2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/string.h>
21 #include <linux/slab.h>
22 #include <linux/kernel.h>
23 #include <linux/interrupt.h>
24 #include <linux/blkdev.h>
25 #include <linux/delay.h>
27 #include <scsi/scsi.h>
28 #include <scsi/scsi_dbg.h>
29 #include <scsi/scsi_device.h>
30 #include <scsi/scsi_eh.h>
31 #include <scsi/scsi_host.h>
32 #include <scsi/scsi_ioctl.h>
33 #include <scsi/scsi_request.h>
35 #include "scsi_priv.h"
36 #include "scsi_logging.h"
38 #define SENSE_TIMEOUT (10*HZ)
39 #define START_UNIT_TIMEOUT (30*HZ)
42 * These should *probably* be handled by the host itself.
43 * Since it is allowed to sleep, it probably should.
45 #define BUS_RESET_SETTLE_TIME (10)
46 #define HOST_RESET_SETTLE_TIME (10)
48 /* called with shost->host_lock held */
49 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
51 if (shost
->host_busy
== shost
->host_failed
) {
53 SCSI_LOG_ERROR_RECOVERY(5,
54 printk("Waking error handler thread\n"));
59 * scsi_eh_scmd_add - add scsi cmd to error handling.
60 * @scmd: scmd to run eh on.
61 * @eh_flag: optional SCSI_EH flag.
66 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
68 struct Scsi_Host
*shost
= scmd
->device
->host
;
71 if (shost
->eh_wait
== NULL
)
74 spin_lock_irqsave(shost
->host_lock
, flags
);
76 scsi_eh_eflags_set(scmd
, eh_flag
);
78 * FIXME: Can we stop setting owner and state.
80 scmd
->state
= SCSI_STATE_FAILED
;
81 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
82 set_bit(SHOST_RECOVERY
, &shost
->shost_state
);
84 scsi_eh_wakeup(shost
);
85 spin_unlock_irqrestore(shost
->host_lock
, flags
);
90 * scsi_add_timer - Start timeout timer for a single scsi command.
91 * @scmd: scsi command that is about to start running.
92 * @timeout: amount of time to allow this command to run.
93 * @complete: timeout function to call if timer isn't canceled.
96 * This should be turned into an inline function. Each scsi command
97 * has its own timer, and as it is added to the queue, we set up the
98 * timer. When the command completes, we cancel the timer.
100 void scsi_add_timer(struct scsi_cmnd
*scmd
, int timeout
,
101 void (*complete
)(struct scsi_cmnd
*))
105 * If the clock was already running for this command, then
106 * first delete the timer. The timer handling code gets rather
107 * confused if we don't do this.
109 if (scmd
->eh_timeout
.function
)
110 del_timer(&scmd
->eh_timeout
);
112 scmd
->eh_timeout
.data
= (unsigned long)scmd
;
113 scmd
->eh_timeout
.expires
= jiffies
+ timeout
;
114 scmd
->eh_timeout
.function
= (void (*)(unsigned long)) complete
;
116 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
117 " %d, (%p)\n", __FUNCTION__
,
118 scmd
, timeout
, complete
));
120 add_timer(&scmd
->eh_timeout
);
122 EXPORT_SYMBOL(scsi_add_timer
);
125 * scsi_delete_timer - Delete/cancel timer for a given function.
126 * @scmd: Cmd that we are canceling timer for
129 * This should be turned into an inline function.
132 * 1 if we were able to detach the timer. 0 if we blew it, and the
133 * timer function has already started to run.
135 int scsi_delete_timer(struct scsi_cmnd
*scmd
)
139 rtn
= del_timer(&scmd
->eh_timeout
);
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
142 " rtn: %d\n", __FUNCTION__
,
145 scmd
->eh_timeout
.data
= (unsigned long)NULL
;
146 scmd
->eh_timeout
.function
= NULL
;
150 EXPORT_SYMBOL(scsi_delete_timer
);
153 * scsi_times_out - Timeout function for normal scsi commands.
154 * @scmd: Cmd that is timing out.
157 * We do not need to lock this. There is the potential for a race
158 * only in that the normal completion handling might run, but if the
159 * normal completion function determines that the timer has already
160 * fired, then it mustn't do anything.
162 void scsi_times_out(struct scsi_cmnd
*scmd
)
164 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
166 if (scmd
->device
->host
->hostt
->eh_timed_out
)
167 switch (scmd
->device
->host
->hostt
->eh_timed_out(scmd
)) {
172 /* This allows a single retry even of a command
173 * with allowed == 0 */
174 if (scmd
->retries
++ > scmd
->allowed
)
176 scsi_add_timer(scmd
, scmd
->timeout_per_command
,
183 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
184 panic("Error handler thread not present at %p %p %s %d",
185 scmd
, scmd
->device
->host
, __FILE__
, __LINE__
);
190 * scsi_block_when_processing_errors - Prevent cmds from being queued.
191 * @sdev: Device on which we are performing recovery.
194 * We block until the host is out of error recovery, and then check to
195 * see whether the host or the device is offline.
198 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
200 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
204 wait_event(sdev
->host
->host_wait
, (!test_bit(SHOST_RECOVERY
, &sdev
->host
->shost_state
)));
206 online
= scsi_device_online(sdev
);
208 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__
,
213 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
215 #ifdef CONFIG_SCSI_LOGGING
217 * scsi_eh_prt_fail_stats - Log info on failures.
218 * @shost: scsi host being recovered.
219 * @work_q: Queue of scsi cmds to process.
221 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
222 struct list_head
*work_q
)
224 struct scsi_cmnd
*scmd
;
225 struct scsi_device
*sdev
;
226 int total_failures
= 0;
229 int devices_failed
= 0;
231 shost_for_each_device(sdev
, shost
) {
232 list_for_each_entry(scmd
, work_q
, eh_entry
) {
233 if (scmd
->device
== sdev
) {
235 if (scsi_eh_eflags_chk(scmd
,
243 if (cmd_cancel
|| cmd_failed
) {
244 SCSI_LOG_ERROR_RECOVERY(3,
245 printk("%s: %d:%d:%d:%d cmds failed: %d,"
247 __FUNCTION__
, shost
->host_no
,
248 sdev
->channel
, sdev
->id
, sdev
->lun
,
249 cmd_failed
, cmd_cancel
));
256 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
257 " devices require eh work\n",
258 total_failures
, devices_failed
));
263 * scsi_check_sense - Examine scsi cmd sense
264 * @scmd: Cmd to have sense checked.
267 * SUCCESS or FAILED or NEEDS_RETRY
270 * When a deferred error is detected the current command has
271 * not been executed and needs retrying.
273 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
275 struct scsi_sense_hdr sshdr
;
277 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
278 return FAILED
; /* no valid sense data */
280 if (scsi_sense_is_deferred(&sshdr
))
284 * Previous logic looked for FILEMARK, EOM or ILI which are
285 * mainly associated with tapes and returned SUCCESS.
287 if (sshdr
.response_code
== 0x70) {
289 if (scmd
->sense_buffer
[2] & 0xe0)
293 * descriptor format: look for "stream commands sense data
294 * descriptor" (see SSC-3). Assume single sense data
295 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
297 if ((sshdr
.additional_length
> 3) &&
298 (scmd
->sense_buffer
[8] == 0x4) &&
299 (scmd
->sense_buffer
[11] & 0xe0))
303 switch (sshdr
.sense_key
) {
306 case RECOVERED_ERROR
:
307 return /* soft_error */ SUCCESS
;
309 case ABORTED_COMMAND
:
314 * if we are expecting a cc/ua because of a bus reset that we
315 * performed, treat this just as a retry. otherwise this is
316 * information that we should pass up to the upper-level driver
317 * so that we can deal with it there.
319 if (scmd
->device
->expecting_cc_ua
) {
320 scmd
->device
->expecting_cc_ua
= 0;
324 * if the device is in the process of becoming ready, we
327 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
330 * if the device is not started, we need to wake
331 * the error handler to start the motor
333 if (scmd
->device
->allow_restart
&&
334 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
338 /* these three are not supported */
340 case VOLUME_OVERFLOW
:
348 if (scmd
->device
->retry_hwerror
)
353 case ILLEGAL_REQUEST
:
362 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
363 * @scmd: SCSI cmd to examine.
366 * This is *only* called when we are examining the status of commands
367 * queued during error recovery. the main difference here is that we
368 * don't allow for the possibility of retries here, and we are a lot
369 * more restrictive about what we consider acceptable.
371 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
374 * first check the host byte, to see if there is anything in there
375 * that would indicate what we need to do.
377 if (host_byte(scmd
->result
) == DID_RESET
) {
379 * rats. we are already in the error handler, so we now
380 * get to try and figure out what to do next. if the sense
381 * is valid, we have a pretty good idea of what to do.
382 * if not, we mark it as FAILED.
384 return scsi_check_sense(scmd
);
386 if (host_byte(scmd
->result
) != DID_OK
)
390 * next, check the message byte.
392 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
396 * now, check the status byte to see if this indicates
399 switch (status_byte(scmd
->result
)) {
401 case COMMAND_TERMINATED
:
403 case CHECK_CONDITION
:
404 return scsi_check_sense(scmd
);
406 case INTERMEDIATE_GOOD
:
407 case INTERMEDIATE_C_GOOD
:
409 * who knows? FIXME(eric)
414 case RESERVATION_CONFLICT
:
422 * scsi_eh_times_out - timeout function for error handling.
423 * @scmd: Cmd that is timing out.
426 * During error handling, the kernel thread will be sleeping waiting
427 * for some action to complete on the device. our only job is to
428 * record that it timed out, and to wake up the thread.
430 static void scsi_eh_times_out(struct scsi_cmnd
*scmd
)
432 scsi_eh_eflags_set(scmd
, SCSI_EH_REC_TIMEOUT
);
433 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__
,
436 up(scmd
->device
->host
->eh_action
);
440 * scsi_eh_done - Completion function for error handling.
441 * @scmd: Cmd that is done.
443 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
446 * if the timeout handler is already running, then just set the
447 * flag which says we finished late, and return. we have no
448 * way of stopping the timeout handler from running, so we must
449 * always defer to it.
451 if (del_timer(&scmd
->eh_timeout
)) {
452 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
454 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
455 __FUNCTION__
, scmd
, scmd
->result
));
457 up(scmd
->device
->host
->eh_action
);
462 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
463 * @scmd: SCSI Cmd to send.
464 * @timeout: Timeout for cmd.
467 * The initialization of the structures is quite a bit different in
468 * this case, and furthermore, there is a different completion handler
469 * vs scsi_dispatch_cmd.
471 * SUCCESS or FAILED or NEEDS_RETRY
473 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, int timeout
)
475 struct scsi_device
*sdev
= scmd
->device
;
476 struct Scsi_Host
*shost
= sdev
->host
;
477 DECLARE_MUTEX_LOCKED(sem
);
482 * we will use a queued command if possible, otherwise we will
483 * emulate the queuing and calling of completion function ourselves.
485 if (sdev
->scsi_level
<= SCSI_2
)
486 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
487 (sdev
->lun
<< 5 & 0xe0);
489 scsi_add_timer(scmd
, timeout
, scsi_eh_times_out
);
492 * set up the semaphore so we wait for the command to complete.
494 shost
->eh_action
= &sem
;
495 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
497 spin_lock_irqsave(shost
->host_lock
, flags
);
499 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
500 spin_unlock_irqrestore(shost
->host_lock
, flags
);
503 scsi_log_completion(scmd
, SUCCESS
);
505 shost
->eh_action
= NULL
;
508 * see if timeout. if so, tell the host to forget about it.
509 * in other words, we don't want a callback any more.
511 if (scsi_eh_eflags_chk(scmd
, SCSI_EH_REC_TIMEOUT
)) {
512 scsi_eh_eflags_clr(scmd
, SCSI_EH_REC_TIMEOUT
);
515 * as far as the low level driver is
516 * concerned, this command is still active, so
517 * we must give the low level driver a chance
520 * FIXME(eric) - we are not tracking whether we could
521 * abort a timed out command or not. not sure how
522 * we should treat them differently anyways.
524 if (shost
->hostt
->eh_abort_handler
)
525 shost
->hostt
->eh_abort_handler(scmd
);
527 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
531 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
532 __FUNCTION__
, scmd
, rtn
));
535 * now examine the actual status codes to see whether the command
536 * actually did complete normally.
538 if (rtn
== SUCCESS
) {
539 rtn
= scsi_eh_completed_normally(scmd
);
540 SCSI_LOG_ERROR_RECOVERY(3,
541 printk("%s: scsi_eh_completed_normally %x\n",
558 * scsi_request_sense - Request sense data from a particular target.
559 * @scmd: SCSI cmd for request sense.
562 * Some hosts automatically obtain this information, others require
563 * that we obtain it on our own. This function will *not* return until
564 * the command either times out, or it completes.
566 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
568 static unsigned char generic_sense
[6] =
569 {REQUEST_SENSE
, 0, 0, 0, 252, 0};
570 unsigned char *scsi_result
;
574 memcpy(scmd
->cmnd
, generic_sense
, sizeof(generic_sense
));
576 scsi_result
= kmalloc(252, GFP_ATOMIC
| ((scmd
->device
->host
->hostt
->unchecked_isa_dma
) ? __GFP_DMA
: 0));
579 if (unlikely(!scsi_result
)) {
580 printk(KERN_ERR
"%s: cannot allocate scsi_result.\n",
586 * zero the sense buffer. some host adapters automatically always
587 * request sense, so it is not a good idea that
588 * scmd->request_buffer and scmd->sense_buffer point to the same
589 * address (db). 0 is not a valid sense code.
591 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
592 memset(scsi_result
, 0, 252);
594 saved_result
= scmd
->result
;
595 scmd
->request_buffer
= scsi_result
;
596 scmd
->request_bufflen
= 252;
598 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
599 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
602 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
604 /* last chance to have valid sense data */
605 if(!SCSI_SENSE_VALID(scmd
)) {
606 memcpy(scmd
->sense_buffer
, scmd
->request_buffer
,
607 sizeof(scmd
->sense_buffer
));
613 * when we eventually call scsi_finish, we really wish to complete
614 * the original request, so let's restore the original data. (db)
616 scsi_setup_cmd_retry(scmd
);
617 scmd
->result
= saved_result
;
622 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
623 * @scmd: Original SCSI cmd that eh has finished.
624 * @done_q: Queue for processed commands.
627 * We don't want to use the normal command completion while we are are
628 * still handling errors - it may cause other commands to be queued,
629 * and that would disturb what we are doing. thus we really want to
630 * keep a list of pending commands for final completion, and once we
631 * are ready to leave error handling we handle completion for real.
633 static void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
,
634 struct list_head
*done_q
)
636 scmd
->device
->host
->host_failed
--;
637 scmd
->state
= SCSI_STATE_BHQUEUE
;
639 scsi_eh_eflags_clr_all(scmd
);
642 * set this back so that the upper level can correctly free up
645 scsi_setup_cmd_retry(scmd
);
646 list_move_tail(&scmd
->eh_entry
, done_q
);
650 * scsi_eh_get_sense - Get device sense data.
651 * @work_q: Queue of commands to process.
652 * @done_q: Queue of proccessed commands..
655 * See if we need to request sense information. if so, then get it
656 * now, so we have a better idea of what to do.
659 * This has the unfortunate side effect that if a shost adapter does
660 * not automatically request sense information, that we end up shutting
661 * it down before we request it.
663 * All drivers should request sense information internally these days,
664 * so for now all I have to say is tough noogies if you end up in here.
666 * XXX: Long term this code should go away, but that needs an audit of
669 static int scsi_eh_get_sense(struct list_head
*work_q
,
670 struct list_head
*done_q
)
672 struct list_head
*lh
, *lh_sf
;
673 struct scsi_cmnd
*scmd
;
676 list_for_each_safe(lh
, lh_sf
, work_q
) {
677 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
678 if (scsi_eh_eflags_chk(scmd
, SCSI_EH_CANCEL_CMD
) ||
679 SCSI_SENSE_VALID(scmd
))
682 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
686 rtn
= scsi_request_sense(scmd
);
690 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
691 " result %x\n", scmd
,
693 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
695 rtn
= scsi_decide_disposition(scmd
);
698 * if the result was normal, then just pass it along to the
702 /* we don't want this command reissued, just
703 * finished with the sense data, so set
704 * retries to the max allowed to ensure it
705 * won't get reissued */
706 scmd
->retries
= scmd
->allowed
;
707 else if (rtn
!= NEEDS_RETRY
)
710 scsi_eh_finish_cmd(scmd
, done_q
);
713 return list_empty(work_q
);
717 * scsi_try_to_abort_cmd - Ask host to abort a running command.
718 * @scmd: SCSI cmd to abort from Lower Level.
721 * This function will not return until the user's completion function
722 * has been called. there is no timeout on this operation. if the
723 * author of the low-level driver wishes this operation to be timed,
724 * they can provide this facility themselves. helper functions in
725 * scsi_error.c can be supplied to make this easier to do.
727 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
729 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
733 * scsi_done was called just after the command timed out and before
734 * we had a chance to process it. (db)
736 if (scmd
->serial_number
== 0)
738 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
742 * scsi_eh_tur - Send TUR to device.
743 * @scmd: Scsi cmd to send TUR
746 * 0 - Device is ready. 1 - Device NOT ready.
748 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
750 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
751 int retry_cnt
= 1, rtn
;
755 memcpy(scmd
->cmnd
, tur_command
, sizeof(tur_command
));
758 * zero the sense buffer. the scsi spec mandates that any
759 * untransferred sense data should be interpreted as being zero.
761 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
763 saved_result
= scmd
->result
;
764 scmd
->request_buffer
= NULL
;
765 scmd
->request_bufflen
= 0;
767 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
769 scmd
->sc_data_direction
= DMA_NONE
;
771 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
774 * when we eventually call scsi_finish, we really wish to complete
775 * the original request, so let's restore the original data. (db)
777 scsi_setup_cmd_retry(scmd
);
778 scmd
->result
= saved_result
;
781 * hey, we are done. let's look to see what happened.
783 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
784 __FUNCTION__
, scmd
, rtn
));
787 else if (rtn
== NEEDS_RETRY
)
794 * scsi_eh_abort_cmds - abort canceled commands.
795 * @shost: scsi host being recovered.
796 * @eh_done_q: list_head for processed commands.
799 * Try and see whether or not it makes sense to try and abort the
800 * running command. this only works out to be the case if we have one
801 * command that has timed out. if the command simply failed, it makes
802 * no sense to try and abort the command, since as far as the shost
803 * adapter is concerned, it isn't running.
805 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
806 struct list_head
*done_q
)
808 struct list_head
*lh
, *lh_sf
;
809 struct scsi_cmnd
*scmd
;
812 list_for_each_safe(lh
, lh_sf
, work_q
) {
813 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
814 if (!scsi_eh_eflags_chk(scmd
, SCSI_EH_CANCEL_CMD
))
816 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
817 "0x%p\n", current
->comm
,
819 rtn
= scsi_try_to_abort_cmd(scmd
);
820 if (rtn
== SUCCESS
) {
821 scsi_eh_eflags_clr(scmd
, SCSI_EH_CANCEL_CMD
);
822 if (!scsi_device_online(scmd
->device
) ||
823 !scsi_eh_tur(scmd
)) {
824 scsi_eh_finish_cmd(scmd
, done_q
);
828 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
835 return list_empty(work_q
);
839 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
840 * @scmd: SCSI cmd used to send BDR
843 * There is no timeout for this operation. if this operation is
844 * unreliable for a given host, then the host itself needs to put a
845 * timer on it, and set the host back to a consistent state prior to
848 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
852 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
855 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
856 if (rtn
== SUCCESS
) {
857 scmd
->device
->was_reset
= 1;
858 scmd
->device
->expecting_cc_ua
= 1;
865 * scsi_eh_try_stu - Send START_UNIT to device.
866 * @scmd: Scsi cmd to send START_UNIT
869 * 0 - Device is ready. 1 - Device NOT ready.
871 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
873 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
877 if (!scmd
->device
->allow_restart
)
880 memcpy(scmd
->cmnd
, stu_command
, sizeof(stu_command
));
883 * zero the sense buffer. the scsi spec mandates that any
884 * untransferred sense data should be interpreted as being zero.
886 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
888 saved_result
= scmd
->result
;
889 scmd
->request_buffer
= NULL
;
890 scmd
->request_bufflen
= 0;
892 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
894 scmd
->sc_data_direction
= DMA_NONE
;
896 rtn
= scsi_send_eh_cmnd(scmd
, START_UNIT_TIMEOUT
);
899 * when we eventually call scsi_finish, we really wish to complete
900 * the original request, so let's restore the original data. (db)
902 scsi_setup_cmd_retry(scmd
);
903 scmd
->result
= saved_result
;
906 * hey, we are done. let's look to see what happened.
908 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
909 __FUNCTION__
, scmd
, rtn
));
916 * scsi_eh_stu - send START_UNIT if needed
917 * @shost: scsi host being recovered.
918 * @eh_done_q: list_head for processed commands.
921 * If commands are failing due to not ready, initializing command required,
922 * try revalidating the device, which will end up sending a start unit.
924 static int scsi_eh_stu(struct Scsi_Host
*shost
,
925 struct list_head
*work_q
,
926 struct list_head
*done_q
)
928 struct list_head
*lh
, *lh_sf
;
929 struct scsi_cmnd
*scmd
, *stu_scmd
;
930 struct scsi_device
*sdev
;
932 shost_for_each_device(sdev
, shost
) {
934 list_for_each_entry(scmd
, work_q
, eh_entry
)
935 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
936 scsi_check_sense(scmd
) == FAILED
) {
944 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
945 " 0x%p\n", current
->comm
, sdev
));
947 if (!scsi_eh_try_stu(stu_scmd
)) {
948 if (!scsi_device_online(sdev
) ||
949 !scsi_eh_tur(stu_scmd
)) {
950 list_for_each_safe(lh
, lh_sf
, work_q
) {
951 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
952 if (scmd
->device
== sdev
)
953 scsi_eh_finish_cmd(scmd
, done_q
);
957 SCSI_LOG_ERROR_RECOVERY(3,
958 printk("%s: START_UNIT failed to sdev:"
959 " 0x%p\n", current
->comm
, sdev
));
963 return list_empty(work_q
);
968 * scsi_eh_bus_device_reset - send bdr if needed
969 * @shost: scsi host being recovered.
970 * @eh_done_q: list_head for processed commands.
973 * Try a bus device reset. still, look to see whether we have multiple
974 * devices that are jammed or not - if we have multiple devices, it
975 * makes no sense to try bus_device_reset - we really would need to try
976 * a bus_reset instead.
978 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
979 struct list_head
*work_q
,
980 struct list_head
*done_q
)
982 struct list_head
*lh
, *lh_sf
;
983 struct scsi_cmnd
*scmd
, *bdr_scmd
;
984 struct scsi_device
*sdev
;
987 shost_for_each_device(sdev
, shost
) {
989 list_for_each_entry(scmd
, work_q
, eh_entry
)
990 if (scmd
->device
== sdev
) {
998 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
999 " 0x%p\n", current
->comm
,
1001 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1002 if (rtn
== SUCCESS
) {
1003 if (!scsi_device_online(sdev
) ||
1004 !scsi_eh_tur(bdr_scmd
)) {
1005 list_for_each_safe(lh
, lh_sf
,
1007 scmd
= list_entry(lh
, struct
1010 if (scmd
->device
== sdev
)
1011 scsi_eh_finish_cmd(scmd
,
1016 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1024 return list_empty(work_q
);
1028 * scsi_try_bus_reset - ask host to perform a bus reset
1029 * @scmd: SCSI cmd to send bus reset.
1031 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
1033 unsigned long flags
;
1036 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1039 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
1042 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
1044 if (rtn
== SUCCESS
) {
1045 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1046 ssleep(BUS_RESET_SETTLE_TIME
);
1047 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1048 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1049 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1056 * scsi_try_host_reset - ask host adapter to reset itself
1057 * @scmd: SCSI cmd to send hsot reset.
1059 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
1061 unsigned long flags
;
1064 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1067 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
1070 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
1072 if (rtn
== SUCCESS
) {
1073 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1074 ssleep(HOST_RESET_SETTLE_TIME
);
1075 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1076 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1077 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1084 * scsi_eh_bus_reset - send a bus reset
1085 * @shost: scsi host being recovered.
1086 * @eh_done_q: list_head for processed commands.
1088 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1089 struct list_head
*work_q
,
1090 struct list_head
*done_q
)
1092 struct list_head
*lh
, *lh_sf
;
1093 struct scsi_cmnd
*scmd
;
1094 struct scsi_cmnd
*chan_scmd
;
1095 unsigned int channel
;
1099 * we really want to loop over the various channels, and do this on
1100 * a channel by channel basis. we should also check to see if any
1101 * of the failed commands are on soft_reset devices, and if so, skip
1105 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1107 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1108 if (channel
== scmd
->device
->channel
) {
1112 * FIXME add back in some support for
1113 * soft_reset devices.
1120 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1121 " %d\n", current
->comm
,
1123 rtn
= scsi_try_bus_reset(chan_scmd
);
1124 if (rtn
== SUCCESS
) {
1125 list_for_each_safe(lh
, lh_sf
, work_q
) {
1126 scmd
= list_entry(lh
, struct scsi_cmnd
,
1128 if (channel
== scmd
->device
->channel
)
1129 if (!scsi_device_online(scmd
->device
) ||
1131 scsi_eh_finish_cmd(scmd
,
1135 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1136 " failed chan: %d\n",
1141 return list_empty(work_q
);
1145 * scsi_eh_host_reset - send a host reset
1146 * @work_q: list_head for processed commands.
1147 * @done_q: list_head for processed commands.
1149 static int scsi_eh_host_reset(struct list_head
*work_q
,
1150 struct list_head
*done_q
)
1153 struct list_head
*lh
, *lh_sf
;
1154 struct scsi_cmnd
*scmd
;
1156 if (!list_empty(work_q
)) {
1157 scmd
= list_entry(work_q
->next
,
1158 struct scsi_cmnd
, eh_entry
);
1160 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1163 rtn
= scsi_try_host_reset(scmd
);
1164 if (rtn
== SUCCESS
) {
1165 list_for_each_safe(lh
, lh_sf
, work_q
) {
1166 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
1167 if (!scsi_device_online(scmd
->device
) ||
1168 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1170 scsi_eh_finish_cmd(scmd
, done_q
);
1173 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1178 return list_empty(work_q
);
1182 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1183 * @work_q: list_head for processed commands.
1184 * @done_q: list_head for processed commands.
1187 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1188 struct list_head
*done_q
)
1190 struct list_head
*lh
, *lh_sf
;
1191 struct scsi_cmnd
*scmd
;
1193 list_for_each_safe(lh
, lh_sf
, work_q
) {
1194 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
1195 printk(KERN_INFO
"scsi: Device offlined - not"
1196 " ready after error recovery: host"
1197 " %d channel %d id %d lun %d\n",
1198 scmd
->device
->host
->host_no
,
1199 scmd
->device
->channel
,
1202 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1203 if (scsi_eh_eflags_chk(scmd
, SCSI_EH_CANCEL_CMD
)) {
1205 * FIXME: Handle lost cmds.
1208 scsi_eh_finish_cmd(scmd
, done_q
);
1214 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1215 * @scmd: SCSI cmd to examine.
1218 * This is *only* called when we are examining the status after sending
1219 * out the actual data command. any commands that are queued for error
1220 * recovery (e.g. test_unit_ready) do *not* come through here.
1222 * When this routine returns failed, it means the error handler thread
1223 * is woken. In cases where the error code indicates an error that
1224 * doesn't require the error handler read (i.e. we don't need to
1225 * abort/reset), this function should return SUCCESS.
1227 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1232 * if the device is offline, then we clearly just pass the result back
1233 * up to the top level.
1235 if (!scsi_device_online(scmd
->device
)) {
1236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1243 * first check the host byte, to see if there is anything in there
1244 * that would indicate what we need to do.
1246 switch (host_byte(scmd
->result
)) {
1247 case DID_PASSTHROUGH
:
1249 * no matter what, pass this through to the upper layer.
1250 * nuke this special code so that it looks like we are saying
1253 scmd
->result
&= 0xff00ffff;
1257 * looks good. drop through, and check the next byte.
1260 case DID_NO_CONNECT
:
1261 case DID_BAD_TARGET
:
1264 * note - this means that we just report the status back
1265 * to the top level driver, not that we actually think
1266 * that it indicates SUCCESS.
1270 * when the low level driver returns did_soft_error,
1271 * it is responsible for keeping an internal retry counter
1272 * in order to avoid endless loops (db)
1274 * actually this is a bug in this function here. we should
1275 * be mindful of the maximum number of retries specified
1276 * and not get stuck in a loop.
1278 case DID_SOFT_ERROR
:
1284 return ADD_TO_MLQUEUE
;
1287 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1288 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1290 * execute reservation conflict processing code
1301 * when we scan the bus, we get timeout messages for
1302 * these commands if there is no device available.
1303 * other hosts report did_no_connect for the same thing.
1305 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1306 scmd
->cmnd
[0] == INQUIRY
)) {
1318 * next, check the message byte.
1320 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1324 * check the status byte to see if this indicates anything special.
1326 switch (status_byte(scmd
->result
)) {
1329 * the case of trying to send too many commands to a
1330 * tagged queueing device.
1334 * device can't talk to us at the moment. Should only
1335 * occur (SAM-3) when the task queue is empty, so will cause
1336 * the empty queue handling to trigger a stall in the
1339 return ADD_TO_MLQUEUE
;
1341 case COMMAND_TERMINATED
:
1344 case CHECK_CONDITION
:
1345 rtn
= scsi_check_sense(scmd
);
1346 if (rtn
== NEEDS_RETRY
)
1348 /* if rtn == FAILED, we have no sense information;
1349 * returning FAILED will wake the error handler thread
1350 * to collect the sense and redo the decide
1353 case CONDITION_GOOD
:
1354 case INTERMEDIATE_GOOD
:
1355 case INTERMEDIATE_C_GOOD
:
1358 * who knows? FIXME(eric)
1362 case RESERVATION_CONFLICT
:
1363 printk(KERN_INFO
"scsi: reservation conflict: host"
1364 " %d channel %d id %d lun %d\n",
1365 scmd
->device
->host
->host_no
, scmd
->device
->channel
,
1366 scmd
->device
->id
, scmd
->device
->lun
);
1367 return SUCCESS
; /* causes immediate i/o error */
1375 /* we requeue for retry because the error was retryable, and
1376 * the request was not marked fast fail. Note that above,
1377 * even if the request is marked fast fail, we still requeue
1378 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1379 if ((++scmd
->retries
) < scmd
->allowed
1380 && !blk_noretry_request(scmd
->request
)) {
1384 * no more retries - report this one back to upper level.
1391 * scsi_eh_lock_done - done function for eh door lock request
1392 * @scmd: SCSI command block for the door lock request
1395 * We completed the asynchronous door lock request, and it has either
1396 * locked the door or failed. We must free the command structures
1397 * associated with this request.
1399 static void scsi_eh_lock_done(struct scsi_cmnd
*scmd
)
1401 struct scsi_request
*sreq
= scmd
->sc_request
;
1403 scsi_release_request(sreq
);
1408 * scsi_eh_lock_door - Prevent medium removal for the specified device
1409 * @sdev: SCSI device to prevent medium removal
1412 * We must be called from process context; scsi_allocate_request()
1416 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1417 * head of the devices request queue, and continue.
1420 * scsi_allocate_request() may sleep waiting for existing requests to
1421 * be processed. However, since we haven't kicked off any request
1422 * processing for this host, this may deadlock.
1424 * If scsi_allocate_request() fails for what ever reason, we
1425 * completely forget to lock the door.
1427 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1429 struct scsi_request
*sreq
= scsi_allocate_request(sdev
, GFP_KERNEL
);
1431 if (unlikely(!sreq
)) {
1432 printk(KERN_ERR
"%s: request allocate failed,"
1433 "prevent media removal cmd not sent\n", __FUNCTION__
);
1437 sreq
->sr_cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1438 sreq
->sr_cmnd
[1] = 0;
1439 sreq
->sr_cmnd
[2] = 0;
1440 sreq
->sr_cmnd
[3] = 0;
1441 sreq
->sr_cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1442 sreq
->sr_cmnd
[5] = 0;
1443 sreq
->sr_data_direction
= DMA_NONE
;
1444 sreq
->sr_bufflen
= 0;
1445 sreq
->sr_buffer
= NULL
;
1446 sreq
->sr_allowed
= 5;
1447 sreq
->sr_done
= scsi_eh_lock_done
;
1448 sreq
->sr_timeout_per_command
= 10 * HZ
;
1449 sreq
->sr_cmd_len
= COMMAND_SIZE(sreq
->sr_cmnd
[0]);
1451 scsi_insert_special_req(sreq
, 1);
1456 * scsi_restart_operations - restart io operations to the specified host.
1457 * @shost: Host we are restarting.
1460 * When we entered the error handler, we blocked all further i/o to
1461 * this device. we need to 'reverse' this process.
1463 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1465 struct scsi_device
*sdev
;
1468 * If the door was locked, we need to insert a door lock request
1469 * onto the head of the SCSI request queue for the device. There
1470 * is no point trying to lock the door of an off-line device.
1472 shost_for_each_device(sdev
, shost
) {
1473 if (scsi_device_online(sdev
) && sdev
->locked
)
1474 scsi_eh_lock_door(sdev
);
1478 * next free up anything directly waiting upon the host. this
1479 * will be requests for character device operations, and also for
1480 * ioctls to queued block devices.
1482 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1485 clear_bit(SHOST_RECOVERY
, &shost
->shost_state
);
1487 wake_up(&shost
->host_wait
);
1490 * finally we need to re-initiate requests that may be pending. we will
1491 * have had everything blocked while error handling is taking place, and
1492 * now that error recovery is done, we will need to ensure that these
1493 * requests are started.
1495 scsi_run_host_queues(shost
);
1499 * scsi_eh_ready_devs - check device ready state and recover if not.
1500 * @shost: host to be recovered.
1501 * @eh_done_q: list_head for processed commands.
1504 static void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1505 struct list_head
*work_q
,
1506 struct list_head
*done_q
)
1508 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1509 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1510 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1511 if (!scsi_eh_host_reset(work_q
, done_q
))
1512 scsi_eh_offline_sdevs(work_q
, done_q
);
1516 * scsi_eh_flush_done_q - finish processed commands or retry them.
1517 * @done_q: list_head of processed commands.
1520 static void scsi_eh_flush_done_q(struct list_head
*done_q
)
1522 struct list_head
*lh
, *lh_sf
;
1523 struct scsi_cmnd
*scmd
;
1525 list_for_each_safe(lh
, lh_sf
, done_q
) {
1526 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
1528 if (scsi_device_online(scmd
->device
) &&
1529 !blk_noretry_request(scmd
->request
) &&
1530 (++scmd
->retries
< scmd
->allowed
)) {
1531 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1535 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1538 * If just we got sense for the device (called
1539 * scsi_eh_get_sense), scmd->result is already
1540 * set, do not set DRIVER_TIMEOUT.
1543 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1544 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1546 current
->comm
, scmd
));
1547 scsi_finish_command(scmd
);
1553 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1554 * @shost: Host to unjam.
1557 * When we come in here, we *know* that all commands on the bus have
1558 * either completed, failed or timed out. we also know that no further
1559 * commands are being sent to the host, so things are relatively quiet
1560 * and we have freedom to fiddle with things as we wish.
1562 * This is only the *default* implementation. it is possible for
1563 * individual drivers to supply their own version of this function, and
1564 * if the maintainer wishes to do this, it is strongly suggested that
1565 * this function be taken as a template and modified. this function
1566 * was designed to correctly handle problems for about 95% of the
1567 * different cases out there, and it should always provide at least a
1568 * reasonable amount of error recovery.
1570 * Any command marked 'failed' or 'timeout' must eventually have
1571 * scsi_finish_cmd() called for it. we do all of the retry stuff
1572 * here, so when we restart the host after we return it should have an
1575 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1577 unsigned long flags
;
1578 LIST_HEAD(eh_work_q
);
1579 LIST_HEAD(eh_done_q
);
1581 spin_lock_irqsave(shost
->host_lock
, flags
);
1582 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1583 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1585 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1587 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1588 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1589 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1591 scsi_eh_flush_done_q(&eh_done_q
);
1595 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1596 * @data: Host for which we are running.
1599 * This is always run in the context of a kernel thread. The idea is
1600 * that we start this thing up when the kernel starts up (one per host
1601 * that we detect), and it immediately goes to sleep and waits for some
1602 * event (i.e. failure). When this takes place, we have the job of
1603 * trying to unjam the bus and restarting things.
1605 int scsi_error_handler(void *data
)
1607 struct Scsi_Host
*shost
= (struct Scsi_Host
*) data
;
1609 DECLARE_MUTEX_LOCKED(sem
);
1615 daemonize("scsi_eh_%d", shost
->host_no
);
1617 current
->flags
|= PF_NOFREEZE
;
1619 shost
->eh_wait
= &sem
;
1620 shost
->ehandler
= current
;
1623 * Wake up the thread that created us.
1625 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1626 " scsi_eh_%d\n",shost
->host_no
));
1628 complete(shost
->eh_notify
);
1632 * If we get a signal, it means we are supposed to go
1633 * away and die. This typically happens if the user is
1634 * trying to unload a module.
1636 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1638 " sleeping\n",shost
->host_no
));
1641 * Note - we always use down_interruptible with the semaphore
1642 * even if the module was loaded as part of the kernel. The
1643 * reason is that down() will cause this thread to be counted
1644 * in the load average as a running process, and down
1645 * interruptible doesn't. Given that we need to allow this
1646 * thread to die if the driver was loaded as a module, using
1647 * semaphores isn't unreasonable.
1649 down_interruptible(&sem
);
1653 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1654 " scsi_eh_%d waking"
1655 " up\n",shost
->host_no
));
1657 shost
->eh_active
= 1;
1660 * We have a host that is failing for some reason. Figure out
1661 * what we need to do to get it up and online again (if we can).
1662 * If we fail, we end up taking the thing offline.
1664 if (shost
->hostt
->eh_strategy_handler
)
1665 rtn
= shost
->hostt
->eh_strategy_handler(shost
);
1667 scsi_unjam_host(shost
);
1669 shost
->eh_active
= 0;
1672 * Note - if the above fails completely, the action is to take
1673 * individual devices offline and flush the queue of any
1674 * outstanding requests that may have been pending. When we
1675 * restart, we restart any I/O to any other devices on the bus
1676 * which are still online.
1678 scsi_restart_operations(shost
);
1682 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1683 " exiting\n",shost
->host_no
));
1686 * Make sure that nobody tries to wake us up again.
1688 shost
->eh_wait
= NULL
;
1691 * Knock this down too. From this point on, the host is flying
1692 * without a pilot. If this is because the module is being unloaded,
1693 * that's fine. If the user sent a signal to this thing, we are
1694 * potentially in real danger.
1696 shost
->eh_active
= 0;
1697 shost
->ehandler
= NULL
;
1700 * If anyone is waiting for us to exit (i.e. someone trying to unload
1701 * a driver), then wake up that process to let them know we are on
1702 * the way out the door.
1704 complete_and_exit(shost
->eh_notify
, 0);
1709 * Function: scsi_report_bus_reset()
1711 * Purpose: Utility function used by low-level drivers to report that
1712 * they have observed a bus reset on the bus being handled.
1714 * Arguments: shost - Host in question
1715 * channel - channel on which reset was observed.
1719 * Lock status: Host lock must be held.
1721 * Notes: This only needs to be called if the reset is one which
1722 * originates from an unknown location. Resets originated
1723 * by the mid-level itself don't need to call this, but there
1724 * should be no harm.
1726 * The main purpose of this is to make sure that a CHECK_CONDITION
1727 * is properly treated.
1729 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1731 struct scsi_device
*sdev
;
1733 __shost_for_each_device(sdev
, shost
) {
1734 if (channel
== sdev
->channel
) {
1735 sdev
->was_reset
= 1;
1736 sdev
->expecting_cc_ua
= 1;
1740 EXPORT_SYMBOL(scsi_report_bus_reset
);
1743 * Function: scsi_report_device_reset()
1745 * Purpose: Utility function used by low-level drivers to report that
1746 * they have observed a device reset on the device being handled.
1748 * Arguments: shost - Host in question
1749 * channel - channel on which reset was observed
1750 * target - target on which reset was observed
1754 * Lock status: Host lock must be held
1756 * Notes: This only needs to be called if the reset is one which
1757 * originates from an unknown location. Resets originated
1758 * by the mid-level itself don't need to call this, but there
1759 * should be no harm.
1761 * The main purpose of this is to make sure that a CHECK_CONDITION
1762 * is properly treated.
1764 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1766 struct scsi_device
*sdev
;
1768 __shost_for_each_device(sdev
, shost
) {
1769 if (channel
== sdev
->channel
&&
1770 target
== sdev
->id
) {
1771 sdev
->was_reset
= 1;
1772 sdev
->expecting_cc_ua
= 1;
1776 EXPORT_SYMBOL(scsi_report_device_reset
);
1779 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1784 * Function: scsi_reset_provider
1786 * Purpose: Send requested reset to a bus or device at any phase.
1788 * Arguments: device - device to send reset to
1789 * flag - reset type (see scsi.h)
1791 * Returns: SUCCESS/FAILURE.
1793 * Notes: This is used by the SCSI Generic driver to provide
1794 * Bus/Device reset capability.
1797 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1799 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1803 scmd
->request
= &req
;
1804 memset(&scmd
->eh_timeout
, 0, sizeof(scmd
->eh_timeout
));
1805 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
1806 scmd
->state
= SCSI_STATE_INITIALIZING
;
1808 memset(&scmd
->cmnd
, '\0', sizeof(scmd
->cmnd
));
1810 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1812 scmd
->buffer
= NULL
;
1814 scmd
->request_buffer
= NULL
;
1815 scmd
->request_bufflen
= 0;
1819 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1820 scmd
->sc_request
= NULL
;
1821 scmd
->sc_magic
= SCSI_CMND_MAGIC
;
1823 init_timer(&scmd
->eh_timeout
);
1826 * Sometimes the command can get back into the timer chain,
1827 * so use the pid as an identifier.
1832 case SCSI_TRY_RESET_DEVICE
:
1833 rtn
= scsi_try_bus_device_reset(scmd
);
1837 case SCSI_TRY_RESET_BUS
:
1838 rtn
= scsi_try_bus_reset(scmd
);
1842 case SCSI_TRY_RESET_HOST
:
1843 rtn
= scsi_try_host_reset(scmd
);
1849 scsi_next_command(scmd
);
1852 EXPORT_SYMBOL(scsi_reset_provider
);
1855 * scsi_normalize_sense - normalize main elements from either fixed or
1856 * descriptor sense data format into a common format.
1858 * @sense_buffer: byte array containing sense data returned by device
1859 * @sb_len: number of valid bytes in sense_buffer
1860 * @sshdr: pointer to instance of structure that common
1861 * elements are written to.
1864 * The "main elements" from sense data are: response_code, sense_key,
1865 * asc, ascq and additional_length (only for descriptor format).
1867 * Typically this function can be called after a device has
1868 * responded to a SCSI command with the CHECK_CONDITION status.
1871 * 1 if valid sense data information found, else 0;
1873 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1874 struct scsi_sense_hdr
*sshdr
)
1876 if (!sense_buffer
|| !sb_len
|| (sense_buffer
[0] & 0x70) != 0x70)
1879 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1881 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1882 if (sshdr
->response_code
>= 0x72) {
1887 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1889 sshdr
->asc
= sense_buffer
[2];
1891 sshdr
->ascq
= sense_buffer
[3];
1893 sshdr
->additional_length
= sense_buffer
[7];
1899 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1901 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1902 sb_len
: (sense_buffer
[7] + 8);
1904 sshdr
->asc
= sense_buffer
[12];
1906 sshdr
->ascq
= sense_buffer
[13];
1912 EXPORT_SYMBOL(scsi_normalize_sense
);
1914 int scsi_request_normalize_sense(struct scsi_request
*sreq
,
1915 struct scsi_sense_hdr
*sshdr
)
1917 return scsi_normalize_sense(sreq
->sr_sense_buffer
,
1918 sizeof(sreq
->sr_sense_buffer
), sshdr
);
1920 EXPORT_SYMBOL(scsi_request_normalize_sense
);
1922 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1923 struct scsi_sense_hdr
*sshdr
)
1925 return scsi_normalize_sense(cmd
->sense_buffer
,
1926 sizeof(cmd
->sense_buffer
), sshdr
);
1928 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1931 * scsi_sense_desc_find - search for a given descriptor type in
1932 * descriptor sense data format.
1934 * @sense_buffer: byte array of descriptor format sense data
1935 * @sb_len: number of valid bytes in sense_buffer
1936 * @desc_type: value of descriptor type to find
1937 * (e.g. 0 -> information)
1940 * only valid when sense data is in descriptor format
1943 * pointer to start of (first) descriptor if found else NULL
1945 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1948 int add_sen_len
, add_len
, desc_len
, k
;
1951 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1953 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1955 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1956 add_sen_len
: (sb_len
- 8);
1957 descp
= &sense_buffer
[8];
1958 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1960 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1961 desc_len
= add_len
+ 2;
1962 if (descp
[0] == desc_type
)
1964 if (add_len
< 0) // short descriptor ??
1969 EXPORT_SYMBOL(scsi_sense_desc_find
);
1972 * scsi_get_sense_info_fld - attempts to get information field from
1973 * sense data (either fixed or descriptor format)
1975 * @sense_buffer: byte array of sense data
1976 * @sb_len: number of valid bytes in sense_buffer
1977 * @info_out: pointer to 64 integer where 8 or 4 byte information
1978 * field will be placed if found.
1981 * 1 if information field found, 0 if not found.
1983 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
1992 switch (sense_buffer
[0] & 0x7f) {
1995 if (sense_buffer
[0] & 0x80) {
1996 *info_out
= (sense_buffer
[3] << 24) +
1997 (sense_buffer
[4] << 16) +
1998 (sense_buffer
[5] << 8) + sense_buffer
[6];
2004 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2006 if (ucp
&& (0xa == ucp
[1])) {
2008 for (j
= 0; j
< 8; ++j
) {
2021 EXPORT_SYMBOL(scsi_get_sense_info_fld
);