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))
307 /* these three are not supported */
309 case VOLUME_OVERFLOW:
314 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
315 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
316 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
322 if (scmd->device->retry_hwerror)
323 return ADD_TO_MLQUEUE;
327 case ILLEGAL_REQUEST:
335 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
337 struct scsi_host_template *sht = sdev->host->hostt;
338 struct scsi_device *tmp_sdev;
340 if (!sht->change_queue_depth ||
341 sdev->queue_depth >= sdev->max_queue_depth)
344 if (time_before(jiffies,
345 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
348 if (time_before(jiffies,
349 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
353 * Walk all devices of a target and do
356 shost_for_each_device(tmp_sdev, sdev->host) {
357 if (tmp_sdev->channel != sdev->channel ||
358 tmp_sdev->id != sdev->id ||
359 tmp_sdev->queue_depth == sdev->max_queue_depth)
362 * call back into LLD to increase queue_depth by one
363 * with ramp up reason code.
365 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
366 SCSI_QDEPTH_RAMP_UP);
367 sdev->last_queue_ramp_up = jiffies;
371 static void scsi_handle_queue_full(struct scsi_device *sdev)
373 struct scsi_host_template *sht = sdev->host->hostt;
374 struct scsi_device *tmp_sdev;
376 if (!sht->change_queue_depth)
379 shost_for_each_device(tmp_sdev, sdev->host) {
380 if (tmp_sdev->channel != sdev->channel ||
381 tmp_sdev->id != sdev->id)
384 * We do not know the number of commands that were at
385 * the device when we got the queue full so we start
386 * from the highest possible value and work our way down.
388 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
394 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
395 * @scmd: SCSI cmd to examine.
398 * This is *only* called when we are examining the status of commands
399 * queued during error recovery. the main difference here is that we
400 * don't allow for the possibility of retries here, and we are a lot
401 * more restrictive about what we consider acceptable.
403 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
406 * first check the host byte, to see if there is anything in there
407 * that would indicate what we need to do.
409 if (host_byte(scmd->result) == DID_RESET) {
411 * rats. we are already in the error handler, so we now
412 * get to try and figure out what to do next. if the sense
413 * is valid, we have a pretty good idea of what to do.
414 * if not, we mark it as FAILED.
416 return scsi_check_sense(scmd);
418 if (host_byte(scmd->result) != DID_OK)
422 * next, check the message byte.
424 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
428 * now, check the status byte to see if this indicates
431 switch (status_byte(scmd->result)) {
433 scsi_handle_queue_ramp_up(scmd->device);
434 case COMMAND_TERMINATED:
436 case CHECK_CONDITION:
437 return scsi_check_sense(scmd);
439 case INTERMEDIATE_GOOD:
440 case INTERMEDIATE_C_GOOD:
442 * who knows? FIXME(eric)
445 case RESERVATION_CONFLICT:
447 * let issuer deal with this, it could be just fine
451 scsi_handle_queue_full(scmd->device);
461 * scsi_eh_done - Completion function for error handling.
462 * @scmd: Cmd that is done.
464 static void scsi_eh_done(struct scsi_cmnd *scmd)
466 struct completion *eh_action;
468 SCSI_LOG_ERROR_RECOVERY(3,
469 printk("%s scmd: %p result: %x\n",
470 __func__, scmd, scmd->result));
472 eh_action = scmd->device->host->eh_action;
478 * scsi_try_host_reset - ask host adapter to reset itself
479 * @scmd: SCSI cmd to send hsot reset.
481 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
486 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
489 if (!scmd->device->host->hostt->eh_host_reset_handler)
492 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
494 if (rtn == SUCCESS) {
495 if (!scmd->device->host->hostt->skip_settle_delay)
496 ssleep(HOST_RESET_SETTLE_TIME);
497 spin_lock_irqsave(scmd->device->host->host_lock, flags);
498 scsi_report_bus_reset(scmd->device->host,
500 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
507 * scsi_try_bus_reset - ask host to perform a bus reset
508 * @scmd: SCSI cmd to send bus reset.
510 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
515 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
518 if (!scmd->device->host->hostt->eh_bus_reset_handler)
521 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
523 if (rtn == SUCCESS) {
524 if (!scmd->device->host->hostt->skip_settle_delay)
525 ssleep(BUS_RESET_SETTLE_TIME);
526 spin_lock_irqsave(scmd->device->host->host_lock, flags);
527 scsi_report_bus_reset(scmd->device->host,
529 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
535 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
538 sdev->expecting_cc_ua = 1;
542 * scsi_try_target_reset - Ask host to perform a target reset
543 * @scmd: SCSI cmd used to send a target reset
546 * There is no timeout for this operation. if this operation is
547 * unreliable for a given host, then the host itself needs to put a
548 * timer on it, and set the host back to a consistent state prior to
551 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
556 if (!scmd->device->host->hostt->eh_target_reset_handler)
559 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
560 if (rtn == SUCCESS) {
561 spin_lock_irqsave(scmd->device->host->host_lock, flags);
562 __starget_for_each_device(scsi_target(scmd->device), NULL,
563 __scsi_report_device_reset);
564 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
571 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
572 * @scmd: SCSI cmd used to send BDR
575 * There is no timeout for this operation. if this operation is
576 * unreliable for a given host, then the host itself needs to put a
577 * timer on it, and set the host back to a consistent state prior to
580 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
584 if (!scmd->device->host->hostt->eh_device_reset_handler)
587 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
589 __scsi_report_device_reset(scmd->device, NULL);
593 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
595 if (!scmd->device->host->hostt->eh_abort_handler)
598 return scmd->device->host->hostt->eh_abort_handler(scmd);
602 * scsi_try_to_abort_cmd - Ask host to abort a running command.
603 * @scmd: SCSI cmd to abort from Lower Level.
606 * This function will not return until the user's completion function
607 * has been called. there is no timeout on this operation. if the
608 * author of the low-level driver wishes this operation to be timed,
609 * they can provide this facility themselves. helper functions in
610 * scsi_error.c can be supplied to make this easier to do.
612 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
615 * scsi_done was called just after the command timed out and before
616 * we had a chance to process it. (db)
618 if (scmd->serial_number == 0)
620 return __scsi_try_to_abort_cmd(scmd);
623 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
625 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
626 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
627 if (scsi_try_target_reset(scmd) != SUCCESS)
628 if (scsi_try_bus_reset(scmd) != SUCCESS)
629 scsi_try_host_reset(scmd);
633 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
634 * @scmd: SCSI command structure to hijack
635 * @ses: structure to save restore information
636 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
637 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
638 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
640 * This function is used to save a scsi command information before re-execution
641 * as part of the error recovery process. If @sense_bytes is 0 the command
642 * sent must be one that does not transfer any data. If @sense_bytes != 0
643 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
644 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
646 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
647 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
649 struct scsi_device *sdev = scmd->device;
652 * We need saved copies of a number of fields - this is because
653 * error handling may need to overwrite these with different values
654 * to run different commands, and once error handling is complete,
655 * we will need to restore these values prior to running the actual
658 ses->cmd_len = scmd->cmd_len;
659 ses->cmnd = scmd->cmnd;
660 ses->data_direction = scmd->sc_data_direction;
661 ses->sdb = scmd->sdb;
662 ses->next_rq = scmd->request->next_rq;
663 ses->result = scmd->result;
664 ses->underflow = scmd->underflow;
665 ses->prot_op = scmd->prot_op;
667 scmd->prot_op = SCSI_PROT_NORMAL;
668 scmd->cmnd = ses->eh_cmnd;
669 memset(scmd->cmnd, 0, BLK_MAX_CDB);
670 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
671 scmd->request->next_rq = NULL;
674 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
676 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
678 scmd->sdb.table.sgl = &ses->sense_sgl;
679 scmd->sc_data_direction = DMA_FROM_DEVICE;
680 scmd->sdb.table.nents = 1;
681 scmd->cmnd[0] = REQUEST_SENSE;
682 scmd->cmnd[4] = scmd->sdb.length;
683 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
685 scmd->sc_data_direction = DMA_NONE;
687 BUG_ON(cmnd_size > BLK_MAX_CDB);
688 memcpy(scmd->cmnd, cmnd, cmnd_size);
689 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
695 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
696 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
697 (sdev->lun << 5 & 0xe0);
700 * Zero the sense buffer. The scsi spec mandates that any
701 * untransferred sense data should be interpreted as being zero.
703 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
705 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
708 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
709 * @scmd: SCSI command structure to restore
710 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
712 * Undo any damage done by above scsi_eh_prep_cmnd().
714 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
717 * Restore original data
719 scmd->cmd_len = ses->cmd_len;
720 scmd->cmnd = ses->cmnd;
721 scmd->sc_data_direction = ses->data_direction;
722 scmd->sdb = ses->sdb;
723 scmd->request->next_rq = ses->next_rq;
724 scmd->result = ses->result;
725 scmd->underflow = ses->underflow;
726 scmd->prot_op = ses->prot_op;
728 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
731 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
732 * @scmd: SCSI command structure to hijack
734 * @cmnd_size: size in bytes of @cmnd
735 * @timeout: timeout for this request
736 * @sense_bytes: size of sense data to copy or 0
738 * This function is used to send a scsi command down to a target device
739 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
742 * SUCCESS or FAILED or NEEDS_RETRY
744 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
745 int cmnd_size, int timeout, unsigned sense_bytes)
747 struct scsi_device *sdev = scmd->device;
748 struct Scsi_Host *shost = sdev->host;
749 DECLARE_COMPLETION_ONSTACK(done);
750 unsigned long timeleft;
752 struct scsi_eh_save ses;
755 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
756 shost->eh_action = &done;
758 spin_lock_irqsave(shost->host_lock, flags);
760 shost->hostt->queuecommand(scmd, scsi_eh_done);
761 spin_unlock_irqrestore(shost->host_lock, flags);
763 timeleft = wait_for_completion_timeout(&done, timeout);
765 shost->eh_action = NULL;
767 scsi_log_completion(scmd, SUCCESS);
769 SCSI_LOG_ERROR_RECOVERY(3,
770 printk("%s: scmd: %p, timeleft: %ld\n",
771 __func__, scmd, timeleft));
774 * If there is time left scsi_eh_done got called, and we will
775 * examine the actual status codes to see whether the command
776 * actually did complete normally, else tell the host to forget
777 * about this command.
780 rtn = scsi_eh_completed_normally(scmd);
781 SCSI_LOG_ERROR_RECOVERY(3,
782 printk("%s: scsi_eh_completed_normally %x\n",
798 scsi_abort_eh_cmnd(scmd);
802 scsi_eh_restore_cmnd(scmd, &ses);
807 * scsi_request_sense - Request sense data from a particular target.
808 * @scmd: SCSI cmd for request sense.
811 * Some hosts automatically obtain this information, others require
812 * that we obtain it on our own. This function will *not* return until
813 * the command either times out, or it completes.
815 static int scsi_request_sense(struct scsi_cmnd *scmd)
817 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
821 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
822 * @scmd: Original SCSI cmd that eh has finished.
823 * @done_q: Queue for processed commands.
826 * We don't want to use the normal command completion while we are are
827 * still handling errors - it may cause other commands to be queued,
828 * and that would disturb what we are doing. Thus we really want to
829 * keep a list of pending commands for final completion, and once we
830 * are ready to leave error handling we handle completion for real.
832 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
834 scmd->device->host->host_failed--;
836 list_move_tail(&scmd->eh_entry, done_q);
838 EXPORT_SYMBOL(scsi_eh_finish_cmd);
841 * scsi_eh_get_sense - Get device sense data.
842 * @work_q: Queue of commands to process.
843 * @done_q: Queue of processed commands.
846 * See if we need to request sense information. if so, then get it
847 * now, so we have a better idea of what to do.
850 * This has the unfortunate side effect that if a shost adapter does
851 * not automatically request sense information, we end up shutting
852 * it down before we request it.
854 * All drivers should request sense information internally these days,
855 * so for now all I have to say is tough noogies if you end up in here.
857 * XXX: Long term this code should go away, but that needs an audit of
860 int scsi_eh_get_sense(struct list_head *work_q,
861 struct list_head *done_q)
863 struct scsi_cmnd *scmd, *next;
866 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
867 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
868 SCSI_SENSE_VALID(scmd))
871 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
872 "%s: requesting sense\n",
874 rtn = scsi_request_sense(scmd);
878 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
879 " result %x\n", scmd,
881 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
883 rtn = scsi_decide_disposition(scmd);
886 * if the result was normal, then just pass it along to the
890 /* we don't want this command reissued, just
891 * finished with the sense data, so set
892 * retries to the max allowed to ensure it
893 * won't get reissued */
894 scmd->retries = scmd->allowed;
895 else if (rtn != NEEDS_RETRY)
898 scsi_eh_finish_cmd(scmd, done_q);
901 return list_empty(work_q);
903 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
906 * scsi_eh_tur - Send TUR to device.
907 * @scmd: &scsi_cmnd to send TUR
910 * 0 - Device is ready. 1 - Device NOT ready.
912 static int scsi_eh_tur(struct scsi_cmnd *scmd)
914 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
915 int retry_cnt = 1, rtn;
918 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
920 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
921 __func__, scmd, rtn));
936 * scsi_eh_abort_cmds - abort pending commands.
937 * @work_q: &list_head for pending commands.
938 * @done_q: &list_head for processed commands.
941 * Try and see whether or not it makes sense to try and abort the
942 * running command. This only works out to be the case if we have one
943 * command that has timed out. If the command simply failed, it makes
944 * no sense to try and abort the command, since as far as the shost
945 * adapter is concerned, it isn't running.
947 static int scsi_eh_abort_cmds(struct list_head *work_q,
948 struct list_head *done_q)
950 struct scsi_cmnd *scmd, *next;
953 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
954 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
956 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
957 "0x%p\n", current->comm,
959 rtn = scsi_try_to_abort_cmd(scmd);
960 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
961 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
962 if (!scsi_device_online(scmd->device) ||
963 rtn == FAST_IO_FAIL ||
964 !scsi_eh_tur(scmd)) {
965 scsi_eh_finish_cmd(scmd, done_q);
969 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
976 return list_empty(work_q);
980 * scsi_eh_try_stu - Send START_UNIT to device.
981 * @scmd: &scsi_cmnd to send START_UNIT
984 * 0 - Device is ready. 1 - Device NOT ready.
986 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
988 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
990 if (scmd->device->allow_restart) {
991 int i, rtn = NEEDS_RETRY;
993 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
994 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1004 * scsi_eh_stu - send START_UNIT if needed
1005 * @shost: &scsi host being recovered.
1006 * @work_q: &list_head for pending commands.
1007 * @done_q: &list_head for processed commands.
1010 * If commands are failing due to not ready, initializing command required,
1011 * try revalidating the device, which will end up sending a start unit.
1013 static int scsi_eh_stu(struct Scsi_Host *shost,
1014 struct list_head *work_q,
1015 struct list_head *done_q)
1017 struct scsi_cmnd *scmd, *stu_scmd, *next;
1018 struct scsi_device *sdev;
1020 shost_for_each_device(sdev, shost) {
1022 list_for_each_entry(scmd, work_q, eh_entry)
1023 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1024 scsi_check_sense(scmd) == FAILED ) {
1032 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1033 " 0x%p\n", current->comm, sdev));
1035 if (!scsi_eh_try_stu(stu_scmd)) {
1036 if (!scsi_device_online(sdev) ||
1037 !scsi_eh_tur(stu_scmd)) {
1038 list_for_each_entry_safe(scmd, next,
1040 if (scmd->device == sdev)
1041 scsi_eh_finish_cmd(scmd, done_q);
1045 SCSI_LOG_ERROR_RECOVERY(3,
1046 printk("%s: START_UNIT failed to sdev:"
1047 " 0x%p\n", current->comm, sdev));
1051 return list_empty(work_q);
1056 * scsi_eh_bus_device_reset - send bdr if needed
1057 * @shost: scsi host being recovered.
1058 * @work_q: &list_head for pending commands.
1059 * @done_q: &list_head for processed commands.
1062 * Try a bus device reset. Still, look to see whether we have multiple
1063 * devices that are jammed or not - if we have multiple devices, it
1064 * makes no sense to try bus_device_reset - we really would need to try
1065 * a bus_reset instead.
1067 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1068 struct list_head *work_q,
1069 struct list_head *done_q)
1071 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1072 struct scsi_device *sdev;
1075 shost_for_each_device(sdev, shost) {
1077 list_for_each_entry(scmd, work_q, eh_entry)
1078 if (scmd->device == sdev) {
1086 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1087 " 0x%p\n", current->comm,
1089 rtn = scsi_try_bus_device_reset(bdr_scmd);
1090 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1091 if (!scsi_device_online(sdev) ||
1092 rtn == FAST_IO_FAIL ||
1093 !scsi_eh_tur(bdr_scmd)) {
1094 list_for_each_entry_safe(scmd, next,
1096 if (scmd->device == sdev)
1097 scsi_eh_finish_cmd(scmd,
1102 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1110 return list_empty(work_q);
1114 * scsi_eh_target_reset - send target reset if needed
1115 * @shost: scsi host being recovered.
1116 * @work_q: &list_head for pending commands.
1117 * @done_q: &list_head for processed commands.
1120 * Try a target reset.
1122 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1123 struct list_head *work_q,
1124 struct list_head *done_q)
1126 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1127 unsigned int id = 0;
1132 list_for_each_entry(scmd, work_q, eh_entry) {
1133 if (id == scmd_id(scmd)) {
1139 /* not one exactly equal; find the next highest */
1140 list_for_each_entry(scmd, work_q, eh_entry) {
1141 if (scmd_id(scmd) > id &&
1143 scmd_id(tgtr_scmd) > scmd_id(scmd)))
1148 /* no more commands, that's it */
1151 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1153 current->comm, id));
1154 rtn = scsi_try_target_reset(tgtr_scmd);
1155 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1156 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1157 if (id == scmd_id(scmd))
1158 if (!scsi_device_online(scmd->device) ||
1159 rtn == FAST_IO_FAIL ||
1160 !scsi_eh_tur(tgtr_scmd))
1161 scsi_eh_finish_cmd(scmd,
1165 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1168 current->comm, id));
1172 return list_empty(work_q);
1176 * scsi_eh_bus_reset - send a bus reset
1177 * @shost: &scsi host being recovered.
1178 * @work_q: &list_head for pending commands.
1179 * @done_q: &list_head for processed commands.
1181 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1182 struct list_head *work_q,
1183 struct list_head *done_q)
1185 struct scsi_cmnd *scmd, *chan_scmd, *next;
1186 unsigned int channel;
1190 * we really want to loop over the various channels, and do this on
1191 * a channel by channel basis. we should also check to see if any
1192 * of the failed commands are on soft_reset devices, and if so, skip
1196 for (channel = 0; channel <= shost->max_channel; channel++) {
1198 list_for_each_entry(scmd, work_q, eh_entry) {
1199 if (channel == scmd_channel(scmd)) {
1203 * FIXME add back in some support for
1204 * soft_reset devices.
1211 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1212 " %d\n", current->comm,
1214 rtn = scsi_try_bus_reset(chan_scmd);
1215 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1216 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1217 if (channel == scmd_channel(scmd))
1218 if (!scsi_device_online(scmd->device) ||
1219 rtn == FAST_IO_FAIL ||
1221 scsi_eh_finish_cmd(scmd,
1225 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1226 " failed chan: %d\n",
1231 return list_empty(work_q);
1235 * scsi_eh_host_reset - send a host reset
1236 * @work_q: list_head for processed commands.
1237 * @done_q: list_head for processed commands.
1239 static int scsi_eh_host_reset(struct list_head *work_q,
1240 struct list_head *done_q)
1242 struct scsi_cmnd *scmd, *next;
1245 if (!list_empty(work_q)) {
1246 scmd = list_entry(work_q->next,
1247 struct scsi_cmnd, eh_entry);
1249 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1252 rtn = scsi_try_host_reset(scmd);
1253 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1254 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1255 if (!scsi_device_online(scmd->device) ||
1256 rtn == FAST_IO_FAIL ||
1257 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1259 scsi_eh_finish_cmd(scmd, done_q);
1262 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1267 return list_empty(work_q);
1271 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1272 * @work_q: list_head for processed commands.
1273 * @done_q: list_head for processed commands.
1275 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1276 struct list_head *done_q)
1278 struct scsi_cmnd *scmd, *next;
1280 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1281 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1282 "not ready after error recovery\n");
1283 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1284 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1286 * FIXME: Handle lost cmds.
1289 scsi_eh_finish_cmd(scmd, done_q);
1295 * scsi_noretry_cmd - determinte if command should be failed fast
1296 * @scmd: SCSI cmd to examine.
1298 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1300 switch (host_byte(scmd->result)) {
1304 return blk_failfast_transport(scmd->request);
1306 return blk_failfast_dev(scmd->request);
1308 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1309 status_byte(scmd->result) == RESERVATION_CONFLICT)
1312 case DID_SOFT_ERROR:
1313 return blk_failfast_driver(scmd->request);
1316 switch (status_byte(scmd->result)) {
1317 case CHECK_CONDITION:
1319 * assume caller has checked sense and determinted
1320 * the check condition was retryable.
1322 return blk_failfast_dev(scmd->request);
1329 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1330 * @scmd: SCSI cmd to examine.
1333 * This is *only* called when we are examining the status after sending
1334 * out the actual data command. any commands that are queued for error
1335 * recovery (e.g. test_unit_ready) do *not* come through here.
1337 * When this routine returns failed, it means the error handler thread
1338 * is woken. In cases where the error code indicates an error that
1339 * doesn't require the error handler read (i.e. we don't need to
1340 * abort/reset), this function should return SUCCESS.
1342 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1347 * if the device is offline, then we clearly just pass the result back
1348 * up to the top level.
1350 if (!scsi_device_online(scmd->device)) {
1351 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1358 * first check the host byte, to see if there is anything in there
1359 * that would indicate what we need to do.
1361 switch (host_byte(scmd->result)) {
1362 case DID_PASSTHROUGH:
1364 * no matter what, pass this through to the upper layer.
1365 * nuke this special code so that it looks like we are saying
1368 scmd->result &= 0xff00ffff;
1372 * looks good. drop through, and check the next byte.
1375 case DID_NO_CONNECT:
1376 case DID_BAD_TARGET:
1379 * note - this means that we just report the status back
1380 * to the top level driver, not that we actually think
1381 * that it indicates SUCCESS.
1385 * when the low level driver returns did_soft_error,
1386 * it is responsible for keeping an internal retry counter
1387 * in order to avoid endless loops (db)
1389 * actually this is a bug in this function here. we should
1390 * be mindful of the maximum number of retries specified
1391 * and not get stuck in a loop.
1393 case DID_SOFT_ERROR:
1399 return ADD_TO_MLQUEUE;
1400 case DID_TRANSPORT_DISRUPTED:
1402 * LLD/transport was disrupted during processing of the IO.
1403 * The transport class is now blocked/blocking,
1404 * and the transport will decide what to do with the IO
1405 * based on its timers and recovery capablilities if
1406 * there are enough retries.
1409 case DID_TRANSPORT_FAILFAST:
1411 * The transport decided to failfast the IO (most likely
1412 * the fast io fail tmo fired), so send IO directly upwards.
1416 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1417 status_byte(scmd->result) == RESERVATION_CONFLICT)
1419 * execute reservation conflict processing code
1430 * when we scan the bus, we get timeout messages for
1431 * these commands if there is no device available.
1432 * other hosts report did_no_connect for the same thing.
1434 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1435 scmd->cmnd[0] == INQUIRY)) {
1447 * next, check the message byte.
1449 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1453 * check the status byte to see if this indicates anything special.
1455 switch (status_byte(scmd->result)) {
1457 scsi_handle_queue_full(scmd->device);
1459 * the case of trying to send too many commands to a
1460 * tagged queueing device.
1464 * device can't talk to us at the moment. Should only
1465 * occur (SAM-3) when the task queue is empty, so will cause
1466 * the empty queue handling to trigger a stall in the
1469 return ADD_TO_MLQUEUE;
1471 scsi_handle_queue_ramp_up(scmd->device);
1472 case COMMAND_TERMINATED:
1476 case CHECK_CONDITION:
1477 rtn = scsi_check_sense(scmd);
1478 if (rtn == NEEDS_RETRY)
1480 /* if rtn == FAILED, we have no sense information;
1481 * returning FAILED will wake the error handler thread
1482 * to collect the sense and redo the decide
1485 case CONDITION_GOOD:
1486 case INTERMEDIATE_GOOD:
1487 case INTERMEDIATE_C_GOOD:
1490 * who knows? FIXME(eric)
1494 case RESERVATION_CONFLICT:
1495 sdev_printk(KERN_INFO, scmd->device,
1496 "reservation conflict\n");
1497 return SUCCESS; /* causes immediate i/o error */
1505 /* we requeue for retry because the error was retryable, and
1506 * the request was not marked fast fail. Note that above,
1507 * even if the request is marked fast fail, we still requeue
1508 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1509 if ((++scmd->retries) <= scmd->allowed
1510 && !scsi_noretry_cmd(scmd)) {
1514 * no more retries - report this one back to upper level.
1520 static void eh_lock_door_done(struct request *req, int uptodate)
1522 __blk_put_request(req->q, req);
1526 * scsi_eh_lock_door - Prevent medium removal for the specified device
1527 * @sdev: SCSI device to prevent medium removal
1530 * We must be called from process context.
1533 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1534 * head of the devices request queue, and continue.
1536 static void scsi_eh_lock_door(struct scsi_device *sdev)
1538 struct request *req;
1541 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1542 * request becomes available
1544 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1546 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1550 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1553 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1555 req->cmd_type = REQ_TYPE_BLOCK_PC;
1556 req->cmd_flags |= REQ_QUIET;
1557 req->timeout = 10 * HZ;
1560 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1564 * scsi_restart_operations - restart io operations to the specified host.
1565 * @shost: Host we are restarting.
1568 * When we entered the error handler, we blocked all further i/o to
1569 * this device. we need to 'reverse' this process.
1571 static void scsi_restart_operations(struct Scsi_Host *shost)
1573 struct scsi_device *sdev;
1574 unsigned long flags;
1577 * If the door was locked, we need to insert a door lock request
1578 * onto the head of the SCSI request queue for the device. There
1579 * is no point trying to lock the door of an off-line device.
1581 shost_for_each_device(sdev, shost) {
1582 if (scsi_device_online(sdev) && sdev->locked)
1583 scsi_eh_lock_door(sdev);
1587 * next free up anything directly waiting upon the host. this
1588 * will be requests for character device operations, and also for
1589 * ioctls to queued block devices.
1591 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1594 spin_lock_irqsave(shost->host_lock, flags);
1595 if (scsi_host_set_state(shost, SHOST_RUNNING))
1596 if (scsi_host_set_state(shost, SHOST_CANCEL))
1597 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1598 spin_unlock_irqrestore(shost->host_lock, flags);
1600 wake_up(&shost->host_wait);
1603 * finally we need to re-initiate requests that may be pending. we will
1604 * have had everything blocked while error handling is taking place, and
1605 * now that error recovery is done, we will need to ensure that these
1606 * requests are started.
1608 scsi_run_host_queues(shost);
1612 * scsi_eh_ready_devs - check device ready state and recover if not.
1613 * @shost: host to be recovered.
1614 * @work_q: &list_head for pending commands.
1615 * @done_q: &list_head for processed commands.
1617 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1618 struct list_head *work_q,
1619 struct list_head *done_q)
1621 if (!scsi_eh_stu(shost, work_q, done_q))
1622 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1623 if (!scsi_eh_target_reset(shost, work_q, done_q))
1624 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1625 if (!scsi_eh_host_reset(work_q, done_q))
1626 scsi_eh_offline_sdevs(work_q,
1629 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1632 * scsi_eh_flush_done_q - finish processed commands or retry them.
1633 * @done_q: list_head of processed commands.
1635 void scsi_eh_flush_done_q(struct list_head *done_q)
1637 struct scsi_cmnd *scmd, *next;
1639 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1640 list_del_init(&scmd->eh_entry);
1641 if (scsi_device_online(scmd->device) &&
1642 !scsi_noretry_cmd(scmd) &&
1643 (++scmd->retries <= scmd->allowed)) {
1644 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1648 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1651 * If just we got sense for the device (called
1652 * scsi_eh_get_sense), scmd->result is already
1653 * set, do not set DRIVER_TIMEOUT.
1656 scmd->result |= (DRIVER_TIMEOUT << 24);
1657 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1659 current->comm, scmd));
1660 scsi_finish_command(scmd);
1664 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1667 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1668 * @shost: Host to unjam.
1671 * When we come in here, we *know* that all commands on the bus have
1672 * either completed, failed or timed out. we also know that no further
1673 * commands are being sent to the host, so things are relatively quiet
1674 * and we have freedom to fiddle with things as we wish.
1676 * This is only the *default* implementation. it is possible for
1677 * individual drivers to supply their own version of this function, and
1678 * if the maintainer wishes to do this, it is strongly suggested that
1679 * this function be taken as a template and modified. this function
1680 * was designed to correctly handle problems for about 95% of the
1681 * different cases out there, and it should always provide at least a
1682 * reasonable amount of error recovery.
1684 * Any command marked 'failed' or 'timeout' must eventually have
1685 * scsi_finish_cmd() called for it. we do all of the retry stuff
1686 * here, so when we restart the host after we return it should have an
1689 static void scsi_unjam_host(struct Scsi_Host *shost)
1691 unsigned long flags;
1692 LIST_HEAD(eh_work_q);
1693 LIST_HEAD(eh_done_q);
1695 spin_lock_irqsave(shost->host_lock, flags);
1696 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1697 spin_unlock_irqrestore(shost->host_lock, flags);
1699 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1701 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1702 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1703 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1705 scsi_eh_flush_done_q(&eh_done_q);
1709 * scsi_error_handler - SCSI error handler thread
1710 * @data: Host for which we are running.
1713 * This is the main error handling loop. This is run as a kernel thread
1714 * for every SCSI host and handles all error handling activity.
1716 int scsi_error_handler(void *data)
1718 struct Scsi_Host *shost = data;
1721 * We use TASK_INTERRUPTIBLE so that the thread is not
1722 * counted against the load average as a running process.
1723 * We never actually get interrupted because kthread_run
1724 * disables signal delivery for the created thread.
1726 set_current_state(TASK_INTERRUPTIBLE);
1727 while (!kthread_should_stop()) {
1728 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1729 shost->host_failed != shost->host_busy) {
1730 SCSI_LOG_ERROR_RECOVERY(1,
1731 printk("Error handler scsi_eh_%d sleeping\n",
1734 set_current_state(TASK_INTERRUPTIBLE);
1738 __set_current_state(TASK_RUNNING);
1739 SCSI_LOG_ERROR_RECOVERY(1,
1740 printk("Error handler scsi_eh_%d waking up\n",
1744 * We have a host that is failing for some reason. Figure out
1745 * what we need to do to get it up and online again (if we can).
1746 * If we fail, we end up taking the thing offline.
1748 if (shost->transportt->eh_strategy_handler)
1749 shost->transportt->eh_strategy_handler(shost);
1751 scsi_unjam_host(shost);
1754 * Note - if the above fails completely, the action is to take
1755 * individual devices offline and flush the queue of any
1756 * outstanding requests that may have been pending. When we
1757 * restart, we restart any I/O to any other devices on the bus
1758 * which are still online.
1760 scsi_restart_operations(shost);
1761 set_current_state(TASK_INTERRUPTIBLE);
1763 __set_current_state(TASK_RUNNING);
1765 SCSI_LOG_ERROR_RECOVERY(1,
1766 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1767 shost->ehandler = NULL;
1772 * Function: scsi_report_bus_reset()
1774 * Purpose: Utility function used by low-level drivers to report that
1775 * they have observed a bus reset on the bus being handled.
1777 * Arguments: shost - Host in question
1778 * channel - channel on which reset was observed.
1782 * Lock status: Host lock must be held.
1784 * Notes: This only needs to be called if the reset is one which
1785 * originates from an unknown location. Resets originated
1786 * by the mid-level itself don't need to call this, but there
1787 * should be no harm.
1789 * The main purpose of this is to make sure that a CHECK_CONDITION
1790 * is properly treated.
1792 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1794 struct scsi_device *sdev;
1796 __shost_for_each_device(sdev, shost) {
1797 if (channel == sdev_channel(sdev))
1798 __scsi_report_device_reset(sdev, NULL);
1801 EXPORT_SYMBOL(scsi_report_bus_reset);
1804 * Function: scsi_report_device_reset()
1806 * Purpose: Utility function used by low-level drivers to report that
1807 * they have observed a device reset on the device being handled.
1809 * Arguments: shost - Host in question
1810 * channel - channel on which reset was observed
1811 * target - target on which reset was observed
1815 * Lock status: Host lock must be held
1817 * Notes: This only needs to be called if the reset is one which
1818 * originates from an unknown location. Resets originated
1819 * by the mid-level itself don't need to call this, but there
1820 * should be no harm.
1822 * The main purpose of this is to make sure that a CHECK_CONDITION
1823 * is properly treated.
1825 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1827 struct scsi_device *sdev;
1829 __shost_for_each_device(sdev, shost) {
1830 if (channel == sdev_channel(sdev) &&
1831 target == sdev_id(sdev))
1832 __scsi_report_device_reset(sdev, NULL);
1835 EXPORT_SYMBOL(scsi_report_device_reset);
1838 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1843 * Function: scsi_reset_provider
1845 * Purpose: Send requested reset to a bus or device at any phase.
1847 * Arguments: device - device to send reset to
1848 * flag - reset type (see scsi.h)
1850 * Returns: SUCCESS/FAILURE.
1852 * Notes: This is used by the SCSI Generic driver to provide
1853 * Bus/Device reset capability.
1856 scsi_reset_provider(struct scsi_device *dev, int flag)
1858 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1859 struct Scsi_Host *shost = dev->host;
1861 unsigned long flags;
1864 blk_rq_init(NULL, &req);
1865 scmd->request = &req;
1867 scmd->cmnd = req.cmd;
1869 scmd->scsi_done = scsi_reset_provider_done_command;
1870 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1874 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1876 spin_lock_irqsave(shost->host_lock, flags);
1877 shost->tmf_in_progress = 1;
1878 spin_unlock_irqrestore(shost->host_lock, flags);
1881 case SCSI_TRY_RESET_DEVICE:
1882 rtn = scsi_try_bus_device_reset(scmd);
1886 case SCSI_TRY_RESET_TARGET:
1887 rtn = scsi_try_target_reset(scmd);
1891 case SCSI_TRY_RESET_BUS:
1892 rtn = scsi_try_bus_reset(scmd);
1896 case SCSI_TRY_RESET_HOST:
1897 rtn = scsi_try_host_reset(scmd);
1903 spin_lock_irqsave(shost->host_lock, flags);
1904 shost->tmf_in_progress = 0;
1905 spin_unlock_irqrestore(shost->host_lock, flags);
1908 * be sure to wake up anyone who was sleeping or had their queue
1909 * suspended while we performed the TMF.
1911 SCSI_LOG_ERROR_RECOVERY(3,
1912 printk("%s: waking up host to restart after TMF\n",
1915 wake_up(&shost->host_wait);
1917 scsi_run_host_queues(shost);
1919 scsi_next_command(scmd);
1922 EXPORT_SYMBOL(scsi_reset_provider);
1925 * scsi_normalize_sense - normalize main elements from either fixed or
1926 * descriptor sense data format into a common format.
1928 * @sense_buffer: byte array containing sense data returned by device
1929 * @sb_len: number of valid bytes in sense_buffer
1930 * @sshdr: pointer to instance of structure that common
1931 * elements are written to.
1934 * The "main elements" from sense data are: response_code, sense_key,
1935 * asc, ascq and additional_length (only for descriptor format).
1937 * Typically this function can be called after a device has
1938 * responded to a SCSI command with the CHECK_CONDITION status.
1941 * 1 if valid sense data information found, else 0;
1943 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1944 struct scsi_sense_hdr *sshdr)
1946 if (!sense_buffer || !sb_len)
1949 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1951 sshdr->response_code = (sense_buffer[0] & 0x7f);
1953 if (!scsi_sense_valid(sshdr))
1956 if (sshdr->response_code >= 0x72) {
1961 sshdr->sense_key = (sense_buffer[1] & 0xf);
1963 sshdr->asc = sense_buffer[2];
1965 sshdr->ascq = sense_buffer[3];
1967 sshdr->additional_length = sense_buffer[7];
1973 sshdr->sense_key = (sense_buffer[2] & 0xf);
1975 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1976 sb_len : (sense_buffer[7] + 8);
1978 sshdr->asc = sense_buffer[12];
1980 sshdr->ascq = sense_buffer[13];
1986 EXPORT_SYMBOL(scsi_normalize_sense);
1988 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1989 struct scsi_sense_hdr *sshdr)
1991 return scsi_normalize_sense(cmd->sense_buffer,
1992 SCSI_SENSE_BUFFERSIZE, sshdr);
1994 EXPORT_SYMBOL(scsi_command_normalize_sense);
1997 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1998 * @sense_buffer: byte array of descriptor format sense data
1999 * @sb_len: number of valid bytes in sense_buffer
2000 * @desc_type: value of descriptor type to find
2001 * (e.g. 0 -> information)
2004 * only valid when sense data is in descriptor format
2007 * pointer to start of (first) descriptor if found else NULL
2009 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2012 int add_sen_len, add_len, desc_len, k;
2015 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2017 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2019 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2020 add_sen_len : (sb_len - 8);
2021 descp = &sense_buffer[8];
2022 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2024 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2025 desc_len = add_len + 2;
2026 if (descp[0] == desc_type)
2028 if (add_len < 0) // short descriptor ??
2033 EXPORT_SYMBOL(scsi_sense_desc_find);
2036 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2037 * @sense_buffer: byte array of sense data
2038 * @sb_len: number of valid bytes in sense_buffer
2039 * @info_out: pointer to 64 integer where 8 or 4 byte information
2040 * field will be placed if found.
2043 * 1 if information field found, 0 if not found.
2045 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2054 switch (sense_buffer[0] & 0x7f) {
2057 if (sense_buffer[0] & 0x80) {
2058 *info_out = (sense_buffer[3] << 24) +
2059 (sense_buffer[4] << 16) +
2060 (sense_buffer[5] << 8) + sense_buffer[6];
2066 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2068 if (ucp && (0xa == ucp[1])) {
2070 for (j = 0; j < 8; ++j) {
2083 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2086 * scsi_build_sense_buffer - build sense data in a buffer
2087 * @desc: Sense format (non zero == descriptor format,
2088 * 0 == fixed format)
2089 * @buf: Where to build sense data
2091 * @asc: Additional sense code
2092 * @ascq: Additional sense code qualifier
2095 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2098 buf[0] = 0x72; /* descriptor, current */
2104 buf[0] = 0x70; /* fixed, current */
2111 EXPORT_SYMBOL(scsi_build_sense_buffer);