[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_lunenables, &dst->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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_ccnt, &dst->icb_icnt);
ISP_IOXPUT_8(isp, src->icb_icnt, &dst->icb_ccnt);
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
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_ccnt, &dst->icb_ccnt);
ISP_IOXPUT_8(isp, src->icb_icnt, &dst->icb_icnt);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->icb_lunetimeout, &dst->icb_lunetimeout);
ISP_IOXPUT_16(isp, src->icb_reserved1, &dst->icb_reserved1);
ISP_IOXPUT_16(isp, src->icb_xfwoptions, &dst->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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_racctimer, &dst->icb_idelaytimer);
ISP_IOXPUT_8(isp, src->icb_idelaytimer, &dst->icb_racctimer);
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
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->icb_racctimer, &dst->icb_racctimer);
ISP_IOXPUT_8(isp, src->icb_idelaytimer, &dst->icb_idelaytimer);
}
ISP_IOXPUT_16(isp, src->icb_zfwoptions, &dst->icb_zfwoptions);
}
void
isp_put_icb_2400(ispsoftc_t *isp, isp_icb_2400_t *src, isp_icb_2400_t *dst)
{
int i;
ISP_IOXPUT_16(isp, src->icb_version, &dst->icb_version);
ISP_IOXPUT_16(isp, src->icb_reserved0, &dst->icb_reserved0);
ISP_IOXPUT_16(isp, src->icb_maxfrmlen, &dst->icb_maxfrmlen);
ISP_IOXPUT_16(isp, src->icb_execthrottle, &dst->icb_execthrottle);
ISP_IOXPUT_16(isp, src->icb_xchgcnt, &dst->icb_xchgcnt);
ISP_IOXPUT_16(isp, src->icb_hardaddr, &dst->icb_hardaddr);
for (i = 0; i < 8; i++) {
ISP_IOXPUT_8(isp, src->icb_portname[i], &dst->icb_portname[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOXPUT_8(isp, src->icb_nodename[i], &dst->icb_nodename[i]);
}
ISP_IOXPUT_16(isp, src->icb_rspnsin, &dst->icb_rspnsin);
ISP_IOXPUT_16(isp, src->icb_rqstout, &dst->icb_rqstout);
ISP_IOXPUT_16(isp, src->icb_retry_count, &dst->icb_retry_count);
ISP_IOXPUT_16(isp, src->icb_priout, &dst->icb_priout);
ISP_IOXPUT_16(isp, src->icb_rsltqlen, &dst->icb_rsltqlen);
ISP_IOXPUT_16(isp, src->icb_rqstqlen, &dst->icb_rqstqlen);
ISP_IOXPUT_16(isp, src->icb_ldn_nols, &dst->icb_ldn_nols);
ISP_IOXPUT_16(isp, src->icb_prqstqlen, &dst->icb_prqstqlen);
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, src->icb_rqstaddr[i], &dst->icb_rqstaddr[i]);
}
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, src->icb_respaddr[i], &dst->icb_respaddr[i]);
}
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, src->icb_priaddr[i], &dst->icb_priaddr[i]);
}
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, src->icb_reserved1[i],
&dst->icb_reserved1[i]);
}
ISP_IOXPUT_16(isp, src->icb_atio_in, &dst->icb_atio_in);
ISP_IOXPUT_16(isp, src->icb_atioqlen, &dst->icb_atioqlen);
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp, src->icb_atioqaddr[i],
&dst->icb_atioqaddr[i]);
}
ISP_IOXPUT_16(isp, src->icb_idelaytimer, &dst->icb_idelaytimer);
ISP_IOXPUT_16(isp, src->icb_logintime, &dst->icb_logintime);
ISP_IOXPUT_32(isp, src->icb_fwoptions1, &dst->icb_fwoptions1);
ISP_IOXPUT_32(isp, src->icb_fwoptions2, &dst->icb_fwoptions2);
ISP_IOXPUT_32(isp, src->icb_fwoptions3, &dst->icb_fwoptions3);
for (i = 0; i < 12; i++) {
ISP_IOXPUT_16(isp, src->icb_reserved2[i],
&dst->icb_reserved2[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
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_pdb_21xx(ispsoftc_t *isp, isp_pdb_21xx_t *src, isp_pdb_21xx_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);
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
void
isp_get_pdb_24xx(ispsoftc_t *isp, isp_pdb_24xx_t *src, isp_pdb_24xx_t *dst)
{
int i;
ISP_IOXGET_16(isp, &src->pdb_flags, dst->pdb_flags);
ISP_IOXGET_8(isp, &src->pdb_curstate, dst->pdb_curstate);
ISP_IOXGET_8(isp, &src->pdb_laststate, dst->pdb_laststate);
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]);
}
ISP_IOXGET_16(isp, &src->pdb_retry_timer, dst->pdb_retry_timer);
ISP_IOXGET_16(isp, &src->pdb_handle, dst->pdb_handle);
ISP_IOXGET_16(isp, &src->pdb_rcv_dsize, dst->pdb_rcv_dsize);
ISP_IOXGET_16(isp, &src->pdb_reserved0, dst->pdb_reserved0);
ISP_IOXGET_16(isp, &src->pdb_prli_svc0, dst->pdb_prli_svc0);
ISP_IOXGET_16(isp, &src->pdb_prli_svc3, dst->pdb_prli_svc3);
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]);
}
for (i = 0; i < 24; i++) {
ISP_IOXGET_8(isp, &src->pdb_reserved1[i],
dst->pdb_reserved1[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
/*
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
* PLOGI/LOGO IOCB canonicalization
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
*/
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_plogx(ispsoftc_t *isp, isp_plogx_t *src, isp_plogx_t *dst)
{
int i;
isp_get_hdr(isp, &src->plogx_header, &dst->plogx_header);
ISP_IOXGET_32(isp, &src->plogx_handle, dst->plogx_handle);
ISP_IOXGET_16(isp, &src->plogx_status, dst->plogx_status);
ISP_IOXGET_16(isp, &src->plogx_nphdl, dst->plogx_nphdl);
ISP_IOXGET_16(isp, &src->plogx_flags, dst->plogx_flags);
ISP_IOXGET_16(isp, &src->plogx_vphdl, dst->plogx_vphdl);
ISP_IOXGET_16(isp, &src->plogx_portlo, dst->plogx_portlo);
ISP_IOXGET_16(isp, &src->plogx_rspsz_porthi, dst->plogx_rspsz_porthi);
for (i = 0; i < 11; i++) {
ISP_IOXGET_16(isp, &src->plogx_ioparm[i].lo16,
dst->plogx_ioparm[i].lo16);
ISP_IOXGET_16(isp, &src->plogx_ioparm[i].hi16,
dst->plogx_ioparm[i].hi16);
}
}
void
isp_put_plogx(ispsoftc_t *isp, isp_plogx_t *src, isp_plogx_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
int i;
isp_put_hdr(isp, &src->plogx_header, &dst->plogx_header);
ISP_IOXPUT_32(isp, src->plogx_handle, &dst->plogx_handle);
ISP_IOXPUT_16(isp, src->plogx_status, &dst->plogx_status);
ISP_IOXPUT_16(isp, src->plogx_nphdl, &dst->plogx_nphdl);
ISP_IOXPUT_16(isp, src->plogx_flags, &dst->plogx_flags);
ISP_IOXPUT_16(isp, src->plogx_vphdl, &dst->plogx_vphdl);
ISP_IOXPUT_16(isp, src->plogx_portlo, &dst->plogx_portlo);
ISP_IOXPUT_16(isp, src->plogx_rspsz_porthi, &dst->plogx_rspsz_porthi);
for (i = 0; i < 11; i++) {
ISP_IOXPUT_16(isp, src->plogx_ioparm[i].lo16,
&dst->plogx_ioparm[i].lo16);
ISP_IOXPUT_16(isp, src->plogx_ioparm[i].hi16,
&dst->plogx_ioparm[i].hi16);
}
}
/*
* CT Passthru canonicalization
*/
void
isp_get_ct_pt(ispsoftc_t *isp, isp_ct_pt_t *src, isp_ct_pt_t *dst)
{
int i;
isp_get_hdr(isp, &src->ctp_header, &dst->ctp_header);
ISP_IOXGET_32(isp, &src->ctp_handle, dst->ctp_handle);
ISP_IOXGET_16(isp, &src->ctp_status, dst->ctp_status);
ISP_IOXGET_16(isp, &src->ctp_nphdl, dst->ctp_nphdl);
ISP_IOXGET_16(isp, &src->ctp_cmd_cnt, dst->ctp_cmd_cnt);
ISP_IOXGET_16(isp, &src->ctp_vpidx, dst->ctp_vpidx);
ISP_IOXGET_16(isp, &src->ctp_time, dst->ctp_time);
ISP_IOXGET_16(isp, &src->ctp_reserved0, dst->ctp_reserved0);
ISP_IOXGET_16(isp, &src->ctp_rsp_cnt, dst->ctp_rsp_cnt);
for (i = 0; i < 5; i++) {
ISP_IOXGET_16(isp, &src->ctp_reserved1[i],
dst->ctp_reserved1[i]);
}
ISP_IOXGET_32(isp, &src->ctp_rsp_bcnt, dst->ctp_rsp_bcnt);
ISP_IOXGET_32(isp, &src->ctp_cmd_bcnt, dst->ctp_cmd_bcnt);
for (i = 0; i < 2; i++) {
ISP_IOXGET_32(isp, &src->ctp_dataseg[i].ds_base,
dst->ctp_dataseg[i].ds_base);
ISP_IOXGET_32(isp, &src->ctp_dataseg[i].ds_basehi,
dst->ctp_dataseg[i].ds_basehi);
ISP_IOXGET_32(isp, &src->ctp_dataseg[i].ds_count,
dst->ctp_dataseg[i].ds_count);
}
}
Make the SAN login/logout stuff more common between different chipsets and provied an isp_control entry point so that the outer layers can do PLOGI/LOGO explicitly. Add MS IOCB support. This completes the cycle for base support for SMI-S.
2006-11-18 03:53:16 +00:00
void
isp_get_ms(ispsoftc_t *isp, isp_ms_t *src, isp_ms_t *dst)
{
int i;
isp_get_hdr(isp, &src->ms_header, &dst->ms_header);
ISP_IOXGET_32(isp, &src->ms_handle, dst->ms_handle);
ISP_IOXGET_16(isp, &src->ms_nphdl, dst->ms_nphdl);
ISP_IOXGET_16(isp, &src->ms_status, dst->ms_status);
ISP_IOXGET_16(isp, &src->ms_flags, dst->ms_flags);
ISP_IOXGET_16(isp, &src->ms_reserved1, dst->ms_reserved1);
ISP_IOXGET_16(isp, &src->ms_time, dst->ms_time);
ISP_IOXGET_16(isp, &src->ms_cmd_cnt, dst->ms_cmd_cnt);
ISP_IOXGET_16(isp, &src->ms_tot_cnt, dst->ms_tot_cnt);
ISP_IOXGET_8(isp, &src->ms_type, dst->ms_type);
ISP_IOXGET_8(isp, &src->ms_r_ctl, dst->ms_r_ctl);
ISP_IOXGET_16(isp, &src->ms_rxid, dst->ms_rxid);
ISP_IOXGET_16(isp, &src->ms_reserved2, dst->ms_reserved2);
ISP_IOXGET_32(isp, &src->ms_rsp_bcnt, dst->ms_rsp_bcnt);
ISP_IOXGET_32(isp, &src->ms_cmd_bcnt, dst->ms_cmd_bcnt);
for (i = 0; i < 2; i++) {
ISP_IOXGET_32(isp, &src->ms_dataseg[i].ds_base,
dst->ms_dataseg[i].ds_base);
ISP_IOXGET_32(isp, &src->ms_dataseg[i].ds_basehi,
dst->ms_dataseg[i].ds_basehi);
ISP_IOXGET_32(isp, &src->ms_dataseg[i].ds_count,
dst->ms_dataseg[i].ds_count);
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
void
isp_put_ct_pt(ispsoftc_t *isp, isp_ct_pt_t *src, isp_ct_pt_t *dst)
{
int i;
isp_put_hdr(isp, &src->ctp_header, &dst->ctp_header);
ISP_IOXPUT_32(isp, src->ctp_handle, &dst->ctp_handle);
ISP_IOXPUT_16(isp, src->ctp_status, &dst->ctp_status);
ISP_IOXPUT_16(isp, src->ctp_nphdl, &dst->ctp_nphdl);
ISP_IOXPUT_16(isp, src->ctp_cmd_cnt, &dst->ctp_cmd_cnt);
ISP_IOXPUT_16(isp, src->ctp_vpidx, &dst->ctp_vpidx);
ISP_IOXPUT_16(isp, src->ctp_time, &dst->ctp_time);
ISP_IOXPUT_16(isp, src->ctp_reserved0, &dst->ctp_reserved0);
ISP_IOXPUT_16(isp, src->ctp_rsp_cnt, &dst->ctp_rsp_cnt);
for (i = 0; i < 5; i++) {
ISP_IOXPUT_16(isp, src->ctp_reserved1[i],
&dst->ctp_reserved1[i]);
}
ISP_IOXPUT_32(isp, src->ctp_rsp_bcnt, &dst->ctp_rsp_bcnt);
ISP_IOXPUT_32(isp, src->ctp_cmd_bcnt, &dst->ctp_cmd_bcnt);
for (i = 0; i < 2; i++) {
ISP_IOXPUT_32(isp, src->ctp_dataseg[i].ds_base,
&dst->ctp_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->ctp_dataseg[i].ds_basehi,
&dst->ctp_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, src->ctp_dataseg[i].ds_count,
&dst->ctp_dataseg[i].ds_count);
}
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
}
Make the SAN login/logout stuff more common between different chipsets and provied an isp_control entry point so that the outer layers can do PLOGI/LOGO explicitly. Add MS IOCB support. This completes the cycle for base support for SMI-S.
2006-11-18 03:53:16 +00:00
void
isp_put_ms(ispsoftc_t *isp, isp_ms_t *src, isp_ms_t *dst)
{
int i;
isp_put_hdr(isp, &src->ms_header, &dst->ms_header);
ISP_IOXPUT_32(isp, src->ms_handle, &dst->ms_handle);
ISP_IOXPUT_16(isp, src->ms_nphdl, &dst->ms_nphdl);
ISP_IOXPUT_16(isp, src->ms_status, &dst->ms_status);
ISP_IOXPUT_16(isp, src->ms_flags, &dst->ms_flags);
ISP_IOXPUT_16(isp, src->ms_reserved1, &dst->ms_reserved1);
ISP_IOXPUT_16(isp, src->ms_time, &dst->ms_time);
ISP_IOXPUT_16(isp, src->ms_cmd_cnt, &dst->ms_cmd_cnt);
ISP_IOXPUT_16(isp, src->ms_tot_cnt, &dst->ms_tot_cnt);
ISP_IOXPUT_8(isp, src->ms_type, &dst->ms_type);
ISP_IOXPUT_8(isp, src->ms_r_ctl, &dst->ms_r_ctl);
ISP_IOXPUT_16(isp, src->ms_rxid, &dst->ms_rxid);
ISP_IOXPUT_16(isp, src->ms_reserved2, &dst->ms_reserved2);
ISP_IOXPUT_32(isp, src->ms_rsp_bcnt, &dst->ms_rsp_bcnt);
ISP_IOXPUT_32(isp, src->ms_cmd_bcnt, &dst->ms_cmd_bcnt);
for (i = 0; i < 2; i++) {
ISP_IOXPUT_32(isp, src->ms_dataseg[i].ds_base,
&dst->ms_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->ms_dataseg[i].ds_basehi,
&dst->ms_dataseg[i].ds_basehi);
ISP_IOXPUT_32(isp, src->ms_dataseg[i].ds_count,
&dst->ms_dataseg[i].ds_count);
}
}
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
/*
* 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);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->snscb_reserved0, &dst->snscb_reserved0);
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_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);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->snscb_reserved1, &dst->snscb_reserved1);
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_IOXPUT_16(isp, src->snscb_cmd, &dst->snscb_cmd);
ISP_IOXPUT_16(isp, src->snscb_mword_div_2, &dst->snscb_mword_div_2);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->snscb_reserved3, &dst->snscb_reserved3);
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);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->snscb_reserved0, &dst->snscb_reserved0);
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_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);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->snscb_reserved1, &dst->snscb_reserved1);
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_IOXPUT_16(isp, src->snscb_cmd, &dst->snscb_cmd);
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->snscb_reserved2, &dst->snscb_reserved2);
ISP_IOXPUT_32(isp, src->snscb_reserved3, &dst->snscb_reserved3);
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_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]);
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
void
isp_get_els(ispsoftc_t *isp, els_t *src, els_t *dst)
{
int i;
isp_get_hdr(isp, &src->els_hdr, &dst->els_hdr);
ISP_IOXGET_32(isp, &src->els_handle, dst->els_handle);
ISP_IOXGET_16(isp, &src->els_status, dst->els_status);
ISP_IOXGET_16(isp, &src->els_nphdl, dst->els_nphdl);
ISP_IOXGET_16(isp, &src->els_xmit_dsd_count, dst->els_xmit_dsd_count);
ISP_IOXGET_8(isp, &src->els_vphdl, dst->els_vphdl);
ISP_IOXGET_8(isp, &src->els_sof, dst->els_sof);
ISP_IOXGET_32(isp, &src->els_rxid, dst->els_rxid);
ISP_IOXGET_16(isp, &src->els_recv_dsd_count, dst->els_recv_dsd_count);
ISP_IOXGET_8(isp, &src->els_opcode, dst->els_opcode);
ISP_IOXGET_8(isp, &src->els_reserved2, dst->els_reserved1);
ISP_IOXGET_8(isp, &src->els_did_lo, dst->els_did_lo);
ISP_IOXGET_8(isp, &src->els_did_mid, dst->els_did_mid);
ISP_IOXGET_8(isp, &src->els_did_hi, dst->els_did_hi);
ISP_IOXGET_8(isp, &src->els_reserved2, dst->els_reserved2);
ISP_IOXGET_16(isp, &src->els_reserved3, dst->els_reserved3);
ISP_IOXGET_16(isp, &src->els_ctl_flags, dst->els_ctl_flags);
ISP_IOXGET_32(isp, &src->els_bytecnt, dst->els_bytecnt);
ISP_IOXGET_32(isp, &src->els_subcode1, dst->els_subcode1);
ISP_IOXGET_32(isp, &src->els_subcode2, dst->els_subcode2);
for (i = 0; i < 20; i++) {
ISP_IOXGET_8(isp, &src->els_reserved4[i],
dst->els_reserved4[i]);
}
}
void
isp_put_els(ispsoftc_t *isp, els_t *src, els_t *dst)
{
isp_put_hdr(isp, &src->els_hdr, &dst->els_hdr);
ISP_IOXPUT_32(isp, src->els_handle, &dst->els_handle);
ISP_IOXPUT_16(isp, src->els_status, &dst->els_status);
ISP_IOXPUT_16(isp, src->els_nphdl, &dst->els_nphdl);
ISP_IOXPUT_16(isp, src->els_xmit_dsd_count, &dst->els_xmit_dsd_count);
ISP_IOXPUT_8(isp, src->els_vphdl, &dst->els_vphdl);
ISP_IOXPUT_8(isp, src->els_sof, &dst->els_sof);
ISP_IOXPUT_32(isp, src->els_rxid, &dst->els_rxid);
ISP_IOXPUT_16(isp, src->els_recv_dsd_count, &dst->els_recv_dsd_count);
ISP_IOXPUT_8(isp, src->els_opcode, &dst->els_opcode);
ISP_IOXPUT_8(isp, src->els_reserved2, &dst->els_reserved1);
ISP_IOXPUT_8(isp, src->els_did_lo, &dst->els_did_lo);
ISP_IOXPUT_8(isp, src->els_did_mid, &dst->els_did_mid);
ISP_IOXPUT_8(isp, src->els_did_hi, &dst->els_did_hi);
ISP_IOXPUT_8(isp, src->els_reserved2, &dst->els_reserved2);
ISP_IOXPUT_16(isp, src->els_reserved3, &dst->els_reserved3);
ISP_IOXPUT_16(isp, src->els_ctl_flags, &dst->els_ctl_flags);
ISP_IOXPUT_32(isp, src->els_recv_bytecnt, &dst->els_recv_bytecnt);
ISP_IOXPUT_32(isp, src->els_xmit_bytecnt, &dst->els_xmit_bytecnt);
ISP_IOXPUT_32(isp, src->els_xmit_dsd_length, &dst->els_xmit_dsd_length);
ISP_IOXPUT_16(isp, src->els_xmit_dsd_a1500, &dst->els_xmit_dsd_a1500);
ISP_IOXPUT_16(isp, src->els_xmit_dsd_a3116, &dst->els_xmit_dsd_a3116);
ISP_IOXPUT_16(isp, src->els_xmit_dsd_a4732, &dst->els_xmit_dsd_a4732);
ISP_IOXPUT_16(isp, src->els_xmit_dsd_a6348, &dst->els_xmit_dsd_a6348);
ISP_IOXPUT_32(isp, src->els_recv_dsd_length, &dst->els_recv_dsd_length);
ISP_IOXPUT_16(isp, src->els_recv_dsd_a1500, &dst->els_recv_dsd_a1500);
ISP_IOXPUT_16(isp, src->els_recv_dsd_a3116, &dst->els_recv_dsd_a3116);
ISP_IOXPUT_16(isp, src->els_recv_dsd_a4732, &dst->els_recv_dsd_a4732);
ISP_IOXPUT_16(isp, src->els_recv_dsd_a6348, &dst->els_recv_dsd_a6348);
}
/*
* FC Structure Canonicalization
*/
void
isp_get_fc_hdr(ispsoftc_t *isp, fc_hdr_t *src, fc_hdr_t *dst)
{
ISP_IOZGET_8(isp, &src->r_ctl, dst->r_ctl);
ISP_IOZGET_8(isp, &src->d_id[0], dst->d_id[0]);
ISP_IOZGET_8(isp, &src->d_id[1], dst->d_id[1]);
ISP_IOZGET_8(isp, &src->d_id[2], dst->d_id[2]);
ISP_IOZGET_8(isp, &src->cs_ctl, dst->cs_ctl);
ISP_IOZGET_8(isp, &src->s_id[0], dst->s_id[0]);
ISP_IOZGET_8(isp, &src->s_id[1], dst->s_id[1]);
ISP_IOZGET_8(isp, &src->s_id[2], dst->s_id[2]);
ISP_IOZGET_8(isp, &src->type, dst->type);
ISP_IOZGET_8(isp, &src->f_ctl, dst->f_ctl);
ISP_IOZGET_8(isp, &src->seq_id, dst->seq_id);
ISP_IOZGET_8(isp, &src->df_ctl, dst->df_ctl);
ISP_IOZGET_16(isp, &src->seq_cnt, dst->seq_cnt);
/* XXX SOMETHING WAS AND STILL CONTINUES WRONG HERE XXX */
#if 0
ISP_IOZGET_16(isp, &src->ox_id, dst->ox_id);
ISP_IOZGET_16(isp, &src->rx_id, dst->rx_id);
#else
ISP_IOZGET_32(isp, &src->ox_id, dst->parameter);
dst->ox_id = dst->parameter;
dst->rx_id = dst->parameter >> 16;
#endif
ISP_IOZGET_32(isp, &src->parameter, dst->parameter);
}
void
isp_get_fcp_cmnd_iu(ispsoftc_t *isp, fcp_cmnd_iu_t *src, fcp_cmnd_iu_t *dst)
{
int i;
for (i = 0; i < 8; i++) {
ISP_IOZGET_8(isp, &src->fcp_cmnd_lun[i], dst->fcp_cmnd_lun[i]);
}
ISP_IOZGET_8(isp, &src->fcp_cmnd_crn, dst->fcp_cmnd_crn);
ISP_IOZGET_8(isp, &src->fcp_cmnd_task_attribute,
dst->fcp_cmnd_task_attribute);
ISP_IOZGET_8(isp, &src->fcp_cmnd_task_management,
dst->fcp_cmnd_task_management);
ISP_IOZGET_8(isp, &src->fcp_cmnd_alen_datadir,
dst->fcp_cmnd_alen_datadir);
for (i = 0; i < 16; i++) {
ISP_IOZGET_8(isp, &src->cdb_dl.sf.fcp_cmnd_cdb[i],
dst->cdb_dl.sf.fcp_cmnd_cdb[i]);
}
ISP_IOZGET_32(isp, &src->cdb_dl.sf.fcp_cmnd_dl,
dst->cdb_dl.sf.fcp_cmnd_dl);
}
void
isp_put_rft_id(ispsoftc_t *isp, rft_id_t *src, rft_id_t *dst)
{
int i;
isp_put_ct_hdr(isp, &src->rftid_hdr, &dst->rftid_hdr);
ISP_IOZPUT_8(isp, src->rftid_reserved, &dst->rftid_reserved);
for (i = 0; i < 3; i++) {
ISP_IOZPUT_8(isp, src->rftid_portid[i], &dst->rftid_portid[i]);
}
for (i = 0; i < 8; i++) {
ISP_IOZPUT_32(isp, src->rftid_fc4types[i],
&dst->rftid_fc4types[i]);
}
}
void
isp_get_ct_hdr(ispsoftc_t *isp, ct_hdr_t *src, ct_hdr_t *dst)
{
ISP_IOZGET_8(isp, &src->ct_revision, dst->ct_revision);
ISP_IOZGET_8(isp, &src->ct_in_id[0], dst->ct_in_id[0]);
ISP_IOZGET_8(isp, &src->ct_in_id[1], dst->ct_in_id[1]);
ISP_IOZGET_8(isp, &src->ct_in_id[2], dst->ct_in_id[2]);
ISP_IOZGET_8(isp, &src->ct_fcs_type, dst->ct_fcs_type);
ISP_IOZGET_8(isp, &src->ct_fcs_subtype, dst->ct_fcs_subtype);
ISP_IOZGET_8(isp, &src->ct_options, dst->ct_options);
ISP_IOZGET_8(isp, &src->ct_reserved0, dst->ct_reserved0);
ISP_IOZGET_16(isp, &src->ct_cmd_resp, dst->ct_cmd_resp);
ISP_IOZGET_16(isp, &src->ct_bcnt_resid, dst->ct_bcnt_resid);
ISP_IOZGET_8(isp, &src->ct_reserved1, dst->ct_reserved1);
ISP_IOZGET_8(isp, &src->ct_reason, dst->ct_reason);
ISP_IOZGET_8(isp, &src->ct_explanation, dst->ct_explanation);
ISP_IOZGET_8(isp, &src->ct_vunique, dst->ct_vunique);
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
void
isp_put_ct_hdr(ispsoftc_t *isp, ct_hdr_t *src, ct_hdr_t *dst)
{
ISP_IOZPUT_8(isp, src->ct_revision, &dst->ct_revision);
ISP_IOZPUT_8(isp, src->ct_in_id[0], &dst->ct_in_id[0]);
ISP_IOZPUT_8(isp, src->ct_in_id[1], &dst->ct_in_id[1]);
ISP_IOZPUT_8(isp, src->ct_in_id[2], &dst->ct_in_id[2]);
ISP_IOZPUT_8(isp, src->ct_fcs_type, &dst->ct_fcs_type);
ISP_IOZPUT_8(isp, src->ct_fcs_subtype, &dst->ct_fcs_subtype);
ISP_IOZPUT_8(isp, src->ct_options, &dst->ct_options);
ISP_IOZPUT_8(isp, src->ct_reserved0, &dst->ct_reserved0);
ISP_IOZPUT_16(isp, src->ct_cmd_resp, &dst->ct_cmd_resp);
ISP_IOZPUT_16(isp, src->ct_bcnt_resid, &dst->ct_bcnt_resid);
ISP_IOZPUT_8(isp, src->ct_reserved1, &dst->ct_reserved1);
ISP_IOZPUT_8(isp, src->ct_reason, &dst->ct_reason);
ISP_IOZPUT_8(isp, src->ct_explanation, &dst->ct_explanation);
ISP_IOZPUT_8(isp, src->ct_vunique, &dst->ct_vunique);
}
#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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_save_xs_tgt(ispsoftc_t *isp, void *xs, uint32_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
*handlep = (i+1) | 0x8000;
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 *
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_find_xs_tgt(ispsoftc_t *isp, uint32_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
{
Temporarily desupport simultaneous target and initiator mode. When the linux port changes were imported which split the target command list to be separate from the initiator command list and the handle format changed to encode a type in the handle the implications to the function isp_handle_index (which only the NetBSD/OpenBSD/FreeBSD ports use) were overlooked. The fault is twofold: first, the index into the DMA maps in isp_pci is wrong because a target command handle with the type bit left in place caused a bad index (and panic) into dma map. Secondly, the assumption of the array of DMA maps in either PCS or SBUS attachment structures is that there is a linear mapping between handle index and DMA map index. This can no longer be true if there are overlapping index spaces for initiator mode and target mode commands. These changes bandaid around the problem by forcing us to not have simultaneous dual roles and doing the appropriate masking to make sure things are indexed correctly. A longer term fix is being devloped.
2007-04-02 01:04:20 +00:00
if (handle == 0 || IS_TARGET_HANDLE(handle) == 0 ||
(handle & ISP_HANDLE_MASK) > isp->isp_maxcmds) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_prt(isp, ISP_LOGERR, "bad handle in isp_find_xs_tgt");
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 {
Temporarily desupport simultaneous target and initiator mode. When the linux port changes were imported which split the target command list to be separate from the initiator command list and the handle format changed to encode a type in the handle the implications to the function isp_handle_index (which only the NetBSD/OpenBSD/FreeBSD ports use) were overlooked. The fault is twofold: first, the index into the DMA maps in isp_pci is wrong because a target command handle with the type bit left in place caused a bad index (and panic) into dma map. Secondly, the assumption of the array of DMA maps in either PCS or SBUS attachment structures is that there is a linear mapping between handle index and DMA map index. This can no longer be true if there are overlapping index spaces for initiator mode and target mode commands. These changes bandaid around the problem by forcing us to not have simultaneous dual roles and doing the appropriate masking to make sure things are indexed correctly. A longer term fix is being devloped.
2007-04-02 01:04:20 +00:00
return (isp->isp_tgtlist[(handle & ISP_HANDLE_MASK) - 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
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
uint32_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_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) {
Temporarily desupport simultaneous target and initiator mode. When the linux port changes were imported which split the target command list to be separate from the initiator command list and the handle format changed to encode a type in the handle the implications to the function isp_handle_index (which only the NetBSD/OpenBSD/FreeBSD ports use) were overlooked. The fault is twofold: first, the index into the DMA maps in isp_pci is wrong because a target command handle with the type bit left in place caused a bad index (and panic) into dma map. Secondly, the assumption of the array of DMA maps in either PCS or SBUS attachment structures is that there is a linear mapping between handle index and DMA map index. This can no longer be true if there are overlapping index spaces for initiator mode and target mode commands. These changes bandaid around the problem by forcing us to not have simultaneous dual roles and doing the appropriate masking to make sure things are indexed correctly. A longer term fix is being devloped.
2007-04-02 01:04:20 +00:00
return ((i+1) & ISP_HANDLE_MASK);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_destroy_tgt_handle(ispsoftc_t *isp, uint32_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
{
Temporarily desupport simultaneous target and initiator mode. When the linux port changes were imported which split the target command list to be separate from the initiator command list and the handle format changed to encode a type in the handle the implications to the function isp_handle_index (which only the NetBSD/OpenBSD/FreeBSD ports use) were overlooked. The fault is twofold: first, the index into the DMA maps in isp_pci is wrong because a target command handle with the type bit left in place caused a bad index (and panic) into dma map. Secondly, the assumption of the array of DMA maps in either PCS or SBUS attachment structures is that there is a linear mapping between handle index and DMA map index. This can no longer be true if there are overlapping index spaces for initiator mode and target mode commands. These changes bandaid around the problem by forcing us to not have simultaneous dual roles and doing the appropriate masking to make sure things are indexed correctly. A longer term fix is being devloped.
2007-04-02 01:04:20 +00:00
if (handle == 0 || IS_TARGET_HANDLE(handle) == 0 ||
(handle & ISP_HANDLE_MASK) > isp->isp_maxcmds) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_prt(isp, ISP_LOGERR,
"bad handle in isp_destroy_tgt_handle");
} else {
Temporarily desupport simultaneous target and initiator mode. When the linux port changes were imported which split the target command list to be separate from the initiator command list and the handle format changed to encode a type in the handle the implications to the function isp_handle_index (which only the NetBSD/OpenBSD/FreeBSD ports use) were overlooked. The fault is twofold: first, the index into the DMA maps in isp_pci is wrong because a target command handle with the type bit left in place caused a bad index (and panic) into dma map. Secondly, the assumption of the array of DMA maps in either PCS or SBUS attachment structures is that there is a linear mapping between handle index and DMA map index. This can no longer be true if there are overlapping index spaces for initiator mode and target mode commands. These changes bandaid around the problem by forcing us to not have simultaneous dual roles and doing the appropriate masking to make sure things are indexed correctly. A longer term fix is being devloped.
2007-04-02 01:04:20 +00:00
isp->isp_tgtlist[(handle & ISP_HANDLE_MASK) - 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
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_atio(ispsoftc_t *isp, at_entry_t *src, at_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->at_header, &dst->at_header);
ISP_IOXPUT_16(isp, src->at_reserved, &dst->at_reserved);
ISP_IOXPUT_16(isp, src->at_handle, &dst->at_handle);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (ISP_IS_SBUS(isp)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->at_lun, &dst->at_iid);
ISP_IOXPUT_8(isp, src->at_iid, &dst->at_lun);
ISP_IOXPUT_8(isp, src->at_cdblen, &dst->at_tgt);
ISP_IOXPUT_8(isp, src->at_tgt, &dst->at_cdblen);
ISP_IOXPUT_8(isp, src->at_status, &dst->at_scsi_status);
ISP_IOXPUT_8(isp, src->at_scsi_status, &dst->at_status);
ISP_IOXPUT_8(isp, src->at_tag_val, &dst->at_tag_type);
ISP_IOXPUT_8(isp, src->at_tag_type, &dst->at_tag_val);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->at_lun, &dst->at_lun);
ISP_IOXPUT_8(isp, src->at_iid, &dst->at_iid);
ISP_IOXPUT_8(isp, src->at_cdblen, &dst->at_cdblen);
ISP_IOXPUT_8(isp, src->at_tgt, &dst->at_tgt);
ISP_IOXPUT_8(isp, src->at_status, &dst->at_status);
ISP_IOXPUT_8(isp, src->at_scsi_status,
&dst->at_scsi_status);
ISP_IOXPUT_8(isp, src->at_tag_val, &dst->at_tag_val);
ISP_IOXPUT_8(isp, src->at_tag_type, &dst->at_tag_type);
}
ISP_IOXPUT_32(isp, src->at_flags, &dst->at_flags);
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 < ATIO_CDBLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->at_cdb[i], &dst->at_cdb[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
}
for (i = 0; i < QLTM_SENSELEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->at_sense[i], &dst->at_sense[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_atio(ispsoftc_t *isp, at_entry_t *src, at_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->at_header, &dst->at_header);
ISP_IOXGET_16(isp, &src->at_reserved, dst->at_reserved);
ISP_IOXGET_16(isp, &src->at_handle, dst->at_handle);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
if (ISP_IS_SBUS(isp)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->at_lun, dst->at_iid);
ISP_IOXGET_8(isp, &src->at_iid, dst->at_lun);
ISP_IOXGET_8(isp, &src->at_cdblen, dst->at_tgt);
ISP_IOXGET_8(isp, &src->at_tgt, dst->at_cdblen);
ISP_IOXGET_8(isp, &src->at_status, dst->at_scsi_status);
ISP_IOXGET_8(isp, &src->at_scsi_status, dst->at_status);
ISP_IOXGET_8(isp, &src->at_tag_val, dst->at_tag_type);
ISP_IOXGET_8(isp, &src->at_tag_type, dst->at_tag_val);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->at_lun, dst->at_lun);
ISP_IOXGET_8(isp, &src->at_iid, dst->at_iid);
ISP_IOXGET_8(isp, &src->at_cdblen, dst->at_cdblen);
ISP_IOXGET_8(isp, &src->at_tgt, dst->at_tgt);
ISP_IOXGET_8(isp, &src->at_status, dst->at_status);
ISP_IOXGET_8(isp, &src->at_scsi_status,
dst->at_scsi_status);
ISP_IOXGET_8(isp, &src->at_tag_val, dst->at_tag_val);
ISP_IOXGET_8(isp, &src->at_tag_type, dst->at_tag_type);
}
ISP_IOXGET_32(isp, &src->at_flags, dst->at_flags);
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 < ATIO_CDBLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->at_cdb[i], dst->at_cdb[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
}
for (i = 0; i < QLTM_SENSELEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->at_sense[i], dst->at_sense[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_atio2(ispsoftc_t *isp, at2_entry_t *src, at2_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->at_header, &dst->at_header);
ISP_IOXPUT_32(isp, src->at_reserved, &dst->at_reserved);
ISP_IOXPUT_8(isp, src->at_lun, &dst->at_lun);
ISP_IOXPUT_8(isp, src->at_iid, &dst->at_iid);
ISP_IOXPUT_16(isp, src->at_rxid, &dst->at_rxid);
ISP_IOXPUT_16(isp, src->at_flags, &dst->at_flags);
ISP_IOXPUT_16(isp, src->at_status, &dst->at_status);
ISP_IOXPUT_8(isp, src->at_crn, &dst->at_crn);
ISP_IOXPUT_8(isp, src->at_taskcodes, &dst->at_taskcodes);
ISP_IOXPUT_8(isp, src->at_taskflags, &dst->at_taskflags);
ISP_IOXPUT_8(isp, src->at_execodes, &dst->at_execodes);
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 < ATIO2_CDBLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->at_cdb[i], &dst->at_cdb[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->at_datalen, &dst->at_datalen);
ISP_IOXPUT_16(isp, src->at_scclun, &dst->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++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->at_wwpn[i], &dst->at_wwpn[i]);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
}
for (i = 0; i < 6; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->at_reserved2[i],
&dst->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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->at_oxid, &dst->at_oxid);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_atio2e(ispsoftc_t *isp, at2e_entry_t *src, at2e_entry_t *dst)
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->at_header, &dst->at_header);
ISP_IOXPUT_32(isp, src->at_reserved, &dst->at_reserved);
ISP_IOXPUT_16(isp, src->at_iid, &dst->at_iid);
ISP_IOXPUT_16(isp, src->at_rxid, &dst->at_rxid);
ISP_IOXPUT_16(isp, src->at_flags, &dst->at_flags);
ISP_IOXPUT_16(isp, src->at_status, &dst->at_status);
ISP_IOXPUT_8(isp, src->at_crn, &dst->at_crn);
ISP_IOXPUT_8(isp, src->at_taskcodes, &dst->at_taskcodes);
ISP_IOXPUT_8(isp, src->at_taskflags, &dst->at_taskflags);
ISP_IOXPUT_8(isp, src->at_execodes, &dst->at_execodes);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < ATIO2_CDBLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->at_cdb[i], &dst->at_cdb[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->at_datalen, &dst->at_datalen);
ISP_IOXPUT_16(isp, src->at_scclun, &dst->at_scclun);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < 4; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->at_wwpn[i], &dst->at_wwpn[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
}
for (i = 0; i < 6; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->at_reserved2[i],
&dst->at_reserved2[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->at_oxid, &dst->at_oxid);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_atio2(ispsoftc_t *isp, at2_entry_t *src, at2_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->at_header, &dst->at_header);
ISP_IOXGET_32(isp, &src->at_reserved, dst->at_reserved);
ISP_IOXGET_8(isp, &src->at_lun, dst->at_lun);
ISP_IOXGET_8(isp, &src->at_iid, dst->at_iid);
ISP_IOXGET_16(isp, &src->at_rxid, dst->at_rxid);
ISP_IOXGET_16(isp, &src->at_flags, dst->at_flags);
ISP_IOXGET_16(isp, &src->at_status, dst->at_status);
ISP_IOXGET_8(isp, &src->at_crn, dst->at_crn);
ISP_IOXGET_8(isp, &src->at_taskcodes, dst->at_taskcodes);
ISP_IOXGET_8(isp, &src->at_taskflags, dst->at_taskflags);
ISP_IOXGET_8(isp, &src->at_execodes, dst->at_execodes);
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 < ATIO2_CDBLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->at_cdb[i], dst->at_cdb[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_32(isp, &src->at_datalen, dst->at_datalen);
ISP_IOXGET_16(isp, &src->at_scclun, dst->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++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->at_wwpn[i], dst->at_wwpn[i]);
+ A variety of 23XX changes: disable MWI on 2300 based on function code, set an 'isp_port' for the 2312- it's a separate instance, but the NVRAM is shared, and the second port's NVRAM is at offset 256. + Enable RIO operation for LVD SCSI cards. This makes a *big* difference as even under reasonable load we get batched completions of about 30 commands at a time on, say, an ISP1080. + Do 'continuation' mailbox commands- this allows us to specify a work area within the softc and 'continue' repeated mailbox commands. This is more or less on an ad hoc basis and is currently only used for firmware loading (which f/w now loads substantially faster becuase the calling thread is only woken when all the f/w words are loaded- not for each one of the 40000 f/w words that gets loaded). + If we're about to return from isp_intr with a 'bogus interrupt' indication, and we're not a 23XX card, check to see whether the semaphore register is currently *2* (not *1* as it should be) and whether there's an async completion sitting in outgoing mailbox0. This seems to capture cases of lost fast posting and RIO interrupts that the 12160 && 1080 have been known to pump out under extreme load (extreme, as in > 250 active commands). + FC_SCRATCH_ACQUIRE/FC_SCRATCH_RELEASE macros. + Endian correct swizzle/unswizzle of an ATIO2 that has a WWPN in it. MFC after: 1 week
2002-02-04 21:04:25 +00:00
}
for (i = 0; i < 6; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->at_reserved2[i],
dst->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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->at_oxid, dst->at_oxid);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_atio2e(ispsoftc_t *isp, at2e_entry_t *src, at2e_entry_t *dst)
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->at_header, &dst->at_header);
ISP_IOXGET_32(isp, &src->at_reserved, dst->at_reserved);
ISP_IOXGET_16(isp, &src->at_iid, dst->at_iid);
ISP_IOXGET_16(isp, &src->at_rxid, dst->at_rxid);
ISP_IOXGET_16(isp, &src->at_flags, dst->at_flags);
ISP_IOXGET_16(isp, &src->at_status, dst->at_status);
ISP_IOXGET_8(isp, &src->at_crn, dst->at_crn);
ISP_IOXGET_8(isp, &src->at_taskcodes, dst->at_taskcodes);
ISP_IOXGET_8(isp, &src->at_taskflags, dst->at_taskflags);
ISP_IOXGET_8(isp, &src->at_execodes, dst->at_execodes);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < ATIO2_CDBLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->at_cdb[i], dst->at_cdb[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_32(isp, &src->at_datalen, dst->at_datalen);
ISP_IOXGET_16(isp, &src->at_scclun, dst->at_scclun);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < 4; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->at_wwpn[i], dst->at_wwpn[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
}
for (i = 0; i < 6; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->at_reserved2[i],
dst->at_reserved2[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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->at_oxid, dst->at_oxid);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_atio7(ispsoftc_t *isp, at7_entry_t *src, at7_entry_t *dst)
{
ISP_IOXGET_8(isp, &src->at_type, dst->at_type);
ISP_IOXGET_8(isp, &src->at_count, dst->at_count);
ISP_IOXGET_16(isp, &src->at_ta_len, dst->at_ta_len);
ISP_IOXGET_32(isp, &src->at_rxid, dst->at_rxid);
isp_get_fc_hdr(isp, &src->at_hdr, &dst->at_hdr);
isp_get_fcp_cmnd_iu(isp, &src->at_cmnd, &dst->at_cmnd);
}
void
isp_put_ctio(ispsoftc_t *isp, ct_entry_t *src, ct_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXPUT_16(isp, src->ct_syshandle, &dst->ct_syshandle);
ISP_IOXPUT_16(isp, src->ct_fwhandle, &dst->ct_fwhandle);
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->ct_iid, &dst->ct_lun);
ISP_IOXPUT_8(isp, src->ct_lun, &dst->ct_iid);
ISP_IOXPUT_8(isp, src->ct_tgt, &dst->ct_reserved2);
ISP_IOXPUT_8(isp, src->ct_reserved2, &dst->ct_tgt);
ISP_IOXPUT_8(isp, src->ct_status, &dst->ct_scsi_status);
ISP_IOXPUT_8(isp, src->ct_scsi_status, &dst->ct_status);
ISP_IOXPUT_8(isp, src->ct_tag_type, &dst->ct_tag_val);
ISP_IOXPUT_8(isp, src->ct_tag_val, &dst->ct_tag_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
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->ct_iid, &dst->ct_iid);
ISP_IOXPUT_8(isp, src->ct_lun, &dst->ct_lun);
ISP_IOXPUT_8(isp, src->ct_tgt, &dst->ct_tgt);
ISP_IOXPUT_8(isp, src->ct_reserved2, &dst->ct_reserved2);
ISP_IOXPUT_8(isp, src->ct_scsi_status,
&dst->ct_scsi_status);
ISP_IOXPUT_8(isp, src->ct_status, &dst->ct_status);
ISP_IOXPUT_8(isp, src->ct_tag_type, &dst->ct_tag_type);
ISP_IOXPUT_8(isp, src->ct_tag_val, &dst->ct_tag_val);
}
ISP_IOXPUT_32(isp, src->ct_flags, &dst->ct_flags);
ISP_IOXPUT_32(isp, src->ct_xfrlen, &dst->ct_xfrlen);
ISP_IOXPUT_32(isp, src->ct_resid, &dst->ct_resid);
ISP_IOXPUT_16(isp, src->ct_timeout, &dst->ct_timeout);
ISP_IOXPUT_16(isp, src->ct_seg_count, &dst->ct_seg_count);
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 < ISP_RQDSEG; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->ct_dataseg[i].ds_base,
&dst->ct_dataseg[i].ds_base);
ISP_IOXPUT_32(isp, src->ct_dataseg[i].ds_count,
&dst->ct_dataseg[i].ds_count);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_ctio(ispsoftc_t *isp, ct_entry_t *src, ct_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXGET_16(isp, &src->ct_syshandle, dst->ct_syshandle);
ISP_IOXGET_16(isp, &src->ct_fwhandle, dst->ct_fwhandle);
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->ct_lun, dst->ct_iid);
ISP_IOXGET_8(isp, &src->ct_iid, dst->ct_lun);
ISP_IOXGET_8(isp, &src->ct_reserved2, dst->ct_tgt);
ISP_IOXGET_8(isp, &src->ct_tgt, dst->ct_reserved2);
ISP_IOXGET_8(isp, &src->ct_status, dst->ct_scsi_status);
ISP_IOXGET_8(isp, &src->ct_scsi_status, dst->ct_status);
ISP_IOXGET_8(isp, &src->ct_tag_val, dst->ct_tag_type);
ISP_IOXGET_8(isp, &src->ct_tag_type, dst->ct_tag_val);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->ct_lun, dst->ct_lun);
ISP_IOXGET_8(isp, &src->ct_iid, dst->ct_iid);
ISP_IOXGET_8(isp, &src->ct_reserved2, dst->ct_reserved2);
ISP_IOXGET_8(isp, &src->ct_tgt, dst->ct_tgt);
ISP_IOXGET_8(isp, &src->ct_status, dst->ct_status);
ISP_IOXGET_8(isp, &src->ct_scsi_status,
dst->ct_scsi_status);
ISP_IOXGET_8(isp, &src->ct_tag_val, dst->ct_tag_val);
ISP_IOXGET_8(isp, &src->ct_tag_type, dst->ct_tag_type);
}
ISP_IOXGET_32(isp, &src->ct_flags, dst->ct_flags);
ISP_IOXGET_32(isp, &src->ct_xfrlen, dst->ct_xfrlen);
ISP_IOXGET_32(isp, &src->ct_resid, dst->ct_resid);
ISP_IOXGET_16(isp, &src->ct_timeout, dst->ct_timeout);
ISP_IOXGET_16(isp, &src->ct_seg_count, dst->ct_seg_count);
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 < ISP_RQDSEG; i++) {
ISP_IOXGET_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
&src->ct_dataseg[i].ds_base,
dst->ct_dataseg[i].ds_base);
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_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
&src->ct_dataseg[i].ds_count,
dst->ct_dataseg[i].ds_count);
}
}
void
isp_put_ctio2(ispsoftc_t *isp, ct2_entry_t *src, ct2_entry_t *dst)
{
int i;
isp_put_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXPUT_32(isp, src->ct_syshandle, &dst->ct_syshandle);
ISP_IOXPUT_8(isp, src->ct_lun, &dst->ct_lun);
ISP_IOXPUT_8(isp, src->ct_iid, &dst->ct_iid);
ISP_IOXPUT_16(isp, src->ct_rxid, &dst->ct_rxid);
ISP_IOXPUT_16(isp, src->ct_flags, &dst->ct_flags);
ISP_IOXPUT_16(isp, src->ct_timeout, &dst->ct_timeout);
ISP_IOXPUT_16(isp, src->ct_seg_count, &dst->ct_seg_count);
ISP_IOXPUT_32(isp, src->ct_resid, &dst->ct_resid);
ISP_IOXPUT_32(isp, src->ct_reloff, &dst->ct_reloff);
if ((src->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE0) {
ISP_IOXPUT_32(isp, src->rsp.m0._reserved,
&dst->rsp.m0._reserved);
ISP_IOXPUT_16(isp, src->rsp.m0._reserved2,
&dst->rsp.m0._reserved2);
ISP_IOXPUT_16(isp, src->rsp.m0.ct_scsi_status,
&dst->rsp.m0.ct_scsi_status);
ISP_IOXPUT_32(isp, src->rsp.m0.ct_xfrlen,
&dst->rsp.m0.ct_xfrlen);
if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO2) {
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 < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg[i].ds_base,
&dst->rsp.m0.u.ct_dataseg[i].ds_base);
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,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg[i].ds_count,
&dst->rsp.m0.u.ct_dataseg[i].ds_count);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
} else if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
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 < ISP_RQDSEG_T3; i++) {
ISP_IOXPUT_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg64[i].ds_base,
&dst->rsp.m0.u.ct_dataseg64[i].ds_base);
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,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg64[i].ds_basehi,
&dst->rsp.m0.u.ct_dataseg64[i].ds_basehi);
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,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg64[i].ds_count,
&dst->rsp.m0.u.ct_dataseg64[i].ds_count);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
} else if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO4) {
ISP_IOXPUT_16(isp, src->rsp.m0.u.ct_dslist.ds_type,
&dst->rsp.m0.u.ct_dslist.ds_type);
ISP_IOXPUT_32(isp, src->rsp.m0.u.ct_dslist.ds_segment,
&dst->rsp.m0.u.ct_dslist.ds_segment);
ISP_IOXPUT_32(isp, src->rsp.m0.u.ct_dslist.ds_base,
&dst->rsp.m0.u.ct_dslist.ds_base);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
} else if ((src->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE1) {
ISP_IOXPUT_16(isp, src->rsp.m1._reserved,
&dst->rsp.m1._reserved);
ISP_IOXPUT_16(isp, src->rsp.m1._reserved2,
&dst->rsp.m1._reserved2);
ISP_IOXPUT_16(isp, src->rsp.m1.ct_senselen,
&dst->rsp.m1.ct_senselen);
ISP_IOXPUT_16(isp, src->rsp.m1.ct_scsi_status,
&dst->rsp.m1.ct_scsi_status);
ISP_IOXPUT_16(isp, src->rsp.m1.ct_resplen,
&dst->rsp.m1.ct_resplen);
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 < MAXRESPLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->rsp.m1.ct_resp[i],
&dst->rsp.m1.ct_resp[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
}
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->rsp.m2._reserved,
&dst->rsp.m2._reserved);
ISP_IOXPUT_16(isp, src->rsp.m2._reserved2,
&dst->rsp.m2._reserved2);
ISP_IOXPUT_16(isp, src->rsp.m2._reserved3,
&dst->rsp.m2._reserved3);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_datalen,
&dst->rsp.m2.ct_datalen);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_base,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_count,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
}
void
isp_put_ctio2e(ispsoftc_t *isp, ct2e_entry_t *src, ct2e_entry_t *dst)
{
int i;
isp_put_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXPUT_32(isp, src->ct_syshandle, &dst->ct_syshandle);
ISP_IOXPUT_16(isp, src->ct_iid, &dst->ct_iid);
ISP_IOXPUT_16(isp, src->ct_rxid, &dst->ct_rxid);
ISP_IOXPUT_16(isp, src->ct_flags, &dst->ct_flags);
ISP_IOXPUT_16(isp, src->ct_timeout, &dst->ct_timeout);
ISP_IOXPUT_16(isp, src->ct_seg_count, &dst->ct_seg_count);
ISP_IOXPUT_32(isp, src->ct_resid, &dst->ct_resid);
ISP_IOXPUT_32(isp, src->ct_reloff, &dst->ct_reloff);
if ((src->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE0) {
ISP_IOXPUT_32(isp, src->rsp.m0._reserved,
&dst->rsp.m0._reserved);
ISP_IOXPUT_16(isp, src->rsp.m0._reserved2,
&dst->rsp.m0._reserved2);
ISP_IOXPUT_16(isp, src->rsp.m0.ct_scsi_status,
&dst->rsp.m0.ct_scsi_status);
ISP_IOXPUT_32(isp, src->rsp.m0.ct_xfrlen,
&dst->rsp.m0.ct_xfrlen);
if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO2) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXPUT_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg[i].ds_base,
&dst->rsp.m0.u.ct_dataseg[i].ds_base);
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_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg[i].ds_count,
&dst->rsp.m0.u.ct_dataseg[i].ds_count);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
} else if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < ISP_RQDSEG_T3; i++) {
ISP_IOXPUT_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg64[i].ds_base,
&dst->rsp.m0.u.ct_dataseg64[i].ds_base);
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_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg64[i].ds_basehi,
&dst->rsp.m0.u.ct_dataseg64[i].ds_basehi);
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_32(isp,
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
src->rsp.m0.u.ct_dataseg64[i].ds_count,
&dst->rsp.m0.u.ct_dataseg64[i].ds_count);
}
} else if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO4) {
ISP_IOXPUT_16(isp, src->rsp.m0.u.ct_dslist.ds_type,
&dst->rsp.m0.u.ct_dslist.ds_type);
ISP_IOXPUT_32(isp, src->rsp.m0.u.ct_dslist.ds_segment,
&dst->rsp.m0.u.ct_dslist.ds_segment);
ISP_IOXPUT_32(isp, src->rsp.m0.u.ct_dslist.ds_base,
&dst->rsp.m0.u.ct_dslist.ds_base);
}
} else if ((src->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE1) {
ISP_IOXPUT_16(isp, src->rsp.m1._reserved,
&dst->rsp.m1._reserved);
ISP_IOXPUT_16(isp, src->rsp.m1._reserved2,
&dst->rsp.m1._reserved2);
ISP_IOXPUT_16(isp, src->rsp.m1.ct_senselen,
&dst->rsp.m1.ct_senselen);
ISP_IOXPUT_16(isp, src->rsp.m1.ct_scsi_status,
&dst->rsp.m1.ct_scsi_status);
ISP_IOXPUT_16(isp, src->rsp.m1.ct_resplen,
&dst->rsp.m1.ct_resplen);
for (i = 0; i < MAXRESPLEN; i++) {
ISP_IOXPUT_8(isp, src->rsp.m1.ct_resp[i],
&dst->rsp.m1.ct_resp[i]);
}
} else {
ISP_IOXPUT_32(isp, src->rsp.m2._reserved,
&dst->rsp.m2._reserved);
ISP_IOXPUT_16(isp, src->rsp.m2._reserved2,
&dst->rsp.m2._reserved2);
ISP_IOXPUT_16(isp, src->rsp.m2._reserved3,
&dst->rsp.m2._reserved3);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_datalen,
&dst->rsp.m2.ct_datalen);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_base,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_count,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
}
void
isp_put_ctio7(ispsoftc_t *isp, ct7_entry_t *src, ct7_entry_t *dst)
{
int i;
isp_put_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXPUT_32(isp, src->ct_syshandle, &dst->ct_syshandle);
ISP_IOXPUT_16(isp, src->ct_nphdl, &dst->ct_nphdl);
ISP_IOXPUT_16(isp, src->ct_timeout, &dst->ct_timeout);
ISP_IOXPUT_16(isp, src->ct_seg_count, &dst->ct_seg_count);
ISP_IOXPUT_8(isp, src->ct_vpindex, &dst->ct_vpindex);
ISP_IOXPUT_8(isp, src->ct_xflags, &dst->ct_xflags);
ISP_IOXPUT_16(isp, src->ct_iid_lo, &dst->ct_iid_lo);
ISP_IOXPUT_8(isp, src->ct_iid_hi, &dst->ct_iid_hi);
ISP_IOXPUT_8(isp, src->ct_reserved, &dst->ct_reserved);
ISP_IOXPUT_32(isp, src->ct_rxid, &dst->ct_rxid);
ISP_IOXPUT_16(isp, src->ct_senselen, &dst->ct_senselen);
ISP_IOXPUT_16(isp, src->ct_flags, &dst->ct_flags);
ISP_IOXPUT_32(isp, src->ct_resid, &dst->ct_resid);
ISP_IOXPUT_16(isp, src->ct_oxid, &dst->ct_oxid);
ISP_IOXPUT_16(isp, src->ct_scsi_status, &dst->ct_scsi_status);
if ((dst->ct_flags & CT7_FLAG_MMASK) == CT7_FLAG_MODE0) {
ISP_IOXPUT_32(isp, src->rsp.m0.reloff, &dst->rsp.m0.reloff);
ISP_IOXPUT_32(isp, src->rsp.m0.reserved0,
&dst->rsp.m0.reserved0);
ISP_IOXPUT_32(isp, src->rsp.m0.ct_xfrlen,
&dst->rsp.m0.ct_xfrlen);
ISP_IOXPUT_32(isp, src->rsp.m0.reserved1,
&dst->rsp.m0.reserved1);
ISP_IOXPUT_32(isp, src->rsp.m0.ds.ds_base,
&dst->rsp.m0.ds.ds_base);
ISP_IOXPUT_32(isp, src->rsp.m0.ds.ds_basehi,
&dst->rsp.m0.ds.ds_basehi);
ISP_IOXPUT_32(isp, src->rsp.m0.ds.ds_count,
&dst->rsp.m0.ds.ds_count);
} else if ((dst->ct_flags & CT7_FLAG_MMASK) == CT7_FLAG_MODE1) {
ISP_IOXPUT_16(isp, src->rsp.m1.ct_resplen,
&dst->rsp.m1.ct_resplen);
ISP_IOXPUT_16(isp, src->rsp.m1.reserved, &dst->rsp.m1.reserved);
for (i = 0; i < MAXRESPLEN_24XX; i++) {
ISP_IOXPUT_8(isp, src->rsp.m1.ct_resp[i],
&dst->rsp.m1.ct_resp[i]);
}
} else {
ISP_IOXPUT_32(isp, src->rsp.m2.reserved0,
&dst->rsp.m2.reserved0);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_datalen,
&dst->rsp.m2.ct_datalen);
ISP_IOXPUT_32(isp, src->rsp.m2.reserved1,
&dst->rsp.m2.reserved1);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_base,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_basehi,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_basehi);
ISP_IOXPUT_32(isp, src->rsp.m2.ct_fcp_rsp_iudata.ds_count,
&dst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
}
void
isp_get_ctio2(ispsoftc_t *isp, ct2_entry_t *src, ct2_entry_t *dst)
{
int i;
isp_get_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXGET_32(isp, &src->ct_syshandle, dst->ct_syshandle);
ISP_IOXGET_8(isp, &src->ct_lun, dst->ct_lun);
ISP_IOXGET_8(isp, &src->ct_iid, dst->ct_iid);
ISP_IOXGET_16(isp, &src->ct_rxid, dst->ct_rxid);
ISP_IOXGET_16(isp, &src->ct_flags, dst->ct_flags);
ISP_IOXGET_16(isp, &src->ct_status, dst->ct_status);
ISP_IOXGET_16(isp, &src->ct_timeout, dst->ct_timeout);
ISP_IOXGET_16(isp, &src->ct_seg_count, dst->ct_seg_count);
ISP_IOXGET_32(isp, &src->ct_reloff, dst->ct_reloff);
ISP_IOXGET_32(isp, &src->ct_resid, dst->ct_resid);
if ((dst->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE0) {
ISP_IOXGET_32(isp, &src->rsp.m0._reserved,
dst->rsp.m0._reserved);
ISP_IOXGET_16(isp, &src->rsp.m0._reserved2,
dst->rsp.m0._reserved2);
ISP_IOXGET_16(isp, &src->rsp.m0.ct_scsi_status,
dst->rsp.m0.ct_scsi_status);
ISP_IOXGET_32(isp, &src->rsp.m0.ct_xfrlen,
dst->rsp.m0.ct_xfrlen);
if (dst->ct_header.rqs_entry_type == RQSTYPE_CTIO2) {
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg[i].ds_base,
dst->rsp.m0.u.ct_dataseg[i].ds_base);
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg[i].ds_count,
dst->rsp.m0.u.ct_dataseg[i].ds_count);
}
} else if (dst->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
for (i = 0; i < ISP_RQDSEG_T3; i++) {
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg64[i].ds_base,
dst->rsp.m0.u.ct_dataseg64[i].ds_base);
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg64[i].ds_basehi,
dst->rsp.m0.u.ct_dataseg64[i].ds_basehi);
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg64[i].ds_count,
dst->rsp.m0.u.ct_dataseg64[i].ds_count);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
} else if (dst->ct_header.rqs_entry_type == RQSTYPE_CTIO4) {
ISP_IOXGET_16(isp, &src->rsp.m0.u.ct_dslist.ds_type,
dst->rsp.m0.u.ct_dslist.ds_type);
ISP_IOXGET_32(isp, &src->rsp.m0.u.ct_dslist.ds_segment,
dst->rsp.m0.u.ct_dslist.ds_segment);
ISP_IOXGET_32(isp, &src->rsp.m0.u.ct_dslist.ds_base,
dst->rsp.m0.u.ct_dslist.ds_base);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
} else if ((dst->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE1) {
ISP_IOXGET_16(isp, &src->rsp.m1._reserved,
dst->rsp.m1._reserved);
ISP_IOXGET_16(isp, &src->rsp.m1._reserved2,
dst->rsp.m1._reserved2);
ISP_IOXGET_16(isp, &src->rsp.m1.ct_senselen,
dst->rsp.m1.ct_senselen);
ISP_IOXGET_16(isp, &src->rsp.m1.ct_scsi_status,
dst->rsp.m1.ct_scsi_status);
ISP_IOXGET_16(isp, &src->rsp.m1.ct_resplen,
dst->rsp.m1.ct_resplen);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < MAXRESPLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->rsp.m1.ct_resp[i],
dst->rsp.m1.ct_resp[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
}
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_32(isp, &src->rsp.m2._reserved,
dst->rsp.m2._reserved);
ISP_IOXGET_16(isp, &src->rsp.m2._reserved2,
dst->rsp.m2._reserved2);
ISP_IOXGET_16(isp, &src->rsp.m2._reserved3,
dst->rsp.m2._reserved3);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_datalen,
dst->rsp.m2.ct_datalen);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_base,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_count,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_ctio2e(ispsoftc_t *isp, ct2e_entry_t *src, ct2e_entry_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
int i;
isp_get_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXGET_32(isp, &src->ct_syshandle, dst->ct_syshandle);
ISP_IOXGET_16(isp, &src->ct_iid, dst->ct_iid);
ISP_IOXGET_16(isp, &src->ct_rxid, dst->ct_rxid);
ISP_IOXGET_16(isp, &src->ct_flags, dst->ct_flags);
ISP_IOXGET_16(isp, &src->ct_status, dst->ct_status);
ISP_IOXGET_16(isp, &src->ct_timeout, dst->ct_timeout);
ISP_IOXGET_16(isp, &src->ct_seg_count, dst->ct_seg_count);
ISP_IOXGET_32(isp, &src->ct_reloff, dst->ct_reloff);
ISP_IOXGET_32(isp, &src->ct_resid, dst->ct_resid);
if ((dst->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE0) {
ISP_IOXGET_32(isp, &src->rsp.m0._reserved,
dst->rsp.m0._reserved);
ISP_IOXGET_16(isp, &src->rsp.m0._reserved2,
dst->rsp.m0._reserved2);
ISP_IOXGET_16(isp, &src->rsp.m0.ct_scsi_status,
dst->rsp.m0.ct_scsi_status);
ISP_IOXGET_32(isp, &src->rsp.m0.ct_xfrlen,
dst->rsp.m0.ct_xfrlen);
if (src->ct_header.rqs_entry_type == RQSTYPE_CTIO2) {
for (i = 0; i < ISP_RQDSEG_T2; i++) {
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg[i].ds_base,
dst->rsp.m0.u.ct_dataseg[i].ds_base);
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg[i].ds_count,
dst->rsp.m0.u.ct_dataseg[i].ds_count);
}
} else if (dst->ct_header.rqs_entry_type == RQSTYPE_CTIO3) {
for (i = 0; i < ISP_RQDSEG_T3; i++) {
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg64[i].ds_base,
dst->rsp.m0.u.ct_dataseg64[i].ds_base);
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg64[i].ds_basehi,
dst->rsp.m0.u.ct_dataseg64[i].ds_basehi);
ISP_IOXGET_32(isp,
&src->rsp.m0.u.ct_dataseg64[i].ds_count,
dst->rsp.m0.u.ct_dataseg64[i].ds_count);
}
} else if (dst->ct_header.rqs_entry_type == RQSTYPE_CTIO4) {
ISP_IOXGET_16(isp, &src->rsp.m0.u.ct_dslist.ds_type,
dst->rsp.m0.u.ct_dslist.ds_type);
ISP_IOXGET_32(isp, &src->rsp.m0.u.ct_dslist.ds_segment,
dst->rsp.m0.u.ct_dslist.ds_segment);
ISP_IOXGET_32(isp, &src->rsp.m0.u.ct_dslist.ds_base,
dst->rsp.m0.u.ct_dslist.ds_base);
}
} else if ((dst->ct_flags & CT2_FLAG_MMASK) == CT2_FLAG_MODE1) {
ISP_IOXGET_16(isp, &src->rsp.m1._reserved,
dst->rsp.m1._reserved);
ISP_IOXGET_16(isp, &src->rsp.m1._reserved2,
dst->rsp.m1._reserved2);
ISP_IOXGET_16(isp, &src->rsp.m1.ct_senselen,
dst->rsp.m1.ct_senselen);
ISP_IOXGET_16(isp, &src->rsp.m1.ct_scsi_status,
dst->rsp.m1.ct_scsi_status);
ISP_IOXGET_16(isp, &src->rsp.m1.ct_resplen,
dst->rsp.m1.ct_resplen);
for (i = 0; i < MAXRESPLEN; i++) {
ISP_IOXGET_8(isp, &src->rsp.m1.ct_resp[i],
dst->rsp.m1.ct_resp[i]);
}
} else {
ISP_IOXGET_32(isp, &src->rsp.m2._reserved,
dst->rsp.m2._reserved);
ISP_IOXGET_16(isp, &src->rsp.m2._reserved2,
dst->rsp.m2._reserved2);
ISP_IOXGET_16(isp, &src->rsp.m2._reserved3,
dst->rsp.m2._reserved3);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_datalen,
dst->rsp.m2.ct_datalen);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_base,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_count,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_ctio7(ispsoftc_t *isp, ct7_entry_t *src, ct7_entry_t *dst)
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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
int i;
isp_get_hdr(isp, &src->ct_header, &dst->ct_header);
ISP_IOXGET_32(isp, &src->ct_syshandle, dst->ct_syshandle);
ISP_IOXGET_16(isp, &src->ct_nphdl, dst->ct_nphdl);
ISP_IOXGET_16(isp, &src->ct_timeout, dst->ct_timeout);
ISP_IOXGET_16(isp, &src->ct_seg_count, dst->ct_seg_count);
ISP_IOXGET_8(isp, &src->ct_vpindex, dst->ct_vpindex);
ISP_IOXGET_8(isp, &src->ct_xflags, dst->ct_xflags);
ISP_IOXGET_16(isp, &src->ct_iid_lo, dst->ct_iid_lo);
ISP_IOXGET_8(isp, &src->ct_iid_hi, dst->ct_iid_hi);
ISP_IOXGET_8(isp, &src->ct_reserved, dst->ct_reserved);
ISP_IOXGET_32(isp, &src->ct_rxid, dst->ct_rxid);
ISP_IOXGET_16(isp, &src->ct_senselen, dst->ct_senselen);
ISP_IOXGET_16(isp, &src->ct_flags, dst->ct_flags);
ISP_IOXGET_32(isp, &src->ct_resid, dst->ct_resid);
ISP_IOXGET_16(isp, &src->ct_oxid, dst->ct_oxid);
ISP_IOXGET_16(isp, &src->ct_scsi_status, dst->ct_scsi_status);
if ((dst->ct_flags & CT7_FLAG_MMASK) == CT7_FLAG_MODE0) {
ISP_IOXGET_32(isp, &src->rsp.m0.reloff, dst->rsp.m0.reloff);
ISP_IOXGET_32(isp, &src->rsp.m0.reserved0,
dst->rsp.m0.reserved0);
ISP_IOXGET_32(isp, &src->rsp.m0.ct_xfrlen,
dst->rsp.m0.ct_xfrlen);
ISP_IOXGET_32(isp, &src->rsp.m0.reserved1,
dst->rsp.m0.reserved1);
ISP_IOXGET_32(isp, &src->rsp.m0.ds.ds_base,
dst->rsp.m0.ds.ds_base);
ISP_IOXGET_32(isp, &src->rsp.m0.ds.ds_basehi,
dst->rsp.m0.ds.ds_basehi);
ISP_IOXGET_32(isp, &src->rsp.m0.ds.ds_count,
dst->rsp.m0.ds.ds_count);
} else if ((dst->ct_flags & CT7_FLAG_MMASK) == CT7_FLAG_MODE1) {
ISP_IOXGET_16(isp, &src->rsp.m1.ct_resplen,
dst->rsp.m1.ct_resplen);
ISP_IOXGET_16(isp, &src->rsp.m1.reserved, dst->rsp.m1.reserved);
for (i = 0; i < MAXRESPLEN_24XX; i++) {
ISP_IOXGET_8(isp, &src->rsp.m1.ct_resp[i],
dst->rsp.m1.ct_resp[i]);
}
} else {
ISP_IOXGET_32(isp, &src->rsp.m2.reserved0,
dst->rsp.m2.reserved0);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_datalen,
dst->rsp.m2.ct_datalen);
ISP_IOXGET_32(isp, &src->rsp.m2.reserved1,
dst->rsp.m2.reserved1);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_base,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_base);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_basehi,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_basehi);
ISP_IOXGET_32(isp, &src->rsp.m2.ct_fcp_rsp_iudata.ds_count,
dst->rsp.m2.ct_fcp_rsp_iudata.ds_count);
}
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_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &lesrc->le_header, &ledst->le_header);
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, 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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &lesrc->le_header, &ledst->le_header);
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_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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_notify(ispsoftc_t *isp, in_entry_t *src, in_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXPUT_32(isp, src->in_reserved, &dst->in_reserved);
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->in_lun, &dst->in_iid);
ISP_IOXPUT_8(isp, src->in_iid, &dst->in_lun);
ISP_IOXPUT_8(isp, src->in_reserved2, &dst->in_tgt);
ISP_IOXPUT_8(isp, src->in_tgt, &dst->in_reserved2);
ISP_IOXPUT_8(isp, src->in_status, &dst->in_rsvd2);
ISP_IOXPUT_8(isp, src->in_rsvd2, &dst->in_status);
ISP_IOXPUT_8(isp, src->in_tag_val, &dst->in_tag_type);
ISP_IOXPUT_8(isp, src->in_tag_type, &dst->in_tag_val);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->in_lun, &dst->in_lun);
ISP_IOXPUT_8(isp, src->in_iid, &dst->in_iid);
ISP_IOXPUT_8(isp, src->in_reserved2, &dst->in_reserved2);
ISP_IOXPUT_8(isp, src->in_tgt, &dst->in_tgt);
ISP_IOXPUT_8(isp, src->in_status, &dst->in_status);
ISP_IOXPUT_8(isp, src->in_rsvd2, &dst->in_rsvd2);
ISP_IOXPUT_8(isp, src->in_tag_val, &dst->in_tag_val);
ISP_IOXPUT_8(isp, src->in_tag_type, &dst->in_tag_type);
}
ISP_IOXPUT_32(isp, src->in_flags, &dst->in_flags);
ISP_IOXPUT_16(isp, src->in_seqid, &dst->in_seqid);
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 < IN_MSGLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->in_msg[i], &dst->in_msg[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
}
for (i = 0; i < IN_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->in_reserved3[i],
&dst->in_reserved3[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
}
for (i = 0; i < QLTM_SENSELEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->in_sense[i],
&dst->in_sense[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_notify(ispsoftc_t *isp, in_entry_t *src, in_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXGET_32(isp, &src->in_reserved, dst->in_reserved);
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->in_lun, dst->in_iid);
ISP_IOXGET_8(isp, &src->in_iid, dst->in_lun);
ISP_IOXGET_8(isp, &src->in_reserved2, dst->in_tgt);
ISP_IOXGET_8(isp, &src->in_tgt, dst->in_reserved2);
ISP_IOXGET_8(isp, &src->in_status, dst->in_rsvd2);
ISP_IOXGET_8(isp, &src->in_rsvd2, dst->in_status);
ISP_IOXGET_8(isp, &src->in_tag_val, dst->in_tag_type);
ISP_IOXGET_8(isp, &src->in_tag_type, dst->in_tag_val);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->in_lun, dst->in_lun);
ISP_IOXGET_8(isp, &src->in_iid, dst->in_iid);
ISP_IOXGET_8(isp, &src->in_reserved2, dst->in_reserved2);
ISP_IOXGET_8(isp, &src->in_tgt, dst->in_tgt);
ISP_IOXGET_8(isp, &src->in_status, dst->in_status);
ISP_IOXGET_8(isp, &src->in_rsvd2, dst->in_rsvd2);
ISP_IOXGET_8(isp, &src->in_tag_val, dst->in_tag_val);
ISP_IOXGET_8(isp, &src->in_tag_type, dst->in_tag_type);
}
ISP_IOXGET_32(isp, &src->in_flags, dst->in_flags);
ISP_IOXGET_16(isp, &src->in_seqid, dst->in_seqid);
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 < IN_MSGLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->in_msg[i], dst->in_msg[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
}
for (i = 0; i < IN_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->in_reserved3[i],
dst->in_reserved3[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
}
for (i = 0; i < QLTM_SENSELEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->in_sense[i],
dst->in_sense[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_notify_fc(ispsoftc_t *isp, in_fcentry_t *src,
in_fcentry_t *dst)
{
isp_put_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXPUT_32(isp, src->in_reserved, &dst->in_reserved);
ISP_IOXPUT_8(isp, src->in_lun, &dst->in_lun);
ISP_IOXPUT_8(isp, src->in_iid, &dst->in_iid);
ISP_IOXPUT_16(isp, src->in_scclun, &dst->in_scclun);
ISP_IOXPUT_32(isp, src->in_reserved2, &dst->in_reserved2);
ISP_IOXPUT_16(isp, src->in_status, &dst->in_status);
ISP_IOXPUT_16(isp, src->in_task_flags, &dst->in_task_flags);
ISP_IOXPUT_16(isp, src->in_seqid, &dst->in_seqid);
}
void
isp_put_notify_fc_e(ispsoftc_t *isp, in_fcentry_e_t *src,
in_fcentry_e_t *dst)
{
isp_put_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXPUT_32(isp, src->in_reserved, &dst->in_reserved);
ISP_IOXPUT_16(isp, src->in_iid, &dst->in_iid);
ISP_IOXPUT_16(isp, src->in_scclun, &dst->in_scclun);
ISP_IOXPUT_32(isp, src->in_reserved2, &dst->in_reserved2);
ISP_IOXPUT_16(isp, src->in_status, &dst->in_status);
ISP_IOXPUT_16(isp, src->in_task_flags, &dst->in_task_flags);
ISP_IOXPUT_16(isp, src->in_seqid, &dst->in_seqid);
}
void
isp_put_notify_24xx(ispsoftc_t *isp, in_fcentry_24xx_t *src,
in_fcentry_24xx_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
int i;
isp_put_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXPUT_32(isp, src->in_reserved, &dst->in_reserved);
ISP_IOXPUT_16(isp, src->in_nphdl, &dst->in_nphdl);
ISP_IOXPUT_16(isp, src->in_reserved1, &dst->in_reserved1);
ISP_IOXPUT_16(isp, src->in_flags, &dst->in_flags);
ISP_IOXPUT_16(isp, src->in_srr_rxid, &dst->in_srr_rxid);
ISP_IOXPUT_16(isp, src->in_status, &dst->in_status);
ISP_IOXPUT_8(isp, src->in_status_subcode, &dst->in_status_subcode);
ISP_IOXPUT_16(isp, src->in_reserved2, &dst->in_reserved2);
ISP_IOXPUT_32(isp, src->in_rxid, &dst->in_rxid);
ISP_IOXPUT_16(isp, src->in_srr_reloff_hi, &dst->in_srr_reloff_hi);
ISP_IOXPUT_16(isp, src->in_srr_reloff_lo, &dst->in_srr_reloff_lo);
ISP_IOXPUT_16(isp, src->in_srr_iu, &dst->in_srr_iu);
ISP_IOXPUT_16(isp, src->in_srr_oxid, &dst->in_srr_oxid);
for (i = 0; i < 18; i++) {
ISP_IOXPUT_8(isp, src->in_reserved3[i], &dst->in_reserved3[i]);
}
ISP_IOXPUT_8(isp, src->in_reserved4, &dst->in_reserved4);
ISP_IOXPUT_8(isp, src->in_vpindex, &dst->in_vpindex);
ISP_IOXPUT_32(isp, src->in_reserved5, &dst->in_reserved5);
ISP_IOXPUT_16(isp, src->in_portid_lo, &dst->in_portid_lo);
ISP_IOXPUT_8(isp, src->in_portid_hi, &dst->in_portid_hi);
ISP_IOXPUT_8(isp, src->in_reserved6, &dst->in_reserved6);
ISP_IOXPUT_16(isp, src->in_reserved7, &dst->in_reserved7);
ISP_IOXPUT_16(isp, src->in_oxid, &dst->in_oxid);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_notify_fc(ispsoftc_t *isp, in_fcentry_t *src,
in_fcentry_t *dst)
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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXGET_32(isp, &src->in_reserved, dst->in_reserved);
ISP_IOXGET_8(isp, &src->in_lun, dst->in_lun);
ISP_IOXGET_8(isp, &src->in_iid, dst->in_iid);
ISP_IOXGET_16(isp, &src->in_scclun, dst->in_scclun);
ISP_IOXGET_32(isp, &src->in_reserved2, dst->in_reserved2);
ISP_IOXGET_16(isp, &src->in_status, dst->in_status);
ISP_IOXGET_16(isp, &src->in_task_flags, dst->in_task_flags);
ISP_IOXGET_16(isp, &src->in_seqid, dst->in_seqid);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_notify_fc_e(ispsoftc_t *isp, in_fcentry_e_t *src,
in_fcentry_e_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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXGET_32(isp, &src->in_reserved, dst->in_reserved);
ISP_IOXGET_16(isp, &src->in_iid, dst->in_iid);
ISP_IOXGET_16(isp, &src->in_scclun, dst->in_scclun);
ISP_IOXGET_32(isp, &src->in_reserved2, dst->in_reserved2);
ISP_IOXGET_16(isp, &src->in_status, dst->in_status);
ISP_IOXGET_16(isp, &src->in_task_flags, dst->in_task_flags);
ISP_IOXGET_16(isp, &src->in_seqid, dst->in_seqid);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
}
Remove 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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_notify_24xx(ispsoftc_t *isp, in_fcentry_24xx_t *src,
in_fcentry_24xx_t *dst)
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
{
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
int i;
isp_get_hdr(isp, &src->in_header, &dst->in_header);
ISP_IOXGET_32(isp, &src->in_reserved, dst->in_reserved);
ISP_IOXGET_16(isp, &src->in_nphdl, dst->in_nphdl);
ISP_IOXGET_16(isp, &src->in_reserved1, dst->in_reserved1);
ISP_IOXGET_16(isp, &src->in_flags, dst->in_flags);
ISP_IOXGET_16(isp, &src->in_srr_rxid, dst->in_srr_rxid);
ISP_IOXGET_16(isp, &src->in_status, dst->in_status);
ISP_IOXGET_8(isp, &src->in_status_subcode, dst->in_status_subcode);
ISP_IOXGET_16(isp, &src->in_reserved2, dst->in_reserved2);
ISP_IOXGET_32(isp, &src->in_rxid, dst->in_rxid);
ISP_IOXGET_16(isp, &src->in_srr_reloff_hi, dst->in_srr_reloff_hi);
ISP_IOXGET_16(isp, &src->in_srr_reloff_lo, dst->in_srr_reloff_lo);
ISP_IOXGET_16(isp, &src->in_srr_iu, dst->in_srr_iu);
ISP_IOXGET_16(isp, &src->in_srr_oxid, dst->in_srr_oxid);
for (i = 0; i < 18; i++) {
ISP_IOXGET_8(isp, &src->in_reserved3[i], dst->in_reserved3[i]);
}
ISP_IOXGET_8(isp, &src->in_reserved4, dst->in_reserved4);
ISP_IOXGET_8(isp, &src->in_vpindex, dst->in_vpindex);
ISP_IOXGET_32(isp, &src->in_reserved5, dst->in_reserved5);
ISP_IOXGET_16(isp, &src->in_portid_lo, dst->in_portid_lo);
ISP_IOXGET_8(isp, &src->in_portid_hi, dst->in_portid_hi);
ISP_IOXGET_8(isp, &src->in_reserved6, dst->in_reserved6);
ISP_IOXGET_16(isp, &src->in_reserved7, dst->in_reserved7);
ISP_IOXGET_16(isp, &src->in_oxid, dst->in_oxid);
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_notify_ack(ispsoftc_t *isp, na_entry_t *src, na_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXPUT_32(isp, src->na_reserved, &dst->na_reserved);
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->na_lun, &dst->na_iid);
ISP_IOXPUT_8(isp, src->na_iid, &dst->na_lun);
ISP_IOXPUT_8(isp, src->na_status, &dst->na_event);
ISP_IOXPUT_8(isp, src->na_event, &dst->na_status);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_8(isp, src->na_lun, &dst->na_lun);
ISP_IOXPUT_8(isp, src->na_iid, &dst->na_iid);
ISP_IOXPUT_8(isp, src->na_status, &dst->na_status);
ISP_IOXPUT_8(isp, src->na_event, &dst->na_event);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_32(isp, src->na_flags, &dst->na_flags);
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 < NA_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->na_reserved3[i],
&dst->na_reserved3[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_notify_ack(ispsoftc_t *isp, na_entry_t *src, na_entry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXGET_32(isp, &src->na_reserved, dst->na_reserved);
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)) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->na_lun, dst->na_iid);
ISP_IOXGET_8(isp, &src->na_iid, dst->na_lun);
ISP_IOXGET_8(isp, &src->na_status, dst->na_event);
ISP_IOXGET_8(isp, &src->na_event, dst->na_status);
Major restructuring for swizzling to the request queue and unswizzling from the response queue. Instead of the ad hoc ISP_SWIZZLE_REQUEST, we now have a complete set of inline functions in isp_inline.h. Each platform is responsible for providing just one of a set of ISP_IOX_{GET,PUT}{8,16,32} macros. The reason this needs to be done is that we need to have a single set of functions that will work correctly on multiple architectures for both little and big endian machines. It also needs to work correctly in the case that we have the request or response queues in memory that has to be treated specially (e.g., have ddi_dma_sync called on it for Solaris after we update it or before we read from it). It also has to handle the SBus cards (for platforms that have them) which, while on a Big Endian machine, do *not* require *most* of the request/response queue entry fields to be swizzled or unswizzled. One thing that falls out of this is that we no longer build requests in the request queue itself. Instead, we build the request locally (e.g., on the stack) and then as part of the swizzling operation, copy it to the request queue entry we've allocated. I thought long and hard about whether this was too expensive a change to make as it in a lot of cases requires an extra copy. On balance, the flexbility is worth it. With any luck, the entry that we build locally stays in a processor writeback cache (after all, it's only 64 bytes) so that the cost of actually flushing it to the memory area that is the shared queue with the PCI device is not all that expensive. We may examine this again and try to get clever in the future to try and avoid copies. Another change that falls out of this is that MEMORYBARRIER should be taken a lot more seriously. The macro ISP_ADD_REQUEST does a MEMORYBARRIER on the entry being added. But there had been many other places this had been missing. It's now very important that it be done. Additional changes: Fix a longstanding buglet of sorts. When we get an entry via isp_getrqentry, the iptr value that gets returned is the value we intend to eventually plug into the ISP registers as the entry *one past* the last one we've written- *not* the current entry we're updating. All along we've been calling sync functions on the wrong index value. Argh. The 'fix' here is to rename all 'iptr' variables as 'nxti' to remember that this is the 'next' pointer- not the current pointer. Devote a single bit to mboxbsy- and set aside bits for output mbox registers that we need to pick up- we can have at least one command which does not have any defined output registers (MBOX_EXECUTE_FIRMWARE). MFC after: 2 weeks
2001-12-11 00:18:45 +00:00
} else {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_8(isp, &src->na_lun, dst->na_lun);
ISP_IOXGET_8(isp, &src->na_iid, dst->na_iid);
ISP_IOXGET_8(isp, &src->na_status, dst->na_status);
ISP_IOXGET_8(isp, &src->na_event, dst->na_event);
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
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_32(isp, &src->na_flags, dst->na_flags);
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 < NA_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->na_reserved3[i],
dst->na_reserved3[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_notify_ack_fc(ispsoftc_t *isp, na_fcentry_t *src,
na_fcentry_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXPUT_32(isp, src->na_reserved, &dst->na_reserved);
ISP_IOXPUT_8(isp, src->na_reserved1, &dst->na_reserved1);
ISP_IOXPUT_8(isp, src->na_iid, &dst->na_iid);
ISP_IOXPUT_16(isp, src->na_response, &dst->na_response);
ISP_IOXPUT_16(isp, src->na_flags, &dst->na_flags);
ISP_IOXPUT_16(isp, src->na_reserved2, &dst->na_reserved2);
ISP_IOXPUT_16(isp, src->na_status, &dst->na_status);
ISP_IOXPUT_16(isp, src->na_task_flags, &dst->na_task_flags);
ISP_IOXPUT_16(isp, src->na_seqid, &dst->na_seqid);
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 < NA2_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->na_reserved3[i],
&dst->na_reserved3[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_notify_ack_fc_e(ispsoftc_t *isp, na_fcentry_e_t *src,
na_fcentry_e_t *dst)
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXPUT_32(isp, src->na_reserved, &dst->na_reserved);
ISP_IOXPUT_16(isp, src->na_iid, &dst->na_iid);
ISP_IOXPUT_16(isp, src->na_response, &dst->na_response);
ISP_IOXPUT_16(isp, src->na_flags, &dst->na_flags);
ISP_IOXPUT_16(isp, src->na_reserved2, &dst->na_reserved2);
ISP_IOXPUT_16(isp, src->na_status, &dst->na_status);
ISP_IOXPUT_16(isp, src->na_task_flags, &dst->na_task_flags);
ISP_IOXPUT_16(isp, src->na_seqid, &dst->na_seqid);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < NA2_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXPUT_16(isp, src->na_reserved3[i],
&dst->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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_notify_24xx_ack(ispsoftc_t *isp, na_fcentry_24xx_t *src,
na_fcentry_24xx_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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_put_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXPUT_32(isp, src->na_handle, &dst->na_handle);
ISP_IOXPUT_16(isp, src->na_nphdl, &dst->na_nphdl);
ISP_IOXPUT_16(isp, src->na_reserved1, &dst->na_reserved1);
ISP_IOXPUT_16(isp, src->na_flags, &dst->na_flags);
ISP_IOXPUT_16(isp, src->na_srr_rxid, &dst->na_srr_rxid);
ISP_IOXPUT_16(isp, src->na_status, &dst->na_status);
ISP_IOXPUT_8(isp, src->na_status_subcode, &dst->na_status_subcode);
ISP_IOXPUT_16(isp, src->na_reserved2, &dst->na_reserved2);
ISP_IOXPUT_32(isp, src->na_rxid, &dst->na_rxid);
ISP_IOXPUT_16(isp, src->na_srr_reloff_hi, &dst->na_srr_reloff_hi);
ISP_IOXPUT_16(isp, src->na_srr_reloff_lo, &dst->na_srr_reloff_lo);
ISP_IOXPUT_16(isp, src->na_srr_iu, &dst->na_srr_iu);
ISP_IOXPUT_16(isp, src->na_srr_flags, &dst->na_srr_flags);
for (i = 0; i < 18; i++) {
ISP_IOXPUT_8(isp, src->na_reserved3[i], &dst->na_reserved3[i]);
}
ISP_IOXPUT_8(isp, src->na_reserved4, &dst->na_reserved4);
ISP_IOXPUT_8(isp, src->na_vpindex, &dst->na_vpindex);
ISP_IOXPUT_8(isp, src->na_srr_reject_vunique,
&dst->na_srr_reject_vunique);
ISP_IOXPUT_8(isp, src->na_srr_reject_explanation,
&dst->na_srr_reject_explanation);
ISP_IOXPUT_8(isp, src->na_srr_reject_code, &dst->na_srr_reject_code);
ISP_IOXPUT_8(isp, src->na_reserved5, &dst->na_reserved5);
for (i = 0; i < 6; i++) {
ISP_IOXPUT_8(isp, src->na_reserved6[i], &dst->na_reserved6[i]);
}
ISP_IOXPUT_16(isp, src->na_oxid, &dst->na_oxid);
}
void
isp_get_notify_ack_fc(ispsoftc_t *isp, na_fcentry_t *src,
na_fcentry_t *dst)
{
int i;
isp_get_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXGET_32(isp, &src->na_reserved, dst->na_reserved);
ISP_IOXGET_8(isp, &src->na_reserved1, dst->na_reserved1);
ISP_IOXGET_8(isp, &src->na_iid, dst->na_iid);
ISP_IOXGET_16(isp, &src->na_response, dst->na_response);
ISP_IOXGET_16(isp, &src->na_flags, dst->na_flags);
ISP_IOXGET_16(isp, &src->na_reserved2, dst->na_reserved2);
ISP_IOXGET_16(isp, &src->na_status, dst->na_status);
ISP_IOXGET_16(isp, &src->na_task_flags, dst->na_task_flags);
ISP_IOXGET_16(isp, &src->na_seqid, dst->na_seqid);
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 < NA2_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->na_reserved3[i],
dst->na_reserved3[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
}
}
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
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_notify_ack_fc_e(ispsoftc_t *isp, na_fcentry_e_t *src,
na_fcentry_e_t *dst)
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;
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
isp_get_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXGET_32(isp, &src->na_reserved, dst->na_reserved);
ISP_IOXGET_16(isp, &src->na_iid, dst->na_iid);
ISP_IOXGET_16(isp, &src->na_response, dst->na_response);
ISP_IOXGET_16(isp, &src->na_flags, dst->na_flags);
ISP_IOXGET_16(isp, &src->na_reserved2, dst->na_reserved2);
ISP_IOXGET_16(isp, &src->na_status, dst->na_status);
ISP_IOXGET_16(isp, &src->na_task_flags, dst->na_task_flags);
ISP_IOXGET_16(isp, &src->na_seqid, dst->na_seqid);
First of several commits as this driver is dusted off and maybe brought up to date. Principle changes for this reelase is to support 2K Port Login firmware. This allows us to support the 2322 (and 2422 4Gb) cards which only come with the 2K Port Login firmware. The 2322 should now work- but we don't have firmware sets for it in ispfw (as the change to load 2K Port Login f/w hasn't been made- that f/w is so big it has to be loaded in more than one chunk). Other changes are the beginnings of cleaning up some long standing target mode issues. The next changes here will incorporate a lot of bug fixes from others. Finally, some copyright cleanup and attempts to make the parts of the driver that are FreeBSD specific start conforming more to FreeBSD style. MFC after: 1 month
2006-01-23 06:23:37 +00:00
for (i = 0; i < NA2_RSVDLEN; i++) {
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
ISP_IOXGET_16(isp, &src->na_reserved3[i],
dst->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
}
}
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
void
isp_get_notify_ack_24xx(ispsoftc_t *isp, na_fcentry_24xx_t *src,
na_fcentry_24xx_t *dst)
{
int i;
isp_get_hdr(isp, &src->na_header, &dst->na_header);
ISP_IOXGET_32(isp, &src->na_handle, dst->na_handle);
ISP_IOXGET_16(isp, &src->na_nphdl, dst->na_nphdl);
ISP_IOXGET_16(isp, &src->na_reserved1, dst->na_reserved1);
ISP_IOXGET_16(isp, &src->na_flags, dst->na_flags);
ISP_IOXGET_16(isp, &src->na_srr_rxid, dst->na_srr_rxid);
ISP_IOXGET_16(isp, &src->na_status, dst->na_status);
ISP_IOXGET_8(isp, &src->na_status_subcode, dst->na_status_subcode);
ISP_IOXGET_16(isp, &src->na_reserved2, dst->na_reserved2);
ISP_IOXGET_32(isp, &src->na_rxid, dst->na_rxid);
ISP_IOXGET_16(isp, &src->na_srr_reloff_hi, dst->na_srr_reloff_hi);
ISP_IOXGET_16(isp, &src->na_srr_reloff_lo, dst->na_srr_reloff_lo);
ISP_IOXGET_16(isp, &src->na_srr_iu, dst->na_srr_iu);
ISP_IOXGET_16(isp, &src->na_srr_flags, dst->na_srr_flags);
for (i = 0; i < 18; i++) {
ISP_IOXGET_8(isp, &src->na_reserved3[i], dst->na_reserved3[i]);
}
ISP_IOXGET_8(isp, &src->na_reserved4, dst->na_reserved4);
ISP_IOXGET_8(isp, &src->na_vpindex, dst->na_vpindex);
ISP_IOXGET_8(isp, &src->na_srr_reject_vunique,
dst->na_srr_reject_vunique);
ISP_IOXGET_8(isp, &src->na_srr_reject_explanation,
dst->na_srr_reject_explanation);
ISP_IOXGET_8(isp, &src->na_srr_reject_code, dst->na_srr_reject_code);
ISP_IOXGET_8(isp, &src->na_reserved5, dst->na_reserved5);
for (i = 0; i < 6; i++) {
ISP_IOXGET_8(isp, &src->na_reserved6[i], dst->na_reserved6[i]);
}
ISP_IOXGET_16(isp, &src->na_oxid, dst->na_oxid);
}
void
isp_get_abts(ispsoftc_t *isp, abts_t *src, abts_t *dst)
{
int i;
isp_get_hdr(isp, &src->abts_header, &dst->abts_header);
for (i = 0; i < 6; i++) {
ISP_IOXGET_8(isp, &src->abts_reserved0[i],
dst->abts_reserved0[i]);
}
ISP_IOXGET_16(isp, &src->abts_nphdl, dst->abts_nphdl);
ISP_IOXGET_16(isp, &src->abts_reserved1, dst->abts_reserved1);
ISP_IOXGET_16(isp, &src->abts_sof, dst->abts_sof);
ISP_IOXGET_32(isp, &src->abts_rxid_abts, dst->abts_rxid_abts);
ISP_IOXGET_16(isp, &src->abts_did_lo, dst->abts_did_lo);
ISP_IOXGET_8(isp, &src->abts_did_hi, dst->abts_did_hi);
ISP_IOXGET_8(isp, &src->abts_r_ctl, dst->abts_r_ctl);
ISP_IOXGET_16(isp, &src->abts_sid_lo, dst->abts_sid_lo);
ISP_IOXGET_8(isp, &src->abts_sid_hi, dst->abts_sid_hi);
ISP_IOXGET_8(isp, &src->abts_cs_ctl, dst->abts_cs_ctl);
ISP_IOXGET_16(isp, &src->abts_fs_ctl, dst->abts_fs_ctl);
ISP_IOXGET_8(isp, &src->abts_f_ctl, dst->abts_f_ctl);
ISP_IOXGET_8(isp, &src->abts_type, dst->abts_type);
ISP_IOXGET_16(isp, &src->abts_seq_cnt, dst->abts_seq_cnt);
ISP_IOXGET_8(isp, &src->abts_df_ctl, dst->abts_df_ctl);
ISP_IOXGET_8(isp, &src->abts_seq_id, dst->abts_seq_id);
ISP_IOXGET_16(isp, &src->abts_rx_id, dst->abts_rx_id);
ISP_IOXGET_16(isp, &src->abts_ox_id, dst->abts_ox_id);
ISP_IOXGET_32(isp, &src->abts_param, dst->abts_param);
for (i = 0; i < 16; i++) {
ISP_IOXGET_8(isp, &src->abts_reserved2[i],
dst->abts_reserved2[i]);
}
ISP_IOXGET_32(isp, &src->abts_rxid_task, dst->abts_rxid_task);
}
void
isp_put_abts_rsp(ispsoftc_t *isp, abts_rsp_t *src, abts_rsp_t *dst)
{
int i;
isp_put_hdr(isp, &src->abts_rsp_header, &dst->abts_rsp_header);
ISP_IOXPUT_32(isp, src->abts_rsp_handle, &dst->abts_rsp_handle);
ISP_IOXPUT_16(isp, src->abts_rsp_status, &dst->abts_rsp_status);
ISP_IOXPUT_16(isp, src->abts_rsp_nphdl, &dst->abts_rsp_nphdl);
ISP_IOXPUT_16(isp, src->abts_rsp_ctl_flags, &dst->abts_rsp_ctl_flags);
ISP_IOXPUT_16(isp, src->abts_rsp_sof, &dst->abts_rsp_sof);
ISP_IOXPUT_32(isp, src->abts_rsp_rxid_abts, &dst->abts_rsp_rxid_abts);
ISP_IOXPUT_16(isp, src->abts_rsp_did_lo, &dst->abts_rsp_did_lo);
ISP_IOXPUT_8(isp, src->abts_rsp_did_hi, &dst->abts_rsp_did_hi);
ISP_IOXPUT_8(isp, src->abts_rsp_r_ctl, &dst->abts_rsp_r_ctl);
ISP_IOXPUT_16(isp, src->abts_rsp_sid_lo, &dst->abts_rsp_sid_lo);
ISP_IOXPUT_8(isp, src->abts_rsp_sid_hi, &dst->abts_rsp_sid_hi);
ISP_IOXPUT_8(isp, src->abts_rsp_cs_ctl, &dst->abts_rsp_cs_ctl);
ISP_IOXPUT_16(isp, src->abts_rsp_f_ctl_lo, &dst->abts_rsp_f_ctl_lo);
ISP_IOXPUT_8(isp, src->abts_rsp_f_ctl_hi, &dst->abts_rsp_f_ctl_hi);
ISP_IOXPUT_8(isp, src->abts_rsp_type, &dst->abts_rsp_type);
ISP_IOXPUT_16(isp, src->abts_rsp_seq_cnt, &dst->abts_rsp_seq_cnt);
ISP_IOXPUT_8(isp, src->abts_rsp_df_ctl, &dst->abts_rsp_df_ctl);
ISP_IOXPUT_8(isp, src->abts_rsp_seq_id, &dst->abts_rsp_seq_id);
ISP_IOXPUT_16(isp, src->abts_rsp_rx_id, &dst->abts_rsp_rx_id);
ISP_IOXPUT_16(isp, src->abts_rsp_ox_id, &dst->abts_rsp_ox_id);
ISP_IOXPUT_32(isp, src->abts_rsp_param, &dst->abts_rsp_param);
if (src->abts_rsp_r_ctl == BA_ACC) {
ISP_IOXPUT_16(isp, src->abts_rsp_payload.ba_acc.reserved,
&dst->abts_rsp_payload.ba_acc.reserved);
ISP_IOXPUT_8(isp, src->abts_rsp_payload.ba_acc.last_seq_id,
&dst->abts_rsp_payload.ba_acc.last_seq_id);
ISP_IOXPUT_8(isp, src->abts_rsp_payload.ba_acc.seq_id_valid,
&dst->abts_rsp_payload.ba_acc.seq_id_valid);
ISP_IOXPUT_16(isp, src->abts_rsp_payload.ba_acc.aborted_rx_id,
&dst->abts_rsp_payload.ba_acc.aborted_rx_id);
ISP_IOXPUT_16(isp, src->abts_rsp_payload.ba_acc.aborted_ox_id,
&dst->abts_rsp_payload.ba_acc.aborted_ox_id);
ISP_IOXPUT_16(isp, src->abts_rsp_payload.ba_acc.high_seq_cnt,
&dst->abts_rsp_payload.ba_acc.high_seq_cnt);
ISP_IOXPUT_16(isp, src->abts_rsp_payload.ba_acc.low_seq_cnt,
&dst->abts_rsp_payload.ba_acc.low_seq_cnt);
for (i = 0; i < 4; i++) {
ISP_IOXPUT_16(isp,
src->abts_rsp_payload.ba_acc.reserved2[i],
&dst->abts_rsp_payload.ba_acc.reserved2[i]);
}
} else if (src->abts_rsp_r_ctl == BA_RJT) {
ISP_IOXPUT_8(isp, src->abts_rsp_payload.ba_rjt.vendor_unique,
&dst->abts_rsp_payload.ba_rjt.vendor_unique);
ISP_IOXPUT_8(isp, src->abts_rsp_payload.ba_rjt.explanation,
&dst->abts_rsp_payload.ba_rjt.explanation);
ISP_IOXPUT_8(isp, src->abts_rsp_payload.ba_rjt.reason,
&dst->abts_rsp_payload.ba_rjt.reason);
ISP_IOXPUT_8(isp, src->abts_rsp_payload.ba_rjt.reserved,
&dst->abts_rsp_payload.ba_rjt.reserved);
for (i = 0; i < 12; i++) {
ISP_IOXPUT_16(isp,
src->abts_rsp_payload.ba_rjt.reserved2[i],
&dst->abts_rsp_payload.ba_rjt.reserved2[i]);
}
} else {
for (i = 0; i < 16; i++) {
ISP_IOXPUT_8(isp, src->abts_rsp_payload.reserved[i],
&dst->abts_rsp_payload.reserved[i]);
}
}
ISP_IOXPUT_32(isp, src->abts_rsp_rxid_task, &dst->abts_rsp_rxid_task);
}
void
isp_get_abts_rsp(ispsoftc_t *isp, abts_rsp_t *src, abts_rsp_t *dst)
{
int i;
isp_get_hdr(isp, &src->abts_rsp_header, &dst->abts_rsp_header);
ISP_IOXGET_32(isp, &src->abts_rsp_handle, dst->abts_rsp_handle);
ISP_IOXGET_16(isp, &src->abts_rsp_status, dst->abts_rsp_status);
ISP_IOXGET_16(isp, &src->abts_rsp_nphdl, dst->abts_rsp_nphdl);
ISP_IOXGET_16(isp, &src->abts_rsp_ctl_flags, dst->abts_rsp_ctl_flags);
ISP_IOXGET_16(isp, &src->abts_rsp_sof, dst->abts_rsp_sof);
ISP_IOXGET_32(isp, &src->abts_rsp_rxid_abts, dst->abts_rsp_rxid_abts);
ISP_IOXGET_16(isp, &src->abts_rsp_did_lo, dst->abts_rsp_did_lo);
ISP_IOXGET_8(isp, &src->abts_rsp_did_hi, dst->abts_rsp_did_hi);
ISP_IOXGET_8(isp, &src->abts_rsp_r_ctl, dst->abts_rsp_r_ctl);
ISP_IOXGET_16(isp, &src->abts_rsp_sid_lo, dst->abts_rsp_sid_lo);
ISP_IOXGET_8(isp, &src->abts_rsp_sid_hi, dst->abts_rsp_sid_hi);
ISP_IOXGET_8(isp, &src->abts_rsp_cs_ctl, dst->abts_rsp_cs_ctl);
ISP_IOXGET_16(isp, &src->abts_rsp_f_ctl_lo, dst->abts_rsp_f_ctl_lo);
ISP_IOXGET_8(isp, &src->abts_rsp_f_ctl_hi, dst->abts_rsp_f_ctl_hi);
ISP_IOXGET_8(isp, &src->abts_rsp_type, dst->abts_rsp_type);
ISP_IOXGET_16(isp, &src->abts_rsp_seq_cnt, dst->abts_rsp_seq_cnt);
ISP_IOXGET_8(isp, &src->abts_rsp_df_ctl, dst->abts_rsp_df_ctl);
ISP_IOXGET_8(isp, &src->abts_rsp_seq_id, dst->abts_rsp_seq_id);
ISP_IOXGET_16(isp, &src->abts_rsp_rx_id, dst->abts_rsp_rx_id);
ISP_IOXGET_16(isp, &src->abts_rsp_ox_id, dst->abts_rsp_ox_id);
ISP_IOXGET_32(isp, &src->abts_rsp_param, dst->abts_rsp_param);
for (i = 0; i < 8; i++) {
ISP_IOXGET_8(isp, &src->abts_rsp_payload.rsp.reserved[i],
dst->abts_rsp_payload.rsp.reserved[i]);
}
ISP_IOXGET_32(isp, &src->abts_rsp_payload.rsp.subcode1,
dst->abts_rsp_payload.rsp.subcode1);
ISP_IOXGET_32(isp, &src->abts_rsp_payload.rsp.subcode2,
dst->abts_rsp_payload.rsp.subcode2);
ISP_IOXGET_32(isp, &src->abts_rsp_rxid_task, dst->abts_rsp_rxid_task);
}
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 */
Add 4Gb (24XX) support and lay the foundation for a lot of new stuff.
2006-11-02 03:21:32 +00:00
/*
* vim:ts=8:sw=8
*/
Reference in New Issue Copy Permalink