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 <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include "../scsi/scsi_transport_api.h"
45 #include <linux/libata.h>
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
57 ATA_EFLAG_IS_IO = (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60 /* error categories */
63 ATA_ECAT_TOUT_HSM = 2,
65 ATA_ECAT_DUBIOUS_NONE = 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_EH_CMD_DFL_TIMEOUT = 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN = 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT = 10000,
82 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87 /* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
93 static const unsigned long ata_eh_reset_timeouts[] = {
94 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
98 ULONG_MAX, /* > 1 min has elapsed, give up */
101 static const unsigned long ata_eh_identify_timeouts[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
108 static const unsigned long ata_eh_other_timeouts[] = {
109 5000, /* same rationale as identify timeout */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
115 struct ata_eh_cmd_timeout_ent {
117 const unsigned long *timeouts;
120 /* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
132 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
133 static const struct ata_eh_cmd_timeout_ent
134 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
135 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
136 .timeouts = ata_eh_identify_timeouts, },
137 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
138 .timeouts = ata_eh_other_timeouts, },
139 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
140 .timeouts = ata_eh_other_timeouts, },
141 { .commands = CMDS(ATA_CMD_SET_FEATURES),
142 .timeouts = ata_eh_other_timeouts, },
143 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
144 .timeouts = ata_eh_other_timeouts, },
148 static void __ata_port_freeze(struct ata_port *ap);
150 static void ata_eh_handle_port_suspend(struct ata_port *ap);
151 static void ata_eh_handle_port_resume(struct ata_port *ap);
152 #else /* CONFIG_PM */
153 static void ata_eh_handle_port_suspend(struct ata_port *ap)
156 static void ata_eh_handle_port_resume(struct ata_port *ap)
158 #endif /* CONFIG_PM */
160 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
163 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
164 ATA_EH_DESC_LEN - ehi->desc_len,
169 * __ata_ehi_push_desc - push error description without adding separator
171 * @fmt: printf format string
173 * Format string according to @fmt and append it to @ehi->desc.
176 * spin_lock_irqsave(host lock)
178 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
183 __ata_ehi_pushv_desc(ehi, fmt, args);
188 * ata_ehi_push_desc - push error description with separator
190 * @fmt: printf format string
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
196 * spin_lock_irqsave(host lock)
198 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
203 __ata_ehi_push_desc(ehi, ", ");
206 __ata_ehi_pushv_desc(ehi, fmt, args);
211 * ata_ehi_clear_desc - clean error description
217 * spin_lock_irqsave(host lock)
219 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
238 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
242 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
244 if (ap->link.eh_info.desc_len)
245 __ata_ehi_push_desc(&ap->link.eh_info, " ");
248 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
269 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
272 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
274 unsigned long long start, len;
276 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
278 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
281 start = (unsigned long long)pci_resource_start(pdev, bar);
282 len = (unsigned long long)pci_resource_len(pdev, bar);
285 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
287 ata_port_desc(ap, "%s 0x%llx", name,
288 start + (unsigned long long)offset);
291 #endif /* CONFIG_PCI */
293 static int ata_lookup_timeout_table(u8 cmd)
297 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
300 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
313 * Determine timeout for internal command @cmd for @dev.
319 * Determined timeout.
321 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
323 struct ata_eh_context *ehc = &dev->link->eh_context;
324 int ent = ata_lookup_timeout_table(cmd);
328 return ATA_EH_CMD_DFL_TIMEOUT;
330 idx = ehc->cmd_timeout_idx[dev->devno][ent];
331 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
346 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
348 struct ata_eh_context *ehc = &dev->link->eh_context;
349 int ent = ata_lookup_timeout_table(cmd);
355 idx = ehc->cmd_timeout_idx[dev->devno][ent];
356 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
357 ehc->cmd_timeout_idx[dev->devno][ent]++;
360 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
361 unsigned int err_mask)
363 struct ata_ering_entry *ent;
368 ering->cursor %= ATA_ERING_SIZE;
370 ent = &ering->ring[ering->cursor];
371 ent->eflags = eflags;
372 ent->err_mask = err_mask;
373 ent->timestamp = get_jiffies_64();
376 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
378 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
385 static void ata_ering_clear(struct ata_ering *ering)
387 memset(ering, 0, sizeof(*ering));
390 static int ata_ering_map(struct ata_ering *ering,
391 int (*map_fn)(struct ata_ering_entry *, void *),
395 struct ata_ering_entry *ent;
399 ent = &ering->ring[idx];
402 rc = map_fn(ent, arg);
405 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
406 } while (idx != ering->cursor);
411 static unsigned int ata_eh_dev_action(struct ata_device *dev)
413 struct ata_eh_context *ehc = &dev->link->eh_context;
415 return ehc->i.action | ehc->i.dev_action[dev->devno];
418 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
419 struct ata_eh_info *ehi, unsigned int action)
421 struct ata_device *tdev;
424 ehi->action &= ~action;
425 ata_link_for_each_dev(tdev, link)
426 ehi->dev_action[tdev->devno] &= ~action;
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
431 /* break ehi->action into ehi->dev_action */
432 if (ehi->action & action) {
433 ata_link_for_each_dev(tdev, link)
434 ehi->dev_action[tdev->devno] |=
435 ehi->action & action;
436 ehi->action &= ~action;
439 /* turn off the specified per-dev action */
440 ehi->dev_action[dev->devno] &= ~action;
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
455 * TODO: kill this function once old EH is gone.
458 * Called from timer context
461 * EH_HANDLED or EH_NOT_HANDLED
463 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
465 struct Scsi_Host *host = cmd->device->host;
466 struct ata_port *ap = ata_shost_to_port(host);
468 struct ata_queued_cmd *qc;
469 enum blk_eh_timer_return ret;
473 if (ap->ops->error_handler) {
474 ret = BLK_EH_NOT_HANDLED;
478 ret = BLK_EH_HANDLED;
479 spin_lock_irqsave(ap->lock, flags);
480 qc = ata_qc_from_tag(ap, ap->link.active_tag);
482 WARN_ON(qc->scsicmd != cmd);
483 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
484 qc->err_mask |= AC_ERR_TIMEOUT;
485 ret = BLK_EH_NOT_HANDLED;
487 spin_unlock_irqrestore(ap->lock, flags);
490 DPRINTK("EXIT, ret=%d\n", ret);
495 * ata_scsi_error - SCSI layer error handler callback
496 * @host: SCSI host on which error occurred
498 * Handles SCSI-layer-thrown error events.
501 * Inherited from SCSI layer (none, can sleep)
506 void ata_scsi_error(struct Scsi_Host *host)
508 struct ata_port *ap = ata_shost_to_port(host);
514 /* synchronize with port task */
515 ata_port_flush_task(ap);
517 /* synchronize with host lock and sort out timeouts */
519 /* For new EH, all qcs are finished in one of three ways -
520 * normal completion, error completion, and SCSI timeout.
521 * Both cmpletions can race against SCSI timeout. When normal
522 * completion wins, the qc never reaches EH. When error
523 * completion wins, the qc has ATA_QCFLAG_FAILED set.
525 * When SCSI timeout wins, things are a bit more complex.
526 * Normal or error completion can occur after the timeout but
527 * before this point. In such cases, both types of
528 * completions are honored. A scmd is determined to have
529 * timed out iff its associated qc is active and not failed.
531 if (ap->ops->error_handler) {
532 struct scsi_cmnd *scmd, *tmp;
535 spin_lock_irqsave(ap->lock, flags);
537 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
538 struct ata_queued_cmd *qc;
540 for (i = 0; i < ATA_MAX_QUEUE; i++) {
541 qc = __ata_qc_from_tag(ap, i);
542 if (qc->flags & ATA_QCFLAG_ACTIVE &&
547 if (i < ATA_MAX_QUEUE) {
548 /* the scmd has an associated qc */
549 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
550 /* which hasn't failed yet, timeout */
551 qc->err_mask |= AC_ERR_TIMEOUT;
552 qc->flags |= ATA_QCFLAG_FAILED;
556 /* Normal completion occurred after
557 * SCSI timeout but before this point.
558 * Successfully complete it.
560 scmd->retries = scmd->allowed;
561 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
565 /* If we have timed out qcs. They belong to EH from
566 * this point but the state of the controller is
567 * unknown. Freeze the port to make sure the IRQ
568 * handler doesn't diddle with those qcs. This must
569 * be done atomically w.r.t. setting QCFLAG_FAILED.
572 __ata_port_freeze(ap);
574 spin_unlock_irqrestore(ap->lock, flags);
576 /* initialize eh_tries */
577 ap->eh_tries = ATA_EH_MAX_TRIES;
579 spin_unlock_wait(ap->lock);
582 /* invoke error handler */
583 if (ap->ops->error_handler) {
584 struct ata_link *link;
586 /* kill fast drain timer */
587 del_timer_sync(&ap->fastdrain_timer);
589 /* process port resume request */
590 ata_eh_handle_port_resume(ap);
592 /* fetch & clear EH info */
593 spin_lock_irqsave(ap->lock, flags);
595 __ata_port_for_each_link(link, ap) {
596 struct ata_eh_context *ehc = &link->eh_context;
597 struct ata_device *dev;
599 memset(&link->eh_context, 0, sizeof(link->eh_context));
600 link->eh_context.i = link->eh_info;
601 memset(&link->eh_info, 0, sizeof(link->eh_info));
603 ata_link_for_each_dev(dev, link) {
604 int devno = dev->devno;
606 if (!ata_dev_enabled(dev))
609 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
610 if (ata_ncq_enabled(dev))
611 ehc->saved_ncq_enabled |= 1 << devno;
614 /* set last reset timestamp to some time in the past */
615 ehc->last_reset = jiffies - 60 * HZ;
618 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
619 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
620 ap->excl_link = NULL; /* don't maintain exclusion over EH */
622 spin_unlock_irqrestore(ap->lock, flags);
624 /* invoke EH, skip if unloading or suspended */
625 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
626 ap->ops->error_handler(ap);
630 /* process port suspend request */
631 ata_eh_handle_port_suspend(ap);
633 /* Exception might have happend after ->error_handler
634 * recovered the port but before this point. Repeat
637 spin_lock_irqsave(ap->lock, flags);
639 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
640 if (--ap->eh_tries) {
641 spin_unlock_irqrestore(ap->lock, flags);
644 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
645 "tries, giving up\n", ATA_EH_MAX_TRIES);
646 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
649 /* this run is complete, make sure EH info is clear */
650 __ata_port_for_each_link(link, ap)
651 memset(&link->eh_info, 0, sizeof(link->eh_info));
653 /* Clear host_eh_scheduled while holding ap->lock such
654 * that if exception occurs after this point but
655 * before EH completion, SCSI midlayer will
658 host->host_eh_scheduled = 0;
660 spin_unlock_irqrestore(ap->lock, flags);
662 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
663 ap->ops->eng_timeout(ap);
666 /* finish or retry handled scmd's and clean up */
667 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
669 scsi_eh_flush_done_q(&ap->eh_done_q);
672 spin_lock_irqsave(ap->lock, flags);
674 if (ap->pflags & ATA_PFLAG_LOADING)
675 ap->pflags &= ~ATA_PFLAG_LOADING;
676 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
677 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
679 if (ap->pflags & ATA_PFLAG_RECOVERED)
680 ata_port_printk(ap, KERN_INFO, "EH complete\n");
682 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
684 /* tell wait_eh that we're done */
685 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
686 wake_up_all(&ap->eh_wait_q);
688 spin_unlock_irqrestore(ap->lock, flags);
694 * ata_port_wait_eh - Wait for the currently pending EH to complete
695 * @ap: Port to wait EH for
697 * Wait until the currently pending EH is complete.
700 * Kernel thread context (may sleep).
702 void ata_port_wait_eh(struct ata_port *ap)
708 spin_lock_irqsave(ap->lock, flags);
710 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
711 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
712 spin_unlock_irqrestore(ap->lock, flags);
714 spin_lock_irqsave(ap->lock, flags);
716 finish_wait(&ap->eh_wait_q, &wait);
718 spin_unlock_irqrestore(ap->lock, flags);
720 /* make sure SCSI EH is complete */
721 if (scsi_host_in_recovery(ap->scsi_host)) {
727 static int ata_eh_nr_in_flight(struct ata_port *ap)
732 /* count only non-internal commands */
733 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
734 if (ata_qc_from_tag(ap, tag))
740 void ata_eh_fastdrain_timerfn(unsigned long arg)
742 struct ata_port *ap = (void *)arg;
746 spin_lock_irqsave(ap->lock, flags);
748 cnt = ata_eh_nr_in_flight(ap);
754 if (cnt == ap->fastdrain_cnt) {
757 /* No progress during the last interval, tag all
758 * in-flight qcs as timed out and freeze the port.
760 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
761 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
763 qc->err_mask |= AC_ERR_TIMEOUT;
768 /* some qcs have finished, give it another chance */
769 ap->fastdrain_cnt = cnt;
770 ap->fastdrain_timer.expires =
771 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
772 add_timer(&ap->fastdrain_timer);
776 spin_unlock_irqrestore(ap->lock, flags);
780 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
781 * @ap: target ATA port
782 * @fastdrain: activate fast drain
784 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
785 * is non-zero and EH wasn't pending before. Fast drain ensures
786 * that EH kicks in in timely manner.
789 * spin_lock_irqsave(host lock)
791 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
795 /* already scheduled? */
796 if (ap->pflags & ATA_PFLAG_EH_PENDING)
799 ap->pflags |= ATA_PFLAG_EH_PENDING;
804 /* do we have in-flight qcs? */
805 cnt = ata_eh_nr_in_flight(ap);
809 /* activate fast drain */
810 ap->fastdrain_cnt = cnt;
811 ap->fastdrain_timer.expires =
812 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
813 add_timer(&ap->fastdrain_timer);
817 * ata_qc_schedule_eh - schedule qc for error handling
818 * @qc: command to schedule error handling for
820 * Schedule error handling for @qc. EH will kick in as soon as
821 * other commands are drained.
824 * spin_lock_irqsave(host lock)
826 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
828 struct ata_port *ap = qc->ap;
830 WARN_ON(!ap->ops->error_handler);
832 qc->flags |= ATA_QCFLAG_FAILED;
833 ata_eh_set_pending(ap, 1);
835 /* The following will fail if timeout has already expired.
836 * ata_scsi_error() takes care of such scmds on EH entry.
837 * Note that ATA_QCFLAG_FAILED is unconditionally set after
838 * this function completes.
840 blk_abort_request(qc->scsicmd->request);
844 * ata_port_schedule_eh - schedule error handling without a qc
845 * @ap: ATA port to schedule EH for
847 * Schedule error handling for @ap. EH will kick in as soon as
848 * all commands are drained.
851 * spin_lock_irqsave(host lock)
853 void ata_port_schedule_eh(struct ata_port *ap)
855 WARN_ON(!ap->ops->error_handler);
857 if (ap->pflags & ATA_PFLAG_INITIALIZING)
860 ata_eh_set_pending(ap, 1);
861 scsi_schedule_eh(ap->scsi_host);
863 DPRINTK("port EH scheduled\n");
866 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
868 int tag, nr_aborted = 0;
870 WARN_ON(!ap->ops->error_handler);
872 /* we're gonna abort all commands, no need for fast drain */
873 ata_eh_set_pending(ap, 0);
875 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
876 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
878 if (qc && (!link || qc->dev->link == link)) {
879 qc->flags |= ATA_QCFLAG_FAILED;
886 ata_port_schedule_eh(ap);
892 * ata_link_abort - abort all qc's on the link
893 * @link: ATA link to abort qc's for
895 * Abort all active qc's active on @link and schedule EH.
898 * spin_lock_irqsave(host lock)
901 * Number of aborted qc's.
903 int ata_link_abort(struct ata_link *link)
905 return ata_do_link_abort(link->ap, link);
909 * ata_port_abort - abort all qc's on the port
910 * @ap: ATA port to abort qc's for
912 * Abort all active qc's of @ap and schedule EH.
915 * spin_lock_irqsave(host_set lock)
918 * Number of aborted qc's.
920 int ata_port_abort(struct ata_port *ap)
922 return ata_do_link_abort(ap, NULL);
926 * __ata_port_freeze - freeze port
927 * @ap: ATA port to freeze
929 * This function is called when HSM violation or some other
930 * condition disrupts normal operation of the port. Frozen port
931 * is not allowed to perform any operation until the port is
932 * thawed, which usually follows a successful reset.
934 * ap->ops->freeze() callback can be used for freezing the port
935 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
936 * port cannot be frozen hardware-wise, the interrupt handler
937 * must ack and clear interrupts unconditionally while the port
941 * spin_lock_irqsave(host lock)
943 static void __ata_port_freeze(struct ata_port *ap)
945 WARN_ON(!ap->ops->error_handler);
950 ap->pflags |= ATA_PFLAG_FROZEN;
952 DPRINTK("ata%u port frozen\n", ap->print_id);
956 * ata_port_freeze - abort & freeze port
957 * @ap: ATA port to freeze
959 * Abort and freeze @ap.
962 * spin_lock_irqsave(host lock)
965 * Number of aborted commands.
967 int ata_port_freeze(struct ata_port *ap)
971 WARN_ON(!ap->ops->error_handler);
973 nr_aborted = ata_port_abort(ap);
974 __ata_port_freeze(ap);
980 * sata_async_notification - SATA async notification handler
981 * @ap: ATA port where async notification is received
983 * Handler to be called when async notification via SDB FIS is
984 * received. This function schedules EH if necessary.
987 * spin_lock_irqsave(host lock)
990 * 1 if EH is scheduled, 0 otherwise.
992 int sata_async_notification(struct ata_port *ap)
997 if (!(ap->flags & ATA_FLAG_AN))
1000 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1002 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1004 if (!sata_pmp_attached(ap) || rc) {
1005 /* PMP is not attached or SNTF is not available */
1006 if (!sata_pmp_attached(ap)) {
1007 /* PMP is not attached. Check whether ATAPI
1008 * AN is configured. If so, notify media
1011 struct ata_device *dev = ap->link.device;
1013 if ((dev->class == ATA_DEV_ATAPI) &&
1014 (dev->flags & ATA_DFLAG_AN))
1015 ata_scsi_media_change_notify(dev);
1018 /* PMP is attached but SNTF is not available.
1019 * ATAPI async media change notification is
1020 * not used. The PMP must be reporting PHY
1021 * status change, schedule EH.
1023 ata_port_schedule_eh(ap);
1027 /* PMP is attached and SNTF is available */
1028 struct ata_link *link;
1030 /* check and notify ATAPI AN */
1031 ata_port_for_each_link(link, ap) {
1032 if (!(sntf & (1 << link->pmp)))
1035 if ((link->device->class == ATA_DEV_ATAPI) &&
1036 (link->device->flags & ATA_DFLAG_AN))
1037 ata_scsi_media_change_notify(link->device);
1040 /* If PMP is reporting that PHY status of some
1041 * downstream ports has changed, schedule EH.
1043 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1044 ata_port_schedule_eh(ap);
1053 * ata_eh_freeze_port - EH helper to freeze port
1054 * @ap: ATA port to freeze
1061 void ata_eh_freeze_port(struct ata_port *ap)
1063 unsigned long flags;
1065 if (!ap->ops->error_handler)
1068 spin_lock_irqsave(ap->lock, flags);
1069 __ata_port_freeze(ap);
1070 spin_unlock_irqrestore(ap->lock, flags);
1074 * ata_port_thaw_port - EH helper to thaw port
1075 * @ap: ATA port to thaw
1077 * Thaw frozen port @ap.
1082 void ata_eh_thaw_port(struct ata_port *ap)
1084 unsigned long flags;
1086 if (!ap->ops->error_handler)
1089 spin_lock_irqsave(ap->lock, flags);
1091 ap->pflags &= ~ATA_PFLAG_FROZEN;
1096 spin_unlock_irqrestore(ap->lock, flags);
1098 DPRINTK("ata%u port thawed\n", ap->print_id);
1101 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1106 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1108 struct ata_port *ap = qc->ap;
1109 struct scsi_cmnd *scmd = qc->scsicmd;
1110 unsigned long flags;
1112 spin_lock_irqsave(ap->lock, flags);
1113 qc->scsidone = ata_eh_scsidone;
1114 __ata_qc_complete(qc);
1115 WARN_ON(ata_tag_valid(qc->tag));
1116 spin_unlock_irqrestore(ap->lock, flags);
1118 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1122 * ata_eh_qc_complete - Complete an active ATA command from EH
1123 * @qc: Command to complete
1125 * Indicate to the mid and upper layers that an ATA command has
1126 * completed. To be used from EH.
1128 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1130 struct scsi_cmnd *scmd = qc->scsicmd;
1131 scmd->retries = scmd->allowed;
1132 __ata_eh_qc_complete(qc);
1136 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1137 * @qc: Command to retry
1139 * Indicate to the mid and upper layers that an ATA command
1140 * should be retried. To be used from EH.
1142 * SCSI midlayer limits the number of retries to scmd->allowed.
1143 * scmd->retries is decremented for commands which get retried
1144 * due to unrelated failures (qc->err_mask is zero).
1146 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1148 struct scsi_cmnd *scmd = qc->scsicmd;
1149 if (!qc->err_mask && scmd->retries)
1151 __ata_eh_qc_complete(qc);
1155 * ata_eh_detach_dev - detach ATA device
1156 * @dev: ATA device to detach
1163 void ata_eh_detach_dev(struct ata_device *dev)
1165 struct ata_link *link = dev->link;
1166 struct ata_port *ap = link->ap;
1167 unsigned long flags;
1169 ata_dev_disable(dev);
1171 spin_lock_irqsave(ap->lock, flags);
1173 dev->flags &= ~ATA_DFLAG_DETACH;
1175 if (ata_scsi_offline_dev(dev)) {
1176 dev->flags |= ATA_DFLAG_DETACHED;
1177 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1180 /* clear per-dev EH actions */
1181 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1182 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1184 spin_unlock_irqrestore(ap->lock, flags);
1188 * ata_eh_about_to_do - about to perform eh_action
1189 * @link: target ATA link
1190 * @dev: target ATA dev for per-dev action (can be NULL)
1191 * @action: action about to be performed
1193 * Called just before performing EH actions to clear related bits
1194 * in @link->eh_info such that eh actions are not unnecessarily
1200 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1201 unsigned int action)
1203 struct ata_port *ap = link->ap;
1204 struct ata_eh_info *ehi = &link->eh_info;
1205 struct ata_eh_context *ehc = &link->eh_context;
1206 unsigned long flags;
1208 spin_lock_irqsave(ap->lock, flags);
1210 ata_eh_clear_action(link, dev, ehi, action);
1212 /* About to take EH action, set RECOVERED. Ignore actions on
1213 * slave links as master will do them again.
1215 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1216 ap->pflags |= ATA_PFLAG_RECOVERED;
1218 spin_unlock_irqrestore(ap->lock, flags);
1222 * ata_eh_done - EH action complete
1223 * @ap: target ATA port
1224 * @dev: target ATA dev for per-dev action (can be NULL)
1225 * @action: action just completed
1227 * Called right after performing EH actions to clear related bits
1228 * in @link->eh_context.
1233 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1234 unsigned int action)
1236 struct ata_eh_context *ehc = &link->eh_context;
1238 ata_eh_clear_action(link, dev, &ehc->i, action);
1242 * ata_err_string - convert err_mask to descriptive string
1243 * @err_mask: error mask to convert to string
1245 * Convert @err_mask to descriptive string. Errors are
1246 * prioritized according to severity and only the most severe
1247 * error is reported.
1253 * Descriptive string for @err_mask
1255 static const char *ata_err_string(unsigned int err_mask)
1257 if (err_mask & AC_ERR_HOST_BUS)
1258 return "host bus error";
1259 if (err_mask & AC_ERR_ATA_BUS)
1260 return "ATA bus error";
1261 if (err_mask & AC_ERR_TIMEOUT)
1263 if (err_mask & AC_ERR_HSM)
1264 return "HSM violation";
1265 if (err_mask & AC_ERR_SYSTEM)
1266 return "internal error";
1267 if (err_mask & AC_ERR_MEDIA)
1268 return "media error";
1269 if (err_mask & AC_ERR_INVALID)
1270 return "invalid argument";
1271 if (err_mask & AC_ERR_DEV)
1272 return "device error";
1273 return "unknown error";
1277 * ata_read_log_page - read a specific log page
1278 * @dev: target device
1279 * @page: page to read
1280 * @buf: buffer to store read page
1281 * @sectors: number of sectors to read
1283 * Read log page using READ_LOG_EXT command.
1286 * Kernel thread context (may sleep).
1289 * 0 on success, AC_ERR_* mask otherwise.
1291 static unsigned int ata_read_log_page(struct ata_device *dev,
1292 u8 page, void *buf, unsigned int sectors)
1294 struct ata_taskfile tf;
1295 unsigned int err_mask;
1297 DPRINTK("read log page - page %d\n", page);
1299 ata_tf_init(dev, &tf);
1300 tf.command = ATA_CMD_READ_LOG_EXT;
1303 tf.hob_nsect = sectors >> 8;
1304 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1305 tf.protocol = ATA_PROT_PIO;
1307 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1308 buf, sectors * ATA_SECT_SIZE, 0);
1310 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1315 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1316 * @dev: Device to read log page 10h from
1317 * @tag: Resulting tag of the failed command
1318 * @tf: Resulting taskfile registers of the failed command
1320 * Read log page 10h to obtain NCQ error details and clear error
1324 * Kernel thread context (may sleep).
1327 * 0 on success, -errno otherwise.
1329 static int ata_eh_read_log_10h(struct ata_device *dev,
1330 int *tag, struct ata_taskfile *tf)
1332 u8 *buf = dev->link->ap->sector_buf;
1333 unsigned int err_mask;
1337 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1342 for (i = 0; i < ATA_SECT_SIZE; i++)
1345 ata_dev_printk(dev, KERN_WARNING,
1346 "invalid checksum 0x%x on log page 10h\n", csum);
1351 *tag = buf[0] & 0x1f;
1353 tf->command = buf[2];
1354 tf->feature = buf[3];
1358 tf->device = buf[7];
1359 tf->hob_lbal = buf[8];
1360 tf->hob_lbam = buf[9];
1361 tf->hob_lbah = buf[10];
1362 tf->nsect = buf[12];
1363 tf->hob_nsect = buf[13];
1369 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1370 * @dev: target ATAPI device
1371 * @r_sense_key: out parameter for sense_key
1373 * Perform ATAPI TEST_UNIT_READY.
1376 * EH context (may sleep).
1379 * 0 on success, AC_ERR_* mask on failure.
1381 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1383 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1384 struct ata_taskfile tf;
1385 unsigned int err_mask;
1387 ata_tf_init(dev, &tf);
1389 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1390 tf.command = ATA_CMD_PACKET;
1391 tf.protocol = ATAPI_PROT_NODATA;
1393 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1394 if (err_mask == AC_ERR_DEV)
1395 *r_sense_key = tf.feature >> 4;
1400 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1401 * @dev: device to perform REQUEST_SENSE to
1402 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1403 * @dfl_sense_key: default sense key to use
1405 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1406 * SENSE. This function is EH helper.
1409 * Kernel thread context (may sleep).
1412 * 0 on success, AC_ERR_* mask on failure
1414 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1415 u8 *sense_buf, u8 dfl_sense_key)
1417 u8 cdb[ATAPI_CDB_LEN] =
1418 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1419 struct ata_port *ap = dev->link->ap;
1420 struct ata_taskfile tf;
1422 DPRINTK("ATAPI request sense\n");
1424 /* FIXME: is this needed? */
1425 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1427 /* initialize sense_buf with the error register,
1428 * for the case where they are -not- overwritten
1430 sense_buf[0] = 0x70;
1431 sense_buf[2] = dfl_sense_key;
1433 /* some devices time out if garbage left in tf */
1434 ata_tf_init(dev, &tf);
1436 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1437 tf.command = ATA_CMD_PACKET;
1439 /* is it pointless to prefer PIO for "safety reasons"? */
1440 if (ap->flags & ATA_FLAG_PIO_DMA) {
1441 tf.protocol = ATAPI_PROT_DMA;
1442 tf.feature |= ATAPI_PKT_DMA;
1444 tf.protocol = ATAPI_PROT_PIO;
1445 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1449 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1450 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1454 * ata_eh_analyze_serror - analyze SError for a failed port
1455 * @link: ATA link to analyze SError for
1457 * Analyze SError if available and further determine cause of
1463 static void ata_eh_analyze_serror(struct ata_link *link)
1465 struct ata_eh_context *ehc = &link->eh_context;
1466 u32 serror = ehc->i.serror;
1467 unsigned int err_mask = 0, action = 0;
1470 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1471 err_mask |= AC_ERR_ATA_BUS;
1472 action |= ATA_EH_RESET;
1474 if (serror & SERR_PROTOCOL) {
1475 err_mask |= AC_ERR_HSM;
1476 action |= ATA_EH_RESET;
1478 if (serror & SERR_INTERNAL) {
1479 err_mask |= AC_ERR_SYSTEM;
1480 action |= ATA_EH_RESET;
1483 /* Determine whether a hotplug event has occurred. Both
1484 * SError.N/X are considered hotplug events for enabled or
1485 * host links. For disabled PMP links, only N bit is
1486 * considered as X bit is left at 1 for link plugging.
1490 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1491 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1493 hotplug_mask = SERR_PHYRDY_CHG;
1495 if (serror & hotplug_mask)
1496 ata_ehi_hotplugged(&ehc->i);
1498 ehc->i.err_mask |= err_mask;
1499 ehc->i.action |= action;
1503 * ata_eh_analyze_ncq_error - analyze NCQ error
1504 * @link: ATA link to analyze NCQ error for
1506 * Read log page 10h, determine the offending qc and acquire
1507 * error status TF. For NCQ device errors, all LLDDs have to do
1508 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1512 * Kernel thread context (may sleep).
1514 void ata_eh_analyze_ncq_error(struct ata_link *link)
1516 struct ata_port *ap = link->ap;
1517 struct ata_eh_context *ehc = &link->eh_context;
1518 struct ata_device *dev = link->device;
1519 struct ata_queued_cmd *qc;
1520 struct ata_taskfile tf;
1523 /* if frozen, we can't do much */
1524 if (ap->pflags & ATA_PFLAG_FROZEN)
1527 /* is it NCQ device error? */
1528 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1531 /* has LLDD analyzed already? */
1532 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1533 qc = __ata_qc_from_tag(ap, tag);
1535 if (!(qc->flags & ATA_QCFLAG_FAILED))
1542 /* okay, this error is ours */
1543 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1545 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1546 "(errno=%d)\n", rc);
1550 if (!(link->sactive & (1 << tag))) {
1551 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1552 "inactive tag %d\n", tag);
1556 /* we've got the perpetrator, condemn it */
1557 qc = __ata_qc_from_tag(ap, tag);
1558 memcpy(&qc->result_tf, &tf, sizeof(tf));
1559 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1560 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1561 ehc->i.err_mask &= ~AC_ERR_DEV;
1565 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1566 * @qc: qc to analyze
1567 * @tf: Taskfile registers to analyze
1569 * Analyze taskfile of @qc and further determine cause of
1570 * failure. This function also requests ATAPI sense data if
1574 * Kernel thread context (may sleep).
1577 * Determined recovery action
1579 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1580 const struct ata_taskfile *tf)
1582 unsigned int tmp, action = 0;
1583 u8 stat = tf->command, err = tf->feature;
1585 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1586 qc->err_mask |= AC_ERR_HSM;
1587 return ATA_EH_RESET;
1590 if (stat & (ATA_ERR | ATA_DF))
1591 qc->err_mask |= AC_ERR_DEV;
1595 switch (qc->dev->class) {
1598 qc->err_mask |= AC_ERR_ATA_BUS;
1600 qc->err_mask |= AC_ERR_MEDIA;
1602 qc->err_mask |= AC_ERR_INVALID;
1606 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1607 tmp = atapi_eh_request_sense(qc->dev,
1608 qc->scsicmd->sense_buffer,
1609 qc->result_tf.feature >> 4);
1611 /* ATA_QCFLAG_SENSE_VALID is used to
1612 * tell atapi_qc_complete() that sense
1613 * data is already valid.
1615 * TODO: interpret sense data and set
1616 * appropriate err_mask.
1618 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1620 qc->err_mask |= tmp;
1624 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1625 action |= ATA_EH_RESET;
1630 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1635 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1639 base = ATA_ECAT_DUBIOUS_NONE;
1641 if (err_mask & AC_ERR_ATA_BUS)
1642 return base + ATA_ECAT_ATA_BUS;
1644 if (err_mask & AC_ERR_TIMEOUT)
1645 return base + ATA_ECAT_TOUT_HSM;
1647 if (eflags & ATA_EFLAG_IS_IO) {
1648 if (err_mask & AC_ERR_HSM)
1649 return base + ATA_ECAT_TOUT_HSM;
1651 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1652 return base + ATA_ECAT_UNK_DEV;
1658 struct speed_down_verdict_arg {
1661 int nr_errors[ATA_ECAT_NR];
1664 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1666 struct speed_down_verdict_arg *arg = void_arg;
1669 if (ent->timestamp < arg->since)
1672 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1674 arg->nr_errors[cat]++;
1680 * ata_eh_speed_down_verdict - Determine speed down verdict
1681 * @dev: Device of interest
1683 * This function examines error ring of @dev and determines
1684 * whether NCQ needs to be turned off, transfer speed should be
1685 * stepped down, or falling back to PIO is necessary.
1687 * ECAT_ATA_BUS : ATA_BUS error for any command
1689 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1692 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1694 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1695 * data transfer hasn't been verified.
1699 * NCQ_OFF : Turn off NCQ.
1701 * SPEED_DOWN : Speed down transfer speed but don't fall back
1704 * FALLBACK_TO_PIO : Fall back to PIO.
1706 * Even if multiple verdicts are returned, only one action is
1707 * taken per error. An action triggered by non-DUBIOUS errors
1708 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1709 * This is to expedite speed down decisions right after device is
1710 * initially configured.
1712 * The followings are speed down rules. #1 and #2 deal with
1715 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1716 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1718 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1719 * occurred during last 5 mins, NCQ_OFF.
1721 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1722 * ocurred during last 5 mins, FALLBACK_TO_PIO
1724 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1725 * during last 10 mins, NCQ_OFF.
1727 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1728 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1731 * Inherited from caller.
1734 * OR of ATA_EH_SPDN_* flags.
1736 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1738 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1739 u64 j64 = get_jiffies_64();
1740 struct speed_down_verdict_arg arg;
1741 unsigned int verdict = 0;
1743 /* scan past 5 mins of error history */
1744 memset(&arg, 0, sizeof(arg));
1745 arg.since = j64 - min(j64, j5mins);
1746 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1748 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1749 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1750 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1751 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1753 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1754 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1755 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1757 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1758 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1759 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1760 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1762 /* scan past 10 mins of error history */
1763 memset(&arg, 0, sizeof(arg));
1764 arg.since = j64 - min(j64, j10mins);
1765 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1767 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1768 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1769 verdict |= ATA_EH_SPDN_NCQ_OFF;
1771 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1772 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1773 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1774 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1780 * ata_eh_speed_down - record error and speed down if necessary
1781 * @dev: Failed device
1782 * @eflags: mask of ATA_EFLAG_* flags
1783 * @err_mask: err_mask of the error
1785 * Record error and examine error history to determine whether
1786 * adjusting transmission speed is necessary. It also sets
1787 * transmission limits appropriately if such adjustment is
1791 * Kernel thread context (may sleep).
1794 * Determined recovery action.
1796 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1797 unsigned int eflags, unsigned int err_mask)
1799 struct ata_link *link = ata_dev_phys_link(dev);
1801 unsigned int verdict;
1802 unsigned int action = 0;
1804 /* don't bother if Cat-0 error */
1805 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1808 /* record error and determine whether speed down is necessary */
1809 ata_ering_record(&dev->ering, eflags, err_mask);
1810 verdict = ata_eh_speed_down_verdict(dev);
1813 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1814 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1815 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1816 dev->flags |= ATA_DFLAG_NCQ_OFF;
1817 ata_dev_printk(dev, KERN_WARNING,
1818 "NCQ disabled due to excessive errors\n");
1823 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1824 /* speed down SATA link speed if possible */
1825 if (sata_down_spd_limit(link) == 0) {
1826 action |= ATA_EH_RESET;
1830 /* lower transfer mode */
1831 if (dev->spdn_cnt < 2) {
1832 static const int dma_dnxfer_sel[] =
1833 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1834 static const int pio_dnxfer_sel[] =
1835 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1838 if (dev->xfer_shift != ATA_SHIFT_PIO)
1839 sel = dma_dnxfer_sel[dev->spdn_cnt];
1841 sel = pio_dnxfer_sel[dev->spdn_cnt];
1845 if (ata_down_xfermask_limit(dev, sel) == 0) {
1846 action |= ATA_EH_RESET;
1852 /* Fall back to PIO? Slowing down to PIO is meaningless for
1853 * SATA ATA devices. Consider it only for PATA and SATAPI.
1855 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1856 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1857 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1858 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1860 action |= ATA_EH_RESET;
1867 /* device has been slowed down, blow error history */
1868 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1869 ata_ering_clear(&dev->ering);
1874 * ata_eh_link_autopsy - analyze error and determine recovery action
1875 * @link: host link to perform autopsy on
1877 * Analyze why @link failed and determine which recovery actions
1878 * are needed. This function also sets more detailed AC_ERR_*
1879 * values and fills sense data for ATAPI CHECK SENSE.
1882 * Kernel thread context (may sleep).
1884 static void ata_eh_link_autopsy(struct ata_link *link)
1886 struct ata_port *ap = link->ap;
1887 struct ata_eh_context *ehc = &link->eh_context;
1888 struct ata_device *dev;
1889 unsigned int all_err_mask = 0, eflags = 0;
1896 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1899 /* obtain and analyze SError */
1900 rc = sata_scr_read(link, SCR_ERROR, &serror);
1902 ehc->i.serror |= serror;
1903 ata_eh_analyze_serror(link);
1904 } else if (rc != -EOPNOTSUPP) {
1905 /* SError read failed, force reset and probing */
1906 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1907 ehc->i.action |= ATA_EH_RESET;
1908 ehc->i.err_mask |= AC_ERR_OTHER;
1911 /* analyze NCQ failure */
1912 ata_eh_analyze_ncq_error(link);
1914 /* any real error trumps AC_ERR_OTHER */
1915 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1916 ehc->i.err_mask &= ~AC_ERR_OTHER;
1918 all_err_mask |= ehc->i.err_mask;
1920 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1921 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1923 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1924 ata_dev_phys_link(qc->dev) != link)
1927 /* inherit upper level err_mask */
1928 qc->err_mask |= ehc->i.err_mask;
1931 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1933 /* DEV errors are probably spurious in case of ATA_BUS error */
1934 if (qc->err_mask & AC_ERR_ATA_BUS)
1935 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1938 /* any real error trumps unknown error */
1939 if (qc->err_mask & ~AC_ERR_OTHER)
1940 qc->err_mask &= ~AC_ERR_OTHER;
1942 /* SENSE_VALID trumps dev/unknown error and revalidation */
1943 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1944 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1946 /* determine whether the command is worth retrying */
1947 if (!(qc->err_mask & AC_ERR_INVALID) &&
1948 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1949 qc->flags |= ATA_QCFLAG_RETRY;
1951 /* accumulate error info */
1952 ehc->i.dev = qc->dev;
1953 all_err_mask |= qc->err_mask;
1954 if (qc->flags & ATA_QCFLAG_IO)
1955 eflags |= ATA_EFLAG_IS_IO;
1958 /* enforce default EH actions */
1959 if (ap->pflags & ATA_PFLAG_FROZEN ||
1960 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1961 ehc->i.action |= ATA_EH_RESET;
1962 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1963 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
1964 ehc->i.action |= ATA_EH_REVALIDATE;
1966 /* If we have offending qcs and the associated failed device,
1967 * perform per-dev EH action only on the offending device.
1970 ehc->i.dev_action[ehc->i.dev->devno] |=
1971 ehc->i.action & ATA_EH_PERDEV_MASK;
1972 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1975 /* propagate timeout to host link */
1976 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1977 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1979 /* record error and consider speeding down */
1981 if (!dev && ((ata_link_max_devices(link) == 1 &&
1982 ata_dev_enabled(link->device))))
1986 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1987 eflags |= ATA_EFLAG_DUBIOUS_XFER;
1988 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
1995 * ata_eh_autopsy - analyze error and determine recovery action
1996 * @ap: host port to perform autopsy on
1998 * Analyze all links of @ap and determine why they failed and
1999 * which recovery actions are needed.
2002 * Kernel thread context (may sleep).
2004 void ata_eh_autopsy(struct ata_port *ap)
2006 struct ata_link *link;
2008 ata_port_for_each_link(link, ap)
2009 ata_eh_link_autopsy(link);
2011 /* Handle the frigging slave link. Autopsy is done similarly
2012 * but actions and flags are transferred over to the master
2013 * link and handled from there.
2015 if (ap->slave_link) {
2016 struct ata_eh_context *mehc = &ap->link.eh_context;
2017 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2019 /* transfer control flags from master to slave */
2020 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2022 /* perform autopsy on the slave link */
2023 ata_eh_link_autopsy(ap->slave_link);
2025 /* transfer actions from slave to master and clear slave */
2026 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2027 mehc->i.action |= sehc->i.action;
2028 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2029 mehc->i.flags |= sehc->i.flags;
2030 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2033 /* Autopsy of fanout ports can affect host link autopsy.
2034 * Perform host link autopsy last.
2036 if (sata_pmp_attached(ap))
2037 ata_eh_link_autopsy(&ap->link);
2041 * ata_eh_link_report - report error handling to user
2042 * @link: ATA link EH is going on
2044 * Report EH to user.
2049 static void ata_eh_link_report(struct ata_link *link)
2051 struct ata_port *ap = link->ap;
2052 struct ata_eh_context *ehc = &link->eh_context;
2053 const char *frozen, *desc;
2055 int tag, nr_failed = 0;
2057 if (ehc->i.flags & ATA_EHI_QUIET)
2061 if (ehc->i.desc[0] != '\0')
2064 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2065 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2067 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2068 ata_dev_phys_link(qc->dev) != link ||
2069 ((qc->flags & ATA_QCFLAG_QUIET) &&
2070 qc->err_mask == AC_ERR_DEV))
2072 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2078 if (!nr_failed && !ehc->i.err_mask)
2082 if (ap->pflags & ATA_PFLAG_FROZEN)
2085 memset(tries_buf, 0, sizeof(tries_buf));
2086 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2087 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2091 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2092 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2093 ehc->i.err_mask, link->sactive, ehc->i.serror,
2094 ehc->i.action, frozen, tries_buf);
2096 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2098 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2099 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2100 ehc->i.err_mask, link->sactive, ehc->i.serror,
2101 ehc->i.action, frozen, tries_buf);
2103 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2107 ata_link_printk(link, KERN_ERR,
2108 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2109 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2110 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2111 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2112 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2113 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2114 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2115 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2116 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2117 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2118 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2119 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2120 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2121 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2122 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2123 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2124 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2125 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2127 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2128 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2129 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2130 const u8 *cdb = qc->cdb;
2131 char data_buf[20] = "";
2132 char cdb_buf[70] = "";
2134 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2135 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2138 if (qc->dma_dir != DMA_NONE) {
2139 static const char *dma_str[] = {
2140 [DMA_BIDIRECTIONAL] = "bidi",
2141 [DMA_TO_DEVICE] = "out",
2142 [DMA_FROM_DEVICE] = "in",
2144 static const char *prot_str[] = {
2145 [ATA_PROT_PIO] = "pio",
2146 [ATA_PROT_DMA] = "dma",
2147 [ATA_PROT_NCQ] = "ncq",
2148 [ATAPI_PROT_PIO] = "pio",
2149 [ATAPI_PROT_DMA] = "dma",
2152 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2153 prot_str[qc->tf.protocol], qc->nbytes,
2154 dma_str[qc->dma_dir]);
2157 if (ata_is_atapi(qc->tf.protocol))
2158 snprintf(cdb_buf, sizeof(cdb_buf),
2159 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2160 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2161 cdb[0], cdb[1], cdb[2], cdb[3],
2162 cdb[4], cdb[5], cdb[6], cdb[7],
2163 cdb[8], cdb[9], cdb[10], cdb[11],
2164 cdb[12], cdb[13], cdb[14], cdb[15]);
2166 ata_dev_printk(qc->dev, KERN_ERR,
2167 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2169 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2170 "Emask 0x%x (%s)%s\n",
2171 cmd->command, cmd->feature, cmd->nsect,
2172 cmd->lbal, cmd->lbam, cmd->lbah,
2173 cmd->hob_feature, cmd->hob_nsect,
2174 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2175 cmd->device, qc->tag, data_buf, cdb_buf,
2176 res->command, res->feature, res->nsect,
2177 res->lbal, res->lbam, res->lbah,
2178 res->hob_feature, res->hob_nsect,
2179 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2180 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2181 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2183 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2185 if (res->command & ATA_BUSY)
2186 ata_dev_printk(qc->dev, KERN_ERR,
2187 "status: { Busy }\n");
2189 ata_dev_printk(qc->dev, KERN_ERR,
2190 "status: { %s%s%s%s}\n",
2191 res->command & ATA_DRDY ? "DRDY " : "",
2192 res->command & ATA_DF ? "DF " : "",
2193 res->command & ATA_DRQ ? "DRQ " : "",
2194 res->command & ATA_ERR ? "ERR " : "");
2197 if (cmd->command != ATA_CMD_PACKET &&
2198 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2200 ata_dev_printk(qc->dev, KERN_ERR,
2201 "error: { %s%s%s%s}\n",
2202 res->feature & ATA_ICRC ? "ICRC " : "",
2203 res->feature & ATA_UNC ? "UNC " : "",
2204 res->feature & ATA_IDNF ? "IDNF " : "",
2205 res->feature & ATA_ABORTED ? "ABRT " : "");
2210 * ata_eh_report - report error handling to user
2211 * @ap: ATA port to report EH about
2213 * Report EH to user.
2218 void ata_eh_report(struct ata_port *ap)
2220 struct ata_link *link;
2222 __ata_port_for_each_link(link, ap)
2223 ata_eh_link_report(link);
2226 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2227 unsigned int *classes, unsigned long deadline,
2230 struct ata_device *dev;
2233 ata_link_for_each_dev(dev, link)
2234 classes[dev->devno] = ATA_DEV_UNKNOWN;
2236 return reset(link, classes, deadline);
2239 static int ata_eh_followup_srst_needed(struct ata_link *link,
2240 int rc, const unsigned int *classes)
2242 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2246 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2251 int ata_eh_reset(struct ata_link *link, int classify,
2252 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2253 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2255 struct ata_port *ap = link->ap;
2256 struct ata_link *slave = ap->slave_link;
2257 struct ata_eh_context *ehc = &link->eh_context;
2258 struct ata_eh_context *sehc = &slave->eh_context;
2259 unsigned int *classes = ehc->classes;
2260 unsigned int lflags = link->flags;
2261 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2262 int max_tries = 0, try = 0;
2263 struct ata_link *failed_link;
2264 struct ata_device *dev;
2265 unsigned long deadline, now;
2266 ata_reset_fn_t reset;
2267 unsigned long flags;
2274 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2276 if (link->flags & ATA_LFLAG_NO_HRST)
2278 if (link->flags & ATA_LFLAG_NO_SRST)
2282 deadline = ata_deadline(ehc->last_reset, ATA_EH_RESET_COOL_DOWN);
2283 if (time_before(now, deadline))
2284 schedule_timeout_uninterruptible(deadline - now);
2286 spin_lock_irqsave(ap->lock, flags);
2287 ap->pflags |= ATA_PFLAG_RESETTING;
2288 spin_unlock_irqrestore(ap->lock, flags);
2290 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2291 ehc->last_reset = jiffies;
2293 ata_link_for_each_dev(dev, link) {
2294 /* If we issue an SRST then an ATA drive (not ATAPI)
2295 * may change configuration and be in PIO0 timing. If
2296 * we do a hard reset (or are coming from power on)
2297 * this is true for ATA or ATAPI. Until we've set a
2298 * suitable controller mode we should not touch the
2299 * bus as we may be talking too fast.
2301 dev->pio_mode = XFER_PIO_0;
2303 /* If the controller has a pio mode setup function
2304 * then use it to set the chipset to rights. Don't
2305 * touch the DMA setup as that will be dealt with when
2306 * configuring devices.
2308 if (ap->ops->set_piomode)
2309 ap->ops->set_piomode(ap, dev);
2312 /* prefer hardreset */
2314 ehc->i.action &= ~ATA_EH_RESET;
2317 ehc->i.action |= ATA_EH_HARDRESET;
2318 } else if (softreset) {
2320 ehc->i.action |= ATA_EH_SOFTRESET;
2324 unsigned long deadline = ata_deadline(jiffies,
2325 ATA_EH_PRERESET_TIMEOUT);
2328 sehc->i.action &= ~ATA_EH_RESET;
2329 sehc->i.action |= ehc->i.action;
2332 rc = prereset(link, deadline);
2334 /* If present, do prereset on slave link too. Reset
2335 * is skipped iff both master and slave links report
2336 * -ENOENT or clear ATA_EH_RESET.
2338 if (slave && (rc == 0 || rc == -ENOENT)) {
2341 tmp = prereset(slave, deadline);
2345 ehc->i.action |= sehc->i.action;
2349 if (rc == -ENOENT) {
2350 ata_link_printk(link, KERN_DEBUG,
2351 "port disabled. ignoring.\n");
2352 ehc->i.action &= ~ATA_EH_RESET;
2354 ata_link_for_each_dev(dev, link)
2355 classes[dev->devno] = ATA_DEV_NONE;
2359 ata_link_printk(link, KERN_ERR,
2360 "prereset failed (errno=%d)\n", rc);
2364 /* prereset() might have cleared ATA_EH_RESET. If so,
2365 * bang classes and return.
2367 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2368 ata_link_for_each_dev(dev, link)
2369 classes[dev->devno] = ATA_DEV_NONE;
2379 ehc->last_reset = jiffies;
2380 if (ata_is_host_link(link))
2381 ata_eh_freeze_port(ap);
2383 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2387 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2388 reset == softreset ? "soft" : "hard");
2390 /* mark that this EH session started with reset */
2391 if (reset == hardreset)
2392 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2394 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2396 rc = ata_do_reset(link, reset, classes, deadline, true);
2397 if (rc && rc != -EAGAIN) {
2402 /* hardreset slave link if existent */
2403 if (slave && reset == hardreset) {
2407 ata_link_printk(slave, KERN_INFO,
2408 "hard resetting link\n");
2410 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2411 tmp = ata_do_reset(slave, reset, classes, deadline,
2419 failed_link = slave;
2425 /* perform follow-up SRST if necessary */
2426 if (reset == hardreset &&
2427 ata_eh_followup_srst_needed(link, rc, classes)) {
2431 ata_link_printk(link, KERN_ERR,
2432 "follow-up softreset required "
2433 "but no softreset avaliable\n");
2439 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2440 rc = ata_do_reset(link, reset, classes, deadline, true);
2444 ata_link_printk(link, KERN_INFO, "no reset method "
2445 "available, skipping reset\n");
2446 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2447 lflags |= ATA_LFLAG_ASSUME_ATA;
2451 * Post-reset processing
2453 ata_link_for_each_dev(dev, link) {
2454 /* After the reset, the device state is PIO 0 and the
2455 * controller state is undefined. Reset also wakes up
2456 * drives from sleeping mode.
2458 dev->pio_mode = XFER_PIO_0;
2459 dev->flags &= ~ATA_DFLAG_SLEEPING;
2461 if (!ata_phys_link_offline(ata_dev_phys_link(dev))) {
2462 /* apply class override */
2463 if (lflags & ATA_LFLAG_ASSUME_ATA)
2464 classes[dev->devno] = ATA_DEV_ATA;
2465 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2466 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2468 classes[dev->devno] = ATA_DEV_NONE;
2471 /* record current link speed */
2472 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2473 link->sata_spd = (sstatus >> 4) & 0xf;
2474 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2475 slave->sata_spd = (sstatus >> 4) & 0xf;
2478 if (ata_is_host_link(link))
2479 ata_eh_thaw_port(ap);
2481 /* postreset() should clear hardware SError. Although SError
2482 * is cleared during link resume, clearing SError here is
2483 * necessary as some PHYs raise hotplug events after SRST.
2484 * This introduces race condition where hotplug occurs between
2485 * reset and here. This race is mediated by cross checking
2486 * link onlineness and classification result later.
2489 postreset(link, classes);
2491 postreset(slave, classes);
2494 /* clear cached SError */
2495 spin_lock_irqsave(link->ap->lock, flags);
2496 link->eh_info.serror = 0;
2498 slave->eh_info.serror = 0;
2499 spin_unlock_irqrestore(link->ap->lock, flags);
2501 /* Make sure onlineness and classification result correspond.
2502 * Hotplug could have happened during reset and some
2503 * controllers fail to wait while a drive is spinning up after
2504 * being hotplugged causing misdetection. By cross checking
2505 * link onlineness and classification result, those conditions
2506 * can be reliably detected and retried.
2509 ata_link_for_each_dev(dev, link) {
2510 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2511 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2512 classes[dev->devno] = ATA_DEV_NONE;
2513 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2518 if (classify && nr_unknown) {
2519 if (try < max_tries) {
2520 ata_link_printk(link, KERN_WARNING, "link online but "
2521 "device misclassified, retrying\n");
2526 ata_link_printk(link, KERN_WARNING,
2527 "link online but device misclassified, "
2528 "device detection might fail\n");
2531 /* reset successful, schedule revalidation */
2532 ata_eh_done(link, NULL, ATA_EH_RESET);
2534 ata_eh_done(slave, NULL, ATA_EH_RESET);
2535 ehc->last_reset = jiffies;
2536 ehc->i.action |= ATA_EH_REVALIDATE;
2540 /* clear hotplug flag */
2541 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2543 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2545 spin_lock_irqsave(ap->lock, flags);
2546 ap->pflags &= ~ATA_PFLAG_RESETTING;
2547 spin_unlock_irqrestore(ap->lock, flags);
2552 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2553 if (!ata_is_host_link(link) &&
2554 sata_scr_read(link, SCR_STATUS, &sstatus))
2557 if (rc == -ERESTART || try >= max_tries)
2561 if (time_before(now, deadline)) {
2562 unsigned long delta = deadline - now;
2564 ata_link_printk(failed_link, KERN_WARNING,
2565 "reset failed (errno=%d), retrying in %u secs\n",
2566 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2569 delta = schedule_timeout_uninterruptible(delta);
2572 if (try == max_tries - 1) {
2573 sata_down_spd_limit(link);
2575 sata_down_spd_limit(slave);
2576 } else if (rc == -EPIPE)
2577 sata_down_spd_limit(failed_link);
2584 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2586 struct ata_link *link;
2587 struct ata_device *dev;
2588 unsigned long flags;
2591 * This function can be thought of as an extended version of
2592 * ata_eh_about_to_do() specially crafted to accommodate the
2593 * requirements of ATA_EH_PARK handling. Since the EH thread
2594 * does not leave the do {} while () loop in ata_eh_recover as
2595 * long as the timeout for a park request to *one* device on
2596 * the port has not expired, and since we still want to pick
2597 * up park requests to other devices on the same port or
2598 * timeout updates for the same device, we have to pull
2599 * ATA_EH_PARK actions from eh_info into eh_context.i
2600 * ourselves at the beginning of each pass over the loop.
2602 * Additionally, all write accesses to &ap->park_req_pending
2603 * through INIT_COMPLETION() (see below) or complete_all()
2604 * (see ata_scsi_park_store()) are protected by the host lock.
2605 * As a result we have that park_req_pending.done is zero on
2606 * exit from this function, i.e. when ATA_EH_PARK actions for
2607 * *all* devices on port ap have been pulled into the
2608 * respective eh_context structs. If, and only if,
2609 * park_req_pending.done is non-zero by the time we reach
2610 * wait_for_completion_timeout(), another ATA_EH_PARK action
2611 * has been scheduled for at least one of the devices on port
2612 * ap and we have to cycle over the do {} while () loop in
2613 * ata_eh_recover() again.
2616 spin_lock_irqsave(ap->lock, flags);
2617 INIT_COMPLETION(ap->park_req_pending);
2618 ata_port_for_each_link(link, ap) {
2619 ata_link_for_each_dev(dev, link) {
2620 struct ata_eh_info *ehi = &link->eh_info;
2622 link->eh_context.i.dev_action[dev->devno] |=
2623 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2624 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2627 spin_unlock_irqrestore(ap->lock, flags);
2630 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2632 struct ata_eh_context *ehc = &dev->link->eh_context;
2633 struct ata_taskfile tf;
2634 unsigned int err_mask;
2636 ata_tf_init(dev, &tf);
2638 ehc->unloaded_mask |= 1 << dev->devno;
2639 tf.command = ATA_CMD_IDLEIMMEDIATE;
2645 ehc->unloaded_mask &= ~(1 << dev->devno);
2646 tf.command = ATA_CMD_CHK_POWER;
2649 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2650 tf.protocol |= ATA_PROT_NODATA;
2651 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2652 if (park && (err_mask || tf.lbal != 0xc4)) {
2653 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2654 ehc->unloaded_mask &= ~(1 << dev->devno);
2658 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2659 struct ata_device **r_failed_dev)
2661 struct ata_port *ap = link->ap;
2662 struct ata_eh_context *ehc = &link->eh_context;
2663 struct ata_device *dev;
2664 unsigned int new_mask = 0;
2665 unsigned long flags;
2670 /* For PATA drive side cable detection to work, IDENTIFY must
2671 * be done backwards such that PDIAG- is released by the slave
2672 * device before the master device is identified.
2674 ata_link_for_each_dev_reverse(dev, link) {
2675 unsigned int action = ata_eh_dev_action(dev);
2676 unsigned int readid_flags = 0;
2678 if (ehc->i.flags & ATA_EHI_DID_RESET)
2679 readid_flags |= ATA_READID_POSTRESET;
2681 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2682 WARN_ON(dev->class == ATA_DEV_PMP);
2684 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2689 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2690 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2695 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2697 /* Configuration may have changed, reconfigure
2700 ehc->i.flags |= ATA_EHI_SETMODE;
2702 /* schedule the scsi_rescan_device() here */
2703 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2704 } else if (dev->class == ATA_DEV_UNKNOWN &&
2705 ehc->tries[dev->devno] &&
2706 ata_class_enabled(ehc->classes[dev->devno])) {
2707 dev->class = ehc->classes[dev->devno];
2709 if (dev->class == ATA_DEV_PMP)
2710 rc = sata_pmp_attach(dev);
2712 rc = ata_dev_read_id(dev, &dev->class,
2713 readid_flags, dev->id);
2716 new_mask |= 1 << dev->devno;
2719 /* IDENTIFY was issued to non-existent
2720 * device. No need to reset. Just
2721 * thaw and kill the device.
2723 ata_eh_thaw_port(ap);
2724 dev->class = ATA_DEV_UNKNOWN;
2727 dev->class = ATA_DEV_UNKNOWN;
2733 /* PDIAG- should have been released, ask cable type if post-reset */
2734 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2735 if (ap->ops->cable_detect)
2736 ap->cbl = ap->ops->cable_detect(ap);
2740 /* Configure new devices forward such that user doesn't see
2741 * device detection messages backwards.
2743 ata_link_for_each_dev(dev, link) {
2744 if (!(new_mask & (1 << dev->devno)) ||
2745 dev->class == ATA_DEV_PMP)
2748 ehc->i.flags |= ATA_EHI_PRINTINFO;
2749 rc = ata_dev_configure(dev);
2750 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2754 spin_lock_irqsave(ap->lock, flags);
2755 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2756 spin_unlock_irqrestore(ap->lock, flags);
2758 /* new device discovered, configure xfermode */
2759 ehc->i.flags |= ATA_EHI_SETMODE;
2765 *r_failed_dev = dev;
2766 DPRINTK("EXIT rc=%d\n", rc);
2771 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2772 * @link: link on which timings will be programmed
2773 * @r_failed_dev: out paramter for failed device
2775 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2776 * ata_set_mode() fails, pointer to the failing device is
2777 * returned in @r_failed_dev.
2780 * PCI/etc. bus probe sem.
2783 * 0 on success, negative errno otherwise
2785 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2787 struct ata_port *ap = link->ap;
2788 struct ata_device *dev;
2791 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2792 ata_link_for_each_dev(dev, link) {
2793 if (!ata_dev_enabled(dev))
2796 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2797 struct ata_ering_entry *ent;
2799 ent = ata_ering_top(&dev->ering);
2801 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2805 /* has private set_mode? */
2806 if (ap->ops->set_mode)
2807 rc = ap->ops->set_mode(link, r_failed_dev);
2809 rc = ata_do_set_mode(link, r_failed_dev);
2811 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2812 ata_link_for_each_dev(dev, link) {
2813 struct ata_eh_context *ehc = &link->eh_context;
2814 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2815 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2817 if (!ata_dev_enabled(dev))
2820 if (dev->xfer_mode != saved_xfer_mode ||
2821 ata_ncq_enabled(dev) != saved_ncq)
2822 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2829 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2830 * @dev: ATAPI device to clear UA for
2832 * Resets and other operations can make an ATAPI device raise
2833 * UNIT ATTENTION which causes the next operation to fail. This
2834 * function clears UA.
2837 * EH context (may sleep).
2840 * 0 on success, -errno on failure.
2842 static int atapi_eh_clear_ua(struct ata_device *dev)
2846 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
2847 u8 sense_buffer[SCSI_SENSE_BUFFERSIZE];
2849 unsigned int err_mask;
2851 err_mask = atapi_eh_tur(dev, &sense_key);
2852 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
2853 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
2854 "failed (err_mask=0x%x)\n", err_mask);
2858 if (!err_mask || sense_key != UNIT_ATTENTION)
2861 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
2863 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
2864 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
2869 ata_dev_printk(dev, KERN_WARNING,
2870 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
2875 static int ata_link_nr_enabled(struct ata_link *link)
2877 struct ata_device *dev;
2880 ata_link_for_each_dev(dev, link)
2881 if (ata_dev_enabled(dev))
2886 static int ata_link_nr_vacant(struct ata_link *link)
2888 struct ata_device *dev;
2891 ata_link_for_each_dev(dev, link)
2892 if (dev->class == ATA_DEV_UNKNOWN)
2897 static int ata_eh_skip_recovery(struct ata_link *link)
2899 struct ata_port *ap = link->ap;
2900 struct ata_eh_context *ehc = &link->eh_context;
2901 struct ata_device *dev;
2903 /* skip disabled links */
2904 if (link->flags & ATA_LFLAG_DISABLED)
2907 /* thaw frozen port and recover failed devices */
2908 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2911 /* reset at least once if reset is requested */
2912 if ((ehc->i.action & ATA_EH_RESET) &&
2913 !(ehc->i.flags & ATA_EHI_DID_RESET))
2916 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2917 ata_link_for_each_dev(dev, link) {
2918 if (dev->class == ATA_DEV_UNKNOWN &&
2919 ehc->classes[dev->devno] != ATA_DEV_NONE)
2926 static int ata_eh_schedule_probe(struct ata_device *dev)
2928 struct ata_eh_context *ehc = &dev->link->eh_context;
2930 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2931 (ehc->did_probe_mask & (1 << dev->devno)))
2934 ata_eh_detach_dev(dev);
2936 ehc->did_probe_mask |= (1 << dev->devno);
2937 ehc->i.action |= ATA_EH_RESET;
2938 ehc->saved_xfer_mode[dev->devno] = 0;
2939 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
2944 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2946 struct ata_eh_context *ehc = &dev->link->eh_context;
2948 ehc->tries[dev->devno]--;
2952 /* device missing or wrong IDENTIFY data, schedule probing */
2953 ehc->i.probe_mask |= (1 << dev->devno);
2955 /* give it just one more chance */
2956 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2958 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
2959 /* This is the last chance, better to slow
2960 * down than lose it.
2962 sata_down_spd_limit(ata_dev_phys_link(dev));
2963 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2967 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2968 /* disable device if it has used up all its chances */
2969 ata_dev_disable(dev);
2971 /* detach if offline */
2972 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2973 ata_eh_detach_dev(dev);
2975 /* schedule probe if necessary */
2976 if (ata_eh_schedule_probe(dev)) {
2977 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2978 memset(ehc->cmd_timeout_idx[dev->devno], 0,
2979 sizeof(ehc->cmd_timeout_idx[dev->devno]));
2984 ehc->i.action |= ATA_EH_RESET;
2990 * ata_eh_recover - recover host port after error
2991 * @ap: host port to recover
2992 * @prereset: prereset method (can be NULL)
2993 * @softreset: softreset method (can be NULL)
2994 * @hardreset: hardreset method (can be NULL)
2995 * @postreset: postreset method (can be NULL)
2996 * @r_failed_link: out parameter for failed link
2998 * This is the alpha and omega, eum and yang, heart and soul of
2999 * libata exception handling. On entry, actions required to
3000 * recover each link and hotplug requests are recorded in the
3001 * link's eh_context. This function executes all the operations
3002 * with appropriate retrials and fallbacks to resurrect failed
3003 * devices, detach goners and greet newcomers.
3006 * Kernel thread context (may sleep).
3009 * 0 on success, -errno on failure.
3011 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3012 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3013 ata_postreset_fn_t postreset,
3014 struct ata_link **r_failed_link)
3016 struct ata_link *link;
3017 struct ata_device *dev;
3020 unsigned long flags, deadline;
3024 /* prep for recovery */
3025 ata_port_for_each_link(link, ap) {
3026 struct ata_eh_context *ehc = &link->eh_context;
3028 /* re-enable link? */
3029 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3030 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3031 spin_lock_irqsave(ap->lock, flags);
3032 link->flags &= ~ATA_LFLAG_DISABLED;
3033 spin_unlock_irqrestore(ap->lock, flags);
3034 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3037 ata_link_for_each_dev(dev, link) {
3038 if (link->flags & ATA_LFLAG_NO_RETRY)
3039 ehc->tries[dev->devno] = 1;
3041 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3043 /* collect port action mask recorded in dev actions */
3044 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3045 ~ATA_EH_PERDEV_MASK;
3046 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3048 /* process hotplug request */
3049 if (dev->flags & ATA_DFLAG_DETACH)
3050 ata_eh_detach_dev(dev);
3052 /* schedule probe if necessary */
3053 if (!ata_dev_enabled(dev))
3054 ata_eh_schedule_probe(dev);
3062 /* if UNLOADING, finish immediately */
3063 if (ap->pflags & ATA_PFLAG_UNLOADING)
3067 ata_port_for_each_link(link, ap) {
3068 struct ata_eh_context *ehc = &link->eh_context;
3070 /* skip EH if possible. */
3071 if (ata_eh_skip_recovery(link))
3074 ata_link_for_each_dev(dev, link)
3075 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3079 ata_port_for_each_link(link, ap) {
3080 struct ata_eh_context *ehc = &link->eh_context;
3082 if (!(ehc->i.action & ATA_EH_RESET))
3085 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3086 prereset, softreset, hardreset, postreset);
3088 ata_link_printk(link, KERN_ERR,
3089 "reset failed, giving up\n");
3098 * clears ATA_EH_PARK in eh_info and resets
3099 * ap->park_req_pending
3101 ata_eh_pull_park_action(ap);
3104 ata_port_for_each_link(link, ap) {
3105 ata_link_for_each_dev(dev, link) {
3106 struct ata_eh_context *ehc = &link->eh_context;
3109 if (dev->class != ATA_DEV_ATA)
3111 if (!(ehc->i.dev_action[dev->devno] &
3114 tmp = dev->unpark_deadline;
3115 if (time_before(deadline, tmp))
3117 else if (time_before_eq(tmp, jiffies))
3119 if (ehc->unloaded_mask & (1 << dev->devno))
3122 ata_eh_park_issue_cmd(dev, 1);
3127 if (time_before_eq(deadline, now))
3130 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3133 ata_port_for_each_link(link, ap) {
3134 ata_link_for_each_dev(dev, link) {
3135 if (!(link->eh_context.unloaded_mask &
3139 ata_eh_park_issue_cmd(dev, 0);
3140 ata_eh_done(link, dev, ATA_EH_PARK);
3145 ata_port_for_each_link(link, ap) {
3146 struct ata_eh_context *ehc = &link->eh_context;
3148 /* revalidate existing devices and attach new ones */
3149 rc = ata_eh_revalidate_and_attach(link, &dev);
3153 /* if PMP got attached, return, pmp EH will take care of it */
3154 if (link->device->class == ATA_DEV_PMP) {
3159 /* configure transfer mode if necessary */
3160 if (ehc->i.flags & ATA_EHI_SETMODE) {
3161 rc = ata_set_mode(link, &dev);
3164 ehc->i.flags &= ~ATA_EHI_SETMODE;
3167 /* If reset has been issued, clear UA to avoid
3168 * disrupting the current users of the device.
3170 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3171 ata_link_for_each_dev(dev, link) {
3172 if (dev->class != ATA_DEV_ATAPI)
3174 rc = atapi_eh_clear_ua(dev);
3180 /* configure link power saving */
3181 if (ehc->i.action & ATA_EH_LPM)
3182 ata_link_for_each_dev(dev, link)
3183 ata_dev_enable_pm(dev, ap->pm_policy);
3185 /* this link is okay now */
3191 ata_eh_handle_dev_fail(dev, rc);
3193 if (ap->pflags & ATA_PFLAG_FROZEN) {
3194 /* PMP reset requires working host port.
3195 * Can't retry if it's frozen.
3197 if (sata_pmp_attached(ap))
3207 if (rc && r_failed_link)
3208 *r_failed_link = link;
3210 DPRINTK("EXIT, rc=%d\n", rc);
3215 * ata_eh_finish - finish up EH
3216 * @ap: host port to finish EH for
3218 * Recovery is complete. Clean up EH states and retry or finish
3224 void ata_eh_finish(struct ata_port *ap)
3228 /* retry or finish qcs */
3229 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3230 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3232 if (!(qc->flags & ATA_QCFLAG_FAILED))
3236 /* FIXME: Once EH migration is complete,
3237 * generate sense data in this function,
3238 * considering both err_mask and tf.
3240 if (qc->flags & ATA_QCFLAG_RETRY)
3241 ata_eh_qc_retry(qc);
3243 ata_eh_qc_complete(qc);
3245 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3246 ata_eh_qc_complete(qc);
3248 /* feed zero TF to sense generation */
3249 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3250 ata_eh_qc_retry(qc);
3255 /* make sure nr_active_links is zero after EH */
3256 WARN_ON(ap->nr_active_links);
3257 ap->nr_active_links = 0;
3261 * ata_do_eh - do standard error handling
3262 * @ap: host port to handle error for
3264 * @prereset: prereset method (can be NULL)
3265 * @softreset: softreset method (can be NULL)
3266 * @hardreset: hardreset method (can be NULL)
3267 * @postreset: postreset method (can be NULL)
3269 * Perform standard error handling sequence.
3272 * Kernel thread context (may sleep).
3274 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3275 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3276 ata_postreset_fn_t postreset)
3278 struct ata_device *dev;
3284 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3287 ata_link_for_each_dev(dev, &ap->link)
3288 ata_dev_disable(dev);
3295 * ata_std_error_handler - standard error handler
3296 * @ap: host port to handle error for
3298 * Standard error handler
3301 * Kernel thread context (may sleep).
3303 void ata_std_error_handler(struct ata_port *ap)
3305 struct ata_port_operations *ops = ap->ops;
3306 ata_reset_fn_t hardreset = ops->hardreset;
3308 /* ignore built-in hardreset if SCR access is not available */
3309 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3312 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3317 * ata_eh_handle_port_suspend - perform port suspend operation
3318 * @ap: port to suspend
3323 * Kernel thread context (may sleep).
3325 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3327 unsigned long flags;
3330 /* are we suspending? */
3331 spin_lock_irqsave(ap->lock, flags);
3332 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3333 ap->pm_mesg.event == PM_EVENT_ON) {
3334 spin_unlock_irqrestore(ap->lock, flags);
3337 spin_unlock_irqrestore(ap->lock, flags);
3339 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3341 /* tell ACPI we're suspending */
3342 rc = ata_acpi_on_suspend(ap);
3347 ata_eh_freeze_port(ap);
3349 if (ap->ops->port_suspend)
3350 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3352 ata_acpi_set_state(ap, PMSG_SUSPEND);
3355 spin_lock_irqsave(ap->lock, flags);
3357 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3359 ap->pflags |= ATA_PFLAG_SUSPENDED;
3360 else if (ap->pflags & ATA_PFLAG_FROZEN)
3361 ata_port_schedule_eh(ap);
3363 if (ap->pm_result) {
3364 *ap->pm_result = rc;
3365 ap->pm_result = NULL;
3368 spin_unlock_irqrestore(ap->lock, flags);
3374 * ata_eh_handle_port_resume - perform port resume operation
3375 * @ap: port to resume
3380 * Kernel thread context (may sleep).
3382 static void ata_eh_handle_port_resume(struct ata_port *ap)
3384 unsigned long flags;
3387 /* are we resuming? */
3388 spin_lock_irqsave(ap->lock, flags);
3389 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3390 ap->pm_mesg.event != PM_EVENT_ON) {
3391 spin_unlock_irqrestore(ap->lock, flags);
3394 spin_unlock_irqrestore(ap->lock, flags);
3396 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3398 ata_acpi_set_state(ap, PMSG_ON);
3400 if (ap->ops->port_resume)
3401 rc = ap->ops->port_resume(ap);
3403 /* tell ACPI that we're resuming */
3404 ata_acpi_on_resume(ap);
3407 spin_lock_irqsave(ap->lock, flags);
3408 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3409 if (ap->pm_result) {
3410 *ap->pm_result = rc;
3411 ap->pm_result = NULL;
3413 spin_unlock_irqrestore(ap->lock, flags);
3415 #endif /* CONFIG_PM */