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
->owner
= SCSI_OWNER_ERROR_HANDLER
;
81 scmd
->state
= SCSI_STATE_FAILED
;
82 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
83 set_bit(SHOST_RECOVERY
, &shost
->shost_state
);
85 scsi_eh_wakeup(shost
);
86 spin_unlock_irqrestore(shost
->host_lock
, flags
);
91 * scsi_add_timer - Start timeout timer for a single scsi command.
92 * @scmd: scsi command that is about to start running.
93 * @timeout: amount of time to allow this command to run.
94 * @complete: timeout function to call if timer isn't canceled.
97 * This should be turned into an inline function. Each scsi command
98 * has its own timer, and as it is added to the queue, we set up the
99 * timer. When the command completes, we cancel the timer.
101 void scsi_add_timer(struct scsi_cmnd
*scmd
, int timeout
,
102 void (*complete
)(struct scsi_cmnd
*))
106 * If the clock was already running for this command, then
107 * first delete the timer. The timer handling code gets rather
108 * confused if we don't do this.
110 if (scmd
->eh_timeout
.function
)
111 del_timer(&scmd
->eh_timeout
);
113 scmd
->eh_timeout
.data
= (unsigned long)scmd
;
114 scmd
->eh_timeout
.expires
= jiffies
+ timeout
;
115 scmd
->eh_timeout
.function
= (void (*)(unsigned long)) complete
;
117 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
118 " %d, (%p)\n", __FUNCTION__
,
119 scmd
, timeout
, complete
));
121 add_timer(&scmd
->eh_timeout
);
123 EXPORT_SYMBOL(scsi_add_timer
);
126 * scsi_delete_timer - Delete/cancel timer for a given function.
127 * @scmd: Cmd that we are canceling timer for
130 * This should be turned into an inline function.
133 * 1 if we were able to detach the timer. 0 if we blew it, and the
134 * timer function has already started to run.
136 int scsi_delete_timer(struct scsi_cmnd
*scmd
)
140 rtn
= del_timer(&scmd
->eh_timeout
);
142 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
143 " rtn: %d\n", __FUNCTION__
,
146 scmd
->eh_timeout
.data
= (unsigned long)NULL
;
147 scmd
->eh_timeout
.function
= NULL
;
151 EXPORT_SYMBOL(scsi_delete_timer
);
154 * scsi_times_out - Timeout function for normal scsi commands.
155 * @scmd: Cmd that is timing out.
158 * We do not need to lock this. There is the potential for a race
159 * only in that the normal completion handling might run, but if the
160 * normal completion function determines that the timer has already
161 * fired, then it mustn't do anything.
163 void scsi_times_out(struct scsi_cmnd
*scmd
)
165 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
167 if (scmd
->device
->host
->hostt
->eh_timed_out
)
168 switch (scmd
->device
->host
->hostt
->eh_timed_out(scmd
)) {
173 /* This allows a single retry even of a command
174 * with allowed == 0 */
175 if (scmd
->retries
++ > scmd
->allowed
)
177 scsi_add_timer(scmd
, scmd
->timeout_per_command
,
184 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
185 panic("Error handler thread not present at %p %p %s %d",
186 scmd
, scmd
->device
->host
, __FILE__
, __LINE__
);
191 * scsi_block_when_processing_errors - Prevent cmds from being queued.
192 * @sdev: Device on which we are performing recovery.
195 * We block until the host is out of error recovery, and then check to
196 * see whether the host or the device is offline.
199 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
201 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
205 wait_event(sdev
->host
->host_wait
, (!test_bit(SHOST_RECOVERY
, &sdev
->host
->shost_state
)));
207 online
= scsi_device_online(sdev
);
209 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__
,
214 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
216 #ifdef CONFIG_SCSI_LOGGING
218 * scsi_eh_prt_fail_stats - Log info on failures.
219 * @shost: scsi host being recovered.
220 * @work_q: Queue of scsi cmds to process.
222 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
223 struct list_head
*work_q
)
225 struct scsi_cmnd
*scmd
;
226 struct scsi_device
*sdev
;
227 int total_failures
= 0;
230 int devices_failed
= 0;
232 shost_for_each_device(sdev
, shost
) {
233 list_for_each_entry(scmd
, work_q
, eh_entry
) {
234 if (scmd
->device
== sdev
) {
236 if (scsi_eh_eflags_chk(scmd
,
244 if (cmd_cancel
|| cmd_failed
) {
245 SCSI_LOG_ERROR_RECOVERY(3,
246 printk("%s: %d:%d:%d:%d cmds failed: %d,"
248 __FUNCTION__
, shost
->host_no
,
249 sdev
->channel
, sdev
->id
, sdev
->lun
,
250 cmd_failed
, cmd_cancel
));
257 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
258 " devices require eh work\n",
259 total_failures
, devices_failed
));
264 * scsi_check_sense - Examine scsi cmd sense
265 * @scmd: Cmd to have sense checked.
268 * SUCCESS or FAILED or NEEDS_RETRY
271 * When a deferred error is detected the current command has
272 * not been executed and needs retrying.
274 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
276 struct scsi_sense_hdr sshdr
;
278 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
279 return FAILED
; /* no valid sense data */
281 if (scsi_sense_is_deferred(&sshdr
))
285 * Previous logic looked for FILEMARK, EOM or ILI which are
286 * mainly associated with tapes and returned SUCCESS.
288 if (sshdr
.response_code
== 0x70) {
290 if (scmd
->sense_buffer
[2] & 0xe0)
294 * descriptor format: look for "stream commands sense data
295 * descriptor" (see SSC-3). Assume single sense data
296 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
298 if ((sshdr
.additional_length
> 3) &&
299 (scmd
->sense_buffer
[8] == 0x4) &&
300 (scmd
->sense_buffer
[11] & 0xe0))
304 switch (sshdr
.sense_key
) {
307 case RECOVERED_ERROR
:
308 return /* soft_error */ SUCCESS
;
310 case ABORTED_COMMAND
:
315 * if we are expecting a cc/ua because of a bus reset that we
316 * performed, treat this just as a retry. otherwise this is
317 * information that we should pass up to the upper-level driver
318 * so that we can deal with it there.
320 if (scmd
->device
->expecting_cc_ua
) {
321 scmd
->device
->expecting_cc_ua
= 0;
325 * if the device is in the process of becoming ready, we
328 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
331 * if the device is not started, we need to wake
332 * the error handler to start the motor
334 if (scmd
->device
->allow_restart
&&
335 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
339 /* these three are not supported */
341 case VOLUME_OVERFLOW
:
349 if (scmd
->device
->retry_hwerror
)
354 case ILLEGAL_REQUEST
:
363 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
364 * @scmd: SCSI cmd to examine.
367 * This is *only* called when we are examining the status of commands
368 * queued during error recovery. the main difference here is that we
369 * don't allow for the possibility of retries here, and we are a lot
370 * more restrictive about what we consider acceptable.
372 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
375 * first check the host byte, to see if there is anything in there
376 * that would indicate what we need to do.
378 if (host_byte(scmd
->result
) == DID_RESET
) {
380 * rats. we are already in the error handler, so we now
381 * get to try and figure out what to do next. if the sense
382 * is valid, we have a pretty good idea of what to do.
383 * if not, we mark it as FAILED.
385 return scsi_check_sense(scmd
);
387 if (host_byte(scmd
->result
) != DID_OK
)
391 * next, check the message byte.
393 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
397 * now, check the status byte to see if this indicates
400 switch (status_byte(scmd
->result
)) {
402 case COMMAND_TERMINATED
:
404 case CHECK_CONDITION
:
405 return scsi_check_sense(scmd
);
407 case INTERMEDIATE_GOOD
:
408 case INTERMEDIATE_C_GOOD
:
410 * who knows? FIXME(eric)
415 case RESERVATION_CONFLICT
:
423 * scsi_eh_times_out - timeout function for error handling.
424 * @scmd: Cmd that is timing out.
427 * During error handling, the kernel thread will be sleeping waiting
428 * for some action to complete on the device. our only job is to
429 * record that it timed out, and to wake up the thread.
431 static void scsi_eh_times_out(struct scsi_cmnd
*scmd
)
433 scsi_eh_eflags_set(scmd
, SCSI_EH_REC_TIMEOUT
);
434 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__
,
437 up(scmd
->device
->host
->eh_action
);
441 * scsi_eh_done - Completion function for error handling.
442 * @scmd: Cmd that is done.
444 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
447 * if the timeout handler is already running, then just set the
448 * flag which says we finished late, and return. we have no
449 * way of stopping the timeout handler from running, so we must
450 * always defer to it.
452 if (del_timer(&scmd
->eh_timeout
)) {
453 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
454 scmd
->owner
= SCSI_OWNER_ERROR_HANDLER
;
456 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
457 __FUNCTION__
, scmd
, scmd
->result
));
459 up(scmd
->device
->host
->eh_action
);
464 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
465 * @scmd: SCSI Cmd to send.
466 * @timeout: Timeout for cmd.
469 * The initialization of the structures is quite a bit different in
470 * this case, and furthermore, there is a different completion handler
471 * vs scsi_dispatch_cmd.
473 * SUCCESS or FAILED or NEEDS_RETRY
475 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, int timeout
)
477 struct scsi_device
*sdev
= scmd
->device
;
478 struct Scsi_Host
*shost
= sdev
->host
;
479 DECLARE_MUTEX_LOCKED(sem
);
484 * we will use a queued command if possible, otherwise we will
485 * emulate the queuing and calling of completion function ourselves.
487 scmd
->owner
= SCSI_OWNER_LOWLEVEL
;
489 if (sdev
->scsi_level
<= SCSI_2
)
490 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
491 (sdev
->lun
<< 5 & 0xe0);
493 scsi_add_timer(scmd
, timeout
, scsi_eh_times_out
);
496 * set up the semaphore so we wait for the command to complete.
498 shost
->eh_action
= &sem
;
499 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
501 spin_lock_irqsave(shost
->host_lock
, flags
);
503 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
504 spin_unlock_irqrestore(shost
->host_lock
, flags
);
507 scsi_log_completion(scmd
, SUCCESS
);
509 shost
->eh_action
= NULL
;
512 * see if timeout. if so, tell the host to forget about it.
513 * in other words, we don't want a callback any more.
515 if (scsi_eh_eflags_chk(scmd
, SCSI_EH_REC_TIMEOUT
)) {
516 scsi_eh_eflags_clr(scmd
, SCSI_EH_REC_TIMEOUT
);
517 scmd
->owner
= SCSI_OWNER_LOWLEVEL
;
520 * as far as the low level driver is
521 * concerned, this command is still active, so
522 * we must give the low level driver a chance
525 * FIXME(eric) - we are not tracking whether we could
526 * abort a timed out command or not. not sure how
527 * we should treat them differently anyways.
529 if (shost
->hostt
->eh_abort_handler
)
530 shost
->hostt
->eh_abort_handler(scmd
);
532 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
533 scmd
->owner
= SCSI_OWNER_ERROR_HANDLER
;
538 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
539 __FUNCTION__
, scmd
, rtn
));
542 * now examine the actual status codes to see whether the command
543 * actually did complete normally.
545 if (rtn
== SUCCESS
) {
546 rtn
= scsi_eh_completed_normally(scmd
);
547 SCSI_LOG_ERROR_RECOVERY(3,
548 printk("%s: scsi_eh_completed_normally %x\n",
565 * scsi_request_sense - Request sense data from a particular target.
566 * @scmd: SCSI cmd for request sense.
569 * Some hosts automatically obtain this information, others require
570 * that we obtain it on our own. This function will *not* return until
571 * the command either times out, or it completes.
573 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
575 static unsigned char generic_sense
[6] =
576 {REQUEST_SENSE
, 0, 0, 0, 252, 0};
577 unsigned char *scsi_result
;
581 memcpy(scmd
->cmnd
, generic_sense
, sizeof(generic_sense
));
583 scsi_result
= kmalloc(252, GFP_ATOMIC
| ((scmd
->device
->host
->hostt
->unchecked_isa_dma
) ? __GFP_DMA
: 0));
586 if (unlikely(!scsi_result
)) {
587 printk(KERN_ERR
"%s: cannot allocate scsi_result.\n",
593 * zero the sense buffer. some host adapters automatically always
594 * request sense, so it is not a good idea that
595 * scmd->request_buffer and scmd->sense_buffer point to the same
596 * address (db). 0 is not a valid sense code.
598 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
599 memset(scsi_result
, 0, 252);
601 saved_result
= scmd
->result
;
602 scmd
->request_buffer
= scsi_result
;
603 scmd
->request_bufflen
= 252;
605 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
606 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
609 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
611 /* last chance to have valid sense data */
612 if(!SCSI_SENSE_VALID(scmd
)) {
613 memcpy(scmd
->sense_buffer
, scmd
->request_buffer
,
614 sizeof(scmd
->sense_buffer
));
620 * when we eventually call scsi_finish, we really wish to complete
621 * the original request, so let's restore the original data. (db)
623 scsi_setup_cmd_retry(scmd
);
624 scmd
->result
= saved_result
;
629 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
630 * @scmd: Original SCSI cmd that eh has finished.
631 * @done_q: Queue for processed commands.
634 * We don't want to use the normal command completion while we are are
635 * still handling errors - it may cause other commands to be queued,
636 * and that would disturb what we are doing. thus we really want to
637 * keep a list of pending commands for final completion, and once we
638 * are ready to leave error handling we handle completion for real.
640 static void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
,
641 struct list_head
*done_q
)
643 scmd
->device
->host
->host_failed
--;
644 scmd
->state
= SCSI_STATE_BHQUEUE
;
646 scsi_eh_eflags_clr_all(scmd
);
649 * set this back so that the upper level can correctly free up
652 scsi_setup_cmd_retry(scmd
);
653 list_move_tail(&scmd
->eh_entry
, done_q
);
657 * scsi_eh_get_sense - Get device sense data.
658 * @work_q: Queue of commands to process.
659 * @done_q: Queue of proccessed commands..
662 * See if we need to request sense information. if so, then get it
663 * now, so we have a better idea of what to do.
666 * This has the unfortunate side effect that if a shost adapter does
667 * not automatically request sense information, that we end up shutting
668 * it down before we request it.
670 * All drivers should request sense information internally these days,
671 * so for now all I have to say is tough noogies if you end up in here.
673 * XXX: Long term this code should go away, but that needs an audit of
676 static int scsi_eh_get_sense(struct list_head
*work_q
,
677 struct list_head
*done_q
)
679 struct list_head
*lh
, *lh_sf
;
680 struct scsi_cmnd
*scmd
;
683 list_for_each_safe(lh
, lh_sf
, work_q
) {
684 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
685 if (scsi_eh_eflags_chk(scmd
, SCSI_EH_CANCEL_CMD
) ||
686 SCSI_SENSE_VALID(scmd
))
689 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
693 rtn
= scsi_request_sense(scmd
);
697 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
698 " result %x\n", scmd
,
700 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
702 rtn
= scsi_decide_disposition(scmd
);
705 * if the result was normal, then just pass it along to the
709 /* we don't want this command reissued, just
710 * finished with the sense data, so set
711 * retries to the max allowed to ensure it
712 * won't get reissued */
713 scmd
->retries
= scmd
->allowed
;
714 else if (rtn
!= NEEDS_RETRY
)
717 scsi_eh_finish_cmd(scmd
, done_q
);
720 return list_empty(work_q
);
724 * scsi_try_to_abort_cmd - Ask host to abort a running command.
725 * @scmd: SCSI cmd to abort from Lower Level.
728 * This function will not return until the user's completion function
729 * has been called. there is no timeout on this operation. if the
730 * author of the low-level driver wishes this operation to be timed,
731 * they can provide this facility themselves. helper functions in
732 * scsi_error.c can be supplied to make this easier to do.
734 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
736 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
740 * scsi_done was called just after the command timed out and before
741 * we had a chance to process it. (db)
743 if (scmd
->serial_number
== 0)
746 scmd
->owner
= SCSI_OWNER_LOWLEVEL
;
748 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
752 * scsi_eh_tur - Send TUR to device.
753 * @scmd: Scsi cmd to send TUR
756 * 0 - Device is ready. 1 - Device NOT ready.
758 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
760 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
761 int retry_cnt
= 1, rtn
;
765 memcpy(scmd
->cmnd
, tur_command
, sizeof(tur_command
));
768 * zero the sense buffer. the scsi spec mandates that any
769 * untransferred sense data should be interpreted as being zero.
771 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
773 saved_result
= scmd
->result
;
774 scmd
->request_buffer
= NULL
;
775 scmd
->request_bufflen
= 0;
777 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
779 scmd
->sc_data_direction
= DMA_NONE
;
781 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
784 * when we eventually call scsi_finish, we really wish to complete
785 * the original request, so let's restore the original data. (db)
787 scsi_setup_cmd_retry(scmd
);
788 scmd
->result
= saved_result
;
791 * hey, we are done. let's look to see what happened.
793 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
794 __FUNCTION__
, scmd
, rtn
));
797 else if (rtn
== NEEDS_RETRY
)
804 * scsi_eh_abort_cmds - abort canceled commands.
805 * @shost: scsi host being recovered.
806 * @eh_done_q: list_head for processed commands.
809 * Try and see whether or not it makes sense to try and abort the
810 * running command. this only works out to be the case if we have one
811 * command that has timed out. if the command simply failed, it makes
812 * no sense to try and abort the command, since as far as the shost
813 * adapter is concerned, it isn't running.
815 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
816 struct list_head
*done_q
)
818 struct list_head
*lh
, *lh_sf
;
819 struct scsi_cmnd
*scmd
;
822 list_for_each_safe(lh
, lh_sf
, work_q
) {
823 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
824 if (!scsi_eh_eflags_chk(scmd
, SCSI_EH_CANCEL_CMD
))
826 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
827 "0x%p\n", current
->comm
,
829 rtn
= scsi_try_to_abort_cmd(scmd
);
830 if (rtn
== SUCCESS
) {
831 scsi_eh_eflags_clr(scmd
, SCSI_EH_CANCEL_CMD
);
832 if (!scsi_device_online(scmd
->device
) ||
833 !scsi_eh_tur(scmd
)) {
834 scsi_eh_finish_cmd(scmd
, done_q
);
838 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
845 return list_empty(work_q
);
849 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
850 * @scmd: SCSI cmd used to send BDR
853 * There is no timeout for this operation. if this operation is
854 * unreliable for a given host, then the host itself needs to put a
855 * timer on it, and set the host back to a consistent state prior to
858 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
862 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
865 scmd
->owner
= SCSI_OWNER_LOWLEVEL
;
867 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
869 if (rtn
== SUCCESS
) {
870 scmd
->device
->was_reset
= 1;
871 scmd
->device
->expecting_cc_ua
= 1;
878 * scsi_eh_try_stu - Send START_UNIT to device.
879 * @scmd: Scsi cmd to send START_UNIT
882 * 0 - Device is ready. 1 - Device NOT ready.
884 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
886 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
890 if (!scmd
->device
->allow_restart
)
893 memcpy(scmd
->cmnd
, stu_command
, sizeof(stu_command
));
896 * zero the sense buffer. the scsi spec mandates that any
897 * untransferred sense data should be interpreted as being zero.
899 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
901 saved_result
= scmd
->result
;
902 scmd
->request_buffer
= NULL
;
903 scmd
->request_bufflen
= 0;
905 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
907 scmd
->sc_data_direction
= DMA_NONE
;
909 rtn
= scsi_send_eh_cmnd(scmd
, START_UNIT_TIMEOUT
);
912 * when we eventually call scsi_finish, we really wish to complete
913 * the original request, so let's restore the original data. (db)
915 scsi_setup_cmd_retry(scmd
);
916 scmd
->result
= saved_result
;
919 * hey, we are done. let's look to see what happened.
921 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
922 __FUNCTION__
, scmd
, rtn
));
929 * scsi_eh_stu - send START_UNIT if needed
930 * @shost: scsi host being recovered.
931 * @eh_done_q: list_head for processed commands.
934 * If commands are failing due to not ready, initializing command required,
935 * try revalidating the device, which will end up sending a start unit.
937 static int scsi_eh_stu(struct Scsi_Host
*shost
,
938 struct list_head
*work_q
,
939 struct list_head
*done_q
)
941 struct list_head
*lh
, *lh_sf
;
942 struct scsi_cmnd
*scmd
, *stu_scmd
;
943 struct scsi_device
*sdev
;
945 shost_for_each_device(sdev
, shost
) {
947 list_for_each_entry(scmd
, work_q
, eh_entry
)
948 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
949 scsi_check_sense(scmd
) == FAILED
) {
957 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
958 " 0x%p\n", current
->comm
, sdev
));
960 if (!scsi_eh_try_stu(stu_scmd
)) {
961 if (!scsi_device_online(sdev
) ||
962 !scsi_eh_tur(stu_scmd
)) {
963 list_for_each_safe(lh
, lh_sf
, work_q
) {
964 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
965 if (scmd
->device
== sdev
)
966 scsi_eh_finish_cmd(scmd
, done_q
);
970 SCSI_LOG_ERROR_RECOVERY(3,
971 printk("%s: START_UNIT failed to sdev:"
972 " 0x%p\n", current
->comm
, sdev
));
976 return list_empty(work_q
);
981 * scsi_eh_bus_device_reset - send bdr if needed
982 * @shost: scsi host being recovered.
983 * @eh_done_q: list_head for processed commands.
986 * Try a bus device reset. still, look to see whether we have multiple
987 * devices that are jammed or not - if we have multiple devices, it
988 * makes no sense to try bus_device_reset - we really would need to try
989 * a bus_reset instead.
991 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
992 struct list_head
*work_q
,
993 struct list_head
*done_q
)
995 struct list_head
*lh
, *lh_sf
;
996 struct scsi_cmnd
*scmd
, *bdr_scmd
;
997 struct scsi_device
*sdev
;
1000 shost_for_each_device(sdev
, shost
) {
1002 list_for_each_entry(scmd
, work_q
, eh_entry
)
1003 if (scmd
->device
== sdev
) {
1011 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1012 " 0x%p\n", current
->comm
,
1014 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1015 if (rtn
== SUCCESS
) {
1016 if (!scsi_device_online(sdev
) ||
1017 !scsi_eh_tur(bdr_scmd
)) {
1018 list_for_each_safe(lh
, lh_sf
,
1020 scmd
= list_entry(lh
, struct
1023 if (scmd
->device
== sdev
)
1024 scsi_eh_finish_cmd(scmd
,
1029 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1037 return list_empty(work_q
);
1041 * scsi_try_bus_reset - ask host to perform a bus reset
1042 * @scmd: SCSI cmd to send bus reset.
1044 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
1046 unsigned long flags
;
1049 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1051 scmd
->owner
= SCSI_OWNER_LOWLEVEL
;
1053 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
1056 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1057 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
1058 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1060 if (rtn
== SUCCESS
) {
1061 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1062 ssleep(BUS_RESET_SETTLE_TIME
);
1063 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1064 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1065 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1072 * scsi_try_host_reset - ask host adapter to reset itself
1073 * @scmd: SCSI cmd to send hsot reset.
1075 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
1077 unsigned long flags
;
1080 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1082 scmd
->owner
= SCSI_OWNER_LOWLEVEL
;
1084 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
1087 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1088 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
1089 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1091 if (rtn
== SUCCESS
) {
1092 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1093 ssleep(HOST_RESET_SETTLE_TIME
);
1094 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1095 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1096 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1103 * scsi_eh_bus_reset - send a bus reset
1104 * @shost: scsi host being recovered.
1105 * @eh_done_q: list_head for processed commands.
1107 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1108 struct list_head
*work_q
,
1109 struct list_head
*done_q
)
1111 struct list_head
*lh
, *lh_sf
;
1112 struct scsi_cmnd
*scmd
;
1113 struct scsi_cmnd
*chan_scmd
;
1114 unsigned int channel
;
1118 * we really want to loop over the various channels, and do this on
1119 * a channel by channel basis. we should also check to see if any
1120 * of the failed commands are on soft_reset devices, and if so, skip
1124 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1126 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1127 if (channel
== scmd
->device
->channel
) {
1131 * FIXME add back in some support for
1132 * soft_reset devices.
1139 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1140 " %d\n", current
->comm
,
1142 rtn
= scsi_try_bus_reset(chan_scmd
);
1143 if (rtn
== SUCCESS
) {
1144 list_for_each_safe(lh
, lh_sf
, work_q
) {
1145 scmd
= list_entry(lh
, struct scsi_cmnd
,
1147 if (channel
== scmd
->device
->channel
)
1148 if (!scsi_device_online(scmd
->device
) ||
1150 scsi_eh_finish_cmd(scmd
,
1154 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1155 " failed chan: %d\n",
1160 return list_empty(work_q
);
1164 * scsi_eh_host_reset - send a host reset
1165 * @work_q: list_head for processed commands.
1166 * @done_q: list_head for processed commands.
1168 static int scsi_eh_host_reset(struct list_head
*work_q
,
1169 struct list_head
*done_q
)
1172 struct list_head
*lh
, *lh_sf
;
1173 struct scsi_cmnd
*scmd
;
1175 if (!list_empty(work_q
)) {
1176 scmd
= list_entry(work_q
->next
,
1177 struct scsi_cmnd
, eh_entry
);
1179 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1182 rtn
= scsi_try_host_reset(scmd
);
1183 if (rtn
== SUCCESS
) {
1184 list_for_each_safe(lh
, lh_sf
, work_q
) {
1185 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
1186 if (!scsi_device_online(scmd
->device
) ||
1187 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1189 scsi_eh_finish_cmd(scmd
, done_q
);
1192 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1197 return list_empty(work_q
);
1201 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1202 * @work_q: list_head for processed commands.
1203 * @done_q: list_head for processed commands.
1206 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1207 struct list_head
*done_q
)
1209 struct list_head
*lh
, *lh_sf
;
1210 struct scsi_cmnd
*scmd
;
1212 list_for_each_safe(lh
, lh_sf
, work_q
) {
1213 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
1214 printk(KERN_INFO
"scsi: Device offlined - not"
1215 " ready after error recovery: host"
1216 " %d channel %d id %d lun %d\n",
1217 scmd
->device
->host
->host_no
,
1218 scmd
->device
->channel
,
1221 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1222 if (scsi_eh_eflags_chk(scmd
, SCSI_EH_CANCEL_CMD
)) {
1224 * FIXME: Handle lost cmds.
1227 scsi_eh_finish_cmd(scmd
, done_q
);
1233 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1234 * @scmd: SCSI cmd to examine.
1237 * This is *only* called when we are examining the status after sending
1238 * out the actual data command. any commands that are queued for error
1239 * recovery (e.g. test_unit_ready) do *not* come through here.
1241 * When this routine returns failed, it means the error handler thread
1242 * is woken. In cases where the error code indicates an error that
1243 * doesn't require the error handler read (i.e. we don't need to
1244 * abort/reset), this function should return SUCCESS.
1246 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1251 * if the device is offline, then we clearly just pass the result back
1252 * up to the top level.
1254 if (!scsi_device_online(scmd
->device
)) {
1255 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1262 * first check the host byte, to see if there is anything in there
1263 * that would indicate what we need to do.
1265 switch (host_byte(scmd
->result
)) {
1266 case DID_PASSTHROUGH
:
1268 * no matter what, pass this through to the upper layer.
1269 * nuke this special code so that it looks like we are saying
1272 scmd
->result
&= 0xff00ffff;
1276 * looks good. drop through, and check the next byte.
1279 case DID_NO_CONNECT
:
1280 case DID_BAD_TARGET
:
1283 * note - this means that we just report the status back
1284 * to the top level driver, not that we actually think
1285 * that it indicates SUCCESS.
1289 * when the low level driver returns did_soft_error,
1290 * it is responsible for keeping an internal retry counter
1291 * in order to avoid endless loops (db)
1293 * actually this is a bug in this function here. we should
1294 * be mindful of the maximum number of retries specified
1295 * and not get stuck in a loop.
1297 case DID_SOFT_ERROR
:
1303 return ADD_TO_MLQUEUE
;
1306 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1307 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1309 * execute reservation conflict processing code
1320 * when we scan the bus, we get timeout messages for
1321 * these commands if there is no device available.
1322 * other hosts report did_no_connect for the same thing.
1324 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1325 scmd
->cmnd
[0] == INQUIRY
)) {
1337 * next, check the message byte.
1339 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1343 * check the status byte to see if this indicates anything special.
1345 switch (status_byte(scmd
->result
)) {
1348 * the case of trying to send too many commands to a
1349 * tagged queueing device.
1353 * device can't talk to us at the moment. Should only
1354 * occur (SAM-3) when the task queue is empty, so will cause
1355 * the empty queue handling to trigger a stall in the
1358 return ADD_TO_MLQUEUE
;
1360 case COMMAND_TERMINATED
:
1363 case CHECK_CONDITION
:
1364 rtn
= scsi_check_sense(scmd
);
1365 if (rtn
== NEEDS_RETRY
)
1367 /* if rtn == FAILED, we have no sense information;
1368 * returning FAILED will wake the error handler thread
1369 * to collect the sense and redo the decide
1372 case CONDITION_GOOD
:
1373 case INTERMEDIATE_GOOD
:
1374 case INTERMEDIATE_C_GOOD
:
1377 * who knows? FIXME(eric)
1381 case RESERVATION_CONFLICT
:
1382 printk(KERN_INFO
"scsi: reservation conflict: host"
1383 " %d channel %d id %d lun %d\n",
1384 scmd
->device
->host
->host_no
, scmd
->device
->channel
,
1385 scmd
->device
->id
, scmd
->device
->lun
);
1386 return SUCCESS
; /* causes immediate i/o error */
1394 /* we requeue for retry because the error was retryable, and
1395 * the request was not marked fast fail. Note that above,
1396 * even if the request is marked fast fail, we still requeue
1397 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1398 if ((++scmd
->retries
) < scmd
->allowed
1399 && !blk_noretry_request(scmd
->request
)) {
1403 * no more retries - report this one back to upper level.
1410 * scsi_eh_lock_done - done function for eh door lock request
1411 * @scmd: SCSI command block for the door lock request
1414 * We completed the asynchronous door lock request, and it has either
1415 * locked the door or failed. We must free the command structures
1416 * associated with this request.
1418 static void scsi_eh_lock_done(struct scsi_cmnd
*scmd
)
1420 struct scsi_request
*sreq
= scmd
->sc_request
;
1422 scsi_release_request(sreq
);
1427 * scsi_eh_lock_door - Prevent medium removal for the specified device
1428 * @sdev: SCSI device to prevent medium removal
1431 * We must be called from process context; scsi_allocate_request()
1435 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1436 * head of the devices request queue, and continue.
1439 * scsi_allocate_request() may sleep waiting for existing requests to
1440 * be processed. However, since we haven't kicked off any request
1441 * processing for this host, this may deadlock.
1443 * If scsi_allocate_request() fails for what ever reason, we
1444 * completely forget to lock the door.
1446 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1448 struct scsi_request
*sreq
= scsi_allocate_request(sdev
, GFP_KERNEL
);
1450 if (unlikely(!sreq
)) {
1451 printk(KERN_ERR
"%s: request allocate failed,"
1452 "prevent media removal cmd not sent\n", __FUNCTION__
);
1456 sreq
->sr_cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1457 sreq
->sr_cmnd
[1] = 0;
1458 sreq
->sr_cmnd
[2] = 0;
1459 sreq
->sr_cmnd
[3] = 0;
1460 sreq
->sr_cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1461 sreq
->sr_cmnd
[5] = 0;
1462 sreq
->sr_data_direction
= DMA_NONE
;
1463 sreq
->sr_bufflen
= 0;
1464 sreq
->sr_buffer
= NULL
;
1465 sreq
->sr_allowed
= 5;
1466 sreq
->sr_done
= scsi_eh_lock_done
;
1467 sreq
->sr_timeout_per_command
= 10 * HZ
;
1468 sreq
->sr_cmd_len
= COMMAND_SIZE(sreq
->sr_cmnd
[0]);
1470 scsi_insert_special_req(sreq
, 1);
1475 * scsi_restart_operations - restart io operations to the specified host.
1476 * @shost: Host we are restarting.
1479 * When we entered the error handler, we blocked all further i/o to
1480 * this device. we need to 'reverse' this process.
1482 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1484 struct scsi_device
*sdev
;
1487 * If the door was locked, we need to insert a door lock request
1488 * onto the head of the SCSI request queue for the device. There
1489 * is no point trying to lock the door of an off-line device.
1491 shost_for_each_device(sdev
, shost
) {
1492 if (scsi_device_online(sdev
) && sdev
->locked
)
1493 scsi_eh_lock_door(sdev
);
1497 * next free up anything directly waiting upon the host. this
1498 * will be requests for character device operations, and also for
1499 * ioctls to queued block devices.
1501 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1504 clear_bit(SHOST_RECOVERY
, &shost
->shost_state
);
1506 wake_up(&shost
->host_wait
);
1509 * finally we need to re-initiate requests that may be pending. we will
1510 * have had everything blocked while error handling is taking place, and
1511 * now that error recovery is done, we will need to ensure that these
1512 * requests are started.
1514 scsi_run_host_queues(shost
);
1518 * scsi_eh_ready_devs - check device ready state and recover if not.
1519 * @shost: host to be recovered.
1520 * @eh_done_q: list_head for processed commands.
1523 static void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1524 struct list_head
*work_q
,
1525 struct list_head
*done_q
)
1527 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1528 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1529 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1530 if (!scsi_eh_host_reset(work_q
, done_q
))
1531 scsi_eh_offline_sdevs(work_q
, done_q
);
1535 * scsi_eh_flush_done_q - finish processed commands or retry them.
1536 * @done_q: list_head of processed commands.
1539 static void scsi_eh_flush_done_q(struct list_head
*done_q
)
1541 struct list_head
*lh
, *lh_sf
;
1542 struct scsi_cmnd
*scmd
;
1544 list_for_each_safe(lh
, lh_sf
, done_q
) {
1545 scmd
= list_entry(lh
, struct scsi_cmnd
, eh_entry
);
1547 if (scsi_device_online(scmd
->device
) &&
1548 !blk_noretry_request(scmd
->request
) &&
1549 (++scmd
->retries
< scmd
->allowed
)) {
1550 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1554 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1557 * If just we got sense for the device (called
1558 * scsi_eh_get_sense), scmd->result is already
1559 * set, do not set DRIVER_TIMEOUT.
1562 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1563 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1565 current
->comm
, scmd
));
1566 scsi_finish_command(scmd
);
1572 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1573 * @shost: Host to unjam.
1576 * When we come in here, we *know* that all commands on the bus have
1577 * either completed, failed or timed out. we also know that no further
1578 * commands are being sent to the host, so things are relatively quiet
1579 * and we have freedom to fiddle with things as we wish.
1581 * This is only the *default* implementation. it is possible for
1582 * individual drivers to supply their own version of this function, and
1583 * if the maintainer wishes to do this, it is strongly suggested that
1584 * this function be taken as a template and modified. this function
1585 * was designed to correctly handle problems for about 95% of the
1586 * different cases out there, and it should always provide at least a
1587 * reasonable amount of error recovery.
1589 * Any command marked 'failed' or 'timeout' must eventually have
1590 * scsi_finish_cmd() called for it. we do all of the retry stuff
1591 * here, so when we restart the host after we return it should have an
1594 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1596 unsigned long flags
;
1597 LIST_HEAD(eh_work_q
);
1598 LIST_HEAD(eh_done_q
);
1600 spin_lock_irqsave(shost
->host_lock
, flags
);
1601 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1602 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1604 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1606 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1607 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1608 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1610 scsi_eh_flush_done_q(&eh_done_q
);
1614 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1615 * @data: Host for which we are running.
1618 * This is always run in the context of a kernel thread. The idea is
1619 * that we start this thing up when the kernel starts up (one per host
1620 * that we detect), and it immediately goes to sleep and waits for some
1621 * event (i.e. failure). When this takes place, we have the job of
1622 * trying to unjam the bus and restarting things.
1624 int scsi_error_handler(void *data
)
1626 struct Scsi_Host
*shost
= (struct Scsi_Host
*) data
;
1628 DECLARE_MUTEX_LOCKED(sem
);
1634 daemonize("scsi_eh_%d", shost
->host_no
);
1636 current
->flags
|= PF_NOFREEZE
;
1638 shost
->eh_wait
= &sem
;
1639 shost
->ehandler
= current
;
1642 * Wake up the thread that created us.
1644 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1645 " scsi_eh_%d\n",shost
->host_no
));
1647 complete(shost
->eh_notify
);
1651 * If we get a signal, it means we are supposed to go
1652 * away and die. This typically happens if the user is
1653 * trying to unload a module.
1655 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1657 " sleeping\n",shost
->host_no
));
1660 * Note - we always use down_interruptible with the semaphore
1661 * even if the module was loaded as part of the kernel. The
1662 * reason is that down() will cause this thread to be counted
1663 * in the load average as a running process, and down
1664 * interruptible doesn't. Given that we need to allow this
1665 * thread to die if the driver was loaded as a module, using
1666 * semaphores isn't unreasonable.
1668 down_interruptible(&sem
);
1672 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1673 " scsi_eh_%d waking"
1674 " up\n",shost
->host_no
));
1676 shost
->eh_active
= 1;
1679 * We have a host that is failing for some reason. Figure out
1680 * what we need to do to get it up and online again (if we can).
1681 * If we fail, we end up taking the thing offline.
1683 if (shost
->hostt
->eh_strategy_handler
)
1684 rtn
= shost
->hostt
->eh_strategy_handler(shost
);
1686 scsi_unjam_host(shost
);
1688 shost
->eh_active
= 0;
1691 * Note - if the above fails completely, the action is to take
1692 * individual devices offline and flush the queue of any
1693 * outstanding requests that may have been pending. When we
1694 * restart, we restart any I/O to any other devices on the bus
1695 * which are still online.
1697 scsi_restart_operations(shost
);
1701 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1702 " exiting\n",shost
->host_no
));
1705 * Make sure that nobody tries to wake us up again.
1707 shost
->eh_wait
= NULL
;
1710 * Knock this down too. From this point on, the host is flying
1711 * without a pilot. If this is because the module is being unloaded,
1712 * that's fine. If the user sent a signal to this thing, we are
1713 * potentially in real danger.
1715 shost
->eh_active
= 0;
1716 shost
->ehandler
= NULL
;
1719 * If anyone is waiting for us to exit (i.e. someone trying to unload
1720 * a driver), then wake up that process to let them know we are on
1721 * the way out the door.
1723 complete_and_exit(shost
->eh_notify
, 0);
1728 * Function: scsi_report_bus_reset()
1730 * Purpose: Utility function used by low-level drivers to report that
1731 * they have observed a bus reset on the bus being handled.
1733 * Arguments: shost - Host in question
1734 * channel - channel on which reset was observed.
1738 * Lock status: Host lock must be held.
1740 * Notes: This only needs to be called if the reset is one which
1741 * originates from an unknown location. Resets originated
1742 * by the mid-level itself don't need to call this, but there
1743 * should be no harm.
1745 * The main purpose of this is to make sure that a CHECK_CONDITION
1746 * is properly treated.
1748 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1750 struct scsi_device
*sdev
;
1752 __shost_for_each_device(sdev
, shost
) {
1753 if (channel
== sdev
->channel
) {
1754 sdev
->was_reset
= 1;
1755 sdev
->expecting_cc_ua
= 1;
1759 EXPORT_SYMBOL(scsi_report_bus_reset
);
1762 * Function: scsi_report_device_reset()
1764 * Purpose: Utility function used by low-level drivers to report that
1765 * they have observed a device reset on the device being handled.
1767 * Arguments: shost - Host in question
1768 * channel - channel on which reset was observed
1769 * target - target on which reset was observed
1773 * Lock status: Host lock must be held
1775 * Notes: This only needs to be called if the reset is one which
1776 * originates from an unknown location. Resets originated
1777 * by the mid-level itself don't need to call this, but there
1778 * should be no harm.
1780 * The main purpose of this is to make sure that a CHECK_CONDITION
1781 * is properly treated.
1783 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1785 struct scsi_device
*sdev
;
1787 __shost_for_each_device(sdev
, shost
) {
1788 if (channel
== sdev
->channel
&&
1789 target
== sdev
->id
) {
1790 sdev
->was_reset
= 1;
1791 sdev
->expecting_cc_ua
= 1;
1795 EXPORT_SYMBOL(scsi_report_device_reset
);
1798 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1803 * Function: scsi_reset_provider
1805 * Purpose: Send requested reset to a bus or device at any phase.
1807 * Arguments: device - device to send reset to
1808 * flag - reset type (see scsi.h)
1810 * Returns: SUCCESS/FAILURE.
1812 * Notes: This is used by the SCSI Generic driver to provide
1813 * Bus/Device reset capability.
1816 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1818 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1822 scmd
->request
= &req
;
1823 memset(&scmd
->eh_timeout
, 0, sizeof(scmd
->eh_timeout
));
1824 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
1825 scmd
->state
= SCSI_STATE_INITIALIZING
;
1826 scmd
->owner
= SCSI_OWNER_MIDLEVEL
;
1828 memset(&scmd
->cmnd
, '\0', sizeof(scmd
->cmnd
));
1830 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1832 scmd
->buffer
= NULL
;
1834 scmd
->request_buffer
= NULL
;
1835 scmd
->request_bufflen
= 0;
1836 scmd
->abort_reason
= DID_ABORT
;
1840 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1841 scmd
->sc_request
= NULL
;
1842 scmd
->sc_magic
= SCSI_CMND_MAGIC
;
1844 init_timer(&scmd
->eh_timeout
);
1847 * Sometimes the command can get back into the timer chain,
1848 * so use the pid as an identifier.
1853 case SCSI_TRY_RESET_DEVICE
:
1854 rtn
= scsi_try_bus_device_reset(scmd
);
1858 case SCSI_TRY_RESET_BUS
:
1859 rtn
= scsi_try_bus_reset(scmd
);
1863 case SCSI_TRY_RESET_HOST
:
1864 rtn
= scsi_try_host_reset(scmd
);
1870 scsi_next_command(scmd
);
1873 EXPORT_SYMBOL(scsi_reset_provider
);
1876 * scsi_normalize_sense - normalize main elements from either fixed or
1877 * descriptor sense data format into a common format.
1879 * @sense_buffer: byte array containing sense data returned by device
1880 * @sb_len: number of valid bytes in sense_buffer
1881 * @sshdr: pointer to instance of structure that common
1882 * elements are written to.
1885 * The "main elements" from sense data are: response_code, sense_key,
1886 * asc, ascq and additional_length (only for descriptor format).
1888 * Typically this function can be called after a device has
1889 * responded to a SCSI command with the CHECK_CONDITION status.
1892 * 1 if valid sense data information found, else 0;
1894 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1895 struct scsi_sense_hdr
*sshdr
)
1897 if (!sense_buffer
|| !sb_len
|| (sense_buffer
[0] & 0x70) != 0x70)
1900 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1902 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1903 if (sshdr
->response_code
>= 0x72) {
1908 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1910 sshdr
->asc
= sense_buffer
[2];
1912 sshdr
->ascq
= sense_buffer
[3];
1914 sshdr
->additional_length
= sense_buffer
[7];
1920 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1922 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1923 sb_len
: (sense_buffer
[7] + 8);
1925 sshdr
->asc
= sense_buffer
[12];
1927 sshdr
->ascq
= sense_buffer
[13];
1933 EXPORT_SYMBOL(scsi_normalize_sense
);
1935 int scsi_request_normalize_sense(struct scsi_request
*sreq
,
1936 struct scsi_sense_hdr
*sshdr
)
1938 return scsi_normalize_sense(sreq
->sr_sense_buffer
,
1939 sizeof(sreq
->sr_sense_buffer
), sshdr
);
1941 EXPORT_SYMBOL(scsi_request_normalize_sense
);
1943 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1944 struct scsi_sense_hdr
*sshdr
)
1946 return scsi_normalize_sense(cmd
->sense_buffer
,
1947 sizeof(cmd
->sense_buffer
), sshdr
);
1949 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1952 * scsi_sense_desc_find - search for a given descriptor type in
1953 * descriptor sense data format.
1955 * @sense_buffer: byte array of descriptor format sense data
1956 * @sb_len: number of valid bytes in sense_buffer
1957 * @desc_type: value of descriptor type to find
1958 * (e.g. 0 -> information)
1961 * only valid when sense data is in descriptor format
1964 * pointer to start of (first) descriptor if found else NULL
1966 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1969 int add_sen_len
, add_len
, desc_len
, k
;
1972 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1974 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1976 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1977 add_sen_len
: (sb_len
- 8);
1978 descp
= &sense_buffer
[8];
1979 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1981 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1982 desc_len
= add_len
+ 2;
1983 if (descp
[0] == desc_type
)
1985 if (add_len
< 0) // short descriptor ??
1990 EXPORT_SYMBOL(scsi_sense_desc_find
);
1993 * scsi_get_sense_info_fld - attempts to get information field from
1994 * sense data (either fixed or descriptor format)
1996 * @sense_buffer: byte array of sense data
1997 * @sb_len: number of valid bytes in sense_buffer
1998 * @info_out: pointer to 64 integer where 8 or 4 byte information
1999 * field will be placed if found.
2002 * 1 if information field found, 0 if not found.
2004 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2013 switch (sense_buffer
[0] & 0x7f) {
2016 if (sense_buffer
[0] & 0x80) {
2017 *info_out
= (sense_buffer
[3] << 24) +
2018 (sense_buffer
[4] << 16) +
2019 (sense_buffer
[5] << 8) + sense_buffer
[6];
2025 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2027 if (ucp
&& (0xa == ucp
[1])) {
2029 for (j
= 0; j
< 8; ++j
) {
2042 EXPORT_SYMBOL(scsi_get_sense_info_fld
);