Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
NetBSD/sys/dev/ic/isp_netbsd.c

992 lines
24 KiB
C
Raw Normal View History

Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/* $NetBSD: isp_netbsd.c,v 1.44 2001/05/25 21:45:55 mjacob Exp $ */
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
/*
* This driver, which is contained in NetBSD in the files:
*
* sys/dev/ic/isp.c
Fix pathnames in comment.
2000-12-23 04:37:57 +03:00
* sys/dev/ic/isp_inline.h
* sys/dev/ic/isp_netbsd.c
* sys/dev/ic/isp_netbsd.h
* sys/dev/ic/isp_target.c
* sys/dev/ic/isp_target.h
* sys/dev/ic/isp_tpublic.h
* sys/dev/ic/ispmbox.h
* sys/dev/ic/ispreg.h
* sys/dev/ic/ispvar.h
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
* sys/microcode/isp/asm_sbus.h
* sys/microcode/isp/asm_1040.h
* sys/microcode/isp/asm_1080.h
* sys/microcode/isp/asm_12160.h
* sys/microcode/isp/asm_2100.h
* sys/microcode/isp/asm_2200.h
* sys/pci/isp_pci.c
* sys/sbus/isp_sbus.c
*
* Is being actively maintained by Matthew Jacob (mjacob@netbsd.org).
* This driver also is shared source with FreeBSD, OpenBSD, Linux, Solaris,
* Linux versions. This tends to be an interesting maintenance problem.
*
* Please coordinate with Matthew Jacob on changes you wish to make here.
*/
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
/*
* Platform (NetBSD) dependent common attachment code for Qlogic adapters.
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
* Matthew Jacob <mjacob@nas.nasa.gov>
*/
/*
* Copyright (C) 1997, 1998, 1999 National Aeronautics & Space Administration
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
* notice, this list of conditions and the following disclaimer.
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote products
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
* derived from this software without specific prior written permission
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
*
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
*/
#include <dev/ic/isp_netbsd.h>
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
#include <sys/scsiio.h>
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
Gah. Bad port from FreeBSD code (thanks Robert Elz) for setting up default timeout for watchdogging commands.
2000-07-07 07:14:53 +04:00
/*
* Set a timeout for the watchdogging of a command.
*
* The dimensional analysis is
*
* milliseconds * (seconds/millisecond) * (ticks/second) = ticks
*
* =
*
* (milliseconds / 1000) * hz = ticks
*
*
* For timeouts less than 1 second, we'll get zero. Because of this, and
* because we want to establish *our* timeout to be longer than what the
* firmware might do, we just add 3 seconds at the back end.
*/
#define _XT(xs) ((((xs)->timeout/1000) * hz) + (3 * hz))
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
static void ispminphys_1020(struct buf *);
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
static void ispminphys(struct buf *);
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
static INLINE void ispcmd(struct ispsoftc *, XS_T *);
static void isprequest(struct scsipi_channel *, scsipi_adapter_req_t, void *);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
static int
ispioctl(struct scsipi_channel *, u_long, caddr_t, int, struct proc *);
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
static void isp_polled_cmd(struct ispsoftc *, XS_T *);
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
static void isp_dog(void *);
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
static void isp_create_fc_worker(void *);
static void isp_fc_worker(void *);
Do SCSI Bus resets in this layer (now). Don't do it for Fibre Channel yet (we get LOOP DOWN events, and we'll hang on that at this time). Add other isp_async cases- ISPASYNC_LOOP_DOWN and ISPASYNC_LOOP_UP. DOWN will cause internal queuing until UP, whereupon a timeout will fire up any pending xfers. It doesn't really keep commands from getting destroyed by loop down events, but at least minimizes the damage. This was much easier to implement with CAM.
1999-02-09 03:42:22 +03:00
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
/*
* Complete attachment of hardware, include subdevices.
*/
void
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
isp_attach(struct ispsoftc *isp)
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
{
isp->isp_state = ISP_RUNSTATE;
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._adapter.adapt_dev = &isp->isp_osinfo._dev;
isp->isp_osinfo._adapter.adapt_nchannels = IS_DUALBUS(isp) ? 2 : 1;
isp->isp_osinfo._adapter.adapt_openings = isp->isp_maxcmds;
/*
* It's not stated whether max_periph is limited by SPI
* tag uage, but let's assume that it is.
*/
isp->isp_osinfo._adapter.adapt_max_periph = min(isp->isp_maxcmds, 255);
isp->isp_osinfo._adapter.adapt_ioctl = ispioctl;
isp->isp_osinfo._adapter.adapt_request = isprequest;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
if (isp->isp_type <= ISP_HA_SCSI_1020A) {
isp->isp_osinfo._adapter.adapt_minphys = ispminphys_1020;
} else {
isp->isp_osinfo._adapter.adapt_minphys = ispminphys;
}
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._chan.chan_adapter = &isp->isp_osinfo._adapter;
isp->isp_osinfo._chan.chan_bustype = &scsi_bustype;
isp->isp_osinfo._chan.chan_channel = 0;
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
/*
* Until the midlayer is fixed to use REPORT LUNS, limit to 8 luns.
*/
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._chan.chan_nluns = min(isp->isp_maxluns, 8);
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
if (IS_FC(isp)) {
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._chan.chan_ntargets = MAX_FC_TARG;
isp->isp_osinfo._chan.chan_id = MAX_FC_TARG;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
isp->isp_osinfo.threadwork = 1;
/*
* Note that isp_create_fc_worker won't get called
* until much much later (after proc0 is created).
*/
kthread_create(isp_create_fc_worker, isp);
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
} else {
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
int bus = 0;
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
sdparam *sdp = isp->isp_param;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._chan.chan_ntargets = MAX_TARGETS;
isp->isp_osinfo._chan.chan_id = sdp->isp_initiator_id;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
isp->isp_osinfo.discovered[0] = 1 << sdp->isp_initiator_id;
Dual LVD (1280) support.
1999-12-16 08:35:42 +03:00
if (IS_DUALBUS(isp)) {
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._chan_b = isp->isp_osinfo._chan;
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
sdp++;
some comments added
1999-10-17 05:23:21 +04:00
isp->isp_osinfo.discovered[1] =
1 << sdp->isp_initiator_id;
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo._chan_b.chan_id = sdp->isp_initiator_id;
isp->isp_osinfo._chan_b.chan_channel = 1;
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
}
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
ISP_LOCK(isp);
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
(void) isp_control(isp, ISPCTL_RESET_BUS, &bus);
Dual LVD (1280) support.
1999-12-16 08:35:42 +03:00
if (IS_DUALBUS(isp)) {
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
bus++;
(void) isp_control(isp, ISPCTL_RESET_BUS, &bus);
}
Remove ispcmd_slow routine- SCBUSACCEL routine will set/clear tag/wide/sync. Make some changes about where some things sit in the softc.
2000-08-09 02:58:30 +04:00
ISP_UNLOCK(isp);
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
}
Do SCSI Bus resets in this layer (now). Don't do it for Fibre Channel yet (we get LOOP DOWN events, and we'll hang on that at this time). Add other isp_async cases- ISPASYNC_LOOP_DOWN and ISPASYNC_LOOP_UP. DOWN will cause internal queuing until UP, whereupon a timeout will fire up any pending xfers. It doesn't really keep commands from getting destroyed by loop down events, but at least minimizes the damage. This was much easier to implement with CAM.
1999-02-09 03:42:22 +03:00
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
/*
* After this point, we'll be doing the new configuration
* schema which allows interrups, so we can do tsleep/wakeup
* for mailbox stuff at that point.
*/
isp->isp_osinfo.no_mbox_ints = 0;
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
/*
* And attach children (if any).
*/
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
config_found((void *)isp, &isp->isp_chanA, scsiprint);
Dual LVD (1280) support.
1999-12-16 08:35:42 +03:00
if (IS_DUALBUS(isp)) {
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
config_found((void *)isp, &isp->isp_chanB, scsiprint);
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
}
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
}
/*
* minphys our xfers
*/
static void
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ispminphys_1020(struct buf *bp)
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
{
if (bp->b_bcount >= (1 << 24)) {
bp->b_bcount = (1 << 24);
}
minphys(bp);
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
static void
ispminphys(struct buf *bp)
{
if (bp->b_bcount >= (1 << 30)) {
bp->b_bcount = (1 << 30);
}
minphys(bp);
}
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
static int
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
ispioctl(struct scsipi_channel *chan, u_long cmd, caddr_t addr, int flag,
struct proc *p)
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
{
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
struct ispsoftc *isp = (void *)chan->chan_adapter->adapt_dev;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
int retval = ENOTTY;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
switch (cmd) {
case SCBUSIORESET:
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ISP_LOCK(isp);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
if (isp_control(isp, ISPCTL_RESET_BUS, &chan->chan_channel))
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
retval = EIO;
else
retval = 0;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ISP_UNLOCK(isp);
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
break;
Add initial implementation of ISP_SDBLEV, ISP_RESETHBA, ISP_FC_RESCAN, ISP_FC_LIP and ISP_FC_GETDINFO ioctls.
2001-04-11 01:52:00 +04:00
case ISP_SDBLEV:
{
int olddblev = isp->isp_dblev;
isp->isp_dblev = *(int *)addr;
*(int *)addr = olddblev;
retval = 0;
break;
}
case ISP_RESETHBA:
ISP_LOCK(isp);
isp_reinit(isp);
ISP_UNLOCK(isp);
retval = 0;
break;
case ISP_FC_RESCAN:
if (IS_FC(isp)) {
ISP_LOCK(isp);
if (isp_fc_runstate(isp, 5 * 1000000)) {
retval = EIO;
} else {
retval = 0;
}
ISP_UNLOCK(isp);
}
break;
case ISP_FC_LIP:
if (IS_FC(isp)) {
ISP_LOCK(isp);
if (isp_control(isp, ISPCTL_SEND_LIP, 0)) {
retval = EIO;
} else {
retval = 0;
}
ISP_UNLOCK(isp);
}
break;
case ISP_FC_GETDINFO:
{
struct isp_fc_device *ifc = (struct isp_fc_device *) addr;
struct lportdb *lp;
if (ifc->loopid < 0 || ifc->loopid >= MAX_FC_TARG) {
retval = EINVAL;
break;
}
ISP_LOCK(isp);
lp = &FCPARAM(isp)->portdb[ifc->loopid];
if (lp->valid) {
ifc->loopid = lp->loopid;
ifc->portid = lp->portid;
ifc->node_wwn = lp->node_wwn;
ifc->port_wwn = lp->port_wwn;
retval = 0;
} else {
retval = ENODEV;
}
ISP_UNLOCK(isp);
break;
}
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
default:
break;
}
return (retval);
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
static INLINE void
ispcmd(struct ispsoftc *isp, XS_T *xs)
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
{
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ISP_LOCK(isp);
if (isp->isp_state < ISP_RUNSTATE) {
DISABLE_INTS(isp);
isp_init(isp);
if (isp->isp_state != ISP_INITSTATE) {
ENABLE_INTS(isp);
ISP_UNLOCK(isp);
XS_SETERR(xs, HBA_BOTCH);
scsipi_done(xs);
return;
}
isp->isp_state = ISP_RUNSTATE;
ENABLE_INTS(isp);
}
/*
* Handle the case of a FC card where the FC thread hasn't
* fired up yet and we have loop state to clean up. If we
* can't clear things up and we've never seen loop up, bounce
* the command.
*/
if (IS_FC(isp) && isp->isp_osinfo.threadwork &&
isp->isp_osinfo.thread == 0) {
volatile u_int8_t ombi = isp->isp_osinfo.no_mbox_ints;
int delay_time;
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
if (xs->xs_control & XS_CTL_POLL) {
isp->isp_osinfo.no_mbox_ints = 1;
}
if (isp->isp_osinfo.loop_checked == 0) {
delay_time = 10 * 1000000;
isp->isp_osinfo.loop_checked = 1;
} else {
delay_time = 250000;
}
if (isp_fc_runstate(isp, delay_time) != 0) {
if (xs->xs_control & XS_CTL_POLL) {
isp->isp_osinfo.no_mbox_ints = ombi;
}
if (FCPARAM(isp)->loop_seen_once == 0) {
XS_SETERR(xs, HBA_SELTIMEOUT);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
scsipi_done(xs);
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ISP_UNLOCK(isp);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
return;
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/*
* Otherwise, fall thru to be queued up for later.
*/
} else {
int wasblocked =
(isp->isp_osinfo.blocked || isp->isp_osinfo.paused);
isp->isp_osinfo.threadwork = 0;
isp->isp_osinfo.blocked =
isp->isp_osinfo.paused = 0;
isp_prt(isp, ISP_LOGALL, "ispcmd, manual runstate, (freeze count %d)", isp->isp_chanA.chan_qfreeze);
if (wasblocked) {
scsipi_channel_thaw(&isp->isp_chanA, 1);
}
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
}
Update for SCSIPI changes. Note that capabilities update is disabled until Matt Jacob has a change to update the driver for the latest firmware, etc. since update device parameters seems to fail once the system is really up and running, and eventually causes the controller to wedge. This may be due to a firmware bug. Per discussion with Matt.
1999-10-01 03:06:18 +04:00
if (xs->xs_control & XS_CTL_POLL) {
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp->isp_osinfo.no_mbox_ints = ombi;
Do SCSI Bus resets in this layer (now). Don't do it for Fibre Channel yet (we get LOOP DOWN events, and we'll hang on that at this time). Add other isp_async cases- ISPASYNC_LOOP_DOWN and ISPASYNC_LOOP_UP. DOWN will cause internal queuing until UP, whereupon a timeout will fire up any pending xfers. It doesn't really keep commands from getting destroyed by loop down events, but at least minimizes the damage. This was much easier to implement with CAM.
1999-02-09 03:42:22 +03:00
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
if (isp->isp_osinfo.paused) {
isp_prt(isp, ISP_LOGWARN, "I/O while paused");
xs->error = XS_RESOURCE_SHORTAGE;
scsipi_done(xs);
ISP_UNLOCK(isp);
return;
}
if (isp->isp_osinfo.blocked) {
isp_prt(isp, ISP_LOGWARN, "I/O while blocked");
xs->error = XS_REQUEUE;
scsipi_done(xs);
ISP_UNLOCK(isp);
return;
}
if (xs->xs_control & XS_CTL_POLL) {
volatile u_int8_t ombi = isp->isp_osinfo.no_mbox_ints;
isp->isp_osinfo.no_mbox_ints = 1;
isp_polled_cmd(isp, xs);
isp->isp_osinfo.no_mbox_ints = ombi;
ISP_UNLOCK(isp);
return;
}
switch (isp_start(xs)) {
case CMD_QUEUED:
if (xs->timeout) {
callout_reset(&xs->xs_callout, _XT(xs), isp_dog, xs);
}
break;
case CMD_EAGAIN:
isp->isp_osinfo.paused = 1;
xs->error = XS_RESOURCE_SHORTAGE;
scsipi_channel_freeze(&isp->isp_chanA, 1);
if (IS_DUALBUS(isp)) {
scsipi_channel_freeze(&isp->isp_chanB, 1);
}
scsipi_done(xs);
break;
case CMD_RQLATER:
/*
* We can only get RQLATER from FC devices (1 channel only)
*
* Also, if we've never seen loop up, bounce the command
* (somebody has booted with no FC cable connected)
*/
if (FCPARAM(isp)->loop_seen_once == 0) {
XS_SETERR(xs, HBA_SELTIMEOUT);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
scsipi_done(xs);
break;
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
if (isp->isp_osinfo.blocked == 0) {
isp->isp_osinfo.blocked = 1;
scsipi_channel_freeze(&isp->isp_chanA, 1);
isp_prt(isp, ISP_LOGALL, "ispcmd, RQLATER, (freeze count %d)", isp->isp_chanA.chan_qfreeze);
}
xs->error = XS_REQUEUE;
scsipi_done(xs);
break;
case CMD_COMPLETE:
scsipi_done(xs);
break;
}
ISP_UNLOCK(isp);
}
static void
isprequest(struct scsipi_channel *chan, scsipi_adapter_req_t req, void *arg)
{
struct ispsoftc *isp = (void *)chan->chan_adapter->adapt_dev;
switch (req) {
case ADAPTER_REQ_RUN_XFER:
ispcmd(isp, (XS_T *) arg);
break;
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
case ADAPTER_REQ_GROW_RESOURCES:
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/* Not supported. */
break;
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
case ADAPTER_REQ_SET_XFER_MODE:
if (IS_SCSI(isp)) {
struct scsipi_xfer_mode *xm = arg;
int dflags = 0;
sdparam *sdp = SDPARAM(isp);
sdp += chan->chan_channel;
if (xm->xm_mode & PERIPH_CAP_TQING)
dflags |= DPARM_TQING;
if (xm->xm_mode & PERIPH_CAP_WIDE16)
dflags |= DPARM_WIDE;
if (xm->xm_mode & PERIPH_CAP_SYNC)
dflags |= DPARM_SYNC;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ISP_LOCK(isp);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
sdp->isp_devparam[xm->xm_target].dev_flags |= dflags;
dflags = sdp->isp_devparam[xm->xm_target].dev_flags;
sdp->isp_devparam[xm->xm_target].dev_update = 1;
isp->isp_update |= (1 << chan->chan_channel);
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
ISP_UNLOCK(isp);
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
isp_prt(isp, ISP_LOGDEBUG1,
"ispioctl: device flags 0x%x for %d.%d.X",
dflags, chan->chan_channel, xm->xm_target);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
break;
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
}
default:
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
break;
}
}
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
static void
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
isp_polled_cmd(struct ispsoftc *isp, XS_T *xs)
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
{
int result;
int infinite = 0, mswait;
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
result = isp_start(xs);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
switch (result) {
case CMD_QUEUED:
break;
case CMD_RQLATER:
if (XS_NOERR(xs)) {
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
xs->error = XS_REQUEUE;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
case CMD_EAGAIN:
if (XS_NOERR(xs)) {
xs->error = XS_RESOURCE_SHORTAGE;
}
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
/* FALLTHROUGH */
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
case CMD_COMPLETE:
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
scsipi_done(xs);
return;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
/*
* If we can't use interrupts, poll on completion.
*/
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
if ((mswait = XS_TIME(xs)) == 0)
infinite = 1;
while (mswait || infinite) {
if (isp_intr((void *)isp)) {
if (XS_CMD_DONE_P(xs)) {
break;
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
}
}
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
USEC_DELAY(1000);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
mswait -= 1;
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
/*
* If no other error occurred but we didn't finish,
* something bad happened.
*/
if (XS_CMD_DONE_P(xs) == 0) {
if (isp_control(isp, ISPCTL_ABORT_CMD, xs)) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_reinit(isp);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
}
if (XS_NOERR(xs)) {
XS_SETERR(xs, HBA_BOTCH);
}
}
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
scsipi_done(xs);
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
void
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
isp_done(XS_T *xs)
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
{
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
XS_CMD_S_DONE(xs);
if (XS_CMD_WDOG_P(xs) == 0) {
struct ispsoftc *isp = XS_ISP(xs);
callout_stop(&xs->xs_callout);
if (XS_CMD_GRACE_P(xs)) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGDEBUG1,
"finished command on borrowed time");
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
XS_CMD_S_CLEAR(xs);
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/*
* Fixup- if we get a QFULL, we need
* to set XS_BUSY as the error.
*/
if (xs->status == SCSI_QUEUE_FULL) {
xs->error = XS_BUSY;
}
if (isp->isp_osinfo.paused) {
isp->isp_osinfo.paused = 0;
scsipi_channel_timed_thaw(&isp->isp_chanA);
if (IS_DUALBUS(isp)) {
scsipi_channel_timed_thaw(&isp->isp_chanB);
}
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
scsipi_done(xs);
NetBSD OS specific routines and definitions.
1998-07-15 23:44:04 +04:00
}
}
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
static void
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
isp_dog(void *arg)
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
{
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
XS_T *xs = arg;
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
struct ispsoftc *isp = XS_ISP(xs);
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
u_int16_t handle;
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
ISP_ILOCK(isp);
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
/*
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
* We've decided this command is dead. Make sure we're not trying
* to kill a command that's already dead by getting it's handle and
* and seeing whether it's still alive.
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
*/
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
handle = isp_find_handle(isp, xs);
if (handle) {
u_int16_t r, r1, i;
if (XS_CMD_DONE_P(xs)) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGDEBUG1,
"watchdog found done cmd (handle 0x%x)", handle);
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
ISP_IUNLOCK(isp);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
return;
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
if (XS_CMD_WDOG_P(xs)) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGDEBUG1,
"recursive watchdog (handle 0x%x)", handle);
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
ISP_IUNLOCK(isp);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
return;
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
}
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
XS_CMD_S_WDOG(xs);
i = 0;
do {
r = ISP_READ(isp, BIU_ISR);
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
USEC_DELAY(1);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
r1 = ISP_READ(isp, BIU_ISR);
} while (r != r1 && ++i < 1000);
if (INT_PENDING(isp, r) && isp_intr(isp) && XS_CMD_DONE_P(xs)) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGDEBUG1, "watchdog cleanup (%x, %x)",
handle, r);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
XS_CMD_C_WDOG(xs);
isp_done(xs);
} else if (XS_CMD_GRACE_P(xs)) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGDEBUG1, "watchdog timeout (%x, %x)",
handle, r);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
/*
* Make sure the command is *really* dead before we
* release the handle (and DMA resources) for reuse.
*/
(void) isp_control(isp, ISPCTL_ABORT_CMD, arg);
/*
* After this point, the comamnd is really dead.
*/
if (XS_XFRLEN(xs)) {
ISP_DMAFREE(isp, xs, handle);
}
isp_destroy_handle(isp, handle);
XS_SETERR(xs, XS_TIMEOUT);
XS_CMD_S_CLEAR(xs);
isp_done(xs);
} else {
u_int16_t iptr, optr;
ispreq_t *mp;
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGDEBUG2,
"possible command timeout (%x, %x)", handle, r);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
XS_CMD_C_WDOG(xs);
callout_reset(&xs->xs_callout, hz, isp_dog, xs);
if (isp_getrqentry(isp, &iptr, &optr, (void **) &mp)) {
Finally fix this driver to be sensible about the ENDIAN dance. It's not quite simply a question of the Qlogic being little endian and having to have stuff swapped on big endian machines- it also has to do with the fact that the SBus and PCI DMA layouts are wierd with respect to this. At any rate, now finally fixed- works on Mac G4, tested it on a SS10 for sparc, checked on alpha to see if I've broken anything, and as soon as I get another spare afternoon I'll finally install a sparc64 version which should just work (as it'll be like the Mac).
2000-12-09 11:06:31 +03:00
ISP_UNLOCK(isp);
Redo watchdogs to handle cases of false death of commands- time each command now.. Get rid of SCCLUn stuff. Use an isp_done routine to handle cases of watchdog and isp_done racing to completion.
2000-07-06 02:23:05 +04:00
return;
}
XS_CMD_S_GRACE(xs);
MEMZERO((void *) mp, sizeof (*mp));
mp->req_header.rqs_entry_count = 1;
mp->req_header.rqs_entry_type = RQSTYPE_MARKER;
mp->req_modifier = SYNC_ALL;
mp->req_target = XS_CHANNEL(xs) << 7;
ISP_SWIZZLE_REQUEST(isp, mp);
ISP_ADD_REQUEST(isp, iptr);
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
}
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
} else {
isp_prt(isp, ISP_LOGDEBUG0, "watchdog with no command");
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
}
Add a maintenance note. Change all splbio's to use the ISP_LOCK/ISP_UNLOCK or ISP_ILOCK/ISP_IUNLOCK macros.
2000-08-14 11:08:12 +04:00
ISP_IUNLOCK(isp);
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/*
* Fibre Channel state cleanup thread
*/
static void
isp_create_fc_worker(void *arg)
{
struct ispsoftc *isp = arg;
if (kthread_create1(isp_fc_worker, isp, &isp->isp_osinfo.thread,
"%s:fc_thrd", isp->isp_name)) {
isp_prt(isp, ISP_LOGERR, "unable to create FC worker thread");
panic("isp_create_fc_worker");
}
}
static void
isp_fc_worker(void *arg)
{
void scsipi_run_queue(struct scsipi_channel *);
struct ispsoftc *isp = arg;
for (;;) {
int s;
/*
* Note we do *not* use the ISP_LOCK/ISP_UNLOCK macros here.
*/
s = splbio();
while (isp->isp_osinfo.threadwork) {
isp->isp_osinfo.threadwork = 0;
if (isp_fc_runstate(isp, 10 * 1000000) == 0) {
break;
}
isp->isp_osinfo.threadwork = 1;
splx(s);
delay(500 * 1000);
s = splbio();
}
if (FCPARAM(isp)->isp_fwstate != FW_READY ||
FCPARAM(isp)->isp_loopstate != LOOP_READY) {
isp_prt(isp, ISP_LOGINFO, "isp_fc_runstate in vain");
isp->isp_osinfo.threadwork = 1;
splx(s);
continue;
}
if (isp->isp_osinfo.blocked) {
isp->isp_osinfo.blocked = 0;
isp_prt(isp, /* ISP_LOGDEBUG0 */ ISP_LOGALL, "restarting queues (freeze count %d)", isp->isp_chanA.chan_qfreeze);
scsipi_channel_thaw(&isp->isp_chanA, 1);
}
if (isp->isp_osinfo.thread == NULL)
break;
(void) tsleep(&isp->isp_osinfo.thread, PRIBIO, "fcclnup", 0);
splx(s);
}
/* In case parent is waiting for us to exit. */
wakeup(&isp->isp_osinfo.thread);
kthread_exit(0);
}
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
/*
* Free any associated resources prior to decommissioning and
* set the card to a known state (so it doesn't wake up and kick
* us when we aren't expecting it to).
*
* Locks are held before coming here.
*/
void
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
isp_uninit(struct ispsoftc *isp)
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
{
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_lock(isp);
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
/*
* Leave with interrupts disabled.
*/
DISABLE_INTS(isp);
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_unlock(isp);
clean up headers; move uninit/watch to outer layers
1998-12-28 22:10:43 +03:00
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 10:31:50 +03:00
int
Handles are now 16 bits. ANSIfy.
2001-03-14 08:44:21 +03:00
isp_async(struct ispsoftc *isp, ispasync_t cmd, void *arg)
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 10:31:50 +03:00
{
Do a fairly large internal restructuring to accomodate dual-bus host adapters (e.g., the 1240). Include the new 1080/1240 NVRAM layout reading code. Some moderately significant mailbox changes were necessary also to accomodate a second channel.
1999-05-12 22:59:23 +04:00
int bus, tgt;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 10:31:50 +03:00
switch (cmd) {
case ISPASYNC_NEW_TGT_PARAMS:
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
if (IS_SCSI(isp) && isp->isp_dblev) {
sdparam *sdp = isp->isp_param;
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
int flags;
struct scsipi_xfer_mode xm;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
tgt = *((int *) arg);
bus = (tgt >> 16) & 0xffff;
tgt &= 0xffff;
Dual LVD (1280) support.
1999-12-16 08:35:42 +03:00
sdp += bus;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
flags = sdp->isp_devparam[tgt].cur_dflags;
Dual LVD (1280) support.
1999-12-16 08:35:42 +03:00
Pull up the thorpej_scsipi branch to main branch. This is a completely rewritten scsipi_xfer execution engine, and the associated changes to HBA drivers. Overview of changes & features: - All xfers are queued in the mid-layer, rather than doing so in an ad-hoc fashion in individual adapter drivers. - Adapter/channel resource management in the mid-layer, avoids even trying to start running an xfer if the adapter/channel doesn't have the resources. - Better communication between the mid-layer and the adapters. - Asynchronous event notification mechanism from adapter to mid-layer and peripherals. - Better peripheral queue management: freeze/thaw, sorted requeueing during recovery, etc. - Clean separation of peripherals, adapters, and adapter channels (no more scsipi_link). - Kernel thread for each scsipi_channel makes error recovery much easier (no more dealing with interrupt context when recovering from an error). - Mid-layer support for tagged queueing: commands can have the tag type set explicitly, tag IDs are allocated in the mid-layer (thus eliminating the need to use buggy tag ID allocation schemes in many adapter drivers). - support for QUEUE FULL and CHECK CONDITION status in mid-layer; the command will be requeued, or a REQUEST SENSE will be sent as appropriate. Just before the merge syssrc has been tagged with thorpej_scsipi_beforemerge
2001-04-25 21:53:04 +04:00
xm.xm_mode = 0;
xm.xm_period = sdp->isp_devparam[tgt].cur_period;
xm.xm_offset = sdp->isp_devparam[tgt].cur_offset;
xm.xm_target = tgt;
if ((flags & DPARM_SYNC) && xm.xm_period && xm.xm_offset)
xm.xm_mode |= PERIPH_CAP_SYNC;
if (flags & DPARM_WIDE)
xm.xm_mode |= PERIPH_CAP_WIDE16;
if (flags & DPARM_TQING)
xm.xm_mode |= PERIPH_CAP_TQING;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
scsipi_async_event(bus? &isp->isp_chanB : &isp->isp_chanA,
ASYNC_EVENT_XFER_MODE, &xm);
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 10:31:50 +03:00
break;
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
}
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
case ISPASYNC_BUS_RESET:
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
bus = *((int *) arg);
scsipi_async_event(bus? &isp->isp_chanB : &isp->isp_chanA,
ASYNC_EVENT_RESET, NULL);
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGINFO, "SCSI bus %d reset detected", bus);
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
break;
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
case ISPASYNC_LIP:
/*
* Don't do queue freezes or blockage until we have the
* thread running that can unfreeze/unblock us.
*/
if (isp->isp_osinfo.blocked == 0) {
if (isp->isp_osinfo.thread) {
isp->isp_osinfo.blocked = 1;
scsipi_channel_freeze(&isp->isp_chanA, 1);
}
}
isp_prt(isp, ISP_LOGINFO, "LIP Received");
break;
case ISPASYNC_LOOP_RESET:
/*
* Don't do queue freezes or blockage until we have the
* thread running that can unfreeze/unblock us.
*/
if (isp->isp_osinfo.blocked == 0) {
if (isp->isp_osinfo.thread) {
isp->isp_osinfo.blocked = 1;
scsipi_channel_freeze(&isp->isp_chanA, 1);
}
}
isp_prt(isp, ISP_LOGINFO, "Loop Reset Received");
break;
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
case ISPASYNC_LOOP_DOWN:
/*
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
* Don't do queue freezes or blockage until we have the
* thread running that can unfreeze/unblock us.
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
*/
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
if (isp->isp_osinfo.blocked == 0) {
if (isp->isp_osinfo.thread) {
isp->isp_osinfo.blocked = 1;
scsipi_channel_freeze(&isp->isp_chanA, 1);
}
}
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGINFO, "Loop DOWN");
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
break;
case ISPASYNC_LOOP_UP:
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/*
* Let the subsequent ISPASYNC_CHANGE_NOTIFY invoke
* the FC worker thread. When the FC worker thread
* is done, let *it* call scsipi_channel_thaw...
*/
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGINFO, "Loop UP");
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
break;
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
case ISPASYNC_PROMENADE:
Add in an ioctl entry point so scsictl mediated bus resets will work. Redo how we start commands- do a 'slow' start function which then looks to see when we're done the configuration process at which point it *then* enables sync/wide mode. Set the max openings amount to the true max openings- not a synthetic. Add a timeout driven command requeue function so that Loop Down events well freeze things until a later point in time where they might be restarted.
1999-10-14 06:31:11 +04:00
if (IS_FC(isp) && isp->isp_dblev) {
Apply nearly all of Bill Sommerfeld's -Wformat patches. The one change I didn't take is the %llu format- I can't have a common across multiple platform module assume a %ll argument capability- which really pointed out that I shouldn't be trying to *print* something which could long long.
2000-12-29 01:27:46 +03:00
const char fmt[] = "Target %d (Loop 0x%x) Port ID 0x%x "
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
"(role %s) %s\n Port WWN 0x%08x%08x\n Node WWN 0x%08x%08x";
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
const static char *roles[4] = {
It's 'role None', not 'role No'. Also, pay attention to the difference in the ISPASYNC_CHANGE_NOTIFY between loop and fabric changes.
2001-05-16 07:52:10 +04:00
"None", "Target", "Initiator", "Target/Initiator"
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
};
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
fcparam *fcp = isp->isp_param;
int tgt = *((int *) arg);
struct lportdb *lp = &fcp->portdb[tgt];
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGINFO, fmt, tgt, lp->loopid, lp->portid,
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
roles[lp->roles & 0x3],
(lp->valid)? "Arrived" : "Departed",
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
(u_int32_t) (lp->port_wwn >> 32),
(u_int32_t) (lp->port_wwn & 0xffffffffLL),
(u_int32_t) (lp->node_wwn >> 32),
(u_int32_t) (lp->node_wwn & 0xffffffffLL));
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
break;
}
case ISPASYNC_CHANGE_NOTIFY:
It's 'role None', not 'role No'. Also, pay attention to the difference in the ISPASYNC_CHANGE_NOTIFY between loop and fabric changes.
2001-05-16 07:52:10 +04:00
if (arg == ISPASYNC_CHANGE_PDB) {
isp_prt(isp, ISP_LOGINFO, "Port Database Changed");
} else if (arg == ISPASYNC_CHANGE_SNS) {
ISPASYNC_PDB_CHANGED->ISPASYNC_LOGGED_INOUT. And also, ISPASYNC_CHANGE_NOTIFY is for both local loop and fabric cases now.
2001-01-09 21:54:53 +03:00
isp_prt(isp, ISP_LOGINFO,
"Name Server Database Changed");
}
Create a kernel thread for Fibre Channel cards. This is the new way of doing business- modulo some startup spasms and peculiarities of the way kthreads are started (*after* configuration, weird) and some strangeness with the freeze/thaw code, what now happens is that any of Loop Down, LIP, Loop Reset or Port Datbase or Name Server Database Changed ASYNC events cause the queues to freeze for this channel. The arrival of a Loop UP is not relevant. What *is* relevant is that the Port Datbase or Name Server Changed async event indicate that it's okay to go and (re)evaluate the state of the FC link and (re)probe local loop and fabric membership. We have a kthread do this because it's *sooooo* much nicer to be able to sleep while doing the 130-250 mailbox commands it'll take to re-evaluate things. When the state is well known again, we can unfreeze the channel queues. Then, as commands start arriving, we simply can start them or bounce them with XS_SELTIMEOUT (if the device in question has gone away). Previously, we did lazy evaluation, which meant that if a change occurred, we would wait until the very *next* command to go rebuild stuff. The reason this is not sensible is: a) Even with sleeping, you can hang up your system because you might be making some poor stat(2) call pay the price of re-evaluating the whole fabric. b) If we ever really want to get to dynamic attachment/detachment, we should find out sooner, rather than later, where things get to. Split off ispminphys_1020 from ispminphys- a 1020 has a 24 bit limit- not anything newer. Re-enable LIPs and Loop Resets as async events- this allows the outer layer to set policy about them. Roll platform major && minor. Remove bogus waitq (no longer used). Remove callout entry in softc (no longer used). Define some shorthands for channels. Clean up a variety of cruft left over from the thorpej_scsipi changeover.
2001-05-26 01:45:55 +04:00
/*
* We can set blocked here because we know it's now okay
* to try and run isp_fc_runstate (in order to build loop
* state). But we don't try and freeze the midlayer's queue
* if we have no thread that we can wake to later unfreeze
* it.
*/
if (isp->isp_osinfo.blocked == 0) {
isp->isp_osinfo.blocked = 1;
if (isp->isp_osinfo.thread) {
scsipi_channel_freeze(&isp->isp_chanA, 1);
}
}
/*
* Note that we have work for the thread to do, and
* if the thread is here already, wake it up.
*/
isp->isp_osinfo.threadwork++;
if (isp->isp_osinfo.thread) {
wakeup(&isp->isp_osinfo.thread);
} else {
isp_prt(isp, ISP_LOGDEBUG1, "no FC thread yet");
}
Update the driver with some infrastructure for the 1080. Fix an embarrassing clock botch bug. Additional infrastructure for PDB change stuff.
1999-03-17 09:15:47 +03:00
break;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
case ISPASYNC_FABRIC_DEV:
{
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
int target, lrange;
struct lportdb *lp = NULL;
char *pt;
sns_ganrsp_t *resp = (sns_ganrsp_t *) arg;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
u_int32_t portid;
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
u_int64_t wwpn, wwnn;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
fcparam *fcp = isp->isp_param;
portid =
(((u_int32_t) resp->snscb_port_id[0]) << 16) |
(((u_int32_t) resp->snscb_port_id[1]) << 8) |
(((u_int32_t) resp->snscb_port_id[2]));
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
wwpn =
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
(((u_int64_t)resp->snscb_portname[0]) << 56) |
(((u_int64_t)resp->snscb_portname[1]) << 48) |
(((u_int64_t)resp->snscb_portname[2]) << 40) |
(((u_int64_t)resp->snscb_portname[3]) << 32) |
(((u_int64_t)resp->snscb_portname[4]) << 24) |
(((u_int64_t)resp->snscb_portname[5]) << 16) |
(((u_int64_t)resp->snscb_portname[6]) << 8) |
(((u_int64_t)resp->snscb_portname[7]));
Add in commented-out hiwater measurement. Remove egregious older bug which had us refusing to log into fabric devices unless they were NL ports. Whuff.
2000-10-16 09:15:55 +04:00
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
wwnn =
(((u_int64_t)resp->snscb_nodename[0]) << 56) |
(((u_int64_t)resp->snscb_nodename[1]) << 48) |
(((u_int64_t)resp->snscb_nodename[2]) << 40) |
(((u_int64_t)resp->snscb_nodename[3]) << 32) |
(((u_int64_t)resp->snscb_nodename[4]) << 24) |
(((u_int64_t)resp->snscb_nodename[5]) << 16) |
(((u_int64_t)resp->snscb_nodename[6]) << 8) |
(((u_int64_t)resp->snscb_nodename[7]));
if (portid == 0 || wwpn == 0) {
break;
}
Add in commented-out hiwater measurement. Remove egregious older bug which had us refusing to log into fabric devices unless they were NL ports. Whuff.
2000-10-16 09:15:55 +04:00
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
switch (resp->snscb_port_type) {
case 1:
pt = " N_Port";
break;
case 2:
pt = " NL_Port";
break;
case 3:
pt = "F/NL_Port";
break;
case 0x7f:
pt = " Nx_Port";
break;
case 0x81:
pt = " F_port";
break;
case 0x82:
pt = " FL_Port";
break;
case 0x84:
pt = " E_port";
break;
default:
pt = "?";
break;
}
isp_prt(isp, ISP_LOGINFO,
"%s @ 0x%x, Node 0x%08x%08x Port %08x%08x",
pt, portid, ((u_int32_t) (wwnn >> 32)), ((u_int32_t) wwnn),
((u_int32_t) (wwpn >> 32)), ((u_int32_t) wwpn));
/*
* We're only interested in SCSI_FCP types (for now)
*/
if ((resp->snscb_fc4_types[2] & 1) == 0) {
break;
}
if (fcp->isp_topo != TOPO_F_PORT)
lrange = FC_SNS_ID+1;
else
lrange = 0;
/*
* Is it already in our list?
*/
for (target = lrange; target < MAX_FC_TARG; target++) {
if (target >= FL_PORT_ID && target <= FC_SNS_ID) {
continue;
}
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
lp = &fcp->portdb[target];
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
if (lp->port_wwn == wwpn && lp->node_wwn == wwnn) {
lp->fabric_dev = 1;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
break;
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
}
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
}
if (target < MAX_FC_TARG) {
break;
}
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
for (target = lrange; target < MAX_FC_TARG; target++) {
if (target >= FL_PORT_ID && target <= FC_SNS_ID) {
continue;
}
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
lp = &fcp->portdb[target];
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
if (lp->port_wwn == 0) {
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
break;
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
}
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
}
if (target == MAX_FC_TARG) {
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
isp_prt(isp, ISP_LOGWARN,
"no more space for fabric devices");
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
break;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
}
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
lp->node_wwn = wwnn;
lp->port_wwn = wwpn;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
lp->portid = portid;
Do some cleanup based upon adapter role- mainly not enabling interrupts if we're ISP_ROLE_NONE. Change ISPASYNC_LOGGED_INOUT to ISPASYNC_PROMENADE. Make sure we note if something is a fabric device.
2001-02-13 02:30:12 +03:00
lp->fabric_dev = 1;
Change to approved NASA/Ames copyright. Add Fabric support. Fix SCCLUN support. Add code that tries to track LoopID shifting.
1999-07-06 00:31:35 +04:00
break;
}
Implement and use Fast Posting for both parallel && fibre. Redo a bit of the startup code. Implement a call to outer framework function so that asynchronous events can be handled (e.g., speed negotiation, target mode). Roll internal release tags.
1999-01-30 10:31:50 +03:00
default:
break;
}
return (0);
}
Core version 2.0 (platform version 1.0) rewrite of ISP driver. Some interace cleanups, some new common functions. The major impact that will be noticeable right away is that if you boot with not Fibre connected to the FC cards, you no longer hang indefinitely.
2000-08-02 03:55:09 +04:00
#include <machine/stdarg.h>
void
isp_prt(struct ispsoftc *isp, int level, const char *fmt, ...)
{
va_list ap;
if (level != ISP_LOGALL && (level & isp->isp_dblev) == 0) {
return;
}
printf("%s: ", isp->isp_name);
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
printf("\n");
}