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/gfp.h>
20 #include <linux/timer.h>
21 #include <linux/string.h>
22 #include <linux/kernel.h>
23 #include <linux/freezer.h>
24 #include <linux/kthread.h>
25 #include <linux/interrupt.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
38 #include "scsi_priv.h"
39 #include "scsi_logging.h"
40 #include "scsi_transport_api.h"
42 #define SENSE_TIMEOUT (10*HZ)
45 * These should *probably* be handled by the host itself.
46 * Since it is allowed to sleep, it probably should.
48 #define BUS_RESET_SETTLE_TIME (10)
49 #define HOST_RESET_SETTLE_TIME (10)
51 /* called with shost->host_lock held */
52 void scsi_eh_wakeup(struct Scsi_Host
*shost
)
54 if (shost
->host_busy
== shost
->host_failed
) {
55 wake_up_process(shost
->ehandler
);
56 SCSI_LOG_ERROR_RECOVERY(5,
57 printk("Waking error handler thread\n"));
62 * scsi_schedule_eh - schedule EH for SCSI host
63 * @shost: SCSI host to invoke error handling on.
65 * Schedule SCSI EH without scmd.
67 void scsi_schedule_eh(struct Scsi_Host
*shost
)
71 spin_lock_irqsave(shost
->host_lock
, flags
);
73 if (scsi_host_set_state(shost
, SHOST_RECOVERY
) == 0 ||
74 scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
) == 0) {
75 shost
->host_eh_scheduled
++;
76 scsi_eh_wakeup(shost
);
79 spin_unlock_irqrestore(shost
->host_lock
, flags
);
81 EXPORT_SYMBOL_GPL(scsi_schedule_eh
);
84 * scsi_eh_scmd_add - add scsi cmd to error handling.
85 * @scmd: scmd to run eh on.
86 * @eh_flag: optional SCSI_EH flag.
91 int scsi_eh_scmd_add(struct scsi_cmnd
*scmd
, int eh_flag
)
93 struct Scsi_Host
*shost
= scmd
->device
->host
;
100 spin_lock_irqsave(shost
->host_lock
, flags
);
101 if (scsi_host_set_state(shost
, SHOST_RECOVERY
))
102 if (scsi_host_set_state(shost
, SHOST_CANCEL_RECOVERY
))
106 scmd
->eh_eflags
|= eh_flag
;
107 list_add_tail(&scmd
->eh_entry
, &shost
->eh_cmd_q
);
108 shost
->host_failed
++;
109 scsi_eh_wakeup(shost
);
111 spin_unlock_irqrestore(shost
->host_lock
, flags
);
116 * scsi_times_out - Timeout function for normal scsi commands.
117 * @req: request that is timing out.
120 * We do not need to lock this. There is the potential for a race
121 * only in that the normal completion handling might run, but if the
122 * normal completion function determines that the timer has already
123 * fired, then it mustn't do anything.
125 enum blk_eh_timer_return
scsi_times_out(struct request
*req
)
127 struct scsi_cmnd
*scmd
= req
->special
;
128 enum blk_eh_timer_return rtn
= BLK_EH_NOT_HANDLED
;
130 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
132 if (scmd
->device
->host
->transportt
->eh_timed_out
)
133 rtn
= scmd
->device
->host
->transportt
->eh_timed_out(scmd
);
134 else if (scmd
->device
->host
->hostt
->eh_timed_out
)
135 rtn
= scmd
->device
->host
->hostt
->eh_timed_out(scmd
);
137 if (unlikely(rtn
== BLK_EH_NOT_HANDLED
&&
138 !scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
139 scmd
->result
|= DID_TIME_OUT
<< 16;
140 rtn
= BLK_EH_HANDLED
;
147 * scsi_block_when_processing_errors - Prevent cmds from being queued.
148 * @sdev: Device on which we are performing recovery.
151 * We block until the host is out of error recovery, and then check to
152 * see whether the host or the device is offline.
155 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
157 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
161 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
163 online
= scsi_device_online(sdev
);
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__
,
170 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
172 #ifdef CONFIG_SCSI_LOGGING
174 * scsi_eh_prt_fail_stats - Log info on failures.
175 * @shost: scsi host being recovered.
176 * @work_q: Queue of scsi cmds to process.
178 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
179 struct list_head
*work_q
)
181 struct scsi_cmnd
*scmd
;
182 struct scsi_device
*sdev
;
183 int total_failures
= 0;
186 int devices_failed
= 0;
188 shost_for_each_device(sdev
, shost
) {
189 list_for_each_entry(scmd
, work_q
, eh_entry
) {
190 if (scmd
->device
== sdev
) {
192 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
199 if (cmd_cancel
|| cmd_failed
) {
200 SCSI_LOG_ERROR_RECOVERY(3,
201 sdev_printk(KERN_INFO
, sdev
,
202 "%s: cmds failed: %d, cancel: %d\n",
203 __func__
, cmd_failed
,
211 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
212 " devices require eh work\n",
213 total_failures
, devices_failed
));
218 * scsi_check_sense - Examine scsi cmd sense
219 * @scmd: Cmd to have sense checked.
222 * SUCCESS or FAILED or NEEDS_RETRY
225 * When a deferred error is detected the current command has
226 * not been executed and needs retrying.
228 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
230 struct scsi_device
*sdev
= scmd
->device
;
231 struct scsi_sense_hdr sshdr
;
233 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
234 return FAILED
; /* no valid sense data */
236 if (scsi_sense_is_deferred(&sshdr
))
239 if (sdev
->scsi_dh_data
&& sdev
->scsi_dh_data
->scsi_dh
&&
240 sdev
->scsi_dh_data
->scsi_dh
->check_sense
) {
243 rc
= sdev
->scsi_dh_data
->scsi_dh
->check_sense(sdev
, &sshdr
);
244 if (rc
!= SCSI_RETURN_NOT_HANDLED
)
246 /* handler does not care. Drop down to default handling */
250 * Previous logic looked for FILEMARK, EOM or ILI which are
251 * mainly associated with tapes and returned SUCCESS.
253 if (sshdr
.response_code
== 0x70) {
255 if (scmd
->sense_buffer
[2] & 0xe0)
259 * descriptor format: look for "stream commands sense data
260 * descriptor" (see SSC-3). Assume single sense data
261 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
263 if ((sshdr
.additional_length
> 3) &&
264 (scmd
->sense_buffer
[8] == 0x4) &&
265 (scmd
->sense_buffer
[11] & 0xe0))
269 switch (sshdr
.sense_key
) {
272 case RECOVERED_ERROR
:
273 return /* soft_error */ SUCCESS
;
275 case ABORTED_COMMAND
:
276 if (sshdr
.asc
== 0x10) /* DIF */
283 * if we are expecting a cc/ua because of a bus reset that we
284 * performed, treat this just as a retry. otherwise this is
285 * information that we should pass up to the upper-level driver
286 * so that we can deal with it there.
288 if (scmd
->device
->expecting_cc_ua
) {
289 scmd
->device
->expecting_cc_ua
= 0;
293 * if the device is in the process of becoming ready, we
296 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
299 * if the device is not started, we need to wake
300 * the error handler to start the motor
302 if (scmd
->device
->allow_restart
&&
303 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
306 if (blk_barrier_rq(scmd
->request
))
308 * barrier requests should always retry on UA
309 * otherwise block will get a spurious error
314 * for normal (non barrier) commands, pass the
315 * UA upwards for a determination in the
316 * completion functions
320 /* these three are not supported */
322 case VOLUME_OVERFLOW
:
327 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
328 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
329 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
335 if (scmd
->device
->retry_hwerror
)
336 return ADD_TO_MLQUEUE
;
340 case ILLEGAL_REQUEST
:
348 static void scsi_handle_queue_ramp_up(struct scsi_device
*sdev
)
350 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
351 struct scsi_device
*tmp_sdev
;
353 if (!sht
->change_queue_depth
||
354 sdev
->queue_depth
>= sdev
->max_queue_depth
)
357 if (time_before(jiffies
,
358 sdev
->last_queue_ramp_up
+ sdev
->queue_ramp_up_period
))
361 if (time_before(jiffies
,
362 sdev
->last_queue_full_time
+ sdev
->queue_ramp_up_period
))
366 * Walk all devices of a target and do
369 shost_for_each_device(tmp_sdev
, sdev
->host
) {
370 if (tmp_sdev
->channel
!= sdev
->channel
||
371 tmp_sdev
->id
!= sdev
->id
||
372 tmp_sdev
->queue_depth
== sdev
->max_queue_depth
)
375 * call back into LLD to increase queue_depth by one
376 * with ramp up reason code.
378 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
+ 1,
379 SCSI_QDEPTH_RAMP_UP
);
380 sdev
->last_queue_ramp_up
= jiffies
;
384 static void scsi_handle_queue_full(struct scsi_device
*sdev
)
386 struct scsi_host_template
*sht
= sdev
->host
->hostt
;
387 struct scsi_device
*tmp_sdev
;
389 if (!sht
->change_queue_depth
)
392 shost_for_each_device(tmp_sdev
, sdev
->host
) {
393 if (tmp_sdev
->channel
!= sdev
->channel
||
394 tmp_sdev
->id
!= sdev
->id
)
397 * We do not know the number of commands that were at
398 * the device when we got the queue full so we start
399 * from the highest possible value and work our way down.
401 sht
->change_queue_depth(tmp_sdev
, tmp_sdev
->queue_depth
- 1,
407 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
408 * @scmd: SCSI cmd to examine.
411 * This is *only* called when we are examining the status of commands
412 * queued during error recovery. the main difference here is that we
413 * don't allow for the possibility of retries here, and we are a lot
414 * more restrictive about what we consider acceptable.
416 static int scsi_eh_completed_normally(struct scsi_cmnd
*scmd
)
419 * first check the host byte, to see if there is anything in there
420 * that would indicate what we need to do.
422 if (host_byte(scmd
->result
) == DID_RESET
) {
424 * rats. we are already in the error handler, so we now
425 * get to try and figure out what to do next. if the sense
426 * is valid, we have a pretty good idea of what to do.
427 * if not, we mark it as FAILED.
429 return scsi_check_sense(scmd
);
431 if (host_byte(scmd
->result
) != DID_OK
)
435 * next, check the message byte.
437 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
441 * now, check the status byte to see if this indicates
444 switch (status_byte(scmd
->result
)) {
446 scsi_handle_queue_ramp_up(scmd
->device
);
447 case COMMAND_TERMINATED
:
449 case CHECK_CONDITION
:
450 return scsi_check_sense(scmd
);
452 case INTERMEDIATE_GOOD
:
453 case INTERMEDIATE_C_GOOD
:
455 * who knows? FIXME(eric)
458 case RESERVATION_CONFLICT
:
460 * let issuer deal with this, it could be just fine
464 scsi_handle_queue_full(scmd
->device
);
474 * scsi_eh_done - Completion function for error handling.
475 * @scmd: Cmd that is done.
477 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
479 struct completion
*eh_action
;
481 SCSI_LOG_ERROR_RECOVERY(3,
482 printk("%s scmd: %p result: %x\n",
483 __func__
, scmd
, scmd
->result
));
485 eh_action
= scmd
->device
->host
->eh_action
;
491 * scsi_try_host_reset - ask host adapter to reset itself
492 * @scmd: SCSI cmd to send hsot reset.
494 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
499 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
502 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
505 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
507 if (rtn
== SUCCESS
) {
508 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
509 ssleep(HOST_RESET_SETTLE_TIME
);
510 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
511 scsi_report_bus_reset(scmd
->device
->host
,
513 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
520 * scsi_try_bus_reset - ask host to perform a bus reset
521 * @scmd: SCSI cmd to send bus reset.
523 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
528 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
531 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
534 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
536 if (rtn
== SUCCESS
) {
537 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
538 ssleep(BUS_RESET_SETTLE_TIME
);
539 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
540 scsi_report_bus_reset(scmd
->device
->host
,
542 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
548 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
551 sdev
->expecting_cc_ua
= 1;
555 * scsi_try_target_reset - Ask host to perform a target reset
556 * @scmd: SCSI cmd used to send a target reset
559 * There is no timeout for this operation. if this operation is
560 * unreliable for a given host, then the host itself needs to put a
561 * timer on it, and set the host back to a consistent state prior to
564 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
569 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
572 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
573 if (rtn
== SUCCESS
) {
574 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
575 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
576 __scsi_report_device_reset
);
577 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
584 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
585 * @scmd: SCSI cmd used to send BDR
588 * There is no timeout for this operation. if this operation is
589 * unreliable for a given host, then the host itself needs to put a
590 * timer on it, and set the host back to a consistent state prior to
593 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
597 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
600 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
602 __scsi_report_device_reset(scmd
->device
, NULL
);
606 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
608 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
611 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
615 * scsi_try_to_abort_cmd - Ask host to abort a running command.
616 * @scmd: SCSI cmd to abort from Lower Level.
619 * This function will not return until the user's completion function
620 * has been called. there is no timeout on this operation. if the
621 * author of the low-level driver wishes this operation to be timed,
622 * they can provide this facility themselves. helper functions in
623 * scsi_error.c can be supplied to make this easier to do.
625 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
628 * scsi_done was called just after the command timed out and before
629 * we had a chance to process it. (db)
631 if (scmd
->serial_number
== 0)
633 return __scsi_try_to_abort_cmd(scmd
);
636 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
638 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
639 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
640 if (scsi_try_target_reset(scmd
) != SUCCESS
)
641 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
642 scsi_try_host_reset(scmd
);
646 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
647 * @scmd: SCSI command structure to hijack
648 * @ses: structure to save restore information
649 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
650 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
651 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
653 * This function is used to save a scsi command information before re-execution
654 * as part of the error recovery process. If @sense_bytes is 0 the command
655 * sent must be one that does not transfer any data. If @sense_bytes != 0
656 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
657 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
659 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
660 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
662 struct scsi_device
*sdev
= scmd
->device
;
665 * We need saved copies of a number of fields - this is because
666 * error handling may need to overwrite these with different values
667 * to run different commands, and once error handling is complete,
668 * we will need to restore these values prior to running the actual
671 ses
->cmd_len
= scmd
->cmd_len
;
672 ses
->cmnd
= scmd
->cmnd
;
673 ses
->data_direction
= scmd
->sc_data_direction
;
674 ses
->sdb
= scmd
->sdb
;
675 ses
->next_rq
= scmd
->request
->next_rq
;
676 ses
->result
= scmd
->result
;
677 ses
->underflow
= scmd
->underflow
;
678 ses
->prot_op
= scmd
->prot_op
;
680 scmd
->prot_op
= SCSI_PROT_NORMAL
;
681 scmd
->cmnd
= ses
->eh_cmnd
;
682 memset(scmd
->cmnd
, 0, BLK_MAX_CDB
);
683 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
684 scmd
->request
->next_rq
= NULL
;
687 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
689 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
691 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
692 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
693 scmd
->sdb
.table
.nents
= 1;
694 scmd
->cmnd
[0] = REQUEST_SENSE
;
695 scmd
->cmnd
[4] = scmd
->sdb
.length
;
696 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
698 scmd
->sc_data_direction
= DMA_NONE
;
700 BUG_ON(cmnd_size
> BLK_MAX_CDB
);
701 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
702 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
708 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
709 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
710 (sdev
->lun
<< 5 & 0xe0);
713 * Zero the sense buffer. The scsi spec mandates that any
714 * untransferred sense data should be interpreted as being zero.
716 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
718 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
721 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
722 * @scmd: SCSI command structure to restore
723 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
725 * Undo any damage done by above scsi_eh_prep_cmnd().
727 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
730 * Restore original data
732 scmd
->cmd_len
= ses
->cmd_len
;
733 scmd
->cmnd
= ses
->cmnd
;
734 scmd
->sc_data_direction
= ses
->data_direction
;
735 scmd
->sdb
= ses
->sdb
;
736 scmd
->request
->next_rq
= ses
->next_rq
;
737 scmd
->result
= ses
->result
;
738 scmd
->underflow
= ses
->underflow
;
739 scmd
->prot_op
= ses
->prot_op
;
741 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
744 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
745 * @scmd: SCSI command structure to hijack
747 * @cmnd_size: size in bytes of @cmnd
748 * @timeout: timeout for this request
749 * @sense_bytes: size of sense data to copy or 0
751 * This function is used to send a scsi command down to a target device
752 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
755 * SUCCESS or FAILED or NEEDS_RETRY
757 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
758 int cmnd_size
, int timeout
, unsigned sense_bytes
)
760 struct scsi_device
*sdev
= scmd
->device
;
761 struct Scsi_Host
*shost
= sdev
->host
;
762 DECLARE_COMPLETION_ONSTACK(done
);
763 unsigned long timeleft
;
765 struct scsi_eh_save ses
;
768 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
769 shost
->eh_action
= &done
;
771 spin_lock_irqsave(shost
->host_lock
, flags
);
773 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
774 spin_unlock_irqrestore(shost
->host_lock
, flags
);
776 timeleft
= wait_for_completion_timeout(&done
, timeout
);
778 shost
->eh_action
= NULL
;
780 scsi_log_completion(scmd
, SUCCESS
);
782 SCSI_LOG_ERROR_RECOVERY(3,
783 printk("%s: scmd: %p, timeleft: %ld\n",
784 __func__
, scmd
, timeleft
));
787 * If there is time left scsi_eh_done got called, and we will
788 * examine the actual status codes to see whether the command
789 * actually did complete normally, else tell the host to forget
790 * about this command.
793 rtn
= scsi_eh_completed_normally(scmd
);
794 SCSI_LOG_ERROR_RECOVERY(3,
795 printk("%s: scsi_eh_completed_normally %x\n",
811 scsi_abort_eh_cmnd(scmd
);
815 scsi_eh_restore_cmnd(scmd
, &ses
);
820 * scsi_request_sense - Request sense data from a particular target.
821 * @scmd: SCSI cmd for request sense.
824 * Some hosts automatically obtain this information, others require
825 * that we obtain it on our own. This function will *not* return until
826 * the command either times out, or it completes.
828 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
830 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
834 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
835 * @scmd: Original SCSI cmd that eh has finished.
836 * @done_q: Queue for processed commands.
839 * We don't want to use the normal command completion while we are are
840 * still handling errors - it may cause other commands to be queued,
841 * and that would disturb what we are doing. Thus we really want to
842 * keep a list of pending commands for final completion, and once we
843 * are ready to leave error handling we handle completion for real.
845 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
847 scmd
->device
->host
->host_failed
--;
849 list_move_tail(&scmd
->eh_entry
, done_q
);
851 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
854 * scsi_eh_get_sense - Get device sense data.
855 * @work_q: Queue of commands to process.
856 * @done_q: Queue of processed commands.
859 * See if we need to request sense information. if so, then get it
860 * now, so we have a better idea of what to do.
863 * This has the unfortunate side effect that if a shost adapter does
864 * not automatically request sense information, we end up shutting
865 * it down before we request it.
867 * All drivers should request sense information internally these days,
868 * so for now all I have to say is tough noogies if you end up in here.
870 * XXX: Long term this code should go away, but that needs an audit of
873 int scsi_eh_get_sense(struct list_head
*work_q
,
874 struct list_head
*done_q
)
876 struct scsi_cmnd
*scmd
, *next
;
879 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
880 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
881 SCSI_SENSE_VALID(scmd
))
884 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
885 "%s: requesting sense\n",
887 rtn
= scsi_request_sense(scmd
);
891 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
892 " result %x\n", scmd
,
894 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
896 rtn
= scsi_decide_disposition(scmd
);
899 * if the result was normal, then just pass it along to the
903 /* we don't want this command reissued, just
904 * finished with the sense data, so set
905 * retries to the max allowed to ensure it
906 * won't get reissued */
907 scmd
->retries
= scmd
->allowed
;
908 else if (rtn
!= NEEDS_RETRY
)
911 scsi_eh_finish_cmd(scmd
, done_q
);
914 return list_empty(work_q
);
916 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
919 * scsi_eh_tur - Send TUR to device.
920 * @scmd: &scsi_cmnd to send TUR
923 * 0 - Device is ready. 1 - Device NOT ready.
925 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
927 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
928 int retry_cnt
= 1, rtn
;
931 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
933 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
934 __func__
, scmd
, rtn
));
949 * scsi_eh_abort_cmds - abort pending commands.
950 * @work_q: &list_head for pending commands.
951 * @done_q: &list_head for processed commands.
954 * Try and see whether or not it makes sense to try and abort the
955 * running command. This only works out to be the case if we have one
956 * command that has timed out. If the command simply failed, it makes
957 * no sense to try and abort the command, since as far as the shost
958 * adapter is concerned, it isn't running.
960 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
961 struct list_head
*done_q
)
963 struct scsi_cmnd
*scmd
, *next
;
966 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
967 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
969 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
970 "0x%p\n", current
->comm
,
972 rtn
= scsi_try_to_abort_cmd(scmd
);
973 if (rtn
== SUCCESS
) {
974 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
975 if (!scsi_device_online(scmd
->device
) ||
976 !scsi_eh_tur(scmd
)) {
977 scsi_eh_finish_cmd(scmd
, done_q
);
981 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
988 return list_empty(work_q
);
992 * scsi_eh_try_stu - Send START_UNIT to device.
993 * @scmd: &scsi_cmnd to send START_UNIT
996 * 0 - Device is ready. 1 - Device NOT ready.
998 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
1000 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
1002 if (scmd
->device
->allow_restart
) {
1003 int i
, rtn
= NEEDS_RETRY
;
1005 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
1006 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6, scmd
->device
->request_queue
->rq_timeout
, 0);
1016 * scsi_eh_stu - send START_UNIT if needed
1017 * @shost: &scsi host being recovered.
1018 * @work_q: &list_head for pending commands.
1019 * @done_q: &list_head for processed commands.
1022 * If commands are failing due to not ready, initializing command required,
1023 * try revalidating the device, which will end up sending a start unit.
1025 static int scsi_eh_stu(struct Scsi_Host
*shost
,
1026 struct list_head
*work_q
,
1027 struct list_head
*done_q
)
1029 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1030 struct scsi_device
*sdev
;
1032 shost_for_each_device(sdev
, shost
) {
1034 list_for_each_entry(scmd
, work_q
, eh_entry
)
1035 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1036 scsi_check_sense(scmd
) == FAILED
) {
1044 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1045 " 0x%p\n", current
->comm
, sdev
));
1047 if (!scsi_eh_try_stu(stu_scmd
)) {
1048 if (!scsi_device_online(sdev
) ||
1049 !scsi_eh_tur(stu_scmd
)) {
1050 list_for_each_entry_safe(scmd
, next
,
1052 if (scmd
->device
== sdev
)
1053 scsi_eh_finish_cmd(scmd
, done_q
);
1057 SCSI_LOG_ERROR_RECOVERY(3,
1058 printk("%s: START_UNIT failed to sdev:"
1059 " 0x%p\n", current
->comm
, sdev
));
1063 return list_empty(work_q
);
1068 * scsi_eh_bus_device_reset - send bdr if needed
1069 * @shost: scsi host being recovered.
1070 * @work_q: &list_head for pending commands.
1071 * @done_q: &list_head for processed commands.
1074 * Try a bus device reset. Still, look to see whether we have multiple
1075 * devices that are jammed or not - if we have multiple devices, it
1076 * makes no sense to try bus_device_reset - we really would need to try
1077 * a bus_reset instead.
1079 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1080 struct list_head
*work_q
,
1081 struct list_head
*done_q
)
1083 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1084 struct scsi_device
*sdev
;
1087 shost_for_each_device(sdev
, shost
) {
1089 list_for_each_entry(scmd
, work_q
, eh_entry
)
1090 if (scmd
->device
== sdev
) {
1098 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1099 " 0x%p\n", current
->comm
,
1101 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1102 if (rtn
== SUCCESS
) {
1103 if (!scsi_device_online(sdev
) ||
1104 !scsi_eh_tur(bdr_scmd
)) {
1105 list_for_each_entry_safe(scmd
, next
,
1107 if (scmd
->device
== sdev
)
1108 scsi_eh_finish_cmd(scmd
,
1113 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1121 return list_empty(work_q
);
1125 * scsi_eh_target_reset - send target reset if needed
1126 * @shost: scsi host being recovered.
1127 * @work_q: &list_head for pending commands.
1128 * @done_q: &list_head for processed commands.
1131 * Try a target reset.
1133 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1134 struct list_head
*work_q
,
1135 struct list_head
*done_q
)
1137 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1138 unsigned int id
= 0;
1143 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1144 if (id
== scmd_id(scmd
)) {
1150 /* not one exactly equal; find the next highest */
1151 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1152 if (scmd_id(scmd
) > id
&&
1154 scmd_id(tgtr_scmd
) > scmd_id(scmd
)))
1159 /* no more commands, that's it */
1162 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1164 current
->comm
, id
));
1165 rtn
= scsi_try_target_reset(tgtr_scmd
);
1166 if (rtn
== SUCCESS
) {
1167 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1168 if (id
== scmd_id(scmd
))
1169 if (!scsi_device_online(scmd
->device
) ||
1170 !scsi_eh_tur(tgtr_scmd
))
1171 scsi_eh_finish_cmd(scmd
,
1175 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1178 current
->comm
, id
));
1182 return list_empty(work_q
);
1186 * scsi_eh_bus_reset - send a bus reset
1187 * @shost: &scsi host being recovered.
1188 * @work_q: &list_head for pending commands.
1189 * @done_q: &list_head for processed commands.
1191 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1192 struct list_head
*work_q
,
1193 struct list_head
*done_q
)
1195 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1196 unsigned int channel
;
1200 * we really want to loop over the various channels, and do this on
1201 * a channel by channel basis. we should also check to see if any
1202 * of the failed commands are on soft_reset devices, and if so, skip
1206 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1208 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1209 if (channel
== scmd_channel(scmd
)) {
1213 * FIXME add back in some support for
1214 * soft_reset devices.
1221 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1222 " %d\n", current
->comm
,
1224 rtn
= scsi_try_bus_reset(chan_scmd
);
1225 if (rtn
== SUCCESS
) {
1226 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1227 if (channel
== scmd_channel(scmd
))
1228 if (!scsi_device_online(scmd
->device
) ||
1230 scsi_eh_finish_cmd(scmd
,
1234 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1235 " failed chan: %d\n",
1240 return list_empty(work_q
);
1244 * scsi_eh_host_reset - send a host reset
1245 * @work_q: list_head for processed commands.
1246 * @done_q: list_head for processed commands.
1248 static int scsi_eh_host_reset(struct list_head
*work_q
,
1249 struct list_head
*done_q
)
1251 struct scsi_cmnd
*scmd
, *next
;
1254 if (!list_empty(work_q
)) {
1255 scmd
= list_entry(work_q
->next
,
1256 struct scsi_cmnd
, eh_entry
);
1258 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1261 rtn
= scsi_try_host_reset(scmd
);
1262 if (rtn
== SUCCESS
) {
1263 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1264 if (!scsi_device_online(scmd
->device
) ||
1265 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1267 scsi_eh_finish_cmd(scmd
, done_q
);
1270 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1275 return list_empty(work_q
);
1279 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1280 * @work_q: list_head for processed commands.
1281 * @done_q: list_head for processed commands.
1283 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1284 struct list_head
*done_q
)
1286 struct scsi_cmnd
*scmd
, *next
;
1288 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1289 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1290 "not ready after error recovery\n");
1291 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1292 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1294 * FIXME: Handle lost cmds.
1297 scsi_eh_finish_cmd(scmd
, done_q
);
1303 * scsi_noretry_cmd - determinte if command should be failed fast
1304 * @scmd: SCSI cmd to examine.
1306 int scsi_noretry_cmd(struct scsi_cmnd
*scmd
)
1308 switch (host_byte(scmd
->result
)) {
1312 return blk_failfast_transport(scmd
->request
);
1314 return blk_failfast_dev(scmd
->request
);
1316 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1317 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1320 case DID_SOFT_ERROR
:
1321 return blk_failfast_driver(scmd
->request
);
1324 switch (status_byte(scmd
->result
)) {
1325 case CHECK_CONDITION
:
1327 * assume caller has checked sense and determinted
1328 * the check condition was retryable.
1330 return blk_failfast_dev(scmd
->request
);
1337 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1338 * @scmd: SCSI cmd to examine.
1341 * This is *only* called when we are examining the status after sending
1342 * out the actual data command. any commands that are queued for error
1343 * recovery (e.g. test_unit_ready) do *not* come through here.
1345 * When this routine returns failed, it means the error handler thread
1346 * is woken. In cases where the error code indicates an error that
1347 * doesn't require the error handler read (i.e. we don't need to
1348 * abort/reset), this function should return SUCCESS.
1350 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1355 * if the device is offline, then we clearly just pass the result back
1356 * up to the top level.
1358 if (!scsi_device_online(scmd
->device
)) {
1359 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1366 * first check the host byte, to see if there is anything in there
1367 * that would indicate what we need to do.
1369 switch (host_byte(scmd
->result
)) {
1370 case DID_PASSTHROUGH
:
1372 * no matter what, pass this through to the upper layer.
1373 * nuke this special code so that it looks like we are saying
1376 scmd
->result
&= 0xff00ffff;
1380 * looks good. drop through, and check the next byte.
1383 case DID_NO_CONNECT
:
1384 case DID_BAD_TARGET
:
1387 * note - this means that we just report the status back
1388 * to the top level driver, not that we actually think
1389 * that it indicates SUCCESS.
1393 * when the low level driver returns did_soft_error,
1394 * it is responsible for keeping an internal retry counter
1395 * in order to avoid endless loops (db)
1397 * actually this is a bug in this function here. we should
1398 * be mindful of the maximum number of retries specified
1399 * and not get stuck in a loop.
1401 case DID_SOFT_ERROR
:
1407 return ADD_TO_MLQUEUE
;
1408 case DID_TRANSPORT_DISRUPTED
:
1410 * LLD/transport was disrupted during processing of the IO.
1411 * The transport class is now blocked/blocking,
1412 * and the transport will decide what to do with the IO
1413 * based on its timers and recovery capablilities if
1414 * there are enough retries.
1417 case DID_TRANSPORT_FAILFAST
:
1419 * The transport decided to failfast the IO (most likely
1420 * the fast io fail tmo fired), so send IO directly upwards.
1424 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1425 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1427 * execute reservation conflict processing code
1438 * when we scan the bus, we get timeout messages for
1439 * these commands if there is no device available.
1440 * other hosts report did_no_connect for the same thing.
1442 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1443 scmd
->cmnd
[0] == INQUIRY
)) {
1455 * next, check the message byte.
1457 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1461 * check the status byte to see if this indicates anything special.
1463 switch (status_byte(scmd
->result
)) {
1465 scsi_handle_queue_full(scmd
->device
);
1467 * the case of trying to send too many commands to a
1468 * tagged queueing device.
1472 * device can't talk to us at the moment. Should only
1473 * occur (SAM-3) when the task queue is empty, so will cause
1474 * the empty queue handling to trigger a stall in the
1477 return ADD_TO_MLQUEUE
;
1479 scsi_handle_queue_ramp_up(scmd
->device
);
1480 case COMMAND_TERMINATED
:
1484 case CHECK_CONDITION
:
1485 rtn
= scsi_check_sense(scmd
);
1486 if (rtn
== NEEDS_RETRY
)
1488 /* if rtn == FAILED, we have no sense information;
1489 * returning FAILED will wake the error handler thread
1490 * to collect the sense and redo the decide
1493 case CONDITION_GOOD
:
1494 case INTERMEDIATE_GOOD
:
1495 case INTERMEDIATE_C_GOOD
:
1498 * who knows? FIXME(eric)
1502 case RESERVATION_CONFLICT
:
1503 sdev_printk(KERN_INFO
, scmd
->device
,
1504 "reservation conflict\n");
1505 return SUCCESS
; /* causes immediate i/o error */
1513 /* we requeue for retry because the error was retryable, and
1514 * the request was not marked fast fail. Note that above,
1515 * even if the request is marked fast fail, we still requeue
1516 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1517 if ((++scmd
->retries
) <= scmd
->allowed
1518 && !scsi_noretry_cmd(scmd
)) {
1522 * no more retries - report this one back to upper level.
1528 static void eh_lock_door_done(struct request
*req
, int uptodate
)
1530 __blk_put_request(req
->q
, req
);
1534 * scsi_eh_lock_door - Prevent medium removal for the specified device
1535 * @sdev: SCSI device to prevent medium removal
1538 * We must be called from process context.
1541 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1542 * head of the devices request queue, and continue.
1544 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1546 struct request
*req
;
1549 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1550 * request becomes available
1552 req
= blk_get_request(sdev
->request_queue
, READ
, GFP_KERNEL
);
1554 req
->cmd
[0] = ALLOW_MEDIUM_REMOVAL
;
1558 req
->cmd
[4] = SCSI_REMOVAL_PREVENT
;
1561 req
->cmd_len
= COMMAND_SIZE(req
->cmd
[0]);
1563 req
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1564 req
->cmd_flags
|= REQ_QUIET
;
1565 req
->timeout
= 10 * HZ
;
1568 blk_execute_rq_nowait(req
->q
, NULL
, req
, 1, eh_lock_door_done
);
1572 * scsi_restart_operations - restart io operations to the specified host.
1573 * @shost: Host we are restarting.
1576 * When we entered the error handler, we blocked all further i/o to
1577 * this device. we need to 'reverse' this process.
1579 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1581 struct scsi_device
*sdev
;
1582 unsigned long flags
;
1585 * If the door was locked, we need to insert a door lock request
1586 * onto the head of the SCSI request queue for the device. There
1587 * is no point trying to lock the door of an off-line device.
1589 shost_for_each_device(sdev
, shost
) {
1590 if (scsi_device_online(sdev
) && sdev
->locked
)
1591 scsi_eh_lock_door(sdev
);
1595 * next free up anything directly waiting upon the host. this
1596 * will be requests for character device operations, and also for
1597 * ioctls to queued block devices.
1599 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1602 spin_lock_irqsave(shost
->host_lock
, flags
);
1603 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1604 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1605 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1606 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1608 wake_up(&shost
->host_wait
);
1611 * finally we need to re-initiate requests that may be pending. we will
1612 * have had everything blocked while error handling is taking place, and
1613 * now that error recovery is done, we will need to ensure that these
1614 * requests are started.
1616 scsi_run_host_queues(shost
);
1620 * scsi_eh_ready_devs - check device ready state and recover if not.
1621 * @shost: host to be recovered.
1622 * @work_q: &list_head for pending commands.
1623 * @done_q: &list_head for processed commands.
1625 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1626 struct list_head
*work_q
,
1627 struct list_head
*done_q
)
1629 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1630 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1631 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1632 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1633 if (!scsi_eh_host_reset(work_q
, done_q
))
1634 scsi_eh_offline_sdevs(work_q
,
1637 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1640 * scsi_eh_flush_done_q - finish processed commands or retry them.
1641 * @done_q: list_head of processed commands.
1643 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1645 struct scsi_cmnd
*scmd
, *next
;
1647 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1648 list_del_init(&scmd
->eh_entry
);
1649 if (scsi_device_online(scmd
->device
) &&
1650 !scsi_noretry_cmd(scmd
) &&
1651 (++scmd
->retries
<= scmd
->allowed
)) {
1652 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1656 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1659 * If just we got sense for the device (called
1660 * scsi_eh_get_sense), scmd->result is already
1661 * set, do not set DRIVER_TIMEOUT.
1664 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1665 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1667 current
->comm
, scmd
));
1668 scsi_finish_command(scmd
);
1672 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1675 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1676 * @shost: Host to unjam.
1679 * When we come in here, we *know* that all commands on the bus have
1680 * either completed, failed or timed out. we also know that no further
1681 * commands are being sent to the host, so things are relatively quiet
1682 * and we have freedom to fiddle with things as we wish.
1684 * This is only the *default* implementation. it is possible for
1685 * individual drivers to supply their own version of this function, and
1686 * if the maintainer wishes to do this, it is strongly suggested that
1687 * this function be taken as a template and modified. this function
1688 * was designed to correctly handle problems for about 95% of the
1689 * different cases out there, and it should always provide at least a
1690 * reasonable amount of error recovery.
1692 * Any command marked 'failed' or 'timeout' must eventually have
1693 * scsi_finish_cmd() called for it. we do all of the retry stuff
1694 * here, so when we restart the host after we return it should have an
1697 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1699 unsigned long flags
;
1700 LIST_HEAD(eh_work_q
);
1701 LIST_HEAD(eh_done_q
);
1703 spin_lock_irqsave(shost
->host_lock
, flags
);
1704 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1705 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1707 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1709 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1710 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1711 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1713 scsi_eh_flush_done_q(&eh_done_q
);
1717 * scsi_error_handler - SCSI error handler thread
1718 * @data: Host for which we are running.
1721 * This is the main error handling loop. This is run as a kernel thread
1722 * for every SCSI host and handles all error handling activity.
1724 int scsi_error_handler(void *data
)
1726 struct Scsi_Host
*shost
= data
;
1729 * We use TASK_INTERRUPTIBLE so that the thread is not
1730 * counted against the load average as a running process.
1731 * We never actually get interrupted because kthread_run
1732 * disables signal delivery for the created thread.
1734 set_current_state(TASK_INTERRUPTIBLE
);
1735 while (!kthread_should_stop()) {
1736 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1737 shost
->host_failed
!= shost
->host_busy
) {
1738 SCSI_LOG_ERROR_RECOVERY(1,
1739 printk("Error handler scsi_eh_%d sleeping\n",
1742 set_current_state(TASK_INTERRUPTIBLE
);
1746 __set_current_state(TASK_RUNNING
);
1747 SCSI_LOG_ERROR_RECOVERY(1,
1748 printk("Error handler scsi_eh_%d waking up\n",
1752 * We have a host that is failing for some reason. Figure out
1753 * what we need to do to get it up and online again (if we can).
1754 * If we fail, we end up taking the thing offline.
1756 if (shost
->transportt
->eh_strategy_handler
)
1757 shost
->transportt
->eh_strategy_handler(shost
);
1759 scsi_unjam_host(shost
);
1762 * Note - if the above fails completely, the action is to take
1763 * individual devices offline and flush the queue of any
1764 * outstanding requests that may have been pending. When we
1765 * restart, we restart any I/O to any other devices on the bus
1766 * which are still online.
1768 scsi_restart_operations(shost
);
1769 set_current_state(TASK_INTERRUPTIBLE
);
1771 __set_current_state(TASK_RUNNING
);
1773 SCSI_LOG_ERROR_RECOVERY(1,
1774 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1775 shost
->ehandler
= NULL
;
1780 * Function: scsi_report_bus_reset()
1782 * Purpose: Utility function used by low-level drivers to report that
1783 * they have observed a bus reset on the bus being handled.
1785 * Arguments: shost - Host in question
1786 * channel - channel on which reset was observed.
1790 * Lock status: Host lock must be held.
1792 * Notes: This only needs to be called if the reset is one which
1793 * originates from an unknown location. Resets originated
1794 * by the mid-level itself don't need to call this, but there
1795 * should be no harm.
1797 * The main purpose of this is to make sure that a CHECK_CONDITION
1798 * is properly treated.
1800 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1802 struct scsi_device
*sdev
;
1804 __shost_for_each_device(sdev
, shost
) {
1805 if (channel
== sdev_channel(sdev
))
1806 __scsi_report_device_reset(sdev
, NULL
);
1809 EXPORT_SYMBOL(scsi_report_bus_reset
);
1812 * Function: scsi_report_device_reset()
1814 * Purpose: Utility function used by low-level drivers to report that
1815 * they have observed a device reset on the device being handled.
1817 * Arguments: shost - Host in question
1818 * channel - channel on which reset was observed
1819 * target - target on which reset was observed
1823 * Lock status: Host lock must be held
1825 * Notes: This only needs to be called if the reset is one which
1826 * originates from an unknown location. Resets originated
1827 * by the mid-level itself don't need to call this, but there
1828 * should be no harm.
1830 * The main purpose of this is to make sure that a CHECK_CONDITION
1831 * is properly treated.
1833 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1835 struct scsi_device
*sdev
;
1837 __shost_for_each_device(sdev
, shost
) {
1838 if (channel
== sdev_channel(sdev
) &&
1839 target
== sdev_id(sdev
))
1840 __scsi_report_device_reset(sdev
, NULL
);
1843 EXPORT_SYMBOL(scsi_report_device_reset
);
1846 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1851 * Function: scsi_reset_provider
1853 * Purpose: Send requested reset to a bus or device at any phase.
1855 * Arguments: device - device to send reset to
1856 * flag - reset type (see scsi.h)
1858 * Returns: SUCCESS/FAILURE.
1860 * Notes: This is used by the SCSI Generic driver to provide
1861 * Bus/Device reset capability.
1864 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1866 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1867 struct Scsi_Host
*shost
= dev
->host
;
1869 unsigned long flags
;
1872 blk_rq_init(NULL
, &req
);
1873 scmd
->request
= &req
;
1875 scmd
->cmnd
= req
.cmd
;
1877 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1878 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1882 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1884 spin_lock_irqsave(shost
->host_lock
, flags
);
1885 shost
->tmf_in_progress
= 1;
1886 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1889 case SCSI_TRY_RESET_DEVICE
:
1890 rtn
= scsi_try_bus_device_reset(scmd
);
1894 case SCSI_TRY_RESET_TARGET
:
1895 rtn
= scsi_try_target_reset(scmd
);
1899 case SCSI_TRY_RESET_BUS
:
1900 rtn
= scsi_try_bus_reset(scmd
);
1904 case SCSI_TRY_RESET_HOST
:
1905 rtn
= scsi_try_host_reset(scmd
);
1911 spin_lock_irqsave(shost
->host_lock
, flags
);
1912 shost
->tmf_in_progress
= 0;
1913 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1916 * be sure to wake up anyone who was sleeping or had their queue
1917 * suspended while we performed the TMF.
1919 SCSI_LOG_ERROR_RECOVERY(3,
1920 printk("%s: waking up host to restart after TMF\n",
1923 wake_up(&shost
->host_wait
);
1925 scsi_run_host_queues(shost
);
1927 scsi_next_command(scmd
);
1930 EXPORT_SYMBOL(scsi_reset_provider
);
1933 * scsi_normalize_sense - normalize main elements from either fixed or
1934 * descriptor sense data format into a common format.
1936 * @sense_buffer: byte array containing sense data returned by device
1937 * @sb_len: number of valid bytes in sense_buffer
1938 * @sshdr: pointer to instance of structure that common
1939 * elements are written to.
1942 * The "main elements" from sense data are: response_code, sense_key,
1943 * asc, ascq and additional_length (only for descriptor format).
1945 * Typically this function can be called after a device has
1946 * responded to a SCSI command with the CHECK_CONDITION status.
1949 * 1 if valid sense data information found, else 0;
1951 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1952 struct scsi_sense_hdr
*sshdr
)
1954 if (!sense_buffer
|| !sb_len
)
1957 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1959 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1961 if (!scsi_sense_valid(sshdr
))
1964 if (sshdr
->response_code
>= 0x72) {
1969 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1971 sshdr
->asc
= sense_buffer
[2];
1973 sshdr
->ascq
= sense_buffer
[3];
1975 sshdr
->additional_length
= sense_buffer
[7];
1981 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1983 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1984 sb_len
: (sense_buffer
[7] + 8);
1986 sshdr
->asc
= sense_buffer
[12];
1988 sshdr
->ascq
= sense_buffer
[13];
1994 EXPORT_SYMBOL(scsi_normalize_sense
);
1996 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1997 struct scsi_sense_hdr
*sshdr
)
1999 return scsi_normalize_sense(cmd
->sense_buffer
,
2000 SCSI_SENSE_BUFFERSIZE
, sshdr
);
2002 EXPORT_SYMBOL(scsi_command_normalize_sense
);
2005 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2006 * @sense_buffer: byte array of descriptor format sense data
2007 * @sb_len: number of valid bytes in sense_buffer
2008 * @desc_type: value of descriptor type to find
2009 * (e.g. 0 -> information)
2012 * only valid when sense data is in descriptor format
2015 * pointer to start of (first) descriptor if found else NULL
2017 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
2020 int add_sen_len
, add_len
, desc_len
, k
;
2023 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
2025 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
2027 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
2028 add_sen_len
: (sb_len
- 8);
2029 descp
= &sense_buffer
[8];
2030 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
2032 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
2033 desc_len
= add_len
+ 2;
2034 if (descp
[0] == desc_type
)
2036 if (add_len
< 0) // short descriptor ??
2041 EXPORT_SYMBOL(scsi_sense_desc_find
);
2044 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2045 * @sense_buffer: byte array of sense data
2046 * @sb_len: number of valid bytes in sense_buffer
2047 * @info_out: pointer to 64 integer where 8 or 4 byte information
2048 * field will be placed if found.
2051 * 1 if information field found, 0 if not found.
2053 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
2062 switch (sense_buffer
[0] & 0x7f) {
2065 if (sense_buffer
[0] & 0x80) {
2066 *info_out
= (sense_buffer
[3] << 24) +
2067 (sense_buffer
[4] << 16) +
2068 (sense_buffer
[5] << 8) + sense_buffer
[6];
2074 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
2076 if (ucp
&& (0xa == ucp
[1])) {
2078 for (j
= 0; j
< 8; ++j
) {
2091 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
2094 * scsi_build_sense_buffer - build sense data in a buffer
2095 * @desc: Sense format (non zero == descriptor format,
2096 * 0 == fixed format)
2097 * @buf: Where to build sense data
2099 * @asc: Additional sense code
2100 * @ascq: Additional sense code qualifier
2103 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2106 buf
[0] = 0x72; /* descriptor, current */
2112 buf
[0] = 0x70; /* fixed, current */
2119 EXPORT_SYMBOL(scsi_build_sense_buffer
);