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 scmd
->eh_eflags
|= eh_flag
;
77 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
78 scsi_host_set_state(shost
, SHOST_RECOVERY
);
80 scsi_eh_wakeup(shost
);
81 spin_unlock_irqrestore(shost
->host_lock
, flags
);
86 * scsi_add_timer - Start timeout timer for a single scsi command.
87 * @scmd: scsi command that is about to start running.
88 * @timeout: amount of time to allow this command to run.
89 * @complete: timeout function to call if timer isn't canceled.
92 * This should be turned into an inline function. Each scsi command
93 * has its own timer, and as it is added to the queue, we set up the
94 * timer. When the command completes, we cancel the timer.
96 void scsi_add_timer(struct scsi_cmnd
*scmd
, int timeout
,
97 void (*complete
)(struct scsi_cmnd
*))
101 * If the clock was already running for this command, then
102 * first delete the timer. The timer handling code gets rather
103 * confused if we don't do this.
105 if (scmd
->eh_timeout
.function
)
106 del_timer(&scmd
->eh_timeout
);
108 scmd
->eh_timeout
.data
= (unsigned long)scmd
;
109 scmd
->eh_timeout
.expires
= jiffies
+ timeout
;
110 scmd
->eh_timeout
.function
= (void (*)(unsigned long)) complete
;
112 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
113 " %d, (%p)\n", __FUNCTION__
,
114 scmd
, timeout
, complete
));
116 add_timer(&scmd
->eh_timeout
);
118 EXPORT_SYMBOL(scsi_add_timer
);
121 * scsi_delete_timer - Delete/cancel timer for a given function.
122 * @scmd: Cmd that we are canceling timer for
125 * This should be turned into an inline function.
128 * 1 if we were able to detach the timer. 0 if we blew it, and the
129 * timer function has already started to run.
131 int scsi_delete_timer(struct scsi_cmnd
*scmd
)
135 rtn
= del_timer(&scmd
->eh_timeout
);
137 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
138 " rtn: %d\n", __FUNCTION__
,
141 scmd
->eh_timeout
.data
= (unsigned long)NULL
;
142 scmd
->eh_timeout
.function
= NULL
;
146 EXPORT_SYMBOL(scsi_delete_timer
);
149 * scsi_times_out - Timeout function for normal scsi commands.
150 * @scmd: Cmd that is timing out.
153 * We do not need to lock this. There is the potential for a race
154 * only in that the normal completion handling might run, but if the
155 * normal completion function determines that the timer has already
156 * fired, then it mustn't do anything.
158 void scsi_times_out(struct scsi_cmnd
*scmd
)
160 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
162 if (scmd
->device
->host
->hostt
->eh_timed_out
)
163 switch (scmd
->device
->host
->hostt
->eh_timed_out(scmd
)) {
168 /* This allows a single retry even of a command
169 * with allowed == 0 */
170 if (scmd
->retries
++ > scmd
->allowed
)
172 scsi_add_timer(scmd
, scmd
->timeout_per_command
,
179 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
180 panic("Error handler thread not present at %p %p %s %d",
181 scmd
, scmd
->device
->host
, __FILE__
, __LINE__
);
186 * scsi_block_when_processing_errors - Prevent cmds from being queued.
187 * @sdev: Device on which we are performing recovery.
190 * We block until the host is out of error recovery, and then check to
191 * see whether the host or the device is offline.
194 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
196 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
200 wait_event(sdev
->host
->host_wait
, (sdev
->host
->shost_state
!=
203 online
= scsi_device_online(sdev
);
205 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__
,
210 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
212 #ifdef CONFIG_SCSI_LOGGING
214 * scsi_eh_prt_fail_stats - Log info on failures.
215 * @shost: scsi host being recovered.
216 * @work_q: Queue of scsi cmds to process.
218 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
219 struct list_head
*work_q
)
221 struct scsi_cmnd
*scmd
;
222 struct scsi_device
*sdev
;
223 int total_failures
= 0;
226 int devices_failed
= 0;
228 shost_for_each_device(sdev
, shost
) {
229 list_for_each_entry(scmd
, work_q
, eh_entry
) {
230 if (scmd
->device
== sdev
) {
232 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
239 if (cmd_cancel
|| cmd_failed
) {
240 SCSI_LOG_ERROR_RECOVERY(3,
241 printk("%s: %d:%d:%d:%d cmds failed: %d,"
243 __FUNCTION__
, shost
->host_no
,
244 sdev
->channel
, sdev
->id
, sdev
->lun
,
245 cmd_failed
, cmd_cancel
));
252 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
253 " devices require eh work\n",
254 total_failures
, devices_failed
));
259 * scsi_check_sense - Examine scsi cmd sense
260 * @scmd: Cmd to have sense checked.
263 * SUCCESS or FAILED or NEEDS_RETRY
266 * When a deferred error is detected the current command has
267 * not been executed and needs retrying.
269 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
271 struct scsi_sense_hdr sshdr
;
273 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
274 return FAILED
; /* no valid sense data */
276 if (scsi_sense_is_deferred(&sshdr
))
280 * Previous logic looked for FILEMARK, EOM or ILI which are
281 * mainly associated with tapes and returned SUCCESS.
283 if (sshdr
.response_code
== 0x70) {
285 if (scmd
->sense_buffer
[2] & 0xe0)
289 * descriptor format: look for "stream commands sense data
290 * descriptor" (see SSC-3). Assume single sense data
291 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
293 if ((sshdr
.additional_length
> 3) &&
294 (scmd
->sense_buffer
[8] == 0x4) &&
295 (scmd
->sense_buffer
[11] & 0xe0))
299 switch (sshdr
.sense_key
) {
302 case RECOVERED_ERROR
:
303 return /* soft_error */ SUCCESS
;
305 case ABORTED_COMMAND
:
310 * if we are expecting a cc/ua because of a bus reset that we
311 * performed, treat this just as a retry. otherwise this is
312 * information that we should pass up to the upper-level driver
313 * so that we can deal with it there.
315 if (scmd
->device
->expecting_cc_ua
) {
316 scmd
->device
->expecting_cc_ua
= 0;
320 * if the device is in the process of becoming ready, we
323 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
326 * if the device is not started, we need to wake
327 * the error handler to start the motor
329 if (scmd
->device
->allow_restart
&&
330 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
334 /* these three are not supported */
336 case VOLUME_OVERFLOW
:
344 if (scmd
->device
->retry_hwerror
)
349 case ILLEGAL_REQUEST
:
358 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
359 * @scmd: SCSI cmd to examine.
362 * This is *only* called when we are examining the status of commands
363 * queued during error recovery. the main difference here is that we
364 * don't allow for the possibility of retries here, and we are a lot
365 * more restrictive about what we consider acceptable.
367 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
370 * first check the host byte, to see if there is anything in there
371 * that would indicate what we need to do.
373 if (host_byte(scmd
->result
) == DID_RESET
) {
375 * rats. we are already in the error handler, so we now
376 * get to try and figure out what to do next. if the sense
377 * is valid, we have a pretty good idea of what to do.
378 * if not, we mark it as FAILED.
380 return scsi_check_sense(scmd
);
382 if (host_byte(scmd
->result
) != DID_OK
)
386 * next, check the message byte.
388 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
392 * now, check the status byte to see if this indicates
395 switch (status_byte(scmd
->result
)) {
397 case COMMAND_TERMINATED
:
399 case CHECK_CONDITION
:
400 return scsi_check_sense(scmd
);
402 case INTERMEDIATE_GOOD
:
403 case INTERMEDIATE_C_GOOD
:
405 * who knows? FIXME(eric)
410 case RESERVATION_CONFLICT
:
418 * scsi_eh_times_out - timeout function for error handling.
419 * @scmd: Cmd that is timing out.
422 * During error handling, the kernel thread will be sleeping waiting
423 * for some action to complete on the device. our only job is to
424 * record that it timed out, and to wake up the thread.
426 static void scsi_eh_times_out(struct scsi_cmnd
*scmd
)
428 scmd
->eh_eflags
|= SCSI_EH_REC_TIMEOUT
;
429 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd:%p\n", __FUNCTION__
,
432 up(scmd
->device
->host
->eh_action
);
436 * scsi_eh_done - Completion function for error handling.
437 * @scmd: Cmd that is done.
439 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
442 * if the timeout handler is already running, then just set the
443 * flag which says we finished late, and return. we have no
444 * way of stopping the timeout handler from running, so we must
445 * always defer to it.
447 if (del_timer(&scmd
->eh_timeout
)) {
448 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
450 SCSI_LOG_ERROR_RECOVERY(3, printk("%s scmd: %p result: %x\n",
451 __FUNCTION__
, scmd
, scmd
->result
));
453 up(scmd
->device
->host
->eh_action
);
458 * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery.
459 * @scmd: SCSI Cmd to send.
460 * @timeout: Timeout for cmd.
463 * The initialization of the structures is quite a bit different in
464 * this case, and furthermore, there is a different completion handler
465 * vs scsi_dispatch_cmd.
467 * SUCCESS or FAILED or NEEDS_RETRY
469 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, int timeout
)
471 struct scsi_device
*sdev
= scmd
->device
;
472 struct Scsi_Host
*shost
= sdev
->host
;
473 DECLARE_MUTEX_LOCKED(sem
);
478 * we will use a queued command if possible, otherwise we will
479 * emulate the queuing and calling of completion function ourselves.
481 if (sdev
->scsi_level
<= SCSI_2
)
482 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
483 (sdev
->lun
<< 5 & 0xe0);
485 scsi_add_timer(scmd
, timeout
, scsi_eh_times_out
);
488 * set up the semaphore so we wait for the command to complete.
490 shost
->eh_action
= &sem
;
491 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
493 spin_lock_irqsave(shost
->host_lock
, flags
);
495 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
496 spin_unlock_irqrestore(shost
->host_lock
, flags
);
499 scsi_log_completion(scmd
, SUCCESS
);
501 shost
->eh_action
= NULL
;
504 * see if timeout. if so, tell the host to forget about it.
505 * in other words, we don't want a callback any more.
507 if (scmd
->eh_eflags
& SCSI_EH_REC_TIMEOUT
) {
508 scmd
->eh_eflags
&= ~SCSI_EH_REC_TIMEOUT
;
511 * as far as the low level driver is
512 * concerned, this command is still active, so
513 * we must give the low level driver a chance
516 * FIXME(eric) - we are not tracking whether we could
517 * abort a timed out command or not. not sure how
518 * we should treat them differently anyways.
520 if (shost
->hostt
->eh_abort_handler
)
521 shost
->hostt
->eh_abort_handler(scmd
);
523 scmd
->request
->rq_status
= RQ_SCSI_DONE
;
527 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd: %p, rtn:%x\n",
528 __FUNCTION__
, scmd
, rtn
));
531 * now examine the actual status codes to see whether the command
532 * actually did complete normally.
534 if (rtn
== SUCCESS
) {
535 rtn
= scsi_eh_completed_normally(scmd
);
536 SCSI_LOG_ERROR_RECOVERY(3,
537 printk("%s: scsi_eh_completed_normally %x\n",
554 * scsi_request_sense - Request sense data from a particular target.
555 * @scmd: SCSI cmd for request sense.
558 * Some hosts automatically obtain this information, others require
559 * that we obtain it on our own. This function will *not* return until
560 * the command either times out, or it completes.
562 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
564 static unsigned char generic_sense
[6] =
565 {REQUEST_SENSE
, 0, 0, 0, 252, 0};
566 unsigned char *scsi_result
;
570 memcpy(scmd
->cmnd
, generic_sense
, sizeof(generic_sense
));
572 scsi_result
= kmalloc(252, GFP_ATOMIC
| ((scmd
->device
->host
->hostt
->unchecked_isa_dma
) ? __GFP_DMA
: 0));
575 if (unlikely(!scsi_result
)) {
576 printk(KERN_ERR
"%s: cannot allocate scsi_result.\n",
582 * zero the sense buffer. some host adapters automatically always
583 * request sense, so it is not a good idea that
584 * scmd->request_buffer and scmd->sense_buffer point to the same
585 * address (db). 0 is not a valid sense code.
587 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
588 memset(scsi_result
, 0, 252);
590 saved_result
= scmd
->result
;
591 scmd
->request_buffer
= scsi_result
;
592 scmd
->request_bufflen
= 252;
594 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
595 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
598 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
600 /* last chance to have valid sense data */
601 if(!SCSI_SENSE_VALID(scmd
)) {
602 memcpy(scmd
->sense_buffer
, scmd
->request_buffer
,
603 sizeof(scmd
->sense_buffer
));
609 * when we eventually call scsi_finish, we really wish to complete
610 * the original request, so let's restore the original data. (db)
612 scsi_setup_cmd_retry(scmd
);
613 scmd
->result
= saved_result
;
618 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
619 * @scmd: Original SCSI cmd that eh has finished.
620 * @done_q: Queue for processed commands.
623 * We don't want to use the normal command completion while we are are
624 * still handling errors - it may cause other commands to be queued,
625 * and that would disturb what we are doing. thus we really want to
626 * keep a list of pending commands for final completion, and once we
627 * are ready to leave error handling we handle completion for real.
629 static void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
,
630 struct list_head
*done_q
)
632 scmd
->device
->host
->host_failed
--;
636 * set this back so that the upper level can correctly free up
639 scsi_setup_cmd_retry(scmd
);
640 list_move_tail(&scmd
->eh_entry
, done_q
);
644 * scsi_eh_get_sense - Get device sense data.
645 * @work_q: Queue of commands to process.
646 * @done_q: Queue of proccessed commands..
649 * See if we need to request sense information. if so, then get it
650 * now, so we have a better idea of what to do.
653 * This has the unfortunate side effect that if a shost adapter does
654 * not automatically request sense information, that we end up shutting
655 * it down before we request it.
657 * All drivers should request sense information internally these days,
658 * so for now all I have to say is tough noogies if you end up in here.
660 * XXX: Long term this code should go away, but that needs an audit of
663 static int scsi_eh_get_sense(struct list_head
*work_q
,
664 struct list_head
*done_q
)
666 struct scsi_cmnd
*scmd
, *next
;
669 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
670 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
671 SCSI_SENSE_VALID(scmd
))
674 SCSI_LOG_ERROR_RECOVERY(2, printk("%s: requesting sense"
678 rtn
= scsi_request_sense(scmd
);
682 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
683 " result %x\n", scmd
,
685 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
687 rtn
= scsi_decide_disposition(scmd
);
690 * if the result was normal, then just pass it along to the
694 /* we don't want this command reissued, just
695 * finished with the sense data, so set
696 * retries to the max allowed to ensure it
697 * won't get reissued */
698 scmd
->retries
= scmd
->allowed
;
699 else if (rtn
!= NEEDS_RETRY
)
702 scsi_eh_finish_cmd(scmd
, done_q
);
705 return list_empty(work_q
);
709 * scsi_try_to_abort_cmd - Ask host to abort a running command.
710 * @scmd: SCSI cmd to abort from Lower Level.
713 * This function will not return until the user's completion function
714 * has been called. there is no timeout on this operation. if the
715 * author of the low-level driver wishes this operation to be timed,
716 * they can provide this facility themselves. helper functions in
717 * scsi_error.c can be supplied to make this easier to do.
719 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
721 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
725 * scsi_done was called just after the command timed out and before
726 * we had a chance to process it. (db)
728 if (scmd
->serial_number
== 0)
730 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
734 * scsi_eh_tur - Send TUR to device.
735 * @scmd: Scsi cmd to send TUR
738 * 0 - Device is ready. 1 - Device NOT ready.
740 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
742 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
743 int retry_cnt
= 1, rtn
;
747 memcpy(scmd
->cmnd
, tur_command
, sizeof(tur_command
));
750 * zero the sense buffer. the scsi spec mandates that any
751 * untransferred sense data should be interpreted as being zero.
753 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
755 saved_result
= scmd
->result
;
756 scmd
->request_buffer
= NULL
;
757 scmd
->request_bufflen
= 0;
759 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
761 scmd
->sc_data_direction
= DMA_NONE
;
763 rtn
= scsi_send_eh_cmnd(scmd
, SENSE_TIMEOUT
);
766 * when we eventually call scsi_finish, we really wish to complete
767 * the original request, so let's restore the original data. (db)
769 scsi_setup_cmd_retry(scmd
);
770 scmd
->result
= saved_result
;
773 * hey, we are done. let's look to see what happened.
775 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
776 __FUNCTION__
, scmd
, rtn
));
779 else if (rtn
== NEEDS_RETRY
) {
788 * scsi_eh_abort_cmds - abort canceled commands.
789 * @shost: scsi host being recovered.
790 * @eh_done_q: list_head for processed commands.
793 * Try and see whether or not it makes sense to try and abort the
794 * running command. this only works out to be the case if we have one
795 * command that has timed out. if the command simply failed, it makes
796 * no sense to try and abort the command, since as far as the shost
797 * adapter is concerned, it isn't running.
799 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
800 struct list_head
*done_q
)
802 struct scsi_cmnd
*scmd
, *next
;
805 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
806 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
808 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
809 "0x%p\n", current
->comm
,
811 rtn
= scsi_try_to_abort_cmd(scmd
);
812 if (rtn
== SUCCESS
) {
813 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
814 if (!scsi_device_online(scmd
->device
) ||
815 !scsi_eh_tur(scmd
)) {
816 scsi_eh_finish_cmd(scmd
, done_q
);
820 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
827 return list_empty(work_q
);
831 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
832 * @scmd: SCSI cmd used to send BDR
835 * There is no timeout for this operation. if this operation is
836 * unreliable for a given host, then the host itself needs to put a
837 * timer on it, and set the host back to a consistent state prior to
840 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
844 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
847 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
848 if (rtn
== SUCCESS
) {
849 scmd
->device
->was_reset
= 1;
850 scmd
->device
->expecting_cc_ua
= 1;
857 * scsi_eh_try_stu - Send START_UNIT to device.
858 * @scmd: Scsi cmd to send START_UNIT
861 * 0 - Device is ready. 1 - Device NOT ready.
863 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
865 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
869 if (!scmd
->device
->allow_restart
)
872 memcpy(scmd
->cmnd
, stu_command
, sizeof(stu_command
));
875 * zero the sense buffer. the scsi spec mandates that any
876 * untransferred sense data should be interpreted as being zero.
878 memset(scmd
->sense_buffer
, 0, sizeof(scmd
->sense_buffer
));
880 saved_result
= scmd
->result
;
881 scmd
->request_buffer
= NULL
;
882 scmd
->request_bufflen
= 0;
884 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
886 scmd
->sc_data_direction
= DMA_NONE
;
888 rtn
= scsi_send_eh_cmnd(scmd
, START_UNIT_TIMEOUT
);
891 * when we eventually call scsi_finish, we really wish to complete
892 * the original request, so let's restore the original data. (db)
894 scsi_setup_cmd_retry(scmd
);
895 scmd
->result
= saved_result
;
898 * hey, we are done. let's look to see what happened.
900 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
901 __FUNCTION__
, scmd
, rtn
));
908 * scsi_eh_stu - send START_UNIT if needed
909 * @shost: scsi host being recovered.
910 * @eh_done_q: list_head for processed commands.
913 * If commands are failing due to not ready, initializing command required,
914 * try revalidating the device, which will end up sending a start unit.
916 static int scsi_eh_stu(struct Scsi_Host
*shost
,
917 struct list_head
*work_q
,
918 struct list_head
*done_q
)
920 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
921 struct scsi_device
*sdev
;
923 shost_for_each_device(sdev
, shost
) {
925 list_for_each_entry(scmd
, work_q
, eh_entry
)
926 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
927 scsi_check_sense(scmd
) == FAILED
) {
935 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
936 " 0x%p\n", current
->comm
, sdev
));
938 if (!scsi_eh_try_stu(stu_scmd
)) {
939 if (!scsi_device_online(sdev
) ||
940 !scsi_eh_tur(stu_scmd
)) {
941 list_for_each_entry_safe(scmd
, next
,
943 if (scmd
->device
== sdev
)
944 scsi_eh_finish_cmd(scmd
, done_q
);
948 SCSI_LOG_ERROR_RECOVERY(3,
949 printk("%s: START_UNIT failed to sdev:"
950 " 0x%p\n", current
->comm
, sdev
));
954 return list_empty(work_q
);
959 * scsi_eh_bus_device_reset - send bdr if needed
960 * @shost: scsi host being recovered.
961 * @eh_done_q: list_head for processed commands.
964 * Try a bus device reset. still, look to see whether we have multiple
965 * devices that are jammed or not - if we have multiple devices, it
966 * makes no sense to try bus_device_reset - we really would need to try
967 * a bus_reset instead.
969 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
970 struct list_head
*work_q
,
971 struct list_head
*done_q
)
973 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
974 struct scsi_device
*sdev
;
977 shost_for_each_device(sdev
, shost
) {
979 list_for_each_entry(scmd
, work_q
, eh_entry
)
980 if (scmd
->device
== sdev
) {
988 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
989 " 0x%p\n", current
->comm
,
991 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
992 if (rtn
== SUCCESS
) {
993 if (!scsi_device_online(sdev
) ||
994 !scsi_eh_tur(bdr_scmd
)) {
995 list_for_each_entry_safe(scmd
, next
,
997 if (scmd
->device
== sdev
)
998 scsi_eh_finish_cmd(scmd
,
1003 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1011 return list_empty(work_q
);
1015 * scsi_try_bus_reset - ask host to perform a bus reset
1016 * @scmd: SCSI cmd to send bus reset.
1018 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
1020 unsigned long flags
;
1023 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
1026 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
1029 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
1031 if (rtn
== SUCCESS
) {
1032 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1033 ssleep(BUS_RESET_SETTLE_TIME
);
1034 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1035 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1036 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1043 * scsi_try_host_reset - ask host adapter to reset itself
1044 * @scmd: SCSI cmd to send hsot reset.
1046 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
1048 unsigned long flags
;
1051 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
1054 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
1057 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
1059 if (rtn
== SUCCESS
) {
1060 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
1061 ssleep(HOST_RESET_SETTLE_TIME
);
1062 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
1063 scsi_report_bus_reset(scmd
->device
->host
, scmd
->device
->channel
);
1064 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
1071 * scsi_eh_bus_reset - send a bus reset
1072 * @shost: scsi host being recovered.
1073 * @eh_done_q: list_head for processed commands.
1075 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1076 struct list_head
*work_q
,
1077 struct list_head
*done_q
)
1079 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1080 unsigned int channel
;
1084 * we really want to loop over the various channels, and do this on
1085 * a channel by channel basis. we should also check to see if any
1086 * of the failed commands are on soft_reset devices, and if so, skip
1090 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1092 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1093 if (channel
== scmd
->device
->channel
) {
1097 * FIXME add back in some support for
1098 * soft_reset devices.
1105 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1106 " %d\n", current
->comm
,
1108 rtn
= scsi_try_bus_reset(chan_scmd
);
1109 if (rtn
== SUCCESS
) {
1110 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1111 if (channel
== scmd
->device
->channel
)
1112 if (!scsi_device_online(scmd
->device
) ||
1114 scsi_eh_finish_cmd(scmd
,
1118 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1119 " failed chan: %d\n",
1124 return list_empty(work_q
);
1128 * scsi_eh_host_reset - send a host reset
1129 * @work_q: list_head for processed commands.
1130 * @done_q: list_head for processed commands.
1132 static int scsi_eh_host_reset(struct list_head
*work_q
,
1133 struct list_head
*done_q
)
1135 struct scsi_cmnd
*scmd
, *next
;
1138 if (!list_empty(work_q
)) {
1139 scmd
= list_entry(work_q
->next
,
1140 struct scsi_cmnd
, eh_entry
);
1142 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1145 rtn
= scsi_try_host_reset(scmd
);
1146 if (rtn
== SUCCESS
) {
1147 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1148 if (!scsi_device_online(scmd
->device
) ||
1149 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1151 scsi_eh_finish_cmd(scmd
, done_q
);
1154 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1159 return list_empty(work_q
);
1163 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1164 * @work_q: list_head for processed commands.
1165 * @done_q: list_head for processed commands.
1168 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1169 struct list_head
*done_q
)
1171 struct scsi_cmnd
*scmd
, *next
;
1173 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1174 printk(KERN_INFO
"scsi: Device offlined - not"
1175 " ready after error recovery: host"
1176 " %d channel %d id %d lun %d\n",
1177 scmd
->device
->host
->host_no
,
1178 scmd
->device
->channel
,
1181 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1182 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1184 * FIXME: Handle lost cmds.
1187 scsi_eh_finish_cmd(scmd
, done_q
);
1193 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1194 * @scmd: SCSI cmd to examine.
1197 * This is *only* called when we are examining the status after sending
1198 * out the actual data command. any commands that are queued for error
1199 * recovery (e.g. test_unit_ready) do *not* come through here.
1201 * When this routine returns failed, it means the error handler thread
1202 * is woken. In cases where the error code indicates an error that
1203 * doesn't require the error handler read (i.e. we don't need to
1204 * abort/reset), this function should return SUCCESS.
1206 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1211 * if the device is offline, then we clearly just pass the result back
1212 * up to the top level.
1214 if (!scsi_device_online(scmd
->device
)) {
1215 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1222 * first check the host byte, to see if there is anything in there
1223 * that would indicate what we need to do.
1225 switch (host_byte(scmd
->result
)) {
1226 case DID_PASSTHROUGH
:
1228 * no matter what, pass this through to the upper layer.
1229 * nuke this special code so that it looks like we are saying
1232 scmd
->result
&= 0xff00ffff;
1236 * looks good. drop through, and check the next byte.
1239 case DID_NO_CONNECT
:
1240 case DID_BAD_TARGET
:
1243 * note - this means that we just report the status back
1244 * to the top level driver, not that we actually think
1245 * that it indicates SUCCESS.
1249 * when the low level driver returns did_soft_error,
1250 * it is responsible for keeping an internal retry counter
1251 * in order to avoid endless loops (db)
1253 * actually this is a bug in this function here. we should
1254 * be mindful of the maximum number of retries specified
1255 * and not get stuck in a loop.
1257 case DID_SOFT_ERROR
:
1263 return ADD_TO_MLQUEUE
;
1266 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1267 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1269 * execute reservation conflict processing code
1280 * when we scan the bus, we get timeout messages for
1281 * these commands if there is no device available.
1282 * other hosts report did_no_connect for the same thing.
1284 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1285 scmd
->cmnd
[0] == INQUIRY
)) {
1297 * next, check the message byte.
1299 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1303 * check the status byte to see if this indicates anything special.
1305 switch (status_byte(scmd
->result
)) {
1308 * the case of trying to send too many commands to a
1309 * tagged queueing device.
1313 * device can't talk to us at the moment. Should only
1314 * occur (SAM-3) when the task queue is empty, so will cause
1315 * the empty queue handling to trigger a stall in the
1318 return ADD_TO_MLQUEUE
;
1320 case COMMAND_TERMINATED
:
1323 case CHECK_CONDITION
:
1324 rtn
= scsi_check_sense(scmd
);
1325 if (rtn
== NEEDS_RETRY
)
1327 /* if rtn == FAILED, we have no sense information;
1328 * returning FAILED will wake the error handler thread
1329 * to collect the sense and redo the decide
1332 case CONDITION_GOOD
:
1333 case INTERMEDIATE_GOOD
:
1334 case INTERMEDIATE_C_GOOD
:
1337 * who knows? FIXME(eric)
1341 case RESERVATION_CONFLICT
:
1342 printk(KERN_INFO
"scsi: reservation conflict: host"
1343 " %d channel %d id %d lun %d\n",
1344 scmd
->device
->host
->host_no
, scmd
->device
->channel
,
1345 scmd
->device
->id
, scmd
->device
->lun
);
1346 return SUCCESS
; /* causes immediate i/o error */
1354 /* we requeue for retry because the error was retryable, and
1355 * the request was not marked fast fail. Note that above,
1356 * even if the request is marked fast fail, we still requeue
1357 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1358 if ((++scmd
->retries
) < scmd
->allowed
1359 && !blk_noretry_request(scmd
->request
)) {
1363 * no more retries - report this one back to upper level.
1370 * scsi_eh_lock_done - done function for eh door lock request
1371 * @scmd: SCSI command block for the door lock request
1374 * We completed the asynchronous door lock request, and it has either
1375 * locked the door or failed. We must free the command structures
1376 * associated with this request.
1378 static void scsi_eh_lock_done(struct scsi_cmnd
*scmd
)
1380 struct scsi_request
*sreq
= scmd
->sc_request
;
1382 scsi_release_request(sreq
);
1387 * scsi_eh_lock_door - Prevent medium removal for the specified device
1388 * @sdev: SCSI device to prevent medium removal
1391 * We must be called from process context; scsi_allocate_request()
1395 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1396 * head of the devices request queue, and continue.
1399 * scsi_allocate_request() may sleep waiting for existing requests to
1400 * be processed. However, since we haven't kicked off any request
1401 * processing for this host, this may deadlock.
1403 * If scsi_allocate_request() fails for what ever reason, we
1404 * completely forget to lock the door.
1406 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1408 struct scsi_request
*sreq
= scsi_allocate_request(sdev
, GFP_KERNEL
);
1410 if (unlikely(!sreq
)) {
1411 printk(KERN_ERR
"%s: request allocate failed,"
1412 "prevent media removal cmd not sent\n", __FUNCTION__
);
1416 sreq
->sr_cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1417 sreq
->sr_cmnd
[1] = 0;
1418 sreq
->sr_cmnd
[2] = 0;
1419 sreq
->sr_cmnd
[3] = 0;
1420 sreq
->sr_cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1421 sreq
->sr_cmnd
[5] = 0;
1422 sreq
->sr_data_direction
= DMA_NONE
;
1423 sreq
->sr_bufflen
= 0;
1424 sreq
->sr_buffer
= NULL
;
1425 sreq
->sr_allowed
= 5;
1426 sreq
->sr_done
= scsi_eh_lock_done
;
1427 sreq
->sr_timeout_per_command
= 10 * HZ
;
1428 sreq
->sr_cmd_len
= COMMAND_SIZE(sreq
->sr_cmnd
[0]);
1430 scsi_insert_special_req(sreq
, 1);
1435 * scsi_restart_operations - restart io operations to the specified host.
1436 * @shost: Host we are restarting.
1439 * When we entered the error handler, we blocked all further i/o to
1440 * this device. we need to 'reverse' this process.
1442 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1444 struct scsi_device
*sdev
;
1447 * If the door was locked, we need to insert a door lock request
1448 * onto the head of the SCSI request queue for the device. There
1449 * is no point trying to lock the door of an off-line device.
1451 shost_for_each_device(sdev
, shost
) {
1452 if (scsi_device_online(sdev
) && sdev
->locked
)
1453 scsi_eh_lock_door(sdev
);
1457 * next free up anything directly waiting upon the host. this
1458 * will be requests for character device operations, and also for
1459 * ioctls to queued block devices.
1461 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1464 scsi_host_set_state(shost
, SHOST_RUNNING
);
1466 wake_up(&shost
->host_wait
);
1469 * finally we need to re-initiate requests that may be pending. we will
1470 * have had everything blocked while error handling is taking place, and
1471 * now that error recovery is done, we will need to ensure that these
1472 * requests are started.
1474 scsi_run_host_queues(shost
);
1478 * scsi_eh_ready_devs - check device ready state and recover if not.
1479 * @shost: host to be recovered.
1480 * @eh_done_q: list_head for processed commands.
1483 static void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1484 struct list_head
*work_q
,
1485 struct list_head
*done_q
)
1487 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1488 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1489 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1490 if (!scsi_eh_host_reset(work_q
, done_q
))
1491 scsi_eh_offline_sdevs(work_q
, done_q
);
1495 * scsi_eh_flush_done_q - finish processed commands or retry them.
1496 * @done_q: list_head of processed commands.
1499 static void scsi_eh_flush_done_q(struct list_head
*done_q
)
1501 struct scsi_cmnd
*scmd
, *next
;
1503 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1504 list_del_init(&scmd
->eh_entry
);
1505 if (scsi_device_online(scmd
->device
) &&
1506 !blk_noretry_request(scmd
->request
) &&
1507 (++scmd
->retries
< scmd
->allowed
)) {
1508 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1512 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1515 * If just we got sense for the device (called
1516 * scsi_eh_get_sense), scmd->result is already
1517 * set, do not set DRIVER_TIMEOUT.
1520 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1521 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1523 current
->comm
, scmd
));
1524 scsi_finish_command(scmd
);
1530 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1531 * @shost: Host to unjam.
1534 * When we come in here, we *know* that all commands on the bus have
1535 * either completed, failed or timed out. we also know that no further
1536 * commands are being sent to the host, so things are relatively quiet
1537 * and we have freedom to fiddle with things as we wish.
1539 * This is only the *default* implementation. it is possible for
1540 * individual drivers to supply their own version of this function, and
1541 * if the maintainer wishes to do this, it is strongly suggested that
1542 * this function be taken as a template and modified. this function
1543 * was designed to correctly handle problems for about 95% of the
1544 * different cases out there, and it should always provide at least a
1545 * reasonable amount of error recovery.
1547 * Any command marked 'failed' or 'timeout' must eventually have
1548 * scsi_finish_cmd() called for it. we do all of the retry stuff
1549 * here, so when we restart the host after we return it should have an
1552 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1554 unsigned long flags
;
1555 LIST_HEAD(eh_work_q
);
1556 LIST_HEAD(eh_done_q
);
1558 spin_lock_irqsave(shost
->host_lock
, flags
);
1559 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1560 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1562 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1564 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1565 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1566 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1568 scsi_eh_flush_done_q(&eh_done_q
);
1572 * scsi_error_handler - Handle errors/timeouts of SCSI cmds.
1573 * @data: Host for which we are running.
1576 * This is always run in the context of a kernel thread. The idea is
1577 * that we start this thing up when the kernel starts up (one per host
1578 * that we detect), and it immediately goes to sleep and waits for some
1579 * event (i.e. failure). When this takes place, we have the job of
1580 * trying to unjam the bus and restarting things.
1582 int scsi_error_handler(void *data
)
1584 struct Scsi_Host
*shost
= (struct Scsi_Host
*) data
;
1586 DECLARE_MUTEX_LOCKED(sem
);
1592 daemonize("scsi_eh_%d", shost
->host_no
);
1594 current
->flags
|= PF_NOFREEZE
;
1596 shost
->eh_wait
= &sem
;
1597 shost
->ehandler
= current
;
1600 * Wake up the thread that created us.
1602 SCSI_LOG_ERROR_RECOVERY(3, printk("Wake up parent of"
1603 " scsi_eh_%d\n",shost
->host_no
));
1605 complete(shost
->eh_notify
);
1609 * If we get a signal, it means we are supposed to go
1610 * away and die. This typically happens if the user is
1611 * trying to unload a module.
1613 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1615 " sleeping\n",shost
->host_no
));
1618 * Note - we always use down_interruptible with the semaphore
1619 * even if the module was loaded as part of the kernel. The
1620 * reason is that down() will cause this thread to be counted
1621 * in the load average as a running process, and down
1622 * interruptible doesn't. Given that we need to allow this
1623 * thread to die if the driver was loaded as a module, using
1624 * semaphores isn't unreasonable.
1626 down_interruptible(&sem
);
1630 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler"
1631 " scsi_eh_%d waking"
1632 " up\n",shost
->host_no
));
1634 shost
->eh_active
= 1;
1637 * We have a host that is failing for some reason. Figure out
1638 * what we need to do to get it up and online again (if we can).
1639 * If we fail, we end up taking the thing offline.
1641 if (shost
->hostt
->eh_strategy_handler
)
1642 rtn
= shost
->hostt
->eh_strategy_handler(shost
);
1644 scsi_unjam_host(shost
);
1646 shost
->eh_active
= 0;
1649 * Note - if the above fails completely, the action is to take
1650 * individual devices offline and flush the queue of any
1651 * outstanding requests that may have been pending. When we
1652 * restart, we restart any I/O to any other devices on the bus
1653 * which are still online.
1655 scsi_restart_operations(shost
);
1659 SCSI_LOG_ERROR_RECOVERY(1, printk("Error handler scsi_eh_%d"
1660 " exiting\n",shost
->host_no
));
1663 * Make sure that nobody tries to wake us up again.
1665 shost
->eh_wait
= NULL
;
1668 * Knock this down too. From this point on, the host is flying
1669 * without a pilot. If this is because the module is being unloaded,
1670 * that's fine. If the user sent a signal to this thing, we are
1671 * potentially in real danger.
1673 shost
->eh_active
= 0;
1674 shost
->ehandler
= NULL
;
1677 * If anyone is waiting for us to exit (i.e. someone trying to unload
1678 * a driver), then wake up that process to let them know we are on
1679 * the way out the door.
1681 complete_and_exit(shost
->eh_notify
, 0);
1686 * Function: scsi_report_bus_reset()
1688 * Purpose: Utility function used by low-level drivers to report that
1689 * they have observed a bus reset on the bus being handled.
1691 * Arguments: shost - Host in question
1692 * channel - channel on which reset was observed.
1696 * Lock status: Host lock must be held.
1698 * Notes: This only needs to be called if the reset is one which
1699 * originates from an unknown location. Resets originated
1700 * by the mid-level itself don't need to call this, but there
1701 * should be no harm.
1703 * The main purpose of this is to make sure that a CHECK_CONDITION
1704 * is properly treated.
1706 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1708 struct scsi_device
*sdev
;
1710 __shost_for_each_device(sdev
, shost
) {
1711 if (channel
== sdev
->channel
) {
1712 sdev
->was_reset
= 1;
1713 sdev
->expecting_cc_ua
= 1;
1717 EXPORT_SYMBOL(scsi_report_bus_reset
);
1720 * Function: scsi_report_device_reset()
1722 * Purpose: Utility function used by low-level drivers to report that
1723 * they have observed a device reset on the device being handled.
1725 * Arguments: shost - Host in question
1726 * channel - channel on which reset was observed
1727 * target - target on which reset was observed
1731 * Lock status: Host lock must be held
1733 * Notes: This only needs to be called if the reset is one which
1734 * originates from an unknown location. Resets originated
1735 * by the mid-level itself don't need to call this, but there
1736 * should be no harm.
1738 * The main purpose of this is to make sure that a CHECK_CONDITION
1739 * is properly treated.
1741 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1743 struct scsi_device
*sdev
;
1745 __shost_for_each_device(sdev
, shost
) {
1746 if (channel
== sdev
->channel
&&
1747 target
== sdev
->id
) {
1748 sdev
->was_reset
= 1;
1749 sdev
->expecting_cc_ua
= 1;
1753 EXPORT_SYMBOL(scsi_report_device_reset
);
1756 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1761 * Function: scsi_reset_provider
1763 * Purpose: Send requested reset to a bus or device at any phase.
1765 * Arguments: device - device to send reset to
1766 * flag - reset type (see scsi.h)
1768 * Returns: SUCCESS/FAILURE.
1770 * Notes: This is used by the SCSI Generic driver to provide
1771 * Bus/Device reset capability.
1774 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1776 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1780 scmd
->request
= &req
;
1781 memset(&scmd
->eh_timeout
, 0, sizeof(scmd
->eh_timeout
));
1782 scmd
->request
->rq_status
= RQ_SCSI_BUSY
;
1784 memset(&scmd
->cmnd
, '\0', sizeof(scmd
->cmnd
));
1786 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1788 scmd
->buffer
= NULL
;
1790 scmd
->request_buffer
= NULL
;
1791 scmd
->request_bufflen
= 0;
1795 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1796 scmd
->sc_request
= NULL
;
1797 scmd
->sc_magic
= SCSI_CMND_MAGIC
;
1799 init_timer(&scmd
->eh_timeout
);
1802 * Sometimes the command can get back into the timer chain,
1803 * so use the pid as an identifier.
1808 case SCSI_TRY_RESET_DEVICE
:
1809 rtn
= scsi_try_bus_device_reset(scmd
);
1813 case SCSI_TRY_RESET_BUS
:
1814 rtn
= scsi_try_bus_reset(scmd
);
1818 case SCSI_TRY_RESET_HOST
:
1819 rtn
= scsi_try_host_reset(scmd
);
1825 scsi_next_command(scmd
);
1828 EXPORT_SYMBOL(scsi_reset_provider
);
1831 * scsi_normalize_sense - normalize main elements from either fixed or
1832 * descriptor sense data format into a common format.
1834 * @sense_buffer: byte array containing sense data returned by device
1835 * @sb_len: number of valid bytes in sense_buffer
1836 * @sshdr: pointer to instance of structure that common
1837 * elements are written to.
1840 * The "main elements" from sense data are: response_code, sense_key,
1841 * asc, ascq and additional_length (only for descriptor format).
1843 * Typically this function can be called after a device has
1844 * responded to a SCSI command with the CHECK_CONDITION status.
1847 * 1 if valid sense data information found, else 0;
1849 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1850 struct scsi_sense_hdr
*sshdr
)
1852 if (!sense_buffer
|| !sb_len
|| (sense_buffer
[0] & 0x70) != 0x70)
1855 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1857 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1858 if (sshdr
->response_code
>= 0x72) {
1863 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1865 sshdr
->asc
= sense_buffer
[2];
1867 sshdr
->ascq
= sense_buffer
[3];
1869 sshdr
->additional_length
= sense_buffer
[7];
1875 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1877 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1878 sb_len
: (sense_buffer
[7] + 8);
1880 sshdr
->asc
= sense_buffer
[12];
1882 sshdr
->ascq
= sense_buffer
[13];
1888 EXPORT_SYMBOL(scsi_normalize_sense
);
1890 int scsi_request_normalize_sense(struct scsi_request
*sreq
,
1891 struct scsi_sense_hdr
*sshdr
)
1893 return scsi_normalize_sense(sreq
->sr_sense_buffer
,
1894 sizeof(sreq
->sr_sense_buffer
), sshdr
);
1896 EXPORT_SYMBOL(scsi_request_normalize_sense
);
1898 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1899 struct scsi_sense_hdr
*sshdr
)
1901 return scsi_normalize_sense(cmd
->sense_buffer
,
1902 sizeof(cmd
->sense_buffer
), sshdr
);
1904 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1907 * scsi_sense_desc_find - search for a given descriptor type in
1908 * descriptor sense data format.
1910 * @sense_buffer: byte array of descriptor format sense data
1911 * @sb_len: number of valid bytes in sense_buffer
1912 * @desc_type: value of descriptor type to find
1913 * (e.g. 0 -> information)
1916 * only valid when sense data is in descriptor format
1919 * pointer to start of (first) descriptor if found else NULL
1921 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1924 int add_sen_len
, add_len
, desc_len
, k
;
1927 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1929 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1931 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1932 add_sen_len
: (sb_len
- 8);
1933 descp
= &sense_buffer
[8];
1934 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1936 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1937 desc_len
= add_len
+ 2;
1938 if (descp
[0] == desc_type
)
1940 if (add_len
< 0) // short descriptor ??
1945 EXPORT_SYMBOL(scsi_sense_desc_find
);
1948 * scsi_get_sense_info_fld - attempts to get information field from
1949 * sense data (either fixed or descriptor format)
1951 * @sense_buffer: byte array of sense data
1952 * @sb_len: number of valid bytes in sense_buffer
1953 * @info_out: pointer to 64 integer where 8 or 4 byte information
1954 * field will be placed if found.
1957 * 1 if information field found, 0 if not found.
1959 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
1968 switch (sense_buffer
[0] & 0x7f) {
1971 if (sense_buffer
[0] & 0x80) {
1972 *info_out
= (sense_buffer
[3] << 24) +
1973 (sense_buffer
[4] << 16) +
1974 (sense_buffer
[5] << 8) + sense_buffer
[6];
1980 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
1982 if (ucp
&& (0xa == ucp
[1])) {
1984 for (j
= 0; j
< 8; ++j
) {
1997 EXPORT_SYMBOL(scsi_get_sense_info_fld
);