2 * scsi_error.c Copyright (C) 1997 Eric Youngdale
4 * SCSI error/timeout handling
5 * Initial versions: Eric Youngdale. Based upon conversations with
6 * Leonard Zubkoff and David Miller at Linux Expo,
7 * ideas originating from all over the place.
9 * Restructured scsi_unjam_host and associated functions.
10 * September 04, 2002 Mike Anderson (andmike@us.ibm.com)
12 * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
14 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/string.h>
21 #include <linux/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
41 #define SENSE_TIMEOUT (10*HZ)
44 * These should *probably* be handled by the host itself.
45 * Since it is allowed to sleep, it probably should.
47 #define BUS_RESET_SETTLE_TIME (10)
48 #define HOST_RESET_SETTLE_TIME (10)
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host *shost)
53 if (shost->host_busy == shost->host_failed) {
54 wake_up_process(shost->ehandler);
55 SCSI_LOG_ERROR_RECOVERY(5,
56 printk("Waking error handler thread\n"));
61 * scsi_schedule_eh - schedule EH for SCSI host
62 * @shost: SCSI host to invoke error handling on.
64 * Schedule SCSI EH without scmd.
66 void scsi_schedule_eh(struct Scsi_Host *shost)
70 spin_lock_irqsave(shost->host_lock, flags);
72 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
73 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
74 shost->host_eh_scheduled++;
75 scsi_eh_wakeup(shost);
78 spin_unlock_irqrestore(shost->host_lock, flags);
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
83 * scsi_eh_scmd_add - add scsi cmd to error handling.
84 * @scmd: scmd to run eh on.
85 * @eh_flag: optional SCSI_EH flag.
90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
92 struct Scsi_Host *shost = scmd->device->host;
99 spin_lock_irqsave(shost->host_lock, flags);
100 if (scsi_host_set_state(shost, SHOST_RECOVERY))
101 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
105 scmd->eh_eflags |= eh_flag;
106 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
107 shost->host_failed++;
108 scsi_eh_wakeup(shost);
110 spin_unlock_irqrestore(shost->host_lock, flags);
115 * scsi_add_timer - Start timeout timer for a single scsi command.
116 * @scmd: scsi command that is about to start running.
117 * @timeout: amount of time to allow this command to run.
118 * @complete: timeout function to call if timer isn't canceled.
121 * This should be turned into an inline function. Each scsi command
122 * has its own timer, and as it is added to the queue, we set up the
123 * timer. When the command completes, we cancel the timer.
125 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
126 void (*complete)(struct scsi_cmnd *))
130 * If the clock was already running for this command, then
131 * first delete the timer. The timer handling code gets rather
132 * confused if we don't do this.
134 if (scmd->eh_timeout.function)
135 del_timer(&scmd->eh_timeout);
137 scmd->eh_timeout.data = (unsigned long)scmd;
138 scmd->eh_timeout.expires = jiffies + timeout;
139 scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
141 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142 " %d, (%p)\n", __FUNCTION__,
143 scmd, timeout, complete));
145 add_timer(&scmd->eh_timeout);
149 * scsi_delete_timer - Delete/cancel timer for a given function.
150 * @scmd: Cmd that we are canceling timer for
153 * This should be turned into an inline function.
156 * 1 if we were able to detach the timer. 0 if we blew it, and the
157 * timer function has already started to run.
159 int scsi_delete_timer(struct scsi_cmnd *scmd)
163 rtn = del_timer(&scmd->eh_timeout);
165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166 " rtn: %d\n", __FUNCTION__,
169 scmd->eh_timeout.data = (unsigned long)NULL;
170 scmd->eh_timeout.function = NULL;
176 * scsi_times_out - Timeout function for normal scsi commands.
177 * @scmd: Cmd that is timing out.
180 * We do not need to lock this. There is the potential for a race
181 * only in that the normal completion handling might run, but if the
182 * normal completion function determines that the timer has already
183 * fired, then it mustn't do anything.
185 void scsi_times_out(struct scsi_cmnd *scmd)
187 enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
189 scsi_log_completion(scmd, TIMEOUT_ERROR);
191 if (scmd->device->host->transportt->eh_timed_out)
192 eh_timed_out = scmd->device->host->transportt->eh_timed_out;
193 else if (scmd->device->host->hostt->eh_timed_out)
194 eh_timed_out = scmd->device->host->hostt->eh_timed_out;
199 switch (eh_timed_out(scmd)) {
204 scsi_add_timer(scmd, scmd->timeout_per_command,
211 if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
212 scmd->result |= DID_TIME_OUT << 16;
218 * scsi_block_when_processing_errors - Prevent cmds from being queued.
219 * @sdev: Device on which we are performing recovery.
222 * We block until the host is out of error recovery, and then check to
223 * see whether the host or the device is offline.
226 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
228 int scsi_block_when_processing_errors(struct scsi_device *sdev)
232 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
234 online = scsi_device_online(sdev);
236 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__,
241 EXPORT_SYMBOL(scsi_block_when_processing_errors);
243 #ifdef CONFIG_SCSI_LOGGING
245 * scsi_eh_prt_fail_stats - Log info on failures.
246 * @shost: scsi host being recovered.
247 * @work_q: Queue of scsi cmds to process.
249 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
250 struct list_head *work_q)
252 struct scsi_cmnd *scmd;
253 struct scsi_device *sdev;
254 int total_failures = 0;
257 int devices_failed = 0;
259 shost_for_each_device(sdev, shost) {
260 list_for_each_entry(scmd, work_q, eh_entry) {
261 if (scmd->device == sdev) {
263 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
270 if (cmd_cancel || cmd_failed) {
271 SCSI_LOG_ERROR_RECOVERY(3,
272 sdev_printk(KERN_INFO, sdev,
273 "%s: cmds failed: %d, cancel: %d\n",
274 __FUNCTION__, cmd_failed,
282 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
283 " devices require eh work\n",
284 total_failures, devices_failed));
289 * scsi_check_sense - Examine scsi cmd sense
290 * @scmd: Cmd to have sense checked.
293 * SUCCESS or FAILED or NEEDS_RETRY
296 * When a deferred error is detected the current command has
297 * not been executed and needs retrying.
299 static int scsi_check_sense(struct scsi_cmnd *scmd)
301 struct scsi_device *sdev = scmd->device;
302 struct scsi_sense_hdr sshdr;
304 if (! scsi_command_normalize_sense(scmd, &sshdr))
305 return FAILED; /* no valid sense data */
307 if (scsi_sense_is_deferred(&sshdr))
310 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
311 sdev->scsi_dh_data->scsi_dh->check_sense) {
314 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
315 if (rc != SCSI_RETURN_NOT_HANDLED)
317 /* handler does not care. Drop down to default handling */
321 * Previous logic looked for FILEMARK, EOM or ILI which are
322 * mainly associated with tapes and returned SUCCESS.
324 if (sshdr.response_code == 0x70) {
326 if (scmd->sense_buffer[2] & 0xe0)
330 * descriptor format: look for "stream commands sense data
331 * descriptor" (see SSC-3). Assume single sense data
332 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
334 if ((sshdr.additional_length > 3) &&
335 (scmd->sense_buffer[8] == 0x4) &&
336 (scmd->sense_buffer[11] & 0xe0))
340 switch (sshdr.sense_key) {
343 case RECOVERED_ERROR:
344 return /* soft_error */ SUCCESS;
346 case ABORTED_COMMAND:
347 if (sshdr.asc == 0x10) /* DIF */
354 * if we are expecting a cc/ua because of a bus reset that we
355 * performed, treat this just as a retry. otherwise this is
356 * information that we should pass up to the upper-level driver
357 * so that we can deal with it there.
359 if (scmd->device->expecting_cc_ua) {
360 scmd->device->expecting_cc_ua = 0;
364 * if the device is in the process of becoming ready, we
367 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
370 * if the device is not started, we need to wake
371 * the error handler to start the motor
373 if (scmd->device->allow_restart &&
374 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
378 /* these three are not supported */
380 case VOLUME_OVERFLOW:
385 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
386 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
387 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
393 if (scmd->device->retry_hwerror)
398 case ILLEGAL_REQUEST:
407 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
408 * @scmd: SCSI cmd to examine.
411 * This is *only* called when we are examining the status of commands
412 * queued during error recovery. the main difference here is that we
413 * don't allow for the possibility of retries here, and we are a lot
414 * more restrictive about what we consider acceptable.
416 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
419 * first check the host byte, to see if there is anything in there
420 * that would indicate what we need to do.
422 if (host_byte(scmd->result) == DID_RESET) {
424 * rats. we are already in the error handler, so we now
425 * get to try and figure out what to do next. if the sense
426 * is valid, we have a pretty good idea of what to do.
427 * if not, we mark it as FAILED.
429 return scsi_check_sense(scmd);
431 if (host_byte(scmd->result) != DID_OK)
435 * next, check the message byte.
437 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
441 * now, check the status byte to see if this indicates
444 switch (status_byte(scmd->result)) {
446 case COMMAND_TERMINATED:
448 case CHECK_CONDITION:
449 return scsi_check_sense(scmd);
451 case INTERMEDIATE_GOOD:
452 case INTERMEDIATE_C_GOOD:
454 * who knows? FIXME(eric)
459 case RESERVATION_CONFLICT:
467 * scsi_eh_done - Completion function for error handling.
468 * @scmd: Cmd that is done.
470 static void scsi_eh_done(struct scsi_cmnd *scmd)
472 struct completion *eh_action;
474 SCSI_LOG_ERROR_RECOVERY(3,
475 printk("%s scmd: %p result: %x\n",
476 __FUNCTION__, scmd, scmd->result));
478 eh_action = scmd->device->host->eh_action;
484 * scsi_try_host_reset - ask host adapter to reset itself
485 * @scmd: SCSI cmd to send hsot reset.
487 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
492 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
495 if (!scmd->device->host->hostt->eh_host_reset_handler)
498 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
500 if (rtn == SUCCESS) {
501 if (!scmd->device->host->hostt->skip_settle_delay)
502 ssleep(HOST_RESET_SETTLE_TIME);
503 spin_lock_irqsave(scmd->device->host->host_lock, flags);
504 scsi_report_bus_reset(scmd->device->host,
506 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
513 * scsi_try_bus_reset - ask host to perform a bus reset
514 * @scmd: SCSI cmd to send bus reset.
516 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
521 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
524 if (!scmd->device->host->hostt->eh_bus_reset_handler)
527 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
529 if (rtn == SUCCESS) {
530 if (!scmd->device->host->hostt->skip_settle_delay)
531 ssleep(BUS_RESET_SETTLE_TIME);
532 spin_lock_irqsave(scmd->device->host->host_lock, flags);
533 scsi_report_bus_reset(scmd->device->host,
535 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
541 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
544 sdev->expecting_cc_ua = 1;
548 * scsi_try_target_reset - Ask host to perform a target reset
549 * @scmd: SCSI cmd used to send a target reset
552 * There is no timeout for this operation. if this operation is
553 * unreliable for a given host, then the host itself needs to put a
554 * timer on it, and set the host back to a consistent state prior to
557 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
562 if (!scmd->device->host->hostt->eh_target_reset_handler)
565 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
566 if (rtn == SUCCESS) {
567 spin_lock_irqsave(scmd->device->host->host_lock, flags);
568 __starget_for_each_device(scsi_target(scmd->device), NULL,
569 __scsi_report_device_reset);
570 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
577 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
578 * @scmd: SCSI cmd used to send BDR
581 * There is no timeout for this operation. if this operation is
582 * unreliable for a given host, then the host itself needs to put a
583 * timer on it, and set the host back to a consistent state prior to
586 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
590 if (!scmd->device->host->hostt->eh_device_reset_handler)
593 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
595 __scsi_report_device_reset(scmd->device, NULL);
599 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
601 if (!scmd->device->host->hostt->eh_abort_handler)
604 return scmd->device->host->hostt->eh_abort_handler(scmd);
608 * scsi_try_to_abort_cmd - Ask host to abort a running command.
609 * @scmd: SCSI cmd to abort from Lower Level.
612 * This function will not return until the user's completion function
613 * has been called. there is no timeout on this operation. if the
614 * author of the low-level driver wishes this operation to be timed,
615 * they can provide this facility themselves. helper functions in
616 * scsi_error.c can be supplied to make this easier to do.
618 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
621 * scsi_done was called just after the command timed out and before
622 * we had a chance to process it. (db)
624 if (scmd->serial_number == 0)
626 return __scsi_try_to_abort_cmd(scmd);
629 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
631 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
632 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
633 if (scsi_try_target_reset(scmd) != SUCCESS)
634 if (scsi_try_bus_reset(scmd) != SUCCESS)
635 scsi_try_host_reset(scmd);
639 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
640 * @scmd: SCSI command structure to hijack
641 * @ses: structure to save restore information
642 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
643 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
644 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
646 * This function is used to save a scsi command information before re-execution
647 * as part of the error recovery process. If @sense_bytes is 0 the command
648 * sent must be one that does not transfer any data. If @sense_bytes != 0
649 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
650 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
652 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
653 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
655 struct scsi_device *sdev = scmd->device;
658 * We need saved copies of a number of fields - this is because
659 * error handling may need to overwrite these with different values
660 * to run different commands, and once error handling is complete,
661 * we will need to restore these values prior to running the actual
664 ses->cmd_len = scmd->cmd_len;
665 ses->cmnd = scmd->cmnd;
666 ses->data_direction = scmd->sc_data_direction;
667 ses->sdb = scmd->sdb;
668 ses->next_rq = scmd->request->next_rq;
669 ses->result = scmd->result;
670 ses->prot_op = scmd->prot_op;
672 scmd->prot_op = SCSI_PROT_NORMAL;
673 scmd->cmnd = ses->eh_cmnd;
674 memset(scmd->cmnd, 0, BLK_MAX_CDB);
675 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
676 scmd->request->next_rq = NULL;
679 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
681 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
683 scmd->sdb.table.sgl = &ses->sense_sgl;
684 scmd->sc_data_direction = DMA_FROM_DEVICE;
685 scmd->sdb.table.nents = 1;
686 scmd->cmnd[0] = REQUEST_SENSE;
687 scmd->cmnd[4] = scmd->sdb.length;
688 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
690 scmd->sc_data_direction = DMA_NONE;
692 BUG_ON(cmnd_size > BLK_MAX_CDB);
693 memcpy(scmd->cmnd, cmnd, cmnd_size);
694 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
700 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
701 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
702 (sdev->lun << 5 & 0xe0);
705 * Zero the sense buffer. The scsi spec mandates that any
706 * untransferred sense data should be interpreted as being zero.
708 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
710 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
713 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
714 * @scmd: SCSI command structure to restore
715 * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd
717 * Undo any damage done by above scsi_prep_eh_cmnd().
719 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
722 * Restore original data
724 scmd->cmd_len = ses->cmd_len;
725 scmd->cmnd = ses->cmnd;
726 scmd->sc_data_direction = ses->data_direction;
727 scmd->sdb = ses->sdb;
728 scmd->request->next_rq = ses->next_rq;
729 scmd->result = ses->result;
730 scmd->prot_op = ses->prot_op;
732 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
735 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
736 * @scmd: SCSI command structure to hijack
738 * @cmnd_size: size in bytes of @cmnd
739 * @timeout: timeout for this request
740 * @sense_bytes: size of sense data to copy or 0
742 * This function is used to send a scsi command down to a target device
743 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
746 * SUCCESS or FAILED or NEEDS_RETRY
748 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
749 int cmnd_size, int timeout, unsigned sense_bytes)
751 struct scsi_device *sdev = scmd->device;
752 struct Scsi_Host *shost = sdev->host;
753 DECLARE_COMPLETION_ONSTACK(done);
754 unsigned long timeleft;
756 struct scsi_eh_save ses;
759 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
760 shost->eh_action = &done;
762 spin_lock_irqsave(shost->host_lock, flags);
764 shost->hostt->queuecommand(scmd, scsi_eh_done);
765 spin_unlock_irqrestore(shost->host_lock, flags);
767 timeleft = wait_for_completion_timeout(&done, timeout);
769 shost->eh_action = NULL;
771 scsi_log_completion(scmd, SUCCESS);
773 SCSI_LOG_ERROR_RECOVERY(3,
774 printk("%s: scmd: %p, timeleft: %ld\n",
775 __FUNCTION__, scmd, timeleft));
778 * If there is time left scsi_eh_done got called, and we will
779 * examine the actual status codes to see whether the command
780 * actually did complete normally, else tell the host to forget
781 * about this command.
784 rtn = scsi_eh_completed_normally(scmd);
785 SCSI_LOG_ERROR_RECOVERY(3,
786 printk("%s: scsi_eh_completed_normally %x\n",
799 scsi_abort_eh_cmnd(scmd);
803 scsi_eh_restore_cmnd(scmd, &ses);
808 * scsi_request_sense - Request sense data from a particular target.
809 * @scmd: SCSI cmd for request sense.
812 * Some hosts automatically obtain this information, others require
813 * that we obtain it on our own. This function will *not* return until
814 * the command either times out, or it completes.
816 static int scsi_request_sense(struct scsi_cmnd *scmd)
818 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
822 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
823 * @scmd: Original SCSI cmd that eh has finished.
824 * @done_q: Queue for processed commands.
827 * We don't want to use the normal command completion while we are are
828 * still handling errors - it may cause other commands to be queued,
829 * and that would disturb what we are doing. Thus we really want to
830 * keep a list of pending commands for final completion, and once we
831 * are ready to leave error handling we handle completion for real.
833 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
835 scmd->device->host->host_failed--;
837 list_move_tail(&scmd->eh_entry, done_q);
839 EXPORT_SYMBOL(scsi_eh_finish_cmd);
842 * scsi_eh_get_sense - Get device sense data.
843 * @work_q: Queue of commands to process.
844 * @done_q: Queue of processed commands.
847 * See if we need to request sense information. if so, then get it
848 * now, so we have a better idea of what to do.
851 * This has the unfortunate side effect that if a shost adapter does
852 * not automatically request sense information, we end up shutting
853 * it down before we request it.
855 * All drivers should request sense information internally these days,
856 * so for now all I have to say is tough noogies if you end up in here.
858 * XXX: Long term this code should go away, but that needs an audit of
861 int scsi_eh_get_sense(struct list_head *work_q,
862 struct list_head *done_q)
864 struct scsi_cmnd *scmd, *next;
867 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
868 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
869 SCSI_SENSE_VALID(scmd))
872 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
873 "%s: requesting sense\n",
875 rtn = scsi_request_sense(scmd);
879 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
880 " result %x\n", scmd,
882 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
884 rtn = scsi_decide_disposition(scmd);
887 * if the result was normal, then just pass it along to the
891 /* we don't want this command reissued, just
892 * finished with the sense data, so set
893 * retries to the max allowed to ensure it
894 * won't get reissued */
895 scmd->retries = scmd->allowed;
896 else if (rtn != NEEDS_RETRY)
899 scsi_eh_finish_cmd(scmd, done_q);
902 return list_empty(work_q);
904 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
907 * scsi_eh_tur - Send TUR to device.
908 * @scmd: &scsi_cmnd to send TUR
911 * 0 - Device is ready. 1 - Device NOT ready.
913 static int scsi_eh_tur(struct scsi_cmnd *scmd)
915 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
916 int retry_cnt = 1, rtn;
919 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
921 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
922 __FUNCTION__, scmd, rtn));
937 * scsi_eh_abort_cmds - abort pending commands.
938 * @work_q: &list_head for pending commands.
939 * @done_q: &list_head for processed commands.
942 * Try and see whether or not it makes sense to try and abort the
943 * running command. This only works out to be the case if we have one
944 * command that has timed out. If the command simply failed, it makes
945 * no sense to try and abort the command, since as far as the shost
946 * adapter is concerned, it isn't running.
948 static int scsi_eh_abort_cmds(struct list_head *work_q,
949 struct list_head *done_q)
951 struct scsi_cmnd *scmd, *next;
954 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
955 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
957 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
958 "0x%p\n", current->comm,
960 rtn = scsi_try_to_abort_cmd(scmd);
961 if (rtn == SUCCESS) {
962 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
963 if (!scsi_device_online(scmd->device) ||
964 !scsi_eh_tur(scmd)) {
965 scsi_eh_finish_cmd(scmd, done_q);
969 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
976 return list_empty(work_q);
980 * scsi_eh_try_stu - Send START_UNIT to device.
981 * @scmd: &scsi_cmnd to send START_UNIT
984 * 0 - Device is ready. 1 - Device NOT ready.
986 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
988 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
990 if (scmd->device->allow_restart) {
991 int i, rtn = NEEDS_RETRY;
993 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
994 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
995 scmd->device->timeout, 0);
1005 * scsi_eh_stu - send START_UNIT if needed
1006 * @shost: &scsi host being recovered.
1007 * @work_q: &list_head for pending commands.
1008 * @done_q: &list_head for processed commands.
1011 * If commands are failing due to not ready, initializing command required,
1012 * try revalidating the device, which will end up sending a start unit.
1014 static int scsi_eh_stu(struct Scsi_Host *shost,
1015 struct list_head *work_q,
1016 struct list_head *done_q)
1018 struct scsi_cmnd *scmd, *stu_scmd, *next;
1019 struct scsi_device *sdev;
1021 shost_for_each_device(sdev, shost) {
1023 list_for_each_entry(scmd, work_q, eh_entry)
1024 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1025 scsi_check_sense(scmd) == FAILED ) {
1033 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1034 " 0x%p\n", current->comm, sdev));
1036 if (!scsi_eh_try_stu(stu_scmd)) {
1037 if (!scsi_device_online(sdev) ||
1038 !scsi_eh_tur(stu_scmd)) {
1039 list_for_each_entry_safe(scmd, next,
1041 if (scmd->device == sdev)
1042 scsi_eh_finish_cmd(scmd, done_q);
1046 SCSI_LOG_ERROR_RECOVERY(3,
1047 printk("%s: START_UNIT failed to sdev:"
1048 " 0x%p\n", current->comm, sdev));
1052 return list_empty(work_q);
1057 * scsi_eh_bus_device_reset - send bdr if needed
1058 * @shost: scsi host being recovered.
1059 * @work_q: &list_head for pending commands.
1060 * @done_q: &list_head for processed commands.
1063 * Try a bus device reset. Still, look to see whether we have multiple
1064 * devices that are jammed or not - if we have multiple devices, it
1065 * makes no sense to try bus_device_reset - we really would need to try
1066 * a bus_reset instead.
1068 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1069 struct list_head *work_q,
1070 struct list_head *done_q)
1072 struct scsi_cmnd *scmd, *bdr_scmd, *next;
1073 struct scsi_device *sdev;
1076 shost_for_each_device(sdev, shost) {
1078 list_for_each_entry(scmd, work_q, eh_entry)
1079 if (scmd->device == sdev) {
1087 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1088 " 0x%p\n", current->comm,
1090 rtn = scsi_try_bus_device_reset(bdr_scmd);
1091 if (rtn == SUCCESS) {
1092 if (!scsi_device_online(sdev) ||
1093 !scsi_eh_tur(bdr_scmd)) {
1094 list_for_each_entry_safe(scmd, next,
1096 if (scmd->device == sdev)
1097 scsi_eh_finish_cmd(scmd,
1102 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1110 return list_empty(work_q);
1114 * scsi_eh_target_reset - send target reset if needed
1115 * @shost: scsi host being recovered.
1116 * @work_q: &list_head for pending commands.
1117 * @done_q: &list_head for processed commands.
1120 * Try a target reset.
1122 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1123 struct list_head *work_q,
1124 struct list_head *done_q)
1126 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1130 for (id = 0; id <= shost->max_id; id++) {
1132 list_for_each_entry(scmd, work_q, eh_entry) {
1133 if (id == scmd_id(scmd)) {
1141 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1143 current->comm, id));
1144 rtn = scsi_try_target_reset(tgtr_scmd);
1145 if (rtn == SUCCESS) {
1146 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1147 if (id == scmd_id(scmd))
1148 if (!scsi_device_online(scmd->device) ||
1149 !scsi_eh_tur(tgtr_scmd))
1150 scsi_eh_finish_cmd(scmd,
1154 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1157 current->comm, id));
1160 return list_empty(work_q);
1164 * scsi_eh_bus_reset - send a bus reset
1165 * @shost: &scsi host being recovered.
1166 * @work_q: &list_head for pending commands.
1167 * @done_q: &list_head for processed commands.
1169 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1170 struct list_head *work_q,
1171 struct list_head *done_q)
1173 struct scsi_cmnd *scmd, *chan_scmd, *next;
1174 unsigned int channel;
1178 * we really want to loop over the various channels, and do this on
1179 * a channel by channel basis. we should also check to see if any
1180 * of the failed commands are on soft_reset devices, and if so, skip
1184 for (channel = 0; channel <= shost->max_channel; channel++) {
1186 list_for_each_entry(scmd, work_q, eh_entry) {
1187 if (channel == scmd_channel(scmd)) {
1191 * FIXME add back in some support for
1192 * soft_reset devices.
1199 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1200 " %d\n", current->comm,
1202 rtn = scsi_try_bus_reset(chan_scmd);
1203 if (rtn == SUCCESS) {
1204 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1205 if (channel == scmd_channel(scmd))
1206 if (!scsi_device_online(scmd->device) ||
1208 scsi_eh_finish_cmd(scmd,
1212 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1213 " failed chan: %d\n",
1218 return list_empty(work_q);
1222 * scsi_eh_host_reset - send a host reset
1223 * @work_q: list_head for processed commands.
1224 * @done_q: list_head for processed commands.
1226 static int scsi_eh_host_reset(struct list_head *work_q,
1227 struct list_head *done_q)
1229 struct scsi_cmnd *scmd, *next;
1232 if (!list_empty(work_q)) {
1233 scmd = list_entry(work_q->next,
1234 struct scsi_cmnd, eh_entry);
1236 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1239 rtn = scsi_try_host_reset(scmd);
1240 if (rtn == SUCCESS) {
1241 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1242 if (!scsi_device_online(scmd->device) ||
1243 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1245 scsi_eh_finish_cmd(scmd, done_q);
1248 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1253 return list_empty(work_q);
1257 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1258 * @work_q: list_head for processed commands.
1259 * @done_q: list_head for processed commands.
1261 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1262 struct list_head *done_q)
1264 struct scsi_cmnd *scmd, *next;
1266 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1267 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1268 "not ready after error recovery\n");
1269 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1270 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1272 * FIXME: Handle lost cmds.
1275 scsi_eh_finish_cmd(scmd, done_q);
1281 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1282 * @scmd: SCSI cmd to examine.
1285 * This is *only* called when we are examining the status after sending
1286 * out the actual data command. any commands that are queued for error
1287 * recovery (e.g. test_unit_ready) do *not* come through here.
1289 * When this routine returns failed, it means the error handler thread
1290 * is woken. In cases where the error code indicates an error that
1291 * doesn't require the error handler read (i.e. we don't need to
1292 * abort/reset), this function should return SUCCESS.
1294 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1299 * if the device is offline, then we clearly just pass the result back
1300 * up to the top level.
1302 if (!scsi_device_online(scmd->device)) {
1303 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1310 * first check the host byte, to see if there is anything in there
1311 * that would indicate what we need to do.
1313 switch (host_byte(scmd->result)) {
1314 case DID_PASSTHROUGH:
1316 * no matter what, pass this through to the upper layer.
1317 * nuke this special code so that it looks like we are saying
1320 scmd->result &= 0xff00ffff;
1324 * looks good. drop through, and check the next byte.
1327 case DID_NO_CONNECT:
1328 case DID_BAD_TARGET:
1331 * note - this means that we just report the status back
1332 * to the top level driver, not that we actually think
1333 * that it indicates SUCCESS.
1337 * when the low level driver returns did_soft_error,
1338 * it is responsible for keeping an internal retry counter
1339 * in order to avoid endless loops (db)
1341 * actually this is a bug in this function here. we should
1342 * be mindful of the maximum number of retries specified
1343 * and not get stuck in a loop.
1345 case DID_SOFT_ERROR:
1351 return ADD_TO_MLQUEUE;
1354 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1355 status_byte(scmd->result) == RESERVATION_CONFLICT)
1357 * execute reservation conflict processing code
1368 * when we scan the bus, we get timeout messages for
1369 * these commands if there is no device available.
1370 * other hosts report did_no_connect for the same thing.
1372 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1373 scmd->cmnd[0] == INQUIRY)) {
1385 * next, check the message byte.
1387 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1391 * check the status byte to see if this indicates anything special.
1393 switch (status_byte(scmd->result)) {
1396 * the case of trying to send too many commands to a
1397 * tagged queueing device.
1401 * device can't talk to us at the moment. Should only
1402 * occur (SAM-3) when the task queue is empty, so will cause
1403 * the empty queue handling to trigger a stall in the
1406 return ADD_TO_MLQUEUE;
1408 case COMMAND_TERMINATED:
1411 case CHECK_CONDITION:
1412 rtn = scsi_check_sense(scmd);
1413 if (rtn == NEEDS_RETRY)
1415 /* if rtn == FAILED, we have no sense information;
1416 * returning FAILED will wake the error handler thread
1417 * to collect the sense and redo the decide
1420 case CONDITION_GOOD:
1421 case INTERMEDIATE_GOOD:
1422 case INTERMEDIATE_C_GOOD:
1425 * who knows? FIXME(eric)
1429 case RESERVATION_CONFLICT:
1430 sdev_printk(KERN_INFO, scmd->device,
1431 "reservation conflict\n");
1432 return SUCCESS; /* causes immediate i/o error */
1440 /* we requeue for retry because the error was retryable, and
1441 * the request was not marked fast fail. Note that above,
1442 * even if the request is marked fast fail, we still requeue
1443 * for queue congestion conditions (QUEUE_FULL or BUSY) */
1444 if ((++scmd->retries) <= scmd->allowed
1445 && !blk_noretry_request(scmd->request)) {
1449 * no more retries - report this one back to upper level.
1456 * scsi_eh_lock_door - Prevent medium removal for the specified device
1457 * @sdev: SCSI device to prevent medium removal
1460 * We must be called from process context; scsi_allocate_request()
1464 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1465 * head of the devices request queue, and continue.
1468 * scsi_allocate_request() may sleep waiting for existing requests to
1469 * be processed. However, since we haven't kicked off any request
1470 * processing for this host, this may deadlock.
1472 * If scsi_allocate_request() fails for what ever reason, we
1473 * completely forget to lock the door.
1475 static void scsi_eh_lock_door(struct scsi_device *sdev)
1477 unsigned char cmnd[MAX_COMMAND_SIZE];
1479 cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1483 cmnd[4] = SCSI_REMOVAL_PREVENT;
1486 scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
1487 5, NULL, NULL, GFP_KERNEL);
1492 * scsi_restart_operations - restart io operations to the specified host.
1493 * @shost: Host we are restarting.
1496 * When we entered the error handler, we blocked all further i/o to
1497 * this device. we need to 'reverse' this process.
1499 static void scsi_restart_operations(struct Scsi_Host *shost)
1501 struct scsi_device *sdev;
1502 unsigned long flags;
1505 * If the door was locked, we need to insert a door lock request
1506 * onto the head of the SCSI request queue for the device. There
1507 * is no point trying to lock the door of an off-line device.
1509 shost_for_each_device(sdev, shost) {
1510 if (scsi_device_online(sdev) && sdev->locked)
1511 scsi_eh_lock_door(sdev);
1515 * next free up anything directly waiting upon the host. this
1516 * will be requests for character device operations, and also for
1517 * ioctls to queued block devices.
1519 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1522 spin_lock_irqsave(shost->host_lock, flags);
1523 if (scsi_host_set_state(shost, SHOST_RUNNING))
1524 if (scsi_host_set_state(shost, SHOST_CANCEL))
1525 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1526 spin_unlock_irqrestore(shost->host_lock, flags);
1528 wake_up(&shost->host_wait);
1531 * finally we need to re-initiate requests that may be pending. we will
1532 * have had everything blocked while error handling is taking place, and
1533 * now that error recovery is done, we will need to ensure that these
1534 * requests are started.
1536 scsi_run_host_queues(shost);
1540 * scsi_eh_ready_devs - check device ready state and recover if not.
1541 * @shost: host to be recovered.
1542 * @work_q: &list_head for pending commands.
1543 * @done_q: &list_head for processed commands.
1545 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1546 struct list_head *work_q,
1547 struct list_head *done_q)
1549 if (!scsi_eh_stu(shost, work_q, done_q))
1550 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1551 if (!scsi_eh_target_reset(shost, work_q, done_q))
1552 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1553 if (!scsi_eh_host_reset(work_q, done_q))
1554 scsi_eh_offline_sdevs(work_q,
1557 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1560 * scsi_eh_flush_done_q - finish processed commands or retry them.
1561 * @done_q: list_head of processed commands.
1563 void scsi_eh_flush_done_q(struct list_head *done_q)
1565 struct scsi_cmnd *scmd, *next;
1567 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1568 list_del_init(&scmd->eh_entry);
1569 if (scsi_device_online(scmd->device) &&
1570 !blk_noretry_request(scmd->request) &&
1571 (++scmd->retries <= scmd->allowed)) {
1572 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1576 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1579 * If just we got sense for the device (called
1580 * scsi_eh_get_sense), scmd->result is already
1581 * set, do not set DRIVER_TIMEOUT.
1584 scmd->result |= (DRIVER_TIMEOUT << 24);
1585 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1587 current->comm, scmd));
1588 scsi_finish_command(scmd);
1592 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1595 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1596 * @shost: Host to unjam.
1599 * When we come in here, we *know* that all commands on the bus have
1600 * either completed, failed or timed out. we also know that no further
1601 * commands are being sent to the host, so things are relatively quiet
1602 * and we have freedom to fiddle with things as we wish.
1604 * This is only the *default* implementation. it is possible for
1605 * individual drivers to supply their own version of this function, and
1606 * if the maintainer wishes to do this, it is strongly suggested that
1607 * this function be taken as a template and modified. this function
1608 * was designed to correctly handle problems for about 95% of the
1609 * different cases out there, and it should always provide at least a
1610 * reasonable amount of error recovery.
1612 * Any command marked 'failed' or 'timeout' must eventually have
1613 * scsi_finish_cmd() called for it. we do all of the retry stuff
1614 * here, so when we restart the host after we return it should have an
1617 static void scsi_unjam_host(struct Scsi_Host *shost)
1619 unsigned long flags;
1620 LIST_HEAD(eh_work_q);
1621 LIST_HEAD(eh_done_q);
1623 spin_lock_irqsave(shost->host_lock, flags);
1624 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1625 spin_unlock_irqrestore(shost->host_lock, flags);
1627 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1629 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1630 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1631 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1633 scsi_eh_flush_done_q(&eh_done_q);
1637 * scsi_error_handler - SCSI error handler thread
1638 * @data: Host for which we are running.
1641 * This is the main error handling loop. This is run as a kernel thread
1642 * for every SCSI host and handles all error handling activity.
1644 int scsi_error_handler(void *data)
1646 struct Scsi_Host *shost = data;
1649 * We use TASK_INTERRUPTIBLE so that the thread is not
1650 * counted against the load average as a running process.
1651 * We never actually get interrupted because kthread_run
1652 * disables singal delivery for the created thread.
1654 set_current_state(TASK_INTERRUPTIBLE);
1655 while (!kthread_should_stop()) {
1656 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1657 shost->host_failed != shost->host_busy) {
1658 SCSI_LOG_ERROR_RECOVERY(1,
1659 printk("Error handler scsi_eh_%d sleeping\n",
1662 set_current_state(TASK_INTERRUPTIBLE);
1666 __set_current_state(TASK_RUNNING);
1667 SCSI_LOG_ERROR_RECOVERY(1,
1668 printk("Error handler scsi_eh_%d waking up\n",
1672 * We have a host that is failing for some reason. Figure out
1673 * what we need to do to get it up and online again (if we can).
1674 * If we fail, we end up taking the thing offline.
1676 if (shost->transportt->eh_strategy_handler)
1677 shost->transportt->eh_strategy_handler(shost);
1679 scsi_unjam_host(shost);
1682 * Note - if the above fails completely, the action is to take
1683 * individual devices offline and flush the queue of any
1684 * outstanding requests that may have been pending. When we
1685 * restart, we restart any I/O to any other devices on the bus
1686 * which are still online.
1688 scsi_restart_operations(shost);
1689 set_current_state(TASK_INTERRUPTIBLE);
1691 __set_current_state(TASK_RUNNING);
1693 SCSI_LOG_ERROR_RECOVERY(1,
1694 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1695 shost->ehandler = NULL;
1700 * Function: scsi_report_bus_reset()
1702 * Purpose: Utility function used by low-level drivers to report that
1703 * they have observed a bus reset on the bus being handled.
1705 * Arguments: shost - Host in question
1706 * channel - channel on which reset was observed.
1710 * Lock status: Host lock must be held.
1712 * Notes: This only needs to be called if the reset is one which
1713 * originates from an unknown location. Resets originated
1714 * by the mid-level itself don't need to call this, but there
1715 * should be no harm.
1717 * The main purpose of this is to make sure that a CHECK_CONDITION
1718 * is properly treated.
1720 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1722 struct scsi_device *sdev;
1724 __shost_for_each_device(sdev, shost) {
1725 if (channel == sdev_channel(sdev))
1726 __scsi_report_device_reset(sdev, NULL);
1729 EXPORT_SYMBOL(scsi_report_bus_reset);
1732 * Function: scsi_report_device_reset()
1734 * Purpose: Utility function used by low-level drivers to report that
1735 * they have observed a device reset on the device being handled.
1737 * Arguments: shost - Host in question
1738 * channel - channel on which reset was observed
1739 * target - target on which reset was observed
1743 * Lock status: Host lock must be held
1745 * Notes: This only needs to be called if the reset is one which
1746 * originates from an unknown location. Resets originated
1747 * by the mid-level itself don't need to call this, but there
1748 * should be no harm.
1750 * The main purpose of this is to make sure that a CHECK_CONDITION
1751 * is properly treated.
1753 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1755 struct scsi_device *sdev;
1757 __shost_for_each_device(sdev, shost) {
1758 if (channel == sdev_channel(sdev) &&
1759 target == sdev_id(sdev))
1760 __scsi_report_device_reset(sdev, NULL);
1763 EXPORT_SYMBOL(scsi_report_device_reset);
1766 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1771 * Function: scsi_reset_provider
1773 * Purpose: Send requested reset to a bus or device at any phase.
1775 * Arguments: device - device to send reset to
1776 * flag - reset type (see scsi.h)
1778 * Returns: SUCCESS/FAILURE.
1780 * Notes: This is used by the SCSI Generic driver to provide
1781 * Bus/Device reset capability.
1784 scsi_reset_provider(struct scsi_device *dev, int flag)
1786 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1787 struct Scsi_Host *shost = dev->host;
1789 unsigned long flags;
1792 blk_rq_init(NULL, &req);
1793 scmd->request = &req;
1794 memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1796 scmd->cmnd = req.cmd;
1798 scmd->scsi_done = scsi_reset_provider_done_command;
1799 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1803 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
1805 init_timer(&scmd->eh_timeout);
1807 spin_lock_irqsave(shost->host_lock, flags);
1808 shost->tmf_in_progress = 1;
1809 spin_unlock_irqrestore(shost->host_lock, flags);
1812 case SCSI_TRY_RESET_DEVICE:
1813 rtn = scsi_try_bus_device_reset(scmd);
1817 case SCSI_TRY_RESET_TARGET:
1818 rtn = scsi_try_target_reset(scmd);
1822 case SCSI_TRY_RESET_BUS:
1823 rtn = scsi_try_bus_reset(scmd);
1827 case SCSI_TRY_RESET_HOST:
1828 rtn = scsi_try_host_reset(scmd);
1834 spin_lock_irqsave(shost->host_lock, flags);
1835 shost->tmf_in_progress = 0;
1836 spin_unlock_irqrestore(shost->host_lock, flags);
1839 * be sure to wake up anyone who was sleeping or had their queue
1840 * suspended while we performed the TMF.
1842 SCSI_LOG_ERROR_RECOVERY(3,
1843 printk("%s: waking up host to restart after TMF\n",
1846 wake_up(&shost->host_wait);
1848 scsi_run_host_queues(shost);
1850 scsi_next_command(scmd);
1853 EXPORT_SYMBOL(scsi_reset_provider);
1856 * scsi_normalize_sense - normalize main elements from either fixed or
1857 * descriptor sense data format into a common format.
1859 * @sense_buffer: byte array containing sense data returned by device
1860 * @sb_len: number of valid bytes in sense_buffer
1861 * @sshdr: pointer to instance of structure that common
1862 * elements are written to.
1865 * The "main elements" from sense data are: response_code, sense_key,
1866 * asc, ascq and additional_length (only for descriptor format).
1868 * Typically this function can be called after a device has
1869 * responded to a SCSI command with the CHECK_CONDITION status.
1872 * 1 if valid sense data information found, else 0;
1874 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1875 struct scsi_sense_hdr *sshdr)
1877 if (!sense_buffer || !sb_len)
1880 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1882 sshdr->response_code = (sense_buffer[0] & 0x7f);
1884 if (!scsi_sense_valid(sshdr))
1887 if (sshdr->response_code >= 0x72) {
1892 sshdr->sense_key = (sense_buffer[1] & 0xf);
1894 sshdr->asc = sense_buffer[2];
1896 sshdr->ascq = sense_buffer[3];
1898 sshdr->additional_length = sense_buffer[7];
1904 sshdr->sense_key = (sense_buffer[2] & 0xf);
1906 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1907 sb_len : (sense_buffer[7] + 8);
1909 sshdr->asc = sense_buffer[12];
1911 sshdr->ascq = sense_buffer[13];
1917 EXPORT_SYMBOL(scsi_normalize_sense);
1919 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1920 struct scsi_sense_hdr *sshdr)
1922 return scsi_normalize_sense(cmd->sense_buffer,
1923 SCSI_SENSE_BUFFERSIZE, sshdr);
1925 EXPORT_SYMBOL(scsi_command_normalize_sense);
1928 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1929 * @sense_buffer: byte array of descriptor format sense data
1930 * @sb_len: number of valid bytes in sense_buffer
1931 * @desc_type: value of descriptor type to find
1932 * (e.g. 0 -> information)
1935 * only valid when sense data is in descriptor format
1938 * pointer to start of (first) descriptor if found else NULL
1940 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1943 int add_sen_len, add_len, desc_len, k;
1946 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1948 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1950 add_sen_len = (add_sen_len < (sb_len - 8)) ?
1951 add_sen_len : (sb_len - 8);
1952 descp = &sense_buffer[8];
1953 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1955 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1956 desc_len = add_len + 2;
1957 if (descp[0] == desc_type)
1959 if (add_len < 0) // short descriptor ??
1964 EXPORT_SYMBOL(scsi_sense_desc_find);
1967 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1968 * @sense_buffer: byte array of sense data
1969 * @sb_len: number of valid bytes in sense_buffer
1970 * @info_out: pointer to 64 integer where 8 or 4 byte information
1971 * field will be placed if found.
1974 * 1 if information field found, 0 if not found.
1976 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1985 switch (sense_buffer[0] & 0x7f) {
1988 if (sense_buffer[0] & 0x80) {
1989 *info_out = (sense_buffer[3] << 24) +
1990 (sense_buffer[4] << 16) +
1991 (sense_buffer[5] << 8) + sense_buffer[6];
1997 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1999 if (ucp && (0xa == ucp[1])) {
2001 for (j = 0; j < 8; ++j) {
2014 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2017 * scsi_build_sense_buffer - build sense data in a buffer
2018 * @desc: Sense format (non zero == descriptor format,
2019 * 0 == fixed format)
2020 * @buf: Where to build sense data
2022 * @asc: Additional sense code
2023 * @ascq: Additional sense code qualifier
2026 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2029 buf[0] = 0x72; /* descriptor, current */
2035 buf[0] = 0x70; /* fixed, current */
2042 EXPORT_SYMBOL(scsi_build_sense_buffer);