2 * libata-eh.c - libata error handling
4 * Maintained by: Tejun Heo <tj@kernel.org>
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/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
47 #include <linux/libata.h>
49 #include <trace/events/libata.h>
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
60 ATA_EFLAG_IS_IO = (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
62 ATA_EFLAG_OLD_ER = (1 << 31),
64 /* error categories */
67 ATA_ECAT_TOUT_HSM = 2,
69 ATA_ECAT_DUBIOUS_NONE = 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
75 ATA_EH_CMD_DFL_TIMEOUT = 5000,
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN = 5000,
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
85 ATA_EH_PRERESET_TIMEOUT = 10000,
86 ATA_EH_FASTDRAIN_INTERVAL = 3000,
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS = 2,
95 /* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
102 static const unsigned long ata_eh_reset_timeouts[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX, /* > 1 min has elapsed, give up */
110 static const unsigned long ata_eh_identify_timeouts[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
117 static const unsigned long ata_eh_flush_timeouts[] = {
118 15000, /* be generous with flush */
120 30000, /* and even more generous */
124 static const unsigned long ata_eh_other_timeouts[] = {
125 5000, /* same rationale as identify timeout */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
131 struct ata_eh_cmd_timeout_ent {
133 const unsigned long *timeouts;
136 /* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
148 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
149 static const struct ata_eh_cmd_timeout_ent
150 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
151 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
152 .timeouts = ata_eh_identify_timeouts, },
153 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_SET_FEATURES),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
160 .timeouts = ata_eh_other_timeouts, },
161 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
162 .timeouts = ata_eh_flush_timeouts },
166 static void __ata_port_freeze(struct ata_port *ap);
168 static void ata_eh_handle_port_suspend(struct ata_port *ap);
169 static void ata_eh_handle_port_resume(struct ata_port *ap);
170 #else /* CONFIG_PM */
171 static void ata_eh_handle_port_suspend(struct ata_port *ap)
174 static void ata_eh_handle_port_resume(struct ata_port *ap)
176 #endif /* CONFIG_PM */
178 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
181 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
182 ATA_EH_DESC_LEN - ehi->desc_len,
187 * __ata_ehi_push_desc - push error description without adding separator
189 * @fmt: printf format string
191 * Format string according to @fmt and append it to @ehi->desc.
194 * spin_lock_irqsave(host lock)
196 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
201 __ata_ehi_pushv_desc(ehi, fmt, args);
206 * ata_ehi_push_desc - push error description with separator
208 * @fmt: printf format string
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
214 * spin_lock_irqsave(host lock)
216 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
221 __ata_ehi_push_desc(ehi, ", ");
224 __ata_ehi_pushv_desc(ehi, fmt, args);
229 * ata_ehi_clear_desc - clean error description
235 * spin_lock_irqsave(host lock)
237 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
256 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
287 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
292 unsigned long long start, len;
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
309 #endif /* CONFIG_PCI */
311 static int ata_lookup_timeout_table(u8 cmd)
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
331 * Determine timeout for internal command @cmd for @dev.
337 * Determined timeout.
339 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
346 return ATA_EH_CMD_DFL_TIMEOUT;
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
364 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
378 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
381 struct ata_ering_entry *ent;
386 ering->cursor %= ATA_ERING_SIZE;
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
394 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
403 int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
408 struct ata_ering_entry *ent;
412 ent = &ering->ring[idx];
415 rc = map_fn(ent, arg);
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
424 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
426 ent->eflags |= ATA_EFLAG_OLD_ER;
430 static void ata_ering_clear(struct ata_ering *ering)
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
435 static unsigned int ata_eh_dev_action(struct ata_device *dev)
437 struct ata_eh_context *ehc = &dev->link->eh_context;
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
442 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
445 struct ata_device *tdev;
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
479 void ata_eh_acquire(struct ata_port *ap)
481 mutex_lock(&ap->host->eh_mutex);
482 WARN_ON_ONCE(ap->host->eh_owner);
483 ap->host->eh_owner = current;
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
496 void ata_eh_release(struct ata_port *ap)
498 WARN_ON_ONCE(ap->host->eh_owner != current);
499 ap->host->eh_owner = NULL;
500 mutex_unlock(&ap->host->eh_mutex);
504 * ata_scsi_timed_out - SCSI layer time out callback
505 * @cmd: timed out SCSI command
507 * Handles SCSI layer timeout. We race with normal completion of
508 * the qc for @cmd. If the qc is already gone, we lose and let
509 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
510 * timed out and EH should be invoked. Prevent ata_qc_complete()
511 * from finishing it by setting EH_SCHEDULED and return
514 * TODO: kill this function once old EH is gone.
517 * Called from timer context
520 * EH_HANDLED or EH_NOT_HANDLED
522 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
524 struct Scsi_Host *host = cmd->device->host;
525 struct ata_port *ap = ata_shost_to_port(host);
527 struct ata_queued_cmd *qc;
528 enum blk_eh_timer_return ret;
532 if (ap->ops->error_handler) {
533 ret = BLK_EH_NOT_HANDLED;
537 ret = BLK_EH_HANDLED;
538 spin_lock_irqsave(ap->lock, flags);
539 qc = ata_qc_from_tag(ap, ap->link.active_tag);
541 WARN_ON(qc->scsicmd != cmd);
542 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
543 qc->err_mask |= AC_ERR_TIMEOUT;
544 ret = BLK_EH_NOT_HANDLED;
546 spin_unlock_irqrestore(ap->lock, flags);
549 DPRINTK("EXIT, ret=%d\n", ret);
552 EXPORT_SYMBOL(ata_scsi_timed_out);
554 static void ata_eh_unload(struct ata_port *ap)
556 struct ata_link *link;
557 struct ata_device *dev;
560 /* Restore SControl IPM and SPD for the next driver and
561 * disable attached devices.
563 ata_for_each_link(link, ap, PMP_FIRST) {
564 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
565 ata_for_each_dev(dev, link, ALL)
566 ata_dev_disable(dev);
569 /* freeze and set UNLOADED */
570 spin_lock_irqsave(ap->lock, flags);
572 ata_port_freeze(ap); /* won't be thawed */
573 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
574 ap->pflags |= ATA_PFLAG_UNLOADED;
576 spin_unlock_irqrestore(ap->lock, flags);
580 * ata_scsi_error - SCSI layer error handler callback
581 * @host: SCSI host on which error occurred
583 * Handles SCSI-layer-thrown error events.
586 * Inherited from SCSI layer (none, can sleep)
591 void ata_scsi_error(struct Scsi_Host *host)
593 struct ata_port *ap = ata_shost_to_port(host);
595 LIST_HEAD(eh_work_q);
599 spin_lock_irqsave(host->host_lock, flags);
600 list_splice_init(&host->eh_cmd_q, &eh_work_q);
601 spin_unlock_irqrestore(host->host_lock, flags);
603 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
605 /* If we timed raced normal completion and there is nothing to
606 recover nr_timedout == 0 why exactly are we doing error recovery ? */
607 ata_scsi_port_error_handler(host, ap);
609 /* finish or retry handled scmd's and clean up */
610 WARN_ON(!list_empty(&eh_work_q));
616 * ata_scsi_cmd_error_handler - error callback for a list of commands
617 * @host: scsi host containing the port
618 * @ap: ATA port within the host
619 * @eh_work_q: list of commands to process
621 * process the given list of commands and return those finished to the
622 * ap->eh_done_q. This function is the first part of the libata error
623 * handler which processes a given list of failed commands.
625 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
626 struct list_head *eh_work_q)
631 /* make sure sff pio task is not running */
632 ata_sff_flush_pio_task(ap);
634 /* synchronize with host lock and sort out timeouts */
636 /* For new EH, all qcs are finished in one of three ways -
637 * normal completion, error completion, and SCSI timeout.
638 * Both completions can race against SCSI timeout. When normal
639 * completion wins, the qc never reaches EH. When error
640 * completion wins, the qc has ATA_QCFLAG_FAILED set.
642 * When SCSI timeout wins, things are a bit more complex.
643 * Normal or error completion can occur after the timeout but
644 * before this point. In such cases, both types of
645 * completions are honored. A scmd is determined to have
646 * timed out iff its associated qc is active and not failed.
648 if (ap->ops->error_handler) {
649 struct scsi_cmnd *scmd, *tmp;
652 spin_lock_irqsave(ap->lock, flags);
654 /* This must occur under the ap->lock as we don't want
655 a polled recovery to race the real interrupt handler
657 The lost_interrupt handler checks for any completed but
658 non-notified command and completes much like an IRQ handler.
660 We then fall into the error recovery code which will treat
661 this as if normal completion won the race */
663 if (ap->ops->lost_interrupt)
664 ap->ops->lost_interrupt(ap);
666 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
667 struct ata_queued_cmd *qc;
669 for (i = 0; i < ATA_MAX_QUEUE; i++) {
670 qc = __ata_qc_from_tag(ap, i);
671 if (qc->flags & ATA_QCFLAG_ACTIVE &&
676 if (i < ATA_MAX_QUEUE) {
677 /* the scmd has an associated qc */
678 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
679 /* which hasn't failed yet, timeout */
680 qc->err_mask |= AC_ERR_TIMEOUT;
681 qc->flags |= ATA_QCFLAG_FAILED;
685 /* Normal completion occurred after
686 * SCSI timeout but before this point.
687 * Successfully complete it.
689 scmd->retries = scmd->allowed;
690 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
694 /* If we have timed out qcs. They belong to EH from
695 * this point but the state of the controller is
696 * unknown. Freeze the port to make sure the IRQ
697 * handler doesn't diddle with those qcs. This must
698 * be done atomically w.r.t. setting QCFLAG_FAILED.
701 __ata_port_freeze(ap);
703 spin_unlock_irqrestore(ap->lock, flags);
705 /* initialize eh_tries */
706 ap->eh_tries = ATA_EH_MAX_TRIES;
708 spin_unlock_wait(ap->lock);
711 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
714 * ata_scsi_port_error_handler - recover the port after the commands
715 * @host: SCSI host containing the port
718 * Handle the recovery of the port @ap after all the commands
719 * have been recovered.
721 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
725 /* invoke error handler */
726 if (ap->ops->error_handler) {
727 struct ata_link *link;
729 /* acquire EH ownership */
732 /* kill fast drain timer */
733 del_timer_sync(&ap->fastdrain_timer);
735 /* process port resume request */
736 ata_eh_handle_port_resume(ap);
738 /* fetch & clear EH info */
739 spin_lock_irqsave(ap->lock, flags);
741 ata_for_each_link(link, ap, HOST_FIRST) {
742 struct ata_eh_context *ehc = &link->eh_context;
743 struct ata_device *dev;
745 memset(&link->eh_context, 0, sizeof(link->eh_context));
746 link->eh_context.i = link->eh_info;
747 memset(&link->eh_info, 0, sizeof(link->eh_info));
749 ata_for_each_dev(dev, link, ENABLED) {
750 int devno = dev->devno;
752 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
753 if (ata_ncq_enabled(dev))
754 ehc->saved_ncq_enabled |= 1 << devno;
758 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
759 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
760 ap->excl_link = NULL; /* don't maintain exclusion over EH */
762 spin_unlock_irqrestore(ap->lock, flags);
764 /* invoke EH, skip if unloading or suspended */
765 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
766 ap->ops->error_handler(ap);
768 /* if unloading, commence suicide */
769 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
770 !(ap->pflags & ATA_PFLAG_UNLOADED))
775 /* process port suspend request */
776 ata_eh_handle_port_suspend(ap);
778 /* Exception might have happened after ->error_handler
779 * recovered the port but before this point. Repeat
782 spin_lock_irqsave(ap->lock, flags);
784 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
785 if (--ap->eh_tries) {
786 spin_unlock_irqrestore(ap->lock, flags);
790 "EH pending after %d tries, giving up\n",
792 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
795 /* this run is complete, make sure EH info is clear */
796 ata_for_each_link(link, ap, HOST_FIRST)
797 memset(&link->eh_info, 0, sizeof(link->eh_info));
799 /* end eh (clear host_eh_scheduled) while holding
800 * ap->lock such that if exception occurs after this
801 * point but before EH completion, SCSI midlayer will
806 spin_unlock_irqrestore(ap->lock, flags);
809 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
810 ap->ops->eng_timeout(ap);
813 scsi_eh_flush_done_q(&ap->eh_done_q);
816 spin_lock_irqsave(ap->lock, flags);
818 if (ap->pflags & ATA_PFLAG_LOADING)
819 ap->pflags &= ~ATA_PFLAG_LOADING;
820 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
821 schedule_delayed_work(&ap->hotplug_task, 0);
823 if (ap->pflags & ATA_PFLAG_RECOVERED)
824 ata_port_info(ap, "EH complete\n");
826 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
828 /* tell wait_eh that we're done */
829 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
830 wake_up_all(&ap->eh_wait_q);
832 spin_unlock_irqrestore(ap->lock, flags);
834 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
837 * ata_port_wait_eh - Wait for the currently pending EH to complete
838 * @ap: Port to wait EH for
840 * Wait until the currently pending EH is complete.
843 * Kernel thread context (may sleep).
845 void ata_port_wait_eh(struct ata_port *ap)
851 spin_lock_irqsave(ap->lock, flags);
853 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
854 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
855 spin_unlock_irqrestore(ap->lock, flags);
857 spin_lock_irqsave(ap->lock, flags);
859 finish_wait(&ap->eh_wait_q, &wait);
861 spin_unlock_irqrestore(ap->lock, flags);
863 /* make sure SCSI EH is complete */
864 if (scsi_host_in_recovery(ap->scsi_host)) {
869 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
871 static int ata_eh_nr_in_flight(struct ata_port *ap)
876 /* count only non-internal commands */
877 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
878 if (ata_qc_from_tag(ap, tag))
884 void ata_eh_fastdrain_timerfn(unsigned long arg)
886 struct ata_port *ap = (void *)arg;
890 spin_lock_irqsave(ap->lock, flags);
892 cnt = ata_eh_nr_in_flight(ap);
898 if (cnt == ap->fastdrain_cnt) {
901 /* No progress during the last interval, tag all
902 * in-flight qcs as timed out and freeze the port.
904 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
905 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
907 qc->err_mask |= AC_ERR_TIMEOUT;
912 /* some qcs have finished, give it another chance */
913 ap->fastdrain_cnt = cnt;
914 ap->fastdrain_timer.expires =
915 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
916 add_timer(&ap->fastdrain_timer);
920 spin_unlock_irqrestore(ap->lock, flags);
924 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
925 * @ap: target ATA port
926 * @fastdrain: activate fast drain
928 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
929 * is non-zero and EH wasn't pending before. Fast drain ensures
930 * that EH kicks in in timely manner.
933 * spin_lock_irqsave(host lock)
935 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
939 /* already scheduled? */
940 if (ap->pflags & ATA_PFLAG_EH_PENDING)
943 ap->pflags |= ATA_PFLAG_EH_PENDING;
948 /* do we have in-flight qcs? */
949 cnt = ata_eh_nr_in_flight(ap);
953 /* activate fast drain */
954 ap->fastdrain_cnt = cnt;
955 ap->fastdrain_timer.expires =
956 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
957 add_timer(&ap->fastdrain_timer);
961 * ata_qc_schedule_eh - schedule qc for error handling
962 * @qc: command to schedule error handling for
964 * Schedule error handling for @qc. EH will kick in as soon as
965 * other commands are drained.
968 * spin_lock_irqsave(host lock)
970 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
972 struct ata_port *ap = qc->ap;
973 struct request_queue *q = qc->scsicmd->device->request_queue;
976 WARN_ON(!ap->ops->error_handler);
978 qc->flags |= ATA_QCFLAG_FAILED;
979 ata_eh_set_pending(ap, 1);
981 /* The following will fail if timeout has already expired.
982 * ata_scsi_error() takes care of such scmds on EH entry.
983 * Note that ATA_QCFLAG_FAILED is unconditionally set after
984 * this function completes.
986 spin_lock_irqsave(q->queue_lock, flags);
987 blk_abort_request(qc->scsicmd->request);
988 spin_unlock_irqrestore(q->queue_lock, flags);
992 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
993 * @ap: ATA port to schedule EH for
995 * LOCKING: inherited from ata_port_schedule_eh
996 * spin_lock_irqsave(host lock)
998 void ata_std_sched_eh(struct ata_port *ap)
1000 WARN_ON(!ap->ops->error_handler);
1002 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1005 ata_eh_set_pending(ap, 1);
1006 scsi_schedule_eh(ap->scsi_host);
1008 DPRINTK("port EH scheduled\n");
1010 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1013 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1014 * @ap: ATA port to end EH for
1016 * In the libata object model there is a 1:1 mapping of ata_port to
1017 * shost, so host fields can be directly manipulated under ap->lock, in
1018 * the libsas case we need to hold a lock at the ha->level to coordinate
1022 * spin_lock_irqsave(host lock)
1024 void ata_std_end_eh(struct ata_port *ap)
1026 struct Scsi_Host *host = ap->scsi_host;
1028 host->host_eh_scheduled = 0;
1030 EXPORT_SYMBOL(ata_std_end_eh);
1034 * ata_port_schedule_eh - schedule error handling without a qc
1035 * @ap: ATA port to schedule EH for
1037 * Schedule error handling for @ap. EH will kick in as soon as
1038 * all commands are drained.
1041 * spin_lock_irqsave(host lock)
1043 void ata_port_schedule_eh(struct ata_port *ap)
1045 /* see: ata_std_sched_eh, unless you know better */
1046 ap->ops->sched_eh(ap);
1049 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1051 int tag, nr_aborted = 0;
1053 WARN_ON(!ap->ops->error_handler);
1055 /* we're gonna abort all commands, no need for fast drain */
1056 ata_eh_set_pending(ap, 0);
1058 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1059 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1061 if (qc && (!link || qc->dev->link == link)) {
1062 qc->flags |= ATA_QCFLAG_FAILED;
1063 ata_qc_complete(qc);
1069 ata_port_schedule_eh(ap);
1075 * ata_link_abort - abort all qc's on the link
1076 * @link: ATA link to abort qc's for
1078 * Abort all active qc's active on @link and schedule EH.
1081 * spin_lock_irqsave(host lock)
1084 * Number of aborted qc's.
1086 int ata_link_abort(struct ata_link *link)
1088 return ata_do_link_abort(link->ap, link);
1092 * ata_port_abort - abort all qc's on the port
1093 * @ap: ATA port to abort qc's for
1095 * Abort all active qc's of @ap and schedule EH.
1098 * spin_lock_irqsave(host_set lock)
1101 * Number of aborted qc's.
1103 int ata_port_abort(struct ata_port *ap)
1105 return ata_do_link_abort(ap, NULL);
1109 * __ata_port_freeze - freeze port
1110 * @ap: ATA port to freeze
1112 * This function is called when HSM violation or some other
1113 * condition disrupts normal operation of the port. Frozen port
1114 * is not allowed to perform any operation until the port is
1115 * thawed, which usually follows a successful reset.
1117 * ap->ops->freeze() callback can be used for freezing the port
1118 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1119 * port cannot be frozen hardware-wise, the interrupt handler
1120 * must ack and clear interrupts unconditionally while the port
1124 * spin_lock_irqsave(host lock)
1126 static void __ata_port_freeze(struct ata_port *ap)
1128 WARN_ON(!ap->ops->error_handler);
1130 if (ap->ops->freeze)
1131 ap->ops->freeze(ap);
1133 ap->pflags |= ATA_PFLAG_FROZEN;
1135 DPRINTK("ata%u port frozen\n", ap->print_id);
1139 * ata_port_freeze - abort & freeze port
1140 * @ap: ATA port to freeze
1142 * Abort and freeze @ap. The freeze operation must be called
1143 * first, because some hardware requires special operations
1144 * before the taskfile registers are accessible.
1147 * spin_lock_irqsave(host lock)
1150 * Number of aborted commands.
1152 int ata_port_freeze(struct ata_port *ap)
1156 WARN_ON(!ap->ops->error_handler);
1158 __ata_port_freeze(ap);
1159 nr_aborted = ata_port_abort(ap);
1165 * sata_async_notification - SATA async notification handler
1166 * @ap: ATA port where async notification is received
1168 * Handler to be called when async notification via SDB FIS is
1169 * received. This function schedules EH if necessary.
1172 * spin_lock_irqsave(host lock)
1175 * 1 if EH is scheduled, 0 otherwise.
1177 int sata_async_notification(struct ata_port *ap)
1182 if (!(ap->flags & ATA_FLAG_AN))
1185 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1187 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1189 if (!sata_pmp_attached(ap) || rc) {
1190 /* PMP is not attached or SNTF is not available */
1191 if (!sata_pmp_attached(ap)) {
1192 /* PMP is not attached. Check whether ATAPI
1193 * AN is configured. If so, notify media
1196 struct ata_device *dev = ap->link.device;
1198 if ((dev->class == ATA_DEV_ATAPI) &&
1199 (dev->flags & ATA_DFLAG_AN))
1200 ata_scsi_media_change_notify(dev);
1203 /* PMP is attached but SNTF is not available.
1204 * ATAPI async media change notification is
1205 * not used. The PMP must be reporting PHY
1206 * status change, schedule EH.
1208 ata_port_schedule_eh(ap);
1212 /* PMP is attached and SNTF is available */
1213 struct ata_link *link;
1215 /* check and notify ATAPI AN */
1216 ata_for_each_link(link, ap, EDGE) {
1217 if (!(sntf & (1 << link->pmp)))
1220 if ((link->device->class == ATA_DEV_ATAPI) &&
1221 (link->device->flags & ATA_DFLAG_AN))
1222 ata_scsi_media_change_notify(link->device);
1225 /* If PMP is reporting that PHY status of some
1226 * downstream ports has changed, schedule EH.
1228 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1229 ata_port_schedule_eh(ap);
1238 * ata_eh_freeze_port - EH helper to freeze port
1239 * @ap: ATA port to freeze
1246 void ata_eh_freeze_port(struct ata_port *ap)
1248 unsigned long flags;
1250 if (!ap->ops->error_handler)
1253 spin_lock_irqsave(ap->lock, flags);
1254 __ata_port_freeze(ap);
1255 spin_unlock_irqrestore(ap->lock, flags);
1259 * ata_port_thaw_port - EH helper to thaw port
1260 * @ap: ATA port to thaw
1262 * Thaw frozen port @ap.
1267 void ata_eh_thaw_port(struct ata_port *ap)
1269 unsigned long flags;
1271 if (!ap->ops->error_handler)
1274 spin_lock_irqsave(ap->lock, flags);
1276 ap->pflags &= ~ATA_PFLAG_FROZEN;
1281 spin_unlock_irqrestore(ap->lock, flags);
1283 DPRINTK("ata%u port thawed\n", ap->print_id);
1286 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1291 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1293 struct ata_port *ap = qc->ap;
1294 struct scsi_cmnd *scmd = qc->scsicmd;
1295 unsigned long flags;
1297 spin_lock_irqsave(ap->lock, flags);
1298 qc->scsidone = ata_eh_scsidone;
1299 __ata_qc_complete(qc);
1300 WARN_ON(ata_tag_valid(qc->tag));
1301 spin_unlock_irqrestore(ap->lock, flags);
1303 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1307 * ata_eh_qc_complete - Complete an active ATA command from EH
1308 * @qc: Command to complete
1310 * Indicate to the mid and upper layers that an ATA command has
1311 * completed. To be used from EH.
1313 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1315 struct scsi_cmnd *scmd = qc->scsicmd;
1316 scmd->retries = scmd->allowed;
1317 __ata_eh_qc_complete(qc);
1321 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1322 * @qc: Command to retry
1324 * Indicate to the mid and upper layers that an ATA command
1325 * should be retried. To be used from EH.
1327 * SCSI midlayer limits the number of retries to scmd->allowed.
1328 * scmd->allowed is incremented for commands which get retried
1329 * due to unrelated failures (qc->err_mask is zero).
1331 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1333 struct scsi_cmnd *scmd = qc->scsicmd;
1336 __ata_eh_qc_complete(qc);
1340 * ata_dev_disable - disable ATA device
1341 * @dev: ATA device to disable
1348 void ata_dev_disable(struct ata_device *dev)
1350 if (!ata_dev_enabled(dev))
1353 if (ata_msg_drv(dev->link->ap))
1354 ata_dev_warn(dev, "disabled\n");
1355 ata_acpi_on_disable(dev);
1356 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1359 /* From now till the next successful probe, ering is used to
1360 * track probe failures. Clear accumulated device error info.
1362 ata_ering_clear(&dev->ering);
1366 * ata_eh_detach_dev - detach ATA device
1367 * @dev: ATA device to detach
1374 void ata_eh_detach_dev(struct ata_device *dev)
1376 struct ata_link *link = dev->link;
1377 struct ata_port *ap = link->ap;
1378 struct ata_eh_context *ehc = &link->eh_context;
1379 unsigned long flags;
1381 ata_dev_disable(dev);
1383 spin_lock_irqsave(ap->lock, flags);
1385 dev->flags &= ~ATA_DFLAG_DETACH;
1387 if (ata_scsi_offline_dev(dev)) {
1388 dev->flags |= ATA_DFLAG_DETACHED;
1389 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1392 /* clear per-dev EH info */
1393 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1394 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1395 ehc->saved_xfer_mode[dev->devno] = 0;
1396 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1398 spin_unlock_irqrestore(ap->lock, flags);
1402 * ata_eh_about_to_do - about to perform eh_action
1403 * @link: target ATA link
1404 * @dev: target ATA dev for per-dev action (can be NULL)
1405 * @action: action about to be performed
1407 * Called just before performing EH actions to clear related bits
1408 * in @link->eh_info such that eh actions are not unnecessarily
1414 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1415 unsigned int action)
1417 struct ata_port *ap = link->ap;
1418 struct ata_eh_info *ehi = &link->eh_info;
1419 struct ata_eh_context *ehc = &link->eh_context;
1420 unsigned long flags;
1422 spin_lock_irqsave(ap->lock, flags);
1424 ata_eh_clear_action(link, dev, ehi, action);
1426 /* About to take EH action, set RECOVERED. Ignore actions on
1427 * slave links as master will do them again.
1429 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1430 ap->pflags |= ATA_PFLAG_RECOVERED;
1432 spin_unlock_irqrestore(ap->lock, flags);
1436 * ata_eh_done - EH action complete
1437 * @ap: target ATA port
1438 * @dev: target ATA dev for per-dev action (can be NULL)
1439 * @action: action just completed
1441 * Called right after performing EH actions to clear related bits
1442 * in @link->eh_context.
1447 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1448 unsigned int action)
1450 struct ata_eh_context *ehc = &link->eh_context;
1452 ata_eh_clear_action(link, dev, &ehc->i, action);
1456 * ata_err_string - convert err_mask to descriptive string
1457 * @err_mask: error mask to convert to string
1459 * Convert @err_mask to descriptive string. Errors are
1460 * prioritized according to severity and only the most severe
1461 * error is reported.
1467 * Descriptive string for @err_mask
1469 static const char *ata_err_string(unsigned int err_mask)
1471 if (err_mask & AC_ERR_HOST_BUS)
1472 return "host bus error";
1473 if (err_mask & AC_ERR_ATA_BUS)
1474 return "ATA bus error";
1475 if (err_mask & AC_ERR_TIMEOUT)
1477 if (err_mask & AC_ERR_HSM)
1478 return "HSM violation";
1479 if (err_mask & AC_ERR_SYSTEM)
1480 return "internal error";
1481 if (err_mask & AC_ERR_MEDIA)
1482 return "media error";
1483 if (err_mask & AC_ERR_INVALID)
1484 return "invalid argument";
1485 if (err_mask & AC_ERR_DEV)
1486 return "device error";
1487 return "unknown error";
1491 * ata_read_log_page - read a specific log page
1492 * @dev: target device
1494 * @page: page to read
1495 * @buf: buffer to store read page
1496 * @sectors: number of sectors to read
1498 * Read log page using READ_LOG_EXT command.
1501 * Kernel thread context (may sleep).
1504 * 0 on success, AC_ERR_* mask otherwise.
1506 unsigned int ata_read_log_page(struct ata_device *dev, u8 log,
1507 u8 page, void *buf, unsigned int sectors)
1509 unsigned long ap_flags = dev->link->ap->flags;
1510 struct ata_taskfile tf;
1511 unsigned int err_mask;
1514 DPRINTK("read log page - log 0x%x, page 0x%x\n", log, page);
1517 * Return error without actually issuing the command on controllers
1518 * which e.g. lockup on a read log page.
1520 if (ap_flags & ATA_FLAG_NO_LOG_PAGE)
1524 ata_tf_init(dev, &tf);
1525 if (dev->dma_mode && ata_id_has_read_log_dma_ext(dev->id) &&
1526 !(dev->horkage & ATA_HORKAGE_NO_NCQ_LOG)) {
1527 tf.command = ATA_CMD_READ_LOG_DMA_EXT;
1528 tf.protocol = ATA_PROT_DMA;
1531 tf.command = ATA_CMD_READ_LOG_EXT;
1532 tf.protocol = ATA_PROT_PIO;
1538 tf.hob_nsect = sectors >> 8;
1539 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1541 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1542 buf, sectors * ATA_SECT_SIZE, 0);
1544 if (err_mask && dma) {
1545 dev->horkage |= ATA_HORKAGE_NO_NCQ_LOG;
1546 ata_dev_warn(dev, "READ LOG DMA EXT failed, trying unqueued\n");
1550 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1555 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1556 * @dev: Device to read log page 10h from
1557 * @tag: Resulting tag of the failed command
1558 * @tf: Resulting taskfile registers of the failed command
1560 * Read log page 10h to obtain NCQ error details and clear error
1564 * Kernel thread context (may sleep).
1567 * 0 on success, -errno otherwise.
1569 static int ata_eh_read_log_10h(struct ata_device *dev,
1570 int *tag, struct ata_taskfile *tf)
1572 u8 *buf = dev->link->ap->sector_buf;
1573 unsigned int err_mask;
1577 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1582 for (i = 0; i < ATA_SECT_SIZE; i++)
1585 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1591 *tag = buf[0] & 0x1f;
1593 tf->command = buf[2];
1594 tf->feature = buf[3];
1598 tf->device = buf[7];
1599 tf->hob_lbal = buf[8];
1600 tf->hob_lbam = buf[9];
1601 tf->hob_lbah = buf[10];
1602 tf->nsect = buf[12];
1603 tf->hob_nsect = buf[13];
1604 if (ata_id_has_ncq_autosense(dev->id))
1605 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1611 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1612 * @dev: target ATAPI device
1613 * @r_sense_key: out parameter for sense_key
1615 * Perform ATAPI TEST_UNIT_READY.
1618 * EH context (may sleep).
1621 * 0 on success, AC_ERR_* mask on failure.
1623 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1625 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1626 struct ata_taskfile tf;
1627 unsigned int err_mask;
1629 ata_tf_init(dev, &tf);
1631 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1632 tf.command = ATA_CMD_PACKET;
1633 tf.protocol = ATAPI_PROT_NODATA;
1635 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1636 if (err_mask == AC_ERR_DEV)
1637 *r_sense_key = tf.feature >> 4;
1642 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1643 * @dev: device to perform REQUEST_SENSE_SENSE_DATA_EXT to
1644 * @cmd: scsi command for which the sense code should be set
1646 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1647 * SENSE. This function is an EH helper.
1650 * Kernel thread context (may sleep).
1652 static void ata_eh_request_sense(struct ata_queued_cmd *qc,
1653 struct scsi_cmnd *cmd)
1655 struct ata_device *dev = qc->dev;
1656 struct ata_taskfile tf;
1657 unsigned int err_mask;
1659 if (qc->ap->pflags & ATA_PFLAG_FROZEN) {
1660 ata_dev_warn(dev, "sense data available but port frozen\n");
1664 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1667 if (!ata_id_sense_reporting_enabled(dev->id)) {
1668 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1672 DPRINTK("ATA request sense\n");
1674 ata_tf_init(dev, &tf);
1675 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1676 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1677 tf.command = ATA_CMD_REQ_SENSE_DATA;
1678 tf.protocol = ATA_PROT_NODATA;
1680 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1681 /* Ignore err_mask; ATA_ERR might be set */
1682 if (tf.command & ATA_SENSE) {
1683 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1684 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1686 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1687 tf.command, err_mask);
1692 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1693 * @dev: device to perform REQUEST_SENSE to
1694 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1695 * @dfl_sense_key: default sense key to use
1697 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1698 * SENSE. This function is EH helper.
1701 * Kernel thread context (may sleep).
1704 * 0 on success, AC_ERR_* mask on failure
1706 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1707 u8 *sense_buf, u8 dfl_sense_key)
1709 u8 cdb[ATAPI_CDB_LEN] =
1710 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1711 struct ata_port *ap = dev->link->ap;
1712 struct ata_taskfile tf;
1714 DPRINTK("ATAPI request sense\n");
1716 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1718 /* initialize sense_buf with the error register,
1719 * for the case where they are -not- overwritten
1721 sense_buf[0] = 0x70;
1722 sense_buf[2] = dfl_sense_key;
1724 /* some devices time out if garbage left in tf */
1725 ata_tf_init(dev, &tf);
1727 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1728 tf.command = ATA_CMD_PACKET;
1730 /* is it pointless to prefer PIO for "safety reasons"? */
1731 if (ap->flags & ATA_FLAG_PIO_DMA) {
1732 tf.protocol = ATAPI_PROT_DMA;
1733 tf.feature |= ATAPI_PKT_DMA;
1735 tf.protocol = ATAPI_PROT_PIO;
1736 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1740 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1741 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1745 * ata_eh_analyze_serror - analyze SError for a failed port
1746 * @link: ATA link to analyze SError for
1748 * Analyze SError if available and further determine cause of
1754 static void ata_eh_analyze_serror(struct ata_link *link)
1756 struct ata_eh_context *ehc = &link->eh_context;
1757 u32 serror = ehc->i.serror;
1758 unsigned int err_mask = 0, action = 0;
1761 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1762 err_mask |= AC_ERR_ATA_BUS;
1763 action |= ATA_EH_RESET;
1765 if (serror & SERR_PROTOCOL) {
1766 err_mask |= AC_ERR_HSM;
1767 action |= ATA_EH_RESET;
1769 if (serror & SERR_INTERNAL) {
1770 err_mask |= AC_ERR_SYSTEM;
1771 action |= ATA_EH_RESET;
1774 /* Determine whether a hotplug event has occurred. Both
1775 * SError.N/X are considered hotplug events for enabled or
1776 * host links. For disabled PMP links, only N bit is
1777 * considered as X bit is left at 1 for link plugging.
1779 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1780 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1781 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1782 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1784 hotplug_mask = SERR_PHYRDY_CHG;
1786 if (serror & hotplug_mask)
1787 ata_ehi_hotplugged(&ehc->i);
1789 ehc->i.err_mask |= err_mask;
1790 ehc->i.action |= action;
1794 * ata_eh_analyze_ncq_error - analyze NCQ error
1795 * @link: ATA link to analyze NCQ error for
1797 * Read log page 10h, determine the offending qc and acquire
1798 * error status TF. For NCQ device errors, all LLDDs have to do
1799 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1803 * Kernel thread context (may sleep).
1805 void ata_eh_analyze_ncq_error(struct ata_link *link)
1807 struct ata_port *ap = link->ap;
1808 struct ata_eh_context *ehc = &link->eh_context;
1809 struct ata_device *dev = link->device;
1810 struct ata_queued_cmd *qc;
1811 struct ata_taskfile tf;
1814 /* if frozen, we can't do much */
1815 if (ap->pflags & ATA_PFLAG_FROZEN)
1818 /* is it NCQ device error? */
1819 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1822 /* has LLDD analyzed already? */
1823 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1824 qc = __ata_qc_from_tag(ap, tag);
1826 if (!(qc->flags & ATA_QCFLAG_FAILED))
1833 /* okay, this error is ours */
1834 memset(&tf, 0, sizeof(tf));
1835 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1837 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1842 if (!(link->sactive & (1 << tag))) {
1843 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1848 /* we've got the perpetrator, condemn it */
1849 qc = __ata_qc_from_tag(ap, tag);
1850 memcpy(&qc->result_tf, &tf, sizeof(tf));
1851 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1852 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1853 if ((qc->result_tf.command & ATA_SENSE) || qc->result_tf.auxiliary) {
1854 char sense_key, asc, ascq;
1856 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1857 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1858 ascq = qc->result_tf.auxiliary & 0xff;
1859 ata_scsi_set_sense(dev, qc->scsicmd, sense_key, asc, ascq);
1860 ata_scsi_set_sense_information(dev, qc->scsicmd,
1862 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1865 ehc->i.err_mask &= ~AC_ERR_DEV;
1869 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1870 * @qc: qc to analyze
1871 * @tf: Taskfile registers to analyze
1873 * Analyze taskfile of @qc and further determine cause of
1874 * failure. This function also requests ATAPI sense data if
1878 * Kernel thread context (may sleep).
1881 * Determined recovery action
1883 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1884 const struct ata_taskfile *tf)
1886 unsigned int tmp, action = 0;
1887 u8 stat = tf->command, err = tf->feature;
1889 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1890 qc->err_mask |= AC_ERR_HSM;
1891 return ATA_EH_RESET;
1894 if (stat & (ATA_ERR | ATA_DF)) {
1895 qc->err_mask |= AC_ERR_DEV;
1897 * Sense data reporting does not work if the
1898 * device fault bit is set.
1906 switch (qc->dev->class) {
1909 if (stat & ATA_SENSE)
1910 ata_eh_request_sense(qc, qc->scsicmd);
1912 qc->err_mask |= AC_ERR_ATA_BUS;
1913 if (err & (ATA_UNC | ATA_AMNF))
1914 qc->err_mask |= AC_ERR_MEDIA;
1916 qc->err_mask |= AC_ERR_INVALID;
1920 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1921 tmp = atapi_eh_request_sense(qc->dev,
1922 qc->scsicmd->sense_buffer,
1923 qc->result_tf.feature >> 4);
1925 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1927 qc->err_mask |= tmp;
1931 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1932 int ret = scsi_check_sense(qc->scsicmd);
1934 * SUCCESS here means that the sense code could
1935 * evaluated and should be passed to the upper layers
1936 * for correct evaluation.
1937 * FAILED means the sense code could not interpreted
1938 * and the device would need to be reset.
1939 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1940 * command would need to be retried.
1942 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1943 qc->flags |= ATA_QCFLAG_RETRY;
1944 qc->err_mask |= AC_ERR_OTHER;
1945 } else if (ret != SUCCESS) {
1946 qc->err_mask |= AC_ERR_HSM;
1949 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1950 action |= ATA_EH_RESET;
1955 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1960 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1964 base = ATA_ECAT_DUBIOUS_NONE;
1966 if (err_mask & AC_ERR_ATA_BUS)
1967 return base + ATA_ECAT_ATA_BUS;
1969 if (err_mask & AC_ERR_TIMEOUT)
1970 return base + ATA_ECAT_TOUT_HSM;
1972 if (eflags & ATA_EFLAG_IS_IO) {
1973 if (err_mask & AC_ERR_HSM)
1974 return base + ATA_ECAT_TOUT_HSM;
1976 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1977 return base + ATA_ECAT_UNK_DEV;
1983 struct speed_down_verdict_arg {
1986 int nr_errors[ATA_ECAT_NR];
1989 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1991 struct speed_down_verdict_arg *arg = void_arg;
1994 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1997 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1999 arg->nr_errors[cat]++;
2005 * ata_eh_speed_down_verdict - Determine speed down verdict
2006 * @dev: Device of interest
2008 * This function examines error ring of @dev and determines
2009 * whether NCQ needs to be turned off, transfer speed should be
2010 * stepped down, or falling back to PIO is necessary.
2012 * ECAT_ATA_BUS : ATA_BUS error for any command
2014 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
2017 * ECAT_UNK_DEV : Unknown DEV error for IO commands
2019 * ECAT_DUBIOUS_* : Identical to above three but occurred while
2020 * data transfer hasn't been verified.
2024 * NCQ_OFF : Turn off NCQ.
2026 * SPEED_DOWN : Speed down transfer speed but don't fall back
2029 * FALLBACK_TO_PIO : Fall back to PIO.
2031 * Even if multiple verdicts are returned, only one action is
2032 * taken per error. An action triggered by non-DUBIOUS errors
2033 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
2034 * This is to expedite speed down decisions right after device is
2035 * initially configured.
2037 * The followings are speed down rules. #1 and #2 deal with
2040 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
2041 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
2043 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
2044 * occurred during last 5 mins, NCQ_OFF.
2046 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
2047 * occurred during last 5 mins, FALLBACK_TO_PIO
2049 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
2050 * during last 10 mins, NCQ_OFF.
2052 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
2053 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
2056 * Inherited from caller.
2059 * OR of ATA_EH_SPDN_* flags.
2061 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
2063 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
2064 u64 j64 = get_jiffies_64();
2065 struct speed_down_verdict_arg arg;
2066 unsigned int verdict = 0;
2068 /* scan past 5 mins of error history */
2069 memset(&arg, 0, sizeof(arg));
2070 arg.since = j64 - min(j64, j5mins);
2071 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2073 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
2074 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
2075 verdict |= ATA_EH_SPDN_SPEED_DOWN |
2076 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
2078 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
2079 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
2080 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
2082 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2083 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2084 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2085 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
2087 /* scan past 10 mins of error history */
2088 memset(&arg, 0, sizeof(arg));
2089 arg.since = j64 - min(j64, j10mins);
2090 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2092 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2093 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
2094 verdict |= ATA_EH_SPDN_NCQ_OFF;
2096 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2097 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
2098 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2099 verdict |= ATA_EH_SPDN_SPEED_DOWN;
2105 * ata_eh_speed_down - record error and speed down if necessary
2106 * @dev: Failed device
2107 * @eflags: mask of ATA_EFLAG_* flags
2108 * @err_mask: err_mask of the error
2110 * Record error and examine error history to determine whether
2111 * adjusting transmission speed is necessary. It also sets
2112 * transmission limits appropriately if such adjustment is
2116 * Kernel thread context (may sleep).
2119 * Determined recovery action.
2121 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2122 unsigned int eflags, unsigned int err_mask)
2124 struct ata_link *link = ata_dev_phys_link(dev);
2126 unsigned int verdict;
2127 unsigned int action = 0;
2129 /* don't bother if Cat-0 error */
2130 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2133 /* record error and determine whether speed down is necessary */
2134 ata_ering_record(&dev->ering, eflags, err_mask);
2135 verdict = ata_eh_speed_down_verdict(dev);
2138 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2139 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2140 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2141 dev->flags |= ATA_DFLAG_NCQ_OFF;
2142 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2147 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2148 /* speed down SATA link speed if possible */
2149 if (sata_down_spd_limit(link, 0) == 0) {
2150 action |= ATA_EH_RESET;
2154 /* lower transfer mode */
2155 if (dev->spdn_cnt < 2) {
2156 static const int dma_dnxfer_sel[] =
2157 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2158 static const int pio_dnxfer_sel[] =
2159 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2162 if (dev->xfer_shift != ATA_SHIFT_PIO)
2163 sel = dma_dnxfer_sel[dev->spdn_cnt];
2165 sel = pio_dnxfer_sel[dev->spdn_cnt];
2169 if (ata_down_xfermask_limit(dev, sel) == 0) {
2170 action |= ATA_EH_RESET;
2176 /* Fall back to PIO? Slowing down to PIO is meaningless for
2177 * SATA ATA devices. Consider it only for PATA and SATAPI.
2179 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2180 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2181 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2182 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2184 action |= ATA_EH_RESET;
2191 /* device has been slowed down, blow error history */
2192 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2193 ata_ering_clear(&dev->ering);
2198 * ata_eh_worth_retry - analyze error and decide whether to retry
2199 * @qc: qc to possibly retry
2201 * Look at the cause of the error and decide if a retry
2202 * might be useful or not. We don't want to retry media errors
2203 * because the drive itself has probably already taken 10-30 seconds
2204 * doing its own internal retries before reporting the failure.
2206 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2208 if (qc->err_mask & AC_ERR_MEDIA)
2209 return 0; /* don't retry media errors */
2210 if (qc->flags & ATA_QCFLAG_IO)
2211 return 1; /* otherwise retry anything from fs stack */
2212 if (qc->err_mask & AC_ERR_INVALID)
2213 return 0; /* don't retry these */
2214 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2218 * ata_eh_link_autopsy - analyze error and determine recovery action
2219 * @link: host link to perform autopsy on
2221 * Analyze why @link failed and determine which recovery actions
2222 * are needed. This function also sets more detailed AC_ERR_*
2223 * values and fills sense data for ATAPI CHECK SENSE.
2226 * Kernel thread context (may sleep).
2228 static void ata_eh_link_autopsy(struct ata_link *link)
2230 struct ata_port *ap = link->ap;
2231 struct ata_eh_context *ehc = &link->eh_context;
2232 struct ata_device *dev;
2233 unsigned int all_err_mask = 0, eflags = 0;
2240 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2243 /* obtain and analyze SError */
2244 rc = sata_scr_read(link, SCR_ERROR, &serror);
2246 ehc->i.serror |= serror;
2247 ata_eh_analyze_serror(link);
2248 } else if (rc != -EOPNOTSUPP) {
2249 /* SError read failed, force reset and probing */
2250 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2251 ehc->i.action |= ATA_EH_RESET;
2252 ehc->i.err_mask |= AC_ERR_OTHER;
2255 /* analyze NCQ failure */
2256 ata_eh_analyze_ncq_error(link);
2258 /* any real error trumps AC_ERR_OTHER */
2259 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2260 ehc->i.err_mask &= ~AC_ERR_OTHER;
2262 all_err_mask |= ehc->i.err_mask;
2264 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2265 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2267 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2268 ata_dev_phys_link(qc->dev) != link)
2271 /* inherit upper level err_mask */
2272 qc->err_mask |= ehc->i.err_mask;
2275 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2277 /* DEV errors are probably spurious in case of ATA_BUS error */
2278 if (qc->err_mask & AC_ERR_ATA_BUS)
2279 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2282 /* any real error trumps unknown error */
2283 if (qc->err_mask & ~AC_ERR_OTHER)
2284 qc->err_mask &= ~AC_ERR_OTHER;
2286 /* SENSE_VALID trumps dev/unknown error and revalidation */
2287 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2288 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2290 /* determine whether the command is worth retrying */
2291 if (ata_eh_worth_retry(qc))
2292 qc->flags |= ATA_QCFLAG_RETRY;
2294 /* accumulate error info */
2295 ehc->i.dev = qc->dev;
2296 all_err_mask |= qc->err_mask;
2297 if (qc->flags & ATA_QCFLAG_IO)
2298 eflags |= ATA_EFLAG_IS_IO;
2299 trace_ata_eh_link_autopsy_qc(qc);
2302 /* enforce default EH actions */
2303 if (ap->pflags & ATA_PFLAG_FROZEN ||
2304 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2305 ehc->i.action |= ATA_EH_RESET;
2306 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2307 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2308 ehc->i.action |= ATA_EH_REVALIDATE;
2310 /* If we have offending qcs and the associated failed device,
2311 * perform per-dev EH action only on the offending device.
2314 ehc->i.dev_action[ehc->i.dev->devno] |=
2315 ehc->i.action & ATA_EH_PERDEV_MASK;
2316 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2319 /* propagate timeout to host link */
2320 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2321 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2323 /* record error and consider speeding down */
2325 if (!dev && ((ata_link_max_devices(link) == 1 &&
2326 ata_dev_enabled(link->device))))
2330 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2331 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2332 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2334 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2339 * ata_eh_autopsy - analyze error and determine recovery action
2340 * @ap: host port to perform autopsy on
2342 * Analyze all links of @ap and determine why they failed and
2343 * which recovery actions are needed.
2346 * Kernel thread context (may sleep).
2348 void ata_eh_autopsy(struct ata_port *ap)
2350 struct ata_link *link;
2352 ata_for_each_link(link, ap, EDGE)
2353 ata_eh_link_autopsy(link);
2355 /* Handle the frigging slave link. Autopsy is done similarly
2356 * but actions and flags are transferred over to the master
2357 * link and handled from there.
2359 if (ap->slave_link) {
2360 struct ata_eh_context *mehc = &ap->link.eh_context;
2361 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2363 /* transfer control flags from master to slave */
2364 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2366 /* perform autopsy on the slave link */
2367 ata_eh_link_autopsy(ap->slave_link);
2369 /* transfer actions from slave to master and clear slave */
2370 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2371 mehc->i.action |= sehc->i.action;
2372 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2373 mehc->i.flags |= sehc->i.flags;
2374 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2377 /* Autopsy of fanout ports can affect host link autopsy.
2378 * Perform host link autopsy last.
2380 if (sata_pmp_attached(ap))
2381 ata_eh_link_autopsy(&ap->link);
2385 * ata_get_cmd_descript - get description for ATA command
2386 * @command: ATA command code to get description for
2388 * Return a textual description of the given command, or NULL if the
2389 * command is not known.
2394 const char *ata_get_cmd_descript(u8 command)
2396 #ifdef CONFIG_ATA_VERBOSE_ERROR
2402 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2403 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2404 { ATA_CMD_STANDBY, "STANDBY" },
2405 { ATA_CMD_IDLE, "IDLE" },
2406 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2407 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2408 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2409 { ATA_CMD_NOP, "NOP" },
2410 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2411 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2412 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2413 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2414 { ATA_CMD_SERVICE, "SERVICE" },
2415 { ATA_CMD_READ, "READ DMA" },
2416 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2417 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2418 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2419 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2420 { ATA_CMD_WRITE, "WRITE DMA" },
2421 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2422 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2423 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2424 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2425 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2426 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2427 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2428 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2429 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2430 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2431 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2432 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2433 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2434 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2435 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2436 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2437 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2438 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2439 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2440 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2441 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2442 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2443 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2444 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2445 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2446 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2447 { ATA_CMD_SLEEP, "SLEEP" },
2448 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2449 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2450 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2451 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2452 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2453 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2454 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2455 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2456 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2457 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2458 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2459 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2460 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2461 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2462 { ATA_CMD_PMP_READ, "READ BUFFER" },
2463 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2464 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2465 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2466 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2467 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2468 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2469 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2470 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2471 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2472 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2473 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2474 { ATA_CMD_SMART, "SMART" },
2475 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2476 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2477 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2478 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2479 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2480 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2481 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2482 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2483 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2484 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2485 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2486 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2487 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2488 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2489 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2490 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2491 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2492 { ATA_CMD_RESTORE, "RECALIBRATE" },
2493 { 0, NULL } /* terminate list */
2497 for (i = 0; cmd_descr[i].text; i++)
2498 if (cmd_descr[i].command == command)
2499 return cmd_descr[i].text;
2504 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2507 * ata_eh_link_report - report error handling to user
2508 * @link: ATA link EH is going on
2510 * Report EH to user.
2515 static void ata_eh_link_report(struct ata_link *link)
2517 struct ata_port *ap = link->ap;
2518 struct ata_eh_context *ehc = &link->eh_context;
2519 const char *frozen, *desc;
2520 char tries_buf[6] = "";
2521 int tag, nr_failed = 0;
2523 if (ehc->i.flags & ATA_EHI_QUIET)
2527 if (ehc->i.desc[0] != '\0')
2530 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2531 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2533 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2534 ata_dev_phys_link(qc->dev) != link ||
2535 ((qc->flags & ATA_QCFLAG_QUIET) &&
2536 qc->err_mask == AC_ERR_DEV))
2538 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2544 if (!nr_failed && !ehc->i.err_mask)
2548 if (ap->pflags & ATA_PFLAG_FROZEN)
2551 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2552 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2556 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2557 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2558 ehc->i.err_mask, link->sactive, ehc->i.serror,
2559 ehc->i.action, frozen, tries_buf);
2561 ata_dev_err(ehc->i.dev, "%s\n", desc);
2563 ata_link_err(link, "exception Emask 0x%x "
2564 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2565 ehc->i.err_mask, link->sactive, ehc->i.serror,
2566 ehc->i.action, frozen, tries_buf);
2568 ata_link_err(link, "%s\n", desc);
2571 #ifdef CONFIG_ATA_VERBOSE_ERROR
2574 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2575 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2576 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2577 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2578 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2579 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2580 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2581 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2582 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2583 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2584 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2585 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2586 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2587 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2588 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2589 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2590 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2591 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2594 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2595 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2596 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2597 char data_buf[20] = "";
2598 char cdb_buf[70] = "";
2600 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2601 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2604 if (qc->dma_dir != DMA_NONE) {
2605 static const char *dma_str[] = {
2606 [DMA_BIDIRECTIONAL] = "bidi",
2607 [DMA_TO_DEVICE] = "out",
2608 [DMA_FROM_DEVICE] = "in",
2610 static const char *prot_str[] = {
2611 [ATA_PROT_UNKNOWN] = "unknown",
2612 [ATA_PROT_NODATA] = "nodata",
2613 [ATA_PROT_PIO] = "pio",
2614 [ATA_PROT_DMA] = "dma",
2615 [ATA_PROT_NCQ] = "ncq dma",
2616 [ATA_PROT_NCQ_NODATA] = "ncq nodata",
2617 [ATAPI_PROT_NODATA] = "nodata",
2618 [ATAPI_PROT_PIO] = "pio",
2619 [ATAPI_PROT_DMA] = "dma",
2622 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2623 prot_str[qc->tf.protocol], qc->nbytes,
2624 dma_str[qc->dma_dir]);
2627 if (ata_is_atapi(qc->tf.protocol)) {
2628 const u8 *cdb = qc->cdb;
2629 size_t cdb_len = qc->dev->cdb_len;
2632 cdb = qc->scsicmd->cmnd;
2633 cdb_len = qc->scsicmd->cmd_len;
2635 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2638 const char *descr = ata_get_cmd_descript(cmd->command);
2640 ata_dev_err(qc->dev, "failed command: %s\n",
2644 ata_dev_err(qc->dev,
2645 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2647 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2648 "Emask 0x%x (%s)%s\n",
2649 cmd->command, cmd->feature, cmd->nsect,
2650 cmd->lbal, cmd->lbam, cmd->lbah,
2651 cmd->hob_feature, cmd->hob_nsect,
2652 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2653 cmd->device, qc->tag, data_buf, cdb_buf,
2654 res->command, res->feature, res->nsect,
2655 res->lbal, res->lbam, res->lbah,
2656 res->hob_feature, res->hob_nsect,
2657 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2658 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2659 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2661 #ifdef CONFIG_ATA_VERBOSE_ERROR
2662 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2663 ATA_SENSE | ATA_ERR)) {
2664 if (res->command & ATA_BUSY)
2665 ata_dev_err(qc->dev, "status: { Busy }\n");
2667 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2668 res->command & ATA_DRDY ? "DRDY " : "",
2669 res->command & ATA_DF ? "DF " : "",
2670 res->command & ATA_DRQ ? "DRQ " : "",
2671 res->command & ATA_SENSE ? "SENSE " : "",
2672 res->command & ATA_ERR ? "ERR " : "");
2675 if (cmd->command != ATA_CMD_PACKET &&
2676 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2677 ATA_IDNF | ATA_ABORTED)))
2678 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2679 res->feature & ATA_ICRC ? "ICRC " : "",
2680 res->feature & ATA_UNC ? "UNC " : "",
2681 res->feature & ATA_AMNF ? "AMNF " : "",
2682 res->feature & ATA_IDNF ? "IDNF " : "",
2683 res->feature & ATA_ABORTED ? "ABRT " : "");
2689 * ata_eh_report - report error handling to user
2690 * @ap: ATA port to report EH about
2692 * Report EH to user.
2697 void ata_eh_report(struct ata_port *ap)
2699 struct ata_link *link;
2701 ata_for_each_link(link, ap, HOST_FIRST)
2702 ata_eh_link_report(link);
2705 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2706 unsigned int *classes, unsigned long deadline,
2709 struct ata_device *dev;
2712 ata_for_each_dev(dev, link, ALL)
2713 classes[dev->devno] = ATA_DEV_UNKNOWN;
2715 return reset(link, classes, deadline);
2718 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2720 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2724 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2729 int ata_eh_reset(struct ata_link *link, int classify,
2730 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2731 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2733 struct ata_port *ap = link->ap;
2734 struct ata_link *slave = ap->slave_link;
2735 struct ata_eh_context *ehc = &link->eh_context;
2736 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2737 unsigned int *classes = ehc->classes;
2738 unsigned int lflags = link->flags;
2739 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2740 int max_tries = 0, try = 0;
2741 struct ata_link *failed_link;
2742 struct ata_device *dev;
2743 unsigned long deadline, now;
2744 ata_reset_fn_t reset;
2745 unsigned long flags;
2752 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2754 if (link->flags & ATA_LFLAG_RST_ONCE)
2756 if (link->flags & ATA_LFLAG_NO_HRST)
2758 if (link->flags & ATA_LFLAG_NO_SRST)
2761 /* make sure each reset attempt is at least COOL_DOWN apart */
2762 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2764 WARN_ON(time_after(ehc->last_reset, now));
2765 deadline = ata_deadline(ehc->last_reset,
2766 ATA_EH_RESET_COOL_DOWN);
2767 if (time_before(now, deadline))
2768 schedule_timeout_uninterruptible(deadline - now);
2771 spin_lock_irqsave(ap->lock, flags);
2772 ap->pflags |= ATA_PFLAG_RESETTING;
2773 spin_unlock_irqrestore(ap->lock, flags);
2775 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2777 ata_for_each_dev(dev, link, ALL) {
2778 /* If we issue an SRST then an ATA drive (not ATAPI)
2779 * may change configuration and be in PIO0 timing. If
2780 * we do a hard reset (or are coming from power on)
2781 * this is true for ATA or ATAPI. Until we've set a
2782 * suitable controller mode we should not touch the
2783 * bus as we may be talking too fast.
2785 dev->pio_mode = XFER_PIO_0;
2786 dev->dma_mode = 0xff;
2788 /* If the controller has a pio mode setup function
2789 * then use it to set the chipset to rights. Don't
2790 * touch the DMA setup as that will be dealt with when
2791 * configuring devices.
2793 if (ap->ops->set_piomode)
2794 ap->ops->set_piomode(ap, dev);
2797 /* prefer hardreset */
2799 ehc->i.action &= ~ATA_EH_RESET;
2802 ehc->i.action |= ATA_EH_HARDRESET;
2803 } else if (softreset) {
2805 ehc->i.action |= ATA_EH_SOFTRESET;
2809 unsigned long deadline = ata_deadline(jiffies,
2810 ATA_EH_PRERESET_TIMEOUT);
2813 sehc->i.action &= ~ATA_EH_RESET;
2814 sehc->i.action |= ehc->i.action;
2817 rc = prereset(link, deadline);
2819 /* If present, do prereset on slave link too. Reset
2820 * is skipped iff both master and slave links report
2821 * -ENOENT or clear ATA_EH_RESET.
2823 if (slave && (rc == 0 || rc == -ENOENT)) {
2826 tmp = prereset(slave, deadline);
2830 ehc->i.action |= sehc->i.action;
2834 if (rc == -ENOENT) {
2835 ata_link_dbg(link, "port disabled--ignoring\n");
2836 ehc->i.action &= ~ATA_EH_RESET;
2838 ata_for_each_dev(dev, link, ALL)
2839 classes[dev->devno] = ATA_DEV_NONE;
2844 "prereset failed (errno=%d)\n",
2849 /* prereset() might have cleared ATA_EH_RESET. If so,
2850 * bang classes, thaw and return.
2852 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2853 ata_for_each_dev(dev, link, ALL)
2854 classes[dev->devno] = ATA_DEV_NONE;
2855 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2856 ata_is_host_link(link))
2857 ata_eh_thaw_port(ap);
2867 if (ata_is_host_link(link))
2868 ata_eh_freeze_port(ap);
2870 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2874 ata_link_info(link, "%s resetting link\n",
2875 reset == softreset ? "soft" : "hard");
2877 /* mark that this EH session started with reset */
2878 ehc->last_reset = jiffies;
2879 if (reset == hardreset)
2880 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2882 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2884 rc = ata_do_reset(link, reset, classes, deadline, true);
2885 if (rc && rc != -EAGAIN) {
2890 /* hardreset slave link if existent */
2891 if (slave && reset == hardreset) {
2895 ata_link_info(slave, "hard resetting link\n");
2897 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2898 tmp = ata_do_reset(slave, reset, classes, deadline,
2906 failed_link = slave;
2912 /* perform follow-up SRST if necessary */
2913 if (reset == hardreset &&
2914 ata_eh_followup_srst_needed(link, rc)) {
2919 "follow-up softreset required but no softreset available\n");
2925 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2926 rc = ata_do_reset(link, reset, classes, deadline, true);
2935 "no reset method available, skipping reset\n");
2936 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2937 lflags |= ATA_LFLAG_ASSUME_ATA;
2941 * Post-reset processing
2943 ata_for_each_dev(dev, link, ALL) {
2944 /* After the reset, the device state is PIO 0 and the
2945 * controller state is undefined. Reset also wakes up
2946 * drives from sleeping mode.
2948 dev->pio_mode = XFER_PIO_0;
2949 dev->flags &= ~ATA_DFLAG_SLEEPING;
2951 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2954 /* apply class override */
2955 if (lflags & ATA_LFLAG_ASSUME_ATA)
2956 classes[dev->devno] = ATA_DEV_ATA;
2957 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2958 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2961 /* record current link speed */
2962 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2963 link->sata_spd = (sstatus >> 4) & 0xf;
2964 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2965 slave->sata_spd = (sstatus >> 4) & 0xf;
2968 if (ata_is_host_link(link))
2969 ata_eh_thaw_port(ap);
2971 /* postreset() should clear hardware SError. Although SError
2972 * is cleared during link resume, clearing SError here is
2973 * necessary as some PHYs raise hotplug events after SRST.
2974 * This introduces race condition where hotplug occurs between
2975 * reset and here. This race is mediated by cross checking
2976 * link onlineness and classification result later.
2979 postreset(link, classes);
2981 postreset(slave, classes);
2985 * Some controllers can't be frozen very well and may set spurious
2986 * error conditions during reset. Clear accumulated error
2987 * information and re-thaw the port if frozen. As reset is the
2988 * final recovery action and we cross check link onlineness against
2989 * device classification later, no hotplug event is lost by this.
2991 spin_lock_irqsave(link->ap->lock, flags);
2992 memset(&link->eh_info, 0, sizeof(link->eh_info));
2994 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2995 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2996 spin_unlock_irqrestore(link->ap->lock, flags);
2998 if (ap->pflags & ATA_PFLAG_FROZEN)
2999 ata_eh_thaw_port(ap);
3002 * Make sure onlineness and classification result correspond.
3003 * Hotplug could have happened during reset and some
3004 * controllers fail to wait while a drive is spinning up after
3005 * being hotplugged causing misdetection. By cross checking
3006 * link on/offlineness and classification result, those
3007 * conditions can be reliably detected and retried.
3010 ata_for_each_dev(dev, link, ALL) {
3011 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
3012 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
3013 ata_dev_dbg(dev, "link online but device misclassified\n");
3014 classes[dev->devno] = ATA_DEV_NONE;
3017 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3018 if (ata_class_enabled(classes[dev->devno]))
3020 "link offline, clearing class %d to NONE\n",
3021 classes[dev->devno]);
3022 classes[dev->devno] = ATA_DEV_NONE;
3023 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
3025 "link status unknown, clearing UNKNOWN to NONE\n");
3026 classes[dev->devno] = ATA_DEV_NONE;
3030 if (classify && nr_unknown) {
3031 if (try < max_tries) {
3033 "link online but %d devices misclassified, retrying\n",
3040 "link online but %d devices misclassified, "
3041 "device detection might fail\n", nr_unknown);
3044 /* reset successful, schedule revalidation */
3045 ata_eh_done(link, NULL, ATA_EH_RESET);
3047 ata_eh_done(slave, NULL, ATA_EH_RESET);
3048 ehc->last_reset = jiffies; /* update to completion time */
3049 ehc->i.action |= ATA_EH_REVALIDATE;
3050 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
3054 /* clear hotplug flag */
3055 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3057 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3059 spin_lock_irqsave(ap->lock, flags);
3060 ap->pflags &= ~ATA_PFLAG_RESETTING;
3061 spin_unlock_irqrestore(ap->lock, flags);
3066 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
3067 if (!ata_is_host_link(link) &&
3068 sata_scr_read(link, SCR_STATUS, &sstatus))
3071 if (try >= max_tries) {
3073 * Thaw host port even if reset failed, so that the port
3074 * can be retried on the next phy event. This risks
3075 * repeated EH runs but seems to be a better tradeoff than
3076 * shutting down a port after a botched hotplug attempt.
3078 if (ata_is_host_link(link))
3079 ata_eh_thaw_port(ap);
3084 if (time_before(now, deadline)) {
3085 unsigned long delta = deadline - now;
3087 ata_link_warn(failed_link,
3088 "reset failed (errno=%d), retrying in %u secs\n",
3089 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3093 delta = schedule_timeout_uninterruptible(delta);
3098 * While disks spinup behind PMP, some controllers fail sending SRST.
3099 * They need to be reset - as well as the PMP - before retrying.
3101 if (rc == -ERESTART) {
3102 if (ata_is_host_link(link))
3103 ata_eh_thaw_port(ap);
3107 if (try == max_tries - 1) {
3108 sata_down_spd_limit(link, 0);
3110 sata_down_spd_limit(slave, 0);
3111 } else if (rc == -EPIPE)
3112 sata_down_spd_limit(failed_link, 0);
3119 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3121 struct ata_link *link;
3122 struct ata_device *dev;
3123 unsigned long flags;
3126 * This function can be thought of as an extended version of
3127 * ata_eh_about_to_do() specially crafted to accommodate the
3128 * requirements of ATA_EH_PARK handling. Since the EH thread
3129 * does not leave the do {} while () loop in ata_eh_recover as
3130 * long as the timeout for a park request to *one* device on
3131 * the port has not expired, and since we still want to pick
3132 * up park requests to other devices on the same port or
3133 * timeout updates for the same device, we have to pull
3134 * ATA_EH_PARK actions from eh_info into eh_context.i
3135 * ourselves at the beginning of each pass over the loop.
3137 * Additionally, all write accesses to &ap->park_req_pending
3138 * through reinit_completion() (see below) or complete_all()
3139 * (see ata_scsi_park_store()) are protected by the host lock.
3140 * As a result we have that park_req_pending.done is zero on
3141 * exit from this function, i.e. when ATA_EH_PARK actions for
3142 * *all* devices on port ap have been pulled into the
3143 * respective eh_context structs. If, and only if,
3144 * park_req_pending.done is non-zero by the time we reach
3145 * wait_for_completion_timeout(), another ATA_EH_PARK action
3146 * has been scheduled for at least one of the devices on port
3147 * ap and we have to cycle over the do {} while () loop in
3148 * ata_eh_recover() again.
3151 spin_lock_irqsave(ap->lock, flags);
3152 reinit_completion(&ap->park_req_pending);
3153 ata_for_each_link(link, ap, EDGE) {
3154 ata_for_each_dev(dev, link, ALL) {
3155 struct ata_eh_info *ehi = &link->eh_info;
3157 link->eh_context.i.dev_action[dev->devno] |=
3158 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3159 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3162 spin_unlock_irqrestore(ap->lock, flags);
3165 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3167 struct ata_eh_context *ehc = &dev->link->eh_context;
3168 struct ata_taskfile tf;
3169 unsigned int err_mask;
3171 ata_tf_init(dev, &tf);
3173 ehc->unloaded_mask |= 1 << dev->devno;
3174 tf.command = ATA_CMD_IDLEIMMEDIATE;
3180 ehc->unloaded_mask &= ~(1 << dev->devno);
3181 tf.command = ATA_CMD_CHK_POWER;
3184 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3185 tf.protocol = ATA_PROT_NODATA;
3186 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3187 if (park && (err_mask || tf.lbal != 0xc4)) {
3188 ata_dev_err(dev, "head unload failed!\n");
3189 ehc->unloaded_mask &= ~(1 << dev->devno);
3193 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3194 struct ata_device **r_failed_dev)
3196 struct ata_port *ap = link->ap;
3197 struct ata_eh_context *ehc = &link->eh_context;
3198 struct ata_device *dev;
3199 unsigned int new_mask = 0;
3200 unsigned long flags;
3205 /* For PATA drive side cable detection to work, IDENTIFY must
3206 * be done backwards such that PDIAG- is released by the slave
3207 * device before the master device is identified.
3209 ata_for_each_dev(dev, link, ALL_REVERSE) {
3210 unsigned int action = ata_eh_dev_action(dev);
3211 unsigned int readid_flags = 0;
3213 if (ehc->i.flags & ATA_EHI_DID_RESET)
3214 readid_flags |= ATA_READID_POSTRESET;
3216 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3217 WARN_ON(dev->class == ATA_DEV_PMP);
3219 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3224 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3225 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3230 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3232 /* Configuration may have changed, reconfigure
3235 ehc->i.flags |= ATA_EHI_SETMODE;
3237 /* schedule the scsi_rescan_device() here */
3238 schedule_work(&(ap->scsi_rescan_task));
3239 } else if (dev->class == ATA_DEV_UNKNOWN &&
3240 ehc->tries[dev->devno] &&
3241 ata_class_enabled(ehc->classes[dev->devno])) {
3242 /* Temporarily set dev->class, it will be
3243 * permanently set once all configurations are
3244 * complete. This is necessary because new
3245 * device configuration is done in two
3248 dev->class = ehc->classes[dev->devno];
3250 if (dev->class == ATA_DEV_PMP)
3251 rc = sata_pmp_attach(dev);
3253 rc = ata_dev_read_id(dev, &dev->class,
3254 readid_flags, dev->id);
3256 /* read_id might have changed class, store and reset */
3257 ehc->classes[dev->devno] = dev->class;
3258 dev->class = ATA_DEV_UNKNOWN;
3262 /* clear error info accumulated during probe */
3263 ata_ering_clear(&dev->ering);
3264 new_mask |= 1 << dev->devno;
3267 /* IDENTIFY was issued to non-existent
3268 * device. No need to reset. Just
3269 * thaw and ignore the device.
3271 ata_eh_thaw_port(ap);
3279 /* PDIAG- should have been released, ask cable type if post-reset */
3280 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3281 if (ap->ops->cable_detect)
3282 ap->cbl = ap->ops->cable_detect(ap);
3286 /* Configure new devices forward such that user doesn't see
3287 * device detection messages backwards.
3289 ata_for_each_dev(dev, link, ALL) {
3290 if (!(new_mask & (1 << dev->devno)))
3293 dev->class = ehc->classes[dev->devno];
3295 if (dev->class == ATA_DEV_PMP)
3298 ehc->i.flags |= ATA_EHI_PRINTINFO;
3299 rc = ata_dev_configure(dev);
3300 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3302 dev->class = ATA_DEV_UNKNOWN;
3306 spin_lock_irqsave(ap->lock, flags);
3307 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3308 spin_unlock_irqrestore(ap->lock, flags);
3310 /* new device discovered, configure xfermode */
3311 ehc->i.flags |= ATA_EHI_SETMODE;
3317 *r_failed_dev = dev;
3318 DPRINTK("EXIT rc=%d\n", rc);
3323 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3324 * @link: link on which timings will be programmed
3325 * @r_failed_dev: out parameter for failed device
3327 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3328 * ata_set_mode() fails, pointer to the failing device is
3329 * returned in @r_failed_dev.
3332 * PCI/etc. bus probe sem.
3335 * 0 on success, negative errno otherwise
3337 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3339 struct ata_port *ap = link->ap;
3340 struct ata_device *dev;
3343 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3344 ata_for_each_dev(dev, link, ENABLED) {
3345 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3346 struct ata_ering_entry *ent;
3348 ent = ata_ering_top(&dev->ering);
3350 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3354 /* has private set_mode? */
3355 if (ap->ops->set_mode)
3356 rc = ap->ops->set_mode(link, r_failed_dev);
3358 rc = ata_do_set_mode(link, r_failed_dev);
3360 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3361 ata_for_each_dev(dev, link, ENABLED) {
3362 struct ata_eh_context *ehc = &link->eh_context;
3363 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3364 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3366 if (dev->xfer_mode != saved_xfer_mode ||
3367 ata_ncq_enabled(dev) != saved_ncq)
3368 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3375 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3376 * @dev: ATAPI device to clear UA for
3378 * Resets and other operations can make an ATAPI device raise
3379 * UNIT ATTENTION which causes the next operation to fail. This
3380 * function clears UA.
3383 * EH context (may sleep).
3386 * 0 on success, -errno on failure.
3388 static int atapi_eh_clear_ua(struct ata_device *dev)
3392 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3393 u8 *sense_buffer = dev->link->ap->sector_buf;
3395 unsigned int err_mask;
3397 err_mask = atapi_eh_tur(dev, &sense_key);
3398 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3400 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3405 if (!err_mask || sense_key != UNIT_ATTENTION)
3408 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3410 ata_dev_warn(dev, "failed to clear "
3411 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3416 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3423 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3424 * @dev: ATA device which may need FLUSH retry
3426 * If @dev failed FLUSH, it needs to be reported upper layer
3427 * immediately as it means that @dev failed to remap and already
3428 * lost at least a sector and further FLUSH retrials won't make
3429 * any difference to the lost sector. However, if FLUSH failed
3430 * for other reasons, for example transmission error, FLUSH needs
3433 * This function determines whether FLUSH failure retry is
3434 * necessary and performs it if so.
3437 * 0 if EH can continue, -errno if EH needs to be repeated.
3439 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3441 struct ata_link *link = dev->link;
3442 struct ata_port *ap = link->ap;
3443 struct ata_queued_cmd *qc;
3444 struct ata_taskfile tf;
3445 unsigned int err_mask;
3448 /* did flush fail for this device? */
3449 if (!ata_tag_valid(link->active_tag))
3452 qc = __ata_qc_from_tag(ap, link->active_tag);
3453 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3454 qc->tf.command != ATA_CMD_FLUSH))
3457 /* if the device failed it, it should be reported to upper layers */
3458 if (qc->err_mask & AC_ERR_DEV)
3461 /* flush failed for some other reason, give it another shot */
3462 ata_tf_init(dev, &tf);
3464 tf.command = qc->tf.command;
3465 tf.flags |= ATA_TFLAG_DEVICE;
3466 tf.protocol = ATA_PROT_NODATA;
3468 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3469 tf.command, qc->err_mask);
3471 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3474 * FLUSH is complete but there's no way to
3475 * successfully complete a failed command from EH.
3476 * Making sure retry is allowed at least once and
3477 * retrying it should do the trick - whatever was in
3478 * the cache is already on the platter and this won't
3479 * cause infinite loop.
3481 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3483 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3487 /* if device failed it, report it to upper layers */
3488 if (err_mask & AC_ERR_DEV) {
3489 qc->err_mask |= AC_ERR_DEV;
3491 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3499 * ata_eh_set_lpm - configure SATA interface power management
3500 * @link: link to configure power management
3501 * @policy: the link power management policy
3502 * @r_failed_dev: out parameter for failed device
3504 * Enable SATA Interface power management. This will enable
3505 * Device Interface Power Management (DIPM) for min_power
3506 * policy, and then call driver specific callbacks for
3507 * enabling Host Initiated Power management.
3513 * 0 on success, -errno on failure.
3515 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3516 struct ata_device **r_failed_dev)
3518 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3519 struct ata_eh_context *ehc = &link->eh_context;
3520 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3521 enum ata_lpm_policy old_policy = link->lpm_policy;
3522 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3523 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3524 unsigned int err_mask;
3527 /* if the link or host doesn't do LPM, noop */
3528 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3532 * DIPM is enabled only for MIN_POWER as some devices
3533 * misbehave when the host NACKs transition to SLUMBER. Order
3534 * device and link configurations such that the host always
3535 * allows DIPM requests.
3537 ata_for_each_dev(dev, link, ENABLED) {
3538 bool hipm = ata_id_has_hipm(dev->id);
3539 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3541 /* find the first enabled and LPM enabled devices */
3545 if (!lpm_dev && (hipm || dipm))
3548 hints &= ~ATA_LPM_EMPTY;
3550 hints &= ~ATA_LPM_HIPM;
3552 /* disable DIPM before changing link config */
3553 if (policy != ATA_LPM_MIN_POWER && dipm) {
3554 err_mask = ata_dev_set_feature(dev,
3555 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3556 if (err_mask && err_mask != AC_ERR_DEV) {
3558 "failed to disable DIPM, Emask 0x%x\n",
3567 rc = ap->ops->set_lpm(link, policy, hints);
3568 if (!rc && ap->slave_link)
3569 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3571 rc = sata_pmp_set_lpm(link, policy, hints);
3574 * Attribute link config failure to the first (LPM) enabled
3575 * device on the link.
3578 if (rc == -EOPNOTSUPP) {
3579 link->flags |= ATA_LFLAG_NO_LPM;
3582 dev = lpm_dev ? lpm_dev : link_dev;
3587 * Low level driver acked the transition. Issue DIPM command
3588 * with the new policy set.
3590 link->lpm_policy = policy;
3591 if (ap && ap->slave_link)
3592 ap->slave_link->lpm_policy = policy;
3594 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3595 ata_for_each_dev(dev, link, ENABLED) {
3596 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3597 ata_id_has_dipm(dev->id)) {
3598 err_mask = ata_dev_set_feature(dev,
3599 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3600 if (err_mask && err_mask != AC_ERR_DEV) {
3602 "failed to enable DIPM, Emask 0x%x\n",
3610 link->last_lpm_change = jiffies;
3611 link->flags |= ATA_LFLAG_CHANGED;
3616 /* restore the old policy */
3617 link->lpm_policy = old_policy;
3618 if (ap && ap->slave_link)
3619 ap->slave_link->lpm_policy = old_policy;
3621 /* if no device or only one more chance is left, disable LPM */
3622 if (!dev || ehc->tries[dev->devno] <= 2) {
3623 ata_link_warn(link, "disabling LPM on the link\n");
3624 link->flags |= ATA_LFLAG_NO_LPM;
3627 *r_failed_dev = dev;
3631 int ata_link_nr_enabled(struct ata_link *link)
3633 struct ata_device *dev;
3636 ata_for_each_dev(dev, link, ENABLED)
3641 static int ata_link_nr_vacant(struct ata_link *link)
3643 struct ata_device *dev;
3646 ata_for_each_dev(dev, link, ALL)
3647 if (dev->class == ATA_DEV_UNKNOWN)
3652 static int ata_eh_skip_recovery(struct ata_link *link)
3654 struct ata_port *ap = link->ap;
3655 struct ata_eh_context *ehc = &link->eh_context;
3656 struct ata_device *dev;
3658 /* skip disabled links */
3659 if (link->flags & ATA_LFLAG_DISABLED)
3662 /* skip if explicitly requested */
3663 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3666 /* thaw frozen port and recover failed devices */
3667 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3670 /* reset at least once if reset is requested */
3671 if ((ehc->i.action & ATA_EH_RESET) &&
3672 !(ehc->i.flags & ATA_EHI_DID_RESET))
3675 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3676 ata_for_each_dev(dev, link, ALL) {
3677 if (dev->class == ATA_DEV_UNKNOWN &&
3678 ehc->classes[dev->devno] != ATA_DEV_NONE)
3685 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3687 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3688 u64 now = get_jiffies_64();
3689 int *trials = void_arg;
3691 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3692 (ent->timestamp < now - min(now, interval)))
3699 static int ata_eh_schedule_probe(struct ata_device *dev)
3701 struct ata_eh_context *ehc = &dev->link->eh_context;
3702 struct ata_link *link = ata_dev_phys_link(dev);
3705 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3706 (ehc->did_probe_mask & (1 << dev->devno)))
3709 ata_eh_detach_dev(dev);
3711 ehc->did_probe_mask |= (1 << dev->devno);
3712 ehc->i.action |= ATA_EH_RESET;
3713 ehc->saved_xfer_mode[dev->devno] = 0;
3714 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3716 /* the link maybe in a deep sleep, wake it up */
3717 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3718 if (ata_is_host_link(link))
3719 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3722 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3726 /* Record and count probe trials on the ering. The specific
3727 * error mask used is irrelevant. Because a successful device
3728 * detection clears the ering, this count accumulates only if
3729 * there are consecutive failed probes.
3731 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3732 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3733 * forced to 1.5Gbps.
3735 * This is to work around cases where failed link speed
3736 * negotiation results in device misdetection leading to
3737 * infinite DEVXCHG or PHRDY CHG events.
3739 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3740 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3742 if (trials > ATA_EH_PROBE_TRIALS)
3743 sata_down_spd_limit(link, 1);
3748 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3750 struct ata_eh_context *ehc = &dev->link->eh_context;
3752 /* -EAGAIN from EH routine indicates retry without prejudice.
3753 * The requester is responsible for ensuring forward progress.
3756 ehc->tries[dev->devno]--;
3760 /* device missing or wrong IDENTIFY data, schedule probing */
3761 ehc->i.probe_mask |= (1 << dev->devno);
3763 /* give it just one more chance */
3764 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3766 if (ehc->tries[dev->devno] == 1) {
3767 /* This is the last chance, better to slow
3768 * down than lose it.
3770 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3771 if (dev->pio_mode > XFER_PIO_0)
3772 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3776 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3777 /* disable device if it has used up all its chances */
3778 ata_dev_disable(dev);
3780 /* detach if offline */
3781 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3782 ata_eh_detach_dev(dev);
3784 /* schedule probe if necessary */
3785 if (ata_eh_schedule_probe(dev)) {
3786 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3787 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3788 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3793 ehc->i.action |= ATA_EH_RESET;
3799 * ata_eh_recover - recover host port after error
3800 * @ap: host port to recover
3801 * @prereset: prereset method (can be NULL)
3802 * @softreset: softreset method (can be NULL)
3803 * @hardreset: hardreset method (can be NULL)
3804 * @postreset: postreset method (can be NULL)
3805 * @r_failed_link: out parameter for failed link
3807 * This is the alpha and omega, eum and yang, heart and soul of
3808 * libata exception handling. On entry, actions required to
3809 * recover each link and hotplug requests are recorded in the
3810 * link's eh_context. This function executes all the operations
3811 * with appropriate retrials and fallbacks to resurrect failed
3812 * devices, detach goners and greet newcomers.
3815 * Kernel thread context (may sleep).
3818 * 0 on success, -errno on failure.
3820 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3821 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3822 ata_postreset_fn_t postreset,
3823 struct ata_link **r_failed_link)
3825 struct ata_link *link;
3826 struct ata_device *dev;
3828 unsigned long flags, deadline;
3832 /* prep for recovery */
3833 ata_for_each_link(link, ap, EDGE) {
3834 struct ata_eh_context *ehc = &link->eh_context;
3836 /* re-enable link? */
3837 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3838 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3839 spin_lock_irqsave(ap->lock, flags);
3840 link->flags &= ~ATA_LFLAG_DISABLED;
3841 spin_unlock_irqrestore(ap->lock, flags);
3842 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3845 ata_for_each_dev(dev, link, ALL) {
3846 if (link->flags & ATA_LFLAG_NO_RETRY)
3847 ehc->tries[dev->devno] = 1;
3849 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3851 /* collect port action mask recorded in dev actions */
3852 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3853 ~ATA_EH_PERDEV_MASK;
3854 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3856 /* process hotplug request */
3857 if (dev->flags & ATA_DFLAG_DETACH)
3858 ata_eh_detach_dev(dev);
3860 /* schedule probe if necessary */
3861 if (!ata_dev_enabled(dev))
3862 ata_eh_schedule_probe(dev);
3869 /* if UNLOADING, finish immediately */
3870 if (ap->pflags & ATA_PFLAG_UNLOADING)
3874 ata_for_each_link(link, ap, EDGE) {
3875 struct ata_eh_context *ehc = &link->eh_context;
3877 /* skip EH if possible. */
3878 if (ata_eh_skip_recovery(link))
3881 ata_for_each_dev(dev, link, ALL)
3882 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3886 ata_for_each_link(link, ap, EDGE) {
3887 struct ata_eh_context *ehc = &link->eh_context;
3889 if (!(ehc->i.action & ATA_EH_RESET))
3892 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3893 prereset, softreset, hardreset, postreset);
3895 ata_link_err(link, "reset failed, giving up\n");
3904 * clears ATA_EH_PARK in eh_info and resets
3905 * ap->park_req_pending
3907 ata_eh_pull_park_action(ap);
3910 ata_for_each_link(link, ap, EDGE) {
3911 ata_for_each_dev(dev, link, ALL) {
3912 struct ata_eh_context *ehc = &link->eh_context;
3915 if (dev->class != ATA_DEV_ATA &&
3916 dev->class != ATA_DEV_ZAC)
3918 if (!(ehc->i.dev_action[dev->devno] &
3921 tmp = dev->unpark_deadline;
3922 if (time_before(deadline, tmp))
3924 else if (time_before_eq(tmp, jiffies))
3926 if (ehc->unloaded_mask & (1 << dev->devno))
3929 ata_eh_park_issue_cmd(dev, 1);
3934 if (time_before_eq(deadline, now))
3938 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3942 ata_for_each_link(link, ap, EDGE) {
3943 ata_for_each_dev(dev, link, ALL) {
3944 if (!(link->eh_context.unloaded_mask &
3948 ata_eh_park_issue_cmd(dev, 0);
3949 ata_eh_done(link, dev, ATA_EH_PARK);
3955 ata_for_each_link(link, ap, PMP_FIRST) {
3956 struct ata_eh_context *ehc = &link->eh_context;
3958 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3961 /* revalidate existing devices and attach new ones */
3962 rc = ata_eh_revalidate_and_attach(link, &dev);
3966 /* if PMP got attached, return, pmp EH will take care of it */
3967 if (link->device->class == ATA_DEV_PMP) {
3972 /* configure transfer mode if necessary */
3973 if (ehc->i.flags & ATA_EHI_SETMODE) {
3974 rc = ata_set_mode(link, &dev);
3977 ehc->i.flags &= ~ATA_EHI_SETMODE;
3980 /* If reset has been issued, clear UA to avoid
3981 * disrupting the current users of the device.
3983 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3984 ata_for_each_dev(dev, link, ALL) {
3985 if (dev->class != ATA_DEV_ATAPI)
3987 rc = atapi_eh_clear_ua(dev);
3990 if (zpodd_dev_enabled(dev))
3991 zpodd_post_poweron(dev);
3995 /* retry flush if necessary */
3996 ata_for_each_dev(dev, link, ALL) {
3997 if (dev->class != ATA_DEV_ATA &&
3998 dev->class != ATA_DEV_ZAC)
4000 rc = ata_eh_maybe_retry_flush(dev);
4006 /* configure link power saving */
4007 if (link->lpm_policy != ap->target_lpm_policy) {
4008 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
4013 /* this link is okay now */
4020 ata_eh_handle_dev_fail(dev, rc);
4022 if (ap->pflags & ATA_PFLAG_FROZEN) {
4023 /* PMP reset requires working host port.
4024 * Can't retry if it's frozen.
4026 if (sata_pmp_attached(ap))
4036 if (rc && r_failed_link)
4037 *r_failed_link = link;
4039 DPRINTK("EXIT, rc=%d\n", rc);
4044 * ata_eh_finish - finish up EH
4045 * @ap: host port to finish EH for
4047 * Recovery is complete. Clean up EH states and retry or finish
4053 void ata_eh_finish(struct ata_port *ap)
4057 /* retry or finish qcs */
4058 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
4059 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
4061 if (!(qc->flags & ATA_QCFLAG_FAILED))
4065 /* FIXME: Once EH migration is complete,
4066 * generate sense data in this function,
4067 * considering both err_mask and tf.
4069 if (qc->flags & ATA_QCFLAG_RETRY)
4070 ata_eh_qc_retry(qc);
4072 ata_eh_qc_complete(qc);
4074 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
4075 ata_eh_qc_complete(qc);
4077 /* feed zero TF to sense generation */
4078 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4079 ata_eh_qc_retry(qc);
4084 /* make sure nr_active_links is zero after EH */
4085 WARN_ON(ap->nr_active_links);
4086 ap->nr_active_links = 0;
4090 * ata_do_eh - do standard error handling
4091 * @ap: host port to handle error for
4093 * @prereset: prereset method (can be NULL)
4094 * @softreset: softreset method (can be NULL)
4095 * @hardreset: hardreset method (can be NULL)
4096 * @postreset: postreset method (can be NULL)
4098 * Perform standard error handling sequence.
4101 * Kernel thread context (may sleep).
4103 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4104 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4105 ata_postreset_fn_t postreset)
4107 struct ata_device *dev;
4113 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4116 ata_for_each_dev(dev, &ap->link, ALL)
4117 ata_dev_disable(dev);
4124 * ata_std_error_handler - standard error handler
4125 * @ap: host port to handle error for
4127 * Standard error handler
4130 * Kernel thread context (may sleep).
4132 void ata_std_error_handler(struct ata_port *ap)
4134 struct ata_port_operations *ops = ap->ops;
4135 ata_reset_fn_t hardreset = ops->hardreset;
4137 /* ignore built-in hardreset if SCR access is not available */
4138 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4141 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4146 * ata_eh_handle_port_suspend - perform port suspend operation
4147 * @ap: port to suspend
4152 * Kernel thread context (may sleep).
4154 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4156 unsigned long flags;
4158 struct ata_device *dev;
4160 /* are we suspending? */
4161 spin_lock_irqsave(ap->lock, flags);
4162 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4163 ap->pm_mesg.event & PM_EVENT_RESUME) {
4164 spin_unlock_irqrestore(ap->lock, flags);
4167 spin_unlock_irqrestore(ap->lock, flags);
4169 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4172 * If we have a ZPODD attached, check its zero
4173 * power ready status before the port is frozen.
4174 * Only needed for runtime suspend.
4176 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4177 ata_for_each_dev(dev, &ap->link, ENABLED) {
4178 if (zpodd_dev_enabled(dev))
4179 zpodd_on_suspend(dev);
4183 /* tell ACPI we're suspending */
4184 rc = ata_acpi_on_suspend(ap);
4189 ata_eh_freeze_port(ap);
4191 if (ap->ops->port_suspend)
4192 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4194 ata_acpi_set_state(ap, ap->pm_mesg);
4196 /* update the flags */
4197 spin_lock_irqsave(ap->lock, flags);
4199 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4201 ap->pflags |= ATA_PFLAG_SUSPENDED;
4202 else if (ap->pflags & ATA_PFLAG_FROZEN)
4203 ata_port_schedule_eh(ap);
4205 spin_unlock_irqrestore(ap->lock, flags);
4211 * ata_eh_handle_port_resume - perform port resume operation
4212 * @ap: port to resume
4217 * Kernel thread context (may sleep).
4219 static void ata_eh_handle_port_resume(struct ata_port *ap)
4221 struct ata_link *link;
4222 struct ata_device *dev;
4223 unsigned long flags;
4226 /* are we resuming? */
4227 spin_lock_irqsave(ap->lock, flags);
4228 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4229 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4230 spin_unlock_irqrestore(ap->lock, flags);
4233 spin_unlock_irqrestore(ap->lock, flags);
4235 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4238 * Error timestamps are in jiffies which doesn't run while
4239 * suspended and PHY events during resume isn't too uncommon.
4240 * When the two are combined, it can lead to unnecessary speed
4241 * downs if the machine is suspended and resumed repeatedly.
4242 * Clear error history.
4244 ata_for_each_link(link, ap, HOST_FIRST)
4245 ata_for_each_dev(dev, link, ALL)
4246 ata_ering_clear(&dev->ering);
4248 ata_acpi_set_state(ap, ap->pm_mesg);
4250 if (ap->ops->port_resume)
4251 rc = ap->ops->port_resume(ap);
4253 /* tell ACPI that we're resuming */
4254 ata_acpi_on_resume(ap);
4256 /* update the flags */
4257 spin_lock_irqsave(ap->lock, flags);
4258 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4259 spin_unlock_irqrestore(ap->lock, flags);
4261 #endif /* CONFIG_PM */