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>
28 #include <linux/jiffies.h>
30 #include <scsi/scsi.h>
31 #include <scsi/scsi_cmnd.h>
32 #include <scsi/scsi_dbg.h>
33 #include <scsi/scsi_device.h>
34 #include <scsi/scsi_driver.h>
35 #include <scsi/scsi_eh.h>
36 #include <scsi/scsi_transport.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_ioctl.h>
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
43 #include "scsi_transport_api.h"
45 #include <trace/events/scsi.h>
47 static void scsi_eh_done(struct scsi_cmnd *scmd);
50 * These should *probably* be handled by the host itself.
51 * Since it is allowed to sleep, it probably should.
53 #define BUS_RESET_SETTLE_TIME (10)
54 #define HOST_RESET_SETTLE_TIME (10)
56 static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
57 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
60 /* called with shost->host_lock held */
61 void scsi_eh_wakeup(struct Scsi_Host *shost)
63 if (atomic_read(&shost->host_busy) == shost->host_failed) {
64 trace_scsi_eh_wakeup(shost);
65 wake_up_process(shost->ehandler);
66 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
67 "Waking error handler thread\n"));
72 * scsi_schedule_eh - schedule EH for SCSI host
73 * @shost: SCSI host to invoke error handling on.
75 * Schedule SCSI EH without scmd.
77 void scsi_schedule_eh(struct Scsi_Host *shost)
81 spin_lock_irqsave(shost->host_lock, flags);
83 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
84 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
85 shost->host_eh_scheduled++;
86 scsi_eh_wakeup(shost);
89 spin_unlock_irqrestore(shost->host_lock, flags);
91 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
93 static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
95 if (!shost->last_reset || shost->eh_deadline == -1)
99 * 32bit accesses are guaranteed to be atomic
100 * (on all supported architectures), so instead
101 * of using a spinlock we can as well double check
102 * if eh_deadline has been set to 'off' during the
105 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
106 shost->eh_deadline > -1)
113 * scmd_eh_abort_handler - Handle command aborts
114 * @work: command to be aborted.
117 scmd_eh_abort_handler(struct work_struct *work)
119 struct scsi_cmnd *scmd =
120 container_of(work, struct scsi_cmnd, abort_work.work);
121 struct scsi_device *sdev = scmd->device;
124 if (scsi_host_eh_past_deadline(sdev->host)) {
125 SCSI_LOG_ERROR_RECOVERY(3,
126 scmd_printk(KERN_INFO, scmd,
127 "scmd %p eh timeout, not aborting\n",
130 SCSI_LOG_ERROR_RECOVERY(3,
131 scmd_printk(KERN_INFO, scmd,
132 "aborting command %p\n", scmd));
133 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
134 if (rtn == SUCCESS) {
135 set_host_byte(scmd, DID_TIME_OUT);
136 if (scsi_host_eh_past_deadline(sdev->host)) {
137 SCSI_LOG_ERROR_RECOVERY(3,
138 scmd_printk(KERN_INFO, scmd,
139 "scmd %p eh timeout, "
140 "not retrying aborted "
142 } else if (!scsi_noretry_cmd(scmd) &&
143 (++scmd->retries <= scmd->allowed)) {
144 SCSI_LOG_ERROR_RECOVERY(3,
145 scmd_printk(KERN_WARNING, scmd,
147 "aborted command\n", scmd));
148 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
151 SCSI_LOG_ERROR_RECOVERY(3,
152 scmd_printk(KERN_WARNING, scmd,
154 "aborted command\n", scmd));
155 scsi_finish_command(scmd);
159 SCSI_LOG_ERROR_RECOVERY(3,
160 scmd_printk(KERN_INFO, scmd,
161 "scmd %p abort %s\n", scmd,
162 (rtn == FAST_IO_FAIL) ?
163 "not send" : "failed"));
167 if (!scsi_eh_scmd_add(scmd, 0)) {
168 SCSI_LOG_ERROR_RECOVERY(3,
169 scmd_printk(KERN_WARNING, scmd,
171 "aborted command\n", scmd));
172 set_host_byte(scmd, DID_TIME_OUT);
173 scsi_finish_command(scmd);
178 * scsi_abort_command - schedule a command abort
179 * @scmd: scmd to abort.
181 * We only need to abort commands after a command timeout
184 scsi_abort_command(struct scsi_cmnd *scmd)
186 struct scsi_device *sdev = scmd->device;
187 struct Scsi_Host *shost = sdev->host;
190 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
192 * Retry after abort failed, escalate to next level.
194 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
195 SCSI_LOG_ERROR_RECOVERY(3,
196 scmd_printk(KERN_INFO, scmd,
197 "scmd %p previous abort failed\n", scmd));
198 BUG_ON(delayed_work_pending(&scmd->abort_work));
203 * Do not try a command abort if
204 * SCSI EH has already started.
206 spin_lock_irqsave(shost->host_lock, flags);
207 if (scsi_host_in_recovery(shost)) {
208 spin_unlock_irqrestore(shost->host_lock, flags);
209 SCSI_LOG_ERROR_RECOVERY(3,
210 scmd_printk(KERN_INFO, scmd,
211 "scmd %p not aborting, host in recovery\n",
216 if (shost->eh_deadline != -1 && !shost->last_reset)
217 shost->last_reset = jiffies;
218 spin_unlock_irqrestore(shost->host_lock, flags);
220 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
221 SCSI_LOG_ERROR_RECOVERY(3,
222 scmd_printk(KERN_INFO, scmd,
223 "scmd %p abort scheduled\n", scmd));
224 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
229 * scsi_eh_scmd_add - add scsi cmd to error handling.
230 * @scmd: scmd to run eh on.
231 * @eh_flag: optional SCSI_EH flag.
236 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
238 struct Scsi_Host *shost = scmd->device->host;
242 if (!shost->ehandler)
245 spin_lock_irqsave(shost->host_lock, flags);
246 if (scsi_host_set_state(shost, SHOST_RECOVERY))
247 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
250 if (shost->eh_deadline != -1 && !shost->last_reset)
251 shost->last_reset = jiffies;
254 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
255 eh_flag &= ~SCSI_EH_CANCEL_CMD;
256 scmd->eh_eflags |= eh_flag;
257 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
258 shost->host_failed++;
259 scsi_eh_wakeup(shost);
261 spin_unlock_irqrestore(shost->host_lock, flags);
266 * scsi_times_out - Timeout function for normal scsi commands.
267 * @req: request that is timing out.
270 * We do not need to lock this. There is the potential for a race
271 * only in that the normal completion handling might run, but if the
272 * normal completion function determines that the timer has already
273 * fired, then it mustn't do anything.
275 enum blk_eh_timer_return scsi_times_out(struct request *req)
277 struct scsi_cmnd *scmd = req->special;
278 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
279 struct Scsi_Host *host = scmd->device->host;
281 trace_scsi_dispatch_cmd_timeout(scmd);
282 scsi_log_completion(scmd, TIMEOUT_ERROR);
284 if (host->eh_deadline != -1 && !host->last_reset)
285 host->last_reset = jiffies;
287 if (host->transportt->eh_timed_out)
288 rtn = host->transportt->eh_timed_out(scmd);
289 else if (host->hostt->eh_timed_out)
290 rtn = host->hostt->eh_timed_out(scmd);
292 if (rtn == BLK_EH_NOT_HANDLED) {
293 if (!host->hostt->no_async_abort &&
294 scsi_abort_command(scmd) == SUCCESS)
295 return BLK_EH_NOT_HANDLED;
297 set_host_byte(scmd, DID_TIME_OUT);
298 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
299 rtn = BLK_EH_HANDLED;
306 * scsi_block_when_processing_errors - Prevent cmds from being queued.
307 * @sdev: Device on which we are performing recovery.
310 * We block until the host is out of error recovery, and then check to
311 * see whether the host or the device is offline.
314 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
316 int scsi_block_when_processing_errors(struct scsi_device *sdev)
320 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
322 online = scsi_device_online(sdev);
324 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
325 "%s: rtn: %d\n", __func__, online));
329 EXPORT_SYMBOL(scsi_block_when_processing_errors);
331 #ifdef CONFIG_SCSI_LOGGING
333 * scsi_eh_prt_fail_stats - Log info on failures.
334 * @shost: scsi host being recovered.
335 * @work_q: Queue of scsi cmds to process.
337 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
338 struct list_head *work_q)
340 struct scsi_cmnd *scmd;
341 struct scsi_device *sdev;
342 int total_failures = 0;
345 int devices_failed = 0;
347 shost_for_each_device(sdev, shost) {
348 list_for_each_entry(scmd, work_q, eh_entry) {
349 if (scmd->device == sdev) {
351 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
358 if (cmd_cancel || cmd_failed) {
359 SCSI_LOG_ERROR_RECOVERY(3,
360 shost_printk(KERN_INFO, shost,
361 "%s: cmds failed: %d, cancel: %d\n",
362 __func__, cmd_failed,
370 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
371 "Total of %d commands on %d"
372 " devices require eh work\n",
373 total_failures, devices_failed));
378 * scsi_report_lun_change - Set flag on all *other* devices on the same target
379 * to indicate that a UNIT ATTENTION is expected.
380 * @sdev: Device reporting the UNIT ATTENTION
382 static void scsi_report_lun_change(struct scsi_device *sdev)
384 sdev->sdev_target->expecting_lun_change = 1;
388 * scsi_report_sense - Examine scsi sense information and log messages for
389 * certain conditions, also issue uevents for some of them.
390 * @sdev: Device reporting the sense code
391 * @sshdr: sshdr to be examined
393 static void scsi_report_sense(struct scsi_device *sdev,
394 struct scsi_sense_hdr *sshdr)
396 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
398 if (sshdr->sense_key == UNIT_ATTENTION) {
399 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
400 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
401 sdev_printk(KERN_WARNING, sdev,
402 "Inquiry data has changed");
403 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
404 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
405 scsi_report_lun_change(sdev);
406 sdev_printk(KERN_WARNING, sdev,
407 "Warning! Received an indication that the "
408 "LUN assignments on this target have "
409 "changed. The Linux SCSI layer does not "
410 "automatically remap LUN assignments.\n");
411 } else if (sshdr->asc == 0x3f)
412 sdev_printk(KERN_WARNING, sdev,
413 "Warning! Received an indication that the "
414 "operating parameters on this target have "
415 "changed. The Linux SCSI layer does not "
416 "automatically adjust these parameters.\n");
418 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
419 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
420 sdev_printk(KERN_WARNING, sdev,
421 "Warning! Received an indication that the "
422 "LUN reached a thin provisioning soft "
426 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
427 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
428 sdev_printk(KERN_WARNING, sdev,
429 "Mode parameters changed");
430 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
431 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
432 sdev_printk(KERN_WARNING, sdev,
433 "Capacity data has changed");
434 } else if (sshdr->asc == 0x2a)
435 sdev_printk(KERN_WARNING, sdev,
436 "Parameters changed");
439 if (evt_type != SDEV_EVT_MAXBITS) {
440 set_bit(evt_type, sdev->pending_events);
441 schedule_work(&sdev->event_work);
446 * scsi_check_sense - Examine scsi cmd sense
447 * @scmd: Cmd to have sense checked.
450 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
453 * When a deferred error is detected the current command has
454 * not been executed and needs retrying.
456 static int scsi_check_sense(struct scsi_cmnd *scmd)
458 struct scsi_device *sdev = scmd->device;
459 struct scsi_sense_hdr sshdr;
461 if (! scsi_command_normalize_sense(scmd, &sshdr))
462 return FAILED; /* no valid sense data */
464 scsi_report_sense(sdev, &sshdr);
466 if (scsi_sense_is_deferred(&sshdr))
469 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
470 sdev->scsi_dh_data->scsi_dh->check_sense) {
473 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
474 if (rc != SCSI_RETURN_NOT_HANDLED)
476 /* handler does not care. Drop down to default handling */
479 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
481 * nasty: for mid-layer issued TURs, we need to return the
482 * actual sense data without any recovery attempt. For eh
483 * issued ones, we need to try to recover and interpret
488 * Previous logic looked for FILEMARK, EOM or ILI which are
489 * mainly associated with tapes and returned SUCCESS.
491 if (sshdr.response_code == 0x70) {
493 if (scmd->sense_buffer[2] & 0xe0)
497 * descriptor format: look for "stream commands sense data
498 * descriptor" (see SSC-3). Assume single sense data
499 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
501 if ((sshdr.additional_length > 3) &&
502 (scmd->sense_buffer[8] == 0x4) &&
503 (scmd->sense_buffer[11] & 0xe0))
507 switch (sshdr.sense_key) {
510 case RECOVERED_ERROR:
511 return /* soft_error */ SUCCESS;
513 case ABORTED_COMMAND:
514 if (sshdr.asc == 0x10) /* DIF */
521 * if we are expecting a cc/ua because of a bus reset that we
522 * performed, treat this just as a retry. otherwise this is
523 * information that we should pass up to the upper-level driver
524 * so that we can deal with it there.
526 if (scmd->device->expecting_cc_ua) {
528 * Because some device does not queue unit
529 * attentions correctly, we carefully check
530 * additional sense code and qualifier so as
531 * not to squash media change unit attention.
533 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
534 scmd->device->expecting_cc_ua = 0;
539 * we might also expect a cc/ua if another LUN on the target
540 * reported a UA with an ASC/ASCQ of 3F 0E -
541 * REPORTED LUNS DATA HAS CHANGED.
543 if (scmd->device->sdev_target->expecting_lun_change &&
544 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
547 * if the device is in the process of becoming ready, we
550 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
553 * if the device is not started, we need to wake
554 * the error handler to start the motor
556 if (scmd->device->allow_restart &&
557 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
560 * Pass the UA upwards for a determination in the completion
565 /* these are not supported */
567 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
568 /* Thin provisioning hard threshold reached */
569 set_host_byte(scmd, DID_ALLOC_FAILURE);
573 case VOLUME_OVERFLOW:
576 set_host_byte(scmd, DID_TARGET_FAILURE);
580 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
581 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
582 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
583 set_host_byte(scmd, DID_MEDIUM_ERROR);
589 if (scmd->device->retry_hwerror)
590 return ADD_TO_MLQUEUE;
592 set_host_byte(scmd, DID_TARGET_FAILURE);
594 case ILLEGAL_REQUEST:
595 if (sshdr.asc == 0x20 || /* Invalid command operation code */
596 sshdr.asc == 0x21 || /* Logical block address out of range */
597 sshdr.asc == 0x24 || /* Invalid field in cdb */
598 sshdr.asc == 0x26) { /* Parameter value invalid */
599 set_host_byte(scmd, DID_TARGET_FAILURE);
608 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
610 struct scsi_host_template *sht = sdev->host->hostt;
611 struct scsi_device *tmp_sdev;
613 if (!sht->change_queue_depth ||
614 sdev->queue_depth >= sdev->max_queue_depth)
617 if (time_before(jiffies,
618 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
621 if (time_before(jiffies,
622 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
626 * Walk all devices of a target and do
629 shost_for_each_device(tmp_sdev, sdev->host) {
630 if (tmp_sdev->channel != sdev->channel ||
631 tmp_sdev->id != sdev->id ||
632 tmp_sdev->queue_depth == sdev->max_queue_depth)
635 * call back into LLD to increase queue_depth by one
636 * with ramp up reason code.
638 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
639 SCSI_QDEPTH_RAMP_UP);
640 sdev->last_queue_ramp_up = jiffies;
644 static void scsi_handle_queue_full(struct scsi_device *sdev)
646 struct scsi_host_template *sht = sdev->host->hostt;
647 struct scsi_device *tmp_sdev;
649 if (!sht->change_queue_depth)
652 shost_for_each_device(tmp_sdev, sdev->host) {
653 if (tmp_sdev->channel != sdev->channel ||
654 tmp_sdev->id != sdev->id)
657 * We do not know the number of commands that were at
658 * the device when we got the queue full so we start
659 * from the highest possible value and work our way down.
661 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
667 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
668 * @scmd: SCSI cmd to examine.
671 * This is *only* called when we are examining the status of commands
672 * queued during error recovery. the main difference here is that we
673 * don't allow for the possibility of retries here, and we are a lot
674 * more restrictive about what we consider acceptable.
676 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
679 * first check the host byte, to see if there is anything in there
680 * that would indicate what we need to do.
682 if (host_byte(scmd->result) == DID_RESET) {
684 * rats. we are already in the error handler, so we now
685 * get to try and figure out what to do next. if the sense
686 * is valid, we have a pretty good idea of what to do.
687 * if not, we mark it as FAILED.
689 return scsi_check_sense(scmd);
691 if (host_byte(scmd->result) != DID_OK)
695 * next, check the message byte.
697 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
701 * now, check the status byte to see if this indicates
704 switch (status_byte(scmd->result)) {
706 scsi_handle_queue_ramp_up(scmd->device);
707 case COMMAND_TERMINATED:
709 case CHECK_CONDITION:
710 return scsi_check_sense(scmd);
712 case INTERMEDIATE_GOOD:
713 case INTERMEDIATE_C_GOOD:
715 * who knows? FIXME(eric)
718 case RESERVATION_CONFLICT:
719 if (scmd->cmnd[0] == TEST_UNIT_READY)
720 /* it is a success, we probed the device and
723 /* otherwise, we failed to send the command */
726 scsi_handle_queue_full(scmd->device);
737 * scsi_eh_done - Completion function for error handling.
738 * @scmd: Cmd that is done.
740 static void scsi_eh_done(struct scsi_cmnd *scmd)
742 struct completion *eh_action;
744 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
745 "%s scmd: %p result: %x\n",
746 __func__, scmd, scmd->result));
748 eh_action = scmd->device->host->eh_action;
754 * scsi_try_host_reset - ask host adapter to reset itself
755 * @scmd: SCSI cmd to send host reset.
757 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
761 struct Scsi_Host *host = scmd->device->host;
762 struct scsi_host_template *hostt = host->hostt;
764 SCSI_LOG_ERROR_RECOVERY(3,
765 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
767 if (!hostt->eh_host_reset_handler)
770 rtn = hostt->eh_host_reset_handler(scmd);
772 if (rtn == SUCCESS) {
773 if (!hostt->skip_settle_delay)
774 ssleep(HOST_RESET_SETTLE_TIME);
775 spin_lock_irqsave(host->host_lock, flags);
776 scsi_report_bus_reset(host, scmd_channel(scmd));
777 spin_unlock_irqrestore(host->host_lock, flags);
784 * scsi_try_bus_reset - ask host to perform a bus reset
785 * @scmd: SCSI cmd to send bus reset.
787 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
791 struct Scsi_Host *host = scmd->device->host;
792 struct scsi_host_template *hostt = host->hostt;
794 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
795 "%s: Snd Bus RST\n", __func__));
797 if (!hostt->eh_bus_reset_handler)
800 rtn = hostt->eh_bus_reset_handler(scmd);
802 if (rtn == SUCCESS) {
803 if (!hostt->skip_settle_delay)
804 ssleep(BUS_RESET_SETTLE_TIME);
805 spin_lock_irqsave(host->host_lock, flags);
806 scsi_report_bus_reset(host, scmd_channel(scmd));
807 spin_unlock_irqrestore(host->host_lock, flags);
813 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
816 sdev->expecting_cc_ua = 1;
820 * scsi_try_target_reset - Ask host to perform a target reset
821 * @scmd: SCSI cmd used to send a target reset
824 * There is no timeout for this operation. if this operation is
825 * unreliable for a given host, then the host itself needs to put a
826 * timer on it, and set the host back to a consistent state prior to
829 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
833 struct Scsi_Host *host = scmd->device->host;
834 struct scsi_host_template *hostt = host->hostt;
836 if (!hostt->eh_target_reset_handler)
839 rtn = hostt->eh_target_reset_handler(scmd);
840 if (rtn == SUCCESS) {
841 spin_lock_irqsave(host->host_lock, flags);
842 __starget_for_each_device(scsi_target(scmd->device), NULL,
843 __scsi_report_device_reset);
844 spin_unlock_irqrestore(host->host_lock, flags);
851 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
852 * @scmd: SCSI cmd used to send BDR
855 * There is no timeout for this operation. if this operation is
856 * unreliable for a given host, then the host itself needs to put a
857 * timer on it, and set the host back to a consistent state prior to
860 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
863 struct scsi_host_template *hostt = scmd->device->host->hostt;
865 if (!hostt->eh_device_reset_handler)
868 rtn = hostt->eh_device_reset_handler(scmd);
870 __scsi_report_device_reset(scmd->device, NULL);
875 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
876 * @scmd: SCSI cmd used to send a target reset
879 * SUCCESS, FAILED, or FAST_IO_FAIL
882 * SUCCESS does not necessarily indicate that the command
883 * has been aborted; it only indicates that the LLDDs
884 * has cleared all references to that command.
885 * LLDDs should return FAILED only if an abort was required
886 * but could not be executed. LLDDs should return FAST_IO_FAIL
887 * if the device is temporarily unavailable (eg due to a
888 * link down on FibreChannel)
890 static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
891 struct scsi_cmnd *scmd)
893 if (!hostt->eh_abort_handler)
896 return hostt->eh_abort_handler(scmd);
899 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
901 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
902 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
903 if (scsi_try_target_reset(scmd) != SUCCESS)
904 if (scsi_try_bus_reset(scmd) != SUCCESS)
905 scsi_try_host_reset(scmd);
909 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
910 * @scmd: SCSI command structure to hijack
911 * @ses: structure to save restore information
912 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
913 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
914 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
916 * This function is used to save a scsi command information before re-execution
917 * as part of the error recovery process. If @sense_bytes is 0 the command
918 * sent must be one that does not transfer any data. If @sense_bytes != 0
919 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
920 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
922 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
923 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
925 struct scsi_device *sdev = scmd->device;
928 * We need saved copies of a number of fields - this is because
929 * error handling may need to overwrite these with different values
930 * to run different commands, and once error handling is complete,
931 * we will need to restore these values prior to running the actual
934 ses->cmd_len = scmd->cmd_len;
935 ses->cmnd = scmd->cmnd;
936 ses->data_direction = scmd->sc_data_direction;
937 ses->sdb = scmd->sdb;
938 ses->next_rq = scmd->request->next_rq;
939 ses->result = scmd->result;
940 ses->underflow = scmd->underflow;
941 ses->prot_op = scmd->prot_op;
943 scmd->prot_op = SCSI_PROT_NORMAL;
945 scmd->cmnd = ses->eh_cmnd;
946 memset(scmd->cmnd, 0, BLK_MAX_CDB);
947 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
948 scmd->request->next_rq = NULL;
952 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
954 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
956 scmd->sdb.table.sgl = &ses->sense_sgl;
957 scmd->sc_data_direction = DMA_FROM_DEVICE;
958 scmd->sdb.table.nents = 1;
959 scmd->cmnd[0] = REQUEST_SENSE;
960 scmd->cmnd[4] = scmd->sdb.length;
961 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
963 scmd->sc_data_direction = DMA_NONE;
965 BUG_ON(cmnd_size > BLK_MAX_CDB);
966 memcpy(scmd->cmnd, cmnd, cmnd_size);
967 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
973 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
974 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
975 (sdev->lun << 5 & 0xe0);
978 * Zero the sense buffer. The scsi spec mandates that any
979 * untransferred sense data should be interpreted as being zero.
981 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
983 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
986 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
987 * @scmd: SCSI command structure to restore
988 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
990 * Undo any damage done by above scsi_eh_prep_cmnd().
992 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
995 * Restore original data
997 scmd->cmd_len = ses->cmd_len;
998 scmd->cmnd = ses->cmnd;
999 scmd->sc_data_direction = ses->data_direction;
1000 scmd->sdb = ses->sdb;
1001 scmd->request->next_rq = ses->next_rq;
1002 scmd->result = ses->result;
1003 scmd->underflow = ses->underflow;
1004 scmd->prot_op = ses->prot_op;
1006 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1009 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
1010 * @scmd: SCSI command structure to hijack
1011 * @cmnd: CDB to send
1012 * @cmnd_size: size in bytes of @cmnd
1013 * @timeout: timeout for this request
1014 * @sense_bytes: size of sense data to copy or 0
1016 * This function is used to send a scsi command down to a target device
1017 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1020 * SUCCESS or FAILED or NEEDS_RETRY
1022 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1023 int cmnd_size, int timeout, unsigned sense_bytes)
1025 struct scsi_device *sdev = scmd->device;
1026 struct Scsi_Host *shost = sdev->host;
1027 DECLARE_COMPLETION_ONSTACK(done);
1028 unsigned long timeleft = timeout;
1029 struct scsi_eh_save ses;
1030 const unsigned long stall_for = msecs_to_jiffies(100);
1034 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
1035 shost->eh_action = &done;
1037 scsi_log_send(scmd);
1038 scmd->scsi_done = scsi_eh_done;
1039 rtn = shost->hostt->queuecommand(shost, scmd);
1041 if (timeleft > stall_for) {
1042 scsi_eh_restore_cmnd(scmd, &ses);
1043 timeleft -= stall_for;
1044 msleep(jiffies_to_msecs(stall_for));
1047 /* signal not to enter either branch of the if () below */
1051 timeleft = wait_for_completion_timeout(&done, timeout);
1055 shost->eh_action = NULL;
1057 scsi_log_completion(scmd, rtn);
1059 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1060 "%s: scmd: %p, timeleft: %ld\n",
1061 __func__, scmd, timeleft));
1064 * If there is time left scsi_eh_done got called, and we will examine
1065 * the actual status codes to see whether the command actually did
1066 * complete normally, else if we have a zero return and no time left,
1067 * the command must still be pending, so abort it and return FAILED.
1068 * If we never actually managed to issue the command, because
1069 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1070 * value above (so don't execute either branch of the if)
1073 rtn = scsi_eh_completed_normally(scmd);
1074 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1075 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
1082 case ADD_TO_MLQUEUE:
1090 scsi_abort_eh_cmnd(scmd);
1094 scsi_eh_restore_cmnd(scmd, &ses);
1100 * scsi_request_sense - Request sense data from a particular target.
1101 * @scmd: SCSI cmd for request sense.
1104 * Some hosts automatically obtain this information, others require
1105 * that we obtain it on our own. This function will *not* return until
1106 * the command either times out, or it completes.
1108 static int scsi_request_sense(struct scsi_cmnd *scmd)
1110 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
1113 static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1115 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1116 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1117 if (sdrv->eh_action)
1118 rtn = sdrv->eh_action(scmd, rtn);
1124 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1125 * @scmd: Original SCSI cmd that eh has finished.
1126 * @done_q: Queue for processed commands.
1129 * We don't want to use the normal command completion while we are are
1130 * still handling errors - it may cause other commands to be queued,
1131 * and that would disturb what we are doing. Thus we really want to
1132 * keep a list of pending commands for final completion, and once we
1133 * are ready to leave error handling we handle completion for real.
1135 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
1137 scmd->device->host->host_failed--;
1138 scmd->eh_eflags = 0;
1139 list_move_tail(&scmd->eh_entry, done_q);
1141 EXPORT_SYMBOL(scsi_eh_finish_cmd);
1144 * scsi_eh_get_sense - Get device sense data.
1145 * @work_q: Queue of commands to process.
1146 * @done_q: Queue of processed commands.
1149 * See if we need to request sense information. if so, then get it
1150 * now, so we have a better idea of what to do.
1153 * This has the unfortunate side effect that if a shost adapter does
1154 * not automatically request sense information, we end up shutting
1155 * it down before we request it.
1157 * All drivers should request sense information internally these days,
1158 * so for now all I have to say is tough noogies if you end up in here.
1160 * XXX: Long term this code should go away, but that needs an audit of
1163 int scsi_eh_get_sense(struct list_head *work_q,
1164 struct list_head *done_q)
1166 struct scsi_cmnd *scmd, *next;
1167 struct Scsi_Host *shost;
1170 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1171 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
1172 SCSI_SENSE_VALID(scmd))
1175 shost = scmd->device->host;
1176 if (scsi_host_eh_past_deadline(shost)) {
1177 SCSI_LOG_ERROR_RECOVERY(3,
1178 scmd_printk(KERN_INFO, scmd,
1179 "%s: skip request sense, past eh deadline\n",
1183 if (status_byte(scmd->result) != CHECK_CONDITION)
1185 * don't request sense if there's no check condition
1186 * status because the error we're processing isn't one
1187 * that has a sense code (and some devices get
1188 * confused by sense requests out of the blue)
1192 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1193 "%s: requesting sense\n",
1195 rtn = scsi_request_sense(scmd);
1199 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1200 "sense requested for %p result %x\n",
1201 scmd, scmd->result));
1202 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
1204 rtn = scsi_decide_disposition(scmd);
1207 * if the result was normal, then just pass it along to the
1211 /* we don't want this command reissued, just
1212 * finished with the sense data, so set
1213 * retries to the max allowed to ensure it
1214 * won't get reissued */
1215 scmd->retries = scmd->allowed;
1216 else if (rtn != NEEDS_RETRY)
1219 scsi_eh_finish_cmd(scmd, done_q);
1222 return list_empty(work_q);
1224 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
1227 * scsi_eh_tur - Send TUR to device.
1228 * @scmd: &scsi_cmnd to send TUR
1231 * 0 - Device is ready. 1 - Device NOT ready.
1233 static int scsi_eh_tur(struct scsi_cmnd *scmd)
1235 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1236 int retry_cnt = 1, rtn;
1239 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1240 scmd->device->eh_timeout, 0);
1242 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1243 "%s: scmd %p rtn %x\n", __func__, scmd, rtn));
1258 * scsi_eh_test_devices - check if devices are responding from error recovery.
1259 * @cmd_list: scsi commands in error recovery.
1260 * @work_q: queue for commands which still need more error recovery
1261 * @done_q: queue for commands which are finished
1262 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1265 * Tests if devices are in a working state. Commands to devices now in
1266 * a working state are sent to the done_q while commands to devices which
1267 * are still failing to respond are returned to the work_q for more
1270 static int scsi_eh_test_devices(struct list_head *cmd_list,
1271 struct list_head *work_q,
1272 struct list_head *done_q, int try_stu)
1274 struct scsi_cmnd *scmd, *next;
1275 struct scsi_device *sdev;
1278 while (!list_empty(cmd_list)) {
1279 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1280 sdev = scmd->device;
1283 if (scsi_host_eh_past_deadline(sdev->host)) {
1284 /* Push items back onto work_q */
1285 list_splice_init(cmd_list, work_q);
1286 SCSI_LOG_ERROR_RECOVERY(3,
1287 sdev_printk(KERN_INFO, sdev,
1288 "%s: skip test device, past eh deadline",
1294 finish_cmds = !scsi_device_online(scmd->device) ||
1295 (try_stu && !scsi_eh_try_stu(scmd) &&
1296 !scsi_eh_tur(scmd)) ||
1299 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1300 if (scmd->device == sdev) {
1303 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
1304 scsi_eh_finish_cmd(scmd, done_q);
1306 list_move_tail(&scmd->eh_entry, work_q);
1309 return list_empty(work_q);
1314 * scsi_eh_abort_cmds - abort pending commands.
1315 * @work_q: &list_head for pending commands.
1316 * @done_q: &list_head for processed commands.
1319 * Try and see whether or not it makes sense to try and abort the
1320 * running command. This only works out to be the case if we have one
1321 * command that has timed out. If the command simply failed, it makes
1322 * no sense to try and abort the command, since as far as the shost
1323 * adapter is concerned, it isn't running.
1325 static int scsi_eh_abort_cmds(struct list_head *work_q,
1326 struct list_head *done_q)
1328 struct scsi_cmnd *scmd, *next;
1329 LIST_HEAD(check_list);
1331 struct Scsi_Host *shost;
1333 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1334 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
1336 shost = scmd->device->host;
1337 if (scsi_host_eh_past_deadline(shost)) {
1338 list_splice_init(&check_list, work_q);
1339 SCSI_LOG_ERROR_RECOVERY(3,
1340 scmd_printk(KERN_INFO, scmd,
1341 "%s: skip aborting cmd, past eh deadline\n",
1343 return list_empty(work_q);
1345 SCSI_LOG_ERROR_RECOVERY(3,
1346 scmd_printk(KERN_INFO, scmd,
1347 "%s: aborting cmd\n", current->comm));
1348 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
1349 if (rtn == FAILED) {
1350 SCSI_LOG_ERROR_RECOVERY(3,
1351 scmd_printk(KERN_INFO, scmd,
1352 "%s: aborting cmd failed\n",
1354 list_splice_init(&check_list, work_q);
1355 return list_empty(work_q);
1357 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1358 if (rtn == FAST_IO_FAIL)
1359 scsi_eh_finish_cmd(scmd, done_q);
1361 list_move_tail(&scmd->eh_entry, &check_list);
1364 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1368 * scsi_eh_try_stu - Send START_UNIT to device.
1369 * @scmd: &scsi_cmnd to send START_UNIT
1372 * 0 - Device is ready. 1 - Device NOT ready.
1374 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1376 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1378 if (scmd->device->allow_restart) {
1379 int i, rtn = NEEDS_RETRY;
1381 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1382 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1392 * scsi_eh_stu - send START_UNIT if needed
1393 * @shost: &scsi host being recovered.
1394 * @work_q: &list_head for pending commands.
1395 * @done_q: &list_head for processed commands.
1398 * If commands are failing due to not ready, initializing command required,
1399 * try revalidating the device, which will end up sending a start unit.
1401 static int scsi_eh_stu(struct Scsi_Host *shost,
1402 struct list_head *work_q,
1403 struct list_head *done_q)
1405 struct scsi_cmnd *scmd, *stu_scmd, *next;
1406 struct scsi_device *sdev;
1408 shost_for_each_device(sdev, shost) {
1409 if (scsi_host_eh_past_deadline(shost)) {
1410 SCSI_LOG_ERROR_RECOVERY(3,
1411 sdev_printk(KERN_INFO, sdev,
1412 "%s: skip START_UNIT, past eh deadline\n",
1417 list_for_each_entry(scmd, work_q, eh_entry)
1418 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1419 scsi_check_sense(scmd) == FAILED ) {
1427 SCSI_LOG_ERROR_RECOVERY(3,
1428 sdev_printk(KERN_INFO, sdev,
1429 "%s: Sending START_UNIT\n",
1432 if (!scsi_eh_try_stu(stu_scmd)) {
1433 if (!scsi_device_online(sdev) ||
1434 !scsi_eh_tur(stu_scmd)) {
1435 list_for_each_entry_safe(scmd, next,
1437 if (scmd->device == sdev &&
1438 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
1439 scsi_eh_finish_cmd(scmd, done_q);
1443 SCSI_LOG_ERROR_RECOVERY(3,
1444 sdev_printk(KERN_INFO, sdev,
1445 "%s: START_UNIT failed\n",
1450 return list_empty(work_q);
1455 * scsi_eh_bus_device_reset - send bdr if needed
1456 * @shost: scsi host being recovered.
1457 * @work_q: &list_head for pending commands.
1458 * @done_q: &list_head for processed commands.
1461 * Try a bus device reset. Still, look to see whether we have multiple
1462 * devices that are jammed or not - if we have multiple devices, it
1463 * makes no sense to try bus_device_reset - we really would need to try
1464 * a bus_reset instead.
1466 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1467 struct list_head *work_q,
1468 struct list_head *done_q)
1470 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1471 struct scsi_device *sdev;
1474 shost_for_each_device(sdev, shost) {
1475 if (scsi_host_eh_past_deadline(shost)) {
1476 SCSI_LOG_ERROR_RECOVERY(3,
1477 sdev_printk(KERN_INFO, sdev,
1478 "%s: skip BDR, past eh deadline\n",
1483 list_for_each_entry(scmd, work_q, eh_entry)
1484 if (scmd->device == sdev) {
1492 SCSI_LOG_ERROR_RECOVERY(3,
1493 sdev_printk(KERN_INFO, sdev,
1494 "%s: Sending BDR\n", current->comm));
1495 rtn = scsi_try_bus_device_reset(bdr_scmd);
1496 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1497 if (!scsi_device_online(sdev) ||
1498 rtn == FAST_IO_FAIL ||
1499 !scsi_eh_tur(bdr_scmd)) {
1500 list_for_each_entry_safe(scmd, next,
1502 if (scmd->device == sdev &&
1503 scsi_eh_action(scmd, rtn) != FAILED)
1504 scsi_eh_finish_cmd(scmd,
1509 SCSI_LOG_ERROR_RECOVERY(3,
1510 sdev_printk(KERN_INFO, sdev,
1511 "%s: BDR failed\n", current->comm));
1515 return list_empty(work_q);
1519 * scsi_eh_target_reset - send target reset if needed
1520 * @shost: scsi host being recovered.
1521 * @work_q: &list_head for pending commands.
1522 * @done_q: &list_head for processed commands.
1525 * Try a target reset.
1527 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1528 struct list_head *work_q,
1529 struct list_head *done_q)
1531 LIST_HEAD(tmp_list);
1532 LIST_HEAD(check_list);
1534 list_splice_init(work_q, &tmp_list);
1536 while (!list_empty(&tmp_list)) {
1537 struct scsi_cmnd *next, *scmd;
1541 if (scsi_host_eh_past_deadline(shost)) {
1542 /* push back on work queue for further processing */
1543 list_splice_init(&check_list, work_q);
1544 list_splice_init(&tmp_list, work_q);
1545 SCSI_LOG_ERROR_RECOVERY(3,
1546 shost_printk(KERN_INFO, shost,
1547 "%s: Skip target reset, past eh deadline\n",
1549 return list_empty(work_q);
1552 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1555 SCSI_LOG_ERROR_RECOVERY(3,
1556 shost_printk(KERN_INFO, shost,
1557 "%s: Sending target reset to target %d\n",
1558 current->comm, id));
1559 rtn = scsi_try_target_reset(scmd);
1560 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
1561 SCSI_LOG_ERROR_RECOVERY(3,
1562 shost_printk(KERN_INFO, shost,
1563 "%s: Target reset failed"
1565 current->comm, id));
1566 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1567 if (scmd_id(scmd) != id)
1571 list_move_tail(&scmd->eh_entry, &check_list);
1572 else if (rtn == FAST_IO_FAIL)
1573 scsi_eh_finish_cmd(scmd, done_q);
1575 /* push back on work queue for further processing */
1576 list_move(&scmd->eh_entry, work_q);
1580 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1584 * scsi_eh_bus_reset - send a bus reset
1585 * @shost: &scsi host being recovered.
1586 * @work_q: &list_head for pending commands.
1587 * @done_q: &list_head for processed commands.
1589 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1590 struct list_head *work_q,
1591 struct list_head *done_q)
1593 struct scsi_cmnd *scmd, *chan_scmd, *next;
1594 LIST_HEAD(check_list);
1595 unsigned int channel;
1599 * we really want to loop over the various channels, and do this on
1600 * a channel by channel basis. we should also check to see if any
1601 * of the failed commands are on soft_reset devices, and if so, skip
1605 for (channel = 0; channel <= shost->max_channel; channel++) {
1606 if (scsi_host_eh_past_deadline(shost)) {
1607 list_splice_init(&check_list, work_q);
1608 SCSI_LOG_ERROR_RECOVERY(3,
1609 shost_printk(KERN_INFO, shost,
1610 "%s: skip BRST, past eh deadline\n",
1612 return list_empty(work_q);
1616 list_for_each_entry(scmd, work_q, eh_entry) {
1617 if (channel == scmd_channel(scmd)) {
1621 * FIXME add back in some support for
1622 * soft_reset devices.
1629 SCSI_LOG_ERROR_RECOVERY(3,
1630 shost_printk(KERN_INFO, shost,
1631 "%s: Sending BRST chan: %d\n",
1632 current->comm, channel));
1633 rtn = scsi_try_bus_reset(chan_scmd);
1634 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1635 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1636 if (channel == scmd_channel(scmd)) {
1637 if (rtn == FAST_IO_FAIL)
1638 scsi_eh_finish_cmd(scmd,
1641 list_move_tail(&scmd->eh_entry,
1646 SCSI_LOG_ERROR_RECOVERY(3,
1647 shost_printk(KERN_INFO, shost,
1648 "%s: BRST failed chan: %d\n",
1649 current->comm, channel));
1652 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
1656 * scsi_eh_host_reset - send a host reset
1657 * @shost: host to be reset.
1658 * @work_q: &list_head for pending commands.
1659 * @done_q: &list_head for processed commands.
1661 static int scsi_eh_host_reset(struct Scsi_Host *shost,
1662 struct list_head *work_q,
1663 struct list_head *done_q)
1665 struct scsi_cmnd *scmd, *next;
1666 LIST_HEAD(check_list);
1669 if (!list_empty(work_q)) {
1670 scmd = list_entry(work_q->next,
1671 struct scsi_cmnd, eh_entry);
1673 SCSI_LOG_ERROR_RECOVERY(3,
1674 shost_printk(KERN_INFO, shost,
1675 "%s: Sending HRST\n",
1678 rtn = scsi_try_host_reset(scmd);
1679 if (rtn == SUCCESS) {
1680 list_splice_init(work_q, &check_list);
1681 } else if (rtn == FAST_IO_FAIL) {
1682 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1683 scsi_eh_finish_cmd(scmd, done_q);
1686 SCSI_LOG_ERROR_RECOVERY(3,
1687 shost_printk(KERN_INFO, shost,
1688 "%s: HRST failed\n",
1692 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
1696 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1697 * @work_q: &list_head for pending commands.
1698 * @done_q: &list_head for processed commands.
1700 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1701 struct list_head *done_q)
1703 struct scsi_cmnd *scmd, *next;
1705 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1706 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1707 "not ready after error recovery\n");
1708 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1709 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1711 * FIXME: Handle lost cmds.
1714 scsi_eh_finish_cmd(scmd, done_q);
1720 * scsi_noretry_cmd - determine if command should be failed fast
1721 * @scmd: SCSI cmd to examine.
1723 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1725 switch (host_byte(scmd->result)) {
1731 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1733 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1735 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1736 status_byte(scmd->result) == RESERVATION_CONFLICT)
1739 case DID_SOFT_ERROR:
1740 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1743 if (status_byte(scmd->result) != CHECK_CONDITION)
1748 * assume caller has checked sense and determined
1749 * the check condition was retryable.
1751 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1752 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1759 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1760 * @scmd: SCSI cmd to examine.
1763 * This is *only* called when we are examining the status after sending
1764 * out the actual data command. any commands that are queued for error
1765 * recovery (e.g. test_unit_ready) do *not* come through here.
1767 * When this routine returns failed, it means the error handler thread
1768 * is woken. In cases where the error code indicates an error that
1769 * doesn't require the error handler read (i.e. we don't need to
1770 * abort/reset), this function should return SUCCESS.
1772 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1777 * if the device is offline, then we clearly just pass the result back
1778 * up to the top level.
1780 if (!scsi_device_online(scmd->device)) {
1781 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1782 "%s: device offline - report as SUCCESS\n", __func__));
1787 * first check the host byte, to see if there is anything in there
1788 * that would indicate what we need to do.
1790 switch (host_byte(scmd->result)) {
1791 case DID_PASSTHROUGH:
1793 * no matter what, pass this through to the upper layer.
1794 * nuke this special code so that it looks like we are saying
1797 scmd->result &= 0xff00ffff;
1801 * looks good. drop through, and check the next byte.
1805 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1806 set_host_byte(scmd, DID_TIME_OUT);
1809 case DID_NO_CONNECT:
1810 case DID_BAD_TARGET:
1812 * note - this means that we just report the status back
1813 * to the top level driver, not that we actually think
1814 * that it indicates SUCCESS.
1818 * when the low level driver returns did_soft_error,
1819 * it is responsible for keeping an internal retry counter
1820 * in order to avoid endless loops (db)
1822 * actually this is a bug in this function here. we should
1823 * be mindful of the maximum number of retries specified
1824 * and not get stuck in a loop.
1826 case DID_SOFT_ERROR:
1832 return ADD_TO_MLQUEUE;
1833 case DID_TRANSPORT_DISRUPTED:
1835 * LLD/transport was disrupted during processing of the IO.
1836 * The transport class is now blocked/blocking,
1837 * and the transport will decide what to do with the IO
1838 * based on its timers and recovery capablilities if
1839 * there are enough retries.
1842 case DID_TRANSPORT_FAILFAST:
1844 * The transport decided to failfast the IO (most likely
1845 * the fast io fail tmo fired), so send IO directly upwards.
1849 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1850 status_byte(scmd->result) == RESERVATION_CONFLICT)
1852 * execute reservation conflict processing code
1862 * when we scan the bus, we get timeout messages for
1863 * these commands if there is no device available.
1864 * other hosts report did_no_connect for the same thing.
1866 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1867 scmd->cmnd[0] == INQUIRY)) {
1879 * next, check the message byte.
1881 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1885 * check the status byte to see if this indicates anything special.
1887 switch (status_byte(scmd->result)) {
1889 scsi_handle_queue_full(scmd->device);
1891 * the case of trying to send too many commands to a
1892 * tagged queueing device.
1896 * device can't talk to us at the moment. Should only
1897 * occur (SAM-3) when the task queue is empty, so will cause
1898 * the empty queue handling to trigger a stall in the
1901 return ADD_TO_MLQUEUE;
1903 if (scmd->cmnd[0] == REPORT_LUNS)
1904 scmd->device->sdev_target->expecting_lun_change = 0;
1905 scsi_handle_queue_ramp_up(scmd->device);
1906 case COMMAND_TERMINATED:
1910 case CHECK_CONDITION:
1911 rtn = scsi_check_sense(scmd);
1912 if (rtn == NEEDS_RETRY)
1914 /* if rtn == FAILED, we have no sense information;
1915 * returning FAILED will wake the error handler thread
1916 * to collect the sense and redo the decide
1919 case CONDITION_GOOD:
1920 case INTERMEDIATE_GOOD:
1921 case INTERMEDIATE_C_GOOD:
1924 * who knows? FIXME(eric)
1928 case RESERVATION_CONFLICT:
1929 sdev_printk(KERN_INFO, scmd->device,
1930 "reservation conflict\n");
1931 set_host_byte(scmd, DID_NEXUS_FAILURE);
1932 return SUCCESS; /* causes immediate i/o error */
1940 /* we requeue for retry because the error was retryable, and
1941 * the request was not marked fast fail. Note that above,
1942 * even if the request is marked fast fail, we still requeue
1943 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1944 if ((++scmd->retries) <= scmd->allowed
1945 && !scsi_noretry_cmd(scmd)) {
1949 * no more retries - report this one back to upper level.
1955 static void eh_lock_door_done(struct request *req, int uptodate)
1957 __blk_put_request(req->q, req);
1961 * scsi_eh_lock_door - Prevent medium removal for the specified device
1962 * @sdev: SCSI device to prevent medium removal
1965 * We must be called from process context.
1968 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1969 * head of the devices request queue, and continue.
1971 static void scsi_eh_lock_door(struct scsi_device *sdev)
1973 struct request *req;
1976 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1977 * request becomes available
1979 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1983 blk_rq_set_block_pc(req);
1985 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1989 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1992 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1994 req->cmd_flags |= REQ_QUIET;
1995 req->timeout = 10 * HZ;
1998 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
2002 * scsi_restart_operations - restart io operations to the specified host.
2003 * @shost: Host we are restarting.
2006 * When we entered the error handler, we blocked all further i/o to
2007 * this device. we need to 'reverse' this process.
2009 static void scsi_restart_operations(struct Scsi_Host *shost)
2011 struct scsi_device *sdev;
2012 unsigned long flags;
2015 * If the door was locked, we need to insert a door lock request
2016 * onto the head of the SCSI request queue for the device. There
2017 * is no point trying to lock the door of an off-line device.
2019 shost_for_each_device(sdev, shost) {
2020 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
2021 scsi_eh_lock_door(sdev);
2022 sdev->was_reset = 0;
2027 * next free up anything directly waiting upon the host. this
2028 * will be requests for character device operations, and also for
2029 * ioctls to queued block devices.
2031 SCSI_LOG_ERROR_RECOVERY(3,
2032 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
2034 spin_lock_irqsave(shost->host_lock, flags);
2035 if (scsi_host_set_state(shost, SHOST_RUNNING))
2036 if (scsi_host_set_state(shost, SHOST_CANCEL))
2037 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2038 spin_unlock_irqrestore(shost->host_lock, flags);
2040 wake_up(&shost->host_wait);
2043 * finally we need to re-initiate requests that may be pending. we will
2044 * have had everything blocked while error handling is taking place, and
2045 * now that error recovery is done, we will need to ensure that these
2046 * requests are started.
2048 scsi_run_host_queues(shost);
2051 * if eh is active and host_eh_scheduled is pending we need to re-run
2052 * recovery. we do this check after scsi_run_host_queues() to allow
2053 * everything pent up since the last eh run a chance to make forward
2054 * progress before we sync again. Either we'll immediately re-run
2055 * recovery or scsi_device_unbusy() will wake us again when these
2056 * pending commands complete.
2058 spin_lock_irqsave(shost->host_lock, flags);
2059 if (shost->host_eh_scheduled)
2060 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2061 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2062 spin_unlock_irqrestore(shost->host_lock, flags);
2066 * scsi_eh_ready_devs - check device ready state and recover if not.
2067 * @shost: host to be recovered.
2068 * @work_q: &list_head for pending commands.
2069 * @done_q: &list_head for processed commands.
2071 void scsi_eh_ready_devs(struct Scsi_Host *shost,
2072 struct list_head *work_q,
2073 struct list_head *done_q)
2075 if (!scsi_eh_stu(shost, work_q, done_q))
2076 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
2077 if (!scsi_eh_target_reset(shost, work_q, done_q))
2078 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2079 if (!scsi_eh_host_reset(shost, work_q, done_q))
2080 scsi_eh_offline_sdevs(work_q,
2083 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
2086 * scsi_eh_flush_done_q - finish processed commands or retry them.
2087 * @done_q: list_head of processed commands.
2089 void scsi_eh_flush_done_q(struct list_head *done_q)
2091 struct scsi_cmnd *scmd, *next;
2093 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2094 list_del_init(&scmd->eh_entry);
2095 if (scsi_device_online(scmd->device) &&
2096 !scsi_noretry_cmd(scmd) &&
2097 (++scmd->retries <= scmd->allowed)) {
2098 SCSI_LOG_ERROR_RECOVERY(3,
2099 scmd_printk(KERN_INFO, scmd,
2100 "%s: flush retry cmd: %p\n",
2101 current->comm, scmd));
2102 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2105 * If just we got sense for the device (called
2106 * scsi_eh_get_sense), scmd->result is already
2107 * set, do not set DRIVER_TIMEOUT.
2110 scmd->result |= (DRIVER_TIMEOUT << 24);
2111 SCSI_LOG_ERROR_RECOVERY(3,
2112 scmd_printk(KERN_INFO, scmd,
2113 "%s: flush finish cmd: %p\n",
2114 current->comm, scmd));
2115 scsi_finish_command(scmd);
2119 EXPORT_SYMBOL(scsi_eh_flush_done_q);
2122 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2123 * @shost: Host to unjam.
2126 * When we come in here, we *know* that all commands on the bus have
2127 * either completed, failed or timed out. we also know that no further
2128 * commands are being sent to the host, so things are relatively quiet
2129 * and we have freedom to fiddle with things as we wish.
2131 * This is only the *default* implementation. it is possible for
2132 * individual drivers to supply their own version of this function, and
2133 * if the maintainer wishes to do this, it is strongly suggested that
2134 * this function be taken as a template and modified. this function
2135 * was designed to correctly handle problems for about 95% of the
2136 * different cases out there, and it should always provide at least a
2137 * reasonable amount of error recovery.
2139 * Any command marked 'failed' or 'timeout' must eventually have
2140 * scsi_finish_cmd() called for it. we do all of the retry stuff
2141 * here, so when we restart the host after we return it should have an
2144 static void scsi_unjam_host(struct Scsi_Host *shost)
2146 unsigned long flags;
2147 LIST_HEAD(eh_work_q);
2148 LIST_HEAD(eh_done_q);
2150 spin_lock_irqsave(shost->host_lock, flags);
2151 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2152 spin_unlock_irqrestore(shost->host_lock, flags);
2154 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2156 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2157 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2158 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2160 spin_lock_irqsave(shost->host_lock, flags);
2161 if (shost->eh_deadline != -1)
2162 shost->last_reset = 0;
2163 spin_unlock_irqrestore(shost->host_lock, flags);
2164 scsi_eh_flush_done_q(&eh_done_q);
2168 * scsi_error_handler - SCSI error handler thread
2169 * @data: Host for which we are running.
2172 * This is the main error handling loop. This is run as a kernel thread
2173 * for every SCSI host and handles all error handling activity.
2175 int scsi_error_handler(void *data)
2177 struct Scsi_Host *shost = data;
2180 * We use TASK_INTERRUPTIBLE so that the thread is not
2181 * counted against the load average as a running process.
2182 * We never actually get interrupted because kthread_run
2183 * disables signal delivery for the created thread.
2185 while (!kthread_should_stop()) {
2186 set_current_state(TASK_INTERRUPTIBLE);
2187 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
2188 shost->host_failed != atomic_read(&shost->host_busy)) {
2189 SCSI_LOG_ERROR_RECOVERY(1,
2190 shost_printk(KERN_INFO, shost,
2191 "scsi_eh_%d: sleeping\n",
2197 __set_current_state(TASK_RUNNING);
2198 SCSI_LOG_ERROR_RECOVERY(1,
2199 shost_printk(KERN_INFO, shost,
2200 "scsi_eh_%d: waking up %d/%d/%d\n",
2201 shost->host_no, shost->host_eh_scheduled,
2203 atomic_read(&shost->host_busy)));
2206 * We have a host that is failing for some reason. Figure out
2207 * what we need to do to get it up and online again (if we can).
2208 * If we fail, we end up taking the thing offline.
2210 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
2211 SCSI_LOG_ERROR_RECOVERY(1,
2212 shost_printk(KERN_ERR, shost,
2213 "scsi_eh_%d: unable to autoresume\n",
2218 if (shost->transportt->eh_strategy_handler)
2219 shost->transportt->eh_strategy_handler(shost);
2221 scsi_unjam_host(shost);
2224 * Note - if the above fails completely, the action is to take
2225 * individual devices offline and flush the queue of any
2226 * outstanding requests that may have been pending. When we
2227 * restart, we restart any I/O to any other devices on the bus
2228 * which are still online.
2230 scsi_restart_operations(shost);
2231 if (!shost->eh_noresume)
2232 scsi_autopm_put_host(shost);
2234 __set_current_state(TASK_RUNNING);
2236 SCSI_LOG_ERROR_RECOVERY(1,
2237 shost_printk(KERN_INFO, shost,
2238 "Error handler scsi_eh_%d exiting\n",
2240 shost->ehandler = NULL;
2245 * Function: scsi_report_bus_reset()
2247 * Purpose: Utility function used by low-level drivers to report that
2248 * they have observed a bus reset on the bus being handled.
2250 * Arguments: shost - Host in question
2251 * channel - channel on which reset was observed.
2255 * Lock status: Host lock must be held.
2257 * Notes: This only needs to be called if the reset is one which
2258 * originates from an unknown location. Resets originated
2259 * by the mid-level itself don't need to call this, but there
2260 * should be no harm.
2262 * The main purpose of this is to make sure that a CHECK_CONDITION
2263 * is properly treated.
2265 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2267 struct scsi_device *sdev;
2269 __shost_for_each_device(sdev, shost) {
2270 if (channel == sdev_channel(sdev))
2271 __scsi_report_device_reset(sdev, NULL);
2274 EXPORT_SYMBOL(scsi_report_bus_reset);
2277 * Function: scsi_report_device_reset()
2279 * Purpose: Utility function used by low-level drivers to report that
2280 * they have observed a device reset on the device being handled.
2282 * Arguments: shost - Host in question
2283 * channel - channel on which reset was observed
2284 * target - target on which reset was observed
2288 * Lock status: Host lock must be held
2290 * Notes: This only needs to be called if the reset is one which
2291 * originates from an unknown location. Resets originated
2292 * by the mid-level itself don't need to call this, but there
2293 * should be no harm.
2295 * The main purpose of this is to make sure that a CHECK_CONDITION
2296 * is properly treated.
2298 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2300 struct scsi_device *sdev;
2302 __shost_for_each_device(sdev, shost) {
2303 if (channel == sdev_channel(sdev) &&
2304 target == sdev_id(sdev))
2305 __scsi_report_device_reset(sdev, NULL);
2308 EXPORT_SYMBOL(scsi_report_device_reset);
2311 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2316 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2317 * @dev: scsi_device to operate on
2318 * @arg: reset type (see sg.h)
2321 scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
2323 struct scsi_cmnd *scmd;
2324 struct Scsi_Host *shost = dev->host;
2326 unsigned long flags;
2327 int error = 0, rtn, val;
2329 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2332 error = get_user(val, arg);
2336 if (scsi_autopm_get_host(shost) < 0)
2340 if (!get_device(&dev->sdev_gendev))
2341 goto out_put_autopm_host;
2343 scmd = scsi_get_command(dev, GFP_KERNEL);
2345 put_device(&dev->sdev_gendev);
2346 goto out_put_autopm_host;
2349 blk_rq_init(NULL, &req);
2350 scmd->request = &req;
2352 scmd->cmnd = req.cmd;
2354 scmd->scsi_done = scsi_reset_provider_done_command;
2355 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
2359 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
2361 spin_lock_irqsave(shost->host_lock, flags);
2362 shost->tmf_in_progress = 1;
2363 spin_unlock_irqrestore(shost->host_lock, flags);
2365 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2366 case SG_SCSI_RESET_NOTHING:
2369 case SG_SCSI_RESET_DEVICE:
2370 rtn = scsi_try_bus_device_reset(scmd);
2371 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2374 case SG_SCSI_RESET_TARGET:
2375 rtn = scsi_try_target_reset(scmd);
2376 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2379 case SG_SCSI_RESET_BUS:
2380 rtn = scsi_try_bus_reset(scmd);
2381 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
2384 case SG_SCSI_RESET_HOST:
2385 rtn = scsi_try_host_reset(scmd);
2394 error = (rtn == SUCCESS) ? 0 : -EIO;
2396 spin_lock_irqsave(shost->host_lock, flags);
2397 shost->tmf_in_progress = 0;
2398 spin_unlock_irqrestore(shost->host_lock, flags);
2401 * be sure to wake up anyone who was sleeping or had their queue
2402 * suspended while we performed the TMF.
2404 SCSI_LOG_ERROR_RECOVERY(3,
2405 shost_printk(KERN_INFO, shost,
2406 "waking up host to restart after TMF\n"));
2408 wake_up(&shost->host_wait);
2410 scsi_run_host_queues(shost);
2412 scsi_next_command(scmd);
2413 out_put_autopm_host:
2414 scsi_autopm_put_host(shost);
2417 EXPORT_SYMBOL(scsi_ioctl_reset);
2420 * scsi_normalize_sense - normalize main elements from either fixed or
2421 * descriptor sense data format into a common format.
2423 * @sense_buffer: byte array containing sense data returned by device
2424 * @sb_len: number of valid bytes in sense_buffer
2425 * @sshdr: pointer to instance of structure that common
2426 * elements are written to.
2429 * The "main elements" from sense data are: response_code, sense_key,
2430 * asc, ascq and additional_length (only for descriptor format).
2432 * Typically this function can be called after a device has
2433 * responded to a SCSI command with the CHECK_CONDITION status.
2436 * true if valid sense data information found, else false;
2438 bool scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2439 struct scsi_sense_hdr *sshdr)
2441 if (!sense_buffer || !sb_len)
2444 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2446 sshdr->response_code = (sense_buffer[0] & 0x7f);
2448 if (!scsi_sense_valid(sshdr))
2451 if (sshdr->response_code >= 0x72) {
2456 sshdr->sense_key = (sense_buffer[1] & 0xf);
2458 sshdr->asc = sense_buffer[2];
2460 sshdr->ascq = sense_buffer[3];
2462 sshdr->additional_length = sense_buffer[7];
2468 sshdr->sense_key = (sense_buffer[2] & 0xf);
2470 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2471 sb_len : (sense_buffer[7] + 8);
2473 sshdr->asc = sense_buffer[12];
2475 sshdr->ascq = sense_buffer[13];
2481 EXPORT_SYMBOL(scsi_normalize_sense);
2483 bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2484 struct scsi_sense_hdr *sshdr)
2486 return scsi_normalize_sense(cmd->sense_buffer,
2487 SCSI_SENSE_BUFFERSIZE, sshdr);
2489 EXPORT_SYMBOL(scsi_command_normalize_sense);
2492 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2493 * @sense_buffer: byte array of descriptor format sense data
2494 * @sb_len: number of valid bytes in sense_buffer
2495 * @desc_type: value of descriptor type to find
2496 * (e.g. 0 -> information)
2499 * only valid when sense data is in descriptor format
2502 * pointer to start of (first) descriptor if found else NULL
2504 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2507 int add_sen_len, add_len, desc_len, k;
2510 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2512 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2514 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2515 add_sen_len : (sb_len - 8);
2516 descp = &sense_buffer[8];
2517 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2519 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2520 desc_len = add_len + 2;
2521 if (descp[0] == desc_type)
2523 if (add_len < 0) // short descriptor ??
2528 EXPORT_SYMBOL(scsi_sense_desc_find);
2531 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2532 * @sense_buffer: byte array of sense data
2533 * @sb_len: number of valid bytes in sense_buffer
2534 * @info_out: pointer to 64 integer where 8 or 4 byte information
2535 * field will be placed if found.
2538 * 1 if information field found, 0 if not found.
2540 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2549 switch (sense_buffer[0] & 0x7f) {
2552 if (sense_buffer[0] & 0x80) {
2553 *info_out = (sense_buffer[3] << 24) +
2554 (sense_buffer[4] << 16) +
2555 (sense_buffer[5] << 8) + sense_buffer[6];
2561 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2563 if (ucp && (0xa == ucp[1])) {
2565 for (j = 0; j < 8; ++j) {
2578 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2581 * scsi_build_sense_buffer - build sense data in a buffer
2582 * @desc: Sense format (non zero == descriptor format,
2583 * 0 == fixed format)
2584 * @buf: Where to build sense data
2586 * @asc: Additional sense code
2587 * @ascq: Additional sense code qualifier
2590 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2593 buf[0] = 0x72; /* descriptor, current */
2599 buf[0] = 0x70; /* fixed, current */
2606 EXPORT_SYMBOL(scsi_build_sense_buffer);