NetBSD/sys/dev/ic/aic79xx_inline.h

1004 lines
29 KiB
C
Raw Normal View History

/* $NetBSD: aic79xx_inline.h,v 1.13 2006/02/16 20:17:16 perry Exp $ */
/*
* Inline routines shareable across OS platforms.
*
* Copyright (c) 1994-2001 Justin T. Gibbs.
* Copyright (c) 2000-2003 Adaptec Inc.
* 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,
* without modification.
* 2. Redistributions in binary form must reproduce at minimum a disclaimer
* substantially similar to the "NO WARRANTY" disclaimer below
* ("Disclaimer") and any redistribution must be conditioned upon
* including a substantially similar Disclaimer requirement for further
* binary redistribution.
* 3. Neither the names of the above-listed copyright holders nor the names
* of any contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* Alternatively, this software may be distributed under the terms of the
* GNU General Public License ("GPL") version 2 as published by the Free
* Software Foundation.
*
* NO WARRANTY
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* HOLDERS OR CONTRIBUTORS BE LIABLE FOR 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 DAMAGES.
*
* Id: //depot/aic7xxx/aic7xxx/aic79xx_inline.h#51 $
*
* $FreeBSD: src/sys/dev/aic7xxx/aic79xx_inline.h,v 1.12 2003/06/28 04:43:19 gibbs Exp $
*/
/*
* Ported from FreeBSD by Pascal Renauld, Network Storage Solutions, Inc. - April 2003
*/
#ifndef _AIC79XX_INLINE_H_
#define _AIC79XX_INLINE_H_
/******************************** Debugging ***********************************/
static __inline char *ahd_name(struct ahd_softc *);
static __inline char *
ahd_name(struct ahd_softc *ahd)
{
return (ahd->name);
}
/************************ Sequencer Execution Control *************************/
static __inline void ahd_known_modes(struct ahd_softc *, ahd_mode, ahd_mode);
static __inline ahd_mode_state ahd_build_mode_state(struct ahd_softc *,
ahd_mode, ahd_mode);
static __inline void ahd_extract_mode_state(struct ahd_softc *,
ahd_mode_state, ahd_mode *, ahd_mode *);
static __inline void ahd_set_modes(struct ahd_softc *, ahd_mode, ahd_mode);
static __inline void ahd_update_modes(struct ahd_softc *);
static __inline void ahd_assert_modes(struct ahd_softc *, ahd_mode,
ahd_mode, const char *, int);
static __inline ahd_mode_state ahd_save_modes(struct ahd_softc *);
static __inline void ahd_restore_modes(struct ahd_softc *, ahd_mode_state);
static __inline int ahd_is_paused(struct ahd_softc *);
static __inline void ahd_pause(struct ahd_softc *);
static __inline void ahd_unpause(struct ahd_softc *);
static __inline void
ahd_known_modes(struct ahd_softc *ahd, ahd_mode src, ahd_mode dst)
{
ahd->src_mode = src;
ahd->dst_mode = dst;
ahd->saved_src_mode = src;
ahd->saved_dst_mode = dst;
}
static __inline ahd_mode_state
ahd_build_mode_state(struct ahd_softc *ahd, ahd_mode src, ahd_mode dst)
{
return ((src << SRC_MODE_SHIFT) | (dst << DST_MODE_SHIFT));
}
static __inline void
ahd_extract_mode_state(struct ahd_softc *ahd, ahd_mode_state state,
ahd_mode *src, ahd_mode *dst)
{
*src = (state & SRC_MODE) >> SRC_MODE_SHIFT;
*dst = (state & DST_MODE) >> DST_MODE_SHIFT;
}
static __inline void
ahd_set_modes(struct ahd_softc *ahd, ahd_mode src, ahd_mode dst)
{
if (ahd->src_mode == src && ahd->dst_mode == dst)
return;
#ifdef AHD_DEBUG
if (ahd->src_mode == AHD_MODE_UNKNOWN
|| ahd->dst_mode == AHD_MODE_UNKNOWN)
panic("Setting mode prior to saving it.\n");
if ((ahd_debug & AHD_SHOW_MODEPTR) != 0)
printf("%s: Setting mode 0x%x\n", ahd_name(ahd),
ahd_build_mode_state(ahd, src, dst));
#endif
ahd_outb(ahd, MODE_PTR, ahd_build_mode_state(ahd, src, dst));
ahd->src_mode = src;
ahd->dst_mode = dst;
}
static __inline void
ahd_update_modes(struct ahd_softc *ahd)
{
ahd_mode_state mode_ptr;
ahd_mode src;
ahd_mode dst;
mode_ptr = ahd_inb(ahd, MODE_PTR);
#ifdef AHD_DEBUG
if ((ahd_debug & AHD_SHOW_MODEPTR) != 0)
printf("Reading mode 0x%x\n", mode_ptr);
#endif
ahd_extract_mode_state(ahd, mode_ptr, &src, &dst);
ahd_known_modes(ahd, src, dst);
}
static __inline void
ahd_assert_modes(struct ahd_softc *ahd, ahd_mode srcmode,
ahd_mode dstmode, const char *file, int line)
{
#ifdef AHD_DEBUG
if ((srcmode & AHD_MK_MSK(ahd->src_mode)) == 0
|| (dstmode & AHD_MK_MSK(ahd->dst_mode)) == 0) {
panic("%s:%s:%d: Mode assertion failed.\n",
ahd_name(ahd), file, line);
}
#endif
}
static __inline ahd_mode_state
ahd_save_modes(struct ahd_softc *ahd)
{
if (ahd->src_mode == AHD_MODE_UNKNOWN
|| ahd->dst_mode == AHD_MODE_UNKNOWN)
ahd_update_modes(ahd);
return (ahd_build_mode_state(ahd, ahd->src_mode, ahd->dst_mode));
}
static __inline void
ahd_restore_modes(struct ahd_softc *ahd, ahd_mode_state state)
{
ahd_mode src;
ahd_mode dst;
ahd_extract_mode_state(ahd, state, &src, &dst);
ahd_set_modes(ahd, src, dst);
}
#define AHD_ASSERT_MODES(ahd, source, dest) \
ahd_assert_modes(ahd, source, dest, __FILE__, __LINE__);
/*
* Determine whether the sequencer has halted code execution.
* Returns non-zero status if the sequencer is stopped.
*/
static __inline int
ahd_is_paused(struct ahd_softc *ahd)
{
return ((ahd_inb(ahd, HCNTRL) & PAUSE) != 0);
}
/*
* Request that the sequencer stop and wait, indefinitely, for it
* to stop. The sequencer will only acknowledge that it is paused
* once it has reached an instruction boundary and PAUSEDIS is
* cleared in the SEQCTL register. The sequencer may use PAUSEDIS
* for critical sections.
*/
static __inline void
ahd_pause(struct ahd_softc *ahd)
{
ahd_outb(ahd, HCNTRL, ahd->pause);
/*
* Since the sequencer can disable pausing in a critical section, we
* must loop until it actually stops.
*/
while (ahd_is_paused(ahd) == 0)
;
}
/*
* Allow the sequencer to continue program execution.
* We check here to ensure that no additional interrupt
* sources that would cause the sequencer to halt have been
* asserted. If, for example, a SCSI bus reset is detected
* while we are fielding a different, pausing, interrupt type,
* we don't want to release the sequencer before going back
* into our interrupt handler and dealing with this new
* condition.
*/
static __inline void
ahd_unpause(struct ahd_softc *ahd)
{
/*
* Automatically restore our modes to those saved
* prior to the first change of the mode.
*/
if (ahd->saved_src_mode != AHD_MODE_UNKNOWN
&& ahd->saved_dst_mode != AHD_MODE_UNKNOWN) {
if ((ahd->flags & AHD_UPDATE_PEND_CMDS) != 0)
ahd_reset_cmds_pending(ahd);
ahd_set_modes(ahd, ahd->saved_src_mode, ahd->saved_dst_mode);
}
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
if ((ahd_inb(ahd, INTSTAT) & ~CMDCMPLT) == 0)
ahd_outb(ahd, HCNTRL, ahd->unpause);
ahd_known_modes(ahd, AHD_MODE_UNKNOWN, AHD_MODE_UNKNOWN);
}
/*********************** Scatter Gather List Handling *************************/
static __inline void *ahd_sg_setup(struct ahd_softc *, struct scb *,
void *, bus_addr_t, bus_size_t, int);
static __inline void ahd_setup_scb_common(struct ahd_softc *, struct scb *);
static __inline void ahd_setup_data_scb(struct ahd_softc *, struct scb *);
static __inline void ahd_setup_noxfer_scb(struct ahd_softc *, struct scb *);
static __inline void *
ahd_sg_setup(struct ahd_softc *ahd, struct scb *scb,
void *sgptr, bus_addr_t addr, bus_size_t len, int last)
{
scb->sg_count++;
if (sizeof(bus_addr_t) > 4
&& (ahd->flags & AHD_64BIT_ADDRESSING) != 0) {
struct ahd_dma64_seg *sg;
sg = (struct ahd_dma64_seg *)sgptr;
sg->addr = ahd_htole64(addr);
sg->len = ahd_htole32(len | (last ? AHD_DMA_LAST_SEG : 0));
return (sg + 1);
} else {
struct ahd_dma_seg *sg;
sg = (struct ahd_dma_seg *)sgptr;
sg->addr = ahd_htole32(addr & 0xFFFFFFFF);
sg->len = ahd_htole32(len | ((addr >> 8) & 0x7F000000)
| (last ? AHD_DMA_LAST_SEG : 0));
return (sg + 1);
}
}
static __inline void
ahd_setup_scb_common(struct ahd_softc *ahd, struct scb *scb)
{
/* XXX Handle target mode SCBs. */
scb->crc_retry_count = 0;
if ((scb->flags & SCB_PACKETIZED) != 0) {
/* XXX what about ACA?? It is type 4, but TAG_TYPE == 0x3. */
scb->hscb->task_attribute = scb->hscb->control & SCB_TAG_TYPE;
} else {
if (ahd_get_transfer_length(scb) & 0x01)
scb->hscb->task_attribute = SCB_XFERLEN_ODD;
else
scb->hscb->task_attribute = 0;
}
if (scb->hscb->cdb_len <= MAX_CDB_LEN_WITH_SENSE_ADDR
|| (scb->hscb->cdb_len & SCB_CDB_LEN_PTR) != 0)
scb->hscb->shared_data.idata.cdb_plus_saddr.sense_addr =
ahd_htole32(scb->sense_busaddr);
}
static __inline void
ahd_setup_data_scb(struct ahd_softc *ahd, struct scb *scb)
{
/*
* Copy the first SG into the "current" data ponter area.
*/
if ((ahd->flags & AHD_64BIT_ADDRESSING) != 0) {
struct ahd_dma64_seg *sg;
sg = (struct ahd_dma64_seg *)scb->sg_list;
scb->hscb->dataptr = sg->addr;
scb->hscb->datacnt = sg->len;
} else {
struct ahd_dma_seg *sg;
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
uint32_t *dataptr_words;
sg = (struct ahd_dma_seg *)scb->sg_list;
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
dataptr_words = (uint32_t*)&scb->hscb->dataptr;
dataptr_words[0] = sg->addr;
dataptr_words[1] = 0;
if ((ahd->flags & AHD_39BIT_ADDRESSING) != 0) {
uint64_t high_addr;
high_addr = ahd_le32toh(sg->len) & 0x7F000000;
scb->hscb->dataptr |= ahd_htole64(high_addr << 8);
}
scb->hscb->datacnt = sg->len;
}
/*
* Note where to find the SG entries in bus space.
2005-02-27 03:26:58 +03:00
* We also set the full residual flag which the
* sequencer will clear as soon as a data transfer
* occurs.
*/
scb->hscb->sgptr = ahd_htole32(scb->sg_list_busaddr|SG_FULL_RESID);
}
static __inline void
ahd_setup_noxfer_scb(struct ahd_softc *ahd, struct scb *scb)
{
scb->hscb->sgptr = ahd_htole32(SG_LIST_NULL);
scb->hscb->dataptr = 0;
scb->hscb->datacnt = 0;
}
/************************** Memory mapping routines ***************************/
static __inline size_t ahd_sg_size(struct ahd_softc *);
static __inline void *
ahd_sg_bus_to_virt(struct ahd_softc *, struct scb *,
uint32_t);
static __inline uint32_t
ahd_sg_virt_to_bus(struct ahd_softc *, struct scb *,
void *);
static __inline void ahd_sync_scb(struct ahd_softc *, struct scb *, int);
static __inline void ahd_sync_sglist(struct ahd_softc *, struct scb *, int);
static __inline void ahd_sync_sense(struct ahd_softc *, struct scb *, int);
static __inline uint32_t
ahd_targetcmd_offset(struct ahd_softc *, u_int);
static __inline size_t
ahd_sg_size(struct ahd_softc *ahd)
{
if ((ahd->flags & AHD_64BIT_ADDRESSING) != 0)
return (sizeof(struct ahd_dma64_seg));
return (sizeof(struct ahd_dma_seg));
}
static __inline void *
ahd_sg_bus_to_virt(struct ahd_softc *ahd, struct scb *scb, uint32_t sg_busaddr)
{
bus_addr_t sg_offset;
/* sg_list_phys points to entry 1, not 0 */
sg_offset = sg_busaddr - (scb->sg_list_busaddr - ahd_sg_size(ahd));
return ((uint8_t *)scb->sg_list + sg_offset);
}
static __inline uint32_t
ahd_sg_virt_to_bus(struct ahd_softc *ahd, struct scb *scb, void *sg)
{
bus_addr_t sg_offset;
/* sg_list_phys points to entry 1, not 0 */
sg_offset = ((uint8_t *)sg - (uint8_t *)scb->sg_list)
- ahd_sg_size(ahd);
return (scb->sg_list_busaddr + sg_offset);
}
static __inline void
ahd_sync_scb(struct ahd_softc *ahd, struct scb *scb, int op)
{
2005-02-27 03:26:58 +03:00
ahd_dmamap_sync(ahd, ahd->parent_dmat, scb->hscb_map->dmamap,
/*offset*/(uint8_t*)scb->hscb - scb->hscb_map->vaddr,
/*len*/sizeof(*scb->hscb), op);
}
static __inline void
ahd_sync_sglist(struct ahd_softc *ahd, struct scb *scb, int op)
{
if (scb->sg_count == 0)
return;
ahd_dmamap_sync(ahd, ahd->parent_dmat, scb->sg_map->dmamap,
/*offset*/scb->sg_list_busaddr - ahd_sg_size(ahd),
/*len*/ahd_sg_size(ahd) * scb->sg_count, op);
}
static __inline void
ahd_sync_sense(struct ahd_softc *ahd, struct scb *scb, int op)
{
2005-02-27 03:26:58 +03:00
ahd_dmamap_sync(ahd, ahd->parent_dmat,
scb->sense_map->dmamap,
/*offset*/scb->sense_busaddr,
/*len*/AHD_SENSE_BUFSIZE, op);
}
static __inline uint32_t
ahd_targetcmd_offset(struct ahd_softc *ahd, u_int index)
{
return (((uint8_t *)&ahd->targetcmds[index])
- (uint8_t *)ahd->qoutfifo);
}
/*********************** Miscellaneous Support Functions ***********************/
static __inline void ahd_complete_scb(struct ahd_softc *, struct scb *);
static __inline void ahd_update_residual(struct ahd_softc *, struct scb *);
static __inline struct ahd_initiator_tinfo *
ahd_fetch_transinfo(struct ahd_softc *, char, u_int,
u_int, struct ahd_tmode_tstate **);
static __inline uint16_t
ahd_inw(struct ahd_softc *, u_int);
static __inline void ahd_outw(struct ahd_softc *, u_int, u_int);
static __inline uint32_t
ahd_inl(struct ahd_softc *, u_int);
static __inline void ahd_outl(struct ahd_softc *, u_int, uint32_t);
static __inline uint64_t
ahd_inq(struct ahd_softc *, u_int);
static __inline void ahd_outq(struct ahd_softc *, u_int, uint64_t);
static __inline u_int ahd_get_scbptr(struct ahd_softc *);
static __inline void ahd_set_scbptr(struct ahd_softc *, u_int);
static __inline u_int ahd_get_hnscb_qoff(struct ahd_softc *);
static __inline void ahd_set_hnscb_qoff(struct ahd_softc *, u_int);
static __inline u_int ahd_get_hescb_qoff(struct ahd_softc *);
static __inline void ahd_set_hescb_qoff(struct ahd_softc *, u_int);
static __inline u_int ahd_get_snscb_qoff(struct ahd_softc *);
static __inline void ahd_set_snscb_qoff(struct ahd_softc *, u_int);
static __inline u_int ahd_get_sescb_qoff(struct ahd_softc *);
static __inline void ahd_set_sescb_qoff(struct ahd_softc *, u_int);
static __inline u_int ahd_get_sdscb_qoff(struct ahd_softc *);
static __inline void ahd_set_sdscb_qoff(struct ahd_softc *, u_int);
static __inline u_int ahd_inb_scbram(struct ahd_softc *, u_int);
static __inline u_int ahd_inw_scbram(struct ahd_softc *, u_int);
static __inline uint32_t
ahd_inl_scbram(struct ahd_softc *, u_int);
static __inline uint64_t
ahd_inq_scbram(struct ahd_softc *ahd, u_int offset);
static __inline void ahd_swap_with_next_hscb(struct ahd_softc *,
struct scb *);
static __inline void ahd_queue_scb(struct ahd_softc *, struct scb *);
static __inline uint8_t *
ahd_get_sense_buf(struct ahd_softc *, struct scb *);
static __inline uint32_t
ahd_get_sense_bufaddr(struct ahd_softc *, struct scb *);
static __inline void ahd_post_scb(struct ahd_softc *, struct scb *);
static __inline void
ahd_post_scb(struct ahd_softc *ahd, struct scb *scb)
{
uint32_t sgptr;
sgptr = ahd_le32toh(scb->hscb->sgptr);
if ((sgptr & SG_STATUS_VALID) != 0)
ahd_handle_scb_status(ahd, scb);
2005-02-27 03:26:58 +03:00
else
ahd_done(ahd, scb);
}
static __inline void
ahd_complete_scb(struct ahd_softc *ahd, struct scb *scb)
{
uint32_t sgptr;
sgptr = ahd_le32toh(scb->hscb->sgptr);
if ((sgptr & SG_STATUS_VALID) != 0)
ahd_handle_scb_status(ahd, scb);
2005-02-27 03:26:58 +03:00
else
ahd_done(ahd, scb);
}
/*
* Determine whether the sequencer reported a residual
* for this SCB/transaction.
*/
static __inline void
ahd_update_residual(struct ahd_softc *ahd, struct scb *scb)
{
uint32_t sgptr;
sgptr = ahd_le32toh(scb->hscb->sgptr);
if ((sgptr & SG_STATUS_VALID) != 0)
ahd_calc_residual(ahd, scb);
}
/*
* Return pointers to the transfer negotiation information
* for the specified our_id/remote_id pair.
*/
static __inline struct ahd_initiator_tinfo *
ahd_fetch_transinfo(struct ahd_softc *ahd, char channel, u_int our_id,
u_int remote_id, struct ahd_tmode_tstate **tstate)
{
/*
* Transfer data structures are stored from the perspective
* of the target role. Since the parameters for a connection
* in the initiator role to a given target are the same as
* when the roles are reversed, we pretend we are the target.
*/
if (channel == 'B')
our_id += 8;
*tstate = ahd->enabled_targets[our_id];
return (&(*tstate)->transinfo[remote_id]);
}
#define AHD_COPY_COL_IDX(dst, src) \
do { \
dst->hscb->scsiid = src->hscb->scsiid; \
dst->hscb->lun = src->hscb->lun; \
} while (0)
static __inline uint16_t
ahd_inw(struct ahd_softc *ahd, u_int port)
{
return ((ahd_inb(ahd, port+1) << 8) | ahd_inb(ahd, port));
}
static __inline void
ahd_outw(struct ahd_softc *ahd, u_int port, u_int value)
{
ahd_outb(ahd, port, value & 0xFF);
ahd_outb(ahd, port+1, (value >> 8) & 0xFF);
}
static __inline uint32_t
ahd_inl(struct ahd_softc *ahd, u_int port)
{
return ((ahd_inb(ahd, port))
| (ahd_inb(ahd, port+1) << 8)
| (ahd_inb(ahd, port+2) << 16)
| (ahd_inb(ahd, port+3) << 24));
}
static __inline void
ahd_outl(struct ahd_softc *ahd, u_int port, uint32_t value)
{
ahd_outb(ahd, port, (value) & 0xFF);
ahd_outb(ahd, port+1, ((value) >> 8) & 0xFF);
ahd_outb(ahd, port+2, ((value) >> 16) & 0xFF);
ahd_outb(ahd, port+3, ((value) >> 24) & 0xFF);
}
static __inline uint64_t
ahd_inq(struct ahd_softc *ahd, u_int port)
{
return ((ahd_inb(ahd, port))
| (ahd_inb(ahd, port+1) << 8)
| (ahd_inb(ahd, port+2) << 16)
| (ahd_inb(ahd, port+3) << 24)
| (((uint64_t)ahd_inb(ahd, port+4)) << 32)
| (((uint64_t)ahd_inb(ahd, port+5)) << 40)
| (((uint64_t)ahd_inb(ahd, port+6)) << 48)
| (((uint64_t)ahd_inb(ahd, port+7)) << 56));
}
static __inline void
ahd_outq(struct ahd_softc *ahd, u_int port, uint64_t value)
{
ahd_outb(ahd, port, value & 0xFF);
ahd_outb(ahd, port+1, (value >> 8) & 0xFF);
ahd_outb(ahd, port+2, (value >> 16) & 0xFF);
ahd_outb(ahd, port+3, (value >> 24) & 0xFF);
ahd_outb(ahd, port+4, (value >> 32) & 0xFF);
ahd_outb(ahd, port+5, (value >> 40) & 0xFF);
ahd_outb(ahd, port+6, (value >> 48) & 0xFF);
ahd_outb(ahd, port+7, (value >> 56) & 0xFF);
}
static __inline u_int
ahd_get_scbptr(struct ahd_softc *ahd)
{
AHD_ASSERT_MODES(ahd, ~(AHD_MODE_UNKNOWN_MSK|AHD_MODE_CFG_MSK),
~(AHD_MODE_UNKNOWN_MSK|AHD_MODE_CFG_MSK));
return (ahd_inb(ahd, SCBPTR) | (ahd_inb(ahd, SCBPTR + 1) << 8));
}
static __inline void
ahd_set_scbptr(struct ahd_softc *ahd, u_int scbptr)
{
AHD_ASSERT_MODES(ahd, ~(AHD_MODE_UNKNOWN_MSK|AHD_MODE_CFG_MSK),
~(AHD_MODE_UNKNOWN_MSK|AHD_MODE_CFG_MSK));
ahd_outb(ahd, SCBPTR, scbptr & 0xFF);
ahd_outb(ahd, SCBPTR+1, (scbptr >> 8) & 0xFF);
}
static __inline u_int
ahd_get_hnscb_qoff(struct ahd_softc *ahd)
{
return (ahd_inw_atomic(ahd, HNSCB_QOFF));
}
static __inline void
ahd_set_hnscb_qoff(struct ahd_softc *ahd, u_int value)
{
ahd_outw_atomic(ahd, HNSCB_QOFF, value);
}
static __inline u_int
ahd_get_hescb_qoff(struct ahd_softc *ahd)
{
return (ahd_inb(ahd, HESCB_QOFF));
}
static __inline void
ahd_set_hescb_qoff(struct ahd_softc *ahd, u_int value)
{
ahd_outb(ahd, HESCB_QOFF, value);
}
static __inline u_int
ahd_get_snscb_qoff(struct ahd_softc *ahd)
{
u_int oldvalue;
AHD_ASSERT_MODES(ahd, AHD_MODE_CCHAN_MSK, AHD_MODE_CCHAN_MSK);
oldvalue = ahd_inw(ahd, SNSCB_QOFF);
ahd_outw(ahd, SNSCB_QOFF, oldvalue);
return (oldvalue);
}
static __inline void
ahd_set_snscb_qoff(struct ahd_softc *ahd, u_int value)
{
AHD_ASSERT_MODES(ahd, AHD_MODE_CCHAN_MSK, AHD_MODE_CCHAN_MSK);
ahd_outw(ahd, SNSCB_QOFF, value);
}
static __inline u_int
ahd_get_sescb_qoff(struct ahd_softc *ahd)
{
AHD_ASSERT_MODES(ahd, AHD_MODE_CCHAN_MSK, AHD_MODE_CCHAN_MSK);
return (ahd_inb(ahd, SESCB_QOFF));
}
static __inline void
ahd_set_sescb_qoff(struct ahd_softc *ahd, u_int value)
{
AHD_ASSERT_MODES(ahd, AHD_MODE_CCHAN_MSK, AHD_MODE_CCHAN_MSK);
ahd_outb(ahd, SESCB_QOFF, value);
}
static __inline u_int
ahd_get_sdscb_qoff(struct ahd_softc *ahd)
{
AHD_ASSERT_MODES(ahd, AHD_MODE_CCHAN_MSK, AHD_MODE_CCHAN_MSK);
return (ahd_inb(ahd, SDSCB_QOFF) | (ahd_inb(ahd, SDSCB_QOFF + 1) << 8));
}
static __inline void
ahd_set_sdscb_qoff(struct ahd_softc *ahd, u_int value)
{
AHD_ASSERT_MODES(ahd, AHD_MODE_CCHAN_MSK, AHD_MODE_CCHAN_MSK);
ahd_outb(ahd, SDSCB_QOFF, value & 0xFF);
ahd_outb(ahd, SDSCB_QOFF+1, (value >> 8) & 0xFF);
}
static __inline u_int
ahd_inb_scbram(struct ahd_softc *ahd, u_int offset)
{
u_int value;
/*
* Workaround PCI-X Rev A. hardware bug.
* After a host read of SCB memory, the chip
* may become confused into thinking prefetch
* was required. This starts the discard timer
* running and can cause an unexpected discard
* timer interrupt. The work around is to read
* a normal register prior to the exhaustion of
* the discard timer. The mode pointer register
* has no side effects and so serves well for
* this purpose.
*
* Razor #528
*/
value = ahd_inb(ahd, offset);
if ((ahd->flags & AHD_PCIX_SCBRAM_RD_BUG) != 0)
ahd_inb(ahd, MODE_PTR);
return (value);
}
static __inline u_int
ahd_inw_scbram(struct ahd_softc *ahd, u_int offset)
{
return (ahd_inb_scbram(ahd, offset)
| (ahd_inb_scbram(ahd, offset+1) << 8));
}
static __inline uint32_t
ahd_inl_scbram(struct ahd_softc *ahd, u_int offset)
{
return (ahd_inw_scbram(ahd, offset)
| (ahd_inw_scbram(ahd, offset+2) << 16));
}
static __inline uint64_t
ahd_inq_scbram(struct ahd_softc *ahd, u_int offset)
{
return (ahd_inl_scbram(ahd, offset)
| ((uint64_t)ahd_inl_scbram(ahd, offset+4)) << 32);
}
static __inline struct scb *
ahd_lookup_scb(struct ahd_softc *ahd, u_int tag)
{
struct scb* scb;
if (tag >= AHD_SCB_MAX)
return (NULL);
scb = ahd->scb_data.scbindex[tag];
if (scb != NULL)
ahd_sync_scb(ahd, scb,
BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
return (scb);
}
static __inline void
ahd_swap_with_next_hscb(struct ahd_softc *ahd, struct scb *scb)
{
struct hardware_scb *q_hscb;
struct map_node *q_hscb_map;
uint32_t saved_hscb_busaddr;
/*
* Our queuing method is a bit tricky. The card
* knows in advance which HSCB (by address) to download,
* and we can't disappoint it. To achieve this, the next
* HSCB to download is saved off in ahd->next_queued_hscb.
* When we are called to queue "an arbitrary scb",
* we copy the contents of the incoming HSCB to the one
* the sequencer knows about, swap HSCB pointers and
* finally assign the SCB to the tag indexed location
* in the scb_array. This makes sure that we can still
* locate the correct SCB by SCB_TAG.
*/
q_hscb = ahd->next_queued_hscb;
q_hscb_map = ahd->next_queued_hscb_map;
saved_hscb_busaddr = q_hscb->hscb_busaddr;
memcpy(q_hscb, scb->hscb, sizeof(*scb->hscb));
q_hscb->hscb_busaddr = saved_hscb_busaddr;
q_hscb->next_hscb_busaddr = scb->hscb->hscb_busaddr;
/* Now swap HSCB pointers. */
ahd->next_queued_hscb = scb->hscb;
ahd->next_queued_hscb_map = scb->hscb_map;
scb->hscb = q_hscb;
scb->hscb_map = q_hscb_map;
KASSERT((vaddr_t)scb->hscb >= (vaddr_t)scb->hscb_map->vaddr &&
(vaddr_t)scb->hscb < (vaddr_t)scb->hscb_map->vaddr + PAGE_SIZE);
/* Now define the mapping from tag to SCB in the scbindex */
ahd->scb_data.scbindex[SCB_GET_TAG(scb)] = scb;
}
/*
* Tell the sequencer about a new transaction to execute.
*/
static __inline void
ahd_queue_scb(struct ahd_softc *ahd, struct scb *scb)
{
ahd_swap_with_next_hscb(ahd, scb);
if (SCBID_IS_NULL(SCB_GET_TAG(scb)))
panic("Attempt to queue invalid SCB tag %x\n",
SCB_GET_TAG(scb));
/*
* Keep a history of SCBs we've downloaded in the qinfifo.
*/
ahd->qinfifo[AHD_QIN_WRAP(ahd->qinfifonext)] = SCB_GET_TAG(scb);
ahd->qinfifonext++;
if (scb->sg_count != 0)
ahd_setup_data_scb(ahd, scb);
else
ahd_setup_noxfer_scb(ahd, scb);
ahd_setup_scb_common(ahd, scb);
/*
* Make sure our data is consistent from the
* perspective of the adapter.
*/
ahd_sync_scb(ahd, scb, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
#ifdef AHD_DEBUG
if ((ahd_debug & AHD_SHOW_QUEUE) != 0) {
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
uint64_t host_dataptr;
host_dataptr = ahd_le64toh(scb->hscb->dataptr);
printf("%s: Queueing SCB 0x%x bus addr 0x%x - 0x%x%x/0x%x\n",
ahd_name(ahd),
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
SCB_GET_TAG(scb), ahd_le32toh(scb->hscb->hscb_busaddr),
(u_int)((host_dataptr >> 32) & 0xFFFFFFFF),
(u_int)(host_dataptr & 0xFFFFFFFF),
ahd_le32toh(scb->hscb->datacnt));
}
#endif
/* Tell the adapter about the newly queued SCB */
ahd_set_hnscb_qoff(ahd, ahd->qinfifonext);
}
static __inline uint8_t *
ahd_get_sense_buf(struct ahd_softc *ahd, struct scb *scb)
{
return (scb->sense_data);
}
static __inline uint32_t
ahd_get_sense_bufaddr(struct ahd_softc *ahd, struct scb *scb)
{
return (scb->sense_busaddr);
}
/************************** Interrupt Processing ******************************/
static __inline void ahd_sync_qoutfifo(struct ahd_softc *, int);
static __inline void ahd_sync_tqinfifo(struct ahd_softc *, int);
static __inline u_int ahd_check_cmdcmpltqueues(struct ahd_softc *);
static __inline int ahd_intr(void *);
static __inline void ahd_minphys(struct buf *);
static __inline void
ahd_sync_qoutfifo(struct ahd_softc *ahd, int op)
{
ahd_dmamap_sync(ahd, ahd->parent_dmat, ahd->shared_data_map.dmamap,
/*offset*/0, /*len*/AHD_SCB_MAX * sizeof(uint16_t), op);
}
static __inline void
ahd_sync_tqinfifo(struct ahd_softc *ahd, int op)
{
#ifdef AHD_TARGET_MODE
if ((ahd->flags & AHD_TARGETROLE) != 0) {
ahd_dmamap_sync(ahd, ahd->parent_dmat /*shared_data_dmat*/,
ahd->shared_data_map.dmamap,
ahd_targetcmd_offset(ahd, 0),
sizeof(struct target_cmd) * AHD_TMODE_CMDS,
op);
}
#endif
}
/*
* See if the firmware has posted any completed commands
* into our in-core command complete fifos.
*/
#define AHD_RUN_QOUTFIFO 0x1
#define AHD_RUN_TQINFIFO 0x2
static __inline u_int
ahd_check_cmdcmpltqueues(struct ahd_softc *ahd)
{
u_int retval;
retval = 0;
ahd_dmamap_sync(ahd, ahd->parent_dmat /*shared_data_dmat*/, ahd->shared_data_map.dmamap,
/*offset*/ahd->qoutfifonext, /*len*/2,
BUS_DMASYNC_POSTREAD);
if ((ahd->qoutfifo[ahd->qoutfifonext]
& QOUTFIFO_ENTRY_VALID_LE) == ahd->qoutfifonext_valid_tag)
retval |= AHD_RUN_QOUTFIFO;
#ifdef AHD_TARGET_MODE
if ((ahd->flags & AHD_TARGETROLE) != 0
&& (ahd->flags & AHD_TQINFIFO_BLOCKED) == 0) {
ahd_dmamap_sync(ahd, ahd->parent_dmat /*shared_data_dmat*/,
ahd->shared_data_map.dmamap,
ahd_targetcmd_offset(ahd, ahd->tqinfifofnext),
/*len*/sizeof(struct target_cmd),
BUS_DMASYNC_POSTREAD);
if (ahd->targetcmds[ahd->tqinfifonext].cmd_valid != 0)
retval |= AHD_RUN_TQINFIFO;
}
#endif
return (retval);
}
/*
* Catch an interrupt from the adapter
*/
static __inline int
ahd_intr(void *arg)
{
struct ahd_softc *ahd = (struct ahd_softc*)arg;
u_int intstat;
if ((ahd->pause & INTEN) == 0) {
/*
* Our interrupt is not enabled on the chip
* and may be disabled for re-entrancy reasons,
* so just return. This is likely just a shared
* interrupt.
*/
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
return (0);
}
/*
* Instead of directly reading the interrupt status register,
* infer the cause of the interrupt by checking our in-core
* completion queues. This avoids a costly PCI bus read in
* most cases.
*/
if ((ahd->flags & AHD_ALL_INTERRUPTS) == 0
&& (ahd_check_cmdcmpltqueues(ahd) != 0))
intstat = CMDCMPLT;
else
intstat = ahd_inb(ahd, INTSTAT);
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
if ((intstat & INT_PEND) == 0)
return (0);
if (intstat & CMDCMPLT) {
ahd_outb(ahd, CLRINT, CLRCMDINT);
/*
* Ensure that the chip sees that we've cleared
* this interrupt before we walk the output fifo.
* Otherwise, we may, due to posted bus writes,
* clear the interrupt after we finish the scan,
* and after the sequencer has added new entries
* and asserted the interrupt again.
*/
if ((ahd->bugs & AHD_INTCOLLISION_BUG) != 0) {
if (ahd_is_paused(ahd)) {
/*
* Potentially lost SEQINT.
* If SEQINTCODE is non-zero,
* simulate the SEQINT.
*/
if (ahd_inb(ahd, SEQINTCODE) != NO_SEQINT)
intstat |= SEQINT;
}
} else {
ahd_flush_device_writes(ahd);
}
scsipi_channel_freeze(&ahd->sc_channel, 1);
ahd_run_qoutfifo(ahd);
scsipi_channel_thaw(&ahd->sc_channel, 1);
ahd->cmdcmplt_counts[ahd->cmdcmplt_bucket]++;
ahd->cmdcmplt_total++;
#ifdef AHD_TARGET_MODE
if ((ahd->flags & AHD_TARGETROLE) != 0)
ahd_run_tqinfifo(ahd, /*paused*/FALSE);
#endif
if (intstat == CMDCMPLT)
return 1;
}
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
/*
* Handle statuses that may invalidate our cached
* copy of INTSTAT separately.
*/
if (intstat == 0xFF && (ahd->features & AHD_REMOVABLE) != 0) {
/* Hot eject. Do nothing */
} else if (intstat & HWERRINT) {
ahd_handle_hwerrint(ahd);
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
} else if ((intstat & (PCIINT|SPLTINT)) != 0) {
ahd->bus_intr(ahd);
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
} else {
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
if ((intstat & SEQINT) != 0)
ahd_handle_seqint(ahd, intstat);
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
if ((intstat & SCSIINT) != 0)
ahd_handle_scsiint(ahd, intstat);
}
Apply the following change checked in 2003/05/04 00:20:07 by gibbs to the FreeBSD ahd driver: Correct spelling errors. Switch to handling bad SCSI status as a sequencer interrupt instead of having the kernel proccess these failures via the completion queue. This is done because: o The old scheme required us to pause the sequencer and clear critical sections for each SCB. It seems that these pause actions, if coincident with a sequencer FIFO interrupt, would result in a FIFO interrupt getting lost or directing to the wrong FIFO. This caused hangs when the driver was stressed under high "queue full" loads. o The completion code assumed that it was always called with the sequencer running. This may not be the case in timeout processing where completions occur manually via ahd_pause_and_flushwork(). o With this scheme, the extra expense of clearing critical sections is avoided since the sequencer will only self pause once all pending selections have cleared and it is not in a critical section. aic79xx.c Add code to handle the new BAD_SCB_STATUS sequencer interrupt code. This just redirects the SCB through the already existing ahd_complete_scb() code path. Remove code in ahd_handle_scsi_status() that paused the sequencer, made sure that no selections where pending, and cleared critical sections. Bad status SCBs are now only processed when all of these conditions are true. aic79xx.reg: Add the BAD_SCB_STATUS sequencer interrupt code. aic79xx.seq: When completing an SCB upload to the host, if we are doing this because the SCB contains non-zero SCSI status, defer completing the SCB until there are no pending selection events. When completing these SCBs, use the new BAD_SCB_STATUS sequencer interrupt. For all other uploaded SCBs (currently only for underruns), the SCB is completed via the normal done queue. Additionally, keep the SCB that is currently being uploaded on the COMPLETE_DMA_SCB list until the dma is completed, not just until the DMA is started. This ensures that the DMA is restarted properly should the host disable the DMA transfer for some reason. In our RevA workaround for Maxtor drives, guard against the host pausing us while trying to pause I/O until the first data-valid REQ by clearing the current snapshot so that we can tell if the transfer has completed prior to us noticing the REQINIT status. In cfg4data_intr, shave off an instruction before getting the data path running by adding an entrypoint to the overrun handler to also increment the FIFO use count. In the overrun handler, be sure to clear our LONGJMP address in both exit paths. Perform a few sequencer optimizations. aic79xx.c: Print the full path from the SCB when a packetized status overrun occurs. Remove references to LONGJMP_SCB which is being removed from firmware usage. Print the new SCB_FIFO_USE_COUNT field in the per-SCB section of ahd_dump_card_state(). The SCB_TAG field is now re-used by the sequencer, so it no longer makes sense to reference this field in the kernel driver. aic79xx.h: Re-arrange fields in the hardware SCB from largest size type to smallest. This makes it easier to move fields without changing field alignment. The hardware scb tag field is now down near the "spare" portion of the SCB to facilitate reuse by the sequencer. aic79xx.reg: Remove LONGJMP_ADDR. Rearrange SCB fields to match aic79xx.h. Add SCB_FIFO_USE_COUNT as the first byte of the SCB_TAG field. aic79xx.seq: Add a per-SCB "Fifos in use count" field and use it to determine when it is safe (all data posted) to deliver status back to the host. The old method involved polling one or both FIFOs to verify that the current task did not have pending data. This makes running down the GSFIFO very cheap, so we will empty the GSFIFO in one idle loop pass in all cases. Use this simplification of the completion process to prune down the data FIFO teardown sequencer for packetized transfers. Much more code is now shared between the data residual and transfer complete cases. Correct some issues in the packetized status handler. It used to be possible to CLRCHN our FIFO before status had fully transferred to the host. We also failed to handle NONPACKREQ phases that could occur should a CRC error occur during transmission of the status data packet. Correct a few big endian issues: aic79xx.c: aic79xx_inline.h: aic79xx_pci.c: aic79xx_osm.c: o Always get the SCB's tag via the SCB_GET_TAG acccessor o Add missing use of byte swapping macros when touching hscb fields. o Don't double swap SEEPROM data when it is printed. Correct a big-endian bug. We cannot assign a o When assigning a 32bit LE variable to a 64bit LE variable, we must be explict about how the words of the 64bit LE variable are initialized. Cast to (uint32_t*) to do this. aic79xx.c: In ahd_clear_critical_section(), hit CRLSCSIINT after restoring the interrupt masks to avoid what appears to be a glitch on SCSIINT. Any real SCSIINT status will be persistent and will immidiately reset SCSIINT. This clear should only get rid of spurious SCSIINTs. This glitch was the cause of the "Unexpected PKT busfree" status that occurred under high queue full loads Call ahd_fini_scbdata() after shutdown so that any ahd_chip_init() routine that might access SCB data will not access free'd memory. Reset the bus on an IOERR since the chip doesn't seem to reset to the new voltage level without this. Change offset calculation for scatter gather maps so that the calculation is correct if an integral multiple of sg lists does not fit in the allocation size. Adjust bus dma tag for data buffers based on 39BIT addressing flag in our softc. Use the QFREEZE count to simplify ahd_pause_and_flushworkd(). We can thus rely on the sequencer eventually clearing ENSELO. In ahd_abort_scbs(), fix a bug that could potentially corrupt sequencer state. The saved SCB was being restored in the SCSI mode instead of the saved mode. It turns out that the SCB did not need to be saved at all as the scbptr is already restored by all subroutines called during this function that modify that register. aic79xx.c: aic79xx.h: aic79xx_pci.c: Add support for parsing the seeprom vital product data. The VPD data are currently unused. aic79xx.h: aic79xx.seq: aic79xx_pci.c: Add a firmware workaround to make the LED blink brighter during packetized operations on the H2A. aic79xx_inline.h: The host does not use timer interrupts, so don't gate our decision on whether or not to unpause the sequencer on whether or not a timer interrupt is pending.
2003-08-29 04:09:59 +04:00
return (1);
}
static __inline void
ahd_minphys(bp)
struct buf *bp;
{
/*
* Even though the card can transfer up to 16megs per command
2003-05-03 22:10:37 +04:00
* we are limited by the number of segments in the DMA segment
* list that we can hold. The worst case is that all pages are
* discontinuous physically, hence the "page per segment" limit
* enforced here.
*/
if (bp->b_bcount > AHD_MAXTRANSFER_SIZE) {
bp->b_bcount = AHD_MAXTRANSFER_SIZE;
}
minphys(bp);
}
static __inline u_int32_t scsi_4btoul(u_int8_t *);
static __inline u_int32_t
scsi_4btoul(u_int8_t *bytes)
{
u_int32_t rv;
rv = (bytes[0] << 24) |
(bytes[1] << 16) |
(bytes[2] << 8) |
bytes[3];
return (rv);
}
#endif /* _AIC79XX_INLINE_H_ */