2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_eh.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_cmnd.h>
41 #include "../scsi/scsi_transport_api.h"
43 #include <linux/libata.h>
48 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
49 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
50 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
53 /* Waiting in ->prereset can never be reliable. It's sometimes nice
54 * to wait there but it can't be depended upon; otherwise, we wouldn't
55 * be resetting. Just give it enough time for most drives to spin up.
58 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
59 ATA_EH_FASTDRAIN_INTERVAL = 3 * HZ,
62 /* The following table determines how we sequence resets. Each entry
63 * represents timeout for that try. The first try can be soft or
64 * hardreset. All others are hardreset if available. In most cases
65 * the first reset w/ 10sec timeout should succeed. Following entries
66 * are mostly for error handling, hotplug and retarded devices.
68 static const unsigned long ata_eh_reset_timeouts[] = {
69 10 * HZ, /* most drives spin up by 10sec */
70 10 * HZ, /* > 99% working drives spin up before 20sec */
71 35 * HZ, /* give > 30 secs of idleness for retarded devices */
72 5 * HZ, /* and sweet one last chance */
73 /* > 1 min has elapsed, give up */
76 static void __ata_port_freeze(struct ata_port *ap);
77 static void ata_eh_finish(struct ata_port *ap);
79 static void ata_eh_handle_port_suspend(struct ata_port *ap);
80 static void ata_eh_handle_port_resume(struct ata_port *ap);
82 static void ata_eh_handle_port_suspend(struct ata_port *ap)
85 static void ata_eh_handle_port_resume(struct ata_port *ap)
87 #endif /* CONFIG_PM */
89 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
92 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
93 ATA_EH_DESC_LEN - ehi->desc_len,
98 * __ata_ehi_push_desc - push error description without adding separator
100 * @fmt: printf format string
102 * Format string according to @fmt and append it to @ehi->desc.
105 * spin_lock_irqsave(host lock)
107 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
112 __ata_ehi_pushv_desc(ehi, fmt, args);
117 * ata_ehi_push_desc - push error description with separator
119 * @fmt: printf format string
121 * Format string according to @fmt and append it to @ehi->desc.
122 * If @ehi->desc is not empty, ", " is added in-between.
125 * spin_lock_irqsave(host lock)
127 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
132 __ata_ehi_push_desc(ehi, ", ");
135 __ata_ehi_pushv_desc(ehi, fmt, args);
140 * ata_ehi_clear_desc - clean error description
146 * spin_lock_irqsave(host lock)
148 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
154 static void ata_ering_record(struct ata_ering *ering, int is_io,
155 unsigned int err_mask)
157 struct ata_ering_entry *ent;
162 ering->cursor %= ATA_ERING_SIZE;
164 ent = &ering->ring[ering->cursor];
166 ent->err_mask = err_mask;
167 ent->timestamp = get_jiffies_64();
170 static void ata_ering_clear(struct ata_ering *ering)
172 memset(ering, 0, sizeof(*ering));
175 static int ata_ering_map(struct ata_ering *ering,
176 int (*map_fn)(struct ata_ering_entry *, void *),
180 struct ata_ering_entry *ent;
184 ent = &ering->ring[idx];
187 rc = map_fn(ent, arg);
190 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
191 } while (idx != ering->cursor);
196 static unsigned int ata_eh_dev_action(struct ata_device *dev)
198 struct ata_eh_context *ehc = &dev->link->eh_context;
200 return ehc->i.action | ehc->i.dev_action[dev->devno];
203 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
204 struct ata_eh_info *ehi, unsigned int action)
206 struct ata_device *tdev;
209 ehi->action &= ~action;
210 ata_link_for_each_dev(tdev, link)
211 ehi->dev_action[tdev->devno] &= ~action;
213 /* doesn't make sense for port-wide EH actions */
214 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
216 /* break ehi->action into ehi->dev_action */
217 if (ehi->action & action) {
218 ata_link_for_each_dev(tdev, link)
219 ehi->dev_action[tdev->devno] |=
220 ehi->action & action;
221 ehi->action &= ~action;
224 /* turn off the specified per-dev action */
225 ehi->dev_action[dev->devno] &= ~action;
230 * ata_scsi_timed_out - SCSI layer time out callback
231 * @cmd: timed out SCSI command
233 * Handles SCSI layer timeout. We race with normal completion of
234 * the qc for @cmd. If the qc is already gone, we lose and let
235 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
236 * timed out and EH should be invoked. Prevent ata_qc_complete()
237 * from finishing it by setting EH_SCHEDULED and return
240 * TODO: kill this function once old EH is gone.
243 * Called from timer context
246 * EH_HANDLED or EH_NOT_HANDLED
248 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
250 struct Scsi_Host *host = cmd->device->host;
251 struct ata_port *ap = ata_shost_to_port(host);
253 struct ata_queued_cmd *qc;
254 enum scsi_eh_timer_return ret;
258 if (ap->ops->error_handler) {
259 ret = EH_NOT_HANDLED;
264 spin_lock_irqsave(ap->lock, flags);
265 qc = ata_qc_from_tag(ap, ap->link.active_tag);
267 WARN_ON(qc->scsicmd != cmd);
268 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
269 qc->err_mask |= AC_ERR_TIMEOUT;
270 ret = EH_NOT_HANDLED;
272 spin_unlock_irqrestore(ap->lock, flags);
275 DPRINTK("EXIT, ret=%d\n", ret);
280 * ata_scsi_error - SCSI layer error handler callback
281 * @host: SCSI host on which error occurred
283 * Handles SCSI-layer-thrown error events.
286 * Inherited from SCSI layer (none, can sleep)
291 void ata_scsi_error(struct Scsi_Host *host)
293 struct ata_port *ap = ata_shost_to_port(host);
294 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
299 /* synchronize with port task */
300 ata_port_flush_task(ap);
302 /* synchronize with host lock and sort out timeouts */
304 /* For new EH, all qcs are finished in one of three ways -
305 * normal completion, error completion, and SCSI timeout.
306 * Both cmpletions can race against SCSI timeout. When normal
307 * completion wins, the qc never reaches EH. When error
308 * completion wins, the qc has ATA_QCFLAG_FAILED set.
310 * When SCSI timeout wins, things are a bit more complex.
311 * Normal or error completion can occur after the timeout but
312 * before this point. In such cases, both types of
313 * completions are honored. A scmd is determined to have
314 * timed out iff its associated qc is active and not failed.
316 if (ap->ops->error_handler) {
317 struct scsi_cmnd *scmd, *tmp;
320 spin_lock_irqsave(ap->lock, flags);
322 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
323 struct ata_queued_cmd *qc;
325 for (i = 0; i < ATA_MAX_QUEUE; i++) {
326 qc = __ata_qc_from_tag(ap, i);
327 if (qc->flags & ATA_QCFLAG_ACTIVE &&
332 if (i < ATA_MAX_QUEUE) {
333 /* the scmd has an associated qc */
334 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
335 /* which hasn't failed yet, timeout */
336 qc->err_mask |= AC_ERR_TIMEOUT;
337 qc->flags |= ATA_QCFLAG_FAILED;
341 /* Normal completion occurred after
342 * SCSI timeout but before this point.
343 * Successfully complete it.
345 scmd->retries = scmd->allowed;
346 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
350 /* If we have timed out qcs. They belong to EH from
351 * this point but the state of the controller is
352 * unknown. Freeze the port to make sure the IRQ
353 * handler doesn't diddle with those qcs. This must
354 * be done atomically w.r.t. setting QCFLAG_FAILED.
357 __ata_port_freeze(ap);
359 spin_unlock_irqrestore(ap->lock, flags);
361 spin_unlock_wait(ap->lock);
364 /* invoke error handler */
365 if (ap->ops->error_handler) {
366 /* kill fast drain timer */
367 del_timer_sync(&ap->fastdrain_timer);
369 /* process port resume request */
370 ata_eh_handle_port_resume(ap);
372 /* fetch & clear EH info */
373 spin_lock_irqsave(ap->lock, flags);
375 memset(&ap->link.eh_context, 0, sizeof(ap->link.eh_context));
376 ap->link.eh_context.i = ap->link.eh_info;
377 memset(&ap->link.eh_info, 0, sizeof(ap->link.eh_info));
379 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
380 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
382 spin_unlock_irqrestore(ap->lock, flags);
384 /* invoke EH, skip if unloading or suspended */
385 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
386 ap->ops->error_handler(ap);
390 /* process port suspend request */
391 ata_eh_handle_port_suspend(ap);
393 /* Exception might have happend after ->error_handler
394 * recovered the port but before this point. Repeat
397 spin_lock_irqsave(ap->lock, flags);
399 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
401 ata_port_printk(ap, KERN_INFO,
402 "EH pending after completion, "
403 "repeating EH (cnt=%d)\n", repeat_cnt);
404 spin_unlock_irqrestore(ap->lock, flags);
407 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
408 "tries, giving up\n", ATA_EH_MAX_REPEAT);
409 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
412 /* this run is complete, make sure EH info is clear */
413 memset(&ap->link.eh_info, 0, sizeof(ap->link.eh_info));
415 /* Clear host_eh_scheduled while holding ap->lock such
416 * that if exception occurs after this point but
417 * before EH completion, SCSI midlayer will
420 host->host_eh_scheduled = 0;
422 spin_unlock_irqrestore(ap->lock, flags);
424 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
425 ap->ops->eng_timeout(ap);
428 /* finish or retry handled scmd's and clean up */
429 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
431 scsi_eh_flush_done_q(&ap->eh_done_q);
434 spin_lock_irqsave(ap->lock, flags);
436 if (ap->pflags & ATA_PFLAG_LOADING)
437 ap->pflags &= ~ATA_PFLAG_LOADING;
438 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
439 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
441 if (ap->pflags & ATA_PFLAG_RECOVERED)
442 ata_port_printk(ap, KERN_INFO, "EH complete\n");
444 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
446 /* tell wait_eh that we're done */
447 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
448 wake_up_all(&ap->eh_wait_q);
450 spin_unlock_irqrestore(ap->lock, flags);
456 * ata_port_wait_eh - Wait for the currently pending EH to complete
457 * @ap: Port to wait EH for
459 * Wait until the currently pending EH is complete.
462 * Kernel thread context (may sleep).
464 void ata_port_wait_eh(struct ata_port *ap)
470 spin_lock_irqsave(ap->lock, flags);
472 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
473 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
474 spin_unlock_irqrestore(ap->lock, flags);
476 spin_lock_irqsave(ap->lock, flags);
478 finish_wait(&ap->eh_wait_q, &wait);
480 spin_unlock_irqrestore(ap->lock, flags);
482 /* make sure SCSI EH is complete */
483 if (scsi_host_in_recovery(ap->scsi_host)) {
490 * ata_qc_timeout - Handle timeout of queued command
491 * @qc: Command that timed out
493 * Some part of the kernel (currently, only the SCSI layer)
494 * has noticed that the active command on port @ap has not
495 * completed after a specified length of time. Handle this
496 * condition by disabling DMA (if necessary) and completing
497 * transactions, with error if necessary.
499 * This also handles the case of the "lost interrupt", where
500 * for some reason (possibly hardware bug, possibly driver bug)
501 * an interrupt was not delivered to the driver, even though the
502 * transaction completed successfully.
504 * TODO: kill this function once old EH is gone.
507 * Inherited from SCSI layer (none, can sleep)
509 static void ata_qc_timeout(struct ata_queued_cmd *qc)
511 struct ata_port *ap = qc->ap;
512 u8 host_stat = 0, drv_stat;
517 ap->hsm_task_state = HSM_ST_IDLE;
519 spin_lock_irqsave(ap->lock, flags);
521 switch (qc->tf.protocol) {
524 case ATA_PROT_ATAPI_DMA:
525 host_stat = ap->ops->bmdma_status(ap);
527 /* before we do anything else, clear DMA-Start bit */
528 ap->ops->bmdma_stop(qc);
534 drv_stat = ata_chk_status(ap);
536 /* ack bmdma irq events */
537 ap->ops->irq_clear(ap);
539 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
540 "stat 0x%x host_stat 0x%x\n",
541 qc->tf.command, drv_stat, host_stat);
543 /* complete taskfile transaction */
544 qc->err_mask |= AC_ERR_TIMEOUT;
548 spin_unlock_irqrestore(ap->lock, flags);
550 ata_eh_qc_complete(qc);
556 * ata_eng_timeout - Handle timeout of queued command
557 * @ap: Port on which timed-out command is active
559 * Some part of the kernel (currently, only the SCSI layer)
560 * has noticed that the active command on port @ap has not
561 * completed after a specified length of time. Handle this
562 * condition by disabling DMA (if necessary) and completing
563 * transactions, with error if necessary.
565 * This also handles the case of the "lost interrupt", where
566 * for some reason (possibly hardware bug, possibly driver bug)
567 * an interrupt was not delivered to the driver, even though the
568 * transaction completed successfully.
570 * TODO: kill this function once old EH is gone.
573 * Inherited from SCSI layer (none, can sleep)
575 void ata_eng_timeout(struct ata_port *ap)
579 ata_qc_timeout(ata_qc_from_tag(ap, ap->link.active_tag));
584 static int ata_eh_nr_in_flight(struct ata_port *ap)
589 /* count only non-internal commands */
590 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
591 if (ata_qc_from_tag(ap, tag))
597 void ata_eh_fastdrain_timerfn(unsigned long arg)
599 struct ata_port *ap = (void *)arg;
603 spin_lock_irqsave(ap->lock, flags);
605 cnt = ata_eh_nr_in_flight(ap);
611 if (cnt == ap->fastdrain_cnt) {
614 /* No progress during the last interval, tag all
615 * in-flight qcs as timed out and freeze the port.
617 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
618 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
620 qc->err_mask |= AC_ERR_TIMEOUT;
625 /* some qcs have finished, give it another chance */
626 ap->fastdrain_cnt = cnt;
627 ap->fastdrain_timer.expires =
628 jiffies + ATA_EH_FASTDRAIN_INTERVAL;
629 add_timer(&ap->fastdrain_timer);
633 spin_unlock_irqrestore(ap->lock, flags);
637 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
638 * @ap: target ATA port
639 * @fastdrain: activate fast drain
641 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
642 * is non-zero and EH wasn't pending before. Fast drain ensures
643 * that EH kicks in in timely manner.
646 * spin_lock_irqsave(host lock)
648 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
652 /* already scheduled? */
653 if (ap->pflags & ATA_PFLAG_EH_PENDING)
656 ap->pflags |= ATA_PFLAG_EH_PENDING;
661 /* do we have in-flight qcs? */
662 cnt = ata_eh_nr_in_flight(ap);
666 /* activate fast drain */
667 ap->fastdrain_cnt = cnt;
668 ap->fastdrain_timer.expires = jiffies + ATA_EH_FASTDRAIN_INTERVAL;
669 add_timer(&ap->fastdrain_timer);
673 * ata_qc_schedule_eh - schedule qc for error handling
674 * @qc: command to schedule error handling for
676 * Schedule error handling for @qc. EH will kick in as soon as
677 * other commands are drained.
680 * spin_lock_irqsave(host lock)
682 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
684 struct ata_port *ap = qc->ap;
686 WARN_ON(!ap->ops->error_handler);
688 qc->flags |= ATA_QCFLAG_FAILED;
689 ata_eh_set_pending(ap, 1);
691 /* The following will fail if timeout has already expired.
692 * ata_scsi_error() takes care of such scmds on EH entry.
693 * Note that ATA_QCFLAG_FAILED is unconditionally set after
694 * this function completes.
696 scsi_req_abort_cmd(qc->scsicmd);
700 * ata_port_schedule_eh - schedule error handling without a qc
701 * @ap: ATA port to schedule EH for
703 * Schedule error handling for @ap. EH will kick in as soon as
704 * all commands are drained.
707 * spin_lock_irqsave(host lock)
709 void ata_port_schedule_eh(struct ata_port *ap)
711 WARN_ON(!ap->ops->error_handler);
713 if (ap->pflags & ATA_PFLAG_INITIALIZING)
716 ata_eh_set_pending(ap, 1);
717 scsi_schedule_eh(ap->scsi_host);
719 DPRINTK("port EH scheduled\n");
723 * ata_port_abort - abort all qc's on the port
724 * @ap: ATA port to abort qc's for
726 * Abort all active qc's of @ap and schedule EH.
729 * spin_lock_irqsave(host lock)
732 * Number of aborted qc's.
734 int ata_port_abort(struct ata_port *ap)
736 int tag, nr_aborted = 0;
738 WARN_ON(!ap->ops->error_handler);
740 /* we're gonna abort all commands, no need for fast drain */
741 ata_eh_set_pending(ap, 0);
743 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
744 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
747 qc->flags |= ATA_QCFLAG_FAILED;
754 ata_port_schedule_eh(ap);
760 * __ata_port_freeze - freeze port
761 * @ap: ATA port to freeze
763 * This function is called when HSM violation or some other
764 * condition disrupts normal operation of the port. Frozen port
765 * is not allowed to perform any operation until the port is
766 * thawed, which usually follows a successful reset.
768 * ap->ops->freeze() callback can be used for freezing the port
769 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
770 * port cannot be frozen hardware-wise, the interrupt handler
771 * must ack and clear interrupts unconditionally while the port
775 * spin_lock_irqsave(host lock)
777 static void __ata_port_freeze(struct ata_port *ap)
779 WARN_ON(!ap->ops->error_handler);
784 ap->pflags |= ATA_PFLAG_FROZEN;
786 DPRINTK("ata%u port frozen\n", ap->print_id);
790 * ata_port_freeze - abort & freeze port
791 * @ap: ATA port to freeze
793 * Abort and freeze @ap.
796 * spin_lock_irqsave(host lock)
799 * Number of aborted commands.
801 int ata_port_freeze(struct ata_port *ap)
805 WARN_ON(!ap->ops->error_handler);
807 nr_aborted = ata_port_abort(ap);
808 __ata_port_freeze(ap);
814 * ata_eh_freeze_port - EH helper to freeze port
815 * @ap: ATA port to freeze
822 void ata_eh_freeze_port(struct ata_port *ap)
826 if (!ap->ops->error_handler)
829 spin_lock_irqsave(ap->lock, flags);
830 __ata_port_freeze(ap);
831 spin_unlock_irqrestore(ap->lock, flags);
835 * ata_port_thaw_port - EH helper to thaw port
836 * @ap: ATA port to thaw
838 * Thaw frozen port @ap.
843 void ata_eh_thaw_port(struct ata_port *ap)
847 if (!ap->ops->error_handler)
850 spin_lock_irqsave(ap->lock, flags);
852 ap->pflags &= ~ATA_PFLAG_FROZEN;
857 spin_unlock_irqrestore(ap->lock, flags);
859 DPRINTK("ata%u port thawed\n", ap->print_id);
862 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
867 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
869 struct ata_port *ap = qc->ap;
870 struct scsi_cmnd *scmd = qc->scsicmd;
873 spin_lock_irqsave(ap->lock, flags);
874 qc->scsidone = ata_eh_scsidone;
875 __ata_qc_complete(qc);
876 WARN_ON(ata_tag_valid(qc->tag));
877 spin_unlock_irqrestore(ap->lock, flags);
879 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
883 * ata_eh_qc_complete - Complete an active ATA command from EH
884 * @qc: Command to complete
886 * Indicate to the mid and upper layers that an ATA command has
887 * completed. To be used from EH.
889 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
891 struct scsi_cmnd *scmd = qc->scsicmd;
892 scmd->retries = scmd->allowed;
893 __ata_eh_qc_complete(qc);
897 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
898 * @qc: Command to retry
900 * Indicate to the mid and upper layers that an ATA command
901 * should be retried. To be used from EH.
903 * SCSI midlayer limits the number of retries to scmd->allowed.
904 * scmd->retries is decremented for commands which get retried
905 * due to unrelated failures (qc->err_mask is zero).
907 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
909 struct scsi_cmnd *scmd = qc->scsicmd;
910 if (!qc->err_mask && scmd->retries)
912 __ata_eh_qc_complete(qc);
916 * ata_eh_detach_dev - detach ATA device
917 * @dev: ATA device to detach
924 static void ata_eh_detach_dev(struct ata_device *dev)
926 struct ata_link *link = dev->link;
927 struct ata_port *ap = link->ap;
930 ata_dev_disable(dev);
932 spin_lock_irqsave(ap->lock, flags);
934 dev->flags &= ~ATA_DFLAG_DETACH;
936 if (ata_scsi_offline_dev(dev)) {
937 dev->flags |= ATA_DFLAG_DETACHED;
938 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
941 /* clear per-dev EH actions */
942 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
943 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
945 spin_unlock_irqrestore(ap->lock, flags);
949 * ata_eh_about_to_do - about to perform eh_action
950 * @link: target ATA link
951 * @dev: target ATA dev for per-dev action (can be NULL)
952 * @action: action about to be performed
954 * Called just before performing EH actions to clear related bits
955 * in @link->eh_info such that eh actions are not unnecessarily
961 static void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
964 struct ata_port *ap = link->ap;
965 struct ata_eh_info *ehi = &link->eh_info;
966 struct ata_eh_context *ehc = &link->eh_context;
969 spin_lock_irqsave(ap->lock, flags);
971 /* Reset is represented by combination of actions and EHI
972 * flags. Suck in all related bits before clearing eh_info to
973 * avoid losing requested action.
975 if (action & ATA_EH_RESET_MASK) {
976 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
977 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
979 /* make sure all reset actions are cleared & clear EHI flags */
980 action |= ATA_EH_RESET_MASK;
981 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
984 ata_eh_clear_action(link, dev, ehi, action);
986 if (!(ehc->i.flags & ATA_EHI_QUIET))
987 ap->pflags |= ATA_PFLAG_RECOVERED;
989 spin_unlock_irqrestore(ap->lock, flags);
993 * ata_eh_done - EH action complete
994 * @ap: target ATA port
995 * @dev: target ATA dev for per-dev action (can be NULL)
996 * @action: action just completed
998 * Called right after performing EH actions to clear related bits
999 * in @link->eh_context.
1004 static void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1005 unsigned int action)
1007 struct ata_eh_context *ehc = &link->eh_context;
1009 /* if reset is complete, clear all reset actions & reset modifier */
1010 if (action & ATA_EH_RESET_MASK) {
1011 action |= ATA_EH_RESET_MASK;
1012 ehc->i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
1015 ata_eh_clear_action(link, dev, &ehc->i, action);
1019 * ata_err_string - convert err_mask to descriptive string
1020 * @err_mask: error mask to convert to string
1022 * Convert @err_mask to descriptive string. Errors are
1023 * prioritized according to severity and only the most severe
1024 * error is reported.
1030 * Descriptive string for @err_mask
1032 static const char * ata_err_string(unsigned int err_mask)
1034 if (err_mask & AC_ERR_HOST_BUS)
1035 return "host bus error";
1036 if (err_mask & AC_ERR_ATA_BUS)
1037 return "ATA bus error";
1038 if (err_mask & AC_ERR_TIMEOUT)
1040 if (err_mask & AC_ERR_HSM)
1041 return "HSM violation";
1042 if (err_mask & AC_ERR_SYSTEM)
1043 return "internal error";
1044 if (err_mask & AC_ERR_MEDIA)
1045 return "media error";
1046 if (err_mask & AC_ERR_INVALID)
1047 return "invalid argument";
1048 if (err_mask & AC_ERR_DEV)
1049 return "device error";
1050 return "unknown error";
1054 * ata_read_log_page - read a specific log page
1055 * @dev: target device
1056 * @page: page to read
1057 * @buf: buffer to store read page
1058 * @sectors: number of sectors to read
1060 * Read log page using READ_LOG_EXT command.
1063 * Kernel thread context (may sleep).
1066 * 0 on success, AC_ERR_* mask otherwise.
1068 static unsigned int ata_read_log_page(struct ata_device *dev,
1069 u8 page, void *buf, unsigned int sectors)
1071 struct ata_taskfile tf;
1072 unsigned int err_mask;
1074 DPRINTK("read log page - page %d\n", page);
1076 ata_tf_init(dev, &tf);
1077 tf.command = ATA_CMD_READ_LOG_EXT;
1080 tf.hob_nsect = sectors >> 8;
1081 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1082 tf.protocol = ATA_PROT_PIO;
1084 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1085 buf, sectors * ATA_SECT_SIZE);
1087 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1092 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1093 * @dev: Device to read log page 10h from
1094 * @tag: Resulting tag of the failed command
1095 * @tf: Resulting taskfile registers of the failed command
1097 * Read log page 10h to obtain NCQ error details and clear error
1101 * Kernel thread context (may sleep).
1104 * 0 on success, -errno otherwise.
1106 static int ata_eh_read_log_10h(struct ata_device *dev,
1107 int *tag, struct ata_taskfile *tf)
1109 u8 *buf = dev->link->ap->sector_buf;
1110 unsigned int err_mask;
1114 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1119 for (i = 0; i < ATA_SECT_SIZE; i++)
1122 ata_dev_printk(dev, KERN_WARNING,
1123 "invalid checksum 0x%x on log page 10h\n", csum);
1128 *tag = buf[0] & 0x1f;
1130 tf->command = buf[2];
1131 tf->feature = buf[3];
1135 tf->device = buf[7];
1136 tf->hob_lbal = buf[8];
1137 tf->hob_lbam = buf[9];
1138 tf->hob_lbah = buf[10];
1139 tf->nsect = buf[12];
1140 tf->hob_nsect = buf[13];
1146 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1147 * @dev: device to perform REQUEST_SENSE to
1148 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1150 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1151 * SENSE. This function is EH helper.
1154 * Kernel thread context (may sleep).
1157 * 0 on success, AC_ERR_* mask on failure
1159 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
1161 struct ata_device *dev = qc->dev;
1162 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
1163 struct ata_port *ap = dev->link->ap;
1164 struct ata_taskfile tf;
1165 u8 cdb[ATAPI_CDB_LEN];
1167 DPRINTK("ATAPI request sense\n");
1169 /* FIXME: is this needed? */
1170 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1172 /* initialize sense_buf with the error register,
1173 * for the case where they are -not- overwritten
1175 sense_buf[0] = 0x70;
1176 sense_buf[2] = qc->result_tf.feature >> 4;
1178 /* some devices time out if garbage left in tf */
1179 ata_tf_init(dev, &tf);
1181 memset(cdb, 0, ATAPI_CDB_LEN);
1182 cdb[0] = REQUEST_SENSE;
1183 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1185 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1186 tf.command = ATA_CMD_PACKET;
1188 /* is it pointless to prefer PIO for "safety reasons"? */
1189 if (ap->flags & ATA_FLAG_PIO_DMA) {
1190 tf.protocol = ATA_PROT_ATAPI_DMA;
1191 tf.feature |= ATAPI_PKT_DMA;
1193 tf.protocol = ATA_PROT_ATAPI;
1194 tf.lbam = (8 * 1024) & 0xff;
1195 tf.lbah = (8 * 1024) >> 8;
1198 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1199 sense_buf, SCSI_SENSE_BUFFERSIZE);
1203 * ata_eh_analyze_serror - analyze SError for a failed port
1204 * @ap: ATA port to analyze SError for
1206 * Analyze SError if available and further determine cause of
1212 static void ata_eh_analyze_serror(struct ata_port *ap)
1214 struct ata_eh_context *ehc = &ap->link.eh_context;
1215 u32 serror = ehc->i.serror;
1216 unsigned int err_mask = 0, action = 0;
1218 if (serror & SERR_PERSISTENT) {
1219 err_mask |= AC_ERR_ATA_BUS;
1220 action |= ATA_EH_HARDRESET;
1223 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1224 err_mask |= AC_ERR_ATA_BUS;
1225 action |= ATA_EH_SOFTRESET;
1227 if (serror & SERR_PROTOCOL) {
1228 err_mask |= AC_ERR_HSM;
1229 action |= ATA_EH_SOFTRESET;
1231 if (serror & SERR_INTERNAL) {
1232 err_mask |= AC_ERR_SYSTEM;
1233 action |= ATA_EH_HARDRESET;
1235 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1236 ata_ehi_hotplugged(&ehc->i);
1238 ehc->i.err_mask |= err_mask;
1239 ehc->i.action |= action;
1243 * ata_eh_analyze_ncq_error - analyze NCQ error
1244 * @ap: ATA port to analyze NCQ error for
1246 * Read log page 10h, determine the offending qc and acquire
1247 * error status TF. For NCQ device errors, all LLDDs have to do
1248 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1252 * Kernel thread context (may sleep).
1254 static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1256 struct ata_eh_context *ehc = &ap->link.eh_context;
1257 struct ata_device *dev = ap->link.device;
1258 struct ata_queued_cmd *qc;
1259 struct ata_taskfile tf;
1262 /* if frozen, we can't do much */
1263 if (ap->pflags & ATA_PFLAG_FROZEN)
1266 /* is it NCQ device error? */
1267 if (!ap->link.sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1270 /* has LLDD analyzed already? */
1271 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1272 qc = __ata_qc_from_tag(ap, tag);
1274 if (!(qc->flags & ATA_QCFLAG_FAILED))
1281 /* okay, this error is ours */
1282 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1284 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1285 "(errno=%d)\n", rc);
1289 if (!(ap->link.sactive & (1 << tag))) {
1290 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1291 "inactive tag %d\n", tag);
1295 /* we've got the perpetrator, condemn it */
1296 qc = __ata_qc_from_tag(ap, tag);
1297 memcpy(&qc->result_tf, &tf, sizeof(tf));
1298 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1299 ehc->i.err_mask &= ~AC_ERR_DEV;
1303 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1304 * @qc: qc to analyze
1305 * @tf: Taskfile registers to analyze
1307 * Analyze taskfile of @qc and further determine cause of
1308 * failure. This function also requests ATAPI sense data if
1312 * Kernel thread context (may sleep).
1315 * Determined recovery action
1317 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1318 const struct ata_taskfile *tf)
1320 unsigned int tmp, action = 0;
1321 u8 stat = tf->command, err = tf->feature;
1323 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1324 qc->err_mask |= AC_ERR_HSM;
1325 return ATA_EH_SOFTRESET;
1328 if (stat & (ATA_ERR | ATA_DF))
1329 qc->err_mask |= AC_ERR_DEV;
1333 switch (qc->dev->class) {
1336 qc->err_mask |= AC_ERR_ATA_BUS;
1338 qc->err_mask |= AC_ERR_MEDIA;
1340 qc->err_mask |= AC_ERR_INVALID;
1344 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1345 tmp = atapi_eh_request_sense(qc);
1347 /* ATA_QCFLAG_SENSE_VALID is used to
1348 * tell atapi_qc_complete() that sense
1349 * data is already valid.
1351 * TODO: interpret sense data and set
1352 * appropriate err_mask.
1354 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1356 qc->err_mask |= tmp;
1360 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1361 action |= ATA_EH_SOFTRESET;
1366 static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
1368 if (err_mask & AC_ERR_ATA_BUS)
1371 if (err_mask & AC_ERR_TIMEOUT)
1375 if (err_mask & AC_ERR_HSM)
1378 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1385 struct speed_down_verdict_arg {
1390 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1392 struct speed_down_verdict_arg *arg = void_arg;
1393 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
1395 if (ent->timestamp < arg->since)
1398 arg->nr_errors[cat]++;
1403 * ata_eh_speed_down_verdict - Determine speed down verdict
1404 * @dev: Device of interest
1406 * This function examines error ring of @dev and determines
1407 * whether NCQ needs to be turned off, transfer speed should be
1408 * stepped down, or falling back to PIO is necessary.
1410 * Cat-1 is ATA_BUS error for any command.
1412 * Cat-2 is TIMEOUT for any command or HSM violation for known
1413 * supported commands.
1415 * Cat-3 is is unclassified DEV error for known supported
1418 * NCQ needs to be turned off if there have been more than 3
1419 * Cat-2 + Cat-3 errors during last 10 minutes.
1421 * Speed down is necessary if there have been more than 3 Cat-1 +
1422 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1424 * Falling back to PIO mode is necessary if there have been more
1425 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1428 * Inherited from caller.
1431 * OR of ATA_EH_SPDN_* flags.
1433 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1435 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1436 u64 j64 = get_jiffies_64();
1437 struct speed_down_verdict_arg arg;
1438 unsigned int verdict = 0;
1440 /* scan past 10 mins of error history */
1441 memset(&arg, 0, sizeof(arg));
1442 arg.since = j64 - min(j64, j10mins);
1443 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1445 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1446 verdict |= ATA_EH_SPDN_NCQ_OFF;
1447 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1448 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1450 /* scan past 3 mins of error history */
1451 memset(&arg, 0, sizeof(arg));
1452 arg.since = j64 - min(j64, j5mins);
1453 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1455 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1456 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1462 * ata_eh_speed_down - record error and speed down if necessary
1463 * @dev: Failed device
1464 * @is_io: Did the device fail during normal IO?
1465 * @err_mask: err_mask of the error
1467 * Record error and examine error history to determine whether
1468 * adjusting transmission speed is necessary. It also sets
1469 * transmission limits appropriately if such adjustment is
1473 * Kernel thread context (may sleep).
1476 * Determined recovery action.
1478 static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1479 unsigned int err_mask)
1481 unsigned int verdict;
1482 unsigned int action = 0;
1484 /* don't bother if Cat-0 error */
1485 if (ata_eh_categorize_error(is_io, err_mask) == 0)
1488 /* record error and determine whether speed down is necessary */
1489 ata_ering_record(&dev->ering, is_io, err_mask);
1490 verdict = ata_eh_speed_down_verdict(dev);
1493 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1494 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1495 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1496 dev->flags |= ATA_DFLAG_NCQ_OFF;
1497 ata_dev_printk(dev, KERN_WARNING,
1498 "NCQ disabled due to excessive errors\n");
1503 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1504 /* speed down SATA link speed if possible */
1505 if (sata_down_spd_limit(dev->link) == 0) {
1506 action |= ATA_EH_HARDRESET;
1510 /* lower transfer mode */
1511 if (dev->spdn_cnt < 2) {
1512 static const int dma_dnxfer_sel[] =
1513 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1514 static const int pio_dnxfer_sel[] =
1515 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1518 if (dev->xfer_shift != ATA_SHIFT_PIO)
1519 sel = dma_dnxfer_sel[dev->spdn_cnt];
1521 sel = pio_dnxfer_sel[dev->spdn_cnt];
1525 if (ata_down_xfermask_limit(dev, sel) == 0) {
1526 action |= ATA_EH_SOFTRESET;
1532 /* Fall back to PIO? Slowing down to PIO is meaningless for
1533 * SATA. Consider it only for PATA.
1535 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1536 (dev->link->ap->cbl != ATA_CBL_SATA) &&
1537 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1538 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1540 action |= ATA_EH_SOFTRESET;
1547 /* device has been slowed down, blow error history */
1548 ata_ering_clear(&dev->ering);
1553 * ata_eh_autopsy - analyze error and determine recovery action
1554 * @ap: ATA port to perform autopsy on
1556 * Analyze why @ap failed and determine which recovery action is
1557 * needed. This function also sets more detailed AC_ERR_* values
1558 * and fills sense data for ATAPI CHECK SENSE.
1561 * Kernel thread context (may sleep).
1563 static void ata_eh_autopsy(struct ata_port *ap)
1565 struct ata_link *link = &ap->link;
1566 struct ata_eh_context *ehc = &link->eh_context;
1567 unsigned int all_err_mask = 0;
1574 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1577 /* obtain and analyze SError */
1578 rc = sata_scr_read(link, SCR_ERROR, &serror);
1580 ehc->i.serror |= serror;
1581 ata_eh_analyze_serror(ap);
1582 } else if (rc != -EOPNOTSUPP) {
1583 /* SError read failed, force hardreset and probing */
1584 ata_ehi_schedule_probe(&ehc->i);
1585 ehc->i.action |= ATA_EH_HARDRESET;
1586 ehc->i.err_mask |= AC_ERR_OTHER;
1589 /* analyze NCQ failure */
1590 ata_eh_analyze_ncq_error(ap);
1592 /* any real error trumps AC_ERR_OTHER */
1593 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1594 ehc->i.err_mask &= ~AC_ERR_OTHER;
1596 all_err_mask |= ehc->i.err_mask;
1598 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1599 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1601 if (!(qc->flags & ATA_QCFLAG_FAILED))
1604 /* inherit upper level err_mask */
1605 qc->err_mask |= ehc->i.err_mask;
1608 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1610 /* DEV errors are probably spurious in case of ATA_BUS error */
1611 if (qc->err_mask & AC_ERR_ATA_BUS)
1612 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1615 /* any real error trumps unknown error */
1616 if (qc->err_mask & ~AC_ERR_OTHER)
1617 qc->err_mask &= ~AC_ERR_OTHER;
1619 /* SENSE_VALID trumps dev/unknown error and revalidation */
1620 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1621 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1622 ehc->i.action &= ~ATA_EH_REVALIDATE;
1625 /* accumulate error info */
1626 ehc->i.dev = qc->dev;
1627 all_err_mask |= qc->err_mask;
1628 if (qc->flags & ATA_QCFLAG_IO)
1632 /* enforce default EH actions */
1633 if (ap->pflags & ATA_PFLAG_FROZEN ||
1634 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1635 ehc->i.action |= ATA_EH_SOFTRESET;
1636 else if (all_err_mask)
1637 ehc->i.action |= ATA_EH_REVALIDATE;
1639 /* if we have offending qcs and the associated failed device */
1642 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1645 /* perform per-dev EH action only on the offending device */
1646 ehc->i.dev_action[ehc->i.dev->devno] |=
1647 ehc->i.action & ATA_EH_PERDEV_MASK;
1648 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1655 * ata_eh_report - report error handling to user
1656 * @ap: ATA port EH is going on
1658 * Report EH to user.
1663 static void ata_eh_report(struct ata_port *ap)
1665 struct ata_eh_context *ehc = &ap->link.eh_context;
1666 const char *frozen, *desc;
1667 int tag, nr_failed = 0;
1670 if (ehc->i.desc[0] != '\0')
1673 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1674 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1676 if (!(qc->flags & ATA_QCFLAG_FAILED))
1678 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1684 if (!nr_failed && !ehc->i.err_mask)
1688 if (ap->pflags & ATA_PFLAG_FROZEN)
1692 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1693 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1694 ehc->i.err_mask, ap->link.sactive,
1695 ehc->i.serror, ehc->i.action, frozen);
1697 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
1699 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1700 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1701 ehc->i.err_mask, ap->link.sactive,
1702 ehc->i.serror, ehc->i.action, frozen);
1704 ata_port_printk(ap, KERN_ERR, "%s\n", desc);
1707 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1708 static const char *dma_str[] = {
1709 [DMA_BIDIRECTIONAL] = "bidi",
1710 [DMA_TO_DEVICE] = "out",
1711 [DMA_FROM_DEVICE] = "in",
1714 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1715 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1717 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1720 ata_dev_printk(qc->dev, KERN_ERR,
1721 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1722 "tag %d cdb 0x%x data %u %s\n "
1723 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1724 "Emask 0x%x (%s)%s\n",
1725 cmd->command, cmd->feature, cmd->nsect,
1726 cmd->lbal, cmd->lbam, cmd->lbah,
1727 cmd->hob_feature, cmd->hob_nsect,
1728 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1729 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
1730 dma_str[qc->dma_dir],
1731 res->command, res->feature, res->nsect,
1732 res->lbal, res->lbam, res->lbah,
1733 res->hob_feature, res->hob_nsect,
1734 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1735 res->device, qc->err_mask, ata_err_string(qc->err_mask),
1736 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
1740 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
1741 unsigned int *classes, unsigned long deadline)
1743 struct ata_device *dev;
1746 ata_link_for_each_dev(dev, link)
1747 classes[dev->devno] = ATA_DEV_UNKNOWN;
1749 rc = reset(link, classes, deadline);
1753 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1754 * is complete and convert all ATA_DEV_UNKNOWN to
1757 ata_link_for_each_dev(dev, link)
1758 if (classes[dev->devno] != ATA_DEV_UNKNOWN)
1762 ata_link_for_each_dev(dev, link) {
1763 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
1764 classes[dev->devno] = ATA_DEV_NONE;
1771 static int ata_eh_followup_srst_needed(int rc, int classify,
1772 const unsigned int *classes)
1778 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1783 static int ata_eh_reset(struct ata_link *link, int classify,
1784 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1785 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1787 struct ata_eh_context *ehc = &link->eh_context;
1788 unsigned int *classes = ehc->classes;
1789 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
1791 struct ata_device *dev;
1792 unsigned long deadline;
1793 unsigned int action;
1794 ata_reset_fn_t reset;
1797 /* about to reset */
1798 ata_eh_about_to_do(link, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1800 /* Determine which reset to use and record in ehc->i.action.
1801 * prereset() may examine and modify it.
1803 action = ehc->i.action;
1804 ehc->i.action &= ~ATA_EH_RESET_MASK;
1805 if (softreset && (!hardreset || (!sata_set_spd_needed(link) &&
1806 !(action & ATA_EH_HARDRESET))))
1807 ehc->i.action |= ATA_EH_SOFTRESET;
1809 ehc->i.action |= ATA_EH_HARDRESET;
1812 rc = prereset(link, jiffies + ATA_EH_PRERESET_TIMEOUT);
1814 if (rc == -ENOENT) {
1815 ata_link_printk(link, KERN_DEBUG,
1816 "port disabled. ignoring.\n");
1817 ehc->i.action &= ~ATA_EH_RESET_MASK;
1819 ata_link_for_each_dev(dev, link)
1820 classes[dev->devno] = ATA_DEV_NONE;
1824 ata_link_printk(link, KERN_ERR,
1825 "prereset failed (errno=%d)\n", rc);
1830 /* prereset() might have modified ehc->i.action */
1831 if (ehc->i.action & ATA_EH_HARDRESET)
1833 else if (ehc->i.action & ATA_EH_SOFTRESET)
1836 /* prereset told us not to reset, bang classes and return */
1837 ata_link_for_each_dev(dev, link)
1838 classes[dev->devno] = ATA_DEV_NONE;
1843 /* did prereset() screw up? if so, fix up to avoid oopsing */
1852 deadline = jiffies + ata_eh_reset_timeouts[try++];
1854 /* shut up during boot probing */
1856 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
1857 reset == softreset ? "soft" : "hard");
1859 /* mark that this EH session started with reset */
1860 if (reset == hardreset)
1861 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1863 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
1865 rc = ata_do_reset(link, reset, classes, deadline);
1867 if (reset == hardreset &&
1868 ata_eh_followup_srst_needed(rc, classify, classes)) {
1869 /* okay, let's do follow-up softreset */
1873 ata_link_printk(link, KERN_ERR,
1874 "follow-up softreset required "
1875 "but no softreset avaliable\n");
1880 ata_eh_about_to_do(link, NULL, ATA_EH_RESET_MASK);
1881 rc = ata_do_reset(link, reset, classes, deadline);
1883 if (rc == 0 && classify &&
1884 classes[0] == ATA_DEV_UNKNOWN) {
1885 ata_link_printk(link, KERN_ERR,
1886 "classification failed\n");
1892 if (rc && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
1893 unsigned long now = jiffies;
1895 if (time_before(now, deadline)) {
1896 unsigned long delta = deadline - jiffies;
1898 ata_link_printk(link, KERN_WARNING, "reset failed "
1899 "(errno=%d), retrying in %u secs\n",
1900 rc, (jiffies_to_msecs(delta) + 999) / 1000);
1902 schedule_timeout_uninterruptible(delta);
1906 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
1907 sata_down_spd_limit(link);
1916 /* After the reset, the device state is PIO 0 and the
1917 * controller state is undefined. Record the mode.
1919 ata_link_for_each_dev(dev, link)
1920 dev->pio_mode = XFER_PIO_0;
1922 /* record current link speed */
1923 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
1924 link->sata_spd = (sstatus >> 4) & 0xf;
1927 postreset(link, classes);
1929 /* reset successful, schedule revalidation */
1930 ata_eh_done(link, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1931 ehc->i.action |= ATA_EH_REVALIDATE;
1934 /* clear hotplug flag */
1935 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
1939 static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1940 struct ata_device **r_failed_dev)
1942 struct ata_eh_context *ehc = &ap->link.eh_context;
1943 struct ata_device *dev;
1944 unsigned int new_mask = 0;
1945 unsigned long flags;
1950 /* For PATA drive side cable detection to work, IDENTIFY must
1951 * be done backwards such that PDIAG- is released by the slave
1952 * device before the master device is identified.
1954 ata_link_for_each_dev_reverse(dev, &ap->link) {
1955 unsigned int action = ata_eh_dev_action(dev);
1956 unsigned int readid_flags = 0;
1958 if (ehc->i.flags & ATA_EHI_DID_RESET)
1959 readid_flags |= ATA_READID_POSTRESET;
1961 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
1962 if (ata_link_offline(&ap->link)) {
1967 ata_eh_about_to_do(&ap->link, dev, ATA_EH_REVALIDATE);
1968 rc = ata_dev_revalidate(dev, readid_flags);
1972 ata_eh_done(&ap->link, dev, ATA_EH_REVALIDATE);
1974 /* Configuration may have changed, reconfigure
1977 ehc->i.flags |= ATA_EHI_SETMODE;
1979 /* schedule the scsi_rescan_device() here */
1980 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
1981 } else if (dev->class == ATA_DEV_UNKNOWN &&
1982 ehc->tries[dev->devno] &&
1983 ata_class_enabled(ehc->classes[dev->devno])) {
1984 dev->class = ehc->classes[dev->devno];
1986 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1990 new_mask |= 1 << dev->devno;
1993 /* IDENTIFY was issued to non-existent
1994 * device. No need to reset. Just
1995 * thaw and kill the device.
1997 ata_eh_thaw_port(ap);
1998 dev->class = ATA_DEV_UNKNOWN;
2001 dev->class = ATA_DEV_UNKNOWN;
2007 /* PDIAG- should have been released, ask cable type if post-reset */
2008 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
2009 ap->cbl = ap->ops->cable_detect(ap);
2011 /* Configure new devices forward such that user doesn't see
2012 * device detection messages backwards.
2014 ata_link_for_each_dev(dev, &ap->link) {
2015 if (!(new_mask & (1 << dev->devno)))
2018 ehc->i.flags |= ATA_EHI_PRINTINFO;
2019 rc = ata_dev_configure(dev);
2020 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2024 spin_lock_irqsave(ap->lock, flags);
2025 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2026 spin_unlock_irqrestore(ap->lock, flags);
2028 /* new device discovered, configure xfermode */
2029 ehc->i.flags |= ATA_EHI_SETMODE;
2035 *r_failed_dev = dev;
2036 DPRINTK("EXIT rc=%d\n", rc);
2040 static int ata_port_nr_enabled(struct ata_port *ap)
2042 struct ata_device *dev;
2045 ata_link_for_each_dev(dev, &ap->link)
2046 if (ata_dev_enabled(dev))
2051 static int ata_port_nr_vacant(struct ata_port *ap)
2053 struct ata_device *dev;
2056 ata_link_for_each_dev(dev, &ap->link)
2057 if (dev->class == ATA_DEV_UNKNOWN)
2062 static int ata_eh_skip_recovery(struct ata_port *ap)
2064 struct ata_eh_context *ehc = &ap->link.eh_context;
2065 struct ata_device *dev;
2067 /* thaw frozen port, resume link and recover failed devices */
2068 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
2069 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
2072 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2073 ata_link_for_each_dev(dev, &ap->link) {
2074 if (dev->class == ATA_DEV_UNKNOWN &&
2075 ehc->classes[dev->devno] != ATA_DEV_NONE)
2082 static void ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2084 struct ata_eh_context *ehc = &dev->link->eh_context;
2086 ehc->tries[dev->devno]--;
2090 /* device missing or wrong IDENTIFY data, schedule probing */
2091 ehc->i.probe_mask |= (1 << dev->devno);
2093 /* give it just one more chance */
2094 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2096 if (ehc->tries[dev->devno] == 1) {
2097 /* This is the last chance, better to slow
2098 * down than lose it.
2100 sata_down_spd_limit(dev->link);
2101 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2105 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2106 /* disable device if it has used up all its chances */
2107 ata_dev_disable(dev);
2109 /* detach if offline */
2110 if (ata_link_offline(dev->link))
2111 ata_eh_detach_dev(dev);
2113 /* probe if requested */
2114 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2115 !(ehc->did_probe_mask & (1 << dev->devno))) {
2116 ata_eh_detach_dev(dev);
2119 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2120 ehc->did_probe_mask |= (1 << dev->devno);
2121 ehc->i.action |= ATA_EH_SOFTRESET;
2124 /* soft didn't work? be haaaaard */
2125 if (ehc->i.flags & ATA_EHI_DID_RESET)
2126 ehc->i.action |= ATA_EH_HARDRESET;
2128 ehc->i.action |= ATA_EH_SOFTRESET;
2133 * ata_eh_recover - recover host port after error
2134 * @ap: host port to recover
2135 * @prereset: prereset method (can be NULL)
2136 * @softreset: softreset method (can be NULL)
2137 * @hardreset: hardreset method (can be NULL)
2138 * @postreset: postreset method (can be NULL)
2140 * This is the alpha and omega, eum and yang, heart and soul of
2141 * libata exception handling. On entry, actions required to
2142 * recover the port and hotplug requests are recorded in
2143 * eh_context. This function executes all the operations with
2144 * appropriate retrials and fallbacks to resurrect failed
2145 * devices, detach goners and greet newcomers.
2148 * Kernel thread context (may sleep).
2151 * 0 on success, -errno on failure.
2153 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2154 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2155 ata_postreset_fn_t postreset)
2157 struct ata_eh_context *ehc = &ap->link.eh_context;
2158 struct ata_device *dev;
2163 /* prep for recovery */
2164 ata_link_for_each_dev(dev, &ap->link) {
2165 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2167 /* collect port action mask recorded in dev actions */
2169 ehc->i.dev_action[dev->devno] & ~ATA_EH_PERDEV_MASK;
2170 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
2172 /* process hotplug request */
2173 if (dev->flags & ATA_DFLAG_DETACH)
2174 ata_eh_detach_dev(dev);
2176 if (!ata_dev_enabled(dev) &&
2177 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2178 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2179 ata_eh_detach_dev(dev);
2181 ehc->did_probe_mask |= (1 << dev->devno);
2182 ehc->i.action |= ATA_EH_SOFTRESET;
2189 /* if UNLOADING, finish immediately */
2190 if (ap->pflags & ATA_PFLAG_UNLOADING)
2193 /* skip EH if possible. */
2194 if (ata_eh_skip_recovery(ap))
2197 ata_link_for_each_dev(dev, &ap->link)
2198 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2201 if (ehc->i.action & ATA_EH_RESET_MASK) {
2202 ata_eh_freeze_port(ap);
2204 rc = ata_eh_reset(&ap->link, ata_port_nr_vacant(ap), prereset,
2205 softreset, hardreset, postreset);
2207 ata_port_printk(ap, KERN_ERR,
2208 "reset failed, giving up\n");
2212 ata_eh_thaw_port(ap);
2215 /* revalidate existing devices and attach new ones */
2216 rc = ata_eh_revalidate_and_attach(ap, &dev);
2220 /* configure transfer mode if necessary */
2221 if (ehc->i.flags & ATA_EHI_SETMODE) {
2222 rc = ata_set_mode(ap, &dev);
2225 ehc->i.flags &= ~ATA_EHI_SETMODE;
2231 ata_eh_handle_dev_fail(dev, rc);
2233 if (ata_port_nr_enabled(ap)) {
2234 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2235 "devices, retrying in 5 secs\n");
2238 /* no device left, repeat fast */
2246 ata_link_for_each_dev(dev, &ap->link);
2247 ata_dev_disable(dev);
2250 DPRINTK("EXIT, rc=%d\n", rc);
2255 * ata_eh_finish - finish up EH
2256 * @ap: host port to finish EH for
2258 * Recovery is complete. Clean up EH states and retry or finish
2264 static void ata_eh_finish(struct ata_port *ap)
2268 /* retry or finish qcs */
2269 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2270 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2272 if (!(qc->flags & ATA_QCFLAG_FAILED))
2276 /* FIXME: Once EH migration is complete,
2277 * generate sense data in this function,
2278 * considering both err_mask and tf.
2280 if (qc->err_mask & AC_ERR_INVALID)
2281 ata_eh_qc_complete(qc);
2283 ata_eh_qc_retry(qc);
2285 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2286 ata_eh_qc_complete(qc);
2288 /* feed zero TF to sense generation */
2289 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2290 ata_eh_qc_retry(qc);
2297 * ata_do_eh - do standard error handling
2298 * @ap: host port to handle error for
2299 * @prereset: prereset method (can be NULL)
2300 * @softreset: softreset method (can be NULL)
2301 * @hardreset: hardreset method (can be NULL)
2302 * @postreset: postreset method (can be NULL)
2304 * Perform standard error handling sequence.
2307 * Kernel thread context (may sleep).
2309 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2310 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2311 ata_postreset_fn_t postreset)
2315 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
2321 * ata_eh_handle_port_suspend - perform port suspend operation
2322 * @ap: port to suspend
2327 * Kernel thread context (may sleep).
2329 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2331 unsigned long flags;
2334 /* are we suspending? */
2335 spin_lock_irqsave(ap->lock, flags);
2336 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2337 ap->pm_mesg.event == PM_EVENT_ON) {
2338 spin_unlock_irqrestore(ap->lock, flags);
2341 spin_unlock_irqrestore(ap->lock, flags);
2343 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2345 /* tell ACPI we're suspending */
2346 rc = ata_acpi_on_suspend(ap);
2351 ata_eh_freeze_port(ap);
2353 if (ap->ops->port_suspend)
2354 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2358 spin_lock_irqsave(ap->lock, flags);
2360 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2362 ap->pflags |= ATA_PFLAG_SUSPENDED;
2363 else if (ap->pflags & ATA_PFLAG_FROZEN)
2364 ata_port_schedule_eh(ap);
2366 if (ap->pm_result) {
2367 *ap->pm_result = rc;
2368 ap->pm_result = NULL;
2371 spin_unlock_irqrestore(ap->lock, flags);
2377 * ata_eh_handle_port_resume - perform port resume operation
2378 * @ap: port to resume
2383 * Kernel thread context (may sleep).
2385 static void ata_eh_handle_port_resume(struct ata_port *ap)
2387 unsigned long flags;
2390 /* are we resuming? */
2391 spin_lock_irqsave(ap->lock, flags);
2392 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2393 ap->pm_mesg.event != PM_EVENT_ON) {
2394 spin_unlock_irqrestore(ap->lock, flags);
2397 spin_unlock_irqrestore(ap->lock, flags);
2399 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
2401 if (ap->ops->port_resume)
2402 rc = ap->ops->port_resume(ap);
2404 /* tell ACPI that we're resuming */
2405 ata_acpi_on_resume(ap);
2408 spin_lock_irqsave(ap->lock, flags);
2409 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2410 if (ap->pm_result) {
2411 *ap->pm_result = rc;
2412 ap->pm_result = NULL;
2414 spin_unlock_irqrestore(ap->lock, flags);
2416 #endif /* CONFIG_PM */