1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * libata-eh.c - libata error handling
5 * Copyright 2006 Tejun Heo <htejun@gmail.com>
7 * libata documentation is available via 'make {ps|pdf}docs',
8 * as Documentation/driver-api/libata.rst
10 * Hardware documentation available from http://www.t13.org/ and
11 * http://www.sata-io.org/
14 #include <linux/kernel.h>
15 #include <linux/blkdev.h>
16 #include <linux/export.h>
17 #include <linux/pci.h>
18 #include <scsi/scsi.h>
19 #include <scsi/scsi_host.h>
20 #include <scsi/scsi_eh.h>
21 #include <scsi/scsi_device.h>
22 #include <scsi/scsi_cmnd.h>
23 #include <scsi/scsi_dbg.h>
24 #include "../scsi/scsi_transport_api.h"
26 #include <linux/libata.h>
28 #include <trace/events/libata.h>
32 /* speed down verdicts */
33 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
39 ATA_EFLAG_IS_IO = (1 << 0),
40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
41 ATA_EFLAG_OLD_ER = (1 << 31),
43 /* error categories */
46 ATA_ECAT_TOUT_HSM = 2,
48 ATA_ECAT_DUBIOUS_NONE = 4,
49 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
51 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
54 ATA_EH_CMD_DFL_TIMEOUT = 5000,
56 /* always put at least this amount of time between resets */
57 ATA_EH_RESET_COOL_DOWN = 5000,
59 /* Waiting in ->prereset can never be reliable. It's
60 * sometimes nice to wait there but it can't be depended upon;
61 * otherwise, we wouldn't be resetting. Just give it enough
62 * time for most drives to spin up.
64 ATA_EH_PRERESET_TIMEOUT = 10000,
65 ATA_EH_FASTDRAIN_INTERVAL = 3000,
69 /* probe speed down parameters, see ata_eh_schedule_probe() */
70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
71 ATA_EH_PROBE_TRIALS = 2,
74 /* The following table determines how we sequence resets. Each entry
75 * represents timeout for that try. The first try can be soft or
76 * hardreset. All others are hardreset if available. In most cases
77 * the first reset w/ 10sec timeout should succeed. Following entries
78 * are mostly for error handling, hotplug and those outlier devices that
79 * take an exceptionally long time to recover from reset.
81 static const unsigned int ata_eh_reset_timeouts[] = {
82 10000, /* most drives spin up by 10sec */
83 10000, /* > 99% working drives spin up before 20sec */
84 35000, /* give > 30 secs of idleness for outlier devices */
85 5000, /* and sweet one last chance */
86 UINT_MAX, /* > 1 min has elapsed, give up */
89 static const unsigned int ata_eh_identify_timeouts[] = {
90 5000, /* covers > 99% of successes and not too boring on failures */
91 10000, /* combined time till here is enough even for media access */
92 30000, /* for true idiots */
96 static const unsigned int ata_eh_revalidate_timeouts[] = {
97 15000, /* Some drives are slow to read log pages when waking-up */
98 15000, /* combined time till here is enough even for media access */
102 static const unsigned int ata_eh_flush_timeouts[] = {
103 15000, /* be generous with flush */
105 30000, /* and even more generous */
109 static const unsigned int ata_eh_other_timeouts[] = {
110 5000, /* same rationale as identify timeout */
112 /* but no merciful 30sec for other commands, it just isn't worth it */
116 struct ata_eh_cmd_timeout_ent {
118 const unsigned int *timeouts;
121 /* The following table determines timeouts to use for EH internal
122 * commands. Each table entry is a command class and matches the
123 * commands the entry applies to and the timeout table to use.
125 * On the retry after a command timed out, the next timeout value from
126 * the table is used. If the table doesn't contain further entries,
127 * the last value is used.
129 * ehc->cmd_timeout_idx keeps track of which timeout to use per
130 * command class, so if SET_FEATURES times out on the first try, the
131 * next try will use the second timeout value only for that class.
133 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
134 static const struct ata_eh_cmd_timeout_ent
135 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
136 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
137 .timeouts = ata_eh_identify_timeouts, },
138 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
139 .timeouts = ata_eh_revalidate_timeouts, },
140 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_FEATURES),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
149 .timeouts = ata_eh_flush_timeouts },
153 static void __ata_port_freeze(struct ata_port *ap);
154 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
155 struct ata_device **r_failed_dev);
157 static void ata_eh_handle_port_suspend(struct ata_port *ap);
158 static void ata_eh_handle_port_resume(struct ata_port *ap);
159 #else /* CONFIG_PM */
160 static void ata_eh_handle_port_suspend(struct ata_port *ap)
163 static void ata_eh_handle_port_resume(struct ata_port *ap)
165 #endif /* CONFIG_PM */
167 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
168 const char *fmt, va_list args)
170 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
171 ATA_EH_DESC_LEN - ehi->desc_len,
176 * __ata_ehi_push_desc - push error description without adding separator
178 * @fmt: printf format string
180 * Format string according to @fmt and append it to @ehi->desc.
183 * spin_lock_irqsave(host lock)
185 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
190 __ata_ehi_pushv_desc(ehi, fmt, args);
193 EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
196 * ata_ehi_push_desc - push error description with separator
198 * @fmt: printf format string
200 * Format string according to @fmt and append it to @ehi->desc.
201 * If @ehi->desc is not empty, ", " is added in-between.
204 * spin_lock_irqsave(host lock)
206 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
211 __ata_ehi_push_desc(ehi, ", ");
214 __ata_ehi_pushv_desc(ehi, fmt, args);
217 EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
220 * ata_ehi_clear_desc - clean error description
226 * spin_lock_irqsave(host lock)
228 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
233 EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
236 * ata_port_desc - append port description
237 * @ap: target ATA port
238 * @fmt: printf format string
240 * Format string according to @fmt and append it to port
241 * description. If port description is not empty, " " is added
242 * in-between. This function is to be used while initializing
243 * ata_host. The description is printed on host registration.
248 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
252 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
254 if (ap->link.eh_info.desc_len)
255 __ata_ehi_push_desc(&ap->link.eh_info, " ");
258 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
261 EXPORT_SYMBOL_GPL(ata_port_desc);
265 * ata_port_pbar_desc - append PCI BAR description
266 * @ap: target ATA port
267 * @bar: target PCI BAR
268 * @offset: offset into PCI BAR
269 * @name: name of the area
271 * If @offset is negative, this function formats a string which
272 * contains the name, address, size and type of the BAR and
273 * appends it to the port description. If @offset is zero or
274 * positive, only name and offsetted address is appended.
279 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
282 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
284 unsigned long long start, len;
286 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
288 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
291 start = (unsigned long long)pci_resource_start(pdev, bar);
292 len = (unsigned long long)pci_resource_len(pdev, bar);
295 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
297 ata_port_desc(ap, "%s 0x%llx", name,
298 start + (unsigned long long)offset);
300 EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
301 #endif /* CONFIG_PCI */
303 static int ata_lookup_timeout_table(u8 cmd)
307 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
310 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
319 * ata_internal_cmd_timeout - determine timeout for an internal command
320 * @dev: target device
321 * @cmd: internal command to be issued
323 * Determine timeout for internal command @cmd for @dev.
329 * Determined timeout.
331 unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
333 struct ata_eh_context *ehc = &dev->link->eh_context;
334 int ent = ata_lookup_timeout_table(cmd);
338 return ATA_EH_CMD_DFL_TIMEOUT;
340 idx = ehc->cmd_timeout_idx[dev->devno][ent];
341 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
345 * ata_internal_cmd_timed_out - notification for internal command timeout
346 * @dev: target device
347 * @cmd: internal command which timed out
349 * Notify EH that internal command @cmd for @dev timed out. This
350 * function should be called only for commands whose timeouts are
351 * determined using ata_internal_cmd_timeout().
356 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
358 struct ata_eh_context *ehc = &dev->link->eh_context;
359 int ent = ata_lookup_timeout_table(cmd);
365 idx = ehc->cmd_timeout_idx[dev->devno][ent];
366 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
367 ehc->cmd_timeout_idx[dev->devno][ent]++;
370 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
371 unsigned int err_mask)
373 struct ata_ering_entry *ent;
378 ering->cursor %= ATA_ERING_SIZE;
380 ent = &ering->ring[ering->cursor];
381 ent->eflags = eflags;
382 ent->err_mask = err_mask;
383 ent->timestamp = get_jiffies_64();
386 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
388 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
395 int ata_ering_map(struct ata_ering *ering,
396 int (*map_fn)(struct ata_ering_entry *, void *),
400 struct ata_ering_entry *ent;
404 ent = &ering->ring[idx];
407 rc = map_fn(ent, arg);
410 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
411 } while (idx != ering->cursor);
416 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
418 ent->eflags |= ATA_EFLAG_OLD_ER;
422 static void ata_ering_clear(struct ata_ering *ering)
424 ata_ering_map(ering, ata_ering_clear_cb, NULL);
427 static unsigned int ata_eh_dev_action(struct ata_device *dev)
429 struct ata_eh_context *ehc = &dev->link->eh_context;
431 return ehc->i.action | ehc->i.dev_action[dev->devno];
434 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
435 struct ata_eh_info *ehi, unsigned int action)
437 struct ata_device *tdev;
440 ehi->action &= ~action;
441 ata_for_each_dev(tdev, link, ALL)
442 ehi->dev_action[tdev->devno] &= ~action;
444 /* doesn't make sense for port-wide EH actions */
445 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
447 /* break ehi->action into ehi->dev_action */
448 if (ehi->action & action) {
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] |=
451 ehi->action & action;
452 ehi->action &= ~action;
455 /* turn off the specified per-dev action */
456 ehi->dev_action[dev->devno] &= ~action;
461 * ata_eh_acquire - acquire EH ownership
462 * @ap: ATA port to acquire EH ownership for
464 * Acquire EH ownership for @ap. This is the basic exclusion
465 * mechanism for ports sharing a host. Only one port hanging off
466 * the same host can claim the ownership of EH.
471 void ata_eh_acquire(struct ata_port *ap)
473 mutex_lock(&ap->host->eh_mutex);
474 WARN_ON_ONCE(ap->host->eh_owner);
475 ap->host->eh_owner = current;
479 * ata_eh_release - release EH ownership
480 * @ap: ATA port to release EH ownership for
482 * Release EH ownership for @ap if the caller. The caller must
483 * have acquired EH ownership using ata_eh_acquire() previously.
488 void ata_eh_release(struct ata_port *ap)
490 WARN_ON_ONCE(ap->host->eh_owner != current);
491 ap->host->eh_owner = NULL;
492 mutex_unlock(&ap->host->eh_mutex);
495 static void ata_eh_unload(struct ata_port *ap)
497 struct ata_link *link;
498 struct ata_device *dev;
501 /* Restore SControl IPM and SPD for the next driver and
502 * disable attached devices.
504 ata_for_each_link(link, ap, PMP_FIRST) {
505 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
506 ata_for_each_dev(dev, link, ALL)
507 ata_dev_disable(dev);
510 /* freeze and set UNLOADED */
511 spin_lock_irqsave(ap->lock, flags);
513 ata_port_freeze(ap); /* won't be thawed */
514 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
515 ap->pflags |= ATA_PFLAG_UNLOADED;
517 spin_unlock_irqrestore(ap->lock, flags);
521 * ata_scsi_error - SCSI layer error handler callback
522 * @host: SCSI host on which error occurred
524 * Handles SCSI-layer-thrown error events.
527 * Inherited from SCSI layer (none, can sleep)
532 void ata_scsi_error(struct Scsi_Host *host)
534 struct ata_port *ap = ata_shost_to_port(host);
536 LIST_HEAD(eh_work_q);
538 spin_lock_irqsave(host->host_lock, flags);
539 list_splice_init(&host->eh_cmd_q, &eh_work_q);
540 spin_unlock_irqrestore(host->host_lock, flags);
542 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
544 /* If we timed raced normal completion and there is nothing to
545 recover nr_timedout == 0 why exactly are we doing error recovery ? */
546 ata_scsi_port_error_handler(host, ap);
548 /* finish or retry handled scmd's and clean up */
549 WARN_ON(!list_empty(&eh_work_q));
554 * ata_scsi_cmd_error_handler - error callback for a list of commands
555 * @host: scsi host containing the port
556 * @ap: ATA port within the host
557 * @eh_work_q: list of commands to process
559 * process the given list of commands and return those finished to the
560 * ap->eh_done_q. This function is the first part of the libata error
561 * handler which processes a given list of failed commands.
563 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
564 struct list_head *eh_work_q)
568 struct scsi_cmnd *scmd, *tmp;
571 /* make sure sff pio task is not running */
572 ata_sff_flush_pio_task(ap);
574 /* synchronize with host lock and sort out timeouts */
577 * For EH, all qcs are finished in one of three ways -
578 * normal completion, error completion, and SCSI timeout.
579 * Both completions can race against SCSI timeout. When normal
580 * completion wins, the qc never reaches EH. When error
581 * completion wins, the qc has ATA_QCFLAG_EH set.
583 * When SCSI timeout wins, things are a bit more complex.
584 * Normal or error completion can occur after the timeout but
585 * before this point. In such cases, both types of
586 * completions are honored. A scmd is determined to have
587 * timed out iff its associated qc is active and not failed.
589 spin_lock_irqsave(ap->lock, flags);
592 * This must occur under the ap->lock as we don't want
593 * a polled recovery to race the real interrupt handler
595 * The lost_interrupt handler checks for any completed but
596 * non-notified command and completes much like an IRQ handler.
598 * We then fall into the error recovery code which will treat
599 * this as if normal completion won the race
601 if (ap->ops->lost_interrupt)
602 ap->ops->lost_interrupt(ap);
604 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
605 struct ata_queued_cmd *qc;
607 ata_qc_for_each_raw(ap, qc, i) {
608 if (qc->flags & ATA_QCFLAG_ACTIVE &&
613 if (i < ATA_MAX_QUEUE) {
614 /* the scmd has an associated qc */
615 if (!(qc->flags & ATA_QCFLAG_EH)) {
616 /* which hasn't failed yet, timeout */
617 qc->err_mask |= AC_ERR_TIMEOUT;
618 qc->flags |= ATA_QCFLAG_EH;
622 /* Normal completion occurred after
623 * SCSI timeout but before this point.
624 * Successfully complete it.
626 scmd->retries = scmd->allowed;
627 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
632 * If we have timed out qcs. They belong to EH from
633 * this point but the state of the controller is
634 * unknown. Freeze the port to make sure the IRQ
635 * handler doesn't diddle with those qcs. This must
636 * be done atomically w.r.t. setting ATA_QCFLAG_EH.
639 __ata_port_freeze(ap);
641 /* initialize eh_tries */
642 ap->eh_tries = ATA_EH_MAX_TRIES;
644 spin_unlock_irqrestore(ap->lock, flags);
646 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
649 * ata_scsi_port_error_handler - recover the port after the commands
650 * @host: SCSI host containing the port
653 * Handle the recovery of the port @ap after all the commands
654 * have been recovered.
656 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
659 struct ata_link *link;
661 /* acquire EH ownership */
664 /* kill fast drain timer */
665 del_timer_sync(&ap->fastdrain_timer);
667 /* process port resume request */
668 ata_eh_handle_port_resume(ap);
670 /* fetch & clear EH info */
671 spin_lock_irqsave(ap->lock, flags);
673 ata_for_each_link(link, ap, HOST_FIRST) {
674 struct ata_eh_context *ehc = &link->eh_context;
675 struct ata_device *dev;
677 memset(&link->eh_context, 0, sizeof(link->eh_context));
678 link->eh_context.i = link->eh_info;
679 memset(&link->eh_info, 0, sizeof(link->eh_info));
681 ata_for_each_dev(dev, link, ENABLED) {
682 int devno = dev->devno;
684 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
685 if (ata_ncq_enabled(dev))
686 ehc->saved_ncq_enabled |= 1 << devno;
690 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
691 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
692 ap->excl_link = NULL; /* don't maintain exclusion over EH */
694 spin_unlock_irqrestore(ap->lock, flags);
696 /* invoke EH, skip if unloading or suspended */
697 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
698 ap->ops->error_handler(ap);
700 /* if unloading, commence suicide */
701 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
702 !(ap->pflags & ATA_PFLAG_UNLOADED))
707 /* process port suspend request */
708 ata_eh_handle_port_suspend(ap);
711 * Exception might have happened after ->error_handler recovered the
712 * port but before this point. Repeat EH in such case.
714 spin_lock_irqsave(ap->lock, flags);
716 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
717 if (--ap->eh_tries) {
718 spin_unlock_irqrestore(ap->lock, flags);
722 "EH pending after %d tries, giving up\n",
724 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
727 /* this run is complete, make sure EH info is clear */
728 ata_for_each_link(link, ap, HOST_FIRST)
729 memset(&link->eh_info, 0, sizeof(link->eh_info));
732 * end eh (clear host_eh_scheduled) while holding ap->lock such that if
733 * exception occurs after this point but before EH completion, SCSI
734 * midlayer will re-initiate EH.
738 spin_unlock_irqrestore(ap->lock, flags);
741 scsi_eh_flush_done_q(&ap->eh_done_q);
744 spin_lock_irqsave(ap->lock, flags);
746 if (ap->pflags & ATA_PFLAG_LOADING)
747 ap->pflags &= ~ATA_PFLAG_LOADING;
748 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
749 !(ap->flags & ATA_FLAG_SAS_HOST))
750 schedule_delayed_work(&ap->hotplug_task, 0);
752 if (ap->pflags & ATA_PFLAG_RECOVERED)
753 ata_port_info(ap, "EH complete\n");
755 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
757 /* tell wait_eh that we're done */
758 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
759 wake_up_all(&ap->eh_wait_q);
761 spin_unlock_irqrestore(ap->lock, flags);
763 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
766 * ata_port_wait_eh - Wait for the currently pending EH to complete
767 * @ap: Port to wait EH for
769 * Wait until the currently pending EH is complete.
772 * Kernel thread context (may sleep).
774 void ata_port_wait_eh(struct ata_port *ap)
780 spin_lock_irqsave(ap->lock, flags);
782 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
783 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
784 spin_unlock_irqrestore(ap->lock, flags);
786 spin_lock_irqsave(ap->lock, flags);
788 finish_wait(&ap->eh_wait_q, &wait);
790 spin_unlock_irqrestore(ap->lock, flags);
792 /* make sure SCSI EH is complete */
793 if (scsi_host_in_recovery(ap->scsi_host)) {
798 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
800 static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
802 struct ata_queued_cmd *qc;
806 /* count only non-internal commands */
807 ata_qc_for_each(ap, qc, tag) {
815 void ata_eh_fastdrain_timerfn(struct timer_list *t)
817 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
821 spin_lock_irqsave(ap->lock, flags);
823 cnt = ata_eh_nr_in_flight(ap);
829 if (cnt == ap->fastdrain_cnt) {
830 struct ata_queued_cmd *qc;
833 /* No progress during the last interval, tag all
834 * in-flight qcs as timed out and freeze the port.
836 ata_qc_for_each(ap, qc, tag) {
838 qc->err_mask |= AC_ERR_TIMEOUT;
843 /* some qcs have finished, give it another chance */
844 ap->fastdrain_cnt = cnt;
845 ap->fastdrain_timer.expires =
846 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
847 add_timer(&ap->fastdrain_timer);
851 spin_unlock_irqrestore(ap->lock, flags);
855 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
856 * @ap: target ATA port
857 * @fastdrain: activate fast drain
859 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
860 * is non-zero and EH wasn't pending before. Fast drain ensures
861 * that EH kicks in in timely manner.
864 * spin_lock_irqsave(host lock)
866 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
870 /* already scheduled? */
871 if (ap->pflags & ATA_PFLAG_EH_PENDING)
874 ap->pflags |= ATA_PFLAG_EH_PENDING;
879 /* do we have in-flight qcs? */
880 cnt = ata_eh_nr_in_flight(ap);
884 /* activate fast drain */
885 ap->fastdrain_cnt = cnt;
886 ap->fastdrain_timer.expires =
887 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
888 add_timer(&ap->fastdrain_timer);
892 * ata_qc_schedule_eh - schedule qc for error handling
893 * @qc: command to schedule error handling for
895 * Schedule error handling for @qc. EH will kick in as soon as
896 * other commands are drained.
899 * spin_lock_irqsave(host lock)
901 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
903 struct ata_port *ap = qc->ap;
905 qc->flags |= ATA_QCFLAG_EH;
906 ata_eh_set_pending(ap, 1);
908 /* The following will fail if timeout has already expired.
909 * ata_scsi_error() takes care of such scmds on EH entry.
910 * Note that ATA_QCFLAG_EH is unconditionally set after
911 * this function completes.
913 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
917 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
918 * @ap: ATA port to schedule EH for
920 * LOCKING: inherited from ata_port_schedule_eh
921 * spin_lock_irqsave(host lock)
923 void ata_std_sched_eh(struct ata_port *ap)
925 if (ap->pflags & ATA_PFLAG_INITIALIZING)
928 ata_eh_set_pending(ap, 1);
929 scsi_schedule_eh(ap->scsi_host);
931 trace_ata_std_sched_eh(ap);
933 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
936 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
937 * @ap: ATA port to end EH for
939 * In the libata object model there is a 1:1 mapping of ata_port to
940 * shost, so host fields can be directly manipulated under ap->lock, in
941 * the libsas case we need to hold a lock at the ha->level to coordinate
945 * spin_lock_irqsave(host lock)
947 void ata_std_end_eh(struct ata_port *ap)
949 struct Scsi_Host *host = ap->scsi_host;
951 host->host_eh_scheduled = 0;
953 EXPORT_SYMBOL(ata_std_end_eh);
957 * ata_port_schedule_eh - schedule error handling without a qc
958 * @ap: ATA port to schedule EH for
960 * Schedule error handling for @ap. EH will kick in as soon as
961 * all commands are drained.
964 * spin_lock_irqsave(host lock)
966 void ata_port_schedule_eh(struct ata_port *ap)
968 /* see: ata_std_sched_eh, unless you know better */
969 ap->ops->sched_eh(ap);
971 EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
973 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
975 struct ata_queued_cmd *qc;
976 int tag, nr_aborted = 0;
978 /* we're gonna abort all commands, no need for fast drain */
979 ata_eh_set_pending(ap, 0);
981 /* include internal tag in iteration */
982 ata_qc_for_each_with_internal(ap, qc, tag) {
983 if (qc && (!link || qc->dev->link == link)) {
984 qc->flags |= ATA_QCFLAG_EH;
991 ata_port_schedule_eh(ap);
997 * ata_link_abort - abort all qc's on the link
998 * @link: ATA link to abort qc's for
1000 * Abort all active qc's active on @link and schedule EH.
1003 * spin_lock_irqsave(host lock)
1006 * Number of aborted qc's.
1008 int ata_link_abort(struct ata_link *link)
1010 return ata_do_link_abort(link->ap, link);
1012 EXPORT_SYMBOL_GPL(ata_link_abort);
1015 * ata_port_abort - abort all qc's on the port
1016 * @ap: ATA port to abort qc's for
1018 * Abort all active qc's of @ap and schedule EH.
1021 * spin_lock_irqsave(host_set lock)
1024 * Number of aborted qc's.
1026 int ata_port_abort(struct ata_port *ap)
1028 return ata_do_link_abort(ap, NULL);
1030 EXPORT_SYMBOL_GPL(ata_port_abort);
1033 * __ata_port_freeze - freeze port
1034 * @ap: ATA port to freeze
1036 * This function is called when HSM violation or some other
1037 * condition disrupts normal operation of the port. Frozen port
1038 * is not allowed to perform any operation until the port is
1039 * thawed, which usually follows a successful reset.
1041 * ap->ops->freeze() callback can be used for freezing the port
1042 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1043 * port cannot be frozen hardware-wise, the interrupt handler
1044 * must ack and clear interrupts unconditionally while the port
1048 * spin_lock_irqsave(host lock)
1050 static void __ata_port_freeze(struct ata_port *ap)
1052 if (ap->ops->freeze)
1053 ap->ops->freeze(ap);
1055 ap->pflags |= ATA_PFLAG_FROZEN;
1057 trace_ata_port_freeze(ap);
1061 * ata_port_freeze - abort & freeze port
1062 * @ap: ATA port to freeze
1064 * Abort and freeze @ap. The freeze operation must be called
1065 * first, because some hardware requires special operations
1066 * before the taskfile registers are accessible.
1069 * spin_lock_irqsave(host lock)
1072 * Number of aborted commands.
1074 int ata_port_freeze(struct ata_port *ap)
1076 __ata_port_freeze(ap);
1078 return ata_port_abort(ap);
1080 EXPORT_SYMBOL_GPL(ata_port_freeze);
1083 * ata_eh_freeze_port - EH helper to freeze port
1084 * @ap: ATA port to freeze
1091 void ata_eh_freeze_port(struct ata_port *ap)
1093 unsigned long flags;
1095 spin_lock_irqsave(ap->lock, flags);
1096 __ata_port_freeze(ap);
1097 spin_unlock_irqrestore(ap->lock, flags);
1099 EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1102 * ata_eh_thaw_port - EH helper to thaw port
1103 * @ap: ATA port to thaw
1105 * Thaw frozen port @ap.
1110 void ata_eh_thaw_port(struct ata_port *ap)
1112 unsigned long flags;
1114 spin_lock_irqsave(ap->lock, flags);
1116 ap->pflags &= ~ATA_PFLAG_FROZEN;
1121 spin_unlock_irqrestore(ap->lock, flags);
1123 trace_ata_port_thaw(ap);
1126 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1131 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1133 struct ata_port *ap = qc->ap;
1134 struct scsi_cmnd *scmd = qc->scsicmd;
1135 unsigned long flags;
1137 spin_lock_irqsave(ap->lock, flags);
1138 qc->scsidone = ata_eh_scsidone;
1139 __ata_qc_complete(qc);
1140 WARN_ON(ata_tag_valid(qc->tag));
1141 spin_unlock_irqrestore(ap->lock, flags);
1143 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1147 * ata_eh_qc_complete - Complete an active ATA command from EH
1148 * @qc: Command to complete
1150 * Indicate to the mid and upper layers that an ATA command has
1151 * completed. To be used from EH.
1153 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1155 struct scsi_cmnd *scmd = qc->scsicmd;
1156 scmd->retries = scmd->allowed;
1157 __ata_eh_qc_complete(qc);
1161 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1162 * @qc: Command to retry
1164 * Indicate to the mid and upper layers that an ATA command
1165 * should be retried. To be used from EH.
1167 * SCSI midlayer limits the number of retries to scmd->allowed.
1168 * scmd->allowed is incremented for commands which get retried
1169 * due to unrelated failures (qc->err_mask is zero).
1171 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1173 struct scsi_cmnd *scmd = qc->scsicmd;
1176 __ata_eh_qc_complete(qc);
1180 * ata_dev_disable - disable ATA device
1181 * @dev: ATA device to disable
1188 void ata_dev_disable(struct ata_device *dev)
1190 if (!ata_dev_enabled(dev))
1193 ata_dev_warn(dev, "disable device\n");
1194 ata_acpi_on_disable(dev);
1195 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1198 /* From now till the next successful probe, ering is used to
1199 * track probe failures. Clear accumulated device error info.
1201 ata_ering_clear(&dev->ering);
1203 EXPORT_SYMBOL_GPL(ata_dev_disable);
1206 * ata_eh_detach_dev - detach ATA device
1207 * @dev: ATA device to detach
1214 void ata_eh_detach_dev(struct ata_device *dev)
1216 struct ata_link *link = dev->link;
1217 struct ata_port *ap = link->ap;
1218 struct ata_eh_context *ehc = &link->eh_context;
1219 unsigned long flags;
1221 ata_dev_disable(dev);
1223 spin_lock_irqsave(ap->lock, flags);
1225 dev->flags &= ~ATA_DFLAG_DETACH;
1227 if (ata_scsi_offline_dev(dev)) {
1228 dev->flags |= ATA_DFLAG_DETACHED;
1229 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1232 /* clear per-dev EH info */
1233 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1234 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1235 ehc->saved_xfer_mode[dev->devno] = 0;
1236 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1238 spin_unlock_irqrestore(ap->lock, flags);
1242 * ata_eh_about_to_do - about to perform eh_action
1243 * @link: target ATA link
1244 * @dev: target ATA dev for per-dev action (can be NULL)
1245 * @action: action about to be performed
1247 * Called just before performing EH actions to clear related bits
1248 * in @link->eh_info such that eh actions are not unnecessarily
1254 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1255 unsigned int action)
1257 struct ata_port *ap = link->ap;
1258 struct ata_eh_info *ehi = &link->eh_info;
1259 struct ata_eh_context *ehc = &link->eh_context;
1260 unsigned long flags;
1262 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1264 spin_lock_irqsave(ap->lock, flags);
1266 ata_eh_clear_action(link, dev, ehi, action);
1268 /* About to take EH action, set RECOVERED. Ignore actions on
1269 * slave links as master will do them again.
1271 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1272 ap->pflags |= ATA_PFLAG_RECOVERED;
1274 spin_unlock_irqrestore(ap->lock, flags);
1278 * ata_eh_done - EH action complete
1279 * @link: ATA link for which EH actions are complete
1280 * @dev: target ATA dev for per-dev action (can be NULL)
1281 * @action: action just completed
1283 * Called right after performing EH actions to clear related bits
1284 * in @link->eh_context.
1289 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1290 unsigned int action)
1292 struct ata_eh_context *ehc = &link->eh_context;
1294 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1296 ata_eh_clear_action(link, dev, &ehc->i, action);
1300 * ata_err_string - convert err_mask to descriptive string
1301 * @err_mask: error mask to convert to string
1303 * Convert @err_mask to descriptive string. Errors are
1304 * prioritized according to severity and only the most severe
1305 * error is reported.
1311 * Descriptive string for @err_mask
1313 static const char *ata_err_string(unsigned int err_mask)
1315 if (err_mask & AC_ERR_HOST_BUS)
1316 return "host bus error";
1317 if (err_mask & AC_ERR_ATA_BUS)
1318 return "ATA bus error";
1319 if (err_mask & AC_ERR_TIMEOUT)
1321 if (err_mask & AC_ERR_HSM)
1322 return "HSM violation";
1323 if (err_mask & AC_ERR_SYSTEM)
1324 return "internal error";
1325 if (err_mask & AC_ERR_MEDIA)
1326 return "media error";
1327 if (err_mask & AC_ERR_INVALID)
1328 return "invalid argument";
1329 if (err_mask & AC_ERR_DEV)
1330 return "device error";
1331 if (err_mask & AC_ERR_NCQ)
1333 if (err_mask & AC_ERR_NODEV_HINT)
1334 return "Polling detection error";
1335 return "unknown error";
1339 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1340 * @dev: target ATAPI device
1341 * @r_sense_key: out parameter for sense_key
1343 * Perform ATAPI TEST_UNIT_READY.
1346 * EH context (may sleep).
1349 * 0 on success, AC_ERR_* mask on failure.
1351 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1353 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1354 struct ata_taskfile tf;
1355 unsigned int err_mask;
1357 ata_tf_init(dev, &tf);
1359 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1360 tf.command = ATA_CMD_PACKET;
1361 tf.protocol = ATAPI_PROT_NODATA;
1363 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1364 if (err_mask == AC_ERR_DEV)
1365 *r_sense_key = tf.error >> 4;
1370 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1371 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1373 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1374 * SENSE. This function is an EH helper.
1377 * Kernel thread context (may sleep).
1380 * true if sense data could be fetched, false otherwise.
1382 static bool ata_eh_request_sense(struct ata_queued_cmd *qc)
1384 struct scsi_cmnd *cmd = qc->scsicmd;
1385 struct ata_device *dev = qc->dev;
1386 struct ata_taskfile tf;
1387 unsigned int err_mask;
1389 if (ata_port_is_frozen(qc->ap)) {
1390 ata_dev_warn(dev, "sense data available but port frozen\n");
1394 if (!ata_id_sense_reporting_enabled(dev->id)) {
1395 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1399 ata_tf_init(dev, &tf);
1400 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1401 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1402 tf.command = ATA_CMD_REQ_SENSE_DATA;
1403 tf.protocol = ATA_PROT_NODATA;
1405 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1406 /* Ignore err_mask; ATA_ERR might be set */
1407 if (tf.status & ATA_SENSE) {
1408 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1409 /* Set sense without also setting scsicmd->result */
1410 scsi_build_sense_buffer(dev->flags & ATA_DFLAG_D_SENSE,
1411 cmd->sense_buffer, tf.lbah,
1413 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1417 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1418 tf.status, err_mask);
1425 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1426 * @dev: device to perform REQUEST_SENSE to
1427 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1428 * @dfl_sense_key: default sense key to use
1430 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1431 * SENSE. This function is EH helper.
1434 * Kernel thread context (may sleep).
1437 * 0 on success, AC_ERR_* mask on failure
1439 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1440 u8 *sense_buf, u8 dfl_sense_key)
1442 u8 cdb[ATAPI_CDB_LEN] =
1443 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1444 struct ata_port *ap = dev->link->ap;
1445 struct ata_taskfile tf;
1447 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1449 /* initialize sense_buf with the error register,
1450 * for the case where they are -not- overwritten
1452 sense_buf[0] = 0x70;
1453 sense_buf[2] = dfl_sense_key;
1455 /* some devices time out if garbage left in tf */
1456 ata_tf_init(dev, &tf);
1458 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1459 tf.command = ATA_CMD_PACKET;
1461 /* is it pointless to prefer PIO for "safety reasons"? */
1462 if (ap->flags & ATA_FLAG_PIO_DMA) {
1463 tf.protocol = ATAPI_PROT_DMA;
1464 tf.feature |= ATAPI_PKT_DMA;
1466 tf.protocol = ATAPI_PROT_PIO;
1467 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1471 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1472 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1476 * ata_eh_analyze_serror - analyze SError for a failed port
1477 * @link: ATA link to analyze SError for
1479 * Analyze SError if available and further determine cause of
1485 static void ata_eh_analyze_serror(struct ata_link *link)
1487 struct ata_eh_context *ehc = &link->eh_context;
1488 u32 serror = ehc->i.serror;
1489 unsigned int err_mask = 0, action = 0;
1492 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1493 err_mask |= AC_ERR_ATA_BUS;
1494 action |= ATA_EH_RESET;
1496 if (serror & SERR_PROTOCOL) {
1497 err_mask |= AC_ERR_HSM;
1498 action |= ATA_EH_RESET;
1500 if (serror & SERR_INTERNAL) {
1501 err_mask |= AC_ERR_SYSTEM;
1502 action |= ATA_EH_RESET;
1505 /* Determine whether a hotplug event has occurred. Both
1506 * SError.N/X are considered hotplug events for enabled or
1507 * host links. For disabled PMP links, only N bit is
1508 * considered as X bit is left at 1 for link plugging.
1510 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1511 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1512 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1513 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1515 hotplug_mask = SERR_PHYRDY_CHG;
1517 if (serror & hotplug_mask)
1518 ata_ehi_hotplugged(&ehc->i);
1520 ehc->i.err_mask |= err_mask;
1521 ehc->i.action |= action;
1525 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1526 * @qc: qc to analyze
1528 * Analyze taskfile of @qc and further determine cause of
1529 * failure. This function also requests ATAPI sense data if
1533 * Kernel thread context (may sleep).
1536 * Determined recovery action
1538 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1540 const struct ata_taskfile *tf = &qc->result_tf;
1541 unsigned int tmp, action = 0;
1542 u8 stat = tf->status, err = tf->error;
1544 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1545 qc->err_mask |= AC_ERR_HSM;
1546 return ATA_EH_RESET;
1549 if (stat & (ATA_ERR | ATA_DF)) {
1550 qc->err_mask |= AC_ERR_DEV;
1552 * Sense data reporting does not work if the
1553 * device fault bit is set.
1561 switch (qc->dev->class) {
1565 * Fetch the sense data explicitly if:
1566 * -It was a non-NCQ command that failed, or
1567 * -It was a NCQ command that failed, but the sense data
1568 * was not included in the NCQ command error log
1569 * (i.e. NCQ autosense is not supported by the device).
1571 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) &&
1572 (stat & ATA_SENSE) && ata_eh_request_sense(qc))
1573 set_status_byte(qc->scsicmd, SAM_STAT_CHECK_CONDITION);
1575 qc->err_mask |= AC_ERR_ATA_BUS;
1576 if (err & (ATA_UNC | ATA_AMNF))
1577 qc->err_mask |= AC_ERR_MEDIA;
1579 qc->err_mask |= AC_ERR_INVALID;
1583 if (!ata_port_is_frozen(qc->ap)) {
1584 tmp = atapi_eh_request_sense(qc->dev,
1585 qc->scsicmd->sense_buffer,
1586 qc->result_tf.error >> 4);
1588 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1590 qc->err_mask |= tmp;
1594 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1595 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1597 * SUCCESS here means that the sense code could be
1598 * evaluated and should be passed to the upper layers
1599 * for correct evaluation.
1600 * FAILED means the sense code could not be interpreted
1601 * and the device would need to be reset.
1602 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1603 * command would need to be retried.
1605 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1606 qc->flags |= ATA_QCFLAG_RETRY;
1607 qc->err_mask |= AC_ERR_OTHER;
1608 } else if (ret != SUCCESS) {
1609 qc->err_mask |= AC_ERR_HSM;
1612 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1613 action |= ATA_EH_RESET;
1618 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1623 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1627 base = ATA_ECAT_DUBIOUS_NONE;
1629 if (err_mask & AC_ERR_ATA_BUS)
1630 return base + ATA_ECAT_ATA_BUS;
1632 if (err_mask & AC_ERR_TIMEOUT)
1633 return base + ATA_ECAT_TOUT_HSM;
1635 if (eflags & ATA_EFLAG_IS_IO) {
1636 if (err_mask & AC_ERR_HSM)
1637 return base + ATA_ECAT_TOUT_HSM;
1639 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1640 return base + ATA_ECAT_UNK_DEV;
1646 struct speed_down_verdict_arg {
1649 int nr_errors[ATA_ECAT_NR];
1652 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1654 struct speed_down_verdict_arg *arg = void_arg;
1657 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1660 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1662 arg->nr_errors[cat]++;
1668 * ata_eh_speed_down_verdict - Determine speed down verdict
1669 * @dev: Device of interest
1671 * This function examines error ring of @dev and determines
1672 * whether NCQ needs to be turned off, transfer speed should be
1673 * stepped down, or falling back to PIO is necessary.
1675 * ECAT_ATA_BUS : ATA_BUS error for any command
1677 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1680 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1682 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1683 * data transfer hasn't been verified.
1687 * NCQ_OFF : Turn off NCQ.
1689 * SPEED_DOWN : Speed down transfer speed but don't fall back
1692 * FALLBACK_TO_PIO : Fall back to PIO.
1694 * Even if multiple verdicts are returned, only one action is
1695 * taken per error. An action triggered by non-DUBIOUS errors
1696 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1697 * This is to expedite speed down decisions right after device is
1698 * initially configured.
1700 * The following are speed down rules. #1 and #2 deal with
1703 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1704 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1706 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1707 * occurred during last 5 mins, NCQ_OFF.
1709 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1710 * occurred during last 5 mins, FALLBACK_TO_PIO
1712 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1713 * during last 10 mins, NCQ_OFF.
1715 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1716 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1719 * Inherited from caller.
1722 * OR of ATA_EH_SPDN_* flags.
1724 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1726 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1727 u64 j64 = get_jiffies_64();
1728 struct speed_down_verdict_arg arg;
1729 unsigned int verdict = 0;
1731 /* scan past 5 mins of error history */
1732 memset(&arg, 0, sizeof(arg));
1733 arg.since = j64 - min(j64, j5mins);
1734 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1736 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1737 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1738 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1739 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1741 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1742 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1743 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1745 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1746 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1747 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1748 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1750 /* scan past 10 mins of error history */
1751 memset(&arg, 0, sizeof(arg));
1752 arg.since = j64 - min(j64, j10mins);
1753 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1755 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1756 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1757 verdict |= ATA_EH_SPDN_NCQ_OFF;
1759 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1760 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1761 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1762 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1768 * ata_eh_speed_down - record error and speed down if necessary
1769 * @dev: Failed device
1770 * @eflags: mask of ATA_EFLAG_* flags
1771 * @err_mask: err_mask of the error
1773 * Record error and examine error history to determine whether
1774 * adjusting transmission speed is necessary. It also sets
1775 * transmission limits appropriately if such adjustment is
1779 * Kernel thread context (may sleep).
1782 * Determined recovery action.
1784 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1785 unsigned int eflags, unsigned int err_mask)
1787 struct ata_link *link = ata_dev_phys_link(dev);
1789 unsigned int verdict;
1790 unsigned int action = 0;
1792 /* don't bother if Cat-0 error */
1793 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1796 /* record error and determine whether speed down is necessary */
1797 ata_ering_record(&dev->ering, eflags, err_mask);
1798 verdict = ata_eh_speed_down_verdict(dev);
1801 if ((verdict & ATA_EH_SPDN_NCQ_OFF) && ata_ncq_enabled(dev)) {
1802 dev->flags |= ATA_DFLAG_NCQ_OFF;
1803 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1808 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1809 /* speed down SATA link speed if possible */
1810 if (sata_down_spd_limit(link, 0) == 0) {
1811 action |= ATA_EH_RESET;
1815 /* lower transfer mode */
1816 if (dev->spdn_cnt < 2) {
1817 static const int dma_dnxfer_sel[] =
1818 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1819 static const int pio_dnxfer_sel[] =
1820 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1823 if (dev->xfer_shift != ATA_SHIFT_PIO)
1824 sel = dma_dnxfer_sel[dev->spdn_cnt];
1826 sel = pio_dnxfer_sel[dev->spdn_cnt];
1830 if (ata_down_xfermask_limit(dev, sel) == 0) {
1831 action |= ATA_EH_RESET;
1837 /* Fall back to PIO? Slowing down to PIO is meaningless for
1838 * SATA ATA devices. Consider it only for PATA and SATAPI.
1840 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1841 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1842 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1843 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1845 action |= ATA_EH_RESET;
1852 /* device has been slowed down, blow error history */
1853 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1854 ata_ering_clear(&dev->ering);
1859 * ata_eh_worth_retry - analyze error and decide whether to retry
1860 * @qc: qc to possibly retry
1862 * Look at the cause of the error and decide if a retry
1863 * might be useful or not. We don't want to retry media errors
1864 * because the drive itself has probably already taken 10-30 seconds
1865 * doing its own internal retries before reporting the failure.
1867 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1869 if (qc->err_mask & AC_ERR_MEDIA)
1870 return 0; /* don't retry media errors */
1871 if (qc->flags & ATA_QCFLAG_IO)
1872 return 1; /* otherwise retry anything from fs stack */
1873 if (qc->err_mask & AC_ERR_INVALID)
1874 return 0; /* don't retry these */
1875 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1879 * ata_eh_quiet - check if we need to be quiet about a command error
1882 * Look at the qc flags anbd its scsi command request flags to determine
1883 * if we need to be quiet about the command failure.
1885 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1887 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1888 qc->flags |= ATA_QCFLAG_QUIET;
1889 return qc->flags & ATA_QCFLAG_QUIET;
1892 static int ata_eh_read_sense_success_non_ncq(struct ata_link *link)
1894 struct ata_port *ap = link->ap;
1895 struct ata_queued_cmd *qc;
1897 qc = __ata_qc_from_tag(ap, link->active_tag);
1901 if (!(qc->flags & ATA_QCFLAG_EH) ||
1902 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
1906 if (!ata_eh_request_sense(qc))
1910 * If we have sense data, call scsi_check_sense() in order to set the
1911 * correct SCSI ML byte (if any). No point in checking the return value,
1912 * since the command has already completed successfully.
1914 scsi_check_sense(qc->scsicmd);
1919 static void ata_eh_get_success_sense(struct ata_link *link)
1921 struct ata_eh_context *ehc = &link->eh_context;
1922 struct ata_device *dev = link->device;
1923 struct ata_port *ap = link->ap;
1924 struct ata_queued_cmd *qc;
1927 if (!(ehc->i.dev_action[dev->devno] & ATA_EH_GET_SUCCESS_SENSE))
1930 /* if frozen, we can't do much */
1931 if (ata_port_is_frozen(ap)) {
1933 "successful sense data available but port frozen\n");
1938 * If the link has sactive set, then we have outstanding NCQ commands
1939 * and have to read the Successful NCQ Commands log to get the sense
1940 * data. Otherwise, we are dealing with a non-NCQ command and use
1941 * request sense ext command to retrieve the sense data.
1944 ret = ata_eh_read_sense_success_ncq_log(link);
1946 ret = ata_eh_read_sense_success_non_ncq(link);
1950 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
1955 * If we failed to get sense data for a successful command that ought to
1956 * have sense data, we cannot simply return BLK_STS_OK to user space.
1957 * This is because we can't know if the sense data that we couldn't get
1958 * was actually "DATA CURRENTLY UNAVAILABLE". Reporting such a command
1959 * as success to user space would result in a silent data corruption.
1960 * Thus, add a bogus ABORTED_COMMAND sense data to such commands, such
1961 * that SCSI will report these commands as BLK_STS_IOERR to user space.
1963 ata_qc_for_each_raw(ap, qc, tag) {
1964 if (!(qc->flags & ATA_QCFLAG_EH) ||
1965 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
1967 ata_dev_phys_link(qc->dev) != link)
1970 /* We managed to get sense for this success command, skip. */
1971 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1974 /* This success command did not have any sense data, skip. */
1975 if (!(qc->result_tf.status & ATA_SENSE))
1978 /* This success command had sense data, but we failed to get. */
1979 ata_scsi_set_sense(dev, qc->scsicmd, ABORTED_COMMAND, 0, 0);
1980 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1982 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
1986 * ata_eh_link_autopsy - analyze error and determine recovery action
1987 * @link: host link to perform autopsy on
1989 * Analyze why @link failed and determine which recovery actions
1990 * are needed. This function also sets more detailed AC_ERR_*
1991 * values and fills sense data for ATAPI CHECK SENSE.
1994 * Kernel thread context (may sleep).
1996 static void ata_eh_link_autopsy(struct ata_link *link)
1998 struct ata_port *ap = link->ap;
1999 struct ata_eh_context *ehc = &link->eh_context;
2000 struct ata_queued_cmd *qc;
2001 struct ata_device *dev;
2002 unsigned int all_err_mask = 0, eflags = 0;
2003 int tag, nr_failed = 0, nr_quiet = 0;
2007 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2010 /* obtain and analyze SError */
2011 rc = sata_scr_read(link, SCR_ERROR, &serror);
2013 ehc->i.serror |= serror;
2014 ata_eh_analyze_serror(link);
2015 } else if (rc != -EOPNOTSUPP) {
2016 /* SError read failed, force reset and probing */
2017 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2018 ehc->i.action |= ATA_EH_RESET;
2019 ehc->i.err_mask |= AC_ERR_OTHER;
2022 /* analyze NCQ failure */
2023 ata_eh_analyze_ncq_error(link);
2026 * Check if this was a successful command that simply needs sense data.
2027 * Since the sense data is not part of the completion, we need to fetch
2028 * it using an additional command. Since this can't be done from irq
2029 * context, the sense data for successful commands are fetched by EH.
2031 ata_eh_get_success_sense(link);
2033 /* any real error trumps AC_ERR_OTHER */
2034 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2035 ehc->i.err_mask &= ~AC_ERR_OTHER;
2037 all_err_mask |= ehc->i.err_mask;
2039 ata_qc_for_each_raw(ap, qc, tag) {
2040 if (!(qc->flags & ATA_QCFLAG_EH) ||
2041 qc->flags & ATA_QCFLAG_RETRY ||
2042 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD ||
2043 ata_dev_phys_link(qc->dev) != link)
2046 /* inherit upper level err_mask */
2047 qc->err_mask |= ehc->i.err_mask;
2050 ehc->i.action |= ata_eh_analyze_tf(qc);
2052 /* DEV errors are probably spurious in case of ATA_BUS error */
2053 if (qc->err_mask & AC_ERR_ATA_BUS)
2054 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2057 /* any real error trumps unknown error */
2058 if (qc->err_mask & ~AC_ERR_OTHER)
2059 qc->err_mask &= ~AC_ERR_OTHER;
2062 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2063 * layers will determine whether the command is worth retrying
2064 * based on the sense data and device class/type. Otherwise,
2065 * determine directly if the command is worth retrying using its
2066 * error mask and flags.
2068 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2069 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2070 else if (ata_eh_worth_retry(qc))
2071 qc->flags |= ATA_QCFLAG_RETRY;
2073 /* accumulate error info */
2074 ehc->i.dev = qc->dev;
2075 all_err_mask |= qc->err_mask;
2076 if (qc->flags & ATA_QCFLAG_IO)
2077 eflags |= ATA_EFLAG_IS_IO;
2078 trace_ata_eh_link_autopsy_qc(qc);
2080 /* Count quiet errors */
2081 if (ata_eh_quiet(qc))
2086 /* If all failed commands requested silence, then be quiet */
2087 if (nr_quiet == nr_failed)
2088 ehc->i.flags |= ATA_EHI_QUIET;
2090 /* enforce default EH actions */
2091 if (ata_port_is_frozen(ap) ||
2092 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2093 ehc->i.action |= ATA_EH_RESET;
2094 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2095 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2096 ehc->i.action |= ATA_EH_REVALIDATE;
2098 /* If we have offending qcs and the associated failed device,
2099 * perform per-dev EH action only on the offending device.
2102 ehc->i.dev_action[ehc->i.dev->devno] |=
2103 ehc->i.action & ATA_EH_PERDEV_MASK;
2104 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2107 /* propagate timeout to host link */
2108 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2109 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2111 /* record error and consider speeding down */
2113 if (!dev && ((ata_link_max_devices(link) == 1 &&
2114 ata_dev_enabled(link->device))))
2118 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2119 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2120 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2121 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2126 * ata_eh_autopsy - analyze error and determine recovery action
2127 * @ap: host port to perform autopsy on
2129 * Analyze all links of @ap and determine why they failed and
2130 * which recovery actions are needed.
2133 * Kernel thread context (may sleep).
2135 void ata_eh_autopsy(struct ata_port *ap)
2137 struct ata_link *link;
2139 ata_for_each_link(link, ap, EDGE)
2140 ata_eh_link_autopsy(link);
2142 /* Handle the frigging slave link. Autopsy is done similarly
2143 * but actions and flags are transferred over to the master
2144 * link and handled from there.
2146 if (ap->slave_link) {
2147 struct ata_eh_context *mehc = &ap->link.eh_context;
2148 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2150 /* transfer control flags from master to slave */
2151 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2153 /* perform autopsy on the slave link */
2154 ata_eh_link_autopsy(ap->slave_link);
2156 /* transfer actions from slave to master and clear slave */
2157 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2158 mehc->i.action |= sehc->i.action;
2159 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2160 mehc->i.flags |= sehc->i.flags;
2161 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2164 /* Autopsy of fanout ports can affect host link autopsy.
2165 * Perform host link autopsy last.
2167 if (sata_pmp_attached(ap))
2168 ata_eh_link_autopsy(&ap->link);
2172 * ata_get_cmd_name - get name for ATA command
2173 * @command: ATA command code to get name for
2175 * Return a textual name of the given command or "unknown"
2180 const char *ata_get_cmd_name(u8 command)
2182 #ifdef CONFIG_ATA_VERBOSE_ERROR
2188 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2189 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2190 { ATA_CMD_STANDBY, "STANDBY" },
2191 { ATA_CMD_IDLE, "IDLE" },
2192 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2193 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2194 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2195 { ATA_CMD_NOP, "NOP" },
2196 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2197 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2198 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2199 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2200 { ATA_CMD_SERVICE, "SERVICE" },
2201 { ATA_CMD_READ, "READ DMA" },
2202 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2203 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2204 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2205 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2206 { ATA_CMD_WRITE, "WRITE DMA" },
2207 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2208 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2209 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2210 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2211 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2212 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2213 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2214 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2215 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2216 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2217 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2218 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2219 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2220 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2221 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2222 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2223 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2224 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2225 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2226 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2227 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2228 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2229 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2230 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2231 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2232 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2233 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2234 { ATA_CMD_SLEEP, "SLEEP" },
2235 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2236 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2237 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2238 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2239 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2240 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2241 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2242 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2243 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2244 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2245 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2246 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2247 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2248 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2249 { ATA_CMD_PMP_READ, "READ BUFFER" },
2250 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2251 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2252 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2253 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2254 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2255 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2256 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2257 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2258 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2259 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2260 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2261 { ATA_CMD_SMART, "SMART" },
2262 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2263 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2264 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2265 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2266 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2267 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2268 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2269 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2270 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2271 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2272 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2273 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2274 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2275 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2276 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2277 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2278 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2279 { ATA_CMD_RESTORE, "RECALIBRATE" },
2280 { 0, NULL } /* terminate list */
2284 for (i = 0; cmd_descr[i].text; i++)
2285 if (cmd_descr[i].command == command)
2286 return cmd_descr[i].text;
2291 EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2294 * ata_eh_link_report - report error handling to user
2295 * @link: ATA link EH is going on
2297 * Report EH to user.
2302 static void ata_eh_link_report(struct ata_link *link)
2304 struct ata_port *ap = link->ap;
2305 struct ata_eh_context *ehc = &link->eh_context;
2306 struct ata_queued_cmd *qc;
2307 const char *frozen, *desc;
2308 char tries_buf[6] = "";
2309 int tag, nr_failed = 0;
2311 if (ehc->i.flags & ATA_EHI_QUIET)
2315 if (ehc->i.desc[0] != '\0')
2318 ata_qc_for_each_raw(ap, qc, tag) {
2319 if (!(qc->flags & ATA_QCFLAG_EH) ||
2320 ata_dev_phys_link(qc->dev) != link ||
2321 ((qc->flags & ATA_QCFLAG_QUIET) &&
2322 qc->err_mask == AC_ERR_DEV))
2324 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2330 if (!nr_failed && !ehc->i.err_mask)
2334 if (ata_port_is_frozen(ap))
2337 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2338 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2342 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2343 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2344 ehc->i.err_mask, link->sactive, ehc->i.serror,
2345 ehc->i.action, frozen, tries_buf);
2347 ata_dev_err(ehc->i.dev, "%s\n", desc);
2349 ata_link_err(link, "exception Emask 0x%x "
2350 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2351 ehc->i.err_mask, link->sactive, ehc->i.serror,
2352 ehc->i.action, frozen, tries_buf);
2354 ata_link_err(link, "%s\n", desc);
2357 #ifdef CONFIG_ATA_VERBOSE_ERROR
2360 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2361 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2362 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2363 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2364 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2365 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2366 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2367 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2368 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2369 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2370 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2371 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2372 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2373 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2374 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2375 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2376 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2377 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2380 ata_qc_for_each_raw(ap, qc, tag) {
2381 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2382 char data_buf[20] = "";
2383 char cdb_buf[70] = "";
2385 if (!(qc->flags & ATA_QCFLAG_EH) ||
2386 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2389 if (qc->dma_dir != DMA_NONE) {
2390 static const char *dma_str[] = {
2391 [DMA_BIDIRECTIONAL] = "bidi",
2392 [DMA_TO_DEVICE] = "out",
2393 [DMA_FROM_DEVICE] = "in",
2395 const char *prot_str = NULL;
2397 switch (qc->tf.protocol) {
2398 case ATA_PROT_UNKNOWN:
2399 prot_str = "unknown";
2401 case ATA_PROT_NODATA:
2402 prot_str = "nodata";
2411 prot_str = "ncq dma";
2413 case ATA_PROT_NCQ_NODATA:
2414 prot_str = "ncq nodata";
2416 case ATAPI_PROT_NODATA:
2417 prot_str = "nodata";
2419 case ATAPI_PROT_PIO:
2422 case ATAPI_PROT_DMA:
2426 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2427 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2430 if (ata_is_atapi(qc->tf.protocol)) {
2431 const u8 *cdb = qc->cdb;
2432 size_t cdb_len = qc->dev->cdb_len;
2435 cdb = qc->scsicmd->cmnd;
2436 cdb_len = qc->scsicmd->cmd_len;
2438 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2441 ata_dev_err(qc->dev, "failed command: %s\n",
2442 ata_get_cmd_name(cmd->command));
2444 ata_dev_err(qc->dev,
2445 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2447 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2448 "Emask 0x%x (%s)%s\n",
2449 cmd->command, cmd->feature, cmd->nsect,
2450 cmd->lbal, cmd->lbam, cmd->lbah,
2451 cmd->hob_feature, cmd->hob_nsect,
2452 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2453 cmd->device, qc->tag, data_buf, cdb_buf,
2454 res->status, res->error, res->nsect,
2455 res->lbal, res->lbam, res->lbah,
2456 res->hob_feature, res->hob_nsect,
2457 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2458 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2459 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2461 #ifdef CONFIG_ATA_VERBOSE_ERROR
2462 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2463 ATA_SENSE | ATA_ERR)) {
2464 if (res->status & ATA_BUSY)
2465 ata_dev_err(qc->dev, "status: { Busy }\n");
2467 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2468 res->status & ATA_DRDY ? "DRDY " : "",
2469 res->status & ATA_DF ? "DF " : "",
2470 res->status & ATA_DRQ ? "DRQ " : "",
2471 res->status & ATA_SENSE ? "SENSE " : "",
2472 res->status & ATA_ERR ? "ERR " : "");
2475 if (cmd->command != ATA_CMD_PACKET &&
2476 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2478 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2479 res->error & ATA_ICRC ? "ICRC " : "",
2480 res->error & ATA_UNC ? "UNC " : "",
2481 res->error & ATA_AMNF ? "AMNF " : "",
2482 res->error & ATA_IDNF ? "IDNF " : "",
2483 res->error & ATA_ABORTED ? "ABRT " : "");
2489 * ata_eh_report - report error handling to user
2490 * @ap: ATA port to report EH about
2492 * Report EH to user.
2497 void ata_eh_report(struct ata_port *ap)
2499 struct ata_link *link;
2501 ata_for_each_link(link, ap, HOST_FIRST)
2502 ata_eh_link_report(link);
2505 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2506 unsigned int *classes, unsigned long deadline,
2509 struct ata_device *dev;
2512 ata_for_each_dev(dev, link, ALL)
2513 classes[dev->devno] = ATA_DEV_UNKNOWN;
2515 return reset(link, classes, deadline);
2518 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2520 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2524 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2529 int ata_eh_reset(struct ata_link *link, int classify,
2530 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2531 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2533 struct ata_port *ap = link->ap;
2534 struct ata_link *slave = ap->slave_link;
2535 struct ata_eh_context *ehc = &link->eh_context;
2536 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2537 unsigned int *classes = ehc->classes;
2538 unsigned int lflags = link->flags;
2539 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2540 int max_tries = 0, try = 0;
2541 struct ata_link *failed_link;
2542 struct ata_device *dev;
2543 unsigned long deadline, now;
2544 ata_reset_fn_t reset;
2545 unsigned long flags;
2552 while (ata_eh_reset_timeouts[max_tries] != UINT_MAX)
2554 if (link->flags & ATA_LFLAG_RST_ONCE)
2556 if (link->flags & ATA_LFLAG_NO_HRST)
2558 if (link->flags & ATA_LFLAG_NO_SRST)
2561 /* make sure each reset attempt is at least COOL_DOWN apart */
2562 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2564 WARN_ON(time_after(ehc->last_reset, now));
2565 deadline = ata_deadline(ehc->last_reset,
2566 ATA_EH_RESET_COOL_DOWN);
2567 if (time_before(now, deadline))
2568 schedule_timeout_uninterruptible(deadline - now);
2571 spin_lock_irqsave(ap->lock, flags);
2572 ap->pflags |= ATA_PFLAG_RESETTING;
2573 spin_unlock_irqrestore(ap->lock, flags);
2575 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2577 ata_for_each_dev(dev, link, ALL) {
2578 /* If we issue an SRST then an ATA drive (not ATAPI)
2579 * may change configuration and be in PIO0 timing. If
2580 * we do a hard reset (or are coming from power on)
2581 * this is true for ATA or ATAPI. Until we've set a
2582 * suitable controller mode we should not touch the
2583 * bus as we may be talking too fast.
2585 dev->pio_mode = XFER_PIO_0;
2586 dev->dma_mode = 0xff;
2588 /* If the controller has a pio mode setup function
2589 * then use it to set the chipset to rights. Don't
2590 * touch the DMA setup as that will be dealt with when
2591 * configuring devices.
2593 if (ap->ops->set_piomode)
2594 ap->ops->set_piomode(ap, dev);
2597 /* prefer hardreset */
2599 ehc->i.action &= ~ATA_EH_RESET;
2602 ehc->i.action |= ATA_EH_HARDRESET;
2603 } else if (softreset) {
2605 ehc->i.action |= ATA_EH_SOFTRESET;
2609 unsigned long deadline = ata_deadline(jiffies,
2610 ATA_EH_PRERESET_TIMEOUT);
2613 sehc->i.action &= ~ATA_EH_RESET;
2614 sehc->i.action |= ehc->i.action;
2617 rc = prereset(link, deadline);
2619 /* If present, do prereset on slave link too. Reset
2620 * is skipped iff both master and slave links report
2621 * -ENOENT or clear ATA_EH_RESET.
2623 if (slave && (rc == 0 || rc == -ENOENT)) {
2626 tmp = prereset(slave, deadline);
2630 ehc->i.action |= sehc->i.action;
2634 if (rc == -ENOENT) {
2635 ata_link_dbg(link, "port disabled--ignoring\n");
2636 ehc->i.action &= ~ATA_EH_RESET;
2638 ata_for_each_dev(dev, link, ALL)
2639 classes[dev->devno] = ATA_DEV_NONE;
2644 "prereset failed (errno=%d)\n",
2649 /* prereset() might have cleared ATA_EH_RESET. If so,
2650 * bang classes, thaw and return.
2652 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2653 ata_for_each_dev(dev, link, ALL)
2654 classes[dev->devno] = ATA_DEV_NONE;
2655 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2656 ata_eh_thaw_port(ap);
2666 if (ata_is_host_link(link))
2667 ata_eh_freeze_port(ap);
2669 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2673 ata_link_info(link, "%s resetting link\n",
2674 reset == softreset ? "soft" : "hard");
2676 /* mark that this EH session started with reset */
2677 ehc->last_reset = jiffies;
2678 if (reset == hardreset) {
2679 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2680 trace_ata_link_hardreset_begin(link, classes, deadline);
2682 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2683 trace_ata_link_softreset_begin(link, classes, deadline);
2686 rc = ata_do_reset(link, reset, classes, deadline, true);
2687 if (reset == hardreset)
2688 trace_ata_link_hardreset_end(link, classes, rc);
2690 trace_ata_link_softreset_end(link, classes, rc);
2691 if (rc && rc != -EAGAIN) {
2696 /* hardreset slave link if existent */
2697 if (slave && reset == hardreset) {
2701 ata_link_info(slave, "hard resetting link\n");
2703 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2704 trace_ata_slave_hardreset_begin(slave, classes,
2706 tmp = ata_do_reset(slave, reset, classes, deadline,
2708 trace_ata_slave_hardreset_end(slave, classes, tmp);
2716 failed_link = slave;
2722 /* perform follow-up SRST if necessary */
2723 if (reset == hardreset &&
2724 ata_eh_followup_srst_needed(link, rc)) {
2729 "follow-up softreset required but no softreset available\n");
2735 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2736 trace_ata_link_softreset_begin(link, classes, deadline);
2737 rc = ata_do_reset(link, reset, classes, deadline, true);
2738 trace_ata_link_softreset_end(link, classes, rc);
2747 "no reset method available, skipping reset\n");
2748 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2749 lflags |= ATA_LFLAG_ASSUME_ATA;
2753 * Post-reset processing
2755 ata_for_each_dev(dev, link, ALL) {
2756 /* After the reset, the device state is PIO 0 and the
2757 * controller state is undefined. Reset also wakes up
2758 * drives from sleeping mode.
2760 dev->pio_mode = XFER_PIO_0;
2761 dev->flags &= ~ATA_DFLAG_SLEEPING;
2763 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2766 /* apply class override */
2767 if (lflags & ATA_LFLAG_ASSUME_ATA)
2768 classes[dev->devno] = ATA_DEV_ATA;
2769 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2770 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2773 /* record current link speed */
2774 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2775 link->sata_spd = (sstatus >> 4) & 0xf;
2776 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2777 slave->sata_spd = (sstatus >> 4) & 0xf;
2780 if (ata_is_host_link(link))
2781 ata_eh_thaw_port(ap);
2783 /* postreset() should clear hardware SError. Although SError
2784 * is cleared during link resume, clearing SError here is
2785 * necessary as some PHYs raise hotplug events after SRST.
2786 * This introduces race condition where hotplug occurs between
2787 * reset and here. This race is mediated by cross checking
2788 * link onlineness and classification result later.
2791 postreset(link, classes);
2792 trace_ata_link_postreset(link, classes, rc);
2794 postreset(slave, classes);
2795 trace_ata_slave_postreset(slave, classes, rc);
2799 /* clear cached SError */
2800 spin_lock_irqsave(link->ap->lock, flags);
2801 link->eh_info.serror = 0;
2803 slave->eh_info.serror = 0;
2804 spin_unlock_irqrestore(link->ap->lock, flags);
2807 * Make sure onlineness and classification result correspond.
2808 * Hotplug could have happened during reset and some
2809 * controllers fail to wait while a drive is spinning up after
2810 * being hotplugged causing misdetection. By cross checking
2811 * link on/offlineness and classification result, those
2812 * conditions can be reliably detected and retried.
2815 ata_for_each_dev(dev, link, ALL) {
2816 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2817 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2818 ata_dev_dbg(dev, "link online but device misclassified\n");
2819 classes[dev->devno] = ATA_DEV_NONE;
2822 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2823 if (ata_class_enabled(classes[dev->devno]))
2825 "link offline, clearing class %d to NONE\n",
2826 classes[dev->devno]);
2827 classes[dev->devno] = ATA_DEV_NONE;
2828 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2830 "link status unknown, clearing UNKNOWN to NONE\n");
2831 classes[dev->devno] = ATA_DEV_NONE;
2835 if (classify && nr_unknown) {
2836 if (try < max_tries) {
2838 "link online but %d devices misclassified, retrying\n",
2845 "link online but %d devices misclassified, "
2846 "device detection might fail\n", nr_unknown);
2849 /* reset successful, schedule revalidation */
2850 ata_eh_done(link, NULL, ATA_EH_RESET);
2852 ata_eh_done(slave, NULL, ATA_EH_RESET);
2853 ehc->last_reset = jiffies; /* update to completion time */
2854 ehc->i.action |= ATA_EH_REVALIDATE;
2855 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2859 /* clear hotplug flag */
2860 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2862 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2864 spin_lock_irqsave(ap->lock, flags);
2865 ap->pflags &= ~ATA_PFLAG_RESETTING;
2866 spin_unlock_irqrestore(ap->lock, flags);
2871 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2872 if (!ata_is_host_link(link) &&
2873 sata_scr_read(link, SCR_STATUS, &sstatus))
2876 if (try >= max_tries) {
2878 * Thaw host port even if reset failed, so that the port
2879 * can be retried on the next phy event. This risks
2880 * repeated EH runs but seems to be a better tradeoff than
2881 * shutting down a port after a botched hotplug attempt.
2883 if (ata_is_host_link(link))
2884 ata_eh_thaw_port(ap);
2889 if (time_before(now, deadline)) {
2890 unsigned long delta = deadline - now;
2892 ata_link_warn(failed_link,
2893 "reset failed (errno=%d), retrying in %u secs\n",
2894 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2898 delta = schedule_timeout_uninterruptible(delta);
2903 * While disks spinup behind PMP, some controllers fail sending SRST.
2904 * They need to be reset - as well as the PMP - before retrying.
2906 if (rc == -ERESTART) {
2907 if (ata_is_host_link(link))
2908 ata_eh_thaw_port(ap);
2912 if (try == max_tries - 1) {
2913 sata_down_spd_limit(link, 0);
2915 sata_down_spd_limit(slave, 0);
2916 } else if (rc == -EPIPE)
2917 sata_down_spd_limit(failed_link, 0);
2924 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2926 struct ata_link *link;
2927 struct ata_device *dev;
2928 unsigned long flags;
2931 * This function can be thought of as an extended version of
2932 * ata_eh_about_to_do() specially crafted to accommodate the
2933 * requirements of ATA_EH_PARK handling. Since the EH thread
2934 * does not leave the do {} while () loop in ata_eh_recover as
2935 * long as the timeout for a park request to *one* device on
2936 * the port has not expired, and since we still want to pick
2937 * up park requests to other devices on the same port or
2938 * timeout updates for the same device, we have to pull
2939 * ATA_EH_PARK actions from eh_info into eh_context.i
2940 * ourselves at the beginning of each pass over the loop.
2942 * Additionally, all write accesses to &ap->park_req_pending
2943 * through reinit_completion() (see below) or complete_all()
2944 * (see ata_scsi_park_store()) are protected by the host lock.
2945 * As a result we have that park_req_pending.done is zero on
2946 * exit from this function, i.e. when ATA_EH_PARK actions for
2947 * *all* devices on port ap have been pulled into the
2948 * respective eh_context structs. If, and only if,
2949 * park_req_pending.done is non-zero by the time we reach
2950 * wait_for_completion_timeout(), another ATA_EH_PARK action
2951 * has been scheduled for at least one of the devices on port
2952 * ap and we have to cycle over the do {} while () loop in
2953 * ata_eh_recover() again.
2956 spin_lock_irqsave(ap->lock, flags);
2957 reinit_completion(&ap->park_req_pending);
2958 ata_for_each_link(link, ap, EDGE) {
2959 ata_for_each_dev(dev, link, ALL) {
2960 struct ata_eh_info *ehi = &link->eh_info;
2962 link->eh_context.i.dev_action[dev->devno] |=
2963 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2964 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2967 spin_unlock_irqrestore(ap->lock, flags);
2970 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2972 struct ata_eh_context *ehc = &dev->link->eh_context;
2973 struct ata_taskfile tf;
2974 unsigned int err_mask;
2976 ata_tf_init(dev, &tf);
2978 ehc->unloaded_mask |= 1 << dev->devno;
2979 tf.command = ATA_CMD_IDLEIMMEDIATE;
2985 ehc->unloaded_mask &= ~(1 << dev->devno);
2986 tf.command = ATA_CMD_CHK_POWER;
2989 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2990 tf.protocol = ATA_PROT_NODATA;
2991 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2992 if (park && (err_mask || tf.lbal != 0xc4)) {
2993 ata_dev_err(dev, "head unload failed!\n");
2994 ehc->unloaded_mask &= ~(1 << dev->devno);
2998 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2999 struct ata_device **r_failed_dev)
3001 struct ata_port *ap = link->ap;
3002 struct ata_eh_context *ehc = &link->eh_context;
3003 struct ata_device *dev;
3004 unsigned int new_mask = 0;
3005 unsigned long flags;
3008 /* For PATA drive side cable detection to work, IDENTIFY must
3009 * be done backwards such that PDIAG- is released by the slave
3010 * device before the master device is identified.
3012 ata_for_each_dev(dev, link, ALL_REVERSE) {
3013 unsigned int action = ata_eh_dev_action(dev);
3014 unsigned int readid_flags = 0;
3016 if (ehc->i.flags & ATA_EHI_DID_RESET)
3017 readid_flags |= ATA_READID_POSTRESET;
3019 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3020 WARN_ON(dev->class == ATA_DEV_PMP);
3023 * The link may be in a deep sleep, wake it up.
3025 * If the link is in deep sleep, ata_phys_link_offline()
3026 * will return true, causing the revalidation to fail,
3027 * which leads to a (potentially) needless hard reset.
3029 * ata_eh_recover() will later restore the link policy
3030 * to ap->target_lpm_policy after revalidation is done.
3032 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3033 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
3039 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3044 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3045 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3050 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3052 /* Configuration may have changed, reconfigure
3055 ehc->i.flags |= ATA_EHI_SETMODE;
3057 /* schedule the scsi_rescan_device() here */
3058 schedule_delayed_work(&ap->scsi_rescan_task, 0);
3059 } else if (dev->class == ATA_DEV_UNKNOWN &&
3060 ehc->tries[dev->devno] &&
3061 ata_class_enabled(ehc->classes[dev->devno])) {
3062 /* Temporarily set dev->class, it will be
3063 * permanently set once all configurations are
3064 * complete. This is necessary because new
3065 * device configuration is done in two
3068 dev->class = ehc->classes[dev->devno];
3070 if (dev->class == ATA_DEV_PMP)
3071 rc = sata_pmp_attach(dev);
3073 rc = ata_dev_read_id(dev, &dev->class,
3074 readid_flags, dev->id);
3076 /* read_id might have changed class, store and reset */
3077 ehc->classes[dev->devno] = dev->class;
3078 dev->class = ATA_DEV_UNKNOWN;
3082 /* clear error info accumulated during probe */
3083 ata_ering_clear(&dev->ering);
3084 new_mask |= 1 << dev->devno;
3087 /* IDENTIFY was issued to non-existent
3088 * device. No need to reset. Just
3089 * thaw and ignore the device.
3091 ata_eh_thaw_port(ap);
3099 /* PDIAG- should have been released, ask cable type if post-reset */
3100 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3101 if (ap->ops->cable_detect)
3102 ap->cbl = ap->ops->cable_detect(ap);
3106 /* Configure new devices forward such that user doesn't see
3107 * device detection messages backwards.
3109 ata_for_each_dev(dev, link, ALL) {
3110 if (!(new_mask & (1 << dev->devno)))
3113 dev->class = ehc->classes[dev->devno];
3115 if (dev->class == ATA_DEV_PMP)
3118 ehc->i.flags |= ATA_EHI_PRINTINFO;
3119 rc = ata_dev_configure(dev);
3120 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3122 dev->class = ATA_DEV_UNKNOWN;
3126 spin_lock_irqsave(ap->lock, flags);
3127 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3128 spin_unlock_irqrestore(ap->lock, flags);
3130 /* new device discovered, configure xfermode */
3131 ehc->i.flags |= ATA_EHI_SETMODE;
3137 *r_failed_dev = dev;
3142 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3143 * @link: link on which timings will be programmed
3144 * @r_failed_dev: out parameter for failed device
3146 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3147 * ata_set_mode() fails, pointer to the failing device is
3148 * returned in @r_failed_dev.
3151 * PCI/etc. bus probe sem.
3154 * 0 on success, negative errno otherwise
3156 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3158 struct ata_port *ap = link->ap;
3159 struct ata_device *dev;
3162 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3163 ata_for_each_dev(dev, link, ENABLED) {
3164 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3165 struct ata_ering_entry *ent;
3167 ent = ata_ering_top(&dev->ering);
3169 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3173 /* has private set_mode? */
3174 if (ap->ops->set_mode)
3175 rc = ap->ops->set_mode(link, r_failed_dev);
3177 rc = ata_do_set_mode(link, r_failed_dev);
3179 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3180 ata_for_each_dev(dev, link, ENABLED) {
3181 struct ata_eh_context *ehc = &link->eh_context;
3182 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3183 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3185 if (dev->xfer_mode != saved_xfer_mode ||
3186 ata_ncq_enabled(dev) != saved_ncq)
3187 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3194 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3195 * @dev: ATAPI device to clear UA for
3197 * Resets and other operations can make an ATAPI device raise
3198 * UNIT ATTENTION which causes the next operation to fail. This
3199 * function clears UA.
3202 * EH context (may sleep).
3205 * 0 on success, -errno on failure.
3207 static int atapi_eh_clear_ua(struct ata_device *dev)
3211 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3212 u8 *sense_buffer = dev->link->ap->sector_buf;
3214 unsigned int err_mask;
3216 err_mask = atapi_eh_tur(dev, &sense_key);
3217 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3219 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3224 if (!err_mask || sense_key != UNIT_ATTENTION)
3227 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3229 ata_dev_warn(dev, "failed to clear "
3230 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3235 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3242 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3243 * @dev: ATA device which may need FLUSH retry
3245 * If @dev failed FLUSH, it needs to be reported upper layer
3246 * immediately as it means that @dev failed to remap and already
3247 * lost at least a sector and further FLUSH retrials won't make
3248 * any difference to the lost sector. However, if FLUSH failed
3249 * for other reasons, for example transmission error, FLUSH needs
3252 * This function determines whether FLUSH failure retry is
3253 * necessary and performs it if so.
3256 * 0 if EH can continue, -errno if EH needs to be repeated.
3258 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3260 struct ata_link *link = dev->link;
3261 struct ata_port *ap = link->ap;
3262 struct ata_queued_cmd *qc;
3263 struct ata_taskfile tf;
3264 unsigned int err_mask;
3267 /* did flush fail for this device? */
3268 if (!ata_tag_valid(link->active_tag))
3271 qc = __ata_qc_from_tag(ap, link->active_tag);
3272 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3273 qc->tf.command != ATA_CMD_FLUSH))
3276 /* if the device failed it, it should be reported to upper layers */
3277 if (qc->err_mask & AC_ERR_DEV)
3280 /* flush failed for some other reason, give it another shot */
3281 ata_tf_init(dev, &tf);
3283 tf.command = qc->tf.command;
3284 tf.flags |= ATA_TFLAG_DEVICE;
3285 tf.protocol = ATA_PROT_NODATA;
3287 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3288 tf.command, qc->err_mask);
3290 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3293 * FLUSH is complete but there's no way to
3294 * successfully complete a failed command from EH.
3295 * Making sure retry is allowed at least once and
3296 * retrying it should do the trick - whatever was in
3297 * the cache is already on the platter and this won't
3298 * cause infinite loop.
3300 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3302 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3306 /* if device failed it, report it to upper layers */
3307 if (err_mask & AC_ERR_DEV) {
3308 qc->err_mask |= AC_ERR_DEV;
3310 if (!ata_port_is_frozen(ap))
3318 * ata_eh_set_lpm - configure SATA interface power management
3319 * @link: link to configure power management
3320 * @policy: the link power management policy
3321 * @r_failed_dev: out parameter for failed device
3323 * Enable SATA Interface power management. This will enable
3324 * Device Interface Power Management (DIPM) for min_power and
3325 * medium_power_with_dipm policies, and then call driver specific
3326 * callbacks for enabling Host Initiated Power management.
3332 * 0 on success, -errno on failure.
3334 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3335 struct ata_device **r_failed_dev)
3337 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3338 struct ata_eh_context *ehc = &link->eh_context;
3339 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3340 enum ata_lpm_policy old_policy = link->lpm_policy;
3341 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3342 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3343 unsigned int err_mask;
3346 /* if the link or host doesn't do LPM, noop */
3347 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3348 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3352 * DIPM is enabled only for MIN_POWER as some devices
3353 * misbehave when the host NACKs transition to SLUMBER. Order
3354 * device and link configurations such that the host always
3355 * allows DIPM requests.
3357 ata_for_each_dev(dev, link, ENABLED) {
3358 bool hipm = ata_id_has_hipm(dev->id);
3359 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3361 /* find the first enabled and LPM enabled devices */
3365 if (!lpm_dev && (hipm || dipm))
3368 hints &= ~ATA_LPM_EMPTY;
3370 hints &= ~ATA_LPM_HIPM;
3372 /* disable DIPM before changing link config */
3373 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3374 err_mask = ata_dev_set_feature(dev,
3375 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3376 if (err_mask && err_mask != AC_ERR_DEV) {
3378 "failed to disable DIPM, Emask 0x%x\n",
3387 rc = ap->ops->set_lpm(link, policy, hints);
3388 if (!rc && ap->slave_link)
3389 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3391 rc = sata_pmp_set_lpm(link, policy, hints);
3394 * Attribute link config failure to the first (LPM) enabled
3395 * device on the link.
3398 if (rc == -EOPNOTSUPP) {
3399 link->flags |= ATA_LFLAG_NO_LPM;
3402 dev = lpm_dev ? lpm_dev : link_dev;
3407 * Low level driver acked the transition. Issue DIPM command
3408 * with the new policy set.
3410 link->lpm_policy = policy;
3411 if (ap && ap->slave_link)
3412 ap->slave_link->lpm_policy = policy;
3414 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3415 ata_for_each_dev(dev, link, ENABLED) {
3416 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3417 ata_id_has_dipm(dev->id)) {
3418 err_mask = ata_dev_set_feature(dev,
3419 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3420 if (err_mask && err_mask != AC_ERR_DEV) {
3422 "failed to enable DIPM, Emask 0x%x\n",
3430 link->last_lpm_change = jiffies;
3431 link->flags |= ATA_LFLAG_CHANGED;
3436 /* restore the old policy */
3437 link->lpm_policy = old_policy;
3438 if (ap && ap->slave_link)
3439 ap->slave_link->lpm_policy = old_policy;
3441 /* if no device or only one more chance is left, disable LPM */
3442 if (!dev || ehc->tries[dev->devno] <= 2) {
3443 ata_link_warn(link, "disabling LPM on the link\n");
3444 link->flags |= ATA_LFLAG_NO_LPM;
3447 *r_failed_dev = dev;
3451 int ata_link_nr_enabled(struct ata_link *link)
3453 struct ata_device *dev;
3456 ata_for_each_dev(dev, link, ENABLED)
3461 static int ata_link_nr_vacant(struct ata_link *link)
3463 struct ata_device *dev;
3466 ata_for_each_dev(dev, link, ALL)
3467 if (dev->class == ATA_DEV_UNKNOWN)
3472 static int ata_eh_skip_recovery(struct ata_link *link)
3474 struct ata_port *ap = link->ap;
3475 struct ata_eh_context *ehc = &link->eh_context;
3476 struct ata_device *dev;
3478 /* skip disabled links */
3479 if (link->flags & ATA_LFLAG_DISABLED)
3482 /* skip if explicitly requested */
3483 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3486 /* thaw frozen port and recover failed devices */
3487 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3490 /* reset at least once if reset is requested */
3491 if ((ehc->i.action & ATA_EH_RESET) &&
3492 !(ehc->i.flags & ATA_EHI_DID_RESET))
3495 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3496 ata_for_each_dev(dev, link, ALL) {
3497 if (dev->class == ATA_DEV_UNKNOWN &&
3498 ehc->classes[dev->devno] != ATA_DEV_NONE)
3505 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3507 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3508 u64 now = get_jiffies_64();
3509 int *trials = void_arg;
3511 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3512 (ent->timestamp < now - min(now, interval)))
3519 static int ata_eh_schedule_probe(struct ata_device *dev)
3521 struct ata_eh_context *ehc = &dev->link->eh_context;
3522 struct ata_link *link = ata_dev_phys_link(dev);
3525 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3526 (ehc->did_probe_mask & (1 << dev->devno)))
3529 ata_eh_detach_dev(dev);
3531 ehc->did_probe_mask |= (1 << dev->devno);
3532 ehc->i.action |= ATA_EH_RESET;
3533 ehc->saved_xfer_mode[dev->devno] = 0;
3534 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3536 /* the link maybe in a deep sleep, wake it up */
3537 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3538 if (ata_is_host_link(link))
3539 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3542 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3546 /* Record and count probe trials on the ering. The specific
3547 * error mask used is irrelevant. Because a successful device
3548 * detection clears the ering, this count accumulates only if
3549 * there are consecutive failed probes.
3551 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3552 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3553 * forced to 1.5Gbps.
3555 * This is to work around cases where failed link speed
3556 * negotiation results in device misdetection leading to
3557 * infinite DEVXCHG or PHRDY CHG events.
3559 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3560 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3562 if (trials > ATA_EH_PROBE_TRIALS)
3563 sata_down_spd_limit(link, 1);
3568 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3570 struct ata_eh_context *ehc = &dev->link->eh_context;
3572 /* -EAGAIN from EH routine indicates retry without prejudice.
3573 * The requester is responsible for ensuring forward progress.
3576 ehc->tries[dev->devno]--;
3580 /* device missing or wrong IDENTIFY data, schedule probing */
3581 ehc->i.probe_mask |= (1 << dev->devno);
3584 /* give it just one more chance */
3585 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3588 if (ehc->tries[dev->devno] == 1) {
3589 /* This is the last chance, better to slow
3590 * down than lose it.
3592 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3593 if (dev->pio_mode > XFER_PIO_0)
3594 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3598 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3599 /* disable device if it has used up all its chances */
3600 ata_dev_disable(dev);
3602 /* detach if offline */
3603 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3604 ata_eh_detach_dev(dev);
3606 /* schedule probe if necessary */
3607 if (ata_eh_schedule_probe(dev)) {
3608 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3609 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3610 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3615 ehc->i.action |= ATA_EH_RESET;
3621 * ata_eh_recover - recover host port after error
3622 * @ap: host port to recover
3623 * @prereset: prereset method (can be NULL)
3624 * @softreset: softreset method (can be NULL)
3625 * @hardreset: hardreset method (can be NULL)
3626 * @postreset: postreset method (can be NULL)
3627 * @r_failed_link: out parameter for failed link
3629 * This is the alpha and omega, eum and yang, heart and soul of
3630 * libata exception handling. On entry, actions required to
3631 * recover each link and hotplug requests are recorded in the
3632 * link's eh_context. This function executes all the operations
3633 * with appropriate retrials and fallbacks to resurrect failed
3634 * devices, detach goners and greet newcomers.
3637 * Kernel thread context (may sleep).
3640 * 0 on success, -errno on failure.
3642 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3643 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3644 ata_postreset_fn_t postreset,
3645 struct ata_link **r_failed_link)
3647 struct ata_link *link;
3648 struct ata_device *dev;
3650 unsigned long flags, deadline;
3652 /* prep for recovery */
3653 ata_for_each_link(link, ap, EDGE) {
3654 struct ata_eh_context *ehc = &link->eh_context;
3656 /* re-enable link? */
3657 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3658 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3659 spin_lock_irqsave(ap->lock, flags);
3660 link->flags &= ~ATA_LFLAG_DISABLED;
3661 spin_unlock_irqrestore(ap->lock, flags);
3662 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3665 ata_for_each_dev(dev, link, ALL) {
3666 if (link->flags & ATA_LFLAG_NO_RETRY)
3667 ehc->tries[dev->devno] = 1;
3669 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3671 /* collect port action mask recorded in dev actions */
3672 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3673 ~ATA_EH_PERDEV_MASK;
3674 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3676 /* process hotplug request */
3677 if (dev->flags & ATA_DFLAG_DETACH)
3678 ata_eh_detach_dev(dev);
3680 /* schedule probe if necessary */
3681 if (!ata_dev_enabled(dev))
3682 ata_eh_schedule_probe(dev);
3689 /* if UNLOADING, finish immediately */
3690 if (ap->pflags & ATA_PFLAG_UNLOADING)
3694 ata_for_each_link(link, ap, EDGE) {
3695 struct ata_eh_context *ehc = &link->eh_context;
3697 /* skip EH if possible. */
3698 if (ata_eh_skip_recovery(link))
3701 ata_for_each_dev(dev, link, ALL)
3702 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3706 ata_for_each_link(link, ap, EDGE) {
3707 struct ata_eh_context *ehc = &link->eh_context;
3709 if (!(ehc->i.action & ATA_EH_RESET))
3712 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3713 prereset, softreset, hardreset, postreset);
3715 ata_link_err(link, "reset failed, giving up\n");
3724 * clears ATA_EH_PARK in eh_info and resets
3725 * ap->park_req_pending
3727 ata_eh_pull_park_action(ap);
3730 ata_for_each_link(link, ap, EDGE) {
3731 ata_for_each_dev(dev, link, ALL) {
3732 struct ata_eh_context *ehc = &link->eh_context;
3735 if (dev->class != ATA_DEV_ATA &&
3736 dev->class != ATA_DEV_ZAC)
3738 if (!(ehc->i.dev_action[dev->devno] &
3741 tmp = dev->unpark_deadline;
3742 if (time_before(deadline, tmp))
3744 else if (time_before_eq(tmp, jiffies))
3746 if (ehc->unloaded_mask & (1 << dev->devno))
3749 ata_eh_park_issue_cmd(dev, 1);
3754 if (time_before_eq(deadline, now))
3758 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3762 ata_for_each_link(link, ap, EDGE) {
3763 ata_for_each_dev(dev, link, ALL) {
3764 if (!(link->eh_context.unloaded_mask &
3768 ata_eh_park_issue_cmd(dev, 0);
3769 ata_eh_done(link, dev, ATA_EH_PARK);
3775 ata_for_each_link(link, ap, PMP_FIRST) {
3776 struct ata_eh_context *ehc = &link->eh_context;
3778 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3781 /* revalidate existing devices and attach new ones */
3782 rc = ata_eh_revalidate_and_attach(link, &dev);
3786 /* if PMP got attached, return, pmp EH will take care of it */
3787 if (link->device->class == ATA_DEV_PMP) {
3792 /* configure transfer mode if necessary */
3793 if (ehc->i.flags & ATA_EHI_SETMODE) {
3794 rc = ata_set_mode(link, &dev);
3797 ehc->i.flags &= ~ATA_EHI_SETMODE;
3800 /* If reset has been issued, clear UA to avoid
3801 * disrupting the current users of the device.
3803 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3804 ata_for_each_dev(dev, link, ALL) {
3805 if (dev->class != ATA_DEV_ATAPI)
3807 rc = atapi_eh_clear_ua(dev);
3810 if (zpodd_dev_enabled(dev))
3811 zpodd_post_poweron(dev);
3815 /* retry flush if necessary */
3816 ata_for_each_dev(dev, link, ALL) {
3817 if (dev->class != ATA_DEV_ATA &&
3818 dev->class != ATA_DEV_ZAC)
3820 rc = ata_eh_maybe_retry_flush(dev);
3826 /* configure link power saving */
3827 if (link->lpm_policy != ap->target_lpm_policy) {
3828 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3833 /* this link is okay now */
3840 ata_eh_handle_dev_fail(dev, rc);
3842 if (ata_port_is_frozen(ap)) {
3843 /* PMP reset requires working host port.
3844 * Can't retry if it's frozen.
3846 if (sata_pmp_attached(ap))
3856 if (rc && r_failed_link)
3857 *r_failed_link = link;
3863 * ata_eh_finish - finish up EH
3864 * @ap: host port to finish EH for
3866 * Recovery is complete. Clean up EH states and retry or finish
3872 void ata_eh_finish(struct ata_port *ap)
3874 struct ata_queued_cmd *qc;
3877 /* retry or finish qcs */
3878 ata_qc_for_each_raw(ap, qc, tag) {
3879 if (!(qc->flags & ATA_QCFLAG_EH))
3883 /* FIXME: Once EH migration is complete,
3884 * generate sense data in this function,
3885 * considering both err_mask and tf.
3887 if (qc->flags & ATA_QCFLAG_RETRY) {
3889 * Since qc->err_mask is set, ata_eh_qc_retry()
3890 * will not increment scmd->allowed, so upper
3891 * layer will only retry the command if it has
3892 * not already been retried too many times.
3894 ata_eh_qc_retry(qc);
3896 ata_eh_qc_complete(qc);
3899 if (qc->flags & ATA_QCFLAG_SENSE_VALID ||
3900 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) {
3901 ata_eh_qc_complete(qc);
3903 /* feed zero TF to sense generation */
3904 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3906 * Since qc->err_mask is not set,
3907 * ata_eh_qc_retry() will increment
3908 * scmd->allowed, so upper layer is guaranteed
3909 * to retry the command.
3911 ata_eh_qc_retry(qc);
3916 /* make sure nr_active_links is zero after EH */
3917 WARN_ON(ap->nr_active_links);
3918 ap->nr_active_links = 0;
3922 * ata_do_eh - do standard error handling
3923 * @ap: host port to handle error for
3925 * @prereset: prereset method (can be NULL)
3926 * @softreset: softreset method (can be NULL)
3927 * @hardreset: hardreset method (can be NULL)
3928 * @postreset: postreset method (can be NULL)
3930 * Perform standard error handling sequence.
3933 * Kernel thread context (may sleep).
3935 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3936 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3937 ata_postreset_fn_t postreset)
3939 struct ata_device *dev;
3945 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3948 ata_for_each_dev(dev, &ap->link, ALL)
3949 ata_dev_disable(dev);
3956 * ata_std_error_handler - standard error handler
3957 * @ap: host port to handle error for
3959 * Standard error handler
3962 * Kernel thread context (may sleep).
3964 void ata_std_error_handler(struct ata_port *ap)
3966 struct ata_port_operations *ops = ap->ops;
3967 ata_reset_fn_t hardreset = ops->hardreset;
3969 /* ignore built-in hardreset if SCR access is not available */
3970 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3973 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3975 EXPORT_SYMBOL_GPL(ata_std_error_handler);
3979 * ata_eh_handle_port_suspend - perform port suspend operation
3980 * @ap: port to suspend
3985 * Kernel thread context (may sleep).
3987 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3989 unsigned long flags;
3991 struct ata_device *dev;
3993 /* are we suspending? */
3994 spin_lock_irqsave(ap->lock, flags);
3995 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3996 ap->pm_mesg.event & PM_EVENT_RESUME) {
3997 spin_unlock_irqrestore(ap->lock, flags);
4000 spin_unlock_irqrestore(ap->lock, flags);
4002 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4005 * If we have a ZPODD attached, check its zero
4006 * power ready status before the port is frozen.
4007 * Only needed for runtime suspend.
4009 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4010 ata_for_each_dev(dev, &ap->link, ENABLED) {
4011 if (zpodd_dev_enabled(dev))
4012 zpodd_on_suspend(dev);
4017 ata_eh_freeze_port(ap);
4019 if (ap->ops->port_suspend)
4020 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4022 ata_acpi_set_state(ap, ap->pm_mesg);
4024 /* update the flags */
4025 spin_lock_irqsave(ap->lock, flags);
4027 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4029 ap->pflags |= ATA_PFLAG_SUSPENDED;
4030 else if (ata_port_is_frozen(ap))
4031 ata_port_schedule_eh(ap);
4033 spin_unlock_irqrestore(ap->lock, flags);
4039 * ata_eh_handle_port_resume - perform port resume operation
4040 * @ap: port to resume
4045 * Kernel thread context (may sleep).
4047 static void ata_eh_handle_port_resume(struct ata_port *ap)
4049 struct ata_link *link;
4050 struct ata_device *dev;
4051 unsigned long flags;
4053 /* are we resuming? */
4054 spin_lock_irqsave(ap->lock, flags);
4055 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4056 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4057 spin_unlock_irqrestore(ap->lock, flags);
4060 spin_unlock_irqrestore(ap->lock, flags);
4062 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4065 * Error timestamps are in jiffies which doesn't run while
4066 * suspended and PHY events during resume isn't too uncommon.
4067 * When the two are combined, it can lead to unnecessary speed
4068 * downs if the machine is suspended and resumed repeatedly.
4069 * Clear error history.
4071 ata_for_each_link(link, ap, HOST_FIRST)
4072 ata_for_each_dev(dev, link, ALL)
4073 ata_ering_clear(&dev->ering);
4075 ata_acpi_set_state(ap, ap->pm_mesg);
4077 if (ap->ops->port_resume)
4078 ap->ops->port_resume(ap);
4080 /* tell ACPI that we're resuming */
4081 ata_acpi_on_resume(ap);
4083 /* update the flags */
4084 spin_lock_irqsave(ap->lock, flags);
4085 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4086 spin_unlock_irqrestore(ap->lock, flags);
4088 #endif /* CONFIG_PM */