2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include <scsi/scsi_dbg.h>
44 #include "../scsi/scsi_transport_api.h"
46 #include <linux/libata.h>
51 /* speed down verdicts */
52 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
55 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
58 ATA_EFLAG_IS_IO = (1 << 0),
59 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
61 /* error categories */
64 ATA_ECAT_TOUT_HSM = 2,
66 ATA_ECAT_DUBIOUS_NONE = 4,
67 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
68 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
69 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
72 ATA_EH_CMD_DFL_TIMEOUT = 5000,
74 /* always put at least this amount of time between resets */
75 ATA_EH_RESET_COOL_DOWN = 5000,
77 /* Waiting in ->prereset can never be reliable. It's
78 * sometimes nice to wait there but it can't be depended upon;
79 * otherwise, we wouldn't be resetting. Just give it enough
80 * time for most drives to spin up.
82 ATA_EH_PRERESET_TIMEOUT = 10000,
83 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87 /* probe speed down parameters, see ata_eh_schedule_probe() */
88 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
89 ATA_EH_PROBE_TRIALS = 2,
92 /* The following table determines how we sequence resets. Each entry
93 * represents timeout for that try. The first try can be soft or
94 * hardreset. All others are hardreset if available. In most cases
95 * the first reset w/ 10sec timeout should succeed. Following entries
96 * are mostly for error handling, hotplug and retarded devices.
98 static const unsigned long ata_eh_reset_timeouts[] = {
99 10000, /* most drives spin up by 10sec */
100 10000, /* > 99% working drives spin up before 20sec */
101 35000, /* give > 30 secs of idleness for retarded devices */
102 5000, /* and sweet one last chance */
103 ULONG_MAX, /* > 1 min has elapsed, give up */
106 static const unsigned long ata_eh_identify_timeouts[] = {
107 5000, /* covers > 99% of successes and not too boring on failures */
108 10000, /* combined time till here is enough even for media access */
109 30000, /* for true idiots */
113 static const unsigned long ata_eh_other_timeouts[] = {
114 5000, /* same rationale as identify timeout */
116 /* but no merciful 30sec for other commands, it just isn't worth it */
120 struct ata_eh_cmd_timeout_ent {
122 const unsigned long *timeouts;
125 /* The following table determines timeouts to use for EH internal
126 * commands. Each table entry is a command class and matches the
127 * commands the entry applies to and the timeout table to use.
129 * On the retry after a command timed out, the next timeout value from
130 * the table is used. If the table doesn't contain further entries,
131 * the last value is used.
133 * ehc->cmd_timeout_idx keeps track of which timeout to use per
134 * command class, so if SET_FEATURES times out on the first try, the
135 * next try will use the second timeout value only for that class.
137 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
138 static const struct ata_eh_cmd_timeout_ent
139 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
140 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
141 .timeouts = ata_eh_identify_timeouts, },
142 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_SET_FEATURES),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
149 .timeouts = ata_eh_other_timeouts, },
153 static void __ata_port_freeze(struct ata_port *ap);
155 static void ata_eh_handle_port_suspend(struct ata_port *ap);
156 static void ata_eh_handle_port_resume(struct ata_port *ap);
157 #else /* CONFIG_PM */
158 static void ata_eh_handle_port_suspend(struct ata_port *ap)
161 static void ata_eh_handle_port_resume(struct ata_port *ap)
163 #endif /* CONFIG_PM */
165 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
168 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
169 ATA_EH_DESC_LEN - ehi->desc_len,
174 * __ata_ehi_push_desc - push error description without adding separator
176 * @fmt: printf format string
178 * Format string according to @fmt and append it to @ehi->desc.
181 * spin_lock_irqsave(host lock)
183 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
188 __ata_ehi_pushv_desc(ehi, fmt, args);
193 * ata_ehi_push_desc - push error description with separator
195 * @fmt: printf format string
197 * Format string according to @fmt and append it to @ehi->desc.
198 * If @ehi->desc is not empty, ", " is added in-between.
201 * spin_lock_irqsave(host lock)
203 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
208 __ata_ehi_push_desc(ehi, ", ");
211 __ata_ehi_pushv_desc(ehi, fmt, args);
216 * ata_ehi_clear_desc - clean error description
222 * spin_lock_irqsave(host lock)
224 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
231 * ata_port_desc - append port description
232 * @ap: target ATA port
233 * @fmt: printf format string
235 * Format string according to @fmt and append it to port
236 * description. If port description is not empty, " " is added
237 * in-between. This function is to be used while initializing
238 * ata_host. The description is printed on host registration.
243 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
247 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
249 if (ap->link.eh_info.desc_len)
250 __ata_ehi_push_desc(&ap->link.eh_info, " ");
253 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
260 * ata_port_pbar_desc - append PCI BAR description
261 * @ap: target ATA port
262 * @bar: target PCI BAR
263 * @offset: offset into PCI BAR
264 * @name: name of the area
266 * If @offset is negative, this function formats a string which
267 * contains the name, address, size and type of the BAR and
268 * appends it to the port description. If @offset is zero or
269 * positive, only name and offsetted address is appended.
274 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
277 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
279 unsigned long long start, len;
281 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
283 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
286 start = (unsigned long long)pci_resource_start(pdev, bar);
287 len = (unsigned long long)pci_resource_len(pdev, bar);
290 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
292 ata_port_desc(ap, "%s 0x%llx", name,
293 start + (unsigned long long)offset);
296 #endif /* CONFIG_PCI */
298 static int ata_lookup_timeout_table(u8 cmd)
302 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
305 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
314 * ata_internal_cmd_timeout - determine timeout for an internal command
315 * @dev: target device
316 * @cmd: internal command to be issued
318 * Determine timeout for internal command @cmd for @dev.
324 * Determined timeout.
326 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
328 struct ata_eh_context *ehc = &dev->link->eh_context;
329 int ent = ata_lookup_timeout_table(cmd);
333 return ATA_EH_CMD_DFL_TIMEOUT;
335 idx = ehc->cmd_timeout_idx[dev->devno][ent];
336 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
340 * ata_internal_cmd_timed_out - notification for internal command timeout
341 * @dev: target device
342 * @cmd: internal command which timed out
344 * Notify EH that internal command @cmd for @dev timed out. This
345 * function should be called only for commands whose timeouts are
346 * determined using ata_internal_cmd_timeout().
351 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
353 struct ata_eh_context *ehc = &dev->link->eh_context;
354 int ent = ata_lookup_timeout_table(cmd);
360 idx = ehc->cmd_timeout_idx[dev->devno][ent];
361 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
362 ehc->cmd_timeout_idx[dev->devno][ent]++;
365 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
366 unsigned int err_mask)
368 struct ata_ering_entry *ent;
373 ering->cursor %= ATA_ERING_SIZE;
375 ent = &ering->ring[ering->cursor];
376 ent->eflags = eflags;
377 ent->err_mask = err_mask;
378 ent->timestamp = get_jiffies_64();
381 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
383 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
390 static void ata_ering_clear(struct ata_ering *ering)
392 memset(ering, 0, sizeof(*ering));
395 static int ata_ering_map(struct ata_ering *ering,
396 int (*map_fn)(struct ata_ering_entry *, void *),
400 struct ata_ering_entry *ent;
404 ent = &ering->ring[idx];
407 rc = map_fn(ent, arg);
410 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
411 } while (idx != ering->cursor);
416 static unsigned int ata_eh_dev_action(struct ata_device *dev)
418 struct ata_eh_context *ehc = &dev->link->eh_context;
420 return ehc->i.action | ehc->i.dev_action[dev->devno];
423 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
424 struct ata_eh_info *ehi, unsigned int action)
426 struct ata_device *tdev;
429 ehi->action &= ~action;
430 ata_for_each_dev(tdev, link, ALL)
431 ehi->dev_action[tdev->devno] &= ~action;
433 /* doesn't make sense for port-wide EH actions */
434 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
436 /* break ehi->action into ehi->dev_action */
437 if (ehi->action & action) {
438 ata_for_each_dev(tdev, link, ALL)
439 ehi->dev_action[tdev->devno] |=
440 ehi->action & action;
441 ehi->action &= ~action;
444 /* turn off the specified per-dev action */
445 ehi->dev_action[dev->devno] &= ~action;
450 * ata_scsi_timed_out - SCSI layer time out callback
451 * @cmd: timed out SCSI command
453 * Handles SCSI layer timeout. We race with normal completion of
454 * the qc for @cmd. If the qc is already gone, we lose and let
455 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
456 * timed out and EH should be invoked. Prevent ata_qc_complete()
457 * from finishing it by setting EH_SCHEDULED and return
460 * TODO: kill this function once old EH is gone.
463 * Called from timer context
466 * EH_HANDLED or EH_NOT_HANDLED
468 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
470 struct Scsi_Host *host = cmd->device->host;
471 struct ata_port *ap = ata_shost_to_port(host);
473 struct ata_queued_cmd *qc;
474 enum blk_eh_timer_return ret;
478 if (ap->ops->error_handler) {
479 ret = BLK_EH_NOT_HANDLED;
483 ret = BLK_EH_HANDLED;
484 spin_lock_irqsave(ap->lock, flags);
485 qc = ata_qc_from_tag(ap, ap->link.active_tag);
487 WARN_ON(qc->scsicmd != cmd);
488 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
489 qc->err_mask |= AC_ERR_TIMEOUT;
490 ret = BLK_EH_NOT_HANDLED;
492 spin_unlock_irqrestore(ap->lock, flags);
495 DPRINTK("EXIT, ret=%d\n", ret);
499 static void ata_eh_unload(struct ata_port *ap)
501 struct ata_link *link;
502 struct ata_device *dev;
505 /* Restore SControl IPM and SPD for the next driver and
506 * disable attached devices.
508 ata_for_each_link(link, ap, PMP_FIRST) {
509 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
510 ata_for_each_dev(dev, link, ALL)
511 ata_dev_disable(dev);
514 /* freeze and set UNLOADED */
515 spin_lock_irqsave(ap->lock, flags);
517 ata_port_freeze(ap); /* won't be thawed */
518 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
519 ap->pflags |= ATA_PFLAG_UNLOADED;
521 spin_unlock_irqrestore(ap->lock, flags);
525 * ata_scsi_error - SCSI layer error handler callback
526 * @host: SCSI host on which error occurred
528 * Handles SCSI-layer-thrown error events.
531 * Inherited from SCSI layer (none, can sleep)
536 void ata_scsi_error(struct Scsi_Host *host)
538 struct ata_port *ap = ata_shost_to_port(host);
544 /* synchronize with port task */
545 ata_port_flush_task(ap);
547 /* synchronize with host lock and sort out timeouts */
549 /* For new EH, all qcs are finished in one of three ways -
550 * normal completion, error completion, and SCSI timeout.
551 * Both completions can race against SCSI timeout. When normal
552 * completion wins, the qc never reaches EH. When error
553 * completion wins, the qc has ATA_QCFLAG_FAILED set.
555 * When SCSI timeout wins, things are a bit more complex.
556 * Normal or error completion can occur after the timeout but
557 * before this point. In such cases, both types of
558 * completions are honored. A scmd is determined to have
559 * timed out iff its associated qc is active and not failed.
561 if (ap->ops->error_handler) {
562 struct scsi_cmnd *scmd, *tmp;
565 spin_lock_irqsave(ap->lock, flags);
567 /* This must occur under the ap->lock as we don't want
568 a polled recovery to race the real interrupt handler
570 The lost_interrupt handler checks for any completed but
571 non-notified command and completes much like an IRQ handler.
573 We then fall into the error recovery code which will treat
574 this as if normal completion won the race */
576 if (ap->ops->lost_interrupt)
577 ap->ops->lost_interrupt(ap);
579 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
580 struct ata_queued_cmd *qc;
582 for (i = 0; i < ATA_MAX_QUEUE; i++) {
583 qc = __ata_qc_from_tag(ap, i);
584 if (qc->flags & ATA_QCFLAG_ACTIVE &&
589 if (i < ATA_MAX_QUEUE) {
590 /* the scmd has an associated qc */
591 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
592 /* which hasn't failed yet, timeout */
593 qc->err_mask |= AC_ERR_TIMEOUT;
594 qc->flags |= ATA_QCFLAG_FAILED;
598 /* Normal completion occurred after
599 * SCSI timeout but before this point.
600 * Successfully complete it.
602 scmd->retries = scmd->allowed;
603 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
607 /* If we have timed out qcs. They belong to EH from
608 * this point but the state of the controller is
609 * unknown. Freeze the port to make sure the IRQ
610 * handler doesn't diddle with those qcs. This must
611 * be done atomically w.r.t. setting QCFLAG_FAILED.
614 __ata_port_freeze(ap);
616 spin_unlock_irqrestore(ap->lock, flags);
618 /* initialize eh_tries */
619 ap->eh_tries = ATA_EH_MAX_TRIES;
621 spin_unlock_wait(ap->lock);
623 /* If we timed raced normal completion and there is nothing to
624 recover nr_timedout == 0 why exactly are we doing error recovery ? */
627 /* invoke error handler */
628 if (ap->ops->error_handler) {
629 struct ata_link *link;
631 /* kill fast drain timer */
632 del_timer_sync(&ap->fastdrain_timer);
634 /* process port resume request */
635 ata_eh_handle_port_resume(ap);
637 /* fetch & clear EH info */
638 spin_lock_irqsave(ap->lock, flags);
640 ata_for_each_link(link, ap, HOST_FIRST) {
641 struct ata_eh_context *ehc = &link->eh_context;
642 struct ata_device *dev;
644 memset(&link->eh_context, 0, sizeof(link->eh_context));
645 link->eh_context.i = link->eh_info;
646 memset(&link->eh_info, 0, sizeof(link->eh_info));
648 ata_for_each_dev(dev, link, ENABLED) {
649 int devno = dev->devno;
651 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
652 if (ata_ncq_enabled(dev))
653 ehc->saved_ncq_enabled |= 1 << devno;
657 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
658 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
659 ap->excl_link = NULL; /* don't maintain exclusion over EH */
661 spin_unlock_irqrestore(ap->lock, flags);
663 /* invoke EH, skip if unloading or suspended */
664 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
665 ap->ops->error_handler(ap);
667 /* if unloading, commence suicide */
668 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
669 !(ap->pflags & ATA_PFLAG_UNLOADED))
674 /* process port suspend request */
675 ata_eh_handle_port_suspend(ap);
677 /* Exception might have happend after ->error_handler
678 * recovered the port but before this point. Repeat
681 spin_lock_irqsave(ap->lock, flags);
683 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
684 if (--ap->eh_tries) {
685 spin_unlock_irqrestore(ap->lock, flags);
688 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
689 "tries, giving up\n", ATA_EH_MAX_TRIES);
690 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
693 /* this run is complete, make sure EH info is clear */
694 ata_for_each_link(link, ap, HOST_FIRST)
695 memset(&link->eh_info, 0, sizeof(link->eh_info));
697 /* Clear host_eh_scheduled while holding ap->lock such
698 * that if exception occurs after this point but
699 * before EH completion, SCSI midlayer will
702 host->host_eh_scheduled = 0;
704 spin_unlock_irqrestore(ap->lock, flags);
706 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
707 ap->ops->eng_timeout(ap);
710 /* finish or retry handled scmd's and clean up */
711 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
713 scsi_eh_flush_done_q(&ap->eh_done_q);
716 spin_lock_irqsave(ap->lock, flags);
718 if (ap->pflags & ATA_PFLAG_LOADING)
719 ap->pflags &= ~ATA_PFLAG_LOADING;
720 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
721 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
723 if (ap->pflags & ATA_PFLAG_RECOVERED)
724 ata_port_printk(ap, KERN_INFO, "EH complete\n");
726 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
728 /* tell wait_eh that we're done */
729 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
730 wake_up_all(&ap->eh_wait_q);
732 spin_unlock_irqrestore(ap->lock, flags);
738 * ata_port_wait_eh - Wait for the currently pending EH to complete
739 * @ap: Port to wait EH for
741 * Wait until the currently pending EH is complete.
744 * Kernel thread context (may sleep).
746 void ata_port_wait_eh(struct ata_port *ap)
752 spin_lock_irqsave(ap->lock, flags);
754 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
755 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
756 spin_unlock_irqrestore(ap->lock, flags);
758 spin_lock_irqsave(ap->lock, flags);
760 finish_wait(&ap->eh_wait_q, &wait);
762 spin_unlock_irqrestore(ap->lock, flags);
764 /* make sure SCSI EH is complete */
765 if (scsi_host_in_recovery(ap->scsi_host)) {
771 static int ata_eh_nr_in_flight(struct ata_port *ap)
776 /* count only non-internal commands */
777 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
778 if (ata_qc_from_tag(ap, tag))
784 void ata_eh_fastdrain_timerfn(unsigned long arg)
786 struct ata_port *ap = (void *)arg;
790 spin_lock_irqsave(ap->lock, flags);
792 cnt = ata_eh_nr_in_flight(ap);
798 if (cnt == ap->fastdrain_cnt) {
801 /* No progress during the last interval, tag all
802 * in-flight qcs as timed out and freeze the port.
804 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
805 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
807 qc->err_mask |= AC_ERR_TIMEOUT;
812 /* some qcs have finished, give it another chance */
813 ap->fastdrain_cnt = cnt;
814 ap->fastdrain_timer.expires =
815 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
816 add_timer(&ap->fastdrain_timer);
820 spin_unlock_irqrestore(ap->lock, flags);
824 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
825 * @ap: target ATA port
826 * @fastdrain: activate fast drain
828 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
829 * is non-zero and EH wasn't pending before. Fast drain ensures
830 * that EH kicks in in timely manner.
833 * spin_lock_irqsave(host lock)
835 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
839 /* already scheduled? */
840 if (ap->pflags & ATA_PFLAG_EH_PENDING)
843 ap->pflags |= ATA_PFLAG_EH_PENDING;
848 /* do we have in-flight qcs? */
849 cnt = ata_eh_nr_in_flight(ap);
853 /* activate fast drain */
854 ap->fastdrain_cnt = cnt;
855 ap->fastdrain_timer.expires =
856 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
857 add_timer(&ap->fastdrain_timer);
861 * ata_qc_schedule_eh - schedule qc for error handling
862 * @qc: command to schedule error handling for
864 * Schedule error handling for @qc. EH will kick in as soon as
865 * other commands are drained.
868 * spin_lock_irqsave(host lock)
870 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
872 struct ata_port *ap = qc->ap;
874 WARN_ON(!ap->ops->error_handler);
876 qc->flags |= ATA_QCFLAG_FAILED;
877 ata_eh_set_pending(ap, 1);
879 /* The following will fail if timeout has already expired.
880 * ata_scsi_error() takes care of such scmds on EH entry.
881 * Note that ATA_QCFLAG_FAILED is unconditionally set after
882 * this function completes.
884 blk_abort_request(qc->scsicmd->request);
888 * ata_port_schedule_eh - schedule error handling without a qc
889 * @ap: ATA port to schedule EH for
891 * Schedule error handling for @ap. EH will kick in as soon as
892 * all commands are drained.
895 * spin_lock_irqsave(host lock)
897 void ata_port_schedule_eh(struct ata_port *ap)
899 WARN_ON(!ap->ops->error_handler);
901 if (ap->pflags & ATA_PFLAG_INITIALIZING)
904 ata_eh_set_pending(ap, 1);
905 scsi_schedule_eh(ap->scsi_host);
907 DPRINTK("port EH scheduled\n");
910 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
912 int tag, nr_aborted = 0;
914 WARN_ON(!ap->ops->error_handler);
916 /* we're gonna abort all commands, no need for fast drain */
917 ata_eh_set_pending(ap, 0);
919 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
920 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
922 if (qc && (!link || qc->dev->link == link)) {
923 qc->flags |= ATA_QCFLAG_FAILED;
930 ata_port_schedule_eh(ap);
936 * ata_link_abort - abort all qc's on the link
937 * @link: ATA link to abort qc's for
939 * Abort all active qc's active on @link and schedule EH.
942 * spin_lock_irqsave(host lock)
945 * Number of aborted qc's.
947 int ata_link_abort(struct ata_link *link)
949 return ata_do_link_abort(link->ap, link);
953 * ata_port_abort - abort all qc's on the port
954 * @ap: ATA port to abort qc's for
956 * Abort all active qc's of @ap and schedule EH.
959 * spin_lock_irqsave(host_set lock)
962 * Number of aborted qc's.
964 int ata_port_abort(struct ata_port *ap)
966 return ata_do_link_abort(ap, NULL);
970 * __ata_port_freeze - freeze port
971 * @ap: ATA port to freeze
973 * This function is called when HSM violation or some other
974 * condition disrupts normal operation of the port. Frozen port
975 * is not allowed to perform any operation until the port is
976 * thawed, which usually follows a successful reset.
978 * ap->ops->freeze() callback can be used for freezing the port
979 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
980 * port cannot be frozen hardware-wise, the interrupt handler
981 * must ack and clear interrupts unconditionally while the port
985 * spin_lock_irqsave(host lock)
987 static void __ata_port_freeze(struct ata_port *ap)
989 WARN_ON(!ap->ops->error_handler);
994 ap->pflags |= ATA_PFLAG_FROZEN;
996 DPRINTK("ata%u port frozen\n", ap->print_id);
1000 * ata_port_freeze - abort & freeze port
1001 * @ap: ATA port to freeze
1003 * Abort and freeze @ap. The freeze operation must be called
1004 * first, because some hardware requires special operations
1005 * before the taskfile registers are accessible.
1008 * spin_lock_irqsave(host lock)
1011 * Number of aborted commands.
1013 int ata_port_freeze(struct ata_port *ap)
1017 WARN_ON(!ap->ops->error_handler);
1019 __ata_port_freeze(ap);
1020 nr_aborted = ata_port_abort(ap);
1026 * sata_async_notification - SATA async notification handler
1027 * @ap: ATA port where async notification is received
1029 * Handler to be called when async notification via SDB FIS is
1030 * received. This function schedules EH if necessary.
1033 * spin_lock_irqsave(host lock)
1036 * 1 if EH is scheduled, 0 otherwise.
1038 int sata_async_notification(struct ata_port *ap)
1043 if (!(ap->flags & ATA_FLAG_AN))
1046 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1048 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1050 if (!sata_pmp_attached(ap) || rc) {
1051 /* PMP is not attached or SNTF is not available */
1052 if (!sata_pmp_attached(ap)) {
1053 /* PMP is not attached. Check whether ATAPI
1054 * AN is configured. If so, notify media
1057 struct ata_device *dev = ap->link.device;
1059 if ((dev->class == ATA_DEV_ATAPI) &&
1060 (dev->flags & ATA_DFLAG_AN))
1061 ata_scsi_media_change_notify(dev);
1064 /* PMP is attached but SNTF is not available.
1065 * ATAPI async media change notification is
1066 * not used. The PMP must be reporting PHY
1067 * status change, schedule EH.
1069 ata_port_schedule_eh(ap);
1073 /* PMP is attached and SNTF is available */
1074 struct ata_link *link;
1076 /* check and notify ATAPI AN */
1077 ata_for_each_link(link, ap, EDGE) {
1078 if (!(sntf & (1 << link->pmp)))
1081 if ((link->device->class == ATA_DEV_ATAPI) &&
1082 (link->device->flags & ATA_DFLAG_AN))
1083 ata_scsi_media_change_notify(link->device);
1086 /* If PMP is reporting that PHY status of some
1087 * downstream ports has changed, schedule EH.
1089 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1090 ata_port_schedule_eh(ap);
1099 * ata_eh_freeze_port - EH helper to freeze port
1100 * @ap: ATA port to freeze
1107 void ata_eh_freeze_port(struct ata_port *ap)
1109 unsigned long flags;
1111 if (!ap->ops->error_handler)
1114 spin_lock_irqsave(ap->lock, flags);
1115 __ata_port_freeze(ap);
1116 spin_unlock_irqrestore(ap->lock, flags);
1120 * ata_port_thaw_port - EH helper to thaw port
1121 * @ap: ATA port to thaw
1123 * Thaw frozen port @ap.
1128 void ata_eh_thaw_port(struct ata_port *ap)
1130 unsigned long flags;
1132 if (!ap->ops->error_handler)
1135 spin_lock_irqsave(ap->lock, flags);
1137 ap->pflags &= ~ATA_PFLAG_FROZEN;
1142 spin_unlock_irqrestore(ap->lock, flags);
1144 DPRINTK("ata%u port thawed\n", ap->print_id);
1147 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1152 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1154 struct ata_port *ap = qc->ap;
1155 struct scsi_cmnd *scmd = qc->scsicmd;
1156 unsigned long flags;
1158 spin_lock_irqsave(ap->lock, flags);
1159 qc->scsidone = ata_eh_scsidone;
1160 __ata_qc_complete(qc);
1161 WARN_ON(ata_tag_valid(qc->tag));
1162 spin_unlock_irqrestore(ap->lock, flags);
1164 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1168 * ata_eh_qc_complete - Complete an active ATA command from EH
1169 * @qc: Command to complete
1171 * Indicate to the mid and upper layers that an ATA command has
1172 * completed. To be used from EH.
1174 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1176 struct scsi_cmnd *scmd = qc->scsicmd;
1177 scmd->retries = scmd->allowed;
1178 __ata_eh_qc_complete(qc);
1182 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1183 * @qc: Command to retry
1185 * Indicate to the mid and upper layers that an ATA command
1186 * should be retried. To be used from EH.
1188 * SCSI midlayer limits the number of retries to scmd->allowed.
1189 * scmd->retries is decremented for commands which get retried
1190 * due to unrelated failures (qc->err_mask is zero).
1192 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1194 struct scsi_cmnd *scmd = qc->scsicmd;
1195 if (!qc->err_mask && scmd->retries)
1197 __ata_eh_qc_complete(qc);
1201 * ata_dev_disable - disable ATA device
1202 * @dev: ATA device to disable
1209 void ata_dev_disable(struct ata_device *dev)
1211 if (!ata_dev_enabled(dev))
1214 if (ata_msg_drv(dev->link->ap))
1215 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1216 ata_acpi_on_disable(dev);
1217 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1220 /* From now till the next successful probe, ering is used to
1221 * track probe failures. Clear accumulated device error info.
1223 ata_ering_clear(&dev->ering);
1227 * ata_eh_detach_dev - detach ATA device
1228 * @dev: ATA device to detach
1235 void ata_eh_detach_dev(struct ata_device *dev)
1237 struct ata_link *link = dev->link;
1238 struct ata_port *ap = link->ap;
1239 struct ata_eh_context *ehc = &link->eh_context;
1240 unsigned long flags;
1242 ata_dev_disable(dev);
1244 spin_lock_irqsave(ap->lock, flags);
1246 dev->flags &= ~ATA_DFLAG_DETACH;
1248 if (ata_scsi_offline_dev(dev)) {
1249 dev->flags |= ATA_DFLAG_DETACHED;
1250 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1253 /* clear per-dev EH info */
1254 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1255 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1256 ehc->saved_xfer_mode[dev->devno] = 0;
1257 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1259 spin_unlock_irqrestore(ap->lock, flags);
1263 * ata_eh_about_to_do - about to perform eh_action
1264 * @link: target ATA link
1265 * @dev: target ATA dev for per-dev action (can be NULL)
1266 * @action: action about to be performed
1268 * Called just before performing EH actions to clear related bits
1269 * in @link->eh_info such that eh actions are not unnecessarily
1275 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1276 unsigned int action)
1278 struct ata_port *ap = link->ap;
1279 struct ata_eh_info *ehi = &link->eh_info;
1280 struct ata_eh_context *ehc = &link->eh_context;
1281 unsigned long flags;
1283 spin_lock_irqsave(ap->lock, flags);
1285 ata_eh_clear_action(link, dev, ehi, action);
1287 /* About to take EH action, set RECOVERED. Ignore actions on
1288 * slave links as master will do them again.
1290 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1291 ap->pflags |= ATA_PFLAG_RECOVERED;
1293 spin_unlock_irqrestore(ap->lock, flags);
1297 * ata_eh_done - EH action complete
1298 * @ap: target ATA port
1299 * @dev: target ATA dev for per-dev action (can be NULL)
1300 * @action: action just completed
1302 * Called right after performing EH actions to clear related bits
1303 * in @link->eh_context.
1308 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1309 unsigned int action)
1311 struct ata_eh_context *ehc = &link->eh_context;
1313 ata_eh_clear_action(link, dev, &ehc->i, action);
1317 * ata_err_string - convert err_mask to descriptive string
1318 * @err_mask: error mask to convert to string
1320 * Convert @err_mask to descriptive string. Errors are
1321 * prioritized according to severity and only the most severe
1322 * error is reported.
1328 * Descriptive string for @err_mask
1330 static const char *ata_err_string(unsigned int err_mask)
1332 if (err_mask & AC_ERR_HOST_BUS)
1333 return "host bus error";
1334 if (err_mask & AC_ERR_ATA_BUS)
1335 return "ATA bus error";
1336 if (err_mask & AC_ERR_TIMEOUT)
1338 if (err_mask & AC_ERR_HSM)
1339 return "HSM violation";
1340 if (err_mask & AC_ERR_SYSTEM)
1341 return "internal error";
1342 if (err_mask & AC_ERR_MEDIA)
1343 return "media error";
1344 if (err_mask & AC_ERR_INVALID)
1345 return "invalid argument";
1346 if (err_mask & AC_ERR_DEV)
1347 return "device error";
1348 return "unknown error";
1352 * ata_read_log_page - read a specific log page
1353 * @dev: target device
1354 * @page: page to read
1355 * @buf: buffer to store read page
1356 * @sectors: number of sectors to read
1358 * Read log page using READ_LOG_EXT command.
1361 * Kernel thread context (may sleep).
1364 * 0 on success, AC_ERR_* mask otherwise.
1366 static unsigned int ata_read_log_page(struct ata_device *dev,
1367 u8 page, void *buf, unsigned int sectors)
1369 struct ata_taskfile tf;
1370 unsigned int err_mask;
1372 DPRINTK("read log page - page %d\n", page);
1374 ata_tf_init(dev, &tf);
1375 tf.command = ATA_CMD_READ_LOG_EXT;
1378 tf.hob_nsect = sectors >> 8;
1379 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1380 tf.protocol = ATA_PROT_PIO;
1382 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1383 buf, sectors * ATA_SECT_SIZE, 0);
1385 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1390 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1391 * @dev: Device to read log page 10h from
1392 * @tag: Resulting tag of the failed command
1393 * @tf: Resulting taskfile registers of the failed command
1395 * Read log page 10h to obtain NCQ error details and clear error
1399 * Kernel thread context (may sleep).
1402 * 0 on success, -errno otherwise.
1404 static int ata_eh_read_log_10h(struct ata_device *dev,
1405 int *tag, struct ata_taskfile *tf)
1407 u8 *buf = dev->link->ap->sector_buf;
1408 unsigned int err_mask;
1412 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1417 for (i = 0; i < ATA_SECT_SIZE; i++)
1420 ata_dev_printk(dev, KERN_WARNING,
1421 "invalid checksum 0x%x on log page 10h\n", csum);
1426 *tag = buf[0] & 0x1f;
1428 tf->command = buf[2];
1429 tf->feature = buf[3];
1433 tf->device = buf[7];
1434 tf->hob_lbal = buf[8];
1435 tf->hob_lbam = buf[9];
1436 tf->hob_lbah = buf[10];
1437 tf->nsect = buf[12];
1438 tf->hob_nsect = buf[13];
1444 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1445 * @dev: target ATAPI device
1446 * @r_sense_key: out parameter for sense_key
1448 * Perform ATAPI TEST_UNIT_READY.
1451 * EH context (may sleep).
1454 * 0 on success, AC_ERR_* mask on failure.
1456 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1458 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1459 struct ata_taskfile tf;
1460 unsigned int err_mask;
1462 ata_tf_init(dev, &tf);
1464 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1465 tf.command = ATA_CMD_PACKET;
1466 tf.protocol = ATAPI_PROT_NODATA;
1468 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1469 if (err_mask == AC_ERR_DEV)
1470 *r_sense_key = tf.feature >> 4;
1475 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1476 * @dev: device to perform REQUEST_SENSE to
1477 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1478 * @dfl_sense_key: default sense key to use
1480 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1481 * SENSE. This function is EH helper.
1484 * Kernel thread context (may sleep).
1487 * 0 on success, AC_ERR_* mask on failure
1489 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1490 u8 *sense_buf, u8 dfl_sense_key)
1492 u8 cdb[ATAPI_CDB_LEN] =
1493 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1494 struct ata_port *ap = dev->link->ap;
1495 struct ata_taskfile tf;
1497 DPRINTK("ATAPI request sense\n");
1499 /* FIXME: is this needed? */
1500 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1502 /* initialize sense_buf with the error register,
1503 * for the case where they are -not- overwritten
1505 sense_buf[0] = 0x70;
1506 sense_buf[2] = dfl_sense_key;
1508 /* some devices time out if garbage left in tf */
1509 ata_tf_init(dev, &tf);
1511 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1512 tf.command = ATA_CMD_PACKET;
1514 /* is it pointless to prefer PIO for "safety reasons"? */
1515 if (ap->flags & ATA_FLAG_PIO_DMA) {
1516 tf.protocol = ATAPI_PROT_DMA;
1517 tf.feature |= ATAPI_PKT_DMA;
1519 tf.protocol = ATAPI_PROT_PIO;
1520 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1524 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1525 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1529 * ata_eh_analyze_serror - analyze SError for a failed port
1530 * @link: ATA link to analyze SError for
1532 * Analyze SError if available and further determine cause of
1538 static void ata_eh_analyze_serror(struct ata_link *link)
1540 struct ata_eh_context *ehc = &link->eh_context;
1541 u32 serror = ehc->i.serror;
1542 unsigned int err_mask = 0, action = 0;
1545 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1546 err_mask |= AC_ERR_ATA_BUS;
1547 action |= ATA_EH_RESET;
1549 if (serror & SERR_PROTOCOL) {
1550 err_mask |= AC_ERR_HSM;
1551 action |= ATA_EH_RESET;
1553 if (serror & SERR_INTERNAL) {
1554 err_mask |= AC_ERR_SYSTEM;
1555 action |= ATA_EH_RESET;
1558 /* Determine whether a hotplug event has occurred. Both
1559 * SError.N/X are considered hotplug events for enabled or
1560 * host links. For disabled PMP links, only N bit is
1561 * considered as X bit is left at 1 for link plugging.
1565 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1566 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1568 hotplug_mask = SERR_PHYRDY_CHG;
1570 if (serror & hotplug_mask)
1571 ata_ehi_hotplugged(&ehc->i);
1573 ehc->i.err_mask |= err_mask;
1574 ehc->i.action |= action;
1578 * ata_eh_analyze_ncq_error - analyze NCQ error
1579 * @link: ATA link to analyze NCQ error for
1581 * Read log page 10h, determine the offending qc and acquire
1582 * error status TF. For NCQ device errors, all LLDDs have to do
1583 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1587 * Kernel thread context (may sleep).
1589 void ata_eh_analyze_ncq_error(struct ata_link *link)
1591 struct ata_port *ap = link->ap;
1592 struct ata_eh_context *ehc = &link->eh_context;
1593 struct ata_device *dev = link->device;
1594 struct ata_queued_cmd *qc;
1595 struct ata_taskfile tf;
1598 /* if frozen, we can't do much */
1599 if (ap->pflags & ATA_PFLAG_FROZEN)
1602 /* is it NCQ device error? */
1603 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1606 /* has LLDD analyzed already? */
1607 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1608 qc = __ata_qc_from_tag(ap, tag);
1610 if (!(qc->flags & ATA_QCFLAG_FAILED))
1617 /* okay, this error is ours */
1618 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1620 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1621 "(errno=%d)\n", rc);
1625 if (!(link->sactive & (1 << tag))) {
1626 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1627 "inactive tag %d\n", tag);
1631 /* we've got the perpetrator, condemn it */
1632 qc = __ata_qc_from_tag(ap, tag);
1633 memcpy(&qc->result_tf, &tf, sizeof(tf));
1634 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1635 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1636 ehc->i.err_mask &= ~AC_ERR_DEV;
1640 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1641 * @qc: qc to analyze
1642 * @tf: Taskfile registers to analyze
1644 * Analyze taskfile of @qc and further determine cause of
1645 * failure. This function also requests ATAPI sense data if
1649 * Kernel thread context (may sleep).
1652 * Determined recovery action
1654 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1655 const struct ata_taskfile *tf)
1657 unsigned int tmp, action = 0;
1658 u8 stat = tf->command, err = tf->feature;
1660 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1661 qc->err_mask |= AC_ERR_HSM;
1662 return ATA_EH_RESET;
1665 if (stat & (ATA_ERR | ATA_DF))
1666 qc->err_mask |= AC_ERR_DEV;
1670 switch (qc->dev->class) {
1673 qc->err_mask |= AC_ERR_ATA_BUS;
1675 qc->err_mask |= AC_ERR_MEDIA;
1677 qc->err_mask |= AC_ERR_INVALID;
1681 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1682 tmp = atapi_eh_request_sense(qc->dev,
1683 qc->scsicmd->sense_buffer,
1684 qc->result_tf.feature >> 4);
1686 /* ATA_QCFLAG_SENSE_VALID is used to
1687 * tell atapi_qc_complete() that sense
1688 * data is already valid.
1690 * TODO: interpret sense data and set
1691 * appropriate err_mask.
1693 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1695 qc->err_mask |= tmp;
1699 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1700 action |= ATA_EH_RESET;
1705 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1710 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1714 base = ATA_ECAT_DUBIOUS_NONE;
1716 if (err_mask & AC_ERR_ATA_BUS)
1717 return base + ATA_ECAT_ATA_BUS;
1719 if (err_mask & AC_ERR_TIMEOUT)
1720 return base + ATA_ECAT_TOUT_HSM;
1722 if (eflags & ATA_EFLAG_IS_IO) {
1723 if (err_mask & AC_ERR_HSM)
1724 return base + ATA_ECAT_TOUT_HSM;
1726 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1727 return base + ATA_ECAT_UNK_DEV;
1733 struct speed_down_verdict_arg {
1736 int nr_errors[ATA_ECAT_NR];
1739 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1741 struct speed_down_verdict_arg *arg = void_arg;
1744 if (ent->timestamp < arg->since)
1747 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1749 arg->nr_errors[cat]++;
1755 * ata_eh_speed_down_verdict - Determine speed down verdict
1756 * @dev: Device of interest
1758 * This function examines error ring of @dev and determines
1759 * whether NCQ needs to be turned off, transfer speed should be
1760 * stepped down, or falling back to PIO is necessary.
1762 * ECAT_ATA_BUS : ATA_BUS error for any command
1764 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1767 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1769 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1770 * data transfer hasn't been verified.
1774 * NCQ_OFF : Turn off NCQ.
1776 * SPEED_DOWN : Speed down transfer speed but don't fall back
1779 * FALLBACK_TO_PIO : Fall back to PIO.
1781 * Even if multiple verdicts are returned, only one action is
1782 * taken per error. An action triggered by non-DUBIOUS errors
1783 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1784 * This is to expedite speed down decisions right after device is
1785 * initially configured.
1787 * The followings are speed down rules. #1 and #2 deal with
1790 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1791 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1793 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1794 * occurred during last 5 mins, NCQ_OFF.
1796 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1797 * ocurred during last 5 mins, FALLBACK_TO_PIO
1799 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1800 * during last 10 mins, NCQ_OFF.
1802 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1803 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1806 * Inherited from caller.
1809 * OR of ATA_EH_SPDN_* flags.
1811 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1813 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1814 u64 j64 = get_jiffies_64();
1815 struct speed_down_verdict_arg arg;
1816 unsigned int verdict = 0;
1818 /* scan past 5 mins of error history */
1819 memset(&arg, 0, sizeof(arg));
1820 arg.since = j64 - min(j64, j5mins);
1821 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1823 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1824 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1825 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1826 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1828 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1829 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1830 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1832 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1833 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1834 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1835 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1837 /* scan past 10 mins of error history */
1838 memset(&arg, 0, sizeof(arg));
1839 arg.since = j64 - min(j64, j10mins);
1840 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1842 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1843 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1844 verdict |= ATA_EH_SPDN_NCQ_OFF;
1846 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1847 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1848 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1849 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1855 * ata_eh_speed_down - record error and speed down if necessary
1856 * @dev: Failed device
1857 * @eflags: mask of ATA_EFLAG_* flags
1858 * @err_mask: err_mask of the error
1860 * Record error and examine error history to determine whether
1861 * adjusting transmission speed is necessary. It also sets
1862 * transmission limits appropriately if such adjustment is
1866 * Kernel thread context (may sleep).
1869 * Determined recovery action.
1871 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1872 unsigned int eflags, unsigned int err_mask)
1874 struct ata_link *link = ata_dev_phys_link(dev);
1876 unsigned int verdict;
1877 unsigned int action = 0;
1879 /* don't bother if Cat-0 error */
1880 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1883 /* record error and determine whether speed down is necessary */
1884 ata_ering_record(&dev->ering, eflags, err_mask);
1885 verdict = ata_eh_speed_down_verdict(dev);
1888 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1889 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1890 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1891 dev->flags |= ATA_DFLAG_NCQ_OFF;
1892 ata_dev_printk(dev, KERN_WARNING,
1893 "NCQ disabled due to excessive errors\n");
1898 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1899 /* speed down SATA link speed if possible */
1900 if (sata_down_spd_limit(link, 0) == 0) {
1901 action |= ATA_EH_RESET;
1905 /* lower transfer mode */
1906 if (dev->spdn_cnt < 2) {
1907 static const int dma_dnxfer_sel[] =
1908 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1909 static const int pio_dnxfer_sel[] =
1910 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1913 if (dev->xfer_shift != ATA_SHIFT_PIO)
1914 sel = dma_dnxfer_sel[dev->spdn_cnt];
1916 sel = pio_dnxfer_sel[dev->spdn_cnt];
1920 if (ata_down_xfermask_limit(dev, sel) == 0) {
1921 action |= ATA_EH_RESET;
1927 /* Fall back to PIO? Slowing down to PIO is meaningless for
1928 * SATA ATA devices. Consider it only for PATA and SATAPI.
1930 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1931 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1932 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1933 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1935 action |= ATA_EH_RESET;
1942 /* device has been slowed down, blow error history */
1943 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1944 ata_ering_clear(&dev->ering);
1949 * ata_eh_link_autopsy - analyze error and determine recovery action
1950 * @link: host link to perform autopsy on
1952 * Analyze why @link failed and determine which recovery actions
1953 * are needed. This function also sets more detailed AC_ERR_*
1954 * values and fills sense data for ATAPI CHECK SENSE.
1957 * Kernel thread context (may sleep).
1959 static void ata_eh_link_autopsy(struct ata_link *link)
1961 struct ata_port *ap = link->ap;
1962 struct ata_eh_context *ehc = &link->eh_context;
1963 struct ata_device *dev;
1964 unsigned int all_err_mask = 0, eflags = 0;
1971 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1974 /* obtain and analyze SError */
1975 rc = sata_scr_read(link, SCR_ERROR, &serror);
1977 ehc->i.serror |= serror;
1978 ata_eh_analyze_serror(link);
1979 } else if (rc != -EOPNOTSUPP) {
1980 /* SError read failed, force reset and probing */
1981 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1982 ehc->i.action |= ATA_EH_RESET;
1983 ehc->i.err_mask |= AC_ERR_OTHER;
1986 /* analyze NCQ failure */
1987 ata_eh_analyze_ncq_error(link);
1989 /* any real error trumps AC_ERR_OTHER */
1990 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1991 ehc->i.err_mask &= ~AC_ERR_OTHER;
1993 all_err_mask |= ehc->i.err_mask;
1995 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1996 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1998 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1999 ata_dev_phys_link(qc->dev) != link)
2002 /* inherit upper level err_mask */
2003 qc->err_mask |= ehc->i.err_mask;
2006 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2008 /* DEV errors are probably spurious in case of ATA_BUS error */
2009 if (qc->err_mask & AC_ERR_ATA_BUS)
2010 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2013 /* any real error trumps unknown error */
2014 if (qc->err_mask & ~AC_ERR_OTHER)
2015 qc->err_mask &= ~AC_ERR_OTHER;
2017 /* SENSE_VALID trumps dev/unknown error and revalidation */
2018 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2019 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2021 /* determine whether the command is worth retrying */
2022 if (!(qc->err_mask & AC_ERR_INVALID) &&
2023 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
2024 qc->flags |= ATA_QCFLAG_RETRY;
2026 /* accumulate error info */
2027 ehc->i.dev = qc->dev;
2028 all_err_mask |= qc->err_mask;
2029 if (qc->flags & ATA_QCFLAG_IO)
2030 eflags |= ATA_EFLAG_IS_IO;
2033 /* enforce default EH actions */
2034 if (ap->pflags & ATA_PFLAG_FROZEN ||
2035 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2036 ehc->i.action |= ATA_EH_RESET;
2037 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2038 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2039 ehc->i.action |= ATA_EH_REVALIDATE;
2041 /* If we have offending qcs and the associated failed device,
2042 * perform per-dev EH action only on the offending device.
2045 ehc->i.dev_action[ehc->i.dev->devno] |=
2046 ehc->i.action & ATA_EH_PERDEV_MASK;
2047 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2050 /* propagate timeout to host link */
2051 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2052 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2054 /* record error and consider speeding down */
2056 if (!dev && ((ata_link_max_devices(link) == 1 &&
2057 ata_dev_enabled(link->device))))
2061 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2062 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2063 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2070 * ata_eh_autopsy - analyze error and determine recovery action
2071 * @ap: host port to perform autopsy on
2073 * Analyze all links of @ap and determine why they failed and
2074 * which recovery actions are needed.
2077 * Kernel thread context (may sleep).
2079 void ata_eh_autopsy(struct ata_port *ap)
2081 struct ata_link *link;
2083 ata_for_each_link(link, ap, EDGE)
2084 ata_eh_link_autopsy(link);
2086 /* Handle the frigging slave link. Autopsy is done similarly
2087 * but actions and flags are transferred over to the master
2088 * link and handled from there.
2090 if (ap->slave_link) {
2091 struct ata_eh_context *mehc = &ap->link.eh_context;
2092 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2094 /* transfer control flags from master to slave */
2095 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2097 /* perform autopsy on the slave link */
2098 ata_eh_link_autopsy(ap->slave_link);
2100 /* transfer actions from slave to master and clear slave */
2101 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2102 mehc->i.action |= sehc->i.action;
2103 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2104 mehc->i.flags |= sehc->i.flags;
2105 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2108 /* Autopsy of fanout ports can affect host link autopsy.
2109 * Perform host link autopsy last.
2111 if (sata_pmp_attached(ap))
2112 ata_eh_link_autopsy(&ap->link);
2116 * ata_get_cmd_descript - get description for ATA command
2117 * @command: ATA command code to get description for
2119 * Return a textual description of the given command, or NULL if the
2120 * command is not known.
2125 const char *ata_get_cmd_descript(u8 command)
2127 #ifdef CONFIG_ATA_VERBOSE_ERROR
2133 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2134 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2135 { ATA_CMD_STANDBY, "STANDBY" },
2136 { ATA_CMD_IDLE, "IDLE" },
2137 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2138 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2139 { ATA_CMD_NOP, "NOP" },
2140 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2141 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2142 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2143 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2144 { ATA_CMD_SERVICE, "SERVICE" },
2145 { ATA_CMD_READ, "READ DMA" },
2146 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2147 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2148 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2149 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2150 { ATA_CMD_WRITE, "WRITE DMA" },
2151 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2152 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2153 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2154 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2155 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2156 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2157 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2158 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2159 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2160 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2161 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2162 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2163 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2164 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2165 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2166 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2167 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2168 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2169 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2170 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2171 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2172 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2173 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2174 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2175 { ATA_CMD_SLEEP, "SLEEP" },
2176 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2177 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2178 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2179 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2180 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2181 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2182 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2183 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2184 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2185 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2186 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2187 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2188 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2189 { ATA_CMD_PMP_READ, "READ BUFFER" },
2190 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2191 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2192 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2193 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2194 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2195 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2196 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2197 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2198 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2199 { ATA_CMD_SMART, "SMART" },
2200 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2201 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2202 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2203 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2204 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2205 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2206 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2207 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2208 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2209 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2210 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2211 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2212 { ATA_CMD_RESTORE, "RECALIBRATE" },
2213 { 0, NULL } /* terminate list */
2217 for (i = 0; cmd_descr[i].text; i++)
2218 if (cmd_descr[i].command == command)
2219 return cmd_descr[i].text;
2226 * ata_eh_link_report - report error handling to user
2227 * @link: ATA link EH is going on
2229 * Report EH to user.
2234 static void ata_eh_link_report(struct ata_link *link)
2236 struct ata_port *ap = link->ap;
2237 struct ata_eh_context *ehc = &link->eh_context;
2238 const char *frozen, *desc;
2240 int tag, nr_failed = 0;
2242 if (ehc->i.flags & ATA_EHI_QUIET)
2246 if (ehc->i.desc[0] != '\0')
2249 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2250 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2252 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2253 ata_dev_phys_link(qc->dev) != link ||
2254 ((qc->flags & ATA_QCFLAG_QUIET) &&
2255 qc->err_mask == AC_ERR_DEV))
2257 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2263 if (!nr_failed && !ehc->i.err_mask)
2267 if (ap->pflags & ATA_PFLAG_FROZEN)
2270 memset(tries_buf, 0, sizeof(tries_buf));
2271 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2272 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2276 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2277 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2278 ehc->i.err_mask, link->sactive, ehc->i.serror,
2279 ehc->i.action, frozen, tries_buf);
2281 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2283 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2284 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2285 ehc->i.err_mask, link->sactive, ehc->i.serror,
2286 ehc->i.action, frozen, tries_buf);
2288 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2291 #ifdef CONFIG_ATA_VERBOSE_ERROR
2293 ata_link_printk(link, KERN_ERR,
2294 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2295 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2296 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2297 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2298 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2299 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2300 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2301 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2302 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2303 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2304 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2305 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2306 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2307 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2308 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2309 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2310 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2311 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2314 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2315 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2316 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2317 const u8 *cdb = qc->cdb;
2318 char data_buf[20] = "";
2319 char cdb_buf[70] = "";
2321 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2322 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2325 if (qc->dma_dir != DMA_NONE) {
2326 static const char *dma_str[] = {
2327 [DMA_BIDIRECTIONAL] = "bidi",
2328 [DMA_TO_DEVICE] = "out",
2329 [DMA_FROM_DEVICE] = "in",
2331 static const char *prot_str[] = {
2332 [ATA_PROT_PIO] = "pio",
2333 [ATA_PROT_DMA] = "dma",
2334 [ATA_PROT_NCQ] = "ncq",
2335 [ATAPI_PROT_PIO] = "pio",
2336 [ATAPI_PROT_DMA] = "dma",
2339 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2340 prot_str[qc->tf.protocol], qc->nbytes,
2341 dma_str[qc->dma_dir]);
2344 if (ata_is_atapi(qc->tf.protocol)) {
2346 scsi_print_command(qc->scsicmd);
2348 snprintf(cdb_buf, sizeof(cdb_buf),
2349 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2350 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2351 cdb[0], cdb[1], cdb[2], cdb[3],
2352 cdb[4], cdb[5], cdb[6], cdb[7],
2353 cdb[8], cdb[9], cdb[10], cdb[11],
2354 cdb[12], cdb[13], cdb[14], cdb[15]);
2356 const char *descr = ata_get_cmd_descript(cmd->command);
2358 ata_dev_printk(qc->dev, KERN_ERR,
2359 "failed command: %s\n", descr);
2362 ata_dev_printk(qc->dev, KERN_ERR,
2363 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2365 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2366 "Emask 0x%x (%s)%s\n",
2367 cmd->command, cmd->feature, cmd->nsect,
2368 cmd->lbal, cmd->lbam, cmd->lbah,
2369 cmd->hob_feature, cmd->hob_nsect,
2370 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2371 cmd->device, qc->tag, data_buf, cdb_buf,
2372 res->command, res->feature, res->nsect,
2373 res->lbal, res->lbam, res->lbah,
2374 res->hob_feature, res->hob_nsect,
2375 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2376 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2377 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2379 #ifdef CONFIG_ATA_VERBOSE_ERROR
2380 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2382 if (res->command & ATA_BUSY)
2383 ata_dev_printk(qc->dev, KERN_ERR,
2384 "status: { Busy }\n");
2386 ata_dev_printk(qc->dev, KERN_ERR,
2387 "status: { %s%s%s%s}\n",
2388 res->command & ATA_DRDY ? "DRDY " : "",
2389 res->command & ATA_DF ? "DF " : "",
2390 res->command & ATA_DRQ ? "DRQ " : "",
2391 res->command & ATA_ERR ? "ERR " : "");
2394 if (cmd->command != ATA_CMD_PACKET &&
2395 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2397 ata_dev_printk(qc->dev, KERN_ERR,
2398 "error: { %s%s%s%s}\n",
2399 res->feature & ATA_ICRC ? "ICRC " : "",
2400 res->feature & ATA_UNC ? "UNC " : "",
2401 res->feature & ATA_IDNF ? "IDNF " : "",
2402 res->feature & ATA_ABORTED ? "ABRT " : "");
2408 * ata_eh_report - report error handling to user
2409 * @ap: ATA port to report EH about
2411 * Report EH to user.
2416 void ata_eh_report(struct ata_port *ap)
2418 struct ata_link *link;
2420 ata_for_each_link(link, ap, HOST_FIRST)
2421 ata_eh_link_report(link);
2424 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2425 unsigned int *classes, unsigned long deadline,
2428 struct ata_device *dev;
2431 ata_for_each_dev(dev, link, ALL)
2432 classes[dev->devno] = ATA_DEV_UNKNOWN;
2434 return reset(link, classes, deadline);
2437 static int ata_eh_followup_srst_needed(struct ata_link *link,
2438 int rc, const unsigned int *classes)
2440 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2444 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2449 int ata_eh_reset(struct ata_link *link, int classify,
2450 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2451 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2453 struct ata_port *ap = link->ap;
2454 struct ata_link *slave = ap->slave_link;
2455 struct ata_eh_context *ehc = &link->eh_context;
2456 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2457 unsigned int *classes = ehc->classes;
2458 unsigned int lflags = link->flags;
2459 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2460 int max_tries = 0, try = 0;
2461 struct ata_link *failed_link;
2462 struct ata_device *dev;
2463 unsigned long deadline, now;
2464 ata_reset_fn_t reset;
2465 unsigned long flags;
2472 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2474 if (link->flags & ATA_LFLAG_NO_HRST)
2476 if (link->flags & ATA_LFLAG_NO_SRST)
2479 /* make sure each reset attemp is at least COOL_DOWN apart */
2480 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2482 WARN_ON(time_after(ehc->last_reset, now));
2483 deadline = ata_deadline(ehc->last_reset,
2484 ATA_EH_RESET_COOL_DOWN);
2485 if (time_before(now, deadline))
2486 schedule_timeout_uninterruptible(deadline - now);
2489 spin_lock_irqsave(ap->lock, flags);
2490 ap->pflags |= ATA_PFLAG_RESETTING;
2491 spin_unlock_irqrestore(ap->lock, flags);
2493 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2495 ata_for_each_dev(dev, link, ALL) {
2496 /* If we issue an SRST then an ATA drive (not ATAPI)
2497 * may change configuration and be in PIO0 timing. If
2498 * we do a hard reset (or are coming from power on)
2499 * this is true for ATA or ATAPI. Until we've set a
2500 * suitable controller mode we should not touch the
2501 * bus as we may be talking too fast.
2503 dev->pio_mode = XFER_PIO_0;
2505 /* If the controller has a pio mode setup function
2506 * then use it to set the chipset to rights. Don't
2507 * touch the DMA setup as that will be dealt with when
2508 * configuring devices.
2510 if (ap->ops->set_piomode)
2511 ap->ops->set_piomode(ap, dev);
2514 /* prefer hardreset */
2516 ehc->i.action &= ~ATA_EH_RESET;
2519 ehc->i.action |= ATA_EH_HARDRESET;
2520 } else if (softreset) {
2522 ehc->i.action |= ATA_EH_SOFTRESET;
2526 unsigned long deadline = ata_deadline(jiffies,
2527 ATA_EH_PRERESET_TIMEOUT);
2530 sehc->i.action &= ~ATA_EH_RESET;
2531 sehc->i.action |= ehc->i.action;
2534 rc = prereset(link, deadline);
2536 /* If present, do prereset on slave link too. Reset
2537 * is skipped iff both master and slave links report
2538 * -ENOENT or clear ATA_EH_RESET.
2540 if (slave && (rc == 0 || rc == -ENOENT)) {
2543 tmp = prereset(slave, deadline);
2547 ehc->i.action |= sehc->i.action;
2551 if (rc == -ENOENT) {
2552 ata_link_printk(link, KERN_DEBUG,
2553 "port disabled. ignoring.\n");
2554 ehc->i.action &= ~ATA_EH_RESET;
2556 ata_for_each_dev(dev, link, ALL)
2557 classes[dev->devno] = ATA_DEV_NONE;
2561 ata_link_printk(link, KERN_ERR,
2562 "prereset failed (errno=%d)\n", rc);
2566 /* prereset() might have cleared ATA_EH_RESET. If so,
2567 * bang classes, thaw and return.
2569 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2570 ata_for_each_dev(dev, link, ALL)
2571 classes[dev->devno] = ATA_DEV_NONE;
2572 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2573 ata_is_host_link(link))
2574 ata_eh_thaw_port(ap);
2584 if (ata_is_host_link(link))
2585 ata_eh_freeze_port(ap);
2587 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2591 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2592 reset == softreset ? "soft" : "hard");
2594 /* mark that this EH session started with reset */
2595 ehc->last_reset = jiffies;
2596 if (reset == hardreset)
2597 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2599 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2601 rc = ata_do_reset(link, reset, classes, deadline, true);
2602 if (rc && rc != -EAGAIN) {
2607 /* hardreset slave link if existent */
2608 if (slave && reset == hardreset) {
2612 ata_link_printk(slave, KERN_INFO,
2613 "hard resetting link\n");
2615 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2616 tmp = ata_do_reset(slave, reset, classes, deadline,
2624 failed_link = slave;
2630 /* perform follow-up SRST if necessary */
2631 if (reset == hardreset &&
2632 ata_eh_followup_srst_needed(link, rc, classes)) {
2636 ata_link_printk(link, KERN_ERR,
2637 "follow-up softreset required "
2638 "but no softreset avaliable\n");
2644 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2645 rc = ata_do_reset(link, reset, classes, deadline, true);
2653 ata_link_printk(link, KERN_INFO, "no reset method "
2654 "available, skipping reset\n");
2655 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2656 lflags |= ATA_LFLAG_ASSUME_ATA;
2660 * Post-reset processing
2662 ata_for_each_dev(dev, link, ALL) {
2663 /* After the reset, the device state is PIO 0 and the
2664 * controller state is undefined. Reset also wakes up
2665 * drives from sleeping mode.
2667 dev->pio_mode = XFER_PIO_0;
2668 dev->flags &= ~ATA_DFLAG_SLEEPING;
2670 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2673 /* apply class override */
2674 if (lflags & ATA_LFLAG_ASSUME_ATA)
2675 classes[dev->devno] = ATA_DEV_ATA;
2676 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2677 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2680 /* record current link speed */
2681 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2682 link->sata_spd = (sstatus >> 4) & 0xf;
2683 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2684 slave->sata_spd = (sstatus >> 4) & 0xf;
2687 if (ata_is_host_link(link))
2688 ata_eh_thaw_port(ap);
2690 /* postreset() should clear hardware SError. Although SError
2691 * is cleared during link resume, clearing SError here is
2692 * necessary as some PHYs raise hotplug events after SRST.
2693 * This introduces race condition where hotplug occurs between
2694 * reset and here. This race is mediated by cross checking
2695 * link onlineness and classification result later.
2698 postreset(link, classes);
2700 postreset(slave, classes);
2704 * Some controllers can't be frozen very well and may set
2705 * spuruious error conditions during reset. Clear accumulated
2706 * error information. As reset is the final recovery action,
2707 * nothing is lost by doing this.
2709 spin_lock_irqsave(link->ap->lock, flags);
2710 memset(&link->eh_info, 0, sizeof(link->eh_info));
2712 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2713 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2714 spin_unlock_irqrestore(link->ap->lock, flags);
2717 * Make sure onlineness and classification result correspond.
2718 * Hotplug could have happened during reset and some
2719 * controllers fail to wait while a drive is spinning up after
2720 * being hotplugged causing misdetection. By cross checking
2721 * link on/offlineness and classification result, those
2722 * conditions can be reliably detected and retried.
2725 ata_for_each_dev(dev, link, ALL) {
2726 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2727 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2728 ata_dev_printk(dev, KERN_DEBUG, "link online "
2729 "but device misclassifed\n");
2730 classes[dev->devno] = ATA_DEV_NONE;
2733 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2734 if (ata_class_enabled(classes[dev->devno]))
2735 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2736 "clearing class %d to NONE\n",
2737 classes[dev->devno]);
2738 classes[dev->devno] = ATA_DEV_NONE;
2739 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2740 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2741 "clearing UNKNOWN to NONE\n");
2742 classes[dev->devno] = ATA_DEV_NONE;
2746 if (classify && nr_unknown) {
2747 if (try < max_tries) {
2748 ata_link_printk(link, KERN_WARNING, "link online but "
2749 "%d devices misclassified, retrying\n",
2755 ata_link_printk(link, KERN_WARNING,
2756 "link online but %d devices misclassified, "
2757 "device detection might fail\n", nr_unknown);
2760 /* reset successful, schedule revalidation */
2761 ata_eh_done(link, NULL, ATA_EH_RESET);
2763 ata_eh_done(slave, NULL, ATA_EH_RESET);
2764 ehc->last_reset = jiffies; /* update to completion time */
2765 ehc->i.action |= ATA_EH_REVALIDATE;
2769 /* clear hotplug flag */
2770 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2772 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2774 spin_lock_irqsave(ap->lock, flags);
2775 ap->pflags &= ~ATA_PFLAG_RESETTING;
2776 spin_unlock_irqrestore(ap->lock, flags);
2781 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2782 if (!ata_is_host_link(link) &&
2783 sata_scr_read(link, SCR_STATUS, &sstatus))
2786 if (rc == -ERESTART || try >= max_tries)
2790 if (time_before(now, deadline)) {
2791 unsigned long delta = deadline - now;
2793 ata_link_printk(failed_link, KERN_WARNING,
2794 "reset failed (errno=%d), retrying in %u secs\n",
2795 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2798 delta = schedule_timeout_uninterruptible(delta);
2801 if (try == max_tries - 1) {
2802 sata_down_spd_limit(link, 0);
2804 sata_down_spd_limit(slave, 0);
2805 } else if (rc == -EPIPE)
2806 sata_down_spd_limit(failed_link, 0);
2813 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2815 struct ata_link *link;
2816 struct ata_device *dev;
2817 unsigned long flags;
2820 * This function can be thought of as an extended version of
2821 * ata_eh_about_to_do() specially crafted to accommodate the
2822 * requirements of ATA_EH_PARK handling. Since the EH thread
2823 * does not leave the do {} while () loop in ata_eh_recover as
2824 * long as the timeout for a park request to *one* device on
2825 * the port has not expired, and since we still want to pick
2826 * up park requests to other devices on the same port or
2827 * timeout updates for the same device, we have to pull
2828 * ATA_EH_PARK actions from eh_info into eh_context.i
2829 * ourselves at the beginning of each pass over the loop.
2831 * Additionally, all write accesses to &ap->park_req_pending
2832 * through INIT_COMPLETION() (see below) or complete_all()
2833 * (see ata_scsi_park_store()) are protected by the host lock.
2834 * As a result we have that park_req_pending.done is zero on
2835 * exit from this function, i.e. when ATA_EH_PARK actions for
2836 * *all* devices on port ap have been pulled into the
2837 * respective eh_context structs. If, and only if,
2838 * park_req_pending.done is non-zero by the time we reach
2839 * wait_for_completion_timeout(), another ATA_EH_PARK action
2840 * has been scheduled for at least one of the devices on port
2841 * ap and we have to cycle over the do {} while () loop in
2842 * ata_eh_recover() again.
2845 spin_lock_irqsave(ap->lock, flags);
2846 INIT_COMPLETION(ap->park_req_pending);
2847 ata_for_each_link(link, ap, EDGE) {
2848 ata_for_each_dev(dev, link, ALL) {
2849 struct ata_eh_info *ehi = &link->eh_info;
2851 link->eh_context.i.dev_action[dev->devno] |=
2852 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2853 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2856 spin_unlock_irqrestore(ap->lock, flags);
2859 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2861 struct ata_eh_context *ehc = &dev->link->eh_context;
2862 struct ata_taskfile tf;
2863 unsigned int err_mask;
2865 ata_tf_init(dev, &tf);
2867 ehc->unloaded_mask |= 1 << dev->devno;
2868 tf.command = ATA_CMD_IDLEIMMEDIATE;
2874 ehc->unloaded_mask &= ~(1 << dev->devno);
2875 tf.command = ATA_CMD_CHK_POWER;
2878 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2879 tf.protocol |= ATA_PROT_NODATA;
2880 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2881 if (park && (err_mask || tf.lbal != 0xc4)) {
2882 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2883 ehc->unloaded_mask &= ~(1 << dev->devno);
2887 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2888 struct ata_device **r_failed_dev)
2890 struct ata_port *ap = link->ap;
2891 struct ata_eh_context *ehc = &link->eh_context;
2892 struct ata_device *dev;
2893 unsigned int new_mask = 0;
2894 unsigned long flags;
2899 /* For PATA drive side cable detection to work, IDENTIFY must
2900 * be done backwards such that PDIAG- is released by the slave
2901 * device before the master device is identified.
2903 ata_for_each_dev(dev, link, ALL_REVERSE) {
2904 unsigned int action = ata_eh_dev_action(dev);
2905 unsigned int readid_flags = 0;
2907 if (ehc->i.flags & ATA_EHI_DID_RESET)
2908 readid_flags |= ATA_READID_POSTRESET;
2910 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2911 WARN_ON(dev->class == ATA_DEV_PMP);
2913 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2918 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2919 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2924 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2926 /* Configuration may have changed, reconfigure
2929 ehc->i.flags |= ATA_EHI_SETMODE;
2931 /* schedule the scsi_rescan_device() here */
2932 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2933 } else if (dev->class == ATA_DEV_UNKNOWN &&
2934 ehc->tries[dev->devno] &&
2935 ata_class_enabled(ehc->classes[dev->devno])) {
2936 /* Temporarily set dev->class, it will be
2937 * permanently set once all configurations are
2938 * complete. This is necessary because new
2939 * device configuration is done in two
2942 dev->class = ehc->classes[dev->devno];
2944 if (dev->class == ATA_DEV_PMP)
2945 rc = sata_pmp_attach(dev);
2947 rc = ata_dev_read_id(dev, &dev->class,
2948 readid_flags, dev->id);
2950 /* read_id might have changed class, store and reset */
2951 ehc->classes[dev->devno] = dev->class;
2952 dev->class = ATA_DEV_UNKNOWN;
2956 /* clear error info accumulated during probe */
2957 ata_ering_clear(&dev->ering);
2958 new_mask |= 1 << dev->devno;
2961 /* IDENTIFY was issued to non-existent
2962 * device. No need to reset. Just
2963 * thaw and ignore the device.
2965 ata_eh_thaw_port(ap);
2973 /* PDIAG- should have been released, ask cable type if post-reset */
2974 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2975 if (ap->ops->cable_detect)
2976 ap->cbl = ap->ops->cable_detect(ap);
2980 /* Configure new devices forward such that user doesn't see
2981 * device detection messages backwards.
2983 ata_for_each_dev(dev, link, ALL) {
2984 if (!(new_mask & (1 << dev->devno)))
2987 dev->class = ehc->classes[dev->devno];
2989 if (dev->class == ATA_DEV_PMP)
2992 ehc->i.flags |= ATA_EHI_PRINTINFO;
2993 rc = ata_dev_configure(dev);
2994 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2996 dev->class = ATA_DEV_UNKNOWN;
3000 spin_lock_irqsave(ap->lock, flags);
3001 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3002 spin_unlock_irqrestore(ap->lock, flags);
3004 /* new device discovered, configure xfermode */
3005 ehc->i.flags |= ATA_EHI_SETMODE;
3011 *r_failed_dev = dev;
3012 DPRINTK("EXIT rc=%d\n", rc);
3017 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3018 * @link: link on which timings will be programmed
3019 * @r_failed_dev: out parameter for failed device
3021 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3022 * ata_set_mode() fails, pointer to the failing device is
3023 * returned in @r_failed_dev.
3026 * PCI/etc. bus probe sem.
3029 * 0 on success, negative errno otherwise
3031 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3033 struct ata_port *ap = link->ap;
3034 struct ata_device *dev;
3037 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3038 ata_for_each_dev(dev, link, ENABLED) {
3039 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3040 struct ata_ering_entry *ent;
3042 ent = ata_ering_top(&dev->ering);
3044 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3048 /* has private set_mode? */
3049 if (ap->ops->set_mode)
3050 rc = ap->ops->set_mode(link, r_failed_dev);
3052 rc = ata_do_set_mode(link, r_failed_dev);
3054 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3055 ata_for_each_dev(dev, link, ENABLED) {
3056 struct ata_eh_context *ehc = &link->eh_context;
3057 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3058 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3060 if (dev->xfer_mode != saved_xfer_mode ||
3061 ata_ncq_enabled(dev) != saved_ncq)
3062 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3069 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3070 * @dev: ATAPI device to clear UA for
3072 * Resets and other operations can make an ATAPI device raise
3073 * UNIT ATTENTION which causes the next operation to fail. This
3074 * function clears UA.
3077 * EH context (may sleep).
3080 * 0 on success, -errno on failure.
3082 static int atapi_eh_clear_ua(struct ata_device *dev)
3086 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3087 u8 *sense_buffer = dev->link->ap->sector_buf;
3089 unsigned int err_mask;
3091 err_mask = atapi_eh_tur(dev, &sense_key);
3092 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3093 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3094 "failed (err_mask=0x%x)\n", err_mask);
3098 if (!err_mask || sense_key != UNIT_ATTENTION)
3101 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3103 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3104 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3109 ata_dev_printk(dev, KERN_WARNING,
3110 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3115 static int ata_link_nr_enabled(struct ata_link *link)
3117 struct ata_device *dev;
3120 ata_for_each_dev(dev, link, ENABLED)
3125 static int ata_link_nr_vacant(struct ata_link *link)
3127 struct ata_device *dev;
3130 ata_for_each_dev(dev, link, ALL)
3131 if (dev->class == ATA_DEV_UNKNOWN)
3136 static int ata_eh_skip_recovery(struct ata_link *link)
3138 struct ata_port *ap = link->ap;
3139 struct ata_eh_context *ehc = &link->eh_context;
3140 struct ata_device *dev;
3142 /* skip disabled links */
3143 if (link->flags & ATA_LFLAG_DISABLED)
3146 /* thaw frozen port and recover failed devices */
3147 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3150 /* reset at least once if reset is requested */
3151 if ((ehc->i.action & ATA_EH_RESET) &&
3152 !(ehc->i.flags & ATA_EHI_DID_RESET))
3155 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3156 ata_for_each_dev(dev, link, ALL) {
3157 if (dev->class == ATA_DEV_UNKNOWN &&
3158 ehc->classes[dev->devno] != ATA_DEV_NONE)
3165 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3167 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3168 u64 now = get_jiffies_64();
3169 int *trials = void_arg;
3171 if (ent->timestamp < now - min(now, interval))
3178 static int ata_eh_schedule_probe(struct ata_device *dev)
3180 struct ata_eh_context *ehc = &dev->link->eh_context;
3181 struct ata_link *link = ata_dev_phys_link(dev);
3184 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3185 (ehc->did_probe_mask & (1 << dev->devno)))
3188 ata_eh_detach_dev(dev);
3190 ehc->did_probe_mask |= (1 << dev->devno);
3191 ehc->i.action |= ATA_EH_RESET;
3192 ehc->saved_xfer_mode[dev->devno] = 0;
3193 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3195 /* Record and count probe trials on the ering. The specific
3196 * error mask used is irrelevant. Because a successful device
3197 * detection clears the ering, this count accumulates only if
3198 * there are consecutive failed probes.
3200 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3201 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3202 * forced to 1.5Gbps.
3204 * This is to work around cases where failed link speed
3205 * negotiation results in device misdetection leading to
3206 * infinite DEVXCHG or PHRDY CHG events.
3208 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3209 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3211 if (trials > ATA_EH_PROBE_TRIALS)
3212 sata_down_spd_limit(link, 1);
3217 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3219 struct ata_eh_context *ehc = &dev->link->eh_context;
3221 /* -EAGAIN from EH routine indicates retry without prejudice.
3222 * The requester is responsible for ensuring forward progress.
3225 ehc->tries[dev->devno]--;
3229 /* device missing or wrong IDENTIFY data, schedule probing */
3230 ehc->i.probe_mask |= (1 << dev->devno);
3232 /* give it just one more chance */
3233 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3235 if (ehc->tries[dev->devno] == 1) {
3236 /* This is the last chance, better to slow
3237 * down than lose it.
3239 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3240 if (dev->pio_mode > XFER_PIO_0)
3241 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3245 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3246 /* disable device if it has used up all its chances */
3247 ata_dev_disable(dev);
3249 /* detach if offline */
3250 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3251 ata_eh_detach_dev(dev);
3253 /* schedule probe if necessary */
3254 if (ata_eh_schedule_probe(dev)) {
3255 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3256 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3257 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3262 ehc->i.action |= ATA_EH_RESET;
3268 * ata_eh_recover - recover host port after error
3269 * @ap: host port to recover
3270 * @prereset: prereset method (can be NULL)
3271 * @softreset: softreset method (can be NULL)
3272 * @hardreset: hardreset method (can be NULL)
3273 * @postreset: postreset method (can be NULL)
3274 * @r_failed_link: out parameter for failed link
3276 * This is the alpha and omega, eum and yang, heart and soul of
3277 * libata exception handling. On entry, actions required to
3278 * recover each link and hotplug requests are recorded in the
3279 * link's eh_context. This function executes all the operations
3280 * with appropriate retrials and fallbacks to resurrect failed
3281 * devices, detach goners and greet newcomers.
3284 * Kernel thread context (may sleep).
3287 * 0 on success, -errno on failure.
3289 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3290 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3291 ata_postreset_fn_t postreset,
3292 struct ata_link **r_failed_link)
3294 struct ata_link *link;
3295 struct ata_device *dev;
3298 unsigned long flags, deadline;
3302 /* prep for recovery */
3303 ata_for_each_link(link, ap, EDGE) {
3304 struct ata_eh_context *ehc = &link->eh_context;
3306 /* re-enable link? */
3307 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3308 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3309 spin_lock_irqsave(ap->lock, flags);
3310 link->flags &= ~ATA_LFLAG_DISABLED;
3311 spin_unlock_irqrestore(ap->lock, flags);
3312 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3315 ata_for_each_dev(dev, link, ALL) {
3316 if (link->flags & ATA_LFLAG_NO_RETRY)
3317 ehc->tries[dev->devno] = 1;
3319 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3321 /* collect port action mask recorded in dev actions */
3322 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3323 ~ATA_EH_PERDEV_MASK;
3324 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3326 /* process hotplug request */
3327 if (dev->flags & ATA_DFLAG_DETACH)
3328 ata_eh_detach_dev(dev);
3330 /* schedule probe if necessary */
3331 if (!ata_dev_enabled(dev))
3332 ata_eh_schedule_probe(dev);
3340 /* if UNLOADING, finish immediately */
3341 if (ap->pflags & ATA_PFLAG_UNLOADING)
3345 ata_for_each_link(link, ap, EDGE) {
3346 struct ata_eh_context *ehc = &link->eh_context;
3348 /* skip EH if possible. */
3349 if (ata_eh_skip_recovery(link))
3352 ata_for_each_dev(dev, link, ALL)
3353 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3357 ata_for_each_link(link, ap, EDGE) {
3358 struct ata_eh_context *ehc = &link->eh_context;
3360 if (!(ehc->i.action & ATA_EH_RESET))
3363 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3364 prereset, softreset, hardreset, postreset);
3366 ata_link_printk(link, KERN_ERR,
3367 "reset failed, giving up\n");
3376 * clears ATA_EH_PARK in eh_info and resets
3377 * ap->park_req_pending
3379 ata_eh_pull_park_action(ap);
3382 ata_for_each_link(link, ap, EDGE) {
3383 ata_for_each_dev(dev, link, ALL) {
3384 struct ata_eh_context *ehc = &link->eh_context;
3387 if (dev->class != ATA_DEV_ATA)
3389 if (!(ehc->i.dev_action[dev->devno] &
3392 tmp = dev->unpark_deadline;
3393 if (time_before(deadline, tmp))
3395 else if (time_before_eq(tmp, jiffies))
3397 if (ehc->unloaded_mask & (1 << dev->devno))
3400 ata_eh_park_issue_cmd(dev, 1);
3405 if (time_before_eq(deadline, now))
3408 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3411 ata_for_each_link(link, ap, EDGE) {
3412 ata_for_each_dev(dev, link, ALL) {
3413 if (!(link->eh_context.unloaded_mask &
3417 ata_eh_park_issue_cmd(dev, 0);
3418 ata_eh_done(link, dev, ATA_EH_PARK);
3423 ata_for_each_link(link, ap, EDGE) {
3424 struct ata_eh_context *ehc = &link->eh_context;
3426 /* revalidate existing devices and attach new ones */
3427 rc = ata_eh_revalidate_and_attach(link, &dev);
3431 /* if PMP got attached, return, pmp EH will take care of it */
3432 if (link->device->class == ATA_DEV_PMP) {
3437 /* configure transfer mode if necessary */
3438 if (ehc->i.flags & ATA_EHI_SETMODE) {
3439 rc = ata_set_mode(link, &dev);
3442 ehc->i.flags &= ~ATA_EHI_SETMODE;
3445 /* If reset has been issued, clear UA to avoid
3446 * disrupting the current users of the device.
3448 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3449 ata_for_each_dev(dev, link, ALL) {
3450 if (dev->class != ATA_DEV_ATAPI)
3452 rc = atapi_eh_clear_ua(dev);
3458 /* configure link power saving */
3459 if (ehc->i.action & ATA_EH_LPM)
3460 ata_for_each_dev(dev, link, ALL)
3461 ata_dev_enable_pm(dev, ap->pm_policy);
3463 /* this link is okay now */
3469 ata_eh_handle_dev_fail(dev, rc);
3471 if (ap->pflags & ATA_PFLAG_FROZEN) {
3472 /* PMP reset requires working host port.
3473 * Can't retry if it's frozen.
3475 if (sata_pmp_attached(ap))
3485 if (rc && r_failed_link)
3486 *r_failed_link = link;
3488 DPRINTK("EXIT, rc=%d\n", rc);
3493 * ata_eh_finish - finish up EH
3494 * @ap: host port to finish EH for
3496 * Recovery is complete. Clean up EH states and retry or finish
3502 void ata_eh_finish(struct ata_port *ap)
3506 /* retry or finish qcs */
3507 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3508 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3510 if (!(qc->flags & ATA_QCFLAG_FAILED))
3514 /* FIXME: Once EH migration is complete,
3515 * generate sense data in this function,
3516 * considering both err_mask and tf.
3518 if (qc->flags & ATA_QCFLAG_RETRY)
3519 ata_eh_qc_retry(qc);
3521 ata_eh_qc_complete(qc);
3523 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3524 ata_eh_qc_complete(qc);
3526 /* feed zero TF to sense generation */
3527 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3528 ata_eh_qc_retry(qc);
3533 /* make sure nr_active_links is zero after EH */
3534 WARN_ON(ap->nr_active_links);
3535 ap->nr_active_links = 0;
3539 * ata_do_eh - do standard error handling
3540 * @ap: host port to handle error for
3542 * @prereset: prereset method (can be NULL)
3543 * @softreset: softreset method (can be NULL)
3544 * @hardreset: hardreset method (can be NULL)
3545 * @postreset: postreset method (can be NULL)
3547 * Perform standard error handling sequence.
3550 * Kernel thread context (may sleep).
3552 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3553 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3554 ata_postreset_fn_t postreset)
3556 struct ata_device *dev;
3562 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3565 ata_for_each_dev(dev, &ap->link, ALL)
3566 ata_dev_disable(dev);
3573 * ata_std_error_handler - standard error handler
3574 * @ap: host port to handle error for
3576 * Standard error handler
3579 * Kernel thread context (may sleep).
3581 void ata_std_error_handler(struct ata_port *ap)
3583 struct ata_port_operations *ops = ap->ops;
3584 ata_reset_fn_t hardreset = ops->hardreset;
3586 /* ignore built-in hardreset if SCR access is not available */
3587 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3590 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3595 * ata_eh_handle_port_suspend - perform port suspend operation
3596 * @ap: port to suspend
3601 * Kernel thread context (may sleep).
3603 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3605 unsigned long flags;
3608 /* are we suspending? */
3609 spin_lock_irqsave(ap->lock, flags);
3610 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3611 ap->pm_mesg.event == PM_EVENT_ON) {
3612 spin_unlock_irqrestore(ap->lock, flags);
3615 spin_unlock_irqrestore(ap->lock, flags);
3617 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3619 /* tell ACPI we're suspending */
3620 rc = ata_acpi_on_suspend(ap);
3625 ata_eh_freeze_port(ap);
3627 if (ap->ops->port_suspend)
3628 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3630 ata_acpi_set_state(ap, PMSG_SUSPEND);
3633 spin_lock_irqsave(ap->lock, flags);
3635 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3637 ap->pflags |= ATA_PFLAG_SUSPENDED;
3638 else if (ap->pflags & ATA_PFLAG_FROZEN)
3639 ata_port_schedule_eh(ap);
3641 if (ap->pm_result) {
3642 *ap->pm_result = rc;
3643 ap->pm_result = NULL;
3646 spin_unlock_irqrestore(ap->lock, flags);
3652 * ata_eh_handle_port_resume - perform port resume operation
3653 * @ap: port to resume
3658 * Kernel thread context (may sleep).
3660 static void ata_eh_handle_port_resume(struct ata_port *ap)
3662 struct ata_link *link;
3663 struct ata_device *dev;
3664 unsigned long flags;
3667 /* are we resuming? */
3668 spin_lock_irqsave(ap->lock, flags);
3669 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3670 ap->pm_mesg.event != PM_EVENT_ON) {
3671 spin_unlock_irqrestore(ap->lock, flags);
3674 spin_unlock_irqrestore(ap->lock, flags);
3676 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3679 * Error timestamps are in jiffies which doesn't run while
3680 * suspended and PHY events during resume isn't too uncommon.
3681 * When the two are combined, it can lead to unnecessary speed
3682 * downs if the machine is suspended and resumed repeatedly.
3683 * Clear error history.
3685 ata_for_each_link(link, ap, HOST_FIRST)
3686 ata_for_each_dev(dev, link, ALL)
3687 ata_ering_clear(&dev->ering);
3689 ata_acpi_set_state(ap, PMSG_ON);
3691 if (ap->ops->port_resume)
3692 rc = ap->ops->port_resume(ap);
3694 /* tell ACPI that we're resuming */
3695 ata_acpi_on_resume(ap);
3698 spin_lock_irqsave(ap->lock, flags);
3699 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3700 if (ap->pm_result) {
3701 *ap->pm_result = rc;
3702 ap->pm_result = NULL;
3704 spin_unlock_irqrestore(ap->lock, flags);
3706 #endif /* CONFIG_PM */