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/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
41 #define SENSE_TIMEOUT (10*HZ)
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
53 if (shost
->host_busy
== shost
->host_failed
) {
54 wake_up_process(shost
->ehandler
);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
64 * Schedule SCSI EH without scmd.
66 void scsi_schedule_eh(struct Scsi_Host
*shost
)
70 spin_lock_irqsave(shost
->host_lock
, flags
);
72 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
73 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
74 shost
->host_eh_scheduled
++;
75 scsi_eh_wakeup(shost
);
78 spin_unlock_irqrestore(shost
->host_lock
, flags
);
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
90 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
92 struct Scsi_Host
*shost
= scmd
->device
->host
;
99 spin_lock_irqsave(shost
->host_lock
, flags
);
100 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
101 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
105 scmd
->eh_eflags
|= eh_flag
;
106 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
107 shost
->host_failed
++;
108 scsi_eh_wakeup(shost
);
110 spin_unlock_irqrestore(shost
->host_lock
, flags
);
115 * scsi_times_out - Timeout function for normal scsi commands.
116 * @req: request that is timing out.
119 * We do not need to lock this. There is the potential for a race
120 * only in that the normal completion handling might run, but if the
121 * normal completion function determines that the timer has already
122 * fired, then it mustn't do anything.
124 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
126 struct scsi_cmnd
*scmd
= req
->special
;
127 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
129 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
131 if (scmd
->device
->host
->transportt
->eh_timed_out
)
132 rtn
= scmd
->device
->host
->transportt
->eh_timed_out(scmd
);
133 else if (scmd
->device
->host
->hostt
->eh_timed_out
)
134 rtn
= scmd
->device
->host
->hostt
->eh_timed_out(scmd
);
136 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
137 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
138 scmd
->result
|= DID_TIME_OUT
<< 16;
139 rtn
= BLK_EH_HANDLED
;
146 * scsi_block_when_processing_errors - Prevent cmds from being queued.
147 * @sdev: Device on which we are performing recovery.
150 * We block until the host is out of error recovery, and then check to
151 * see whether the host or the device is offline.
154 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
156 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
160 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
162 online
= scsi_device_online(sdev
);
164 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
169 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
171 #ifdef CONFIG_SCSI_LOGGING
173 * scsi_eh_prt_fail_stats - Log info on failures.
174 * @shost: scsi host being recovered.
175 * @work_q: Queue of scsi cmds to process.
177 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
178 struct list_head
*work_q
)
180 struct scsi_cmnd
*scmd
;
181 struct scsi_device
*sdev
;
182 int total_failures
= 0;
185 int devices_failed
= 0;
187 shost_for_each_device(sdev
, shost
) {
188 list_for_each_entry(scmd
, work_q
, eh_entry
) {
189 if (scmd
->device
== sdev
) {
191 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
198 if (cmd_cancel
|| cmd_failed
) {
199 SCSI_LOG_ERROR_RECOVERY(3,
200 sdev_printk(KERN_INFO
, sdev
,
201 "%s: cmds failed: %d, cancel: %d\n",
202 __func__
, cmd_failed
,
210 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
211 " devices require eh work\n",
212 total_failures
, devices_failed
));
217 * scsi_check_sense - Examine scsi cmd sense
218 * @scmd: Cmd to have sense checked.
221 * SUCCESS or FAILED or NEEDS_RETRY
224 * When a deferred error is detected the current command has
225 * not been executed and needs retrying.
227 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
229 struct scsi_device
*sdev
= scmd
->device
;
230 struct scsi_sense_hdr sshdr
;
232 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
233 return FAILED
; /* no valid sense data */
235 if (scsi_sense_is_deferred(&sshdr
))
238 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
239 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
242 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
243 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
245 /* handler does not care. Drop down to default handling */
249 * Previous logic looked for FILEMARK, EOM or ILI which are
250 * mainly associated with tapes and returned SUCCESS.
252 if (sshdr
.response_code
== 0x70) {
254 if (scmd
->sense_buffer
[2] & 0xe0)
258 * descriptor format: look for "stream commands sense data
259 * descriptor" (see SSC-3). Assume single sense data
260 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
262 if ((sshdr
.additional_length
> 3) &&
263 (scmd
->sense_buffer
[8] == 0x4) &&
264 (scmd
->sense_buffer
[11] & 0xe0))
268 switch (sshdr
.sense_key
) {
271 case RECOVERED_ERROR
:
272 return /* soft_error */ SUCCESS
;
274 case ABORTED_COMMAND
:
275 if (sshdr
.asc
== 0x10) /* DIF */
282 * if we are expecting a cc/ua because of a bus reset that we
283 * performed, treat this just as a retry. otherwise this is
284 * information that we should pass up to the upper-level driver
285 * so that we can deal with it there.
287 if (scmd
->device
->expecting_cc_ua
) {
288 scmd
->device
->expecting_cc_ua
= 0;
292 * if the device is in the process of becoming ready, we
295 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
298 * if the device is not started, we need to wake
299 * the error handler to start the motor
301 if (scmd
->device
->allow_restart
&&
302 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
306 /* these three are not supported */
308 case VOLUME_OVERFLOW
:
313 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
314 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
315 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
321 if (scmd
->device
->retry_hwerror
)
322 return ADD_TO_MLQUEUE
;
326 case ILLEGAL_REQUEST
:
334 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
336 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
337 struct scsi_device
*tmp_sdev
;
339 if (!sht
->change_queue_depth
||
340 sdev
->queue_depth
>= sdev
->max_queue_depth
)
343 if (time_before(jiffies
,
344 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
347 if (time_before(jiffies
,
348 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
352 * Walk all devices of a target and do
355 shost_for_each_device(tmp_sdev
, sdev
->host
) {
356 if (tmp_sdev
->channel
!= sdev
->channel
||
357 tmp_sdev
->id
!= sdev
->id
||
358 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
361 * call back into LLD to increase queue_depth by one
362 * with ramp up reason code.
364 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
365 SCSI_QDEPTH_RAMP_UP
);
366 sdev
->last_queue_ramp_up
= jiffies
;
370 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
372 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
373 struct scsi_device
*tmp_sdev
;
375 if (!sht
->change_queue_depth
)
378 shost_for_each_device(tmp_sdev
, sdev
->host
) {
379 if (tmp_sdev
->channel
!= sdev
->channel
||
380 tmp_sdev
->id
!= sdev
->id
)
383 * We do not know the number of commands that were at
384 * the device when we got the queue full so we start
385 * from the highest possible value and work our way down.
387 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
393 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
394 * @scmd: SCSI cmd to examine.
397 * This is *only* called when we are examining the status of commands
398 * queued during error recovery. the main difference here is that we
399 * don't allow for the possibility of retries here, and we are a lot
400 * more restrictive about what we consider acceptable.
402 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
405 * first check the host byte, to see if there is anything in there
406 * that would indicate what we need to do.
408 if (host_byte(scmd
->result
) == DID_RESET
) {
410 * rats. we are already in the error handler, so we now
411 * get to try and figure out what to do next. if the sense
412 * is valid, we have a pretty good idea of what to do.
413 * if not, we mark it as FAILED.
415 return scsi_check_sense(scmd
);
417 if (host_byte(scmd
->result
) != DID_OK
)
421 * next, check the message byte.
423 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
427 * now, check the status byte to see if this indicates
430 switch (status_byte(scmd
->result
)) {
432 scsi_handle_queue_ramp_up(scmd
->device
);
433 case COMMAND_TERMINATED
:
435 case CHECK_CONDITION
:
436 return scsi_check_sense(scmd
);
438 case INTERMEDIATE_GOOD
:
439 case INTERMEDIATE_C_GOOD
:
441 * who knows? FIXME(eric)
444 case RESERVATION_CONFLICT
:
446 * let issuer deal with this, it could be just fine
450 scsi_handle_queue_full(scmd
->device
);
460 * scsi_eh_done - Completion function for error handling.
461 * @scmd: Cmd that is done.
463 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
465 struct completion
*eh_action
;
467 SCSI_LOG_ERROR_RECOVERY(3,
468 printk("%s scmd: %p result: %x\n",
469 __func__
, scmd
, scmd
->result
));
471 eh_action
= scmd
->device
->host
->eh_action
;
477 * scsi_try_host_reset - ask host adapter to reset itself
478 * @scmd: SCSI cmd to send hsot reset.
480 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
485 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
488 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
491 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
493 if (rtn
== SUCCESS
) {
494 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
495 ssleep(HOST_RESET_SETTLE_TIME
);
496 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
497 scsi_report_bus_reset(scmd
->device
->host
,
499 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
506 * scsi_try_bus_reset - ask host to perform a bus reset
507 * @scmd: SCSI cmd to send bus reset.
509 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
514 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
517 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
520 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
522 if (rtn
== SUCCESS
) {
523 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
524 ssleep(BUS_RESET_SETTLE_TIME
);
525 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
526 scsi_report_bus_reset(scmd
->device
->host
,
528 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
534 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
537 sdev
->expecting_cc_ua
= 1;
541 * scsi_try_target_reset - Ask host to perform a target reset
542 * @scmd: SCSI cmd used to send a target reset
545 * There is no timeout for this operation. if this operation is
546 * unreliable for a given host, then the host itself needs to put a
547 * timer on it, and set the host back to a consistent state prior to
550 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
555 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
558 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
559 if (rtn
== SUCCESS
) {
560 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
561 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
562 __scsi_report_device_reset
);
563 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
570 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
571 * @scmd: SCSI cmd used to send BDR
574 * There is no timeout for this operation. if this operation is
575 * unreliable for a given host, then the host itself needs to put a
576 * timer on it, and set the host back to a consistent state prior to
579 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
583 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
586 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
588 __scsi_report_device_reset(scmd
->device
, NULL
);
592 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
594 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
597 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
601 * scsi_try_to_abort_cmd - Ask host to abort a running command.
602 * @scmd: SCSI cmd to abort from Lower Level.
605 * This function will not return until the user's completion function
606 * has been called. there is no timeout on this operation. if the
607 * author of the low-level driver wishes this operation to be timed,
608 * they can provide this facility themselves. helper functions in
609 * scsi_error.c can be supplied to make this easier to do.
611 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
614 * scsi_done was called just after the command timed out and before
615 * we had a chance to process it. (db)
617 if (scmd
->serial_number
== 0)
619 return __scsi_try_to_abort_cmd(scmd
);
622 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
624 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
625 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
626 if (scsi_try_target_reset(scmd
) != SUCCESS
)
627 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
628 scsi_try_host_reset(scmd
);
632 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
633 * @scmd: SCSI command structure to hijack
634 * @ses: structure to save restore information
635 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
636 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
637 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
639 * This function is used to save a scsi command information before re-execution
640 * as part of the error recovery process. If @sense_bytes is 0 the command
641 * sent must be one that does not transfer any data. If @sense_bytes != 0
642 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
643 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
645 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
646 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
648 struct scsi_device
*sdev
= scmd
->device
;
651 * We need saved copies of a number of fields - this is because
652 * error handling may need to overwrite these with different values
653 * to run different commands, and once error handling is complete,
654 * we will need to restore these values prior to running the actual
657 ses
->cmd_len
= scmd
->cmd_len
;
658 ses
->cmnd
= scmd
->cmnd
;
659 ses
->data_direction
= scmd
->sc_data_direction
;
660 ses
->sdb
= scmd
->sdb
;
661 ses
->next_rq
= scmd
->request
->next_rq
;
662 ses
->result
= scmd
->result
;
663 ses
->underflow
= scmd
->underflow
;
664 ses
->prot_op
= scmd
->prot_op
;
666 scmd
->prot_op
= SCSI_PROT_NORMAL
;
667 scmd
->cmnd
= ses
->eh_cmnd
;
668 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
669 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
670 scmd
->request
->next_rq
= NULL
;
673 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
675 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
677 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
678 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
679 scmd
->sdb
.table
.nents
= 1;
680 scmd
->cmnd
[0] = REQUEST_SENSE
;
681 scmd
->cmnd
[4] = scmd
->sdb
.length
;
682 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
684 scmd
->sc_data_direction
= DMA_NONE
;
686 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
687 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
688 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
694 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
695 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
696 (sdev
->lun
<< 5 & 0xe0);
699 * Zero the sense buffer. The scsi spec mandates that any
700 * untransferred sense data should be interpreted as being zero.
702 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
704 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
707 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
708 * @scmd: SCSI command structure to restore
709 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
711 * Undo any damage done by above scsi_eh_prep_cmnd().
713 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
716 * Restore original data
718 scmd
->cmd_len
= ses
->cmd_len
;
719 scmd
->cmnd
= ses
->cmnd
;
720 scmd
->sc_data_direction
= ses
->data_direction
;
721 scmd
->sdb
= ses
->sdb
;
722 scmd
->request
->next_rq
= ses
->next_rq
;
723 scmd
->result
= ses
->result
;
724 scmd
->underflow
= ses
->underflow
;
725 scmd
->prot_op
= ses
->prot_op
;
727 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
730 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
731 * @scmd: SCSI command structure to hijack
733 * @cmnd_size: size in bytes of @cmnd
734 * @timeout: timeout for this request
735 * @sense_bytes: size of sense data to copy or 0
737 * This function is used to send a scsi command down to a target device
738 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
741 * SUCCESS or FAILED or NEEDS_RETRY
743 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
744 int cmnd_size
, int timeout
, unsigned sense_bytes
)
746 struct scsi_device
*sdev
= scmd
->device
;
747 struct Scsi_Host
*shost
= sdev
->host
;
748 DECLARE_COMPLETION_ONSTACK(done
);
749 unsigned long timeleft
;
751 struct scsi_eh_save ses
;
754 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
755 shost
->eh_action
= &done
;
757 spin_lock_irqsave(shost
->host_lock
, flags
);
759 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
760 spin_unlock_irqrestore(shost
->host_lock
, flags
);
762 timeleft
= wait_for_completion_timeout(&done
, timeout
);
764 shost
->eh_action
= NULL
;
766 scsi_log_completion(scmd
, SUCCESS
);
768 SCSI_LOG_ERROR_RECOVERY(3,
769 printk("%s: scmd: %p, timeleft: %ld\n",
770 __func__
, scmd
, timeleft
));
773 * If there is time left scsi_eh_done got called, and we will
774 * examine the actual status codes to see whether the command
775 * actually did complete normally, else tell the host to forget
776 * about this command.
779 rtn
= scsi_eh_completed_normally(scmd
);
780 SCSI_LOG_ERROR_RECOVERY(3,
781 printk("%s: scsi_eh_completed_normally %x\n",
797 scsi_abort_eh_cmnd(scmd
);
801 scsi_eh_restore_cmnd(scmd
, &ses
);
806 * scsi_request_sense - Request sense data from a particular target.
807 * @scmd: SCSI cmd for request sense.
810 * Some hosts automatically obtain this information, others require
811 * that we obtain it on our own. This function will *not* return until
812 * the command either times out, or it completes.
814 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
816 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
820 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
821 * @scmd: Original SCSI cmd that eh has finished.
822 * @done_q: Queue for processed commands.
825 * We don't want to use the normal command completion while we are are
826 * still handling errors - it may cause other commands to be queued,
827 * and that would disturb what we are doing. Thus we really want to
828 * keep a list of pending commands for final completion, and once we
829 * are ready to leave error handling we handle completion for real.
831 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
833 scmd
->device
->host
->host_failed
--;
835 list_move_tail(&scmd
->eh_entry
, done_q
);
837 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
840 * scsi_eh_get_sense - Get device sense data.
841 * @work_q: Queue of commands to process.
842 * @done_q: Queue of processed commands.
845 * See if we need to request sense information. if so, then get it
846 * now, so we have a better idea of what to do.
849 * This has the unfortunate side effect that if a shost adapter does
850 * not automatically request sense information, we end up shutting
851 * it down before we request it.
853 * All drivers should request sense information internally these days,
854 * so for now all I have to say is tough noogies if you end up in here.
856 * XXX: Long term this code should go away, but that needs an audit of
859 int scsi_eh_get_sense(struct list_head
*work_q
,
860 struct list_head
*done_q
)
862 struct scsi_cmnd
*scmd
, *next
;
865 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
866 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
867 SCSI_SENSE_VALID(scmd
))
870 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
871 "%s: requesting sense\n",
873 rtn
= scsi_request_sense(scmd
);
877 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
878 " result %x\n", scmd
,
880 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
882 rtn
= scsi_decide_disposition(scmd
);
885 * if the result was normal, then just pass it along to the
889 /* we don't want this command reissued, just
890 * finished with the sense data, so set
891 * retries to the max allowed to ensure it
892 * won't get reissued */
893 scmd
->retries
= scmd
->allowed
;
894 else if (rtn
!= NEEDS_RETRY
)
897 scsi_eh_finish_cmd(scmd
, done_q
);
900 return list_empty(work_q
);
902 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
905 * scsi_eh_tur - Send TUR to device.
906 * @scmd: &scsi_cmnd to send TUR
909 * 0 - Device is ready. 1 - Device NOT ready.
911 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
913 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
914 int retry_cnt
= 1, rtn
;
917 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
919 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
920 __func__
, scmd
, rtn
));
935 * scsi_eh_abort_cmds - abort pending commands.
936 * @work_q: &list_head for pending commands.
937 * @done_q: &list_head for processed commands.
940 * Try and see whether or not it makes sense to try and abort the
941 * running command. This only works out to be the case if we have one
942 * command that has timed out. If the command simply failed, it makes
943 * no sense to try and abort the command, since as far as the shost
944 * adapter is concerned, it isn't running.
946 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
947 struct list_head
*done_q
)
949 struct scsi_cmnd
*scmd
, *next
;
952 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
953 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
955 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
956 "0x%p\n", current
->comm
,
958 rtn
= scsi_try_to_abort_cmd(scmd
);
959 if (rtn
== SUCCESS
) {
960 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
961 if (!scsi_device_online(scmd
->device
) ||
962 !scsi_eh_tur(scmd
)) {
963 scsi_eh_finish_cmd(scmd
, done_q
);
967 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
974 return list_empty(work_q
);
978 * scsi_eh_try_stu - Send START_UNIT to device.
979 * @scmd: &scsi_cmnd to send START_UNIT
982 * 0 - Device is ready. 1 - Device NOT ready.
984 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
986 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
988 if (scmd
->device
->allow_restart
) {
989 int i
, rtn
= NEEDS_RETRY
;
991 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
992 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1002 * scsi_eh_stu - send START_UNIT if needed
1003 * @shost: &scsi host being recovered.
1004 * @work_q: &list_head for pending commands.
1005 * @done_q: &list_head for processed commands.
1008 * If commands are failing due to not ready, initializing command required,
1009 * try revalidating the device, which will end up sending a start unit.
1011 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1012 struct list_head
*work_q
,
1013 struct list_head
*done_q
)
1015 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1016 struct scsi_device
*sdev
;
1018 shost_for_each_device(sdev
, shost
) {
1020 list_for_each_entry(scmd
, work_q
, eh_entry
)
1021 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1022 scsi_check_sense(scmd
) == FAILED
) {
1030 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1031 " 0x%p\n", current
->comm
, sdev
));
1033 if (!scsi_eh_try_stu(stu_scmd
)) {
1034 if (!scsi_device_online(sdev
) ||
1035 !scsi_eh_tur(stu_scmd
)) {
1036 list_for_each_entry_safe(scmd
, next
,
1038 if (scmd
->device
== sdev
)
1039 scsi_eh_finish_cmd(scmd
, done_q
);
1043 SCSI_LOG_ERROR_RECOVERY(3,
1044 printk("%s: START_UNIT failed to sdev:"
1045 " 0x%p\n", current
->comm
, sdev
));
1049 return list_empty(work_q
);
1054 * scsi_eh_bus_device_reset - send bdr if needed
1055 * @shost: scsi host being recovered.
1056 * @work_q: &list_head for pending commands.
1057 * @done_q: &list_head for processed commands.
1060 * Try a bus device reset. Still, look to see whether we have multiple
1061 * devices that are jammed or not - if we have multiple devices, it
1062 * makes no sense to try bus_device_reset - we really would need to try
1063 * a bus_reset instead.
1065 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1066 struct list_head
*work_q
,
1067 struct list_head
*done_q
)
1069 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1070 struct scsi_device
*sdev
;
1073 shost_for_each_device(sdev
, shost
) {
1075 list_for_each_entry(scmd
, work_q
, eh_entry
)
1076 if (scmd
->device
== sdev
) {
1084 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1085 " 0x%p\n", current
->comm
,
1087 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1088 if (rtn
== SUCCESS
) {
1089 if (!scsi_device_online(sdev
) ||
1090 !scsi_eh_tur(bdr_scmd
)) {
1091 list_for_each_entry_safe(scmd
, next
,
1093 if (scmd
->device
== sdev
)
1094 scsi_eh_finish_cmd(scmd
,
1099 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1107 return list_empty(work_q
);
1111 * scsi_eh_target_reset - send target reset if needed
1112 * @shost: scsi host being recovered.
1113 * @work_q: &list_head for pending commands.
1114 * @done_q: &list_head for processed commands.
1117 * Try a target reset.
1119 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1120 struct list_head
*work_q
,
1121 struct list_head
*done_q
)
1123 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1124 unsigned int id
= 0;
1129 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1130 if (id
== scmd_id(scmd
)) {
1136 /* not one exactly equal; find the next highest */
1137 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1138 if (scmd_id(scmd
) > id
&&
1140 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1145 /* no more commands, that's it */
1148 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1150 current
->comm
, id
));
1151 rtn
= scsi_try_target_reset(tgtr_scmd
);
1152 if (rtn
== SUCCESS
) {
1153 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1154 if (id
== scmd_id(scmd
))
1155 if (!scsi_device_online(scmd
->device
) ||
1156 !scsi_eh_tur(tgtr_scmd
))
1157 scsi_eh_finish_cmd(scmd
,
1161 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1164 current
->comm
, id
));
1168 return list_empty(work_q
);
1172 * scsi_eh_bus_reset - send a bus reset
1173 * @shost: &scsi host being recovered.
1174 * @work_q: &list_head for pending commands.
1175 * @done_q: &list_head for processed commands.
1177 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1178 struct list_head
*work_q
,
1179 struct list_head
*done_q
)
1181 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1182 unsigned int channel
;
1186 * we really want to loop over the various channels, and do this on
1187 * a channel by channel basis. we should also check to see if any
1188 * of the failed commands are on soft_reset devices, and if so, skip
1192 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1194 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1195 if (channel
== scmd_channel(scmd
)) {
1199 * FIXME add back in some support for
1200 * soft_reset devices.
1207 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1208 " %d\n", current
->comm
,
1210 rtn
= scsi_try_bus_reset(chan_scmd
);
1211 if (rtn
== SUCCESS
) {
1212 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1213 if (channel
== scmd_channel(scmd
))
1214 if (!scsi_device_online(scmd
->device
) ||
1216 scsi_eh_finish_cmd(scmd
,
1220 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1221 " failed chan: %d\n",
1226 return list_empty(work_q
);
1230 * scsi_eh_host_reset - send a host reset
1231 * @work_q: list_head for processed commands.
1232 * @done_q: list_head for processed commands.
1234 static int scsi_eh_host_reset(struct list_head
*work_q
,
1235 struct list_head
*done_q
)
1237 struct scsi_cmnd
*scmd
, *next
;
1240 if (!list_empty(work_q
)) {
1241 scmd
= list_entry(work_q
->next
,
1242 struct scsi_cmnd
, eh_entry
);
1244 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1247 rtn
= scsi_try_host_reset(scmd
);
1248 if (rtn
== SUCCESS
) {
1249 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1250 if (!scsi_device_online(scmd
->device
) ||
1251 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1253 scsi_eh_finish_cmd(scmd
, done_q
);
1256 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1261 return list_empty(work_q
);
1265 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1266 * @work_q: list_head for processed commands.
1267 * @done_q: list_head for processed commands.
1269 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1270 struct list_head
*done_q
)
1272 struct scsi_cmnd
*scmd
, *next
;
1274 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1275 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1276 "not ready after error recovery\n");
1277 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1278 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1280 * FIXME: Handle lost cmds.
1283 scsi_eh_finish_cmd(scmd
, done_q
);
1289 * scsi_noretry_cmd - determinte if command should be failed fast
1290 * @scmd: SCSI cmd to examine.
1292 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1294 switch (host_byte(scmd
->result
)) {
1298 return blk_failfast_transport(scmd
->request
);
1300 return blk_failfast_dev(scmd
->request
);
1302 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1303 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1306 case DID_SOFT_ERROR
:
1307 return blk_failfast_driver(scmd
->request
);
1310 switch (status_byte(scmd
->result
)) {
1311 case CHECK_CONDITION
:
1313 * assume caller has checked sense and determinted
1314 * the check condition was retryable.
1316 return blk_failfast_dev(scmd
->request
);
1323 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1324 * @scmd: SCSI cmd to examine.
1327 * This is *only* called when we are examining the status after sending
1328 * out the actual data command. any commands that are queued for error
1329 * recovery (e.g. test_unit_ready) do *not* come through here.
1331 * When this routine returns failed, it means the error handler thread
1332 * is woken. In cases where the error code indicates an error that
1333 * doesn't require the error handler read (i.e. we don't need to
1334 * abort/reset), this function should return SUCCESS.
1336 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1341 * if the device is offline, then we clearly just pass the result back
1342 * up to the top level.
1344 if (!scsi_device_online(scmd
->device
)) {
1345 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1352 * first check the host byte, to see if there is anything in there
1353 * that would indicate what we need to do.
1355 switch (host_byte(scmd
->result
)) {
1356 case DID_PASSTHROUGH
:
1358 * no matter what, pass this through to the upper layer.
1359 * nuke this special code so that it looks like we are saying
1362 scmd
->result
&= 0xff00ffff;
1366 * looks good. drop through, and check the next byte.
1369 case DID_NO_CONNECT
:
1370 case DID_BAD_TARGET
:
1373 * note - this means that we just report the status back
1374 * to the top level driver, not that we actually think
1375 * that it indicates SUCCESS.
1379 * when the low level driver returns did_soft_error,
1380 * it is responsible for keeping an internal retry counter
1381 * in order to avoid endless loops (db)
1383 * actually this is a bug in this function here. we should
1384 * be mindful of the maximum number of retries specified
1385 * and not get stuck in a loop.
1387 case DID_SOFT_ERROR
:
1393 return ADD_TO_MLQUEUE
;
1394 case DID_TRANSPORT_DISRUPTED
:
1396 * LLD/transport was disrupted during processing of the IO.
1397 * The transport class is now blocked/blocking,
1398 * and the transport will decide what to do with the IO
1399 * based on its timers and recovery capablilities if
1400 * there are enough retries.
1403 case DID_TRANSPORT_FAILFAST
:
1405 * The transport decided to failfast the IO (most likely
1406 * the fast io fail tmo fired), so send IO directly upwards.
1410 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1411 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1413 * execute reservation conflict processing code
1424 * when we scan the bus, we get timeout messages for
1425 * these commands if there is no device available.
1426 * other hosts report did_no_connect for the same thing.
1428 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1429 scmd
->cmnd
[0] == INQUIRY
)) {
1441 * next, check the message byte.
1443 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1447 * check the status byte to see if this indicates anything special.
1449 switch (status_byte(scmd
->result
)) {
1451 scsi_handle_queue_full(scmd
->device
);
1453 * the case of trying to send too many commands to a
1454 * tagged queueing device.
1458 * device can't talk to us at the moment. Should only
1459 * occur (SAM-3) when the task queue is empty, so will cause
1460 * the empty queue handling to trigger a stall in the
1463 return ADD_TO_MLQUEUE
;
1465 scsi_handle_queue_ramp_up(scmd
->device
);
1466 case COMMAND_TERMINATED
:
1470 case CHECK_CONDITION
:
1471 rtn
= scsi_check_sense(scmd
);
1472 if (rtn
== NEEDS_RETRY
)
1474 /* if rtn == FAILED, we have no sense information;
1475 * returning FAILED will wake the error handler thread
1476 * to collect the sense and redo the decide
1479 case CONDITION_GOOD
:
1480 case INTERMEDIATE_GOOD
:
1481 case INTERMEDIATE_C_GOOD
:
1484 * who knows? FIXME(eric)
1488 case RESERVATION_CONFLICT
:
1489 sdev_printk(KERN_INFO
, scmd
->device
,
1490 "reservation conflict\n");
1491 return SUCCESS
; /* causes immediate i/o error */
1499 /* we requeue for retry because the error was retryable, and
1500 * the request was not marked fast fail. Note that above,
1501 * even if the request is marked fast fail, we still requeue
1502 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1503 if ((++scmd
->retries
) <= scmd
->allowed
1504 && !scsi_noretry_cmd(scmd
)) {
1508 * no more retries - report this one back to upper level.
1514 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1516 __blk_put_request(req
->q
, req
);
1520 * scsi_eh_lock_door - Prevent medium removal for the specified device
1521 * @sdev: SCSI device to prevent medium removal
1524 * We must be called from process context.
1527 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1528 * head of the devices request queue, and continue.
1530 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1532 struct request
*req
;
1535 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1536 * request becomes available
1538 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1540 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1544 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1547 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1549 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1550 req
->cmd_flags
|= REQ_QUIET
;
1551 req
->timeout
= 10 * HZ
;
1554 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1558 * scsi_restart_operations - restart io operations to the specified host.
1559 * @shost: Host we are restarting.
1562 * When we entered the error handler, we blocked all further i/o to
1563 * this device. we need to 'reverse' this process.
1565 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1567 struct scsi_device
*sdev
;
1568 unsigned long flags
;
1571 * If the door was locked, we need to insert a door lock request
1572 * onto the head of the SCSI request queue for the device. There
1573 * is no point trying to lock the door of an off-line device.
1575 shost_for_each_device(sdev
, shost
) {
1576 if (scsi_device_online(sdev
) && sdev
->locked
)
1577 scsi_eh_lock_door(sdev
);
1581 * next free up anything directly waiting upon the host. this
1582 * will be requests for character device operations, and also for
1583 * ioctls to queued block devices.
1585 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1588 spin_lock_irqsave(shost
->host_lock
, flags
);
1589 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1590 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1591 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1592 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1594 wake_up(&shost
->host_wait
);
1597 * finally we need to re-initiate requests that may be pending. we will
1598 * have had everything blocked while error handling is taking place, and
1599 * now that error recovery is done, we will need to ensure that these
1600 * requests are started.
1602 scsi_run_host_queues(shost
);
1606 * scsi_eh_ready_devs - check device ready state and recover if not.
1607 * @shost: host to be recovered.
1608 * @work_q: &list_head for pending commands.
1609 * @done_q: &list_head for processed commands.
1611 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1612 struct list_head
*work_q
,
1613 struct list_head
*done_q
)
1615 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1616 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1617 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1618 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1619 if (!scsi_eh_host_reset(work_q
, done_q
))
1620 scsi_eh_offline_sdevs(work_q
,
1623 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1626 * scsi_eh_flush_done_q - finish processed commands or retry them.
1627 * @done_q: list_head of processed commands.
1629 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1631 struct scsi_cmnd
*scmd
, *next
;
1633 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1634 list_del_init(&scmd
->eh_entry
);
1635 if (scsi_device_online(scmd
->device
) &&
1636 !scsi_noretry_cmd(scmd
) &&
1637 (++scmd
->retries
<= scmd
->allowed
)) {
1638 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1642 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1645 * If just we got sense for the device (called
1646 * scsi_eh_get_sense), scmd->result is already
1647 * set, do not set DRIVER_TIMEOUT.
1650 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1651 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1653 current
->comm
, scmd
));
1654 scsi_finish_command(scmd
);
1658 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1661 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1662 * @shost: Host to unjam.
1665 * When we come in here, we *know* that all commands on the bus have
1666 * either completed, failed or timed out. we also know that no further
1667 * commands are being sent to the host, so things are relatively quiet
1668 * and we have freedom to fiddle with things as we wish.
1670 * This is only the *default* implementation. it is possible for
1671 * individual drivers to supply their own version of this function, and
1672 * if the maintainer wishes to do this, it is strongly suggested that
1673 * this function be taken as a template and modified. this function
1674 * was designed to correctly handle problems for about 95% of the
1675 * different cases out there, and it should always provide at least a
1676 * reasonable amount of error recovery.
1678 * Any command marked 'failed' or 'timeout' must eventually have
1679 * scsi_finish_cmd() called for it. we do all of the retry stuff
1680 * here, so when we restart the host after we return it should have an
1683 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1685 unsigned long flags
;
1686 LIST_HEAD(eh_work_q
);
1687 LIST_HEAD(eh_done_q
);
1689 spin_lock_irqsave(shost
->host_lock
, flags
);
1690 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1691 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1693 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1695 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1696 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1697 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1699 scsi_eh_flush_done_q(&eh_done_q
);
1703 * scsi_error_handler - SCSI error handler thread
1704 * @data: Host for which we are running.
1707 * This is the main error handling loop. This is run as a kernel thread
1708 * for every SCSI host and handles all error handling activity.
1710 int scsi_error_handler(void *data
)
1712 struct Scsi_Host
*shost
= data
;
1715 * We use TASK_INTERRUPTIBLE so that the thread is not
1716 * counted against the load average as a running process.
1717 * We never actually get interrupted because kthread_run
1718 * disables signal delivery for the created thread.
1720 set_current_state(TASK_INTERRUPTIBLE
);
1721 while (!kthread_should_stop()) {
1722 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1723 shost
->host_failed
!= shost
->host_busy
) {
1724 SCSI_LOG_ERROR_RECOVERY(1,
1725 printk("Error handler scsi_eh_%d sleeping\n",
1728 set_current_state(TASK_INTERRUPTIBLE
);
1732 __set_current_state(TASK_RUNNING
);
1733 SCSI_LOG_ERROR_RECOVERY(1,
1734 printk("Error handler scsi_eh_%d waking up\n",
1738 * We have a host that is failing for some reason. Figure out
1739 * what we need to do to get it up and online again (if we can).
1740 * If we fail, we end up taking the thing offline.
1742 if (shost
->transportt
->eh_strategy_handler
)
1743 shost
->transportt
->eh_strategy_handler(shost
);
1745 scsi_unjam_host(shost
);
1748 * Note - if the above fails completely, the action is to take
1749 * individual devices offline and flush the queue of any
1750 * outstanding requests that may have been pending. When we
1751 * restart, we restart any I/O to any other devices on the bus
1752 * which are still online.
1754 scsi_restart_operations(shost
);
1755 set_current_state(TASK_INTERRUPTIBLE
);
1757 __set_current_state(TASK_RUNNING
);
1759 SCSI_LOG_ERROR_RECOVERY(1,
1760 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1761 shost
->ehandler
= NULL
;
1766 * Function: scsi_report_bus_reset()
1768 * Purpose: Utility function used by low-level drivers to report that
1769 * they have observed a bus reset on the bus being handled.
1771 * Arguments: shost - Host in question
1772 * channel - channel on which reset was observed.
1776 * Lock status: Host lock must be held.
1778 * Notes: This only needs to be called if the reset is one which
1779 * originates from an unknown location. Resets originated
1780 * by the mid-level itself don't need to call this, but there
1781 * should be no harm.
1783 * The main purpose of this is to make sure that a CHECK_CONDITION
1784 * is properly treated.
1786 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1788 struct scsi_device
*sdev
;
1790 __shost_for_each_device(sdev
, shost
) {
1791 if (channel
== sdev_channel(sdev
))
1792 __scsi_report_device_reset(sdev
, NULL
);
1795 EXPORT_SYMBOL(scsi_report_bus_reset
);
1798 * Function: scsi_report_device_reset()
1800 * Purpose: Utility function used by low-level drivers to report that
1801 * they have observed a device reset on the device being handled.
1803 * Arguments: shost - Host in question
1804 * channel - channel on which reset was observed
1805 * target - target on which reset was observed
1809 * Lock status: Host lock must be held
1811 * Notes: This only needs to be called if the reset is one which
1812 * originates from an unknown location. Resets originated
1813 * by the mid-level itself don't need to call this, but there
1814 * should be no harm.
1816 * The main purpose of this is to make sure that a CHECK_CONDITION
1817 * is properly treated.
1819 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1821 struct scsi_device
*sdev
;
1823 __shost_for_each_device(sdev
, shost
) {
1824 if (channel
== sdev_channel(sdev
) &&
1825 target
== sdev_id(sdev
))
1826 __scsi_report_device_reset(sdev
, NULL
);
1829 EXPORT_SYMBOL(scsi_report_device_reset
);
1832 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1837 * Function: scsi_reset_provider
1839 * Purpose: Send requested reset to a bus or device at any phase.
1841 * Arguments: device - device to send reset to
1842 * flag - reset type (see scsi.h)
1844 * Returns: SUCCESS/FAILURE.
1846 * Notes: This is used by the SCSI Generic driver to provide
1847 * Bus/Device reset capability.
1850 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1852 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1853 struct Scsi_Host
*shost
= dev
->host
;
1855 unsigned long flags
;
1858 blk_rq_init(NULL
, &req
);
1859 scmd
->request
= &req
;
1861 scmd
->cmnd
= req
.cmd
;
1863 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1864 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1868 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1870 spin_lock_irqsave(shost
->host_lock
, flags
);
1871 shost
->tmf_in_progress
= 1;
1872 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1875 case SCSI_TRY_RESET_DEVICE
:
1876 rtn
= scsi_try_bus_device_reset(scmd
);
1880 case SCSI_TRY_RESET_TARGET
:
1881 rtn
= scsi_try_target_reset(scmd
);
1885 case SCSI_TRY_RESET_BUS
:
1886 rtn
= scsi_try_bus_reset(scmd
);
1890 case SCSI_TRY_RESET_HOST
:
1891 rtn
= scsi_try_host_reset(scmd
);
1897 spin_lock_irqsave(shost
->host_lock
, flags
);
1898 shost
->tmf_in_progress
= 0;
1899 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1902 * be sure to wake up anyone who was sleeping or had their queue
1903 * suspended while we performed the TMF.
1905 SCSI_LOG_ERROR_RECOVERY(3,
1906 printk("%s: waking up host to restart after TMF\n",
1909 wake_up(&shost
->host_wait
);
1911 scsi_run_host_queues(shost
);
1913 scsi_next_command(scmd
);
1916 EXPORT_SYMBOL(scsi_reset_provider
);
1919 * scsi_normalize_sense - normalize main elements from either fixed or
1920 * descriptor sense data format into a common format.
1922 * @sense_buffer: byte array containing sense data returned by device
1923 * @sb_len: number of valid bytes in sense_buffer
1924 * @sshdr: pointer to instance of structure that common
1925 * elements are written to.
1928 * The "main elements" from sense data are: response_code, sense_key,
1929 * asc, ascq and additional_length (only for descriptor format).
1931 * Typically this function can be called after a device has
1932 * responded to a SCSI command with the CHECK_CONDITION status.
1935 * 1 if valid sense data information found, else 0;
1937 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1938 struct scsi_sense_hdr
*sshdr
)
1940 if (!sense_buffer
|| !sb_len
)
1943 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1945 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1947 if (!scsi_sense_valid(sshdr
))
1950 if (sshdr
->response_code
>= 0x72) {
1955 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1957 sshdr
->asc
= sense_buffer
[2];
1959 sshdr
->ascq
= sense_buffer
[3];
1961 sshdr
->additional_length
= sense_buffer
[7];
1967 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1969 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1970 sb_len
: (sense_buffer
[7] + 8);
1972 sshdr
->asc
= sense_buffer
[12];
1974 sshdr
->ascq
= sense_buffer
[13];
1980 EXPORT_SYMBOL(scsi_normalize_sense
);
1982 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1983 struct scsi_sense_hdr
*sshdr
)
1985 return scsi_normalize_sense(cmd
->sense_buffer
,
1986 SCSI_SENSE_BUFFERSIZE
, sshdr
);
1988 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1991 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1992 * @sense_buffer: byte array of descriptor format sense data
1993 * @sb_len: number of valid bytes in sense_buffer
1994 * @desc_type: value of descriptor type to find
1995 * (e.g. 0 -> information)
1998 * only valid when sense data is in descriptor format
2001 * pointer to start of (first) descriptor if found else NULL
2003 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2006 int add_sen_len
, add_len
, desc_len
, k
;
2009 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2011 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2013 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2014 add_sen_len
: (sb_len
- 8);
2015 descp
= &sense_buffer
[8];
2016 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2018 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2019 desc_len
= add_len
+ 2;
2020 if (descp
[0] == desc_type
)
2022 if (add_len
< 0) // short descriptor ??
2027 EXPORT_SYMBOL(scsi_sense_desc_find
);
2030 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2031 * @sense_buffer: byte array of sense data
2032 * @sb_len: number of valid bytes in sense_buffer
2033 * @info_out: pointer to 64 integer where 8 or 4 byte information
2034 * field will be placed if found.
2037 * 1 if information field found, 0 if not found.
2039 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2048 switch (sense_buffer
[0] & 0x7f) {
2051 if (sense_buffer
[0] & 0x80) {
2052 *info_out
= (sense_buffer
[3] << 24) +
2053 (sense_buffer
[4] << 16) +
2054 (sense_buffer
[5] << 8) + sense_buffer
[6];
2060 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2062 if (ucp
&& (0xa == ucp
[1])) {
2064 for (j
= 0; j
< 8; ++j
) {
2077 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2080 * scsi_build_sense_buffer - build sense data in a buffer
2081 * @desc: Sense format (non zero == descriptor format,
2082 * 0 == fixed format)
2083 * @buf: Where to build sense data
2085 * @asc: Additional sense code
2086 * @ascq: Additional sense code qualifier
2089 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2092 buf
[0] = 0x72; /* descriptor, current */
2098 buf
[0] = 0x70; /* fixed, current */
2105 EXPORT_SYMBOL(scsi_build_sense_buffer
);