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 },
150 { .commands = CMDS(ATA_CMD_VERIFY),
151 .timeouts = ata_eh_reset_timeouts },
155 static void __ata_port_freeze(struct ata_port *ap);
156 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
157 struct ata_device **r_failed_dev);
159 static void ata_eh_handle_port_suspend(struct ata_port *ap);
160 static void ata_eh_handle_port_resume(struct ata_port *ap);
161 #else /* CONFIG_PM */
162 static void ata_eh_handle_port_suspend(struct ata_port *ap)
165 static void ata_eh_handle_port_resume(struct ata_port *ap)
167 #endif /* CONFIG_PM */
169 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
170 const char *fmt, va_list args)
172 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
173 ATA_EH_DESC_LEN - ehi->desc_len,
178 * __ata_ehi_push_desc - push error description without adding separator
180 * @fmt: printf format string
182 * Format string according to @fmt and append it to @ehi->desc.
185 * spin_lock_irqsave(host lock)
187 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
192 __ata_ehi_pushv_desc(ehi, fmt, args);
195 EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
198 * ata_ehi_push_desc - push error description with separator
200 * @fmt: printf format string
202 * Format string according to @fmt and append it to @ehi->desc.
203 * If @ehi->desc is not empty, ", " is added in-between.
206 * spin_lock_irqsave(host lock)
208 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
213 __ata_ehi_push_desc(ehi, ", ");
216 __ata_ehi_pushv_desc(ehi, fmt, args);
219 EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
222 * ata_ehi_clear_desc - clean error description
228 * spin_lock_irqsave(host lock)
230 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
235 EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
238 * ata_port_desc - append port description
239 * @ap: target ATA port
240 * @fmt: printf format string
242 * Format string according to @fmt and append it to port
243 * description. If port description is not empty, " " is added
244 * in-between. This function is to be used while initializing
245 * ata_host. The description is printed on host registration.
250 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
254 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
256 if (ap->link.eh_info.desc_len)
257 __ata_ehi_push_desc(&ap->link.eh_info, " ");
260 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
263 EXPORT_SYMBOL_GPL(ata_port_desc);
267 * ata_port_pbar_desc - append PCI BAR description
268 * @ap: target ATA port
269 * @bar: target PCI BAR
270 * @offset: offset into PCI BAR
271 * @name: name of the area
273 * If @offset is negative, this function formats a string which
274 * contains the name, address, size and type of the BAR and
275 * appends it to the port description. If @offset is zero or
276 * positive, only name and offsetted address is appended.
281 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
284 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
286 unsigned long long start, len;
288 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
290 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
293 start = (unsigned long long)pci_resource_start(pdev, bar);
294 len = (unsigned long long)pci_resource_len(pdev, bar);
297 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
299 ata_port_desc(ap, "%s 0x%llx", name,
300 start + (unsigned long long)offset);
302 EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
303 #endif /* CONFIG_PCI */
305 static int ata_lookup_timeout_table(u8 cmd)
309 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
312 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
321 * ata_internal_cmd_timeout - determine timeout for an internal command
322 * @dev: target device
323 * @cmd: internal command to be issued
325 * Determine timeout for internal command @cmd for @dev.
331 * Determined timeout.
333 unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
335 struct ata_eh_context *ehc = &dev->link->eh_context;
336 int ent = ata_lookup_timeout_table(cmd);
340 return ATA_EH_CMD_DFL_TIMEOUT;
342 idx = ehc->cmd_timeout_idx[dev->devno][ent];
343 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
347 * ata_internal_cmd_timed_out - notification for internal command timeout
348 * @dev: target device
349 * @cmd: internal command which timed out
351 * Notify EH that internal command @cmd for @dev timed out. This
352 * function should be called only for commands whose timeouts are
353 * determined using ata_internal_cmd_timeout().
358 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
360 struct ata_eh_context *ehc = &dev->link->eh_context;
361 int ent = ata_lookup_timeout_table(cmd);
367 idx = ehc->cmd_timeout_idx[dev->devno][ent];
368 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
369 ehc->cmd_timeout_idx[dev->devno][ent]++;
372 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
373 unsigned int err_mask)
375 struct ata_ering_entry *ent;
380 ering->cursor %= ATA_ERING_SIZE;
382 ent = &ering->ring[ering->cursor];
383 ent->eflags = eflags;
384 ent->err_mask = err_mask;
385 ent->timestamp = get_jiffies_64();
388 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
390 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
397 int ata_ering_map(struct ata_ering *ering,
398 int (*map_fn)(struct ata_ering_entry *, void *),
402 struct ata_ering_entry *ent;
406 ent = &ering->ring[idx];
409 rc = map_fn(ent, arg);
412 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
413 } while (idx != ering->cursor);
418 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
420 ent->eflags |= ATA_EFLAG_OLD_ER;
424 static void ata_ering_clear(struct ata_ering *ering)
426 ata_ering_map(ering, ata_ering_clear_cb, NULL);
429 static unsigned int ata_eh_dev_action(struct ata_device *dev)
431 struct ata_eh_context *ehc = &dev->link->eh_context;
433 return ehc->i.action | ehc->i.dev_action[dev->devno];
436 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
437 struct ata_eh_info *ehi, unsigned int action)
439 struct ata_device *tdev;
442 ehi->action &= ~action;
443 ata_for_each_dev(tdev, link, ALL)
444 ehi->dev_action[tdev->devno] &= ~action;
446 /* doesn't make sense for port-wide EH actions */
447 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
449 /* break ehi->action into ehi->dev_action */
450 if (ehi->action & action) {
451 ata_for_each_dev(tdev, link, ALL)
452 ehi->dev_action[tdev->devno] |=
453 ehi->action & action;
454 ehi->action &= ~action;
457 /* turn off the specified per-dev action */
458 ehi->dev_action[dev->devno] &= ~action;
463 * ata_eh_acquire - acquire EH ownership
464 * @ap: ATA port to acquire EH ownership for
466 * Acquire EH ownership for @ap. This is the basic exclusion
467 * mechanism for ports sharing a host. Only one port hanging off
468 * the same host can claim the ownership of EH.
473 void ata_eh_acquire(struct ata_port *ap)
475 mutex_lock(&ap->host->eh_mutex);
476 WARN_ON_ONCE(ap->host->eh_owner);
477 ap->host->eh_owner = current;
481 * ata_eh_release - release EH ownership
482 * @ap: ATA port to release EH ownership for
484 * Release EH ownership for @ap if the caller. The caller must
485 * have acquired EH ownership using ata_eh_acquire() previously.
490 void ata_eh_release(struct ata_port *ap)
492 WARN_ON_ONCE(ap->host->eh_owner != current);
493 ap->host->eh_owner = NULL;
494 mutex_unlock(&ap->host->eh_mutex);
497 static void ata_eh_dev_disable(struct ata_device *dev)
499 ata_acpi_on_disable(dev);
500 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
503 /* From now till the next successful probe, ering is used to
504 * track probe failures. Clear accumulated device error info.
506 ata_ering_clear(&dev->ering);
509 static void ata_eh_unload(struct ata_port *ap)
511 struct ata_link *link;
512 struct ata_device *dev;
516 * Unless we are restarting, transition all enabled devices to
517 * standby power mode.
519 if (system_state != SYSTEM_RESTART) {
520 ata_for_each_link(link, ap, PMP_FIRST) {
521 ata_for_each_dev(dev, link, ENABLED)
522 ata_dev_power_set_standby(dev);
527 * Restore SControl IPM and SPD for the next driver and
528 * disable attached devices.
530 ata_for_each_link(link, ap, PMP_FIRST) {
531 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
532 ata_for_each_dev(dev, link, ENABLED)
533 ata_eh_dev_disable(dev);
536 /* freeze and set UNLOADED */
537 spin_lock_irqsave(ap->lock, flags);
539 ata_port_freeze(ap); /* won't be thawed */
540 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
541 ap->pflags |= ATA_PFLAG_UNLOADED;
543 spin_unlock_irqrestore(ap->lock, flags);
547 * ata_scsi_error - SCSI layer error handler callback
548 * @host: SCSI host on which error occurred
550 * Handles SCSI-layer-thrown error events.
553 * Inherited from SCSI layer (none, can sleep)
558 void ata_scsi_error(struct Scsi_Host *host)
560 struct ata_port *ap = ata_shost_to_port(host);
562 LIST_HEAD(eh_work_q);
564 spin_lock_irqsave(host->host_lock, flags);
565 list_splice_init(&host->eh_cmd_q, &eh_work_q);
566 spin_unlock_irqrestore(host->host_lock, flags);
568 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
570 /* If we timed raced normal completion and there is nothing to
571 recover nr_timedout == 0 why exactly are we doing error recovery ? */
572 ata_scsi_port_error_handler(host, ap);
574 /* finish or retry handled scmd's and clean up */
575 WARN_ON(!list_empty(&eh_work_q));
580 * ata_scsi_cmd_error_handler - error callback for a list of commands
581 * @host: scsi host containing the port
582 * @ap: ATA port within the host
583 * @eh_work_q: list of commands to process
585 * process the given list of commands and return those finished to the
586 * ap->eh_done_q. This function is the first part of the libata error
587 * handler which processes a given list of failed commands.
589 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
590 struct list_head *eh_work_q)
594 struct scsi_cmnd *scmd, *tmp;
597 /* make sure sff pio task is not running */
598 ata_sff_flush_pio_task(ap);
600 /* synchronize with host lock and sort out timeouts */
603 * For EH, all qcs are finished in one of three ways -
604 * normal completion, error completion, and SCSI timeout.
605 * Both completions can race against SCSI timeout. When normal
606 * completion wins, the qc never reaches EH. When error
607 * completion wins, the qc has ATA_QCFLAG_EH set.
609 * When SCSI timeout wins, things are a bit more complex.
610 * Normal or error completion can occur after the timeout but
611 * before this point. In such cases, both types of
612 * completions are honored. A scmd is determined to have
613 * timed out iff its associated qc is active and not failed.
615 spin_lock_irqsave(ap->lock, flags);
618 * This must occur under the ap->lock as we don't want
619 * a polled recovery to race the real interrupt handler
621 * The lost_interrupt handler checks for any completed but
622 * non-notified command and completes much like an IRQ handler.
624 * We then fall into the error recovery code which will treat
625 * this as if normal completion won the race
627 if (ap->ops->lost_interrupt)
628 ap->ops->lost_interrupt(ap);
630 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
631 struct ata_queued_cmd *qc;
633 ata_qc_for_each_raw(ap, qc, i) {
634 if (qc->flags & ATA_QCFLAG_ACTIVE &&
639 if (i < ATA_MAX_QUEUE) {
640 /* the scmd has an associated qc */
641 if (!(qc->flags & ATA_QCFLAG_EH)) {
642 /* which hasn't failed yet, timeout */
643 qc->err_mask |= AC_ERR_TIMEOUT;
644 qc->flags |= ATA_QCFLAG_EH;
648 /* Normal completion occurred after
649 * SCSI timeout but before this point.
650 * Successfully complete it.
652 scmd->retries = scmd->allowed;
653 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
658 * If we have timed out qcs. They belong to EH from
659 * this point but the state of the controller is
660 * unknown. Freeze the port to make sure the IRQ
661 * handler doesn't diddle with those qcs. This must
662 * be done atomically w.r.t. setting ATA_QCFLAG_EH.
665 __ata_port_freeze(ap);
667 /* initialize eh_tries */
668 ap->eh_tries = ATA_EH_MAX_TRIES;
670 spin_unlock_irqrestore(ap->lock, flags);
672 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
675 * ata_scsi_port_error_handler - recover the port after the commands
676 * @host: SCSI host containing the port
679 * Handle the recovery of the port @ap after all the commands
680 * have been recovered.
682 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
685 struct ata_link *link;
687 /* acquire EH ownership */
690 /* kill fast drain timer */
691 del_timer_sync(&ap->fastdrain_timer);
693 /* process port resume request */
694 ata_eh_handle_port_resume(ap);
696 /* fetch & clear EH info */
697 spin_lock_irqsave(ap->lock, flags);
699 ata_for_each_link(link, ap, HOST_FIRST) {
700 struct ata_eh_context *ehc = &link->eh_context;
701 struct ata_device *dev;
703 memset(&link->eh_context, 0, sizeof(link->eh_context));
704 link->eh_context.i = link->eh_info;
705 memset(&link->eh_info, 0, sizeof(link->eh_info));
707 ata_for_each_dev(dev, link, ENABLED) {
708 int devno = dev->devno;
710 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
711 if (ata_ncq_enabled(dev))
712 ehc->saved_ncq_enabled |= 1 << devno;
714 /* If we are resuming, wake up the device */
715 if (ap->pflags & ATA_PFLAG_RESUMING)
716 ehc->i.dev_action[devno] |= ATA_EH_SET_ACTIVE;
720 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
721 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
722 ap->excl_link = NULL; /* don't maintain exclusion over EH */
724 spin_unlock_irqrestore(ap->lock, flags);
726 /* invoke EH, skip if unloading or suspended */
727 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
728 ap->ops->error_handler(ap);
730 /* if unloading, commence suicide */
731 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
732 !(ap->pflags & ATA_PFLAG_UNLOADED))
737 /* process port suspend request */
738 ata_eh_handle_port_suspend(ap);
741 * Exception might have happened after ->error_handler recovered the
742 * port but before this point. Repeat EH in such case.
744 spin_lock_irqsave(ap->lock, flags);
746 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
747 if (--ap->eh_tries) {
748 spin_unlock_irqrestore(ap->lock, flags);
752 "EH pending after %d tries, giving up\n",
754 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
757 /* this run is complete, make sure EH info is clear */
758 ata_for_each_link(link, ap, HOST_FIRST)
759 memset(&link->eh_info, 0, sizeof(link->eh_info));
762 * end eh (clear host_eh_scheduled) while holding ap->lock such that if
763 * exception occurs after this point but before EH completion, SCSI
764 * midlayer will re-initiate EH.
768 spin_unlock_irqrestore(ap->lock, flags);
771 scsi_eh_flush_done_q(&ap->eh_done_q);
774 spin_lock_irqsave(ap->lock, flags);
776 ap->pflags &= ~ATA_PFLAG_RESUMING;
778 if (ap->pflags & ATA_PFLAG_LOADING)
779 ap->pflags &= ~ATA_PFLAG_LOADING;
780 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
781 !(ap->flags & ATA_FLAG_SAS_HOST))
782 schedule_delayed_work(&ap->hotplug_task, 0);
784 if (ap->pflags & ATA_PFLAG_RECOVERED)
785 ata_port_info(ap, "EH complete\n");
787 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
789 /* tell wait_eh that we're done */
790 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
791 wake_up_all(&ap->eh_wait_q);
793 spin_unlock_irqrestore(ap->lock, flags);
795 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
798 * ata_port_wait_eh - Wait for the currently pending EH to complete
799 * @ap: Port to wait EH for
801 * Wait until the currently pending EH is complete.
804 * Kernel thread context (may sleep).
806 void ata_port_wait_eh(struct ata_port *ap)
812 spin_lock_irqsave(ap->lock, flags);
814 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
815 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
816 spin_unlock_irqrestore(ap->lock, flags);
818 spin_lock_irqsave(ap->lock, flags);
820 finish_wait(&ap->eh_wait_q, &wait);
822 spin_unlock_irqrestore(ap->lock, flags);
824 /* make sure SCSI EH is complete */
825 if (scsi_host_in_recovery(ap->scsi_host)) {
830 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
832 static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
834 struct ata_queued_cmd *qc;
838 /* count only non-internal commands */
839 ata_qc_for_each(ap, qc, tag) {
847 void ata_eh_fastdrain_timerfn(struct timer_list *t)
849 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
853 spin_lock_irqsave(ap->lock, flags);
855 cnt = ata_eh_nr_in_flight(ap);
861 if (cnt == ap->fastdrain_cnt) {
862 struct ata_queued_cmd *qc;
865 /* No progress during the last interval, tag all
866 * in-flight qcs as timed out and freeze the port.
868 ata_qc_for_each(ap, qc, tag) {
870 qc->err_mask |= AC_ERR_TIMEOUT;
875 /* some qcs have finished, give it another chance */
876 ap->fastdrain_cnt = cnt;
877 ap->fastdrain_timer.expires =
878 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
879 add_timer(&ap->fastdrain_timer);
883 spin_unlock_irqrestore(ap->lock, flags);
887 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
888 * @ap: target ATA port
889 * @fastdrain: activate fast drain
891 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
892 * is non-zero and EH wasn't pending before. Fast drain ensures
893 * that EH kicks in in timely manner.
896 * spin_lock_irqsave(host lock)
898 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
902 /* already scheduled? */
903 if (ap->pflags & ATA_PFLAG_EH_PENDING)
906 ap->pflags |= ATA_PFLAG_EH_PENDING;
911 /* do we have in-flight qcs? */
912 cnt = ata_eh_nr_in_flight(ap);
916 /* activate fast drain */
917 ap->fastdrain_cnt = cnt;
918 ap->fastdrain_timer.expires =
919 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
920 add_timer(&ap->fastdrain_timer);
924 * ata_qc_schedule_eh - schedule qc for error handling
925 * @qc: command to schedule error handling for
927 * Schedule error handling for @qc. EH will kick in as soon as
928 * other commands are drained.
931 * spin_lock_irqsave(host lock)
933 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
935 struct ata_port *ap = qc->ap;
937 qc->flags |= ATA_QCFLAG_EH;
938 ata_eh_set_pending(ap, 1);
940 /* The following will fail if timeout has already expired.
941 * ata_scsi_error() takes care of such scmds on EH entry.
942 * Note that ATA_QCFLAG_EH is unconditionally set after
943 * this function completes.
945 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
949 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
950 * @ap: ATA port to schedule EH for
952 * LOCKING: inherited from ata_port_schedule_eh
953 * spin_lock_irqsave(host lock)
955 void ata_std_sched_eh(struct ata_port *ap)
957 if (ap->pflags & ATA_PFLAG_INITIALIZING)
960 ata_eh_set_pending(ap, 1);
961 scsi_schedule_eh(ap->scsi_host);
963 trace_ata_std_sched_eh(ap);
965 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
968 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
969 * @ap: ATA port to end EH for
971 * In the libata object model there is a 1:1 mapping of ata_port to
972 * shost, so host fields can be directly manipulated under ap->lock, in
973 * the libsas case we need to hold a lock at the ha->level to coordinate
977 * spin_lock_irqsave(host lock)
979 void ata_std_end_eh(struct ata_port *ap)
981 struct Scsi_Host *host = ap->scsi_host;
983 host->host_eh_scheduled = 0;
985 EXPORT_SYMBOL(ata_std_end_eh);
989 * ata_port_schedule_eh - schedule error handling without a qc
990 * @ap: ATA port to schedule EH for
992 * Schedule error handling for @ap. EH will kick in as soon as
993 * all commands are drained.
996 * spin_lock_irqsave(host lock)
998 void ata_port_schedule_eh(struct ata_port *ap)
1000 /* see: ata_std_sched_eh, unless you know better */
1001 ap->ops->sched_eh(ap);
1003 EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
1005 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1007 struct ata_queued_cmd *qc;
1008 int tag, nr_aborted = 0;
1010 /* we're gonna abort all commands, no need for fast drain */
1011 ata_eh_set_pending(ap, 0);
1013 /* include internal tag in iteration */
1014 ata_qc_for_each_with_internal(ap, qc, tag) {
1015 if (qc && (!link || qc->dev->link == link)) {
1016 qc->flags |= ATA_QCFLAG_EH;
1017 ata_qc_complete(qc);
1023 ata_port_schedule_eh(ap);
1029 * ata_link_abort - abort all qc's on the link
1030 * @link: ATA link to abort qc's for
1032 * Abort all active qc's active on @link and schedule EH.
1035 * spin_lock_irqsave(host lock)
1038 * Number of aborted qc's.
1040 int ata_link_abort(struct ata_link *link)
1042 return ata_do_link_abort(link->ap, link);
1044 EXPORT_SYMBOL_GPL(ata_link_abort);
1047 * ata_port_abort - abort all qc's on the port
1048 * @ap: ATA port to abort qc's for
1050 * Abort all active qc's of @ap and schedule EH.
1053 * spin_lock_irqsave(host_set lock)
1056 * Number of aborted qc's.
1058 int ata_port_abort(struct ata_port *ap)
1060 return ata_do_link_abort(ap, NULL);
1062 EXPORT_SYMBOL_GPL(ata_port_abort);
1065 * __ata_port_freeze - freeze port
1066 * @ap: ATA port to freeze
1068 * This function is called when HSM violation or some other
1069 * condition disrupts normal operation of the port. Frozen port
1070 * is not allowed to perform any operation until the port is
1071 * thawed, which usually follows a successful reset.
1073 * ap->ops->freeze() callback can be used for freezing the port
1074 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1075 * port cannot be frozen hardware-wise, the interrupt handler
1076 * must ack and clear interrupts unconditionally while the port
1080 * spin_lock_irqsave(host lock)
1082 static void __ata_port_freeze(struct ata_port *ap)
1084 if (ap->ops->freeze)
1085 ap->ops->freeze(ap);
1087 ap->pflags |= ATA_PFLAG_FROZEN;
1089 trace_ata_port_freeze(ap);
1093 * ata_port_freeze - abort & freeze port
1094 * @ap: ATA port to freeze
1096 * Abort and freeze @ap. The freeze operation must be called
1097 * first, because some hardware requires special operations
1098 * before the taskfile registers are accessible.
1101 * spin_lock_irqsave(host lock)
1104 * Number of aborted commands.
1106 int ata_port_freeze(struct ata_port *ap)
1108 __ata_port_freeze(ap);
1110 return ata_port_abort(ap);
1112 EXPORT_SYMBOL_GPL(ata_port_freeze);
1115 * ata_eh_freeze_port - EH helper to freeze port
1116 * @ap: ATA port to freeze
1123 void ata_eh_freeze_port(struct ata_port *ap)
1125 unsigned long flags;
1127 spin_lock_irqsave(ap->lock, flags);
1128 __ata_port_freeze(ap);
1129 spin_unlock_irqrestore(ap->lock, flags);
1131 EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1134 * ata_eh_thaw_port - EH helper to thaw port
1135 * @ap: ATA port to thaw
1137 * Thaw frozen port @ap.
1142 void ata_eh_thaw_port(struct ata_port *ap)
1144 unsigned long flags;
1146 spin_lock_irqsave(ap->lock, flags);
1148 ap->pflags &= ~ATA_PFLAG_FROZEN;
1153 spin_unlock_irqrestore(ap->lock, flags);
1155 trace_ata_port_thaw(ap);
1158 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1163 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1165 struct ata_port *ap = qc->ap;
1166 struct scsi_cmnd *scmd = qc->scsicmd;
1167 unsigned long flags;
1169 spin_lock_irqsave(ap->lock, flags);
1170 qc->scsidone = ata_eh_scsidone;
1171 __ata_qc_complete(qc);
1172 WARN_ON(ata_tag_valid(qc->tag));
1173 spin_unlock_irqrestore(ap->lock, flags);
1175 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1179 * ata_eh_qc_complete - Complete an active ATA command from EH
1180 * @qc: Command to complete
1182 * Indicate to the mid and upper layers that an ATA command has
1183 * completed. To be used from EH.
1185 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1187 struct scsi_cmnd *scmd = qc->scsicmd;
1188 scmd->retries = scmd->allowed;
1189 __ata_eh_qc_complete(qc);
1193 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1194 * @qc: Command to retry
1196 * Indicate to the mid and upper layers that an ATA command
1197 * should be retried. To be used from EH.
1199 * SCSI midlayer limits the number of retries to scmd->allowed.
1200 * scmd->allowed is incremented for commands which get retried
1201 * due to unrelated failures (qc->err_mask is zero).
1203 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1205 struct scsi_cmnd *scmd = qc->scsicmd;
1208 __ata_eh_qc_complete(qc);
1212 * ata_dev_disable - disable ATA device
1213 * @dev: ATA device to disable
1220 void ata_dev_disable(struct ata_device *dev)
1222 if (!ata_dev_enabled(dev))
1225 ata_dev_warn(dev, "disable device\n");
1227 ata_eh_dev_disable(dev);
1229 EXPORT_SYMBOL_GPL(ata_dev_disable);
1232 * ata_eh_detach_dev - detach ATA device
1233 * @dev: ATA device to detach
1240 void ata_eh_detach_dev(struct ata_device *dev)
1242 struct ata_link *link = dev->link;
1243 struct ata_port *ap = link->ap;
1244 struct ata_eh_context *ehc = &link->eh_context;
1245 unsigned long flags;
1248 * If the device is still enabled, transition it to standby power mode
1249 * (i.e. spin down HDDs) and disable it.
1251 if (ata_dev_enabled(dev)) {
1252 ata_dev_power_set_standby(dev);
1253 ata_eh_dev_disable(dev);
1256 spin_lock_irqsave(ap->lock, flags);
1258 dev->flags &= ~ATA_DFLAG_DETACH;
1260 if (ata_scsi_offline_dev(dev)) {
1261 dev->flags |= ATA_DFLAG_DETACHED;
1262 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1265 /* clear per-dev EH info */
1266 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1267 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1268 ehc->saved_xfer_mode[dev->devno] = 0;
1269 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1271 spin_unlock_irqrestore(ap->lock, flags);
1275 * ata_eh_about_to_do - about to perform eh_action
1276 * @link: target ATA link
1277 * @dev: target ATA dev for per-dev action (can be NULL)
1278 * @action: action about to be performed
1280 * Called just before performing EH actions to clear related bits
1281 * in @link->eh_info such that eh actions are not unnecessarily
1287 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1288 unsigned int action)
1290 struct ata_port *ap = link->ap;
1291 struct ata_eh_info *ehi = &link->eh_info;
1292 struct ata_eh_context *ehc = &link->eh_context;
1293 unsigned long flags;
1295 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1297 spin_lock_irqsave(ap->lock, flags);
1299 ata_eh_clear_action(link, dev, ehi, action);
1301 /* About to take EH action, set RECOVERED. Ignore actions on
1302 * slave links as master will do them again.
1304 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1305 ap->pflags |= ATA_PFLAG_RECOVERED;
1307 spin_unlock_irqrestore(ap->lock, flags);
1311 * ata_eh_done - EH action complete
1312 * @link: ATA link for which EH actions are complete
1313 * @dev: target ATA dev for per-dev action (can be NULL)
1314 * @action: action just completed
1316 * Called right after performing EH actions to clear related bits
1317 * in @link->eh_context.
1322 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1323 unsigned int action)
1325 struct ata_eh_context *ehc = &link->eh_context;
1327 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1329 ata_eh_clear_action(link, dev, &ehc->i, action);
1333 * ata_err_string - convert err_mask to descriptive string
1334 * @err_mask: error mask to convert to string
1336 * Convert @err_mask to descriptive string. Errors are
1337 * prioritized according to severity and only the most severe
1338 * error is reported.
1344 * Descriptive string for @err_mask
1346 static const char *ata_err_string(unsigned int err_mask)
1348 if (err_mask & AC_ERR_HOST_BUS)
1349 return "host bus error";
1350 if (err_mask & AC_ERR_ATA_BUS)
1351 return "ATA bus error";
1352 if (err_mask & AC_ERR_TIMEOUT)
1354 if (err_mask & AC_ERR_HSM)
1355 return "HSM violation";
1356 if (err_mask & AC_ERR_SYSTEM)
1357 return "internal error";
1358 if (err_mask & AC_ERR_MEDIA)
1359 return "media error";
1360 if (err_mask & AC_ERR_INVALID)
1361 return "invalid argument";
1362 if (err_mask & AC_ERR_DEV)
1363 return "device error";
1364 if (err_mask & AC_ERR_NCQ)
1366 if (err_mask & AC_ERR_NODEV_HINT)
1367 return "Polling detection error";
1368 return "unknown error";
1372 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1373 * @dev: target ATAPI device
1374 * @r_sense_key: out parameter for sense_key
1376 * Perform ATAPI TEST_UNIT_READY.
1379 * EH context (may sleep).
1382 * 0 on success, AC_ERR_* mask on failure.
1384 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1386 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1387 struct ata_taskfile tf;
1388 unsigned int err_mask;
1390 ata_tf_init(dev, &tf);
1392 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1393 tf.command = ATA_CMD_PACKET;
1394 tf.protocol = ATAPI_PROT_NODATA;
1396 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1397 if (err_mask == AC_ERR_DEV)
1398 *r_sense_key = tf.error >> 4;
1403 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1404 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1406 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1407 * SENSE. This function is an EH helper.
1410 * Kernel thread context (may sleep).
1413 * true if sense data could be fetched, false otherwise.
1415 static bool ata_eh_request_sense(struct ata_queued_cmd *qc)
1417 struct scsi_cmnd *cmd = qc->scsicmd;
1418 struct ata_device *dev = qc->dev;
1419 struct ata_taskfile tf;
1420 unsigned int err_mask;
1422 if (ata_port_is_frozen(qc->ap)) {
1423 ata_dev_warn(dev, "sense data available but port frozen\n");
1427 if (!ata_id_sense_reporting_enabled(dev->id)) {
1428 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1432 ata_tf_init(dev, &tf);
1433 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1434 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1435 tf.command = ATA_CMD_REQ_SENSE_DATA;
1436 tf.protocol = ATA_PROT_NODATA;
1438 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1439 /* Ignore err_mask; ATA_ERR might be set */
1440 if (tf.status & ATA_SENSE) {
1441 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1442 /* Set sense without also setting scsicmd->result */
1443 scsi_build_sense_buffer(dev->flags & ATA_DFLAG_D_SENSE,
1444 cmd->sense_buffer, tf.lbah,
1446 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1450 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1451 tf.status, err_mask);
1458 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1459 * @dev: device to perform REQUEST_SENSE to
1460 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1461 * @dfl_sense_key: default sense key to use
1463 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1464 * SENSE. This function is EH helper.
1467 * Kernel thread context (may sleep).
1470 * 0 on success, AC_ERR_* mask on failure
1472 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1473 u8 *sense_buf, u8 dfl_sense_key)
1475 u8 cdb[ATAPI_CDB_LEN] =
1476 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1477 struct ata_port *ap = dev->link->ap;
1478 struct ata_taskfile tf;
1480 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1482 /* initialize sense_buf with the error register,
1483 * for the case where they are -not- overwritten
1485 sense_buf[0] = 0x70;
1486 sense_buf[2] = dfl_sense_key;
1488 /* some devices time out if garbage left in tf */
1489 ata_tf_init(dev, &tf);
1491 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1492 tf.command = ATA_CMD_PACKET;
1494 /* is it pointless to prefer PIO for "safety reasons"? */
1495 if (ap->flags & ATA_FLAG_PIO_DMA) {
1496 tf.protocol = ATAPI_PROT_DMA;
1497 tf.feature |= ATAPI_PKT_DMA;
1499 tf.protocol = ATAPI_PROT_PIO;
1500 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1504 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1505 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1509 * ata_eh_analyze_serror - analyze SError for a failed port
1510 * @link: ATA link to analyze SError for
1512 * Analyze SError if available and further determine cause of
1518 static void ata_eh_analyze_serror(struct ata_link *link)
1520 struct ata_eh_context *ehc = &link->eh_context;
1521 u32 serror = ehc->i.serror;
1522 unsigned int err_mask = 0, action = 0;
1525 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1526 err_mask |= AC_ERR_ATA_BUS;
1527 action |= ATA_EH_RESET;
1529 if (serror & SERR_PROTOCOL) {
1530 err_mask |= AC_ERR_HSM;
1531 action |= ATA_EH_RESET;
1533 if (serror & SERR_INTERNAL) {
1534 err_mask |= AC_ERR_SYSTEM;
1535 action |= ATA_EH_RESET;
1538 /* Determine whether a hotplug event has occurred. Both
1539 * SError.N/X are considered hotplug events for enabled or
1540 * host links. For disabled PMP links, only N bit is
1541 * considered as X bit is left at 1 for link plugging.
1543 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1544 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1545 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1546 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1548 hotplug_mask = SERR_PHYRDY_CHG;
1550 if (serror & hotplug_mask)
1551 ata_ehi_hotplugged(&ehc->i);
1553 ehc->i.err_mask |= err_mask;
1554 ehc->i.action |= action;
1558 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1559 * @qc: qc to analyze
1561 * Analyze taskfile of @qc and further determine cause of
1562 * failure. This function also requests ATAPI sense data if
1566 * Kernel thread context (may sleep).
1569 * Determined recovery action
1571 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1573 const struct ata_taskfile *tf = &qc->result_tf;
1574 unsigned int tmp, action = 0;
1575 u8 stat = tf->status, err = tf->error;
1577 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1578 qc->err_mask |= AC_ERR_HSM;
1579 return ATA_EH_RESET;
1582 if (stat & (ATA_ERR | ATA_DF)) {
1583 qc->err_mask |= AC_ERR_DEV;
1585 * Sense data reporting does not work if the
1586 * device fault bit is set.
1594 switch (qc->dev->class) {
1598 * Fetch the sense data explicitly if:
1599 * -It was a non-NCQ command that failed, or
1600 * -It was a NCQ command that failed, but the sense data
1601 * was not included in the NCQ command error log
1602 * (i.e. NCQ autosense is not supported by the device).
1604 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) &&
1605 (stat & ATA_SENSE) && ata_eh_request_sense(qc))
1606 set_status_byte(qc->scsicmd, SAM_STAT_CHECK_CONDITION);
1608 qc->err_mask |= AC_ERR_ATA_BUS;
1609 if (err & (ATA_UNC | ATA_AMNF))
1610 qc->err_mask |= AC_ERR_MEDIA;
1612 qc->err_mask |= AC_ERR_INVALID;
1616 if (!ata_port_is_frozen(qc->ap)) {
1617 tmp = atapi_eh_request_sense(qc->dev,
1618 qc->scsicmd->sense_buffer,
1619 qc->result_tf.error >> 4);
1621 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1623 qc->err_mask |= tmp;
1627 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1628 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1630 * SUCCESS here means that the sense code could be
1631 * evaluated and should be passed to the upper layers
1632 * for correct evaluation.
1633 * FAILED means the sense code could not be interpreted
1634 * and the device would need to be reset.
1635 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1636 * command would need to be retried.
1638 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1639 qc->flags |= ATA_QCFLAG_RETRY;
1640 qc->err_mask |= AC_ERR_OTHER;
1641 } else if (ret != SUCCESS) {
1642 qc->err_mask |= AC_ERR_HSM;
1645 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1646 action |= ATA_EH_RESET;
1651 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1656 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1660 base = ATA_ECAT_DUBIOUS_NONE;
1662 if (err_mask & AC_ERR_ATA_BUS)
1663 return base + ATA_ECAT_ATA_BUS;
1665 if (err_mask & AC_ERR_TIMEOUT)
1666 return base + ATA_ECAT_TOUT_HSM;
1668 if (eflags & ATA_EFLAG_IS_IO) {
1669 if (err_mask & AC_ERR_HSM)
1670 return base + ATA_ECAT_TOUT_HSM;
1672 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1673 return base + ATA_ECAT_UNK_DEV;
1679 struct speed_down_verdict_arg {
1682 int nr_errors[ATA_ECAT_NR];
1685 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1687 struct speed_down_verdict_arg *arg = void_arg;
1690 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1693 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1695 arg->nr_errors[cat]++;
1701 * ata_eh_speed_down_verdict - Determine speed down verdict
1702 * @dev: Device of interest
1704 * This function examines error ring of @dev and determines
1705 * whether NCQ needs to be turned off, transfer speed should be
1706 * stepped down, or falling back to PIO is necessary.
1708 * ECAT_ATA_BUS : ATA_BUS error for any command
1710 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1713 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1715 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1716 * data transfer hasn't been verified.
1720 * NCQ_OFF : Turn off NCQ.
1722 * SPEED_DOWN : Speed down transfer speed but don't fall back
1725 * FALLBACK_TO_PIO : Fall back to PIO.
1727 * Even if multiple verdicts are returned, only one action is
1728 * taken per error. An action triggered by non-DUBIOUS errors
1729 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1730 * This is to expedite speed down decisions right after device is
1731 * initially configured.
1733 * The following are speed down rules. #1 and #2 deal with
1736 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1737 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1739 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1740 * occurred during last 5 mins, NCQ_OFF.
1742 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1743 * occurred during last 5 mins, FALLBACK_TO_PIO
1745 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1746 * during last 10 mins, NCQ_OFF.
1748 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1749 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1752 * Inherited from caller.
1755 * OR of ATA_EH_SPDN_* flags.
1757 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1759 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1760 u64 j64 = get_jiffies_64();
1761 struct speed_down_verdict_arg arg;
1762 unsigned int verdict = 0;
1764 /* scan past 5 mins of error history */
1765 memset(&arg, 0, sizeof(arg));
1766 arg.since = j64 - min(j64, j5mins);
1767 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1769 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1770 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1771 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1772 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1774 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1775 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1776 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1778 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1779 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1780 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1781 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1783 /* scan past 10 mins of error history */
1784 memset(&arg, 0, sizeof(arg));
1785 arg.since = j64 - min(j64, j10mins);
1786 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1788 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1789 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1790 verdict |= ATA_EH_SPDN_NCQ_OFF;
1792 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1793 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1794 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1795 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1801 * ata_eh_speed_down - record error and speed down if necessary
1802 * @dev: Failed device
1803 * @eflags: mask of ATA_EFLAG_* flags
1804 * @err_mask: err_mask of the error
1806 * Record error and examine error history to determine whether
1807 * adjusting transmission speed is necessary. It also sets
1808 * transmission limits appropriately if such adjustment is
1812 * Kernel thread context (may sleep).
1815 * Determined recovery action.
1817 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1818 unsigned int eflags, unsigned int err_mask)
1820 struct ata_link *link = ata_dev_phys_link(dev);
1822 unsigned int verdict;
1823 unsigned int action = 0;
1825 /* don't bother if Cat-0 error */
1826 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1829 /* record error and determine whether speed down is necessary */
1830 ata_ering_record(&dev->ering, eflags, err_mask);
1831 verdict = ata_eh_speed_down_verdict(dev);
1834 if ((verdict & ATA_EH_SPDN_NCQ_OFF) && ata_ncq_enabled(dev)) {
1835 dev->flags |= ATA_DFLAG_NCQ_OFF;
1836 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1841 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1842 /* speed down SATA link speed if possible */
1843 if (sata_down_spd_limit(link, 0) == 0) {
1844 action |= ATA_EH_RESET;
1848 /* lower transfer mode */
1849 if (dev->spdn_cnt < 2) {
1850 static const int dma_dnxfer_sel[] =
1851 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1852 static const int pio_dnxfer_sel[] =
1853 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1856 if (dev->xfer_shift != ATA_SHIFT_PIO)
1857 sel = dma_dnxfer_sel[dev->spdn_cnt];
1859 sel = pio_dnxfer_sel[dev->spdn_cnt];
1863 if (ata_down_xfermask_limit(dev, sel) == 0) {
1864 action |= ATA_EH_RESET;
1870 /* Fall back to PIO? Slowing down to PIO is meaningless for
1871 * SATA ATA devices. Consider it only for PATA and SATAPI.
1873 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1874 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1875 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1876 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1878 action |= ATA_EH_RESET;
1885 /* device has been slowed down, blow error history */
1886 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1887 ata_ering_clear(&dev->ering);
1892 * ata_eh_worth_retry - analyze error and decide whether to retry
1893 * @qc: qc to possibly retry
1895 * Look at the cause of the error and decide if a retry
1896 * might be useful or not. We don't want to retry media errors
1897 * because the drive itself has probably already taken 10-30 seconds
1898 * doing its own internal retries before reporting the failure.
1900 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1902 if (qc->err_mask & AC_ERR_MEDIA)
1903 return 0; /* don't retry media errors */
1904 if (qc->flags & ATA_QCFLAG_IO)
1905 return 1; /* otherwise retry anything from fs stack */
1906 if (qc->err_mask & AC_ERR_INVALID)
1907 return 0; /* don't retry these */
1908 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1912 * ata_eh_quiet - check if we need to be quiet about a command error
1915 * Look at the qc flags anbd its scsi command request flags to determine
1916 * if we need to be quiet about the command failure.
1918 static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1920 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1921 qc->flags |= ATA_QCFLAG_QUIET;
1922 return qc->flags & ATA_QCFLAG_QUIET;
1925 static int ata_eh_read_sense_success_non_ncq(struct ata_link *link)
1927 struct ata_port *ap = link->ap;
1928 struct ata_queued_cmd *qc;
1930 qc = __ata_qc_from_tag(ap, link->active_tag);
1934 if (!(qc->flags & ATA_QCFLAG_EH) ||
1935 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
1939 if (!ata_eh_request_sense(qc))
1943 * If we have sense data, call scsi_check_sense() in order to set the
1944 * correct SCSI ML byte (if any). No point in checking the return value,
1945 * since the command has already completed successfully.
1947 scsi_check_sense(qc->scsicmd);
1952 static void ata_eh_get_success_sense(struct ata_link *link)
1954 struct ata_eh_context *ehc = &link->eh_context;
1955 struct ata_device *dev = link->device;
1956 struct ata_port *ap = link->ap;
1957 struct ata_queued_cmd *qc;
1960 if (!(ehc->i.dev_action[dev->devno] & ATA_EH_GET_SUCCESS_SENSE))
1963 /* if frozen, we can't do much */
1964 if (ata_port_is_frozen(ap)) {
1966 "successful sense data available but port frozen\n");
1971 * If the link has sactive set, then we have outstanding NCQ commands
1972 * and have to read the Successful NCQ Commands log to get the sense
1973 * data. Otherwise, we are dealing with a non-NCQ command and use
1974 * request sense ext command to retrieve the sense data.
1977 ret = ata_eh_read_sense_success_ncq_log(link);
1979 ret = ata_eh_read_sense_success_non_ncq(link);
1983 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
1988 * If we failed to get sense data for a successful command that ought to
1989 * have sense data, we cannot simply return BLK_STS_OK to user space.
1990 * This is because we can't know if the sense data that we couldn't get
1991 * was actually "DATA CURRENTLY UNAVAILABLE". Reporting such a command
1992 * as success to user space would result in a silent data corruption.
1993 * Thus, add a bogus ABORTED_COMMAND sense data to such commands, such
1994 * that SCSI will report these commands as BLK_STS_IOERR to user space.
1996 ata_qc_for_each_raw(ap, qc, tag) {
1997 if (!(qc->flags & ATA_QCFLAG_EH) ||
1998 !(qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) ||
2000 ata_dev_phys_link(qc->dev) != link)
2003 /* We managed to get sense for this success command, skip. */
2004 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2007 /* This success command did not have any sense data, skip. */
2008 if (!(qc->result_tf.status & ATA_SENSE))
2011 /* This success command had sense data, but we failed to get. */
2012 ata_scsi_set_sense(dev, qc->scsicmd, ABORTED_COMMAND, 0, 0);
2013 qc->flags |= ATA_QCFLAG_SENSE_VALID;
2015 ata_eh_done(link, dev, ATA_EH_GET_SUCCESS_SENSE);
2019 * ata_eh_link_autopsy - analyze error and determine recovery action
2020 * @link: host link to perform autopsy on
2022 * Analyze why @link failed and determine which recovery actions
2023 * are needed. This function also sets more detailed AC_ERR_*
2024 * values and fills sense data for ATAPI CHECK SENSE.
2027 * Kernel thread context (may sleep).
2029 static void ata_eh_link_autopsy(struct ata_link *link)
2031 struct ata_port *ap = link->ap;
2032 struct ata_eh_context *ehc = &link->eh_context;
2033 struct ata_queued_cmd *qc;
2034 struct ata_device *dev;
2035 unsigned int all_err_mask = 0, eflags = 0;
2036 int tag, nr_failed = 0, nr_quiet = 0;
2040 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2043 /* obtain and analyze SError */
2044 rc = sata_scr_read(link, SCR_ERROR, &serror);
2046 ehc->i.serror |= serror;
2047 ata_eh_analyze_serror(link);
2048 } else if (rc != -EOPNOTSUPP) {
2049 /* SError read failed, force reset and probing */
2050 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2051 ehc->i.action |= ATA_EH_RESET;
2052 ehc->i.err_mask |= AC_ERR_OTHER;
2055 /* analyze NCQ failure */
2056 ata_eh_analyze_ncq_error(link);
2059 * Check if this was a successful command that simply needs sense data.
2060 * Since the sense data is not part of the completion, we need to fetch
2061 * it using an additional command. Since this can't be done from irq
2062 * context, the sense data for successful commands are fetched by EH.
2064 ata_eh_get_success_sense(link);
2066 /* any real error trumps AC_ERR_OTHER */
2067 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2068 ehc->i.err_mask &= ~AC_ERR_OTHER;
2070 all_err_mask |= ehc->i.err_mask;
2072 ata_qc_for_each_raw(ap, qc, tag) {
2073 if (!(qc->flags & ATA_QCFLAG_EH) ||
2074 qc->flags & ATA_QCFLAG_RETRY ||
2075 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD ||
2076 ata_dev_phys_link(qc->dev) != link)
2079 /* inherit upper level err_mask */
2080 qc->err_mask |= ehc->i.err_mask;
2083 ehc->i.action |= ata_eh_analyze_tf(qc);
2085 /* DEV errors are probably spurious in case of ATA_BUS error */
2086 if (qc->err_mask & AC_ERR_ATA_BUS)
2087 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2090 /* any real error trumps unknown error */
2091 if (qc->err_mask & ~AC_ERR_OTHER)
2092 qc->err_mask &= ~AC_ERR_OTHER;
2095 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2096 * layers will determine whether the command is worth retrying
2097 * based on the sense data and device class/type. Otherwise,
2098 * determine directly if the command is worth retrying using its
2099 * error mask and flags.
2101 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2102 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2103 else if (ata_eh_worth_retry(qc))
2104 qc->flags |= ATA_QCFLAG_RETRY;
2106 /* accumulate error info */
2107 ehc->i.dev = qc->dev;
2108 all_err_mask |= qc->err_mask;
2109 if (qc->flags & ATA_QCFLAG_IO)
2110 eflags |= ATA_EFLAG_IS_IO;
2111 trace_ata_eh_link_autopsy_qc(qc);
2113 /* Count quiet errors */
2114 if (ata_eh_quiet(qc))
2119 /* If all failed commands requested silence, then be quiet */
2120 if (nr_quiet == nr_failed)
2121 ehc->i.flags |= ATA_EHI_QUIET;
2123 /* enforce default EH actions */
2124 if (ata_port_is_frozen(ap) ||
2125 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2126 ehc->i.action |= ATA_EH_RESET;
2127 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2128 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2129 ehc->i.action |= ATA_EH_REVALIDATE;
2131 /* If we have offending qcs and the associated failed device,
2132 * perform per-dev EH action only on the offending device.
2135 ehc->i.dev_action[ehc->i.dev->devno] |=
2136 ehc->i.action & ATA_EH_PERDEV_MASK;
2137 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2140 /* propagate timeout to host link */
2141 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2142 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2144 /* record error and consider speeding down */
2146 if (!dev && ((ata_link_max_devices(link) == 1 &&
2147 ata_dev_enabled(link->device))))
2151 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2152 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2153 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2154 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2159 * ata_eh_autopsy - analyze error and determine recovery action
2160 * @ap: host port to perform autopsy on
2162 * Analyze all links of @ap and determine why they failed and
2163 * which recovery actions are needed.
2166 * Kernel thread context (may sleep).
2168 void ata_eh_autopsy(struct ata_port *ap)
2170 struct ata_link *link;
2172 ata_for_each_link(link, ap, EDGE)
2173 ata_eh_link_autopsy(link);
2175 /* Handle the frigging slave link. Autopsy is done similarly
2176 * but actions and flags are transferred over to the master
2177 * link and handled from there.
2179 if (ap->slave_link) {
2180 struct ata_eh_context *mehc = &ap->link.eh_context;
2181 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2183 /* transfer control flags from master to slave */
2184 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2186 /* perform autopsy on the slave link */
2187 ata_eh_link_autopsy(ap->slave_link);
2189 /* transfer actions from slave to master and clear slave */
2190 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2191 mehc->i.action |= sehc->i.action;
2192 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2193 mehc->i.flags |= sehc->i.flags;
2194 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2197 /* Autopsy of fanout ports can affect host link autopsy.
2198 * Perform host link autopsy last.
2200 if (sata_pmp_attached(ap))
2201 ata_eh_link_autopsy(&ap->link);
2205 * ata_get_cmd_name - get name for ATA command
2206 * @command: ATA command code to get name for
2208 * Return a textual name of the given command or "unknown"
2213 const char *ata_get_cmd_name(u8 command)
2215 #ifdef CONFIG_ATA_VERBOSE_ERROR
2221 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2222 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2223 { ATA_CMD_STANDBY, "STANDBY" },
2224 { ATA_CMD_IDLE, "IDLE" },
2225 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2226 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2227 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2228 { ATA_CMD_NOP, "NOP" },
2229 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2230 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2231 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2232 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2233 { ATA_CMD_SERVICE, "SERVICE" },
2234 { ATA_CMD_READ, "READ DMA" },
2235 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2236 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2237 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2238 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2239 { ATA_CMD_WRITE, "WRITE DMA" },
2240 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2241 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2242 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2243 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2244 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2245 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2246 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2247 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2248 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2249 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2250 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2251 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2252 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2253 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2254 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2255 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2256 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2257 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2258 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2259 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2260 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2261 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2262 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2263 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2264 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2265 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2266 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2267 { ATA_CMD_SLEEP, "SLEEP" },
2268 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2269 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2270 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2271 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2272 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2273 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2274 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2275 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2276 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2277 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2278 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2279 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2280 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2281 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2282 { ATA_CMD_PMP_READ, "READ BUFFER" },
2283 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2284 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2285 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2286 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2287 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2288 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2289 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2290 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2291 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2292 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2293 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2294 { ATA_CMD_SMART, "SMART" },
2295 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2296 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2297 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2298 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2299 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2300 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2301 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2302 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2303 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2304 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2305 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2306 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2307 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2308 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2309 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2310 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2311 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2312 { ATA_CMD_RESTORE, "RECALIBRATE" },
2313 { 0, NULL } /* terminate list */
2317 for (i = 0; cmd_descr[i].text; i++)
2318 if (cmd_descr[i].command == command)
2319 return cmd_descr[i].text;
2324 EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2327 * ata_eh_link_report - report error handling to user
2328 * @link: ATA link EH is going on
2330 * Report EH to user.
2335 static void ata_eh_link_report(struct ata_link *link)
2337 struct ata_port *ap = link->ap;
2338 struct ata_eh_context *ehc = &link->eh_context;
2339 struct ata_queued_cmd *qc;
2340 const char *frozen, *desc;
2341 char tries_buf[16] = "";
2342 int tag, nr_failed = 0;
2344 if (ehc->i.flags & ATA_EHI_QUIET)
2348 if (ehc->i.desc[0] != '\0')
2351 ata_qc_for_each_raw(ap, qc, tag) {
2352 if (!(qc->flags & ATA_QCFLAG_EH) ||
2353 ata_dev_phys_link(qc->dev) != link ||
2354 ((qc->flags & ATA_QCFLAG_QUIET) &&
2355 qc->err_mask == AC_ERR_DEV))
2357 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2363 if (!nr_failed && !ehc->i.err_mask)
2367 if (ata_port_is_frozen(ap))
2370 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2371 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2375 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2376 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2377 ehc->i.err_mask, link->sactive, ehc->i.serror,
2378 ehc->i.action, frozen, tries_buf);
2380 ata_dev_err(ehc->i.dev, "%s\n", desc);
2382 ata_link_err(link, "exception Emask 0x%x "
2383 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2384 ehc->i.err_mask, link->sactive, ehc->i.serror,
2385 ehc->i.action, frozen, tries_buf);
2387 ata_link_err(link, "%s\n", desc);
2390 #ifdef CONFIG_ATA_VERBOSE_ERROR
2393 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2394 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2395 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2396 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2397 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2398 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2399 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2400 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2401 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2402 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2403 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2404 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2405 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2406 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2407 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2408 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2409 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2410 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2413 ata_qc_for_each_raw(ap, qc, tag) {
2414 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2415 char data_buf[20] = "";
2416 char cdb_buf[70] = "";
2418 if (!(qc->flags & ATA_QCFLAG_EH) ||
2419 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2422 if (qc->dma_dir != DMA_NONE) {
2423 static const char *dma_str[] = {
2424 [DMA_BIDIRECTIONAL] = "bidi",
2425 [DMA_TO_DEVICE] = "out",
2426 [DMA_FROM_DEVICE] = "in",
2428 const char *prot_str = NULL;
2430 switch (qc->tf.protocol) {
2431 case ATA_PROT_UNKNOWN:
2432 prot_str = "unknown";
2434 case ATA_PROT_NODATA:
2435 prot_str = "nodata";
2444 prot_str = "ncq dma";
2446 case ATA_PROT_NCQ_NODATA:
2447 prot_str = "ncq nodata";
2449 case ATAPI_PROT_NODATA:
2450 prot_str = "nodata";
2452 case ATAPI_PROT_PIO:
2455 case ATAPI_PROT_DMA:
2459 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2460 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2463 if (ata_is_atapi(qc->tf.protocol)) {
2464 const u8 *cdb = qc->cdb;
2465 size_t cdb_len = qc->dev->cdb_len;
2468 cdb = qc->scsicmd->cmnd;
2469 cdb_len = qc->scsicmd->cmd_len;
2471 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2474 ata_dev_err(qc->dev, "failed command: %s\n",
2475 ata_get_cmd_name(cmd->command));
2477 ata_dev_err(qc->dev,
2478 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2480 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2481 "Emask 0x%x (%s)%s\n",
2482 cmd->command, cmd->feature, cmd->nsect,
2483 cmd->lbal, cmd->lbam, cmd->lbah,
2484 cmd->hob_feature, cmd->hob_nsect,
2485 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2486 cmd->device, qc->tag, data_buf, cdb_buf,
2487 res->status, res->error, res->nsect,
2488 res->lbal, res->lbam, res->lbah,
2489 res->hob_feature, res->hob_nsect,
2490 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2491 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2492 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2494 #ifdef CONFIG_ATA_VERBOSE_ERROR
2495 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2496 ATA_SENSE | ATA_ERR)) {
2497 if (res->status & ATA_BUSY)
2498 ata_dev_err(qc->dev, "status: { Busy }\n");
2500 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2501 res->status & ATA_DRDY ? "DRDY " : "",
2502 res->status & ATA_DF ? "DF " : "",
2503 res->status & ATA_DRQ ? "DRQ " : "",
2504 res->status & ATA_SENSE ? "SENSE " : "",
2505 res->status & ATA_ERR ? "ERR " : "");
2508 if (cmd->command != ATA_CMD_PACKET &&
2509 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2511 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2512 res->error & ATA_ICRC ? "ICRC " : "",
2513 res->error & ATA_UNC ? "UNC " : "",
2514 res->error & ATA_AMNF ? "AMNF " : "",
2515 res->error & ATA_IDNF ? "IDNF " : "",
2516 res->error & ATA_ABORTED ? "ABRT " : "");
2522 * ata_eh_report - report error handling to user
2523 * @ap: ATA port to report EH about
2525 * Report EH to user.
2530 void ata_eh_report(struct ata_port *ap)
2532 struct ata_link *link;
2534 ata_for_each_link(link, ap, HOST_FIRST)
2535 ata_eh_link_report(link);
2538 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2539 unsigned int *classes, unsigned long deadline,
2542 struct ata_device *dev;
2545 ata_for_each_dev(dev, link, ALL)
2546 classes[dev->devno] = ATA_DEV_UNKNOWN;
2548 return reset(link, classes, deadline);
2551 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2553 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2557 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2562 int ata_eh_reset(struct ata_link *link, int classify,
2563 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2564 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2566 struct ata_port *ap = link->ap;
2567 struct ata_link *slave = ap->slave_link;
2568 struct ata_eh_context *ehc = &link->eh_context;
2569 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2570 unsigned int *classes = ehc->classes;
2571 unsigned int lflags = link->flags;
2572 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2573 int max_tries = 0, try = 0;
2574 struct ata_link *failed_link;
2575 struct ata_device *dev;
2576 unsigned long deadline, now;
2577 ata_reset_fn_t reset;
2578 unsigned long flags;
2585 while (ata_eh_reset_timeouts[max_tries] != UINT_MAX)
2587 if (link->flags & ATA_LFLAG_RST_ONCE)
2589 if (link->flags & ATA_LFLAG_NO_HRST)
2591 if (link->flags & ATA_LFLAG_NO_SRST)
2594 /* make sure each reset attempt is at least COOL_DOWN apart */
2595 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2597 WARN_ON(time_after(ehc->last_reset, now));
2598 deadline = ata_deadline(ehc->last_reset,
2599 ATA_EH_RESET_COOL_DOWN);
2600 if (time_before(now, deadline))
2601 schedule_timeout_uninterruptible(deadline - now);
2604 spin_lock_irqsave(ap->lock, flags);
2605 ap->pflags |= ATA_PFLAG_RESETTING;
2606 spin_unlock_irqrestore(ap->lock, flags);
2608 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2610 ata_for_each_dev(dev, link, ALL) {
2611 /* If we issue an SRST then an ATA drive (not ATAPI)
2612 * may change configuration and be in PIO0 timing. If
2613 * we do a hard reset (or are coming from power on)
2614 * this is true for ATA or ATAPI. Until we've set a
2615 * suitable controller mode we should not touch the
2616 * bus as we may be talking too fast.
2618 dev->pio_mode = XFER_PIO_0;
2619 dev->dma_mode = 0xff;
2621 /* If the controller has a pio mode setup function
2622 * then use it to set the chipset to rights. Don't
2623 * touch the DMA setup as that will be dealt with when
2624 * configuring devices.
2626 if (ap->ops->set_piomode)
2627 ap->ops->set_piomode(ap, dev);
2630 /* prefer hardreset */
2632 ehc->i.action &= ~ATA_EH_RESET;
2635 ehc->i.action |= ATA_EH_HARDRESET;
2636 } else if (softreset) {
2638 ehc->i.action |= ATA_EH_SOFTRESET;
2642 unsigned long deadline = ata_deadline(jiffies,
2643 ATA_EH_PRERESET_TIMEOUT);
2646 sehc->i.action &= ~ATA_EH_RESET;
2647 sehc->i.action |= ehc->i.action;
2650 rc = prereset(link, deadline);
2652 /* If present, do prereset on slave link too. Reset
2653 * is skipped iff both master and slave links report
2654 * -ENOENT or clear ATA_EH_RESET.
2656 if (slave && (rc == 0 || rc == -ENOENT)) {
2659 tmp = prereset(slave, deadline);
2663 ehc->i.action |= sehc->i.action;
2667 if (rc == -ENOENT) {
2668 ata_link_dbg(link, "port disabled--ignoring\n");
2669 ehc->i.action &= ~ATA_EH_RESET;
2671 ata_for_each_dev(dev, link, ALL)
2672 classes[dev->devno] = ATA_DEV_NONE;
2677 "prereset failed (errno=%d)\n",
2682 /* prereset() might have cleared ATA_EH_RESET. If so,
2683 * bang classes, thaw and return.
2685 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2686 ata_for_each_dev(dev, link, ALL)
2687 classes[dev->devno] = ATA_DEV_NONE;
2688 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2689 ata_eh_thaw_port(ap);
2699 if (ata_is_host_link(link))
2700 ata_eh_freeze_port(ap);
2702 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2706 ata_link_info(link, "%s resetting link\n",
2707 reset == softreset ? "soft" : "hard");
2709 /* mark that this EH session started with reset */
2710 ehc->last_reset = jiffies;
2711 if (reset == hardreset) {
2712 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2713 trace_ata_link_hardreset_begin(link, classes, deadline);
2715 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2716 trace_ata_link_softreset_begin(link, classes, deadline);
2719 rc = ata_do_reset(link, reset, classes, deadline, true);
2720 if (reset == hardreset)
2721 trace_ata_link_hardreset_end(link, classes, rc);
2723 trace_ata_link_softreset_end(link, classes, rc);
2724 if (rc && rc != -EAGAIN) {
2729 /* hardreset slave link if existent */
2730 if (slave && reset == hardreset) {
2734 ata_link_info(slave, "hard resetting link\n");
2736 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2737 trace_ata_slave_hardreset_begin(slave, classes,
2739 tmp = ata_do_reset(slave, reset, classes, deadline,
2741 trace_ata_slave_hardreset_end(slave, classes, tmp);
2749 failed_link = slave;
2755 /* perform follow-up SRST if necessary */
2756 if (reset == hardreset &&
2757 ata_eh_followup_srst_needed(link, rc)) {
2762 "follow-up softreset required but no softreset available\n");
2768 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2769 trace_ata_link_softreset_begin(link, classes, deadline);
2770 rc = ata_do_reset(link, reset, classes, deadline, true);
2771 trace_ata_link_softreset_end(link, classes, rc);
2780 "no reset method available, skipping reset\n");
2781 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2782 lflags |= ATA_LFLAG_ASSUME_ATA;
2786 * Post-reset processing
2788 ata_for_each_dev(dev, link, ALL) {
2789 /* After the reset, the device state is PIO 0 and the
2790 * controller state is undefined. Reset also wakes up
2791 * drives from sleeping mode.
2793 dev->pio_mode = XFER_PIO_0;
2794 dev->flags &= ~ATA_DFLAG_SLEEPING;
2796 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2799 /* apply class override */
2800 if (lflags & ATA_LFLAG_ASSUME_ATA)
2801 classes[dev->devno] = ATA_DEV_ATA;
2802 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2803 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2806 /* record current link speed */
2807 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2808 link->sata_spd = (sstatus >> 4) & 0xf;
2809 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2810 slave->sata_spd = (sstatus >> 4) & 0xf;
2813 if (ata_is_host_link(link))
2814 ata_eh_thaw_port(ap);
2816 /* postreset() should clear hardware SError. Although SError
2817 * is cleared during link resume, clearing SError here is
2818 * necessary as some PHYs raise hotplug events after SRST.
2819 * This introduces race condition where hotplug occurs between
2820 * reset and here. This race is mediated by cross checking
2821 * link onlineness and classification result later.
2824 postreset(link, classes);
2825 trace_ata_link_postreset(link, classes, rc);
2827 postreset(slave, classes);
2828 trace_ata_slave_postreset(slave, classes, rc);
2832 /* clear cached SError */
2833 spin_lock_irqsave(link->ap->lock, flags);
2834 link->eh_info.serror = 0;
2836 slave->eh_info.serror = 0;
2837 spin_unlock_irqrestore(link->ap->lock, flags);
2840 * Make sure onlineness and classification result correspond.
2841 * Hotplug could have happened during reset and some
2842 * controllers fail to wait while a drive is spinning up after
2843 * being hotplugged causing misdetection. By cross checking
2844 * link on/offlineness and classification result, those
2845 * conditions can be reliably detected and retried.
2848 ata_for_each_dev(dev, link, ALL) {
2849 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2850 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2851 ata_dev_dbg(dev, "link online but device misclassified\n");
2852 classes[dev->devno] = ATA_DEV_NONE;
2855 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2856 if (ata_class_enabled(classes[dev->devno]))
2858 "link offline, clearing class %d to NONE\n",
2859 classes[dev->devno]);
2860 classes[dev->devno] = ATA_DEV_NONE;
2861 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2863 "link status unknown, clearing UNKNOWN to NONE\n");
2864 classes[dev->devno] = ATA_DEV_NONE;
2868 if (classify && nr_unknown) {
2869 if (try < max_tries) {
2871 "link online but %d devices misclassified, retrying\n",
2878 "link online but %d devices misclassified, "
2879 "device detection might fail\n", nr_unknown);
2882 /* reset successful, schedule revalidation */
2883 ata_eh_done(link, NULL, ATA_EH_RESET);
2885 ata_eh_done(slave, NULL, ATA_EH_RESET);
2886 ehc->last_reset = jiffies; /* update to completion time */
2887 ehc->i.action |= ATA_EH_REVALIDATE;
2888 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2892 /* clear hotplug flag */
2893 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2895 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2897 spin_lock_irqsave(ap->lock, flags);
2898 ap->pflags &= ~ATA_PFLAG_RESETTING;
2899 spin_unlock_irqrestore(ap->lock, flags);
2904 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2905 if (!ata_is_host_link(link) &&
2906 sata_scr_read(link, SCR_STATUS, &sstatus))
2909 if (try >= max_tries) {
2911 * Thaw host port even if reset failed, so that the port
2912 * can be retried on the next phy event. This risks
2913 * repeated EH runs but seems to be a better tradeoff than
2914 * shutting down a port after a botched hotplug attempt.
2916 if (ata_is_host_link(link))
2917 ata_eh_thaw_port(ap);
2918 ata_link_warn(link, "%s failed\n",
2919 reset == hardreset ? "hardreset" : "softreset");
2924 if (time_before(now, deadline)) {
2925 unsigned long delta = deadline - now;
2927 ata_link_warn(failed_link,
2928 "reset failed (errno=%d), retrying in %u secs\n",
2929 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2933 delta = schedule_timeout_uninterruptible(delta);
2938 * While disks spinup behind PMP, some controllers fail sending SRST.
2939 * They need to be reset - as well as the PMP - before retrying.
2941 if (rc == -ERESTART) {
2942 if (ata_is_host_link(link))
2943 ata_eh_thaw_port(ap);
2947 if (try == max_tries - 1) {
2948 sata_down_spd_limit(link, 0);
2950 sata_down_spd_limit(slave, 0);
2951 } else if (rc == -EPIPE)
2952 sata_down_spd_limit(failed_link, 0);
2959 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2961 struct ata_link *link;
2962 struct ata_device *dev;
2963 unsigned long flags;
2966 * This function can be thought of as an extended version of
2967 * ata_eh_about_to_do() specially crafted to accommodate the
2968 * requirements of ATA_EH_PARK handling. Since the EH thread
2969 * does not leave the do {} while () loop in ata_eh_recover as
2970 * long as the timeout for a park request to *one* device on
2971 * the port has not expired, and since we still want to pick
2972 * up park requests to other devices on the same port or
2973 * timeout updates for the same device, we have to pull
2974 * ATA_EH_PARK actions from eh_info into eh_context.i
2975 * ourselves at the beginning of each pass over the loop.
2977 * Additionally, all write accesses to &ap->park_req_pending
2978 * through reinit_completion() (see below) or complete_all()
2979 * (see ata_scsi_park_store()) are protected by the host lock.
2980 * As a result we have that park_req_pending.done is zero on
2981 * exit from this function, i.e. when ATA_EH_PARK actions for
2982 * *all* devices on port ap have been pulled into the
2983 * respective eh_context structs. If, and only if,
2984 * park_req_pending.done is non-zero by the time we reach
2985 * wait_for_completion_timeout(), another ATA_EH_PARK action
2986 * has been scheduled for at least one of the devices on port
2987 * ap and we have to cycle over the do {} while () loop in
2988 * ata_eh_recover() again.
2991 spin_lock_irqsave(ap->lock, flags);
2992 reinit_completion(&ap->park_req_pending);
2993 ata_for_each_link(link, ap, EDGE) {
2994 ata_for_each_dev(dev, link, ALL) {
2995 struct ata_eh_info *ehi = &link->eh_info;
2997 link->eh_context.i.dev_action[dev->devno] |=
2998 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2999 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3002 spin_unlock_irqrestore(ap->lock, flags);
3005 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3007 struct ata_eh_context *ehc = &dev->link->eh_context;
3008 struct ata_taskfile tf;
3009 unsigned int err_mask;
3011 ata_tf_init(dev, &tf);
3013 ehc->unloaded_mask |= 1 << dev->devno;
3014 tf.command = ATA_CMD_IDLEIMMEDIATE;
3020 ehc->unloaded_mask &= ~(1 << dev->devno);
3021 tf.command = ATA_CMD_CHK_POWER;
3024 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3025 tf.protocol = ATA_PROT_NODATA;
3026 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3027 if (park && (err_mask || tf.lbal != 0xc4)) {
3028 ata_dev_err(dev, "head unload failed!\n");
3029 ehc->unloaded_mask &= ~(1 << dev->devno);
3033 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3034 struct ata_device **r_failed_dev)
3036 struct ata_port *ap = link->ap;
3037 struct ata_eh_context *ehc = &link->eh_context;
3038 struct ata_device *dev;
3039 unsigned int new_mask = 0;
3040 unsigned long flags;
3043 /* For PATA drive side cable detection to work, IDENTIFY must
3044 * be done backwards such that PDIAG- is released by the slave
3045 * device before the master device is identified.
3047 ata_for_each_dev(dev, link, ALL_REVERSE) {
3048 unsigned int action = ata_eh_dev_action(dev);
3049 unsigned int readid_flags = 0;
3051 if (ehc->i.flags & ATA_EHI_DID_RESET)
3052 readid_flags |= ATA_READID_POSTRESET;
3054 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3055 WARN_ON(dev->class == ATA_DEV_PMP);
3058 * The link may be in a deep sleep, wake it up.
3060 * If the link is in deep sleep, ata_phys_link_offline()
3061 * will return true, causing the revalidation to fail,
3062 * which leads to a (potentially) needless hard reset.
3064 * ata_eh_recover() will later restore the link policy
3065 * to ap->target_lpm_policy after revalidation is done.
3067 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3068 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
3074 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3079 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3080 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3085 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3087 /* Configuration may have changed, reconfigure
3090 ehc->i.flags |= ATA_EHI_SETMODE;
3092 /* schedule the scsi_rescan_device() here */
3093 schedule_delayed_work(&ap->scsi_rescan_task, 0);
3094 } else if (dev->class == ATA_DEV_UNKNOWN &&
3095 ehc->tries[dev->devno] &&
3096 ata_class_enabled(ehc->classes[dev->devno])) {
3097 /* Temporarily set dev->class, it will be
3098 * permanently set once all configurations are
3099 * complete. This is necessary because new
3100 * device configuration is done in two
3103 dev->class = ehc->classes[dev->devno];
3105 if (dev->class == ATA_DEV_PMP)
3106 rc = sata_pmp_attach(dev);
3108 rc = ata_dev_read_id(dev, &dev->class,
3109 readid_flags, dev->id);
3111 /* read_id might have changed class, store and reset */
3112 ehc->classes[dev->devno] = dev->class;
3113 dev->class = ATA_DEV_UNKNOWN;
3117 /* clear error info accumulated during probe */
3118 ata_ering_clear(&dev->ering);
3119 new_mask |= 1 << dev->devno;
3122 /* IDENTIFY was issued to non-existent
3123 * device. No need to reset. Just
3124 * thaw and ignore the device.
3126 ata_eh_thaw_port(ap);
3134 /* PDIAG- should have been released, ask cable type if post-reset */
3135 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3136 if (ap->ops->cable_detect)
3137 ap->cbl = ap->ops->cable_detect(ap);
3141 /* Configure new devices forward such that user doesn't see
3142 * device detection messages backwards.
3144 ata_for_each_dev(dev, link, ALL) {
3145 if (!(new_mask & (1 << dev->devno)))
3148 dev->class = ehc->classes[dev->devno];
3150 if (dev->class == ATA_DEV_PMP)
3153 ehc->i.flags |= ATA_EHI_PRINTINFO;
3154 rc = ata_dev_configure(dev);
3155 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3157 dev->class = ATA_DEV_UNKNOWN;
3161 spin_lock_irqsave(ap->lock, flags);
3162 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3163 spin_unlock_irqrestore(ap->lock, flags);
3165 /* new device discovered, configure xfermode */
3166 ehc->i.flags |= ATA_EHI_SETMODE;
3172 *r_failed_dev = dev;
3177 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3178 * @link: link on which timings will be programmed
3179 * @r_failed_dev: out parameter for failed device
3181 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3182 * ata_set_mode() fails, pointer to the failing device is
3183 * returned in @r_failed_dev.
3186 * PCI/etc. bus probe sem.
3189 * 0 on success, negative errno otherwise
3191 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3193 struct ata_port *ap = link->ap;
3194 struct ata_device *dev;
3197 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3198 ata_for_each_dev(dev, link, ENABLED) {
3199 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3200 struct ata_ering_entry *ent;
3202 ent = ata_ering_top(&dev->ering);
3204 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3208 /* has private set_mode? */
3209 if (ap->ops->set_mode)
3210 rc = ap->ops->set_mode(link, r_failed_dev);
3212 rc = ata_do_set_mode(link, r_failed_dev);
3214 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3215 ata_for_each_dev(dev, link, ENABLED) {
3216 struct ata_eh_context *ehc = &link->eh_context;
3217 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3218 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3220 if (dev->xfer_mode != saved_xfer_mode ||
3221 ata_ncq_enabled(dev) != saved_ncq)
3222 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3229 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3230 * @dev: ATAPI device to clear UA for
3232 * Resets and other operations can make an ATAPI device raise
3233 * UNIT ATTENTION which causes the next operation to fail. This
3234 * function clears UA.
3237 * EH context (may sleep).
3240 * 0 on success, -errno on failure.
3242 static int atapi_eh_clear_ua(struct ata_device *dev)
3246 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3247 u8 *sense_buffer = dev->link->ap->sector_buf;
3249 unsigned int err_mask;
3251 err_mask = atapi_eh_tur(dev, &sense_key);
3252 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3254 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3259 if (!err_mask || sense_key != UNIT_ATTENTION)
3262 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3264 ata_dev_warn(dev, "failed to clear "
3265 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3270 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3277 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3278 * @dev: ATA device which may need FLUSH retry
3280 * If @dev failed FLUSH, it needs to be reported upper layer
3281 * immediately as it means that @dev failed to remap and already
3282 * lost at least a sector and further FLUSH retrials won't make
3283 * any difference to the lost sector. However, if FLUSH failed
3284 * for other reasons, for example transmission error, FLUSH needs
3287 * This function determines whether FLUSH failure retry is
3288 * necessary and performs it if so.
3291 * 0 if EH can continue, -errno if EH needs to be repeated.
3293 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3295 struct ata_link *link = dev->link;
3296 struct ata_port *ap = link->ap;
3297 struct ata_queued_cmd *qc;
3298 struct ata_taskfile tf;
3299 unsigned int err_mask;
3302 /* did flush fail for this device? */
3303 if (!ata_tag_valid(link->active_tag))
3306 qc = __ata_qc_from_tag(ap, link->active_tag);
3307 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3308 qc->tf.command != ATA_CMD_FLUSH))
3311 /* if the device failed it, it should be reported to upper layers */
3312 if (qc->err_mask & AC_ERR_DEV)
3315 /* flush failed for some other reason, give it another shot */
3316 ata_tf_init(dev, &tf);
3318 tf.command = qc->tf.command;
3319 tf.flags |= ATA_TFLAG_DEVICE;
3320 tf.protocol = ATA_PROT_NODATA;
3322 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3323 tf.command, qc->err_mask);
3325 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3328 * FLUSH is complete but there's no way to
3329 * successfully complete a failed command from EH.
3330 * Making sure retry is allowed at least once and
3331 * retrying it should do the trick - whatever was in
3332 * the cache is already on the platter and this won't
3333 * cause infinite loop.
3335 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3337 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3341 /* if device failed it, report it to upper layers */
3342 if (err_mask & AC_ERR_DEV) {
3343 qc->err_mask |= AC_ERR_DEV;
3345 if (!ata_port_is_frozen(ap))
3353 * ata_eh_set_lpm - configure SATA interface power management
3354 * @link: link to configure power management
3355 * @policy: the link power management policy
3356 * @r_failed_dev: out parameter for failed device
3358 * Enable SATA Interface power management. This will enable
3359 * Device Interface Power Management (DIPM) for min_power and
3360 * medium_power_with_dipm policies, and then call driver specific
3361 * callbacks for enabling Host Initiated Power management.
3367 * 0 on success, -errno on failure.
3369 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3370 struct ata_device **r_failed_dev)
3372 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3373 struct ata_eh_context *ehc = &link->eh_context;
3374 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3375 enum ata_lpm_policy old_policy = link->lpm_policy;
3376 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3377 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3378 unsigned int err_mask;
3381 /* if the link or host doesn't do LPM, noop */
3382 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3383 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3387 * DIPM is enabled only for MIN_POWER as some devices
3388 * misbehave when the host NACKs transition to SLUMBER. Order
3389 * device and link configurations such that the host always
3390 * allows DIPM requests.
3392 ata_for_each_dev(dev, link, ENABLED) {
3393 bool hipm = ata_id_has_hipm(dev->id);
3394 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3396 /* find the first enabled and LPM enabled devices */
3400 if (!lpm_dev && (hipm || dipm))
3403 hints &= ~ATA_LPM_EMPTY;
3405 hints &= ~ATA_LPM_HIPM;
3407 /* disable DIPM before changing link config */
3408 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3409 err_mask = ata_dev_set_feature(dev,
3410 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3411 if (err_mask && err_mask != AC_ERR_DEV) {
3413 "failed to disable DIPM, Emask 0x%x\n",
3422 rc = ap->ops->set_lpm(link, policy, hints);
3423 if (!rc && ap->slave_link)
3424 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3426 rc = sata_pmp_set_lpm(link, policy, hints);
3429 * Attribute link config failure to the first (LPM) enabled
3430 * device on the link.
3433 if (rc == -EOPNOTSUPP) {
3434 link->flags |= ATA_LFLAG_NO_LPM;
3437 dev = lpm_dev ? lpm_dev : link_dev;
3442 * Low level driver acked the transition. Issue DIPM command
3443 * with the new policy set.
3445 link->lpm_policy = policy;
3446 if (ap && ap->slave_link)
3447 ap->slave_link->lpm_policy = policy;
3449 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3450 ata_for_each_dev(dev, link, ENABLED) {
3451 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3452 ata_id_has_dipm(dev->id)) {
3453 err_mask = ata_dev_set_feature(dev,
3454 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3455 if (err_mask && err_mask != AC_ERR_DEV) {
3457 "failed to enable DIPM, Emask 0x%x\n",
3465 link->last_lpm_change = jiffies;
3466 link->flags |= ATA_LFLAG_CHANGED;
3471 /* restore the old policy */
3472 link->lpm_policy = old_policy;
3473 if (ap && ap->slave_link)
3474 ap->slave_link->lpm_policy = old_policy;
3476 /* if no device or only one more chance is left, disable LPM */
3477 if (!dev || ehc->tries[dev->devno] <= 2) {
3478 ata_link_warn(link, "disabling LPM on the link\n");
3479 link->flags |= ATA_LFLAG_NO_LPM;
3482 *r_failed_dev = dev;
3486 int ata_link_nr_enabled(struct ata_link *link)
3488 struct ata_device *dev;
3491 ata_for_each_dev(dev, link, ENABLED)
3496 static int ata_link_nr_vacant(struct ata_link *link)
3498 struct ata_device *dev;
3501 ata_for_each_dev(dev, link, ALL)
3502 if (dev->class == ATA_DEV_UNKNOWN)
3507 static int ata_eh_skip_recovery(struct ata_link *link)
3509 struct ata_port *ap = link->ap;
3510 struct ata_eh_context *ehc = &link->eh_context;
3511 struct ata_device *dev;
3513 /* skip disabled links */
3514 if (link->flags & ATA_LFLAG_DISABLED)
3517 /* skip if explicitly requested */
3518 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3521 /* thaw frozen port and recover failed devices */
3522 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3525 /* reset at least once if reset is requested */
3526 if ((ehc->i.action & ATA_EH_RESET) &&
3527 !(ehc->i.flags & ATA_EHI_DID_RESET))
3530 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3531 ata_for_each_dev(dev, link, ALL) {
3532 if (dev->class == ATA_DEV_UNKNOWN &&
3533 ehc->classes[dev->devno] != ATA_DEV_NONE)
3540 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3542 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3543 u64 now = get_jiffies_64();
3544 int *trials = void_arg;
3546 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3547 (ent->timestamp < now - min(now, interval)))
3554 static int ata_eh_schedule_probe(struct ata_device *dev)
3556 struct ata_eh_context *ehc = &dev->link->eh_context;
3557 struct ata_link *link = ata_dev_phys_link(dev);
3560 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3561 (ehc->did_probe_mask & (1 << dev->devno)))
3564 ata_eh_detach_dev(dev);
3566 ehc->did_probe_mask |= (1 << dev->devno);
3567 ehc->i.action |= ATA_EH_RESET;
3568 ehc->saved_xfer_mode[dev->devno] = 0;
3569 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3571 /* the link maybe in a deep sleep, wake it up */
3572 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3573 if (ata_is_host_link(link))
3574 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3577 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3581 /* Record and count probe trials on the ering. The specific
3582 * error mask used is irrelevant. Because a successful device
3583 * detection clears the ering, this count accumulates only if
3584 * there are consecutive failed probes.
3586 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3587 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3588 * forced to 1.5Gbps.
3590 * This is to work around cases where failed link speed
3591 * negotiation results in device misdetection leading to
3592 * infinite DEVXCHG or PHRDY CHG events.
3594 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3595 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3597 if (trials > ATA_EH_PROBE_TRIALS)
3598 sata_down_spd_limit(link, 1);
3603 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3605 struct ata_eh_context *ehc = &dev->link->eh_context;
3607 /* -EAGAIN from EH routine indicates retry without prejudice.
3608 * The requester is responsible for ensuring forward progress.
3611 ehc->tries[dev->devno]--;
3615 /* device missing or wrong IDENTIFY data, schedule probing */
3616 ehc->i.probe_mask |= (1 << dev->devno);
3619 /* give it just one more chance */
3620 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3623 if (ehc->tries[dev->devno] == 1) {
3624 /* This is the last chance, better to slow
3625 * down than lose it.
3627 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3628 if (dev->pio_mode > XFER_PIO_0)
3629 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3633 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3634 /* disable device if it has used up all its chances */
3635 ata_dev_disable(dev);
3637 /* detach if offline */
3638 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3639 ata_eh_detach_dev(dev);
3641 /* schedule probe if necessary */
3642 if (ata_eh_schedule_probe(dev)) {
3643 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3644 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3645 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3650 ehc->i.action |= ATA_EH_RESET;
3656 * ata_eh_recover - recover host port after error
3657 * @ap: host port to recover
3658 * @prereset: prereset method (can be NULL)
3659 * @softreset: softreset method (can be NULL)
3660 * @hardreset: hardreset method (can be NULL)
3661 * @postreset: postreset method (can be NULL)
3662 * @r_failed_link: out parameter for failed link
3664 * This is the alpha and omega, eum and yang, heart and soul of
3665 * libata exception handling. On entry, actions required to
3666 * recover each link and hotplug requests are recorded in the
3667 * link's eh_context. This function executes all the operations
3668 * with appropriate retrials and fallbacks to resurrect failed
3669 * devices, detach goners and greet newcomers.
3672 * Kernel thread context (may sleep).
3675 * 0 on success, -errno on failure.
3677 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3678 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3679 ata_postreset_fn_t postreset,
3680 struct ata_link **r_failed_link)
3682 struct ata_link *link;
3683 struct ata_device *dev;
3685 unsigned long flags, deadline;
3687 /* prep for recovery */
3688 ata_for_each_link(link, ap, EDGE) {
3689 struct ata_eh_context *ehc = &link->eh_context;
3691 /* re-enable link? */
3692 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3693 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3694 spin_lock_irqsave(ap->lock, flags);
3695 link->flags &= ~ATA_LFLAG_DISABLED;
3696 spin_unlock_irqrestore(ap->lock, flags);
3697 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3700 ata_for_each_dev(dev, link, ALL) {
3701 if (link->flags & ATA_LFLAG_NO_RETRY)
3702 ehc->tries[dev->devno] = 1;
3704 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3706 /* collect port action mask recorded in dev actions */
3707 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3708 ~ATA_EH_PERDEV_MASK;
3709 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3711 /* process hotplug request */
3712 if (dev->flags & ATA_DFLAG_DETACH)
3713 ata_eh_detach_dev(dev);
3715 /* schedule probe if necessary */
3716 if (!ata_dev_enabled(dev))
3717 ata_eh_schedule_probe(dev);
3724 /* if UNLOADING, finish immediately */
3725 if (ap->pflags & ATA_PFLAG_UNLOADING)
3729 ata_for_each_link(link, ap, EDGE) {
3730 struct ata_eh_context *ehc = &link->eh_context;
3732 /* skip EH if possible. */
3733 if (ata_eh_skip_recovery(link))
3736 ata_for_each_dev(dev, link, ALL)
3737 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3741 ata_for_each_link(link, ap, EDGE) {
3742 struct ata_eh_context *ehc = &link->eh_context;
3744 if (!(ehc->i.action & ATA_EH_RESET))
3747 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3748 prereset, softreset, hardreset, postreset);
3750 ata_link_err(link, "reset failed, giving up\n");
3759 * clears ATA_EH_PARK in eh_info and resets
3760 * ap->park_req_pending
3762 ata_eh_pull_park_action(ap);
3765 ata_for_each_link(link, ap, EDGE) {
3766 ata_for_each_dev(dev, link, ALL) {
3767 struct ata_eh_context *ehc = &link->eh_context;
3770 if (dev->class != ATA_DEV_ATA &&
3771 dev->class != ATA_DEV_ZAC)
3773 if (!(ehc->i.dev_action[dev->devno] &
3776 tmp = dev->unpark_deadline;
3777 if (time_before(deadline, tmp))
3779 else if (time_before_eq(tmp, jiffies))
3781 if (ehc->unloaded_mask & (1 << dev->devno))
3784 ata_eh_park_issue_cmd(dev, 1);
3789 if (time_before_eq(deadline, now))
3793 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3797 ata_for_each_link(link, ap, EDGE) {
3798 ata_for_each_dev(dev, link, ALL) {
3799 if (!(link->eh_context.unloaded_mask &
3803 ata_eh_park_issue_cmd(dev, 0);
3804 ata_eh_done(link, dev, ATA_EH_PARK);
3810 ata_for_each_link(link, ap, PMP_FIRST) {
3811 struct ata_eh_context *ehc = &link->eh_context;
3813 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3816 /* revalidate existing devices and attach new ones */
3817 rc = ata_eh_revalidate_and_attach(link, &dev);
3821 /* if PMP got attached, return, pmp EH will take care of it */
3822 if (link->device->class == ATA_DEV_PMP) {
3827 /* configure transfer mode if necessary */
3828 if (ehc->i.flags & ATA_EHI_SETMODE) {
3829 rc = ata_set_mode(link, &dev);
3832 ehc->i.flags &= ~ATA_EHI_SETMODE;
3835 /* If reset has been issued, clear UA to avoid
3836 * disrupting the current users of the device.
3838 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3839 ata_for_each_dev(dev, link, ALL) {
3840 if (dev->class != ATA_DEV_ATAPI)
3842 rc = atapi_eh_clear_ua(dev);
3845 if (zpodd_dev_enabled(dev))
3846 zpodd_post_poweron(dev);
3851 * Make sure to transition devices to the active power mode
3852 * if needed (e.g. if we were scheduled on system resume).
3854 ata_for_each_dev(dev, link, ENABLED) {
3855 if (ehc->i.dev_action[dev->devno] & ATA_EH_SET_ACTIVE) {
3856 ata_dev_power_set_active(dev);
3857 ata_eh_done(link, dev, ATA_EH_SET_ACTIVE);
3861 /* retry flush if necessary */
3862 ata_for_each_dev(dev, link, ALL) {
3863 if (dev->class != ATA_DEV_ATA &&
3864 dev->class != ATA_DEV_ZAC)
3866 rc = ata_eh_maybe_retry_flush(dev);
3872 /* configure link power saving */
3873 if (link->lpm_policy != ap->target_lpm_policy) {
3874 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3879 /* this link is okay now */
3886 ata_eh_handle_dev_fail(dev, rc);
3888 if (ata_port_is_frozen(ap)) {
3889 /* PMP reset requires working host port.
3890 * Can't retry if it's frozen.
3892 if (sata_pmp_attached(ap))
3902 if (rc && r_failed_link)
3903 *r_failed_link = link;
3909 * ata_eh_finish - finish up EH
3910 * @ap: host port to finish EH for
3912 * Recovery is complete. Clean up EH states and retry or finish
3918 void ata_eh_finish(struct ata_port *ap)
3920 struct ata_queued_cmd *qc;
3923 /* retry or finish qcs */
3924 ata_qc_for_each_raw(ap, qc, tag) {
3925 if (!(qc->flags & ATA_QCFLAG_EH))
3929 /* FIXME: Once EH migration is complete,
3930 * generate sense data in this function,
3931 * considering both err_mask and tf.
3933 if (qc->flags & ATA_QCFLAG_RETRY) {
3935 * Since qc->err_mask is set, ata_eh_qc_retry()
3936 * will not increment scmd->allowed, so upper
3937 * layer will only retry the command if it has
3938 * not already been retried too many times.
3940 ata_eh_qc_retry(qc);
3942 ata_eh_qc_complete(qc);
3945 if (qc->flags & ATA_QCFLAG_SENSE_VALID ||
3946 qc->flags & ATA_QCFLAG_EH_SUCCESS_CMD) {
3947 ata_eh_qc_complete(qc);
3949 /* feed zero TF to sense generation */
3950 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3952 * Since qc->err_mask is not set,
3953 * ata_eh_qc_retry() will increment
3954 * scmd->allowed, so upper layer is guaranteed
3955 * to retry the command.
3957 ata_eh_qc_retry(qc);
3962 /* make sure nr_active_links is zero after EH */
3963 WARN_ON(ap->nr_active_links);
3964 ap->nr_active_links = 0;
3968 * ata_do_eh - do standard error handling
3969 * @ap: host port to handle error for
3971 * @prereset: prereset method (can be NULL)
3972 * @softreset: softreset method (can be NULL)
3973 * @hardreset: hardreset method (can be NULL)
3974 * @postreset: postreset method (can be NULL)
3976 * Perform standard error handling sequence.
3979 * Kernel thread context (may sleep).
3981 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3982 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3983 ata_postreset_fn_t postreset)
3985 struct ata_device *dev;
3991 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3994 ata_for_each_dev(dev, &ap->link, ALL)
3995 ata_dev_disable(dev);
4002 * ata_std_error_handler - standard error handler
4003 * @ap: host port to handle error for
4005 * Standard error handler
4008 * Kernel thread context (may sleep).
4010 void ata_std_error_handler(struct ata_port *ap)
4012 struct ata_port_operations *ops = ap->ops;
4013 ata_reset_fn_t hardreset = ops->hardreset;
4015 /* ignore built-in hardreset if SCR access is not available */
4016 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4019 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4021 EXPORT_SYMBOL_GPL(ata_std_error_handler);
4025 * ata_eh_handle_port_suspend - perform port suspend operation
4026 * @ap: port to suspend
4031 * Kernel thread context (may sleep).
4033 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4035 unsigned long flags;
4037 struct ata_device *dev;
4038 struct ata_link *link;
4040 /* are we suspending? */
4041 spin_lock_irqsave(ap->lock, flags);
4042 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4043 ap->pm_mesg.event & PM_EVENT_RESUME) {
4044 spin_unlock_irqrestore(ap->lock, flags);
4047 spin_unlock_irqrestore(ap->lock, flags);
4049 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4051 /* Set all devices attached to the port in standby mode */
4052 ata_for_each_link(link, ap, HOST_FIRST) {
4053 ata_for_each_dev(dev, link, ENABLED)
4054 ata_dev_power_set_standby(dev);
4058 * If we have a ZPODD attached, check its zero
4059 * power ready status before the port is frozen.
4060 * Only needed for runtime suspend.
4062 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4063 ata_for_each_dev(dev, &ap->link, ENABLED) {
4064 if (zpodd_dev_enabled(dev))
4065 zpodd_on_suspend(dev);
4070 ata_eh_freeze_port(ap);
4072 if (ap->ops->port_suspend)
4073 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4075 ata_acpi_set_state(ap, ap->pm_mesg);
4077 /* update the flags */
4078 spin_lock_irqsave(ap->lock, flags);
4080 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4082 ap->pflags |= ATA_PFLAG_SUSPENDED;
4083 else if (ata_port_is_frozen(ap))
4084 ata_port_schedule_eh(ap);
4086 spin_unlock_irqrestore(ap->lock, flags);
4092 * ata_eh_handle_port_resume - perform port resume operation
4093 * @ap: port to resume
4098 * Kernel thread context (may sleep).
4100 static void ata_eh_handle_port_resume(struct ata_port *ap)
4102 struct ata_link *link;
4103 struct ata_device *dev;
4104 unsigned long flags;
4106 /* are we resuming? */
4107 spin_lock_irqsave(ap->lock, flags);
4108 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4109 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4110 spin_unlock_irqrestore(ap->lock, flags);
4113 spin_unlock_irqrestore(ap->lock, flags);
4115 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4118 * Error timestamps are in jiffies which doesn't run while
4119 * suspended and PHY events during resume isn't too uncommon.
4120 * When the two are combined, it can lead to unnecessary speed
4121 * downs if the machine is suspended and resumed repeatedly.
4122 * Clear error history.
4124 ata_for_each_link(link, ap, HOST_FIRST)
4125 ata_for_each_dev(dev, link, ALL)
4126 ata_ering_clear(&dev->ering);
4128 ata_acpi_set_state(ap, ap->pm_mesg);
4130 if (ap->ops->port_resume)
4131 ap->ops->port_resume(ap);
4133 /* tell ACPI that we're resuming */
4134 ata_acpi_on_resume(ap);
4136 /* update the flags */
4137 spin_lock_irqsave(ap->lock, flags);
4138 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4139 ap->pflags |= ATA_PFLAG_RESUMING;
4140 spin_unlock_irqrestore(ap->lock, flags);
4142 #endif /* CONFIG_PM */