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 #include <trace/events/scsi.h>
44 #define SENSE_TIMEOUT (10*HZ)
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
50 #define BUS_RESET_SETTLE_TIME (10)
51 #define HOST_RESET_SETTLE_TIME (10)
53 /* called with shost->host_lock held */
54 void scsi_eh_wakeup(struct Scsi_Host *shost)
56 if (shost->host_busy == shost->host_failed) {
57 trace_scsi_eh_wakeup(shost);
58 wake_up_process(shost->ehandler);
59 SCSI_LOG_ERROR_RECOVERY(5,
60 printk("Waking error handler thread\n"));
65 * scsi_schedule_eh - schedule EH for SCSI host
66 * @shost: SCSI host to invoke error handling on.
68 * Schedule SCSI EH without scmd.
70 void scsi_schedule_eh(struct Scsi_Host *shost)
74 spin_lock_irqsave(shost->host_lock, flags);
76 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
77 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
78 shost->host_eh_scheduled++;
79 scsi_eh_wakeup(shost);
82 spin_unlock_irqrestore(shost->host_lock, flags);
84 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
87 * scsi_eh_scmd_add - add scsi cmd to error handling.
88 * @scmd: scmd to run eh on.
89 * @eh_flag: optional SCSI_EH flag.
94 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
96 struct Scsi_Host *shost = scmd->device->host;
100 if (!shost->ehandler)
103 spin_lock_irqsave(shost->host_lock, flags);
104 if (scsi_host_set_state(shost, SHOST_RECOVERY))
105 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
109 scmd->eh_eflags |= eh_flag;
110 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
111 shost->host_failed++;
112 scsi_eh_wakeup(shost);
114 spin_unlock_irqrestore(shost->host_lock, flags);
119 * scsi_times_out - Timeout function for normal scsi commands.
120 * @req: request that is timing out.
123 * We do not need to lock this. There is the potential for a race
124 * only in that the normal completion handling might run, but if the
125 * normal completion function determines that the timer has already
126 * fired, then it mustn't do anything.
128 enum blk_eh_timer_return scsi_times_out(struct request *req)
130 struct scsi_cmnd *scmd = req->special;
131 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
133 trace_scsi_dispatch_cmd_timeout(scmd);
134 scsi_log_completion(scmd, TIMEOUT_ERROR);
136 if (scmd->device->host->transportt->eh_timed_out)
137 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
138 else if (scmd->device->host->hostt->eh_timed_out)
139 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
141 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
142 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
143 scmd->result |= DID_TIME_OUT << 16;
144 rtn = BLK_EH_HANDLED;
151 * scsi_block_when_processing_errors - Prevent cmds from being queued.
152 * @sdev: Device on which we are performing recovery.
155 * We block until the host is out of error recovery, and then check to
156 * see whether the host or the device is offline.
159 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
161 int scsi_block_when_processing_errors(struct scsi_device *sdev)
165 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
167 online = scsi_device_online(sdev);
169 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
174 EXPORT_SYMBOL(scsi_block_when_processing_errors);
176 #ifdef CONFIG_SCSI_LOGGING
178 * scsi_eh_prt_fail_stats - Log info on failures.
179 * @shost: scsi host being recovered.
180 * @work_q: Queue of scsi cmds to process.
182 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
183 struct list_head *work_q)
185 struct scsi_cmnd *scmd;
186 struct scsi_device *sdev;
187 int total_failures = 0;
190 int devices_failed = 0;
192 shost_for_each_device(sdev, shost) {
193 list_for_each_entry(scmd, work_q, eh_entry) {
194 if (scmd->device == sdev) {
196 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
203 if (cmd_cancel || cmd_failed) {
204 SCSI_LOG_ERROR_RECOVERY(3,
205 sdev_printk(KERN_INFO, sdev,
206 "%s: cmds failed: %d, cancel: %d\n",
207 __func__, cmd_failed,
215 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
216 " devices require eh work\n",
217 total_failures, devices_failed));
222 * scsi_check_sense - Examine scsi cmd sense
223 * @scmd: Cmd to have sense checked.
226 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
229 * When a deferred error is detected the current command has
230 * not been executed and needs retrying.
232 static int scsi_check_sense(struct scsi_cmnd *scmd)
234 struct scsi_device *sdev = scmd->device;
235 struct scsi_sense_hdr sshdr;
237 if (! scsi_command_normalize_sense(scmd, &sshdr))
238 return FAILED; /* no valid sense data */
240 if (scsi_sense_is_deferred(&sshdr))
243 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
244 sdev->scsi_dh_data->scsi_dh->check_sense) {
247 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
248 if (rc != SCSI_RETURN_NOT_HANDLED)
250 /* handler does not care. Drop down to default handling */
254 * Previous logic looked for FILEMARK, EOM or ILI which are
255 * mainly associated with tapes and returned SUCCESS.
257 if (sshdr.response_code == 0x70) {
259 if (scmd->sense_buffer[2] & 0xe0)
263 * descriptor format: look for "stream commands sense data
264 * descriptor" (see SSC-3). Assume single sense data
265 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
267 if ((sshdr.additional_length > 3) &&
268 (scmd->sense_buffer[8] == 0x4) &&
269 (scmd->sense_buffer[11] & 0xe0))
273 switch (sshdr.sense_key) {
276 case RECOVERED_ERROR:
277 return /* soft_error */ SUCCESS;
279 case ABORTED_COMMAND:
280 if (sshdr.asc == 0x10) /* DIF */
287 * if we are expecting a cc/ua because of a bus reset that we
288 * performed, treat this just as a retry. otherwise this is
289 * information that we should pass up to the upper-level driver
290 * so that we can deal with it there.
292 if (scmd->device->expecting_cc_ua) {
293 scmd->device->expecting_cc_ua = 0;
297 * if the device is in the process of becoming ready, we
300 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
303 * if the device is not started, we need to wake
304 * the error handler to start the motor
306 if (scmd->device->allow_restart &&
307 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
310 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
311 scmd_printk(KERN_WARNING, scmd,
312 "Warning! Received an indication that the "
313 "LUN assignments on this target have "
314 "changed. The Linux SCSI layer does not "
315 "automatically remap LUN assignments.\n");
316 else if (sshdr.asc == 0x3f)
317 scmd_printk(KERN_WARNING, scmd,
318 "Warning! Received an indication that the "
319 "operating parameters on this target have "
320 "changed. The Linux SCSI layer does not "
321 "automatically adjust these parameters.\n");
324 * Pass the UA upwards for a determination in the completion
329 /* these are not supported */
331 case VOLUME_OVERFLOW:
338 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
339 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
340 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
346 if (scmd->device->retry_hwerror)
347 return ADD_TO_MLQUEUE;
351 case ILLEGAL_REQUEST:
357 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
359 struct scsi_host_template *sht = sdev->host->hostt;
360 struct scsi_device *tmp_sdev;
362 if (!sht->change_queue_depth ||
363 sdev->queue_depth >= sdev->max_queue_depth)
366 if (time_before(jiffies,
367 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
370 if (time_before(jiffies,
371 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
375 * Walk all devices of a target and do
378 shost_for_each_device(tmp_sdev, sdev->host) {
379 if (tmp_sdev->channel != sdev->channel ||
380 tmp_sdev->id != sdev->id ||
381 tmp_sdev->queue_depth == sdev->max_queue_depth)
384 * call back into LLD to increase queue_depth by one
385 * with ramp up reason code.
387 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
388 SCSI_QDEPTH_RAMP_UP);
389 sdev->last_queue_ramp_up = jiffies;
393 static void scsi_handle_queue_full(struct scsi_device *sdev)
395 struct scsi_host_template *sht = sdev->host->hostt;
396 struct scsi_device *tmp_sdev;
398 if (!sht->change_queue_depth)
401 shost_for_each_device(tmp_sdev, sdev->host) {
402 if (tmp_sdev->channel != sdev->channel ||
403 tmp_sdev->id != sdev->id)
406 * We do not know the number of commands that were at
407 * the device when we got the queue full so we start
408 * from the highest possible value and work our way down.
410 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
416 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
417 * @scmd: SCSI cmd to examine.
420 * This is *only* called when we are examining the status of commands
421 * queued during error recovery. the main difference here is that we
422 * don't allow for the possibility of retries here, and we are a lot
423 * more restrictive about what we consider acceptable.
425 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
428 * first check the host byte, to see if there is anything in there
429 * that would indicate what we need to do.
431 if (host_byte(scmd->result) == DID_RESET) {
433 * rats. we are already in the error handler, so we now
434 * get to try and figure out what to do next. if the sense
435 * is valid, we have a pretty good idea of what to do.
436 * if not, we mark it as FAILED.
438 return scsi_check_sense(scmd);
440 if (host_byte(scmd->result) != DID_OK)
444 * next, check the message byte.
446 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
450 * now, check the status byte to see if this indicates
453 switch (status_byte(scmd->result)) {
455 scsi_handle_queue_ramp_up(scmd->device);
456 case COMMAND_TERMINATED:
458 case CHECK_CONDITION:
459 return scsi_check_sense(scmd);
461 case INTERMEDIATE_GOOD:
462 case INTERMEDIATE_C_GOOD:
464 * who knows? FIXME(eric)
467 case RESERVATION_CONFLICT:
468 if (scmd->cmnd[0] == TEST_UNIT_READY)
469 /* it is a success, we probed the device and
472 /* otherwise, we failed to send the command */
475 scsi_handle_queue_full(scmd->device);
486 * scsi_eh_done - Completion function for error handling.
487 * @scmd: Cmd that is done.
489 static void scsi_eh_done(struct scsi_cmnd *scmd)
491 struct completion *eh_action;
493 SCSI_LOG_ERROR_RECOVERY(3,
494 printk("%s scmd: %p result: %x\n",
495 __func__, scmd, scmd->result));
497 eh_action = scmd->device->host->eh_action;
503 * scsi_try_host_reset - ask host adapter to reset itself
504 * @scmd: SCSI cmd to send hsot reset.
506 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
511 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
514 if (!scmd->device->host->hostt->eh_host_reset_handler)
517 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
519 if (rtn == SUCCESS) {
520 if (!scmd->device->host->hostt->skip_settle_delay)
521 ssleep(HOST_RESET_SETTLE_TIME);
522 spin_lock_irqsave(scmd->device->host->host_lock, flags);
523 scsi_report_bus_reset(scmd->device->host,
525 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
532 * scsi_try_bus_reset - ask host to perform a bus reset
533 * @scmd: SCSI cmd to send bus reset.
535 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
540 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
543 if (!scmd->device->host->hostt->eh_bus_reset_handler)
546 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
548 if (rtn == SUCCESS) {
549 if (!scmd->device->host->hostt->skip_settle_delay)
550 ssleep(BUS_RESET_SETTLE_TIME);
551 spin_lock_irqsave(scmd->device->host->host_lock, flags);
552 scsi_report_bus_reset(scmd->device->host,
554 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
560 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
563 sdev->expecting_cc_ua = 1;
567 * scsi_try_target_reset - Ask host to perform a target reset
568 * @scmd: SCSI cmd used to send a target reset
571 * There is no timeout for this operation. if this operation is
572 * unreliable for a given host, then the host itself needs to put a
573 * timer on it, and set the host back to a consistent state prior to
576 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
581 if (!scmd->device->host->hostt->eh_target_reset_handler)
584 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
585 if (rtn == SUCCESS) {
586 spin_lock_irqsave(scmd->device->host->host_lock, flags);
587 __starget_for_each_device(scsi_target(scmd->device), NULL,
588 __scsi_report_device_reset);
589 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
596 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
597 * @scmd: SCSI cmd used to send BDR
600 * There is no timeout for this operation. if this operation is
601 * unreliable for a given host, then the host itself needs to put a
602 * timer on it, and set the host back to a consistent state prior to
605 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
609 if (!scmd->device->host->hostt->eh_device_reset_handler)
612 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
614 __scsi_report_device_reset(scmd->device, NULL);
618 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
620 if (!scmd->device->host->hostt->eh_abort_handler)
623 return scmd->device->host->hostt->eh_abort_handler(scmd);
626 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
628 if (scsi_try_to_abort_cmd(scmd) != SUCCESS)
629 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
630 if (scsi_try_target_reset(scmd) != SUCCESS)
631 if (scsi_try_bus_reset(scmd) != SUCCESS)
632 scsi_try_host_reset(scmd);
636 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
637 * @scmd: SCSI command structure to hijack
638 * @ses: structure to save restore information
639 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
640 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
641 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
643 * This function is used to save a scsi command information before re-execution
644 * as part of the error recovery process. If @sense_bytes is 0 the command
645 * sent must be one that does not transfer any data. If @sense_bytes != 0
646 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
647 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
649 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
650 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
652 struct scsi_device *sdev = scmd->device;
655 * We need saved copies of a number of fields - this is because
656 * error handling may need to overwrite these with different values
657 * to run different commands, and once error handling is complete,
658 * we will need to restore these values prior to running the actual
661 ses->cmd_len = scmd->cmd_len;
662 ses->cmnd = scmd->cmnd;
663 ses->data_direction = scmd->sc_data_direction;
664 ses->sdb = scmd->sdb;
665 ses->next_rq = scmd->request->next_rq;
666 ses->result = scmd->result;
667 ses->underflow = scmd->underflow;
668 ses->prot_op = scmd->prot_op;
670 scmd->prot_op = SCSI_PROT_NORMAL;
671 scmd->cmnd = ses->eh_cmnd;
672 memset(scmd->cmnd, 0, BLK_MAX_CDB);
673 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
674 scmd->request->next_rq = NULL;
677 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
679 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
681 scmd->sdb.table.sgl = &ses->sense_sgl;
682 scmd->sc_data_direction = DMA_FROM_DEVICE;
683 scmd->sdb.table.nents = 1;
684 scmd->cmnd[0] = REQUEST_SENSE;
685 scmd->cmnd[4] = scmd->sdb.length;
686 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
688 scmd->sc_data_direction = DMA_NONE;
690 BUG_ON(cmnd_size > BLK_MAX_CDB);
691 memcpy(scmd->cmnd, cmnd, cmnd_size);
692 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
698 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
699 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
700 (sdev->lun << 5 & 0xe0);
703 * Zero the sense buffer. The scsi spec mandates that any
704 * untransferred sense data should be interpreted as being zero.
706 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
708 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
711 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
712 * @scmd: SCSI command structure to restore
713 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
715 * Undo any damage done by above scsi_eh_prep_cmnd().
717 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
720 * Restore original data
722 scmd->cmd_len = ses->cmd_len;
723 scmd->cmnd = ses->cmnd;
724 scmd->sc_data_direction = ses->data_direction;
725 scmd->sdb = ses->sdb;
726 scmd->request->next_rq = ses->next_rq;
727 scmd->result = ses->result;
728 scmd->underflow = ses->underflow;
729 scmd->prot_op = ses->prot_op;
731 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
734 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
735 * @scmd: SCSI command structure to hijack
737 * @cmnd_size: size in bytes of @cmnd
738 * @timeout: timeout for this request
739 * @sense_bytes: size of sense data to copy or 0
741 * This function is used to send a scsi command down to a target device
742 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
745 * SUCCESS or FAILED or NEEDS_RETRY
747 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
748 int cmnd_size, int timeout, unsigned sense_bytes)
750 struct scsi_device *sdev = scmd->device;
751 struct Scsi_Host *shost = sdev->host;
752 DECLARE_COMPLETION_ONSTACK(done);
753 unsigned long timeleft;
754 struct scsi_eh_save ses;
757 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
758 shost->eh_action = &done;
761 scmd->scsi_done = scsi_eh_done;
762 shost->hostt->queuecommand(shost, scmd);
764 timeleft = wait_for_completion_timeout(&done, timeout);
766 shost->eh_action = NULL;
768 scsi_log_completion(scmd, SUCCESS);
770 SCSI_LOG_ERROR_RECOVERY(3,
771 printk("%s: scmd: %p, timeleft: %ld\n",
772 __func__, scmd, timeleft));
775 * If there is time left scsi_eh_done got called, and we will
776 * examine the actual status codes to see whether the command
777 * actually did complete normally, else tell the host to forget
778 * about this command.
781 rtn = scsi_eh_completed_normally(scmd);
782 SCSI_LOG_ERROR_RECOVERY(3,
783 printk("%s: scsi_eh_completed_normally %x\n",
800 scsi_abort_eh_cmnd(scmd);
804 scsi_eh_restore_cmnd(scmd, &ses);
809 * scsi_request_sense - Request sense data from a particular target.
810 * @scmd: SCSI cmd for request sense.
813 * Some hosts automatically obtain this information, others require
814 * that we obtain it on our own. This function will *not* return until
815 * the command either times out, or it completes.
817 static int scsi_request_sense(struct scsi_cmnd *scmd)
819 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
823 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
824 * @scmd: Original SCSI cmd that eh has finished.
825 * @done_q: Queue for processed commands.
828 * We don't want to use the normal command completion while we are are
829 * still handling errors - it may cause other commands to be queued,
830 * and that would disturb what we are doing. Thus we really want to
831 * keep a list of pending commands for final completion, and once we
832 * are ready to leave error handling we handle completion for real.
834 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
836 scmd->device->host->host_failed--;
838 list_move_tail(&scmd->eh_entry, done_q);
840 EXPORT_SYMBOL(scsi_eh_finish_cmd);
843 * scsi_eh_get_sense - Get device sense data.
844 * @work_q: Queue of commands to process.
845 * @done_q: Queue of processed commands.
848 * See if we need to request sense information. if so, then get it
849 * now, so we have a better idea of what to do.
852 * This has the unfortunate side effect that if a shost adapter does
853 * not automatically request sense information, we end up shutting
854 * it down before we request it.
856 * All drivers should request sense information internally these days,
857 * so for now all I have to say is tough noogies if you end up in here.
859 * XXX: Long term this code should go away, but that needs an audit of
862 int scsi_eh_get_sense(struct list_head *work_q,
863 struct list_head *done_q)
865 struct scsi_cmnd *scmd, *next;
868 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
869 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
870 SCSI_SENSE_VALID(scmd))
873 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
874 "%s: requesting sense\n",
876 rtn = scsi_request_sense(scmd);
880 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
881 " result %x\n", scmd,
883 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
885 rtn = scsi_decide_disposition(scmd);
888 * if the result was normal, then just pass it along to the
892 /* we don't want this command reissued, just
893 * finished with the sense data, so set
894 * retries to the max allowed to ensure it
895 * won't get reissued */
896 scmd->retries = scmd->allowed;
897 else if (rtn != NEEDS_RETRY)
900 scsi_eh_finish_cmd(scmd, done_q);
903 return list_empty(work_q);
905 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
908 * scsi_eh_tur - Send TUR to device.
909 * @scmd: &scsi_cmnd to send TUR
912 * 0 - Device is ready. 1 - Device NOT ready.
914 static int scsi_eh_tur(struct scsi_cmnd *scmd)
916 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
917 int retry_cnt = 1, rtn;
920 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
922 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
923 __func__, scmd, rtn));
938 * scsi_eh_abort_cmds - abort pending commands.
939 * @work_q: &list_head for pending commands.
940 * @done_q: &list_head for processed commands.
943 * Try and see whether or not it makes sense to try and abort the
944 * running command. This only works out to be the case if we have one
945 * command that has timed out. If the command simply failed, it makes
946 * no sense to try and abort the command, since as far as the shost
947 * adapter is concerned, it isn't running.
949 static int scsi_eh_abort_cmds(struct list_head *work_q,
950 struct list_head *done_q)
952 struct scsi_cmnd *scmd, *next;
955 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
956 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
958 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
959 "0x%p\n", current->comm,
961 rtn = scsi_try_to_abort_cmd(scmd);
962 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
963 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
964 if (!scsi_device_online(scmd->device) ||
965 rtn == FAST_IO_FAIL ||
966 !scsi_eh_tur(scmd)) {
967 scsi_eh_finish_cmd(scmd, done_q);
971 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
978 return list_empty(work_q);
982 * scsi_eh_try_stu - Send START_UNIT to device.
983 * @scmd: &scsi_cmnd to send START_UNIT
986 * 0 - Device is ready. 1 - Device NOT ready.
988 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
990 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
992 if (scmd->device->allow_restart) {
993 int i, rtn = NEEDS_RETRY;
995 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
996 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1006 * scsi_eh_stu - send START_UNIT if needed
1007 * @shost: &scsi host being recovered.
1008 * @work_q: &list_head for pending commands.
1009 * @done_q: &list_head for processed commands.
1012 * If commands are failing due to not ready, initializing command required,
1013 * try revalidating the device, which will end up sending a start unit.
1015 static int scsi_eh_stu(struct Scsi_Host *shost,
1016 struct list_head *work_q,
1017 struct list_head *done_q)
1019 struct scsi_cmnd *scmd, *stu_scmd, *next;
1020 struct scsi_device *sdev;
1022 shost_for_each_device(sdev, shost) {
1024 list_for_each_entry(scmd, work_q, eh_entry)
1025 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1026 scsi_check_sense(scmd) == FAILED ) {
1034 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1035 " 0x%p\n", current->comm, sdev));
1037 if (!scsi_eh_try_stu(stu_scmd)) {
1038 if (!scsi_device_online(sdev) ||
1039 !scsi_eh_tur(stu_scmd)) {
1040 list_for_each_entry_safe(scmd, next,
1042 if (scmd->device == sdev)
1043 scsi_eh_finish_cmd(scmd, done_q);
1047 SCSI_LOG_ERROR_RECOVERY(3,
1048 printk("%s: START_UNIT failed to sdev:"
1049 " 0x%p\n", current->comm, sdev));
1053 return list_empty(work_q);
1058 * scsi_eh_bus_device_reset - send bdr if needed
1059 * @shost: scsi host being recovered.
1060 * @work_q: &list_head for pending commands.
1061 * @done_q: &list_head for processed commands.
1064 * Try a bus device reset. Still, look to see whether we have multiple
1065 * devices that are jammed or not - if we have multiple devices, it
1066 * makes no sense to try bus_device_reset - we really would need to try
1067 * a bus_reset instead.
1069 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1070 struct list_head *work_q,
1071 struct list_head *done_q)
1073 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1074 struct scsi_device *sdev;
1077 shost_for_each_device(sdev, shost) {
1079 list_for_each_entry(scmd, work_q, eh_entry)
1080 if (scmd->device == sdev) {
1088 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1089 " 0x%p\n", current->comm,
1091 rtn = scsi_try_bus_device_reset(bdr_scmd);
1092 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1093 if (!scsi_device_online(sdev) ||
1094 rtn == FAST_IO_FAIL ||
1095 !scsi_eh_tur(bdr_scmd)) {
1096 list_for_each_entry_safe(scmd, next,
1098 if (scmd->device == sdev)
1099 scsi_eh_finish_cmd(scmd,
1104 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1112 return list_empty(work_q);
1116 * scsi_eh_target_reset - send target reset if needed
1117 * @shost: scsi host being recovered.
1118 * @work_q: &list_head for pending commands.
1119 * @done_q: &list_head for processed commands.
1122 * Try a target reset.
1124 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1125 struct list_head *work_q,
1126 struct list_head *done_q)
1128 LIST_HEAD(tmp_list);
1130 list_splice_init(work_q, &tmp_list);
1132 while (!list_empty(&tmp_list)) {
1133 struct scsi_cmnd *next, *scmd;
1137 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1140 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1142 current->comm, id));
1143 rtn = scsi_try_target_reset(scmd);
1144 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1145 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1148 current->comm, id));
1149 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1150 if (scmd_id(scmd) != id)
1153 if ((rtn == SUCCESS || rtn == FAST_IO_FAIL)
1154 && (!scsi_device_online(scmd->device) ||
1155 rtn == FAST_IO_FAIL || !scsi_eh_tur(scmd)))
1156 scsi_eh_finish_cmd(scmd, done_q);
1158 /* push back on work queue for further processing */
1159 list_move(&scmd->eh_entry, work_q);
1163 return list_empty(work_q);
1167 * scsi_eh_bus_reset - send a bus reset
1168 * @shost: &scsi host being recovered.
1169 * @work_q: &list_head for pending commands.
1170 * @done_q: &list_head for processed commands.
1172 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1173 struct list_head *work_q,
1174 struct list_head *done_q)
1176 struct scsi_cmnd *scmd, *chan_scmd, *next;
1177 unsigned int channel;
1181 * we really want to loop over the various channels, and do this on
1182 * a channel by channel basis. we should also check to see if any
1183 * of the failed commands are on soft_reset devices, and if so, skip
1187 for (channel = 0; channel <= shost->max_channel; channel++) {
1189 list_for_each_entry(scmd, work_q, eh_entry) {
1190 if (channel == scmd_channel(scmd)) {
1194 * FIXME add back in some support for
1195 * soft_reset devices.
1202 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1203 " %d\n", current->comm,
1205 rtn = scsi_try_bus_reset(chan_scmd);
1206 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1207 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1208 if (channel == scmd_channel(scmd))
1209 if (!scsi_device_online(scmd->device) ||
1210 rtn == FAST_IO_FAIL ||
1212 scsi_eh_finish_cmd(scmd,
1216 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1217 " failed chan: %d\n",
1222 return list_empty(work_q);
1226 * scsi_eh_host_reset - send a host reset
1227 * @work_q: list_head for processed commands.
1228 * @done_q: list_head for processed commands.
1230 static int scsi_eh_host_reset(struct list_head *work_q,
1231 struct list_head *done_q)
1233 struct scsi_cmnd *scmd, *next;
1236 if (!list_empty(work_q)) {
1237 scmd = list_entry(work_q->next,
1238 struct scsi_cmnd, eh_entry);
1240 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1243 rtn = scsi_try_host_reset(scmd);
1244 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1245 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1246 if (!scsi_device_online(scmd->device) ||
1247 rtn == FAST_IO_FAIL ||
1248 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1250 scsi_eh_finish_cmd(scmd, done_q);
1253 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1258 return list_empty(work_q);
1262 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1263 * @work_q: list_head for processed commands.
1264 * @done_q: list_head for processed commands.
1266 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1267 struct list_head *done_q)
1269 struct scsi_cmnd *scmd, *next;
1271 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1272 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1273 "not ready after error recovery\n");
1274 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1275 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1277 * FIXME: Handle lost cmds.
1280 scsi_eh_finish_cmd(scmd, done_q);
1286 * scsi_noretry_cmd - determinte if command should be failed fast
1287 * @scmd: SCSI cmd to examine.
1289 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1291 switch (host_byte(scmd->result)) {
1295 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1297 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1299 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1300 status_byte(scmd->result) == RESERVATION_CONFLICT)
1303 case DID_SOFT_ERROR:
1304 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1307 switch (status_byte(scmd->result)) {
1308 case CHECK_CONDITION:
1310 * assume caller has checked sense and determinted
1311 * the check condition was retryable.
1313 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1314 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1322 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1323 * @scmd: SCSI cmd to examine.
1326 * This is *only* called when we are examining the status after sending
1327 * out the actual data command. any commands that are queued for error
1328 * recovery (e.g. test_unit_ready) do *not* come through here.
1330 * When this routine returns failed, it means the error handler thread
1331 * is woken. In cases where the error code indicates an error that
1332 * doesn't require the error handler read (i.e. we don't need to
1333 * abort/reset), this function should return SUCCESS.
1335 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1340 * if the device is offline, then we clearly just pass the result back
1341 * up to the top level.
1343 if (!scsi_device_online(scmd->device)) {
1344 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1351 * first check the host byte, to see if there is anything in there
1352 * that would indicate what we need to do.
1354 switch (host_byte(scmd->result)) {
1355 case DID_PASSTHROUGH:
1357 * no matter what, pass this through to the upper layer.
1358 * nuke this special code so that it looks like we are saying
1361 scmd->result &= 0xff00ffff;
1365 * looks good. drop through, and check the next byte.
1368 case DID_NO_CONNECT:
1369 case DID_BAD_TARGET:
1372 * note - this means that we just report the status back
1373 * to the top level driver, not that we actually think
1374 * that it indicates SUCCESS.
1378 * when the low level driver returns did_soft_error,
1379 * it is responsible for keeping an internal retry counter
1380 * in order to avoid endless loops (db)
1382 * actually this is a bug in this function here. we should
1383 * be mindful of the maximum number of retries specified
1384 * and not get stuck in a loop.
1386 case DID_SOFT_ERROR:
1392 return ADD_TO_MLQUEUE;
1393 case DID_TRANSPORT_DISRUPTED:
1395 * LLD/transport was disrupted during processing of the IO.
1396 * The transport class is now blocked/blocking,
1397 * and the transport will decide what to do with the IO
1398 * based on its timers and recovery capablilities if
1399 * there are enough retries.
1402 case DID_TRANSPORT_FAILFAST:
1404 * The transport decided to failfast the IO (most likely
1405 * the fast io fail tmo fired), so send IO directly upwards.
1409 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1410 status_byte(scmd->result) == RESERVATION_CONFLICT)
1412 * execute reservation conflict processing code
1423 * when we scan the bus, we get timeout messages for
1424 * these commands if there is no device available.
1425 * other hosts report did_no_connect for the same thing.
1427 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1428 scmd->cmnd[0] == INQUIRY)) {
1440 * next, check the message byte.
1442 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1446 * check the status byte to see if this indicates anything special.
1448 switch (status_byte(scmd->result)) {
1450 scsi_handle_queue_full(scmd->device);
1452 * the case of trying to send too many commands to a
1453 * tagged queueing device.
1457 * device can't talk to us at the moment. Should only
1458 * occur (SAM-3) when the task queue is empty, so will cause
1459 * the empty queue handling to trigger a stall in the
1462 return ADD_TO_MLQUEUE;
1464 scsi_handle_queue_ramp_up(scmd->device);
1465 case COMMAND_TERMINATED:
1469 case CHECK_CONDITION:
1470 rtn = scsi_check_sense(scmd);
1471 if (rtn == NEEDS_RETRY)
1473 else if (rtn == TARGET_ERROR) {
1475 * Need to modify host byte to signal a
1476 * permanent target failure
1478 scmd->result |= (DID_TARGET_FAILURE << 16);
1481 /* if rtn == FAILED, we have no sense information;
1482 * returning FAILED will wake the error handler thread
1483 * to collect the sense and redo the decide
1486 case CONDITION_GOOD:
1487 case INTERMEDIATE_GOOD:
1488 case INTERMEDIATE_C_GOOD:
1491 * who knows? FIXME(eric)
1495 case RESERVATION_CONFLICT:
1496 sdev_printk(KERN_INFO, scmd->device,
1497 "reservation conflict\n");
1498 scmd->result |= (DID_NEXUS_FAILURE << 16);
1499 return SUCCESS; /* causes immediate i/o error */
1507 /* we requeue for retry because the error was retryable, and
1508 * the request was not marked fast fail. Note that above,
1509 * even if the request is marked fast fail, we still requeue
1510 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1511 if ((++scmd->retries) <= scmd->allowed
1512 && !scsi_noretry_cmd(scmd)) {
1516 * no more retries - report this one back to upper level.
1522 static void eh_lock_door_done(struct request *req, int uptodate)
1524 __blk_put_request(req->q, req);
1528 * scsi_eh_lock_door - Prevent medium removal for the specified device
1529 * @sdev: SCSI device to prevent medium removal
1532 * We must be called from process context.
1535 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1536 * head of the devices request queue, and continue.
1538 static void scsi_eh_lock_door(struct scsi_device *sdev)
1540 struct request *req;
1543 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1544 * request becomes available
1546 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1548 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1552 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1555 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1557 req->cmd_type = REQ_TYPE_BLOCK_PC;
1558 req->cmd_flags |= REQ_QUIET;
1559 req->timeout = 10 * HZ;
1562 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1566 * scsi_restart_operations - restart io operations to the specified host.
1567 * @shost: Host we are restarting.
1570 * When we entered the error handler, we blocked all further i/o to
1571 * this device. we need to 'reverse' this process.
1573 static void scsi_restart_operations(struct Scsi_Host *shost)
1575 struct scsi_device *sdev;
1576 unsigned long flags;
1579 * If the door was locked, we need to insert a door lock request
1580 * onto the head of the SCSI request queue for the device. There
1581 * is no point trying to lock the door of an off-line device.
1583 shost_for_each_device(sdev, shost) {
1584 if (scsi_device_online(sdev) && sdev->locked)
1585 scsi_eh_lock_door(sdev);
1589 * next free up anything directly waiting upon the host. this
1590 * will be requests for character device operations, and also for
1591 * ioctls to queued block devices.
1593 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1596 spin_lock_irqsave(shost->host_lock, flags);
1597 if (scsi_host_set_state(shost, SHOST_RUNNING))
1598 if (scsi_host_set_state(shost, SHOST_CANCEL))
1599 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1600 spin_unlock_irqrestore(shost->host_lock, flags);
1602 wake_up(&shost->host_wait);
1605 * finally we need to re-initiate requests that may be pending. we will
1606 * have had everything blocked while error handling is taking place, and
1607 * now that error recovery is done, we will need to ensure that these
1608 * requests are started.
1610 scsi_run_host_queues(shost);
1614 * scsi_eh_ready_devs - check device ready state and recover if not.
1615 * @shost: host to be recovered.
1616 * @work_q: &list_head for pending commands.
1617 * @done_q: &list_head for processed commands.
1619 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1620 struct list_head *work_q,
1621 struct list_head *done_q)
1623 if (!scsi_eh_stu(shost, work_q, done_q))
1624 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1625 if (!scsi_eh_target_reset(shost, work_q, done_q))
1626 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1627 if (!scsi_eh_host_reset(work_q, done_q))
1628 scsi_eh_offline_sdevs(work_q,
1631 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1634 * scsi_eh_flush_done_q - finish processed commands or retry them.
1635 * @done_q: list_head of processed commands.
1637 void scsi_eh_flush_done_q(struct list_head *done_q)
1639 struct scsi_cmnd *scmd, *next;
1641 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1642 list_del_init(&scmd->eh_entry);
1643 if (scsi_device_online(scmd->device) &&
1644 !scsi_noretry_cmd(scmd) &&
1645 (++scmd->retries <= scmd->allowed)) {
1646 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1650 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1653 * If just we got sense for the device (called
1654 * scsi_eh_get_sense), scmd->result is already
1655 * set, do not set DRIVER_TIMEOUT.
1658 scmd->result |= (DRIVER_TIMEOUT << 24);
1659 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1661 current->comm, scmd));
1662 scsi_finish_command(scmd);
1666 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1669 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1670 * @shost: Host to unjam.
1673 * When we come in here, we *know* that all commands on the bus have
1674 * either completed, failed or timed out. we also know that no further
1675 * commands are being sent to the host, so things are relatively quiet
1676 * and we have freedom to fiddle with things as we wish.
1678 * This is only the *default* implementation. it is possible for
1679 * individual drivers to supply their own version of this function, and
1680 * if the maintainer wishes to do this, it is strongly suggested that
1681 * this function be taken as a template and modified. this function
1682 * was designed to correctly handle problems for about 95% of the
1683 * different cases out there, and it should always provide at least a
1684 * reasonable amount of error recovery.
1686 * Any command marked 'failed' or 'timeout' must eventually have
1687 * scsi_finish_cmd() called for it. we do all of the retry stuff
1688 * here, so when we restart the host after we return it should have an
1691 static void scsi_unjam_host(struct Scsi_Host *shost)
1693 unsigned long flags;
1694 LIST_HEAD(eh_work_q);
1695 LIST_HEAD(eh_done_q);
1697 spin_lock_irqsave(shost->host_lock, flags);
1698 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1699 spin_unlock_irqrestore(shost->host_lock, flags);
1701 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1703 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1704 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1705 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1707 scsi_eh_flush_done_q(&eh_done_q);
1711 * scsi_error_handler - SCSI error handler thread
1712 * @data: Host for which we are running.
1715 * This is the main error handling loop. This is run as a kernel thread
1716 * for every SCSI host and handles all error handling activity.
1718 int scsi_error_handler(void *data)
1720 struct Scsi_Host *shost = data;
1723 * We use TASK_INTERRUPTIBLE so that the thread is not
1724 * counted against the load average as a running process.
1725 * We never actually get interrupted because kthread_run
1726 * disables signal delivery for the created thread.
1728 set_current_state(TASK_INTERRUPTIBLE);
1729 while (!kthread_should_stop()) {
1730 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1731 shost->host_failed != shost->host_busy) {
1732 SCSI_LOG_ERROR_RECOVERY(1,
1733 printk("Error handler scsi_eh_%d sleeping\n",
1736 set_current_state(TASK_INTERRUPTIBLE);
1740 __set_current_state(TASK_RUNNING);
1741 SCSI_LOG_ERROR_RECOVERY(1,
1742 printk("Error handler scsi_eh_%d waking up\n",
1746 * We have a host that is failing for some reason. Figure out
1747 * what we need to do to get it up and online again (if we can).
1748 * If we fail, we end up taking the thing offline.
1750 if (scsi_autopm_get_host(shost) != 0) {
1751 SCSI_LOG_ERROR_RECOVERY(1,
1752 printk(KERN_ERR "Error handler scsi_eh_%d "
1753 "unable to autoresume\n",
1758 if (shost->transportt->eh_strategy_handler)
1759 shost->transportt->eh_strategy_handler(shost);
1761 scsi_unjam_host(shost);
1764 * Note - if the above fails completely, the action is to take
1765 * individual devices offline and flush the queue of any
1766 * outstanding requests that may have been pending. When we
1767 * restart, we restart any I/O to any other devices on the bus
1768 * which are still online.
1770 scsi_restart_operations(shost);
1771 scsi_autopm_put_host(shost);
1772 set_current_state(TASK_INTERRUPTIBLE);
1774 __set_current_state(TASK_RUNNING);
1776 SCSI_LOG_ERROR_RECOVERY(1,
1777 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1778 shost->ehandler = NULL;
1783 * Function: scsi_report_bus_reset()
1785 * Purpose: Utility function used by low-level drivers to report that
1786 * they have observed a bus reset on the bus being handled.
1788 * Arguments: shost - Host in question
1789 * channel - channel on which reset was observed.
1793 * Lock status: Host lock must be held.
1795 * Notes: This only needs to be called if the reset is one which
1796 * originates from an unknown location. Resets originated
1797 * by the mid-level itself don't need to call this, but there
1798 * should be no harm.
1800 * The main purpose of this is to make sure that a CHECK_CONDITION
1801 * is properly treated.
1803 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1805 struct scsi_device *sdev;
1807 __shost_for_each_device(sdev, shost) {
1808 if (channel == sdev_channel(sdev))
1809 __scsi_report_device_reset(sdev, NULL);
1812 EXPORT_SYMBOL(scsi_report_bus_reset);
1815 * Function: scsi_report_device_reset()
1817 * Purpose: Utility function used by low-level drivers to report that
1818 * they have observed a device reset on the device being handled.
1820 * Arguments: shost - Host in question
1821 * channel - channel on which reset was observed
1822 * target - target on which reset was observed
1826 * Lock status: Host lock must be held
1828 * Notes: This only needs to be called if the reset is one which
1829 * originates from an unknown location. Resets originated
1830 * by the mid-level itself don't need to call this, but there
1831 * should be no harm.
1833 * The main purpose of this is to make sure that a CHECK_CONDITION
1834 * is properly treated.
1836 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1838 struct scsi_device *sdev;
1840 __shost_for_each_device(sdev, shost) {
1841 if (channel == sdev_channel(sdev) &&
1842 target == sdev_id(sdev))
1843 __scsi_report_device_reset(sdev, NULL);
1846 EXPORT_SYMBOL(scsi_report_device_reset);
1849 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1854 * Function: scsi_reset_provider
1856 * Purpose: Send requested reset to a bus or device at any phase.
1858 * Arguments: device - device to send reset to
1859 * flag - reset type (see scsi.h)
1861 * Returns: SUCCESS/FAILURE.
1863 * Notes: This is used by the SCSI Generic driver to provide
1864 * Bus/Device reset capability.
1867 scsi_reset_provider(struct scsi_device *dev, int flag)
1869 struct scsi_cmnd *scmd;
1870 struct Scsi_Host *shost = dev->host;
1872 unsigned long flags;
1875 if (scsi_autopm_get_host(shost) < 0)
1878 scmd = scsi_get_command(dev, GFP_KERNEL);
1879 blk_rq_init(NULL, &req);
1880 scmd->request = &req;
1882 scmd->cmnd = req.cmd;
1884 scmd->scsi_done = scsi_reset_provider_done_command;
1885 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1889 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1891 spin_lock_irqsave(shost->host_lock, flags);
1892 shost->tmf_in_progress = 1;
1893 spin_unlock_irqrestore(shost->host_lock, flags);
1896 case SCSI_TRY_RESET_DEVICE:
1897 rtn = scsi_try_bus_device_reset(scmd);
1901 case SCSI_TRY_RESET_TARGET:
1902 rtn = scsi_try_target_reset(scmd);
1906 case SCSI_TRY_RESET_BUS:
1907 rtn = scsi_try_bus_reset(scmd);
1911 case SCSI_TRY_RESET_HOST:
1912 rtn = scsi_try_host_reset(scmd);
1918 spin_lock_irqsave(shost->host_lock, flags);
1919 shost->tmf_in_progress = 0;
1920 spin_unlock_irqrestore(shost->host_lock, flags);
1923 * be sure to wake up anyone who was sleeping or had their queue
1924 * suspended while we performed the TMF.
1926 SCSI_LOG_ERROR_RECOVERY(3,
1927 printk("%s: waking up host to restart after TMF\n",
1930 wake_up(&shost->host_wait);
1932 scsi_run_host_queues(shost);
1934 scsi_next_command(scmd);
1935 scsi_autopm_put_host(shost);
1938 EXPORT_SYMBOL(scsi_reset_provider);
1941 * scsi_normalize_sense - normalize main elements from either fixed or
1942 * descriptor sense data format into a common format.
1944 * @sense_buffer: byte array containing sense data returned by device
1945 * @sb_len: number of valid bytes in sense_buffer
1946 * @sshdr: pointer to instance of structure that common
1947 * elements are written to.
1950 * The "main elements" from sense data are: response_code, sense_key,
1951 * asc, ascq and additional_length (only for descriptor format).
1953 * Typically this function can be called after a device has
1954 * responded to a SCSI command with the CHECK_CONDITION status.
1957 * 1 if valid sense data information found, else 0;
1959 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1960 struct scsi_sense_hdr *sshdr)
1962 if (!sense_buffer || !sb_len)
1965 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1967 sshdr->response_code = (sense_buffer[0] & 0x7f);
1969 if (!scsi_sense_valid(sshdr))
1972 if (sshdr->response_code >= 0x72) {
1977 sshdr->sense_key = (sense_buffer[1] & 0xf);
1979 sshdr->asc = sense_buffer[2];
1981 sshdr->ascq = sense_buffer[3];
1983 sshdr->additional_length = sense_buffer[7];
1989 sshdr->sense_key = (sense_buffer[2] & 0xf);
1991 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1992 sb_len : (sense_buffer[7] + 8);
1994 sshdr->asc = sense_buffer[12];
1996 sshdr->ascq = sense_buffer[13];
2002 EXPORT_SYMBOL(scsi_normalize_sense);
2004 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2005 struct scsi_sense_hdr *sshdr)
2007 return scsi_normalize_sense(cmd->sense_buffer,
2008 SCSI_SENSE_BUFFERSIZE, sshdr);
2010 EXPORT_SYMBOL(scsi_command_normalize_sense);
2013 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2014 * @sense_buffer: byte array of descriptor format sense data
2015 * @sb_len: number of valid bytes in sense_buffer
2016 * @desc_type: value of descriptor type to find
2017 * (e.g. 0 -> information)
2020 * only valid when sense data is in descriptor format
2023 * pointer to start of (first) descriptor if found else NULL
2025 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2028 int add_sen_len, add_len, desc_len, k;
2031 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2033 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2035 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2036 add_sen_len : (sb_len - 8);
2037 descp = &sense_buffer[8];
2038 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2040 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2041 desc_len = add_len + 2;
2042 if (descp[0] == desc_type)
2044 if (add_len < 0) // short descriptor ??
2049 EXPORT_SYMBOL(scsi_sense_desc_find);
2052 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2053 * @sense_buffer: byte array of sense data
2054 * @sb_len: number of valid bytes in sense_buffer
2055 * @info_out: pointer to 64 integer where 8 or 4 byte information
2056 * field will be placed if found.
2059 * 1 if information field found, 0 if not found.
2061 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2070 switch (sense_buffer[0] & 0x7f) {
2073 if (sense_buffer[0] & 0x80) {
2074 *info_out = (sense_buffer[3] << 24) +
2075 (sense_buffer[4] << 16) +
2076 (sense_buffer[5] << 8) + sense_buffer[6];
2082 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2084 if (ucp && (0xa == ucp[1])) {
2086 for (j = 0; j < 8; ++j) {
2099 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2102 * scsi_build_sense_buffer - build sense data in a buffer
2103 * @desc: Sense format (non zero == descriptor format,
2104 * 0 == fixed format)
2105 * @buf: Where to build sense data
2107 * @asc: Additional sense code
2108 * @ascq: Additional sense code qualifier
2111 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2114 buf[0] = 0x72; /* descriptor, current */
2120 buf[0] = 0x70; /* fixed, current */
2127 EXPORT_SYMBOL(scsi_build_sense_buffer);