2 * libata-eh.c - libata error handling
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
47 #include <linux/libata.h>
49 #include <trace/events/libata.h>
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
60 ATA_EFLAG_IS_IO = (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
62 ATA_EFLAG_OLD_ER = (1 << 31),
64 /* error categories */
67 ATA_ECAT_TOUT_HSM = 2,
69 ATA_ECAT_DUBIOUS_NONE = 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
75 ATA_EH_CMD_DFL_TIMEOUT = 5000,
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN = 5000,
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
85 ATA_EH_PRERESET_TIMEOUT = 10000,
86 ATA_EH_FASTDRAIN_INTERVAL = 3000,
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS = 2,
95 /* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
102 static const unsigned long ata_eh_reset_timeouts[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX, /* > 1 min has elapsed, give up */
110 static const unsigned long ata_eh_identify_timeouts[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
117 static const unsigned long ata_eh_flush_timeouts[] = {
118 15000, /* be generous with flush */
120 30000, /* and even more generous */
124 static const unsigned long ata_eh_other_timeouts[] = {
125 5000, /* same rationale as identify timeout */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
131 struct ata_eh_cmd_timeout_ent {
133 const unsigned long *timeouts;
136 /* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
148 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
149 static const struct ata_eh_cmd_timeout_ent
150 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
151 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
152 .timeouts = ata_eh_identify_timeouts, },
153 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_SET_FEATURES),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
160 .timeouts = ata_eh_other_timeouts, },
161 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
162 .timeouts = ata_eh_flush_timeouts },
166 static void __ata_port_freeze(struct ata_port *ap);
168 static void ata_eh_handle_port_suspend(struct ata_port *ap);
169 static void ata_eh_handle_port_resume(struct ata_port *ap);
170 #else /* CONFIG_PM */
171 static void ata_eh_handle_port_suspend(struct ata_port *ap)
174 static void ata_eh_handle_port_resume(struct ata_port *ap)
176 #endif /* CONFIG_PM */
178 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
181 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
182 ATA_EH_DESC_LEN - ehi->desc_len,
187 * __ata_ehi_push_desc - push error description without adding separator
189 * @fmt: printf format string
191 * Format string according to @fmt and append it to @ehi->desc.
194 * spin_lock_irqsave(host lock)
196 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
201 __ata_ehi_pushv_desc(ehi, fmt, args);
206 * ata_ehi_push_desc - push error description with separator
208 * @fmt: printf format string
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
214 * spin_lock_irqsave(host lock)
216 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
221 __ata_ehi_push_desc(ehi, ", ");
224 __ata_ehi_pushv_desc(ehi, fmt, args);
229 * ata_ehi_clear_desc - clean error description
235 * spin_lock_irqsave(host lock)
237 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
256 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
287 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
292 unsigned long long start, len;
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
309 #endif /* CONFIG_PCI */
311 static int ata_lookup_timeout_table(u8 cmd)
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
331 * Determine timeout for internal command @cmd for @dev.
337 * Determined timeout.
339 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
346 return ATA_EH_CMD_DFL_TIMEOUT;
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
364 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
378 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
381 struct ata_ering_entry *ent;
386 ering->cursor %= ATA_ERING_SIZE;
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
394 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
403 int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
408 struct ata_ering_entry *ent;
412 ent = &ering->ring[idx];
415 rc = map_fn(ent, arg);
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
424 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
426 ent->eflags |= ATA_EFLAG_OLD_ER;
430 static void ata_ering_clear(struct ata_ering *ering)
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
435 static unsigned int ata_eh_dev_action(struct ata_device *dev)
437 struct ata_eh_context *ehc = &dev->link->eh_context;
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
442 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
445 struct ata_device *tdev;
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
479 void ata_eh_acquire(struct ata_port *ap)
481 mutex_lock(&ap->host->eh_mutex);
482 WARN_ON_ONCE(ap->host->eh_owner);
483 ap->host->eh_owner = current;
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
496 void ata_eh_release(struct ata_port *ap)
498 WARN_ON_ONCE(ap->host->eh_owner != current);
499 ap->host->eh_owner = NULL;
500 mutex_unlock(&ap->host->eh_mutex);
504 * ata_scsi_timed_out - SCSI layer time out callback
505 * @cmd: timed out SCSI command
507 * Handles SCSI layer timeout. We race with normal completion of
508 * the qc for @cmd. If the qc is already gone, we lose and let
509 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
510 * timed out and EH should be invoked. Prevent ata_qc_complete()
511 * from finishing it by setting EH_SCHEDULED and return
514 * TODO: kill this function once old EH is gone.
517 * Called from timer context
520 * EH_HANDLED or EH_NOT_HANDLED
522 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
524 struct Scsi_Host *host = cmd->device->host;
525 struct ata_port *ap = ata_shost_to_port(host);
527 struct ata_queued_cmd *qc;
528 enum blk_eh_timer_return ret;
532 if (ap->ops->error_handler) {
533 ret = BLK_EH_NOT_HANDLED;
537 ret = BLK_EH_HANDLED;
538 spin_lock_irqsave(ap->lock, flags);
539 qc = ata_qc_from_tag(ap, ap->link.active_tag);
541 WARN_ON(qc->scsicmd != cmd);
542 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
543 qc->err_mask |= AC_ERR_TIMEOUT;
544 ret = BLK_EH_NOT_HANDLED;
546 spin_unlock_irqrestore(ap->lock, flags);
549 DPRINTK("EXIT, ret=%d\n", ret);
553 static void ata_eh_unload(struct ata_port *ap)
555 struct ata_link *link;
556 struct ata_device *dev;
559 /* Restore SControl IPM and SPD for the next driver and
560 * disable attached devices.
562 ata_for_each_link(link, ap, PMP_FIRST) {
563 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
564 ata_for_each_dev(dev, link, ALL)
565 ata_dev_disable(dev);
568 /* freeze and set UNLOADED */
569 spin_lock_irqsave(ap->lock, flags);
571 ata_port_freeze(ap); /* won't be thawed */
572 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
573 ap->pflags |= ATA_PFLAG_UNLOADED;
575 spin_unlock_irqrestore(ap->lock, flags);
579 * ata_scsi_error - SCSI layer error handler callback
580 * @host: SCSI host on which error occurred
582 * Handles SCSI-layer-thrown error events.
585 * Inherited from SCSI layer (none, can sleep)
590 void ata_scsi_error(struct Scsi_Host *host)
592 struct ata_port *ap = ata_shost_to_port(host);
594 LIST_HEAD(eh_work_q);
598 spin_lock_irqsave(host->host_lock, flags);
599 list_splice_init(&host->eh_cmd_q, &eh_work_q);
600 spin_unlock_irqrestore(host->host_lock, flags);
602 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
604 /* If we timed raced normal completion and there is nothing to
605 recover nr_timedout == 0 why exactly are we doing error recovery ? */
606 ata_scsi_port_error_handler(host, ap);
608 /* finish or retry handled scmd's and clean up */
609 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
615 * ata_scsi_cmd_error_handler - error callback for a list of commands
616 * @host: scsi host containing the port
617 * @ap: ATA port within the host
618 * @eh_work_q: list of commands to process
620 * process the given list of commands and return those finished to the
621 * ap->eh_done_q. This function is the first part of the libata error
622 * handler which processes a given list of failed commands.
624 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
625 struct list_head *eh_work_q)
630 /* make sure sff pio task is not running */
631 ata_sff_flush_pio_task(ap);
633 /* synchronize with host lock and sort out timeouts */
635 /* For new EH, all qcs are finished in one of three ways -
636 * normal completion, error completion, and SCSI timeout.
637 * Both completions can race against SCSI timeout. When normal
638 * completion wins, the qc never reaches EH. When error
639 * completion wins, the qc has ATA_QCFLAG_FAILED set.
641 * When SCSI timeout wins, things are a bit more complex.
642 * Normal or error completion can occur after the timeout but
643 * before this point. In such cases, both types of
644 * completions are honored. A scmd is determined to have
645 * timed out iff its associated qc is active and not failed.
647 if (ap->ops->error_handler) {
648 struct scsi_cmnd *scmd, *tmp;
651 spin_lock_irqsave(ap->lock, flags);
653 /* This must occur under the ap->lock as we don't want
654 a polled recovery to race the real interrupt handler
656 The lost_interrupt handler checks for any completed but
657 non-notified command and completes much like an IRQ handler.
659 We then fall into the error recovery code which will treat
660 this as if normal completion won the race */
662 if (ap->ops->lost_interrupt)
663 ap->ops->lost_interrupt(ap);
665 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
666 struct ata_queued_cmd *qc;
668 for (i = 0; i < ATA_MAX_QUEUE; i++) {
669 qc = __ata_qc_from_tag(ap, i);
670 if (qc->flags & ATA_QCFLAG_ACTIVE &&
675 if (i < ATA_MAX_QUEUE) {
676 /* the scmd has an associated qc */
677 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
678 /* which hasn't failed yet, timeout */
679 qc->err_mask |= AC_ERR_TIMEOUT;
680 qc->flags |= ATA_QCFLAG_FAILED;
684 /* Normal completion occurred after
685 * SCSI timeout but before this point.
686 * Successfully complete it.
688 scmd->retries = scmd->allowed;
689 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
693 /* If we have timed out qcs. They belong to EH from
694 * this point but the state of the controller is
695 * unknown. Freeze the port to make sure the IRQ
696 * handler doesn't diddle with those qcs. This must
697 * be done atomically w.r.t. setting QCFLAG_FAILED.
700 __ata_port_freeze(ap);
702 spin_unlock_irqrestore(ap->lock, flags);
704 /* initialize eh_tries */
705 ap->eh_tries = ATA_EH_MAX_TRIES;
707 spin_unlock_wait(ap->lock);
710 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
713 * ata_scsi_port_error_handler - recover the port after the commands
714 * @host: SCSI host containing the port
717 * Handle the recovery of the port @ap after all the commands
718 * have been recovered.
720 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
724 /* invoke error handler */
725 if (ap->ops->error_handler) {
726 struct ata_link *link;
728 /* acquire EH ownership */
731 /* kill fast drain timer */
732 del_timer_sync(&ap->fastdrain_timer);
734 /* process port resume request */
735 ata_eh_handle_port_resume(ap);
737 /* fetch & clear EH info */
738 spin_lock_irqsave(ap->lock, flags);
740 ata_for_each_link(link, ap, HOST_FIRST) {
741 struct ata_eh_context *ehc = &link->eh_context;
742 struct ata_device *dev;
744 memset(&link->eh_context, 0, sizeof(link->eh_context));
745 link->eh_context.i = link->eh_info;
746 memset(&link->eh_info, 0, sizeof(link->eh_info));
748 ata_for_each_dev(dev, link, ENABLED) {
749 int devno = dev->devno;
751 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
752 if (ata_ncq_enabled(dev))
753 ehc->saved_ncq_enabled |= 1 << devno;
757 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
758 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
759 ap->excl_link = NULL; /* don't maintain exclusion over EH */
761 spin_unlock_irqrestore(ap->lock, flags);
763 /* invoke EH, skip if unloading or suspended */
764 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
765 ap->ops->error_handler(ap);
767 /* if unloading, commence suicide */
768 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
769 !(ap->pflags & ATA_PFLAG_UNLOADED))
774 /* process port suspend request */
775 ata_eh_handle_port_suspend(ap);
777 /* Exception might have happened after ->error_handler
778 * recovered the port but before this point. Repeat
781 spin_lock_irqsave(ap->lock, flags);
783 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
784 if (--ap->eh_tries) {
785 spin_unlock_irqrestore(ap->lock, flags);
789 "EH pending after %d tries, giving up\n",
791 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
794 /* this run is complete, make sure EH info is clear */
795 ata_for_each_link(link, ap, HOST_FIRST)
796 memset(&link->eh_info, 0, sizeof(link->eh_info));
798 /* end eh (clear host_eh_scheduled) while holding
799 * ap->lock such that if exception occurs after this
800 * point but before EH completion, SCSI midlayer will
805 spin_unlock_irqrestore(ap->lock, flags);
808 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
809 ap->ops->eng_timeout(ap);
812 scsi_eh_flush_done_q(&ap->eh_done_q);
815 spin_lock_irqsave(ap->lock, flags);
817 if (ap->pflags & ATA_PFLAG_LOADING)
818 ap->pflags &= ~ATA_PFLAG_LOADING;
819 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
820 schedule_delayed_work(&ap->hotplug_task, 0);
822 if (ap->pflags & ATA_PFLAG_RECOVERED)
823 ata_port_info(ap, "EH complete\n");
825 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
827 /* tell wait_eh that we're done */
828 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
829 wake_up_all(&ap->eh_wait_q);
831 spin_unlock_irqrestore(ap->lock, flags);
833 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
836 * ata_port_wait_eh - Wait for the currently pending EH to complete
837 * @ap: Port to wait EH for
839 * Wait until the currently pending EH is complete.
842 * Kernel thread context (may sleep).
844 void ata_port_wait_eh(struct ata_port *ap)
850 spin_lock_irqsave(ap->lock, flags);
852 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
853 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
854 spin_unlock_irqrestore(ap->lock, flags);
856 spin_lock_irqsave(ap->lock, flags);
858 finish_wait(&ap->eh_wait_q, &wait);
860 spin_unlock_irqrestore(ap->lock, flags);
862 /* make sure SCSI EH is complete */
863 if (scsi_host_in_recovery(ap->scsi_host)) {
868 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
870 static int ata_eh_nr_in_flight(struct ata_port *ap)
875 /* count only non-internal commands */
876 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
877 if (ata_qc_from_tag(ap, tag))
883 void ata_eh_fastdrain_timerfn(unsigned long arg)
885 struct ata_port *ap = (void *)arg;
889 spin_lock_irqsave(ap->lock, flags);
891 cnt = ata_eh_nr_in_flight(ap);
897 if (cnt == ap->fastdrain_cnt) {
900 /* No progress during the last interval, tag all
901 * in-flight qcs as timed out and freeze the port.
903 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
904 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
906 qc->err_mask |= AC_ERR_TIMEOUT;
911 /* some qcs have finished, give it another chance */
912 ap->fastdrain_cnt = cnt;
913 ap->fastdrain_timer.expires =
914 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
915 add_timer(&ap->fastdrain_timer);
919 spin_unlock_irqrestore(ap->lock, flags);
923 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
924 * @ap: target ATA port
925 * @fastdrain: activate fast drain
927 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
928 * is non-zero and EH wasn't pending before. Fast drain ensures
929 * that EH kicks in in timely manner.
932 * spin_lock_irqsave(host lock)
934 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
938 /* already scheduled? */
939 if (ap->pflags & ATA_PFLAG_EH_PENDING)
942 ap->pflags |= ATA_PFLAG_EH_PENDING;
947 /* do we have in-flight qcs? */
948 cnt = ata_eh_nr_in_flight(ap);
952 /* activate fast drain */
953 ap->fastdrain_cnt = cnt;
954 ap->fastdrain_timer.expires =
955 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
956 add_timer(&ap->fastdrain_timer);
960 * ata_qc_schedule_eh - schedule qc for error handling
961 * @qc: command to schedule error handling for
963 * Schedule error handling for @qc. EH will kick in as soon as
964 * other commands are drained.
967 * spin_lock_irqsave(host lock)
969 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
971 struct ata_port *ap = qc->ap;
972 struct request_queue *q = qc->scsicmd->device->request_queue;
975 WARN_ON(!ap->ops->error_handler);
977 qc->flags |= ATA_QCFLAG_FAILED;
978 ata_eh_set_pending(ap, 1);
980 /* The following will fail if timeout has already expired.
981 * ata_scsi_error() takes care of such scmds on EH entry.
982 * Note that ATA_QCFLAG_FAILED is unconditionally set after
983 * this function completes.
985 spin_lock_irqsave(q->queue_lock, flags);
986 blk_abort_request(qc->scsicmd->request);
987 spin_unlock_irqrestore(q->queue_lock, flags);
991 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
992 * @ap: ATA port to schedule EH for
994 * LOCKING: inherited from ata_port_schedule_eh
995 * spin_lock_irqsave(host lock)
997 void ata_std_sched_eh(struct ata_port *ap)
999 WARN_ON(!ap->ops->error_handler);
1001 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1004 ata_eh_set_pending(ap, 1);
1005 scsi_schedule_eh(ap->scsi_host);
1007 DPRINTK("port EH scheduled\n");
1009 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1012 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1013 * @ap: ATA port to end EH for
1015 * In the libata object model there is a 1:1 mapping of ata_port to
1016 * shost, so host fields can be directly manipulated under ap->lock, in
1017 * the libsas case we need to hold a lock at the ha->level to coordinate
1021 * spin_lock_irqsave(host lock)
1023 void ata_std_end_eh(struct ata_port *ap)
1025 struct Scsi_Host *host = ap->scsi_host;
1027 host->host_eh_scheduled = 0;
1029 EXPORT_SYMBOL(ata_std_end_eh);
1033 * ata_port_schedule_eh - schedule error handling without a qc
1034 * @ap: ATA port to schedule EH for
1036 * Schedule error handling for @ap. EH will kick in as soon as
1037 * all commands are drained.
1040 * spin_lock_irqsave(host lock)
1042 void ata_port_schedule_eh(struct ata_port *ap)
1044 /* see: ata_std_sched_eh, unless you know better */
1045 ap->ops->sched_eh(ap);
1048 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1050 int tag, nr_aborted = 0;
1052 WARN_ON(!ap->ops->error_handler);
1054 /* we're gonna abort all commands, no need for fast drain */
1055 ata_eh_set_pending(ap, 0);
1057 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1058 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1060 if (qc && (!link || qc->dev->link == link)) {
1061 qc->flags |= ATA_QCFLAG_FAILED;
1062 ata_qc_complete(qc);
1068 ata_port_schedule_eh(ap);
1074 * ata_link_abort - abort all qc's on the link
1075 * @link: ATA link to abort qc's for
1077 * Abort all active qc's active on @link and schedule EH.
1080 * spin_lock_irqsave(host lock)
1083 * Number of aborted qc's.
1085 int ata_link_abort(struct ata_link *link)
1087 return ata_do_link_abort(link->ap, link);
1091 * ata_port_abort - abort all qc's on the port
1092 * @ap: ATA port to abort qc's for
1094 * Abort all active qc's of @ap and schedule EH.
1097 * spin_lock_irqsave(host_set lock)
1100 * Number of aborted qc's.
1102 int ata_port_abort(struct ata_port *ap)
1104 return ata_do_link_abort(ap, NULL);
1108 * __ata_port_freeze - freeze port
1109 * @ap: ATA port to freeze
1111 * This function is called when HSM violation or some other
1112 * condition disrupts normal operation of the port. Frozen port
1113 * is not allowed to perform any operation until the port is
1114 * thawed, which usually follows a successful reset.
1116 * ap->ops->freeze() callback can be used for freezing the port
1117 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1118 * port cannot be frozen hardware-wise, the interrupt handler
1119 * must ack and clear interrupts unconditionally while the port
1123 * spin_lock_irqsave(host lock)
1125 static void __ata_port_freeze(struct ata_port *ap)
1127 WARN_ON(!ap->ops->error_handler);
1129 if (ap->ops->freeze)
1130 ap->ops->freeze(ap);
1132 ap->pflags |= ATA_PFLAG_FROZEN;
1134 DPRINTK("ata%u port frozen\n", ap->print_id);
1138 * ata_port_freeze - abort & freeze port
1139 * @ap: ATA port to freeze
1141 * Abort and freeze @ap. The freeze operation must be called
1142 * first, because some hardware requires special operations
1143 * before the taskfile registers are accessible.
1146 * spin_lock_irqsave(host lock)
1149 * Number of aborted commands.
1151 int ata_port_freeze(struct ata_port *ap)
1155 WARN_ON(!ap->ops->error_handler);
1157 __ata_port_freeze(ap);
1158 nr_aborted = ata_port_abort(ap);
1164 * sata_async_notification - SATA async notification handler
1165 * @ap: ATA port where async notification is received
1167 * Handler to be called when async notification via SDB FIS is
1168 * received. This function schedules EH if necessary.
1171 * spin_lock_irqsave(host lock)
1174 * 1 if EH is scheduled, 0 otherwise.
1176 int sata_async_notification(struct ata_port *ap)
1181 if (!(ap->flags & ATA_FLAG_AN))
1184 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1186 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1188 if (!sata_pmp_attached(ap) || rc) {
1189 /* PMP is not attached or SNTF is not available */
1190 if (!sata_pmp_attached(ap)) {
1191 /* PMP is not attached. Check whether ATAPI
1192 * AN is configured. If so, notify media
1195 struct ata_device *dev = ap->link.device;
1197 if ((dev->class == ATA_DEV_ATAPI) &&
1198 (dev->flags & ATA_DFLAG_AN))
1199 ata_scsi_media_change_notify(dev);
1202 /* PMP is attached but SNTF is not available.
1203 * ATAPI async media change notification is
1204 * not used. The PMP must be reporting PHY
1205 * status change, schedule EH.
1207 ata_port_schedule_eh(ap);
1211 /* PMP is attached and SNTF is available */
1212 struct ata_link *link;
1214 /* check and notify ATAPI AN */
1215 ata_for_each_link(link, ap, EDGE) {
1216 if (!(sntf & (1 << link->pmp)))
1219 if ((link->device->class == ATA_DEV_ATAPI) &&
1220 (link->device->flags & ATA_DFLAG_AN))
1221 ata_scsi_media_change_notify(link->device);
1224 /* If PMP is reporting that PHY status of some
1225 * downstream ports has changed, schedule EH.
1227 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1228 ata_port_schedule_eh(ap);
1237 * ata_eh_freeze_port - EH helper to freeze port
1238 * @ap: ATA port to freeze
1245 void ata_eh_freeze_port(struct ata_port *ap)
1247 unsigned long flags;
1249 if (!ap->ops->error_handler)
1252 spin_lock_irqsave(ap->lock, flags);
1253 __ata_port_freeze(ap);
1254 spin_unlock_irqrestore(ap->lock, flags);
1258 * ata_port_thaw_port - EH helper to thaw port
1259 * @ap: ATA port to thaw
1261 * Thaw frozen port @ap.
1266 void ata_eh_thaw_port(struct ata_port *ap)
1268 unsigned long flags;
1270 if (!ap->ops->error_handler)
1273 spin_lock_irqsave(ap->lock, flags);
1275 ap->pflags &= ~ATA_PFLAG_FROZEN;
1280 spin_unlock_irqrestore(ap->lock, flags);
1282 DPRINTK("ata%u port thawed\n", ap->print_id);
1285 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1290 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1292 struct ata_port *ap = qc->ap;
1293 struct scsi_cmnd *scmd = qc->scsicmd;
1294 unsigned long flags;
1296 spin_lock_irqsave(ap->lock, flags);
1297 qc->scsidone = ata_eh_scsidone;
1298 __ata_qc_complete(qc);
1299 WARN_ON(ata_tag_valid(qc->tag));
1300 spin_unlock_irqrestore(ap->lock, flags);
1302 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1306 * ata_eh_qc_complete - Complete an active ATA command from EH
1307 * @qc: Command to complete
1309 * Indicate to the mid and upper layers that an ATA command has
1310 * completed. To be used from EH.
1312 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1314 struct scsi_cmnd *scmd = qc->scsicmd;
1315 scmd->retries = scmd->allowed;
1316 __ata_eh_qc_complete(qc);
1320 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1321 * @qc: Command to retry
1323 * Indicate to the mid and upper layers that an ATA command
1324 * should be retried. To be used from EH.
1326 * SCSI midlayer limits the number of retries to scmd->allowed.
1327 * scmd->allowed is incremented for commands which get retried
1328 * due to unrelated failures (qc->err_mask is zero).
1330 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1332 struct scsi_cmnd *scmd = qc->scsicmd;
1335 __ata_eh_qc_complete(qc);
1339 * ata_dev_disable - disable ATA device
1340 * @dev: ATA device to disable
1347 void ata_dev_disable(struct ata_device *dev)
1349 if (!ata_dev_enabled(dev))
1352 if (ata_msg_drv(dev->link->ap))
1353 ata_dev_warn(dev, "disabled\n");
1354 ata_acpi_on_disable(dev);
1355 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1358 /* From now till the next successful probe, ering is used to
1359 * track probe failures. Clear accumulated device error info.
1361 ata_ering_clear(&dev->ering);
1365 * ata_eh_detach_dev - detach ATA device
1366 * @dev: ATA device to detach
1373 void ata_eh_detach_dev(struct ata_device *dev)
1375 struct ata_link *link = dev->link;
1376 struct ata_port *ap = link->ap;
1377 struct ata_eh_context *ehc = &link->eh_context;
1378 unsigned long flags;
1380 ata_dev_disable(dev);
1382 spin_lock_irqsave(ap->lock, flags);
1384 dev->flags &= ~ATA_DFLAG_DETACH;
1386 if (ata_scsi_offline_dev(dev)) {
1387 dev->flags |= ATA_DFLAG_DETACHED;
1388 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1391 /* clear per-dev EH info */
1392 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1393 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1394 ehc->saved_xfer_mode[dev->devno] = 0;
1395 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1397 spin_unlock_irqrestore(ap->lock, flags);
1401 * ata_eh_about_to_do - about to perform eh_action
1402 * @link: target ATA link
1403 * @dev: target ATA dev for per-dev action (can be NULL)
1404 * @action: action about to be performed
1406 * Called just before performing EH actions to clear related bits
1407 * in @link->eh_info such that eh actions are not unnecessarily
1413 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1414 unsigned int action)
1416 struct ata_port *ap = link->ap;
1417 struct ata_eh_info *ehi = &link->eh_info;
1418 struct ata_eh_context *ehc = &link->eh_context;
1419 unsigned long flags;
1421 spin_lock_irqsave(ap->lock, flags);
1423 ata_eh_clear_action(link, dev, ehi, action);
1425 /* About to take EH action, set RECOVERED. Ignore actions on
1426 * slave links as master will do them again.
1428 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1429 ap->pflags |= ATA_PFLAG_RECOVERED;
1431 spin_unlock_irqrestore(ap->lock, flags);
1435 * ata_eh_done - EH action complete
1436 * @ap: target ATA port
1437 * @dev: target ATA dev for per-dev action (can be NULL)
1438 * @action: action just completed
1440 * Called right after performing EH actions to clear related bits
1441 * in @link->eh_context.
1446 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1447 unsigned int action)
1449 struct ata_eh_context *ehc = &link->eh_context;
1451 ata_eh_clear_action(link, dev, &ehc->i, action);
1455 * ata_err_string - convert err_mask to descriptive string
1456 * @err_mask: error mask to convert to string
1458 * Convert @err_mask to descriptive string. Errors are
1459 * prioritized according to severity and only the most severe
1460 * error is reported.
1466 * Descriptive string for @err_mask
1468 static const char *ata_err_string(unsigned int err_mask)
1470 if (err_mask & AC_ERR_HOST_BUS)
1471 return "host bus error";
1472 if (err_mask & AC_ERR_ATA_BUS)
1473 return "ATA bus error";
1474 if (err_mask & AC_ERR_TIMEOUT)
1476 if (err_mask & AC_ERR_HSM)
1477 return "HSM violation";
1478 if (err_mask & AC_ERR_SYSTEM)
1479 return "internal error";
1480 if (err_mask & AC_ERR_MEDIA)
1481 return "media error";
1482 if (err_mask & AC_ERR_INVALID)
1483 return "invalid argument";
1484 if (err_mask & AC_ERR_DEV)
1485 return "device error";
1486 return "unknown error";
1490 * ata_read_log_page - read a specific log page
1491 * @dev: target device
1493 * @page: page to read
1494 * @buf: buffer to store read page
1495 * @sectors: number of sectors to read
1497 * Read log page using READ_LOG_EXT command.
1500 * Kernel thread context (may sleep).
1503 * 0 on success, AC_ERR_* mask otherwise.
1505 unsigned int ata_read_log_page(struct ata_device *dev, u8 log,
1506 u8 page, void *buf, unsigned int sectors)
1508 struct ata_taskfile tf;
1509 unsigned int err_mask;
1512 DPRINTK("read log page - log 0x%x, page 0x%x\n", log, page);
1515 ata_tf_init(dev, &tf);
1516 if (dev->dma_mode && ata_id_has_read_log_dma_ext(dev->id) &&
1517 !(dev->horkage & ATA_HORKAGE_NO_NCQ_LOG)) {
1518 tf.command = ATA_CMD_READ_LOG_DMA_EXT;
1519 tf.protocol = ATA_PROT_DMA;
1522 tf.command = ATA_CMD_READ_LOG_EXT;
1523 tf.protocol = ATA_PROT_PIO;
1529 tf.hob_nsect = sectors >> 8;
1530 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1532 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1533 buf, sectors * ATA_SECT_SIZE, 0);
1535 if (err_mask && dma) {
1536 dev->horkage |= ATA_HORKAGE_NO_NCQ_LOG;
1537 ata_dev_warn(dev, "READ LOG DMA EXT failed, trying unqueued\n");
1541 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1546 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1547 * @dev: Device to read log page 10h from
1548 * @tag: Resulting tag of the failed command
1549 * @tf: Resulting taskfile registers of the failed command
1551 * Read log page 10h to obtain NCQ error details and clear error
1555 * Kernel thread context (may sleep).
1558 * 0 on success, -errno otherwise.
1560 static int ata_eh_read_log_10h(struct ata_device *dev,
1561 int *tag, struct ata_taskfile *tf)
1563 u8 *buf = dev->link->ap->sector_buf;
1564 unsigned int err_mask;
1568 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1573 for (i = 0; i < ATA_SECT_SIZE; i++)
1576 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1582 *tag = buf[0] & 0x1f;
1584 tf->command = buf[2];
1585 tf->feature = buf[3];
1589 tf->device = buf[7];
1590 tf->hob_lbal = buf[8];
1591 tf->hob_lbam = buf[9];
1592 tf->hob_lbah = buf[10];
1593 tf->nsect = buf[12];
1594 tf->hob_nsect = buf[13];
1595 if (ata_id_has_ncq_autosense(dev->id))
1596 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1602 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1603 * @dev: target ATAPI device
1604 * @r_sense_key: out parameter for sense_key
1606 * Perform ATAPI TEST_UNIT_READY.
1609 * EH context (may sleep).
1612 * 0 on success, AC_ERR_* mask on failure.
1614 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1616 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1617 struct ata_taskfile tf;
1618 unsigned int err_mask;
1620 ata_tf_init(dev, &tf);
1622 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1623 tf.command = ATA_CMD_PACKET;
1624 tf.protocol = ATAPI_PROT_NODATA;
1626 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1627 if (err_mask == AC_ERR_DEV)
1628 *r_sense_key = tf.feature >> 4;
1633 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1634 * @dev: device to perform REQUEST_SENSE to
1635 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1636 * @dfl_sense_key: default sense key to use
1638 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1639 * SENSE. This function is EH helper.
1642 * Kernel thread context (may sleep).
1645 * 0 on success, AC_ERR_* mask on failure
1647 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1648 u8 *sense_buf, u8 dfl_sense_key)
1650 u8 cdb[ATAPI_CDB_LEN] =
1651 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1652 struct ata_port *ap = dev->link->ap;
1653 struct ata_taskfile tf;
1655 DPRINTK("ATAPI request sense\n");
1657 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1659 /* initialize sense_buf with the error register,
1660 * for the case where they are -not- overwritten
1662 sense_buf[0] = 0x70;
1663 sense_buf[2] = dfl_sense_key;
1665 /* some devices time out if garbage left in tf */
1666 ata_tf_init(dev, &tf);
1668 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1669 tf.command = ATA_CMD_PACKET;
1671 /* is it pointless to prefer PIO for "safety reasons"? */
1672 if (ap->flags & ATA_FLAG_PIO_DMA) {
1673 tf.protocol = ATAPI_PROT_DMA;
1674 tf.feature |= ATAPI_PKT_DMA;
1676 tf.protocol = ATAPI_PROT_PIO;
1677 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1681 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1682 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1686 * ata_eh_analyze_serror - analyze SError for a failed port
1687 * @link: ATA link to analyze SError for
1689 * Analyze SError if available and further determine cause of
1695 static void ata_eh_analyze_serror(struct ata_link *link)
1697 struct ata_eh_context *ehc = &link->eh_context;
1698 u32 serror = ehc->i.serror;
1699 unsigned int err_mask = 0, action = 0;
1702 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1703 err_mask |= AC_ERR_ATA_BUS;
1704 action |= ATA_EH_RESET;
1706 if (serror & SERR_PROTOCOL) {
1707 err_mask |= AC_ERR_HSM;
1708 action |= ATA_EH_RESET;
1710 if (serror & SERR_INTERNAL) {
1711 err_mask |= AC_ERR_SYSTEM;
1712 action |= ATA_EH_RESET;
1715 /* Determine whether a hotplug event has occurred. Both
1716 * SError.N/X are considered hotplug events for enabled or
1717 * host links. For disabled PMP links, only N bit is
1718 * considered as X bit is left at 1 for link plugging.
1720 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1721 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1722 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1723 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1725 hotplug_mask = SERR_PHYRDY_CHG;
1727 if (serror & hotplug_mask)
1728 ata_ehi_hotplugged(&ehc->i);
1730 ehc->i.err_mask |= err_mask;
1731 ehc->i.action |= action;
1735 * ata_eh_analyze_ncq_error - analyze NCQ error
1736 * @link: ATA link to analyze NCQ error for
1738 * Read log page 10h, determine the offending qc and acquire
1739 * error status TF. For NCQ device errors, all LLDDs have to do
1740 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1744 * Kernel thread context (may sleep).
1746 void ata_eh_analyze_ncq_error(struct ata_link *link)
1748 struct ata_port *ap = link->ap;
1749 struct ata_eh_context *ehc = &link->eh_context;
1750 struct ata_device *dev = link->device;
1751 struct ata_queued_cmd *qc;
1752 struct ata_taskfile tf;
1755 /* if frozen, we can't do much */
1756 if (ap->pflags & ATA_PFLAG_FROZEN)
1759 /* is it NCQ device error? */
1760 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1763 /* has LLDD analyzed already? */
1764 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1765 qc = __ata_qc_from_tag(ap, tag);
1767 if (!(qc->flags & ATA_QCFLAG_FAILED))
1774 /* okay, this error is ours */
1775 memset(&tf, 0, sizeof(tf));
1776 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1778 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1783 if (!(link->sactive & (1 << tag))) {
1784 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1789 /* we've got the perpetrator, condemn it */
1790 qc = __ata_qc_from_tag(ap, tag);
1791 memcpy(&qc->result_tf, &tf, sizeof(tf));
1792 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1793 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1794 if (qc->result_tf.auxiliary) {
1795 char sense_key, asc, ascq;
1797 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1798 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1799 ascq = qc->result_tf.auxiliary & 0xff;
1800 ata_dev_dbg(dev, "NCQ Autosense %02x/%02x/%02x\n",
1801 sense_key, asc, ascq);
1802 ata_scsi_set_sense(qc->scsicmd, sense_key, asc, ascq);
1803 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1806 ehc->i.err_mask &= ~AC_ERR_DEV;
1810 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1811 * @qc: qc to analyze
1812 * @tf: Taskfile registers to analyze
1814 * Analyze taskfile of @qc and further determine cause of
1815 * failure. This function also requests ATAPI sense data if
1819 * Kernel thread context (may sleep).
1822 * Determined recovery action
1824 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1825 const struct ata_taskfile *tf)
1827 unsigned int tmp, action = 0;
1828 u8 stat = tf->command, err = tf->feature;
1830 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1831 qc->err_mask |= AC_ERR_HSM;
1832 return ATA_EH_RESET;
1835 /* Set by NCQ autosense */
1836 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1839 if (stat & (ATA_ERR | ATA_DF))
1840 qc->err_mask |= AC_ERR_DEV;
1844 switch (qc->dev->class) {
1848 qc->err_mask |= AC_ERR_ATA_BUS;
1849 if (err & (ATA_UNC | ATA_AMNF))
1850 qc->err_mask |= AC_ERR_MEDIA;
1852 qc->err_mask |= AC_ERR_INVALID;
1856 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1857 tmp = atapi_eh_request_sense(qc->dev,
1858 qc->scsicmd->sense_buffer,
1859 qc->result_tf.feature >> 4);
1861 /* ATA_QCFLAG_SENSE_VALID is used to
1862 * tell atapi_qc_complete() that sense
1863 * data is already valid.
1865 * TODO: interpret sense data and set
1866 * appropriate err_mask.
1868 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1870 qc->err_mask |= tmp;
1874 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1875 action |= ATA_EH_RESET;
1880 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1885 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1889 base = ATA_ECAT_DUBIOUS_NONE;
1891 if (err_mask & AC_ERR_ATA_BUS)
1892 return base + ATA_ECAT_ATA_BUS;
1894 if (err_mask & AC_ERR_TIMEOUT)
1895 return base + ATA_ECAT_TOUT_HSM;
1897 if (eflags & ATA_EFLAG_IS_IO) {
1898 if (err_mask & AC_ERR_HSM)
1899 return base + ATA_ECAT_TOUT_HSM;
1901 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1902 return base + ATA_ECAT_UNK_DEV;
1908 struct speed_down_verdict_arg {
1911 int nr_errors[ATA_ECAT_NR];
1914 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1916 struct speed_down_verdict_arg *arg = void_arg;
1919 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1922 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1924 arg->nr_errors[cat]++;
1930 * ata_eh_speed_down_verdict - Determine speed down verdict
1931 * @dev: Device of interest
1933 * This function examines error ring of @dev and determines
1934 * whether NCQ needs to be turned off, transfer speed should be
1935 * stepped down, or falling back to PIO is necessary.
1937 * ECAT_ATA_BUS : ATA_BUS error for any command
1939 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1942 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1944 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1945 * data transfer hasn't been verified.
1949 * NCQ_OFF : Turn off NCQ.
1951 * SPEED_DOWN : Speed down transfer speed but don't fall back
1954 * FALLBACK_TO_PIO : Fall back to PIO.
1956 * Even if multiple verdicts are returned, only one action is
1957 * taken per error. An action triggered by non-DUBIOUS errors
1958 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1959 * This is to expedite speed down decisions right after device is
1960 * initially configured.
1962 * The followings are speed down rules. #1 and #2 deal with
1965 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1966 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1968 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1969 * occurred during last 5 mins, NCQ_OFF.
1971 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1972 * occurred during last 5 mins, FALLBACK_TO_PIO
1974 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1975 * during last 10 mins, NCQ_OFF.
1977 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1978 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1981 * Inherited from caller.
1984 * OR of ATA_EH_SPDN_* flags.
1986 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1988 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1989 u64 j64 = get_jiffies_64();
1990 struct speed_down_verdict_arg arg;
1991 unsigned int verdict = 0;
1993 /* scan past 5 mins of error history */
1994 memset(&arg, 0, sizeof(arg));
1995 arg.since = j64 - min(j64, j5mins);
1996 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1998 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1999 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
2000 verdict |= ATA_EH_SPDN_SPEED_DOWN |
2001 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
2003 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
2004 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
2005 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
2007 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2008 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2009 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2010 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
2012 /* scan past 10 mins of error history */
2013 memset(&arg, 0, sizeof(arg));
2014 arg.since = j64 - min(j64, j10mins);
2015 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2017 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2018 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
2019 verdict |= ATA_EH_SPDN_NCQ_OFF;
2021 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2022 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
2023 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2024 verdict |= ATA_EH_SPDN_SPEED_DOWN;
2030 * ata_eh_speed_down - record error and speed down if necessary
2031 * @dev: Failed device
2032 * @eflags: mask of ATA_EFLAG_* flags
2033 * @err_mask: err_mask of the error
2035 * Record error and examine error history to determine whether
2036 * adjusting transmission speed is necessary. It also sets
2037 * transmission limits appropriately if such adjustment is
2041 * Kernel thread context (may sleep).
2044 * Determined recovery action.
2046 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2047 unsigned int eflags, unsigned int err_mask)
2049 struct ata_link *link = ata_dev_phys_link(dev);
2051 unsigned int verdict;
2052 unsigned int action = 0;
2054 /* don't bother if Cat-0 error */
2055 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2058 /* record error and determine whether speed down is necessary */
2059 ata_ering_record(&dev->ering, eflags, err_mask);
2060 verdict = ata_eh_speed_down_verdict(dev);
2063 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2064 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2065 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2066 dev->flags |= ATA_DFLAG_NCQ_OFF;
2067 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2072 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2073 /* speed down SATA link speed if possible */
2074 if (sata_down_spd_limit(link, 0) == 0) {
2075 action |= ATA_EH_RESET;
2079 /* lower transfer mode */
2080 if (dev->spdn_cnt < 2) {
2081 static const int dma_dnxfer_sel[] =
2082 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2083 static const int pio_dnxfer_sel[] =
2084 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2087 if (dev->xfer_shift != ATA_SHIFT_PIO)
2088 sel = dma_dnxfer_sel[dev->spdn_cnt];
2090 sel = pio_dnxfer_sel[dev->spdn_cnt];
2094 if (ata_down_xfermask_limit(dev, sel) == 0) {
2095 action |= ATA_EH_RESET;
2101 /* Fall back to PIO? Slowing down to PIO is meaningless for
2102 * SATA ATA devices. Consider it only for PATA and SATAPI.
2104 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2105 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2106 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2107 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2109 action |= ATA_EH_RESET;
2116 /* device has been slowed down, blow error history */
2117 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2118 ata_ering_clear(&dev->ering);
2123 * ata_eh_worth_retry - analyze error and decide whether to retry
2124 * @qc: qc to possibly retry
2126 * Look at the cause of the error and decide if a retry
2127 * might be useful or not. We don't want to retry media errors
2128 * because the drive itself has probably already taken 10-30 seconds
2129 * doing its own internal retries before reporting the failure.
2131 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2133 if (qc->err_mask & AC_ERR_MEDIA)
2134 return 0; /* don't retry media errors */
2135 if (qc->flags & ATA_QCFLAG_IO)
2136 return 1; /* otherwise retry anything from fs stack */
2137 if (qc->err_mask & AC_ERR_INVALID)
2138 return 0; /* don't retry these */
2139 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2143 * ata_eh_link_autopsy - analyze error and determine recovery action
2144 * @link: host link to perform autopsy on
2146 * Analyze why @link failed and determine which recovery actions
2147 * are needed. This function also sets more detailed AC_ERR_*
2148 * values and fills sense data for ATAPI CHECK SENSE.
2151 * Kernel thread context (may sleep).
2153 static void ata_eh_link_autopsy(struct ata_link *link)
2155 struct ata_port *ap = link->ap;
2156 struct ata_eh_context *ehc = &link->eh_context;
2157 struct ata_device *dev;
2158 unsigned int all_err_mask = 0, eflags = 0;
2165 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2168 /* obtain and analyze SError */
2169 rc = sata_scr_read(link, SCR_ERROR, &serror);
2171 ehc->i.serror |= serror;
2172 ata_eh_analyze_serror(link);
2173 } else if (rc != -EOPNOTSUPP) {
2174 /* SError read failed, force reset and probing */
2175 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2176 ehc->i.action |= ATA_EH_RESET;
2177 ehc->i.err_mask |= AC_ERR_OTHER;
2180 /* analyze NCQ failure */
2181 ata_eh_analyze_ncq_error(link);
2183 /* any real error trumps AC_ERR_OTHER */
2184 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2185 ehc->i.err_mask &= ~AC_ERR_OTHER;
2187 all_err_mask |= ehc->i.err_mask;
2189 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2190 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2192 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2193 ata_dev_phys_link(qc->dev) != link)
2196 /* inherit upper level err_mask */
2197 qc->err_mask |= ehc->i.err_mask;
2200 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2202 /* DEV errors are probably spurious in case of ATA_BUS error */
2203 if (qc->err_mask & AC_ERR_ATA_BUS)
2204 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2207 /* any real error trumps unknown error */
2208 if (qc->err_mask & ~AC_ERR_OTHER)
2209 qc->err_mask &= ~AC_ERR_OTHER;
2211 /* SENSE_VALID trumps dev/unknown error and revalidation */
2212 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2213 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2215 /* determine whether the command is worth retrying */
2216 if (ata_eh_worth_retry(qc))
2217 qc->flags |= ATA_QCFLAG_RETRY;
2219 /* accumulate error info */
2220 ehc->i.dev = qc->dev;
2221 all_err_mask |= qc->err_mask;
2222 if (qc->flags & ATA_QCFLAG_IO)
2223 eflags |= ATA_EFLAG_IS_IO;
2224 trace_ata_eh_link_autopsy_qc(qc);
2227 /* enforce default EH actions */
2228 if (ap->pflags & ATA_PFLAG_FROZEN ||
2229 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2230 ehc->i.action |= ATA_EH_RESET;
2231 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2232 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2233 ehc->i.action |= ATA_EH_REVALIDATE;
2235 /* If we have offending qcs and the associated failed device,
2236 * perform per-dev EH action only on the offending device.
2239 ehc->i.dev_action[ehc->i.dev->devno] |=
2240 ehc->i.action & ATA_EH_PERDEV_MASK;
2241 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2244 /* propagate timeout to host link */
2245 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2246 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2248 /* record error and consider speeding down */
2250 if (!dev && ((ata_link_max_devices(link) == 1 &&
2251 ata_dev_enabled(link->device))))
2255 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2256 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2257 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2259 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2264 * ata_eh_autopsy - analyze error and determine recovery action
2265 * @ap: host port to perform autopsy on
2267 * Analyze all links of @ap and determine why they failed and
2268 * which recovery actions are needed.
2271 * Kernel thread context (may sleep).
2273 void ata_eh_autopsy(struct ata_port *ap)
2275 struct ata_link *link;
2277 ata_for_each_link(link, ap, EDGE)
2278 ata_eh_link_autopsy(link);
2280 /* Handle the frigging slave link. Autopsy is done similarly
2281 * but actions and flags are transferred over to the master
2282 * link and handled from there.
2284 if (ap->slave_link) {
2285 struct ata_eh_context *mehc = &ap->link.eh_context;
2286 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2288 /* transfer control flags from master to slave */
2289 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2291 /* perform autopsy on the slave link */
2292 ata_eh_link_autopsy(ap->slave_link);
2294 /* transfer actions from slave to master and clear slave */
2295 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2296 mehc->i.action |= sehc->i.action;
2297 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2298 mehc->i.flags |= sehc->i.flags;
2299 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2302 /* Autopsy of fanout ports can affect host link autopsy.
2303 * Perform host link autopsy last.
2305 if (sata_pmp_attached(ap))
2306 ata_eh_link_autopsy(&ap->link);
2310 * ata_get_cmd_descript - get description for ATA command
2311 * @command: ATA command code to get description for
2313 * Return a textual description of the given command, or NULL if the
2314 * command is not known.
2319 const char *ata_get_cmd_descript(u8 command)
2321 #ifdef CONFIG_ATA_VERBOSE_ERROR
2327 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2328 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2329 { ATA_CMD_STANDBY, "STANDBY" },
2330 { ATA_CMD_IDLE, "IDLE" },
2331 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2332 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2333 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2334 { ATA_CMD_NOP, "NOP" },
2335 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2336 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2337 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2338 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2339 { ATA_CMD_SERVICE, "SERVICE" },
2340 { ATA_CMD_READ, "READ DMA" },
2341 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2342 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2343 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2344 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2345 { ATA_CMD_WRITE, "WRITE DMA" },
2346 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2347 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2348 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2349 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2350 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2351 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2352 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2353 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2354 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2355 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2356 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2357 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2358 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2359 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2360 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2361 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2362 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2363 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2364 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2365 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2366 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2367 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2368 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2369 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2370 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2371 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2372 { ATA_CMD_SLEEP, "SLEEP" },
2373 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2374 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2375 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2376 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2377 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2378 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2379 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2380 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2381 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2382 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2383 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2384 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2385 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2386 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2387 { ATA_CMD_PMP_READ, "READ BUFFER" },
2388 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2389 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2390 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2391 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2392 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2393 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2394 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2395 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2396 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2397 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2398 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2399 { ATA_CMD_SMART, "SMART" },
2400 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2401 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2402 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2403 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2404 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2405 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2406 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2407 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2408 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2409 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2410 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2411 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2412 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2413 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2414 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2415 { ATA_CMD_RESTORE, "RECALIBRATE" },
2416 { 0, NULL } /* terminate list */
2420 for (i = 0; cmd_descr[i].text; i++)
2421 if (cmd_descr[i].command == command)
2422 return cmd_descr[i].text;
2427 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2430 * ata_eh_link_report - report error handling to user
2431 * @link: ATA link EH is going on
2433 * Report EH to user.
2438 static void ata_eh_link_report(struct ata_link *link)
2440 struct ata_port *ap = link->ap;
2441 struct ata_eh_context *ehc = &link->eh_context;
2442 const char *frozen, *desc;
2443 char tries_buf[6] = "";
2444 int tag, nr_failed = 0;
2446 if (ehc->i.flags & ATA_EHI_QUIET)
2450 if (ehc->i.desc[0] != '\0')
2453 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2454 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2456 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2457 ata_dev_phys_link(qc->dev) != link ||
2458 ((qc->flags & ATA_QCFLAG_QUIET) &&
2459 qc->err_mask == AC_ERR_DEV))
2461 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2467 if (!nr_failed && !ehc->i.err_mask)
2471 if (ap->pflags & ATA_PFLAG_FROZEN)
2474 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2475 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2479 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2480 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2481 ehc->i.err_mask, link->sactive, ehc->i.serror,
2482 ehc->i.action, frozen, tries_buf);
2484 ata_dev_err(ehc->i.dev, "%s\n", desc);
2486 ata_link_err(link, "exception Emask 0x%x "
2487 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2488 ehc->i.err_mask, link->sactive, ehc->i.serror,
2489 ehc->i.action, frozen, tries_buf);
2491 ata_link_err(link, "%s\n", desc);
2494 #ifdef CONFIG_ATA_VERBOSE_ERROR
2497 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2498 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2499 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2500 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2501 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2502 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2503 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2504 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2505 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2506 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2507 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2508 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2509 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2510 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2511 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2512 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2513 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2514 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2517 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2518 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2519 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2520 char data_buf[20] = "";
2521 char cdb_buf[70] = "";
2523 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2524 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2527 if (qc->dma_dir != DMA_NONE) {
2528 static const char *dma_str[] = {
2529 [DMA_BIDIRECTIONAL] = "bidi",
2530 [DMA_TO_DEVICE] = "out",
2531 [DMA_FROM_DEVICE] = "in",
2533 static const char *prot_str[] = {
2534 [ATA_PROT_PIO] = "pio",
2535 [ATA_PROT_DMA] = "dma",
2536 [ATA_PROT_NCQ] = "ncq",
2537 [ATAPI_PROT_PIO] = "pio",
2538 [ATAPI_PROT_DMA] = "dma",
2541 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2542 prot_str[qc->tf.protocol], qc->nbytes,
2543 dma_str[qc->dma_dir]);
2546 if (ata_is_atapi(qc->tf.protocol)) {
2547 const u8 *cdb = qc->cdb;
2548 size_t cdb_len = qc->dev->cdb_len;
2551 cdb = qc->scsicmd->cmnd;
2552 cdb_len = qc->scsicmd->cmd_len;
2554 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2557 const char *descr = ata_get_cmd_descript(cmd->command);
2559 ata_dev_err(qc->dev, "failed command: %s\n",
2563 ata_dev_err(qc->dev,
2564 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2566 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2567 "Emask 0x%x (%s)%s\n",
2568 cmd->command, cmd->feature, cmd->nsect,
2569 cmd->lbal, cmd->lbam, cmd->lbah,
2570 cmd->hob_feature, cmd->hob_nsect,
2571 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2572 cmd->device, qc->tag, data_buf, cdb_buf,
2573 res->command, res->feature, res->nsect,
2574 res->lbal, res->lbam, res->lbah,
2575 res->hob_feature, res->hob_nsect,
2576 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2577 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2578 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2580 #ifdef CONFIG_ATA_VERBOSE_ERROR
2581 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2583 if (res->command & ATA_BUSY)
2584 ata_dev_err(qc->dev, "status: { Busy }\n");
2586 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
2587 res->command & ATA_DRDY ? "DRDY " : "",
2588 res->command & ATA_DF ? "DF " : "",
2589 res->command & ATA_DRQ ? "DRQ " : "",
2590 res->command & ATA_ERR ? "ERR " : "");
2593 if (cmd->command != ATA_CMD_PACKET &&
2594 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2595 ATA_IDNF | ATA_ABORTED)))
2596 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2597 res->feature & ATA_ICRC ? "ICRC " : "",
2598 res->feature & ATA_UNC ? "UNC " : "",
2599 res->feature & ATA_AMNF ? "AMNF " : "",
2600 res->feature & ATA_IDNF ? "IDNF " : "",
2601 res->feature & ATA_ABORTED ? "ABRT " : "");
2607 * ata_eh_report - report error handling to user
2608 * @ap: ATA port to report EH about
2610 * Report EH to user.
2615 void ata_eh_report(struct ata_port *ap)
2617 struct ata_link *link;
2619 ata_for_each_link(link, ap, HOST_FIRST)
2620 ata_eh_link_report(link);
2623 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2624 unsigned int *classes, unsigned long deadline,
2627 struct ata_device *dev;
2630 ata_for_each_dev(dev, link, ALL)
2631 classes[dev->devno] = ATA_DEV_UNKNOWN;
2633 return reset(link, classes, deadline);
2636 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2638 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2642 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2647 int ata_eh_reset(struct ata_link *link, int classify,
2648 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2649 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2651 struct ata_port *ap = link->ap;
2652 struct ata_link *slave = ap->slave_link;
2653 struct ata_eh_context *ehc = &link->eh_context;
2654 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2655 unsigned int *classes = ehc->classes;
2656 unsigned int lflags = link->flags;
2657 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2658 int max_tries = 0, try = 0;
2659 struct ata_link *failed_link;
2660 struct ata_device *dev;
2661 unsigned long deadline, now;
2662 ata_reset_fn_t reset;
2663 unsigned long flags;
2670 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2672 if (link->flags & ATA_LFLAG_RST_ONCE)
2674 if (link->flags & ATA_LFLAG_NO_HRST)
2676 if (link->flags & ATA_LFLAG_NO_SRST)
2679 /* make sure each reset attempt is at least COOL_DOWN apart */
2680 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2682 WARN_ON(time_after(ehc->last_reset, now));
2683 deadline = ata_deadline(ehc->last_reset,
2684 ATA_EH_RESET_COOL_DOWN);
2685 if (time_before(now, deadline))
2686 schedule_timeout_uninterruptible(deadline - now);
2689 spin_lock_irqsave(ap->lock, flags);
2690 ap->pflags |= ATA_PFLAG_RESETTING;
2691 spin_unlock_irqrestore(ap->lock, flags);
2693 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2695 ata_for_each_dev(dev, link, ALL) {
2696 /* If we issue an SRST then an ATA drive (not ATAPI)
2697 * may change configuration and be in PIO0 timing. If
2698 * we do a hard reset (or are coming from power on)
2699 * this is true for ATA or ATAPI. Until we've set a
2700 * suitable controller mode we should not touch the
2701 * bus as we may be talking too fast.
2703 dev->pio_mode = XFER_PIO_0;
2704 dev->dma_mode = 0xff;
2706 /* If the controller has a pio mode setup function
2707 * then use it to set the chipset to rights. Don't
2708 * touch the DMA setup as that will be dealt with when
2709 * configuring devices.
2711 if (ap->ops->set_piomode)
2712 ap->ops->set_piomode(ap, dev);
2715 /* prefer hardreset */
2717 ehc->i.action &= ~ATA_EH_RESET;
2720 ehc->i.action |= ATA_EH_HARDRESET;
2721 } else if (softreset) {
2723 ehc->i.action |= ATA_EH_SOFTRESET;
2727 unsigned long deadline = ata_deadline(jiffies,
2728 ATA_EH_PRERESET_TIMEOUT);
2731 sehc->i.action &= ~ATA_EH_RESET;
2732 sehc->i.action |= ehc->i.action;
2735 rc = prereset(link, deadline);
2737 /* If present, do prereset on slave link too. Reset
2738 * is skipped iff both master and slave links report
2739 * -ENOENT or clear ATA_EH_RESET.
2741 if (slave && (rc == 0 || rc == -ENOENT)) {
2744 tmp = prereset(slave, deadline);
2748 ehc->i.action |= sehc->i.action;
2752 if (rc == -ENOENT) {
2753 ata_link_dbg(link, "port disabled--ignoring\n");
2754 ehc->i.action &= ~ATA_EH_RESET;
2756 ata_for_each_dev(dev, link, ALL)
2757 classes[dev->devno] = ATA_DEV_NONE;
2762 "prereset failed (errno=%d)\n",
2767 /* prereset() might have cleared ATA_EH_RESET. If so,
2768 * bang classes, thaw and return.
2770 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2771 ata_for_each_dev(dev, link, ALL)
2772 classes[dev->devno] = ATA_DEV_NONE;
2773 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2774 ata_is_host_link(link))
2775 ata_eh_thaw_port(ap);
2785 if (ata_is_host_link(link))
2786 ata_eh_freeze_port(ap);
2788 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2792 ata_link_info(link, "%s resetting link\n",
2793 reset == softreset ? "soft" : "hard");
2795 /* mark that this EH session started with reset */
2796 ehc->last_reset = jiffies;
2797 if (reset == hardreset)
2798 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2800 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2802 rc = ata_do_reset(link, reset, classes, deadline, true);
2803 if (rc && rc != -EAGAIN) {
2808 /* hardreset slave link if existent */
2809 if (slave && reset == hardreset) {
2813 ata_link_info(slave, "hard resetting link\n");
2815 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2816 tmp = ata_do_reset(slave, reset, classes, deadline,
2824 failed_link = slave;
2830 /* perform follow-up SRST if necessary */
2831 if (reset == hardreset &&
2832 ata_eh_followup_srst_needed(link, rc)) {
2837 "follow-up softreset required but no softreset available\n");
2843 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2844 rc = ata_do_reset(link, reset, classes, deadline, true);
2853 "no reset method available, skipping reset\n");
2854 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2855 lflags |= ATA_LFLAG_ASSUME_ATA;
2859 * Post-reset processing
2861 ata_for_each_dev(dev, link, ALL) {
2862 /* After the reset, the device state is PIO 0 and the
2863 * controller state is undefined. Reset also wakes up
2864 * drives from sleeping mode.
2866 dev->pio_mode = XFER_PIO_0;
2867 dev->flags &= ~ATA_DFLAG_SLEEPING;
2869 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2872 /* apply class override */
2873 if (lflags & ATA_LFLAG_ASSUME_ATA)
2874 classes[dev->devno] = ATA_DEV_ATA;
2875 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2876 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2879 /* record current link speed */
2880 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2881 link->sata_spd = (sstatus >> 4) & 0xf;
2882 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2883 slave->sata_spd = (sstatus >> 4) & 0xf;
2886 if (ata_is_host_link(link))
2887 ata_eh_thaw_port(ap);
2889 /* postreset() should clear hardware SError. Although SError
2890 * is cleared during link resume, clearing SError here is
2891 * necessary as some PHYs raise hotplug events after SRST.
2892 * This introduces race condition where hotplug occurs between
2893 * reset and here. This race is mediated by cross checking
2894 * link onlineness and classification result later.
2897 postreset(link, classes);
2899 postreset(slave, classes);
2903 * Some controllers can't be frozen very well and may set spurious
2904 * error conditions during reset. Clear accumulated error
2905 * information and re-thaw the port if frozen. As reset is the
2906 * final recovery action and we cross check link onlineness against
2907 * device classification later, no hotplug event is lost by this.
2909 spin_lock_irqsave(link->ap->lock, flags);
2910 memset(&link->eh_info, 0, sizeof(link->eh_info));
2912 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2913 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2914 spin_unlock_irqrestore(link->ap->lock, flags);
2916 if (ap->pflags & ATA_PFLAG_FROZEN)
2917 ata_eh_thaw_port(ap);
2920 * Make sure onlineness and classification result correspond.
2921 * Hotplug could have happened during reset and some
2922 * controllers fail to wait while a drive is spinning up after
2923 * being hotplugged causing misdetection. By cross checking
2924 * link on/offlineness and classification result, those
2925 * conditions can be reliably detected and retried.
2928 ata_for_each_dev(dev, link, ALL) {
2929 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2930 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2931 ata_dev_dbg(dev, "link online but device misclassified\n");
2932 classes[dev->devno] = ATA_DEV_NONE;
2935 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2936 if (ata_class_enabled(classes[dev->devno]))
2938 "link offline, clearing class %d to NONE\n",
2939 classes[dev->devno]);
2940 classes[dev->devno] = ATA_DEV_NONE;
2941 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2943 "link status unknown, clearing UNKNOWN to NONE\n");
2944 classes[dev->devno] = ATA_DEV_NONE;
2948 if (classify && nr_unknown) {
2949 if (try < max_tries) {
2951 "link online but %d devices misclassified, retrying\n",
2958 "link online but %d devices misclassified, "
2959 "device detection might fail\n", nr_unknown);
2962 /* reset successful, schedule revalidation */
2963 ata_eh_done(link, NULL, ATA_EH_RESET);
2965 ata_eh_done(slave, NULL, ATA_EH_RESET);
2966 ehc->last_reset = jiffies; /* update to completion time */
2967 ehc->i.action |= ATA_EH_REVALIDATE;
2968 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2972 /* clear hotplug flag */
2973 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2975 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2977 spin_lock_irqsave(ap->lock, flags);
2978 ap->pflags &= ~ATA_PFLAG_RESETTING;
2979 spin_unlock_irqrestore(ap->lock, flags);
2984 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2985 if (!ata_is_host_link(link) &&
2986 sata_scr_read(link, SCR_STATUS, &sstatus))
2989 if (try >= max_tries) {
2991 * Thaw host port even if reset failed, so that the port
2992 * can be retried on the next phy event. This risks
2993 * repeated EH runs but seems to be a better tradeoff than
2994 * shutting down a port after a botched hotplug attempt.
2996 if (ata_is_host_link(link))
2997 ata_eh_thaw_port(ap);
3002 if (time_before(now, deadline)) {
3003 unsigned long delta = deadline - now;
3005 ata_link_warn(failed_link,
3006 "reset failed (errno=%d), retrying in %u secs\n",
3007 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3011 delta = schedule_timeout_uninterruptible(delta);
3016 * While disks spinup behind PMP, some controllers fail sending SRST.
3017 * They need to be reset - as well as the PMP - before retrying.
3019 if (rc == -ERESTART) {
3020 if (ata_is_host_link(link))
3021 ata_eh_thaw_port(ap);
3025 if (try == max_tries - 1) {
3026 sata_down_spd_limit(link, 0);
3028 sata_down_spd_limit(slave, 0);
3029 } else if (rc == -EPIPE)
3030 sata_down_spd_limit(failed_link, 0);
3037 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3039 struct ata_link *link;
3040 struct ata_device *dev;
3041 unsigned long flags;
3044 * This function can be thought of as an extended version of
3045 * ata_eh_about_to_do() specially crafted to accommodate the
3046 * requirements of ATA_EH_PARK handling. Since the EH thread
3047 * does not leave the do {} while () loop in ata_eh_recover as
3048 * long as the timeout for a park request to *one* device on
3049 * the port has not expired, and since we still want to pick
3050 * up park requests to other devices on the same port or
3051 * timeout updates for the same device, we have to pull
3052 * ATA_EH_PARK actions from eh_info into eh_context.i
3053 * ourselves at the beginning of each pass over the loop.
3055 * Additionally, all write accesses to &ap->park_req_pending
3056 * through reinit_completion() (see below) or complete_all()
3057 * (see ata_scsi_park_store()) are protected by the host lock.
3058 * As a result we have that park_req_pending.done is zero on
3059 * exit from this function, i.e. when ATA_EH_PARK actions for
3060 * *all* devices on port ap have been pulled into the
3061 * respective eh_context structs. If, and only if,
3062 * park_req_pending.done is non-zero by the time we reach
3063 * wait_for_completion_timeout(), another ATA_EH_PARK action
3064 * has been scheduled for at least one of the devices on port
3065 * ap and we have to cycle over the do {} while () loop in
3066 * ata_eh_recover() again.
3069 spin_lock_irqsave(ap->lock, flags);
3070 reinit_completion(&ap->park_req_pending);
3071 ata_for_each_link(link, ap, EDGE) {
3072 ata_for_each_dev(dev, link, ALL) {
3073 struct ata_eh_info *ehi = &link->eh_info;
3075 link->eh_context.i.dev_action[dev->devno] |=
3076 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3077 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3080 spin_unlock_irqrestore(ap->lock, flags);
3083 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3085 struct ata_eh_context *ehc = &dev->link->eh_context;
3086 struct ata_taskfile tf;
3087 unsigned int err_mask;
3089 ata_tf_init(dev, &tf);
3091 ehc->unloaded_mask |= 1 << dev->devno;
3092 tf.command = ATA_CMD_IDLEIMMEDIATE;
3098 ehc->unloaded_mask &= ~(1 << dev->devno);
3099 tf.command = ATA_CMD_CHK_POWER;
3102 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3103 tf.protocol |= ATA_PROT_NODATA;
3104 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3105 if (park && (err_mask || tf.lbal != 0xc4)) {
3106 ata_dev_err(dev, "head unload failed!\n");
3107 ehc->unloaded_mask &= ~(1 << dev->devno);
3111 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3112 struct ata_device **r_failed_dev)
3114 struct ata_port *ap = link->ap;
3115 struct ata_eh_context *ehc = &link->eh_context;
3116 struct ata_device *dev;
3117 unsigned int new_mask = 0;
3118 unsigned long flags;
3123 /* For PATA drive side cable detection to work, IDENTIFY must
3124 * be done backwards such that PDIAG- is released by the slave
3125 * device before the master device is identified.
3127 ata_for_each_dev(dev, link, ALL_REVERSE) {
3128 unsigned int action = ata_eh_dev_action(dev);
3129 unsigned int readid_flags = 0;
3131 if (ehc->i.flags & ATA_EHI_DID_RESET)
3132 readid_flags |= ATA_READID_POSTRESET;
3134 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3135 WARN_ON(dev->class == ATA_DEV_PMP);
3137 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3142 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3143 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3148 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3150 /* Configuration may have changed, reconfigure
3153 ehc->i.flags |= ATA_EHI_SETMODE;
3155 /* schedule the scsi_rescan_device() here */
3156 schedule_work(&(ap->scsi_rescan_task));
3157 } else if (dev->class == ATA_DEV_UNKNOWN &&
3158 ehc->tries[dev->devno] &&
3159 ata_class_enabled(ehc->classes[dev->devno])) {
3160 /* Temporarily set dev->class, it will be
3161 * permanently set once all configurations are
3162 * complete. This is necessary because new
3163 * device configuration is done in two
3166 dev->class = ehc->classes[dev->devno];
3168 if (dev->class == ATA_DEV_PMP)
3169 rc = sata_pmp_attach(dev);
3171 rc = ata_dev_read_id(dev, &dev->class,
3172 readid_flags, dev->id);
3174 /* read_id might have changed class, store and reset */
3175 ehc->classes[dev->devno] = dev->class;
3176 dev->class = ATA_DEV_UNKNOWN;
3180 /* clear error info accumulated during probe */
3181 ata_ering_clear(&dev->ering);
3182 new_mask |= 1 << dev->devno;
3185 /* IDENTIFY was issued to non-existent
3186 * device. No need to reset. Just
3187 * thaw and ignore the device.
3189 ata_eh_thaw_port(ap);
3197 /* PDIAG- should have been released, ask cable type if post-reset */
3198 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3199 if (ap->ops->cable_detect)
3200 ap->cbl = ap->ops->cable_detect(ap);
3204 /* Configure new devices forward such that user doesn't see
3205 * device detection messages backwards.
3207 ata_for_each_dev(dev, link, ALL) {
3208 if (!(new_mask & (1 << dev->devno)))
3211 dev->class = ehc->classes[dev->devno];
3213 if (dev->class == ATA_DEV_PMP)
3216 ehc->i.flags |= ATA_EHI_PRINTINFO;
3217 rc = ata_dev_configure(dev);
3218 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3220 dev->class = ATA_DEV_UNKNOWN;
3224 spin_lock_irqsave(ap->lock, flags);
3225 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3226 spin_unlock_irqrestore(ap->lock, flags);
3228 /* new device discovered, configure xfermode */
3229 ehc->i.flags |= ATA_EHI_SETMODE;
3235 *r_failed_dev = dev;
3236 DPRINTK("EXIT rc=%d\n", rc);
3241 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3242 * @link: link on which timings will be programmed
3243 * @r_failed_dev: out parameter for failed device
3245 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3246 * ata_set_mode() fails, pointer to the failing device is
3247 * returned in @r_failed_dev.
3250 * PCI/etc. bus probe sem.
3253 * 0 on success, negative errno otherwise
3255 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3257 struct ata_port *ap = link->ap;
3258 struct ata_device *dev;
3261 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3262 ata_for_each_dev(dev, link, ENABLED) {
3263 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3264 struct ata_ering_entry *ent;
3266 ent = ata_ering_top(&dev->ering);
3268 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3272 /* has private set_mode? */
3273 if (ap->ops->set_mode)
3274 rc = ap->ops->set_mode(link, r_failed_dev);
3276 rc = ata_do_set_mode(link, r_failed_dev);
3278 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3279 ata_for_each_dev(dev, link, ENABLED) {
3280 struct ata_eh_context *ehc = &link->eh_context;
3281 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3282 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3284 if (dev->xfer_mode != saved_xfer_mode ||
3285 ata_ncq_enabled(dev) != saved_ncq)
3286 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3293 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3294 * @dev: ATAPI device to clear UA for
3296 * Resets and other operations can make an ATAPI device raise
3297 * UNIT ATTENTION which causes the next operation to fail. This
3298 * function clears UA.
3301 * EH context (may sleep).
3304 * 0 on success, -errno on failure.
3306 static int atapi_eh_clear_ua(struct ata_device *dev)
3310 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3311 u8 *sense_buffer = dev->link->ap->sector_buf;
3313 unsigned int err_mask;
3315 err_mask = atapi_eh_tur(dev, &sense_key);
3316 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3318 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3323 if (!err_mask || sense_key != UNIT_ATTENTION)
3326 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3328 ata_dev_warn(dev, "failed to clear "
3329 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3334 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3341 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3342 * @dev: ATA device which may need FLUSH retry
3344 * If @dev failed FLUSH, it needs to be reported upper layer
3345 * immediately as it means that @dev failed to remap and already
3346 * lost at least a sector and further FLUSH retrials won't make
3347 * any difference to the lost sector. However, if FLUSH failed
3348 * for other reasons, for example transmission error, FLUSH needs
3351 * This function determines whether FLUSH failure retry is
3352 * necessary and performs it if so.
3355 * 0 if EH can continue, -errno if EH needs to be repeated.
3357 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3359 struct ata_link *link = dev->link;
3360 struct ata_port *ap = link->ap;
3361 struct ata_queued_cmd *qc;
3362 struct ata_taskfile tf;
3363 unsigned int err_mask;
3366 /* did flush fail for this device? */
3367 if (!ata_tag_valid(link->active_tag))
3370 qc = __ata_qc_from_tag(ap, link->active_tag);
3371 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3372 qc->tf.command != ATA_CMD_FLUSH))
3375 /* if the device failed it, it should be reported to upper layers */
3376 if (qc->err_mask & AC_ERR_DEV)
3379 /* flush failed for some other reason, give it another shot */
3380 ata_tf_init(dev, &tf);
3382 tf.command = qc->tf.command;
3383 tf.flags |= ATA_TFLAG_DEVICE;
3384 tf.protocol = ATA_PROT_NODATA;
3386 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3387 tf.command, qc->err_mask);
3389 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3392 * FLUSH is complete but there's no way to
3393 * successfully complete a failed command from EH.
3394 * Making sure retry is allowed at least once and
3395 * retrying it should do the trick - whatever was in
3396 * the cache is already on the platter and this won't
3397 * cause infinite loop.
3399 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3401 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3405 /* if device failed it, report it to upper layers */
3406 if (err_mask & AC_ERR_DEV) {
3407 qc->err_mask |= AC_ERR_DEV;
3409 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3417 * ata_eh_set_lpm - configure SATA interface power management
3418 * @link: link to configure power management
3419 * @policy: the link power management policy
3420 * @r_failed_dev: out parameter for failed device
3422 * Enable SATA Interface power management. This will enable
3423 * Device Interface Power Management (DIPM) for min_power
3424 * policy, and then call driver specific callbacks for
3425 * enabling Host Initiated Power management.
3431 * 0 on success, -errno on failure.
3433 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3434 struct ata_device **r_failed_dev)
3436 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3437 struct ata_eh_context *ehc = &link->eh_context;
3438 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3439 enum ata_lpm_policy old_policy = link->lpm_policy;
3440 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3441 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3442 unsigned int err_mask;
3445 /* if the link or host doesn't do LPM, noop */
3446 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3450 * DIPM is enabled only for MIN_POWER as some devices
3451 * misbehave when the host NACKs transition to SLUMBER. Order
3452 * device and link configurations such that the host always
3453 * allows DIPM requests.
3455 ata_for_each_dev(dev, link, ENABLED) {
3456 bool hipm = ata_id_has_hipm(dev->id);
3457 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3459 /* find the first enabled and LPM enabled devices */
3463 if (!lpm_dev && (hipm || dipm))
3466 hints &= ~ATA_LPM_EMPTY;
3468 hints &= ~ATA_LPM_HIPM;
3470 /* disable DIPM before changing link config */
3471 if (policy != ATA_LPM_MIN_POWER && dipm) {
3472 err_mask = ata_dev_set_feature(dev,
3473 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3474 if (err_mask && err_mask != AC_ERR_DEV) {
3476 "failed to disable DIPM, Emask 0x%x\n",
3485 rc = ap->ops->set_lpm(link, policy, hints);
3486 if (!rc && ap->slave_link)
3487 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3489 rc = sata_pmp_set_lpm(link, policy, hints);
3492 * Attribute link config failure to the first (LPM) enabled
3493 * device on the link.
3496 if (rc == -EOPNOTSUPP) {
3497 link->flags |= ATA_LFLAG_NO_LPM;
3500 dev = lpm_dev ? lpm_dev : link_dev;
3505 * Low level driver acked the transition. Issue DIPM command
3506 * with the new policy set.
3508 link->lpm_policy = policy;
3509 if (ap && ap->slave_link)
3510 ap->slave_link->lpm_policy = policy;
3512 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3513 ata_for_each_dev(dev, link, ENABLED) {
3514 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3515 ata_id_has_dipm(dev->id)) {
3516 err_mask = ata_dev_set_feature(dev,
3517 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3518 if (err_mask && err_mask != AC_ERR_DEV) {
3520 "failed to enable DIPM, Emask 0x%x\n",
3528 link->last_lpm_change = jiffies;
3529 link->flags |= ATA_LFLAG_CHANGED;
3534 /* restore the old policy */
3535 link->lpm_policy = old_policy;
3536 if (ap && ap->slave_link)
3537 ap->slave_link->lpm_policy = old_policy;
3539 /* if no device or only one more chance is left, disable LPM */
3540 if (!dev || ehc->tries[dev->devno] <= 2) {
3541 ata_link_warn(link, "disabling LPM on the link\n");
3542 link->flags |= ATA_LFLAG_NO_LPM;
3545 *r_failed_dev = dev;
3549 int ata_link_nr_enabled(struct ata_link *link)
3551 struct ata_device *dev;
3554 ata_for_each_dev(dev, link, ENABLED)
3559 static int ata_link_nr_vacant(struct ata_link *link)
3561 struct ata_device *dev;
3564 ata_for_each_dev(dev, link, ALL)
3565 if (dev->class == ATA_DEV_UNKNOWN)
3570 static int ata_eh_skip_recovery(struct ata_link *link)
3572 struct ata_port *ap = link->ap;
3573 struct ata_eh_context *ehc = &link->eh_context;
3574 struct ata_device *dev;
3576 /* skip disabled links */
3577 if (link->flags & ATA_LFLAG_DISABLED)
3580 /* skip if explicitly requested */
3581 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3584 /* thaw frozen port and recover failed devices */
3585 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3588 /* reset at least once if reset is requested */
3589 if ((ehc->i.action & ATA_EH_RESET) &&
3590 !(ehc->i.flags & ATA_EHI_DID_RESET))
3593 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3594 ata_for_each_dev(dev, link, ALL) {
3595 if (dev->class == ATA_DEV_UNKNOWN &&
3596 ehc->classes[dev->devno] != ATA_DEV_NONE)
3603 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3605 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3606 u64 now = get_jiffies_64();
3607 int *trials = void_arg;
3609 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3610 (ent->timestamp < now - min(now, interval)))
3617 static int ata_eh_schedule_probe(struct ata_device *dev)
3619 struct ata_eh_context *ehc = &dev->link->eh_context;
3620 struct ata_link *link = ata_dev_phys_link(dev);
3623 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3624 (ehc->did_probe_mask & (1 << dev->devno)))
3627 ata_eh_detach_dev(dev);
3629 ehc->did_probe_mask |= (1 << dev->devno);
3630 ehc->i.action |= ATA_EH_RESET;
3631 ehc->saved_xfer_mode[dev->devno] = 0;
3632 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3634 /* the link maybe in a deep sleep, wake it up */
3635 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3636 if (ata_is_host_link(link))
3637 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3640 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3644 /* Record and count probe trials on the ering. The specific
3645 * error mask used is irrelevant. Because a successful device
3646 * detection clears the ering, this count accumulates only if
3647 * there are consecutive failed probes.
3649 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3650 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3651 * forced to 1.5Gbps.
3653 * This is to work around cases where failed link speed
3654 * negotiation results in device misdetection leading to
3655 * infinite DEVXCHG or PHRDY CHG events.
3657 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3658 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3660 if (trials > ATA_EH_PROBE_TRIALS)
3661 sata_down_spd_limit(link, 1);
3666 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3668 struct ata_eh_context *ehc = &dev->link->eh_context;
3670 /* -EAGAIN from EH routine indicates retry without prejudice.
3671 * The requester is responsible for ensuring forward progress.
3674 ehc->tries[dev->devno]--;
3678 /* device missing or wrong IDENTIFY data, schedule probing */
3679 ehc->i.probe_mask |= (1 << dev->devno);
3681 /* give it just one more chance */
3682 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3684 if (ehc->tries[dev->devno] == 1) {
3685 /* This is the last chance, better to slow
3686 * down than lose it.
3688 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3689 if (dev->pio_mode > XFER_PIO_0)
3690 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3694 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3695 /* disable device if it has used up all its chances */
3696 ata_dev_disable(dev);
3698 /* detach if offline */
3699 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3700 ata_eh_detach_dev(dev);
3702 /* schedule probe if necessary */
3703 if (ata_eh_schedule_probe(dev)) {
3704 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3705 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3706 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3711 ehc->i.action |= ATA_EH_RESET;
3717 * ata_eh_recover - recover host port after error
3718 * @ap: host port to recover
3719 * @prereset: prereset method (can be NULL)
3720 * @softreset: softreset method (can be NULL)
3721 * @hardreset: hardreset method (can be NULL)
3722 * @postreset: postreset method (can be NULL)
3723 * @r_failed_link: out parameter for failed link
3725 * This is the alpha and omega, eum and yang, heart and soul of
3726 * libata exception handling. On entry, actions required to
3727 * recover each link and hotplug requests are recorded in the
3728 * link's eh_context. This function executes all the operations
3729 * with appropriate retrials and fallbacks to resurrect failed
3730 * devices, detach goners and greet newcomers.
3733 * Kernel thread context (may sleep).
3736 * 0 on success, -errno on failure.
3738 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3739 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3740 ata_postreset_fn_t postreset,
3741 struct ata_link **r_failed_link)
3743 struct ata_link *link;
3744 struct ata_device *dev;
3746 unsigned long flags, deadline;
3750 /* prep for recovery */
3751 ata_for_each_link(link, ap, EDGE) {
3752 struct ata_eh_context *ehc = &link->eh_context;
3754 /* re-enable link? */
3755 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3756 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3757 spin_lock_irqsave(ap->lock, flags);
3758 link->flags &= ~ATA_LFLAG_DISABLED;
3759 spin_unlock_irqrestore(ap->lock, flags);
3760 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3763 ata_for_each_dev(dev, link, ALL) {
3764 if (link->flags & ATA_LFLAG_NO_RETRY)
3765 ehc->tries[dev->devno] = 1;
3767 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3769 /* collect port action mask recorded in dev actions */
3770 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3771 ~ATA_EH_PERDEV_MASK;
3772 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3774 /* process hotplug request */
3775 if (dev->flags & ATA_DFLAG_DETACH)
3776 ata_eh_detach_dev(dev);
3778 /* schedule probe if necessary */
3779 if (!ata_dev_enabled(dev))
3780 ata_eh_schedule_probe(dev);
3787 /* if UNLOADING, finish immediately */
3788 if (ap->pflags & ATA_PFLAG_UNLOADING)
3792 ata_for_each_link(link, ap, EDGE) {
3793 struct ata_eh_context *ehc = &link->eh_context;
3795 /* skip EH if possible. */
3796 if (ata_eh_skip_recovery(link))
3799 ata_for_each_dev(dev, link, ALL)
3800 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3804 ata_for_each_link(link, ap, EDGE) {
3805 struct ata_eh_context *ehc = &link->eh_context;
3807 if (!(ehc->i.action & ATA_EH_RESET))
3810 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3811 prereset, softreset, hardreset, postreset);
3813 ata_link_err(link, "reset failed, giving up\n");
3822 * clears ATA_EH_PARK in eh_info and resets
3823 * ap->park_req_pending
3825 ata_eh_pull_park_action(ap);
3828 ata_for_each_link(link, ap, EDGE) {
3829 ata_for_each_dev(dev, link, ALL) {
3830 struct ata_eh_context *ehc = &link->eh_context;
3833 if (dev->class != ATA_DEV_ATA &&
3834 dev->class != ATA_DEV_ZAC)
3836 if (!(ehc->i.dev_action[dev->devno] &
3839 tmp = dev->unpark_deadline;
3840 if (time_before(deadline, tmp))
3842 else if (time_before_eq(tmp, jiffies))
3844 if (ehc->unloaded_mask & (1 << dev->devno))
3847 ata_eh_park_issue_cmd(dev, 1);
3852 if (time_before_eq(deadline, now))
3856 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3860 ata_for_each_link(link, ap, EDGE) {
3861 ata_for_each_dev(dev, link, ALL) {
3862 if (!(link->eh_context.unloaded_mask &
3866 ata_eh_park_issue_cmd(dev, 0);
3867 ata_eh_done(link, dev, ATA_EH_PARK);
3873 ata_for_each_link(link, ap, PMP_FIRST) {
3874 struct ata_eh_context *ehc = &link->eh_context;
3876 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3879 /* revalidate existing devices and attach new ones */
3880 rc = ata_eh_revalidate_and_attach(link, &dev);
3884 /* if PMP got attached, return, pmp EH will take care of it */
3885 if (link->device->class == ATA_DEV_PMP) {
3890 /* configure transfer mode if necessary */
3891 if (ehc->i.flags & ATA_EHI_SETMODE) {
3892 rc = ata_set_mode(link, &dev);
3895 ehc->i.flags &= ~ATA_EHI_SETMODE;
3898 /* If reset has been issued, clear UA to avoid
3899 * disrupting the current users of the device.
3901 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3902 ata_for_each_dev(dev, link, ALL) {
3903 if (dev->class != ATA_DEV_ATAPI)
3905 rc = atapi_eh_clear_ua(dev);
3908 if (zpodd_dev_enabled(dev))
3909 zpodd_post_poweron(dev);
3913 /* retry flush if necessary */
3914 ata_for_each_dev(dev, link, ALL) {
3915 if (dev->class != ATA_DEV_ATA &&
3916 dev->class != ATA_DEV_ZAC)
3918 rc = ata_eh_maybe_retry_flush(dev);
3924 /* configure link power saving */
3925 if (link->lpm_policy != ap->target_lpm_policy) {
3926 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3931 /* this link is okay now */
3938 ata_eh_handle_dev_fail(dev, rc);
3940 if (ap->pflags & ATA_PFLAG_FROZEN) {
3941 /* PMP reset requires working host port.
3942 * Can't retry if it's frozen.
3944 if (sata_pmp_attached(ap))
3954 if (rc && r_failed_link)
3955 *r_failed_link = link;
3957 DPRINTK("EXIT, rc=%d\n", rc);
3962 * ata_eh_finish - finish up EH
3963 * @ap: host port to finish EH for
3965 * Recovery is complete. Clean up EH states and retry or finish
3971 void ata_eh_finish(struct ata_port *ap)
3975 /* retry or finish qcs */
3976 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3977 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3979 if (!(qc->flags & ATA_QCFLAG_FAILED))
3983 /* FIXME: Once EH migration is complete,
3984 * generate sense data in this function,
3985 * considering both err_mask and tf.
3987 if (qc->flags & ATA_QCFLAG_RETRY)
3988 ata_eh_qc_retry(qc);
3990 ata_eh_qc_complete(qc);
3992 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3993 ata_eh_qc_complete(qc);
3995 /* feed zero TF to sense generation */
3996 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3997 ata_eh_qc_retry(qc);
4002 /* make sure nr_active_links is zero after EH */
4003 WARN_ON(ap->nr_active_links);
4004 ap->nr_active_links = 0;
4008 * ata_do_eh - do standard error handling
4009 * @ap: host port to handle error for
4011 * @prereset: prereset method (can be NULL)
4012 * @softreset: softreset method (can be NULL)
4013 * @hardreset: hardreset method (can be NULL)
4014 * @postreset: postreset method (can be NULL)
4016 * Perform standard error handling sequence.
4019 * Kernel thread context (may sleep).
4021 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4022 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4023 ata_postreset_fn_t postreset)
4025 struct ata_device *dev;
4031 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4034 ata_for_each_dev(dev, &ap->link, ALL)
4035 ata_dev_disable(dev);
4042 * ata_std_error_handler - standard error handler
4043 * @ap: host port to handle error for
4045 * Standard error handler
4048 * Kernel thread context (may sleep).
4050 void ata_std_error_handler(struct ata_port *ap)
4052 struct ata_port_operations *ops = ap->ops;
4053 ata_reset_fn_t hardreset = ops->hardreset;
4055 /* ignore built-in hardreset if SCR access is not available */
4056 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4059 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4064 * ata_eh_handle_port_suspend - perform port suspend operation
4065 * @ap: port to suspend
4070 * Kernel thread context (may sleep).
4072 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4074 unsigned long flags;
4076 struct ata_device *dev;
4078 /* are we suspending? */
4079 spin_lock_irqsave(ap->lock, flags);
4080 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4081 ap->pm_mesg.event & PM_EVENT_RESUME) {
4082 spin_unlock_irqrestore(ap->lock, flags);
4085 spin_unlock_irqrestore(ap->lock, flags);
4087 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4090 * If we have a ZPODD attached, check its zero
4091 * power ready status before the port is frozen.
4092 * Only needed for runtime suspend.
4094 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4095 ata_for_each_dev(dev, &ap->link, ENABLED) {
4096 if (zpodd_dev_enabled(dev))
4097 zpodd_on_suspend(dev);
4101 /* tell ACPI we're suspending */
4102 rc = ata_acpi_on_suspend(ap);
4107 ata_eh_freeze_port(ap);
4109 if (ap->ops->port_suspend)
4110 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4112 ata_acpi_set_state(ap, ap->pm_mesg);
4114 /* update the flags */
4115 spin_lock_irqsave(ap->lock, flags);
4117 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4119 ap->pflags |= ATA_PFLAG_SUSPENDED;
4120 else if (ap->pflags & ATA_PFLAG_FROZEN)
4121 ata_port_schedule_eh(ap);
4123 spin_unlock_irqrestore(ap->lock, flags);
4129 * ata_eh_handle_port_resume - perform port resume operation
4130 * @ap: port to resume
4135 * Kernel thread context (may sleep).
4137 static void ata_eh_handle_port_resume(struct ata_port *ap)
4139 struct ata_link *link;
4140 struct ata_device *dev;
4141 unsigned long flags;
4144 /* are we resuming? */
4145 spin_lock_irqsave(ap->lock, flags);
4146 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4147 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4148 spin_unlock_irqrestore(ap->lock, flags);
4151 spin_unlock_irqrestore(ap->lock, flags);
4153 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4156 * Error timestamps are in jiffies which doesn't run while
4157 * suspended and PHY events during resume isn't too uncommon.
4158 * When the two are combined, it can lead to unnecessary speed
4159 * downs if the machine is suspended and resumed repeatedly.
4160 * Clear error history.
4162 ata_for_each_link(link, ap, HOST_FIRST)
4163 ata_for_each_dev(dev, link, ALL)
4164 ata_ering_clear(&dev->ering);
4166 ata_acpi_set_state(ap, ap->pm_mesg);
4168 if (ap->ops->port_resume)
4169 rc = ap->ops->port_resume(ap);
4171 /* tell ACPI that we're resuming */
4172 ata_acpi_on_resume(ap);
4174 /* update the flags */
4175 spin_lock_irqsave(ap->lock, flags);
4176 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4177 spin_unlock_irqrestore(ap->lock, flags);
4179 #endif /* CONFIG_PM */