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_add_timer - Start timeout timer for a single scsi command.
116 * @scmd: scsi command that is about to start running.
117 * @timeout: amount of time to allow this command to run.
118 * @complete: timeout function to call if timer isn't canceled.
121 * This should be turned into an inline function. Each scsi command
122 * has its own timer, and as it is added to the queue, we set up the
123 * timer. When the command completes, we cancel the timer.
125 void scsi_add_timer(struct scsi_cmnd
*scmd
, int timeout
,
126 void (*complete
)(struct scsi_cmnd
*))
130 * If the clock was already running for this command, then
131 * first delete the timer. The timer handling code gets rather
132 * confused if we don't do this.
134 if (scmd
->eh_timeout
.function
)
135 del_timer(&scmd
->eh_timeout
);
137 scmd
->eh_timeout
.data
= (unsigned long)scmd
;
138 scmd
->eh_timeout
.expires
= jiffies
+ timeout
;
139 scmd
->eh_timeout
.function
= (void (*)(unsigned long)) complete
;
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142 " %d, (%p)\n", __FUNCTION__
,
143 scmd
, timeout
, complete
));
145 add_timer(&scmd
->eh_timeout
);
149 * scsi_delete_timer - Delete/cancel timer for a given function.
150 * @scmd: Cmd that we are canceling timer for
153 * This should be turned into an inline function.
156 * 1 if we were able to detach the timer. 0 if we blew it, and the
157 * timer function has already started to run.
159 int scsi_delete_timer(struct scsi_cmnd
*scmd
)
163 rtn
= del_timer(&scmd
->eh_timeout
);
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166 " rtn: %d\n", __FUNCTION__
,
169 scmd
->eh_timeout
.data
= (unsigned long)NULL
;
170 scmd
->eh_timeout
.function
= NULL
;
176 * scsi_times_out - Timeout function for normal scsi commands.
177 * @scmd: Cmd that is timing out.
180 * We do not need to lock this. There is the potential for a race
181 * only in that the normal completion handling might run, but if the
182 * normal completion function determines that the timer has already
183 * fired, then it mustn't do anything.
185 void scsi_times_out(struct scsi_cmnd
*scmd
)
187 enum scsi_eh_timer_return (* eh_timed_out
)(struct scsi_cmnd
*);
189 scsi_log_completion(scmd
, TIMEOUT_ERROR
);
191 if (scmd
->device
->host
->transportt
->eh_timed_out
)
192 eh_timed_out
= scmd
->device
->host
->transportt
->eh_timed_out
;
193 else if (scmd
->device
->host
->hostt
->eh_timed_out
)
194 eh_timed_out
= scmd
->device
->host
->hostt
->eh_timed_out
;
199 switch (eh_timed_out(scmd
)) {
204 scsi_add_timer(scmd
, scmd
->timeout_per_command
,
211 if (unlikely(!scsi_eh_scmd_add(scmd
, SCSI_EH_CANCEL_CMD
))) {
212 scmd
->result
|= DID_TIME_OUT
<< 16;
218 * scsi_block_when_processing_errors - Prevent cmds from being queued.
219 * @sdev: Device on which we are performing recovery.
222 * We block until the host is out of error recovery, and then check to
223 * see whether the host or the device is offline.
226 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
228 int scsi_block_when_processing_errors(struct scsi_device
*sdev
)
232 wait_event(sdev
->host
->host_wait
, !scsi_host_in_recovery(sdev
->host
));
234 online
= scsi_device_online(sdev
);
236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__
,
241 EXPORT_SYMBOL(scsi_block_when_processing_errors
);
243 #ifdef CONFIG_SCSI_LOGGING
245 * scsi_eh_prt_fail_stats - Log info on failures.
246 * @shost: scsi host being recovered.
247 * @work_q: Queue of scsi cmds to process.
249 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host
*shost
,
250 struct list_head
*work_q
)
252 struct scsi_cmnd
*scmd
;
253 struct scsi_device
*sdev
;
254 int total_failures
= 0;
257 int devices_failed
= 0;
259 shost_for_each_device(sdev
, shost
) {
260 list_for_each_entry(scmd
, work_q
, eh_entry
) {
261 if (scmd
->device
== sdev
) {
263 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
)
270 if (cmd_cancel
|| cmd_failed
) {
271 SCSI_LOG_ERROR_RECOVERY(3,
272 sdev_printk(KERN_INFO
, sdev
,
273 "%s: cmds failed: %d, cancel: %d\n",
274 __FUNCTION__
, cmd_failed
,
282 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
283 " devices require eh work\n",
284 total_failures
, devices_failed
));
289 * scsi_check_sense - Examine scsi cmd sense
290 * @scmd: Cmd to have sense checked.
293 * SUCCESS or FAILED or NEEDS_RETRY
296 * When a deferred error is detected the current command has
297 * not been executed and needs retrying.
299 static int scsi_check_sense(struct scsi_cmnd
*scmd
)
301 struct scsi_sense_hdr sshdr
;
303 if (! scsi_command_normalize_sense(scmd
, &sshdr
))
304 return FAILED
; /* no valid sense data */
306 if (scsi_sense_is_deferred(&sshdr
))
310 * Previous logic looked for FILEMARK, EOM or ILI which are
311 * mainly associated with tapes and returned SUCCESS.
313 if (sshdr
.response_code
== 0x70) {
315 if (scmd
->sense_buffer
[2] & 0xe0)
319 * descriptor format: look for "stream commands sense data
320 * descriptor" (see SSC-3). Assume single sense data
321 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
323 if ((sshdr
.additional_length
> 3) &&
324 (scmd
->sense_buffer
[8] == 0x4) &&
325 (scmd
->sense_buffer
[11] & 0xe0))
329 switch (sshdr
.sense_key
) {
332 case RECOVERED_ERROR
:
333 return /* soft_error */ SUCCESS
;
335 case ABORTED_COMMAND
:
340 * if we are expecting a cc/ua because of a bus reset that we
341 * performed, treat this just as a retry. otherwise this is
342 * information that we should pass up to the upper-level driver
343 * so that we can deal with it there.
345 if (scmd
->device
->expecting_cc_ua
) {
346 scmd
->device
->expecting_cc_ua
= 0;
350 * if the device is in the process of becoming ready, we
353 if ((sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x01))
356 * if the device is not started, we need to wake
357 * the error handler to start the motor
359 if (scmd
->device
->allow_restart
&&
360 (sshdr
.asc
== 0x04) && (sshdr
.ascq
== 0x02))
364 /* these three are not supported */
366 case VOLUME_OVERFLOW
:
371 if (sshdr
.asc
== 0x11 || /* UNRECOVERED READ ERR */
372 sshdr
.asc
== 0x13 || /* AMNF DATA FIELD */
373 sshdr
.asc
== 0x14) { /* RECORD NOT FOUND */
379 if (scmd
->device
->retry_hwerror
)
384 case ILLEGAL_REQUEST
:
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 case COMMAND_TERMINATED
:
434 case CHECK_CONDITION
:
435 return scsi_check_sense(scmd
);
437 case INTERMEDIATE_GOOD
:
438 case INTERMEDIATE_C_GOOD
:
440 * who knows? FIXME(eric)
445 case RESERVATION_CONFLICT
:
453 * scsi_eh_done - Completion function for error handling.
454 * @scmd: Cmd that is done.
456 static void scsi_eh_done(struct scsi_cmnd
*scmd
)
458 struct completion
*eh_action
;
460 SCSI_LOG_ERROR_RECOVERY(3,
461 printk("%s scmd: %p result: %x\n",
462 __FUNCTION__
, scmd
, scmd
->result
));
464 eh_action
= scmd
->device
->host
->eh_action
;
470 * scsi_try_host_reset - ask host adapter to reset itself
471 * @scmd: SCSI cmd to send hsot reset.
473 static int scsi_try_host_reset(struct scsi_cmnd
*scmd
)
478 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
481 if (!scmd
->device
->host
->hostt
->eh_host_reset_handler
)
484 rtn
= scmd
->device
->host
->hostt
->eh_host_reset_handler(scmd
);
486 if (rtn
== SUCCESS
) {
487 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
488 ssleep(HOST_RESET_SETTLE_TIME
);
489 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
490 scsi_report_bus_reset(scmd
->device
->host
,
492 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
499 * scsi_try_bus_reset - ask host to perform a bus reset
500 * @scmd: SCSI cmd to send bus reset.
502 static int scsi_try_bus_reset(struct scsi_cmnd
*scmd
)
507 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
510 if (!scmd
->device
->host
->hostt
->eh_bus_reset_handler
)
513 rtn
= scmd
->device
->host
->hostt
->eh_bus_reset_handler(scmd
);
515 if (rtn
== SUCCESS
) {
516 if (!scmd
->device
->host
->hostt
->skip_settle_delay
)
517 ssleep(BUS_RESET_SETTLE_TIME
);
518 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
519 scsi_report_bus_reset(scmd
->device
->host
,
521 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
527 static void __scsi_report_device_reset(struct scsi_device
*sdev
, void *data
)
530 sdev
->expecting_cc_ua
= 1;
534 * scsi_try_target_reset - Ask host to perform a target reset
535 * @scmd: SCSI cmd used to send a target reset
538 * There is no timeout for this operation. if this operation is
539 * unreliable for a given host, then the host itself needs to put a
540 * timer on it, and set the host back to a consistent state prior to
543 static int scsi_try_target_reset(struct scsi_cmnd
*scmd
)
548 if (!scmd
->device
->host
->hostt
->eh_target_reset_handler
)
551 rtn
= scmd
->device
->host
->hostt
->eh_target_reset_handler(scmd
);
552 if (rtn
== SUCCESS
) {
553 spin_lock_irqsave(scmd
->device
->host
->host_lock
, flags
);
554 __starget_for_each_device(scsi_target(scmd
->device
), NULL
,
555 __scsi_report_device_reset
);
556 spin_unlock_irqrestore(scmd
->device
->host
->host_lock
, flags
);
563 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
564 * @scmd: SCSI cmd used to send BDR
567 * There is no timeout for this operation. if this operation is
568 * unreliable for a given host, then the host itself needs to put a
569 * timer on it, and set the host back to a consistent state prior to
572 static int scsi_try_bus_device_reset(struct scsi_cmnd
*scmd
)
576 if (!scmd
->device
->host
->hostt
->eh_device_reset_handler
)
579 rtn
= scmd
->device
->host
->hostt
->eh_device_reset_handler(scmd
);
581 __scsi_report_device_reset(scmd
->device
, NULL
);
585 static int __scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
587 if (!scmd
->device
->host
->hostt
->eh_abort_handler
)
590 return scmd
->device
->host
->hostt
->eh_abort_handler(scmd
);
594 * scsi_try_to_abort_cmd - Ask host to abort a running command.
595 * @scmd: SCSI cmd to abort from Lower Level.
598 * This function will not return until the user's completion function
599 * has been called. there is no timeout on this operation. if the
600 * author of the low-level driver wishes this operation to be timed,
601 * they can provide this facility themselves. helper functions in
602 * scsi_error.c can be supplied to make this easier to do.
604 static int scsi_try_to_abort_cmd(struct scsi_cmnd
*scmd
)
607 * scsi_done was called just after the command timed out and before
608 * we had a chance to process it. (db)
610 if (scmd
->serial_number
== 0)
612 return __scsi_try_to_abort_cmd(scmd
);
615 static void scsi_abort_eh_cmnd(struct scsi_cmnd
*scmd
)
617 if (__scsi_try_to_abort_cmd(scmd
) != SUCCESS
)
618 if (scsi_try_bus_device_reset(scmd
) != SUCCESS
)
619 if (scsi_try_target_reset(scmd
) != SUCCESS
)
620 if (scsi_try_bus_reset(scmd
) != SUCCESS
)
621 scsi_try_host_reset(scmd
);
625 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
626 * @scmd: SCSI command structure to hijack
627 * @ses: structure to save restore information
628 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
629 * @cmnd_size: size in bytes of @cmnd
630 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
632 * This function is used to save a scsi command information before re-execution
633 * as part of the error recovery process. If @sense_bytes is 0 the command
634 * sent must be one that does not transfer any data. If @sense_bytes != 0
635 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
636 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
638 void scsi_eh_prep_cmnd(struct scsi_cmnd
*scmd
, struct scsi_eh_save
*ses
,
639 unsigned char *cmnd
, int cmnd_size
, unsigned sense_bytes
)
641 struct scsi_device
*sdev
= scmd
->device
;
644 * We need saved copies of a number of fields - this is because
645 * error handling may need to overwrite these with different values
646 * to run different commands, and once error handling is complete,
647 * we will need to restore these values prior to running the actual
650 ses
->cmd_len
= scmd
->cmd_len
;
651 memcpy(ses
->cmnd
, scmd
->cmnd
, sizeof(scmd
->cmnd
));
652 ses
->data_direction
= scmd
->sc_data_direction
;
653 ses
->sdb
= scmd
->sdb
;
654 ses
->next_rq
= scmd
->request
->next_rq
;
655 ses
->result
= scmd
->result
;
657 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
658 scmd
->request
->next_rq
= NULL
;
661 scmd
->sdb
.length
= min_t(unsigned, SCSI_SENSE_BUFFERSIZE
,
663 sg_init_one(&ses
->sense_sgl
, scmd
->sense_buffer
,
665 scmd
->sdb
.table
.sgl
= &ses
->sense_sgl
;
666 scmd
->sc_data_direction
= DMA_FROM_DEVICE
;
667 scmd
->sdb
.table
.nents
= 1;
668 memset(scmd
->cmnd
, 0, sizeof(scmd
->cmnd
));
669 scmd
->cmnd
[0] = REQUEST_SENSE
;
670 scmd
->cmnd
[4] = scmd
->sdb
.length
;
671 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
673 scmd
->sc_data_direction
= DMA_NONE
;
675 memset(scmd
->cmnd
, 0, sizeof(scmd
->cmnd
));
676 memcpy(scmd
->cmnd
, cmnd
, cmnd_size
);
677 scmd
->cmd_len
= COMMAND_SIZE(scmd
->cmnd
[0]);
683 if (sdev
->scsi_level
<= SCSI_2
&& sdev
->scsi_level
!= SCSI_UNKNOWN
)
684 scmd
->cmnd
[1] = (scmd
->cmnd
[1] & 0x1f) |
685 (sdev
->lun
<< 5 & 0xe0);
688 * Zero the sense buffer. The scsi spec mandates that any
689 * untransferred sense data should be interpreted as being zero.
691 memset(scmd
->sense_buffer
, 0, SCSI_SENSE_BUFFERSIZE
);
693 EXPORT_SYMBOL(scsi_eh_prep_cmnd
);
696 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
697 * @scmd: SCSI command structure to restore
698 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
700 * Undo any damage done by above scsi_prep_eh_cmnd().
702 void scsi_eh_restore_cmnd(struct scsi_cmnd
* scmd
, struct scsi_eh_save
*ses
)
705 * Restore original data
707 scmd
->cmd_len
= ses
->cmd_len
;
708 memcpy(scmd
->cmnd
, ses
->cmnd
, sizeof(scmd
->cmnd
));
709 scmd
->sc_data_direction
= ses
->data_direction
;
710 scmd
->sdb
= ses
->sdb
;
711 scmd
->request
->next_rq
= ses
->next_rq
;
712 scmd
->result
= ses
->result
;
714 EXPORT_SYMBOL(scsi_eh_restore_cmnd
);
717 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
718 * @scmd: SCSI command structure to hijack
720 * @cmnd_size: size in bytes of @cmnd
721 * @timeout: timeout for this request
722 * @sense_bytes: size of sense data to copy or 0
724 * This function is used to send a scsi command down to a target device
725 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
728 * SUCCESS or FAILED or NEEDS_RETRY
730 static int scsi_send_eh_cmnd(struct scsi_cmnd
*scmd
, unsigned char *cmnd
,
731 int cmnd_size
, int timeout
, unsigned sense_bytes
)
733 struct scsi_device
*sdev
= scmd
->device
;
734 struct Scsi_Host
*shost
= sdev
->host
;
735 DECLARE_COMPLETION_ONSTACK(done
);
736 unsigned long timeleft
;
738 struct scsi_eh_save ses
;
741 scsi_eh_prep_cmnd(scmd
, &ses
, cmnd
, cmnd_size
, sense_bytes
);
742 shost
->eh_action
= &done
;
744 spin_lock_irqsave(shost
->host_lock
, flags
);
746 shost
->hostt
->queuecommand(scmd
, scsi_eh_done
);
747 spin_unlock_irqrestore(shost
->host_lock
, flags
);
749 timeleft
= wait_for_completion_timeout(&done
, timeout
);
751 shost
->eh_action
= NULL
;
753 scsi_log_completion(scmd
, SUCCESS
);
755 SCSI_LOG_ERROR_RECOVERY(3,
756 printk("%s: scmd: %p, timeleft: %ld\n",
757 __FUNCTION__
, scmd
, timeleft
));
760 * If there is time left scsi_eh_done got called, and we will
761 * examine the actual status codes to see whether the command
762 * actually did complete normally, else tell the host to forget
763 * about this command.
766 rtn
= scsi_eh_completed_normally(scmd
);
767 SCSI_LOG_ERROR_RECOVERY(3,
768 printk("%s: scsi_eh_completed_normally %x\n",
781 scsi_abort_eh_cmnd(scmd
);
785 scsi_eh_restore_cmnd(scmd
, &ses
);
790 * scsi_request_sense - Request sense data from a particular target.
791 * @scmd: SCSI cmd for request sense.
794 * Some hosts automatically obtain this information, others require
795 * that we obtain it on our own. This function will *not* return until
796 * the command either times out, or it completes.
798 static int scsi_request_sense(struct scsi_cmnd
*scmd
)
800 return scsi_send_eh_cmnd(scmd
, NULL
, 0, SENSE_TIMEOUT
, ~0);
804 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
805 * @scmd: Original SCSI cmd that eh has finished.
806 * @done_q: Queue for processed commands.
809 * We don't want to use the normal command completion while we are are
810 * still handling errors - it may cause other commands to be queued,
811 * and that would disturb what we are doing. Thus we really want to
812 * keep a list of pending commands for final completion, and once we
813 * are ready to leave error handling we handle completion for real.
815 void scsi_eh_finish_cmd(struct scsi_cmnd
*scmd
, struct list_head
*done_q
)
817 scmd
->device
->host
->host_failed
--;
819 list_move_tail(&scmd
->eh_entry
, done_q
);
821 EXPORT_SYMBOL(scsi_eh_finish_cmd
);
824 * scsi_eh_get_sense - Get device sense data.
825 * @work_q: Queue of commands to process.
826 * @done_q: Queue of processed commands.
829 * See if we need to request sense information. if so, then get it
830 * now, so we have a better idea of what to do.
833 * This has the unfortunate side effect that if a shost adapter does
834 * not automatically request sense information, we end up shutting
835 * it down before we request it.
837 * All drivers should request sense information internally these days,
838 * so for now all I have to say is tough noogies if you end up in here.
840 * XXX: Long term this code should go away, but that needs an audit of
843 int scsi_eh_get_sense(struct list_head
*work_q
,
844 struct list_head
*done_q
)
846 struct scsi_cmnd
*scmd
, *next
;
849 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
850 if ((scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) ||
851 SCSI_SENSE_VALID(scmd
))
854 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO
, scmd
,
855 "%s: requesting sense\n",
857 rtn
= scsi_request_sense(scmd
);
861 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
862 " result %x\n", scmd
,
864 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd
));
866 rtn
= scsi_decide_disposition(scmd
);
869 * if the result was normal, then just pass it along to the
873 /* we don't want this command reissued, just
874 * finished with the sense data, so set
875 * retries to the max allowed to ensure it
876 * won't get reissued */
877 scmd
->retries
= scmd
->allowed
;
878 else if (rtn
!= NEEDS_RETRY
)
881 scsi_eh_finish_cmd(scmd
, done_q
);
884 return list_empty(work_q
);
886 EXPORT_SYMBOL_GPL(scsi_eh_get_sense
);
889 * scsi_eh_tur - Send TUR to device.
890 * @scmd: &scsi_cmnd to send TUR
893 * 0 - Device is ready. 1 - Device NOT ready.
895 static int scsi_eh_tur(struct scsi_cmnd
*scmd
)
897 static unsigned char tur_command
[6] = {TEST_UNIT_READY
, 0, 0, 0, 0, 0};
898 int retry_cnt
= 1, rtn
;
901 rtn
= scsi_send_eh_cmnd(scmd
, tur_command
, 6, SENSE_TIMEOUT
, 0);
903 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
904 __FUNCTION__
, scmd
, rtn
));
919 * scsi_eh_abort_cmds - abort pending commands.
920 * @work_q: &list_head for pending commands.
921 * @done_q: &list_head for processed commands.
924 * Try and see whether or not it makes sense to try and abort the
925 * running command. This only works out to be the case if we have one
926 * command that has timed out. If the command simply failed, it makes
927 * no sense to try and abort the command, since as far as the shost
928 * adapter is concerned, it isn't running.
930 static int scsi_eh_abort_cmds(struct list_head
*work_q
,
931 struct list_head
*done_q
)
933 struct scsi_cmnd
*scmd
, *next
;
936 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
937 if (!(scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
))
939 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
940 "0x%p\n", current
->comm
,
942 rtn
= scsi_try_to_abort_cmd(scmd
);
943 if (rtn
== SUCCESS
) {
944 scmd
->eh_eflags
&= ~SCSI_EH_CANCEL_CMD
;
945 if (!scsi_device_online(scmd
->device
) ||
946 !scsi_eh_tur(scmd
)) {
947 scsi_eh_finish_cmd(scmd
, done_q
);
951 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
958 return list_empty(work_q
);
962 * scsi_eh_try_stu - Send START_UNIT to device.
963 * @scmd: &scsi_cmnd to send START_UNIT
966 * 0 - Device is ready. 1 - Device NOT ready.
968 static int scsi_eh_try_stu(struct scsi_cmnd
*scmd
)
970 static unsigned char stu_command
[6] = {START_STOP
, 0, 0, 0, 1, 0};
972 if (scmd
->device
->allow_restart
) {
973 int i
, rtn
= NEEDS_RETRY
;
975 for (i
= 0; rtn
== NEEDS_RETRY
&& i
< 2; i
++)
976 rtn
= scsi_send_eh_cmnd(scmd
, stu_command
, 6,
977 scmd
->device
->timeout
, 0);
987 * scsi_eh_stu - send START_UNIT if needed
988 * @shost: &scsi host being recovered.
989 * @work_q: &list_head for pending commands.
990 * @done_q: &list_head for processed commands.
993 * If commands are failing due to not ready, initializing command required,
994 * try revalidating the device, which will end up sending a start unit.
996 static int scsi_eh_stu(struct Scsi_Host
*shost
,
997 struct list_head
*work_q
,
998 struct list_head
*done_q
)
1000 struct scsi_cmnd
*scmd
, *stu_scmd
, *next
;
1001 struct scsi_device
*sdev
;
1003 shost_for_each_device(sdev
, shost
) {
1005 list_for_each_entry(scmd
, work_q
, eh_entry
)
1006 if (scmd
->device
== sdev
&& SCSI_SENSE_VALID(scmd
) &&
1007 scsi_check_sense(scmd
) == FAILED
) {
1015 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1016 " 0x%p\n", current
->comm
, sdev
));
1018 if (!scsi_eh_try_stu(stu_scmd
)) {
1019 if (!scsi_device_online(sdev
) ||
1020 !scsi_eh_tur(stu_scmd
)) {
1021 list_for_each_entry_safe(scmd
, next
,
1023 if (scmd
->device
== sdev
)
1024 scsi_eh_finish_cmd(scmd
, done_q
);
1028 SCSI_LOG_ERROR_RECOVERY(3,
1029 printk("%s: START_UNIT failed to sdev:"
1030 " 0x%p\n", current
->comm
, sdev
));
1034 return list_empty(work_q
);
1039 * scsi_eh_bus_device_reset - send bdr if needed
1040 * @shost: scsi host being recovered.
1041 * @work_q: &list_head for pending commands.
1042 * @done_q: &list_head for processed commands.
1045 * Try a bus device reset. Still, look to see whether we have multiple
1046 * devices that are jammed or not - if we have multiple devices, it
1047 * makes no sense to try bus_device_reset - we really would need to try
1048 * a bus_reset instead.
1050 static int scsi_eh_bus_device_reset(struct Scsi_Host
*shost
,
1051 struct list_head
*work_q
,
1052 struct list_head
*done_q
)
1054 struct scsi_cmnd
*scmd
, *bdr_scmd
, *next
;
1055 struct scsi_device
*sdev
;
1058 shost_for_each_device(sdev
, shost
) {
1060 list_for_each_entry(scmd
, work_q
, eh_entry
)
1061 if (scmd
->device
== sdev
) {
1069 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1070 " 0x%p\n", current
->comm
,
1072 rtn
= scsi_try_bus_device_reset(bdr_scmd
);
1073 if (rtn
== SUCCESS
) {
1074 if (!scsi_device_online(sdev
) ||
1075 !scsi_eh_tur(bdr_scmd
)) {
1076 list_for_each_entry_safe(scmd
, next
,
1078 if (scmd
->device
== sdev
)
1079 scsi_eh_finish_cmd(scmd
,
1084 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1092 return list_empty(work_q
);
1096 * scsi_eh_target_reset - send target reset if needed
1097 * @shost: scsi host being recovered.
1098 * @work_q: &list_head for pending commands.
1099 * @done_q: &list_head for processed commands.
1102 * Try a target reset.
1104 static int scsi_eh_target_reset(struct Scsi_Host
*shost
,
1105 struct list_head
*work_q
,
1106 struct list_head
*done_q
)
1108 struct scsi_cmnd
*scmd
, *tgtr_scmd
, *next
;
1112 for (id
= 0; id
<= shost
->max_id
; id
++) {
1114 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1115 if (id
== scmd_id(scmd
)) {
1123 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1125 current
->comm
, id
));
1126 rtn
= scsi_try_target_reset(tgtr_scmd
);
1127 if (rtn
== SUCCESS
) {
1128 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1129 if (id
== scmd_id(scmd
))
1130 if (!scsi_device_online(scmd
->device
) ||
1131 !scsi_eh_tur(tgtr_scmd
))
1132 scsi_eh_finish_cmd(scmd
,
1136 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1139 current
->comm
, id
));
1142 return list_empty(work_q
);
1146 * scsi_eh_bus_reset - send a bus reset
1147 * @shost: &scsi host being recovered.
1148 * @work_q: &list_head for pending commands.
1149 * @done_q: &list_head for processed commands.
1151 static int scsi_eh_bus_reset(struct Scsi_Host
*shost
,
1152 struct list_head
*work_q
,
1153 struct list_head
*done_q
)
1155 struct scsi_cmnd
*scmd
, *chan_scmd
, *next
;
1156 unsigned int channel
;
1160 * we really want to loop over the various channels, and do this on
1161 * a channel by channel basis. we should also check to see if any
1162 * of the failed commands are on soft_reset devices, and if so, skip
1166 for (channel
= 0; channel
<= shost
->max_channel
; channel
++) {
1168 list_for_each_entry(scmd
, work_q
, eh_entry
) {
1169 if (channel
== scmd_channel(scmd
)) {
1173 * FIXME add back in some support for
1174 * soft_reset devices.
1181 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1182 " %d\n", current
->comm
,
1184 rtn
= scsi_try_bus_reset(chan_scmd
);
1185 if (rtn
== SUCCESS
) {
1186 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1187 if (channel
== scmd_channel(scmd
))
1188 if (!scsi_device_online(scmd
->device
) ||
1190 scsi_eh_finish_cmd(scmd
,
1194 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1195 " failed chan: %d\n",
1200 return list_empty(work_q
);
1204 * scsi_eh_host_reset - send a host reset
1205 * @work_q: list_head for processed commands.
1206 * @done_q: list_head for processed commands.
1208 static int scsi_eh_host_reset(struct list_head
*work_q
,
1209 struct list_head
*done_q
)
1211 struct scsi_cmnd
*scmd
, *next
;
1214 if (!list_empty(work_q
)) {
1215 scmd
= list_entry(work_q
->next
,
1216 struct scsi_cmnd
, eh_entry
);
1218 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1221 rtn
= scsi_try_host_reset(scmd
);
1222 if (rtn
== SUCCESS
) {
1223 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1224 if (!scsi_device_online(scmd
->device
) ||
1225 (!scsi_eh_try_stu(scmd
) && !scsi_eh_tur(scmd
)) ||
1227 scsi_eh_finish_cmd(scmd
, done_q
);
1230 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1235 return list_empty(work_q
);
1239 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1240 * @work_q: list_head for processed commands.
1241 * @done_q: list_head for processed commands.
1243 static void scsi_eh_offline_sdevs(struct list_head
*work_q
,
1244 struct list_head
*done_q
)
1246 struct scsi_cmnd
*scmd
, *next
;
1248 list_for_each_entry_safe(scmd
, next
, work_q
, eh_entry
) {
1249 sdev_printk(KERN_INFO
, scmd
->device
, "Device offlined - "
1250 "not ready after error recovery\n");
1251 scsi_device_set_state(scmd
->device
, SDEV_OFFLINE
);
1252 if (scmd
->eh_eflags
& SCSI_EH_CANCEL_CMD
) {
1254 * FIXME: Handle lost cmds.
1257 scsi_eh_finish_cmd(scmd
, done_q
);
1263 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1264 * @scmd: SCSI cmd to examine.
1267 * This is *only* called when we are examining the status after sending
1268 * out the actual data command. any commands that are queued for error
1269 * recovery (e.g. test_unit_ready) do *not* come through here.
1271 * When this routine returns failed, it means the error handler thread
1272 * is woken. In cases where the error code indicates an error that
1273 * doesn't require the error handler read (i.e. we don't need to
1274 * abort/reset), this function should return SUCCESS.
1276 int scsi_decide_disposition(struct scsi_cmnd
*scmd
)
1281 * if the device is offline, then we clearly just pass the result back
1282 * up to the top level.
1284 if (!scsi_device_online(scmd
->device
)) {
1285 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1292 * first check the host byte, to see if there is anything in there
1293 * that would indicate what we need to do.
1295 switch (host_byte(scmd
->result
)) {
1296 case DID_PASSTHROUGH
:
1298 * no matter what, pass this through to the upper layer.
1299 * nuke this special code so that it looks like we are saying
1302 scmd
->result
&= 0xff00ffff;
1306 * looks good. drop through, and check the next byte.
1309 case DID_NO_CONNECT
:
1310 case DID_BAD_TARGET
:
1313 * note - this means that we just report the status back
1314 * to the top level driver, not that we actually think
1315 * that it indicates SUCCESS.
1319 * when the low level driver returns did_soft_error,
1320 * it is responsible for keeping an internal retry counter
1321 * in order to avoid endless loops (db)
1323 * actually this is a bug in this function here. we should
1324 * be mindful of the maximum number of retries specified
1325 * and not get stuck in a loop.
1327 case DID_SOFT_ERROR
:
1333 return ADD_TO_MLQUEUE
;
1336 if (msg_byte(scmd
->result
) == COMMAND_COMPLETE
&&
1337 status_byte(scmd
->result
) == RESERVATION_CONFLICT
)
1339 * execute reservation conflict processing code
1350 * when we scan the bus, we get timeout messages for
1351 * these commands if there is no device available.
1352 * other hosts report did_no_connect for the same thing.
1354 if ((scmd
->cmnd
[0] == TEST_UNIT_READY
||
1355 scmd
->cmnd
[0] == INQUIRY
)) {
1367 * next, check the message byte.
1369 if (msg_byte(scmd
->result
) != COMMAND_COMPLETE
)
1373 * check the status byte to see if this indicates anything special.
1375 switch (status_byte(scmd
->result
)) {
1378 * the case of trying to send too many commands to a
1379 * tagged queueing device.
1383 * device can't talk to us at the moment. Should only
1384 * occur (SAM-3) when the task queue is empty, so will cause
1385 * the empty queue handling to trigger a stall in the
1388 return ADD_TO_MLQUEUE
;
1390 case COMMAND_TERMINATED
:
1393 case CHECK_CONDITION
:
1394 rtn
= scsi_check_sense(scmd
);
1395 if (rtn
== NEEDS_RETRY
)
1397 /* if rtn == FAILED, we have no sense information;
1398 * returning FAILED will wake the error handler thread
1399 * to collect the sense and redo the decide
1402 case CONDITION_GOOD
:
1403 case INTERMEDIATE_GOOD
:
1404 case INTERMEDIATE_C_GOOD
:
1407 * who knows? FIXME(eric)
1411 case RESERVATION_CONFLICT
:
1412 sdev_printk(KERN_INFO
, scmd
->device
,
1413 "reservation conflict\n");
1414 return SUCCESS
; /* causes immediate i/o error */
1422 /* we requeue for retry because the error was retryable, and
1423 * the request was not marked fast fail. Note that above,
1424 * even if the request is marked fast fail, we still requeue
1425 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1426 if ((++scmd
->retries
) <= scmd
->allowed
1427 && !blk_noretry_request(scmd
->request
)) {
1431 * no more retries - report this one back to upper level.
1438 * scsi_eh_lock_door - Prevent medium removal for the specified device
1439 * @sdev: SCSI device to prevent medium removal
1442 * We must be called from process context; scsi_allocate_request()
1446 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1447 * head of the devices request queue, and continue.
1450 * scsi_allocate_request() may sleep waiting for existing requests to
1451 * be processed. However, since we haven't kicked off any request
1452 * processing for this host, this may deadlock.
1454 * If scsi_allocate_request() fails for what ever reason, we
1455 * completely forget to lock the door.
1457 static void scsi_eh_lock_door(struct scsi_device
*sdev
)
1459 unsigned char cmnd
[MAX_COMMAND_SIZE
];
1461 cmnd
[0] = ALLOW_MEDIUM_REMOVAL
;
1465 cmnd
[4] = SCSI_REMOVAL_PREVENT
;
1468 scsi_execute_async(sdev
, cmnd
, 6, DMA_NONE
, NULL
, 0, 0, 10 * HZ
,
1469 5, NULL
, NULL
, GFP_KERNEL
);
1474 * scsi_restart_operations - restart io operations to the specified host.
1475 * @shost: Host we are restarting.
1478 * When we entered the error handler, we blocked all further i/o to
1479 * this device. we need to 'reverse' this process.
1481 static void scsi_restart_operations(struct Scsi_Host
*shost
)
1483 struct scsi_device
*sdev
;
1484 unsigned long flags
;
1487 * If the door was locked, we need to insert a door lock request
1488 * onto the head of the SCSI request queue for the device. There
1489 * is no point trying to lock the door of an off-line device.
1491 shost_for_each_device(sdev
, shost
) {
1492 if (scsi_device_online(sdev
) && sdev
->locked
)
1493 scsi_eh_lock_door(sdev
);
1497 * next free up anything directly waiting upon the host. this
1498 * will be requests for character device operations, and also for
1499 * ioctls to queued block devices.
1501 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1504 spin_lock_irqsave(shost
->host_lock
, flags
);
1505 if (scsi_host_set_state(shost
, SHOST_RUNNING
))
1506 if (scsi_host_set_state(shost
, SHOST_CANCEL
))
1507 BUG_ON(scsi_host_set_state(shost
, SHOST_DEL
));
1508 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1510 wake_up(&shost
->host_wait
);
1513 * finally we need to re-initiate requests that may be pending. we will
1514 * have had everything blocked while error handling is taking place, and
1515 * now that error recovery is done, we will need to ensure that these
1516 * requests are started.
1518 scsi_run_host_queues(shost
);
1522 * scsi_eh_ready_devs - check device ready state and recover if not.
1523 * @shost: host to be recovered.
1524 * @work_q: &list_head for pending commands.
1525 * @done_q: &list_head for processed commands.
1527 void scsi_eh_ready_devs(struct Scsi_Host
*shost
,
1528 struct list_head
*work_q
,
1529 struct list_head
*done_q
)
1531 if (!scsi_eh_stu(shost
, work_q
, done_q
))
1532 if (!scsi_eh_bus_device_reset(shost
, work_q
, done_q
))
1533 if (!scsi_eh_target_reset(shost
, work_q
, done_q
))
1534 if (!scsi_eh_bus_reset(shost
, work_q
, done_q
))
1535 if (!scsi_eh_host_reset(work_q
, done_q
))
1536 scsi_eh_offline_sdevs(work_q
,
1539 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs
);
1542 * scsi_eh_flush_done_q - finish processed commands or retry them.
1543 * @done_q: list_head of processed commands.
1545 void scsi_eh_flush_done_q(struct list_head
*done_q
)
1547 struct scsi_cmnd
*scmd
, *next
;
1549 list_for_each_entry_safe(scmd
, next
, done_q
, eh_entry
) {
1550 list_del_init(&scmd
->eh_entry
);
1551 if (scsi_device_online(scmd
->device
) &&
1552 !blk_noretry_request(scmd
->request
) &&
1553 (++scmd
->retries
<= scmd
->allowed
)) {
1554 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1558 scsi_queue_insert(scmd
, SCSI_MLQUEUE_EH_RETRY
);
1561 * If just we got sense for the device (called
1562 * scsi_eh_get_sense), scmd->result is already
1563 * set, do not set DRIVER_TIMEOUT.
1566 scmd
->result
|= (DRIVER_TIMEOUT
<< 24);
1567 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1569 current
->comm
, scmd
));
1570 scsi_finish_command(scmd
);
1574 EXPORT_SYMBOL(scsi_eh_flush_done_q
);
1577 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1578 * @shost: Host to unjam.
1581 * When we come in here, we *know* that all commands on the bus have
1582 * either completed, failed or timed out. we also know that no further
1583 * commands are being sent to the host, so things are relatively quiet
1584 * and we have freedom to fiddle with things as we wish.
1586 * This is only the *default* implementation. it is possible for
1587 * individual drivers to supply their own version of this function, and
1588 * if the maintainer wishes to do this, it is strongly suggested that
1589 * this function be taken as a template and modified. this function
1590 * was designed to correctly handle problems for about 95% of the
1591 * different cases out there, and it should always provide at least a
1592 * reasonable amount of error recovery.
1594 * Any command marked 'failed' or 'timeout' must eventually have
1595 * scsi_finish_cmd() called for it. we do all of the retry stuff
1596 * here, so when we restart the host after we return it should have an
1599 static void scsi_unjam_host(struct Scsi_Host
*shost
)
1601 unsigned long flags
;
1602 LIST_HEAD(eh_work_q
);
1603 LIST_HEAD(eh_done_q
);
1605 spin_lock_irqsave(shost
->host_lock
, flags
);
1606 list_splice_init(&shost
->eh_cmd_q
, &eh_work_q
);
1607 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1609 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost
, &eh_work_q
));
1611 if (!scsi_eh_get_sense(&eh_work_q
, &eh_done_q
))
1612 if (!scsi_eh_abort_cmds(&eh_work_q
, &eh_done_q
))
1613 scsi_eh_ready_devs(shost
, &eh_work_q
, &eh_done_q
);
1615 scsi_eh_flush_done_q(&eh_done_q
);
1619 * scsi_error_handler - SCSI error handler thread
1620 * @data: Host for which we are running.
1623 * This is the main error handling loop. This is run as a kernel thread
1624 * for every SCSI host and handles all error handling activity.
1626 int scsi_error_handler(void *data
)
1628 struct Scsi_Host
*shost
= data
;
1631 * We use TASK_INTERRUPTIBLE so that the thread is not
1632 * counted against the load average as a running process.
1633 * We never actually get interrupted because kthread_run
1634 * disables singal delivery for the created thread.
1636 set_current_state(TASK_INTERRUPTIBLE
);
1637 while (!kthread_should_stop()) {
1638 if ((shost
->host_failed
== 0 && shost
->host_eh_scheduled
== 0) ||
1639 shost
->host_failed
!= shost
->host_busy
) {
1640 SCSI_LOG_ERROR_RECOVERY(1,
1641 printk("Error handler scsi_eh_%d sleeping\n",
1644 set_current_state(TASK_INTERRUPTIBLE
);
1648 __set_current_state(TASK_RUNNING
);
1649 SCSI_LOG_ERROR_RECOVERY(1,
1650 printk("Error handler scsi_eh_%d waking up\n",
1654 * We have a host that is failing for some reason. Figure out
1655 * what we need to do to get it up and online again (if we can).
1656 * If we fail, we end up taking the thing offline.
1658 if (shost
->transportt
->eh_strategy_handler
)
1659 shost
->transportt
->eh_strategy_handler(shost
);
1661 scsi_unjam_host(shost
);
1664 * Note - if the above fails completely, the action is to take
1665 * individual devices offline and flush the queue of any
1666 * outstanding requests that may have been pending. When we
1667 * restart, we restart any I/O to any other devices on the bus
1668 * which are still online.
1670 scsi_restart_operations(shost
);
1671 set_current_state(TASK_INTERRUPTIBLE
);
1673 __set_current_state(TASK_RUNNING
);
1675 SCSI_LOG_ERROR_RECOVERY(1,
1676 printk("Error handler scsi_eh_%d exiting\n", shost
->host_no
));
1677 shost
->ehandler
= NULL
;
1682 * Function: scsi_report_bus_reset()
1684 * Purpose: Utility function used by low-level drivers to report that
1685 * they have observed a bus reset on the bus being handled.
1687 * Arguments: shost - Host in question
1688 * channel - channel on which reset was observed.
1692 * Lock status: Host lock must be held.
1694 * Notes: This only needs to be called if the reset is one which
1695 * originates from an unknown location. Resets originated
1696 * by the mid-level itself don't need to call this, but there
1697 * should be no harm.
1699 * The main purpose of this is to make sure that a CHECK_CONDITION
1700 * is properly treated.
1702 void scsi_report_bus_reset(struct Scsi_Host
*shost
, int channel
)
1704 struct scsi_device
*sdev
;
1706 __shost_for_each_device(sdev
, shost
) {
1707 if (channel
== sdev_channel(sdev
))
1708 __scsi_report_device_reset(sdev
, NULL
);
1711 EXPORT_SYMBOL(scsi_report_bus_reset
);
1714 * Function: scsi_report_device_reset()
1716 * Purpose: Utility function used by low-level drivers to report that
1717 * they have observed a device reset on the device being handled.
1719 * Arguments: shost - Host in question
1720 * channel - channel on which reset was observed
1721 * target - target on which reset was observed
1725 * Lock status: Host lock must be held
1727 * Notes: This only needs to be called if the reset is one which
1728 * originates from an unknown location. Resets originated
1729 * by the mid-level itself don't need to call this, but there
1730 * should be no harm.
1732 * The main purpose of this is to make sure that a CHECK_CONDITION
1733 * is properly treated.
1735 void scsi_report_device_reset(struct Scsi_Host
*shost
, int channel
, int target
)
1737 struct scsi_device
*sdev
;
1739 __shost_for_each_device(sdev
, shost
) {
1740 if (channel
== sdev_channel(sdev
) &&
1741 target
== sdev_id(sdev
))
1742 __scsi_report_device_reset(sdev
, NULL
);
1745 EXPORT_SYMBOL(scsi_report_device_reset
);
1748 scsi_reset_provider_done_command(struct scsi_cmnd
*scmd
)
1753 * Function: scsi_reset_provider
1755 * Purpose: Send requested reset to a bus or device at any phase.
1757 * Arguments: device - device to send reset to
1758 * flag - reset type (see scsi.h)
1760 * Returns: SUCCESS/FAILURE.
1762 * Notes: This is used by the SCSI Generic driver to provide
1763 * Bus/Device reset capability.
1766 scsi_reset_provider(struct scsi_device
*dev
, int flag
)
1768 struct scsi_cmnd
*scmd
= scsi_get_command(dev
, GFP_KERNEL
);
1769 struct Scsi_Host
*shost
= dev
->host
;
1771 unsigned long flags
;
1774 scmd
->request
= &req
;
1775 memset(&scmd
->eh_timeout
, 0, sizeof(scmd
->eh_timeout
));
1777 memset(&scmd
->cmnd
, '\0', sizeof(scmd
->cmnd
));
1779 scmd
->scsi_done
= scsi_reset_provider_done_command
;
1780 memset(&scmd
->sdb
, 0, sizeof(scmd
->sdb
));
1784 scmd
->sc_data_direction
= DMA_BIDIRECTIONAL
;
1786 init_timer(&scmd
->eh_timeout
);
1788 spin_lock_irqsave(shost
->host_lock
, flags
);
1789 shost
->tmf_in_progress
= 1;
1790 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1793 case SCSI_TRY_RESET_DEVICE
:
1794 rtn
= scsi_try_bus_device_reset(scmd
);
1798 case SCSI_TRY_RESET_TARGET
:
1799 rtn
= scsi_try_target_reset(scmd
);
1803 case SCSI_TRY_RESET_BUS
:
1804 rtn
= scsi_try_bus_reset(scmd
);
1808 case SCSI_TRY_RESET_HOST
:
1809 rtn
= scsi_try_host_reset(scmd
);
1815 spin_lock_irqsave(shost
->host_lock
, flags
);
1816 shost
->tmf_in_progress
= 0;
1817 spin_unlock_irqrestore(shost
->host_lock
, flags
);
1820 * be sure to wake up anyone who was sleeping or had their queue
1821 * suspended while we performed the TMF.
1823 SCSI_LOG_ERROR_RECOVERY(3,
1824 printk("%s: waking up host to restart after TMF\n",
1827 wake_up(&shost
->host_wait
);
1829 scsi_run_host_queues(shost
);
1831 scsi_next_command(scmd
);
1834 EXPORT_SYMBOL(scsi_reset_provider
);
1837 * scsi_normalize_sense - normalize main elements from either fixed or
1838 * descriptor sense data format into a common format.
1840 * @sense_buffer: byte array containing sense data returned by device
1841 * @sb_len: number of valid bytes in sense_buffer
1842 * @sshdr: pointer to instance of structure that common
1843 * elements are written to.
1846 * The "main elements" from sense data are: response_code, sense_key,
1847 * asc, ascq and additional_length (only for descriptor format).
1849 * Typically this function can be called after a device has
1850 * responded to a SCSI command with the CHECK_CONDITION status.
1853 * 1 if valid sense data information found, else 0;
1855 int scsi_normalize_sense(const u8
*sense_buffer
, int sb_len
,
1856 struct scsi_sense_hdr
*sshdr
)
1858 if (!sense_buffer
|| !sb_len
)
1861 memset(sshdr
, 0, sizeof(struct scsi_sense_hdr
));
1863 sshdr
->response_code
= (sense_buffer
[0] & 0x7f);
1865 if (!scsi_sense_valid(sshdr
))
1868 if (sshdr
->response_code
>= 0x72) {
1873 sshdr
->sense_key
= (sense_buffer
[1] & 0xf);
1875 sshdr
->asc
= sense_buffer
[2];
1877 sshdr
->ascq
= sense_buffer
[3];
1879 sshdr
->additional_length
= sense_buffer
[7];
1885 sshdr
->sense_key
= (sense_buffer
[2] & 0xf);
1887 sb_len
= (sb_len
< (sense_buffer
[7] + 8)) ?
1888 sb_len
: (sense_buffer
[7] + 8);
1890 sshdr
->asc
= sense_buffer
[12];
1892 sshdr
->ascq
= sense_buffer
[13];
1898 EXPORT_SYMBOL(scsi_normalize_sense
);
1900 int scsi_command_normalize_sense(struct scsi_cmnd
*cmd
,
1901 struct scsi_sense_hdr
*sshdr
)
1903 return scsi_normalize_sense(cmd
->sense_buffer
,
1904 SCSI_SENSE_BUFFERSIZE
, sshdr
);
1906 EXPORT_SYMBOL(scsi_command_normalize_sense
);
1909 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1910 * @sense_buffer: byte array of descriptor format sense data
1911 * @sb_len: number of valid bytes in sense_buffer
1912 * @desc_type: value of descriptor type to find
1913 * (e.g. 0 -> information)
1916 * only valid when sense data is in descriptor format
1919 * pointer to start of (first) descriptor if found else NULL
1921 const u8
* scsi_sense_desc_find(const u8
* sense_buffer
, int sb_len
,
1924 int add_sen_len
, add_len
, desc_len
, k
;
1927 if ((sb_len
< 8) || (0 == (add_sen_len
= sense_buffer
[7])))
1929 if ((sense_buffer
[0] < 0x72) || (sense_buffer
[0] > 0x73))
1931 add_sen_len
= (add_sen_len
< (sb_len
- 8)) ?
1932 add_sen_len
: (sb_len
- 8);
1933 descp
= &sense_buffer
[8];
1934 for (desc_len
= 0, k
= 0; k
< add_sen_len
; k
+= desc_len
) {
1936 add_len
= (k
< (add_sen_len
- 1)) ? descp
[1]: -1;
1937 desc_len
= add_len
+ 2;
1938 if (descp
[0] == desc_type
)
1940 if (add_len
< 0) // short descriptor ??
1945 EXPORT_SYMBOL(scsi_sense_desc_find
);
1948 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1949 * @sense_buffer: byte array of sense data
1950 * @sb_len: number of valid bytes in sense_buffer
1951 * @info_out: pointer to 64 integer where 8 or 4 byte information
1952 * field will be placed if found.
1955 * 1 if information field found, 0 if not found.
1957 int scsi_get_sense_info_fld(const u8
* sense_buffer
, int sb_len
,
1966 switch (sense_buffer
[0] & 0x7f) {
1969 if (sense_buffer
[0] & 0x80) {
1970 *info_out
= (sense_buffer
[3] << 24) +
1971 (sense_buffer
[4] << 16) +
1972 (sense_buffer
[5] << 8) + sense_buffer
[6];
1978 ucp
= scsi_sense_desc_find(sense_buffer
, sb_len
,
1980 if (ucp
&& (0xa == ucp
[1])) {
1982 for (j
= 0; j
< 8; ++j
) {
1995 EXPORT_SYMBOL(scsi_get_sense_info_fld
);
1998 * scsi_build_sense_buffer - build sense data in a buffer
1999 * @desc: Sense format (non zero == descriptor format,
2000 * 0 == fixed format)
2001 * @buf: Where to build sense data
2003 * @asc: Additional sense code
2004 * @ascq: Additional sense code qualifier
2007 void scsi_build_sense_buffer(int desc
, u8
*buf
, u8 key
, u8 asc
, u8 ascq
)
2010 buf
[0] = 0x72; /* descriptor, current */
2016 buf
[0] = 0x70; /* fixed, current */
2023 EXPORT_SYMBOL(scsi_build_sense_buffer
);