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/pci.h>
37 #include <scsi/scsi.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_eh.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_cmnd.h>
42 #include "../scsi/scsi_transport_api.h"
44 #include <linux/libata.h>
49 /* speed down verdicts */
50 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
51 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
52 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
53 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
56 ATA_EFLAG_IS_IO = (1 << 0),
57 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
59 /* error categories */
62 ATA_ECAT_TOUT_HSM = 2,
64 ATA_ECAT_DUBIOUS_NONE = 4,
65 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
66 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
67 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 /* Waiting in ->prereset can never be reliable. It's sometimes nice
72 * to wait there but it can't be depended upon; otherwise, we wouldn't
73 * be resetting. Just give it enough time for most drives to spin up.
76 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
77 ATA_EH_FASTDRAIN_INTERVAL = 3 * HZ,
80 /* The following table determines how we sequence resets. Each entry
81 * represents timeout for that try. The first try can be soft or
82 * hardreset. All others are hardreset if available. In most cases
83 * the first reset w/ 10sec timeout should succeed. Following entries
84 * are mostly for error handling, hotplug and retarded devices.
86 static const unsigned long ata_eh_reset_timeouts[] = {
87 10 * HZ, /* most drives spin up by 10sec */
88 10 * HZ, /* > 99% working drives spin up before 20sec */
89 35 * HZ, /* give > 30 secs of idleness for retarded devices */
90 5 * HZ, /* and sweet one last chance */
91 /* > 1 min has elapsed, give up */
94 static void __ata_port_freeze(struct ata_port *ap);
96 static void ata_eh_handle_port_suspend(struct ata_port *ap);
97 static void ata_eh_handle_port_resume(struct ata_port *ap);
99 static void ata_eh_handle_port_suspend(struct ata_port *ap)
102 static void ata_eh_handle_port_resume(struct ata_port *ap)
104 #endif /* CONFIG_PM */
106 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
109 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
110 ATA_EH_DESC_LEN - ehi->desc_len,
115 * __ata_ehi_push_desc - push error description without adding separator
117 * @fmt: printf format string
119 * Format string according to @fmt and append it to @ehi->desc.
122 * spin_lock_irqsave(host lock)
124 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
129 __ata_ehi_pushv_desc(ehi, fmt, args);
134 * ata_ehi_push_desc - push error description with separator
136 * @fmt: printf format string
138 * Format string according to @fmt and append it to @ehi->desc.
139 * If @ehi->desc is not empty, ", " is added in-between.
142 * spin_lock_irqsave(host lock)
144 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
149 __ata_ehi_push_desc(ehi, ", ");
152 __ata_ehi_pushv_desc(ehi, fmt, args);
157 * ata_ehi_clear_desc - clean error description
163 * spin_lock_irqsave(host lock)
165 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
172 * ata_port_desc - append port description
173 * @ap: target ATA port
174 * @fmt: printf format string
176 * Format string according to @fmt and append it to port
177 * description. If port description is not empty, " " is added
178 * in-between. This function is to be used while initializing
179 * ata_host. The description is printed on host registration.
184 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
188 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
190 if (ap->link.eh_info.desc_len)
191 __ata_ehi_push_desc(&ap->link.eh_info, " ");
194 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
201 * ata_port_pbar_desc - append PCI BAR description
202 * @ap: target ATA port
203 * @bar: target PCI BAR
204 * @offset: offset into PCI BAR
205 * @name: name of the area
207 * If @offset is negative, this function formats a string which
208 * contains the name, address, size and type of the BAR and
209 * appends it to the port description. If @offset is zero or
210 * positive, only name and offsetted address is appended.
215 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
218 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
220 unsigned long long start, len;
222 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
224 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
227 start = (unsigned long long)pci_resource_start(pdev, bar);
228 len = (unsigned long long)pci_resource_len(pdev, bar);
231 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
233 ata_port_desc(ap, "%s 0x%llx", name,
234 start + (unsigned long long)offset);
237 #endif /* CONFIG_PCI */
239 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
240 unsigned int err_mask)
242 struct ata_ering_entry *ent;
247 ering->cursor %= ATA_ERING_SIZE;
249 ent = &ering->ring[ering->cursor];
250 ent->eflags = eflags;
251 ent->err_mask = err_mask;
252 ent->timestamp = get_jiffies_64();
255 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
257 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
264 static void ata_ering_clear(struct ata_ering *ering)
266 memset(ering, 0, sizeof(*ering));
269 static int ata_ering_map(struct ata_ering *ering,
270 int (*map_fn)(struct ata_ering_entry *, void *),
274 struct ata_ering_entry *ent;
278 ent = &ering->ring[idx];
281 rc = map_fn(ent, arg);
284 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
285 } while (idx != ering->cursor);
290 static unsigned int ata_eh_dev_action(struct ata_device *dev)
292 struct ata_eh_context *ehc = &dev->link->eh_context;
294 return ehc->i.action | ehc->i.dev_action[dev->devno];
297 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
298 struct ata_eh_info *ehi, unsigned int action)
300 struct ata_device *tdev;
303 ehi->action &= ~action;
304 ata_link_for_each_dev(tdev, link)
305 ehi->dev_action[tdev->devno] &= ~action;
307 /* doesn't make sense for port-wide EH actions */
308 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
310 /* break ehi->action into ehi->dev_action */
311 if (ehi->action & action) {
312 ata_link_for_each_dev(tdev, link)
313 ehi->dev_action[tdev->devno] |=
314 ehi->action & action;
315 ehi->action &= ~action;
318 /* turn off the specified per-dev action */
319 ehi->dev_action[dev->devno] &= ~action;
324 * ata_scsi_timed_out - SCSI layer time out callback
325 * @cmd: timed out SCSI command
327 * Handles SCSI layer timeout. We race with normal completion of
328 * the qc for @cmd. If the qc is already gone, we lose and let
329 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
330 * timed out and EH should be invoked. Prevent ata_qc_complete()
331 * from finishing it by setting EH_SCHEDULED and return
334 * TODO: kill this function once old EH is gone.
337 * Called from timer context
340 * EH_HANDLED or EH_NOT_HANDLED
342 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
344 struct Scsi_Host *host = cmd->device->host;
345 struct ata_port *ap = ata_shost_to_port(host);
347 struct ata_queued_cmd *qc;
348 enum scsi_eh_timer_return ret;
352 if (ap->ops->error_handler) {
353 ret = EH_NOT_HANDLED;
358 spin_lock_irqsave(ap->lock, flags);
359 qc = ata_qc_from_tag(ap, ap->link.active_tag);
361 WARN_ON(qc->scsicmd != cmd);
362 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
363 qc->err_mask |= AC_ERR_TIMEOUT;
364 ret = EH_NOT_HANDLED;
366 spin_unlock_irqrestore(ap->lock, flags);
369 DPRINTK("EXIT, ret=%d\n", ret);
374 * ata_scsi_error - SCSI layer error handler callback
375 * @host: SCSI host on which error occurred
377 * Handles SCSI-layer-thrown error events.
380 * Inherited from SCSI layer (none, can sleep)
385 void ata_scsi_error(struct Scsi_Host *host)
387 struct ata_port *ap = ata_shost_to_port(host);
393 /* synchronize with port task */
394 ata_port_flush_task(ap);
396 /* synchronize with host lock and sort out timeouts */
398 /* For new EH, all qcs are finished in one of three ways -
399 * normal completion, error completion, and SCSI timeout.
400 * Both cmpletions can race against SCSI timeout. When normal
401 * completion wins, the qc never reaches EH. When error
402 * completion wins, the qc has ATA_QCFLAG_FAILED set.
404 * When SCSI timeout wins, things are a bit more complex.
405 * Normal or error completion can occur after the timeout but
406 * before this point. In such cases, both types of
407 * completions are honored. A scmd is determined to have
408 * timed out iff its associated qc is active and not failed.
410 if (ap->ops->error_handler) {
411 struct scsi_cmnd *scmd, *tmp;
414 spin_lock_irqsave(ap->lock, flags);
416 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
417 struct ata_queued_cmd *qc;
419 for (i = 0; i < ATA_MAX_QUEUE; i++) {
420 qc = __ata_qc_from_tag(ap, i);
421 if (qc->flags & ATA_QCFLAG_ACTIVE &&
426 if (i < ATA_MAX_QUEUE) {
427 /* the scmd has an associated qc */
428 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
429 /* which hasn't failed yet, timeout */
430 qc->err_mask |= AC_ERR_TIMEOUT;
431 qc->flags |= ATA_QCFLAG_FAILED;
435 /* Normal completion occurred after
436 * SCSI timeout but before this point.
437 * Successfully complete it.
439 scmd->retries = scmd->allowed;
440 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
444 /* If we have timed out qcs. They belong to EH from
445 * this point but the state of the controller is
446 * unknown. Freeze the port to make sure the IRQ
447 * handler doesn't diddle with those qcs. This must
448 * be done atomically w.r.t. setting QCFLAG_FAILED.
451 __ata_port_freeze(ap);
453 spin_unlock_irqrestore(ap->lock, flags);
455 /* initialize eh_tries */
456 ap->eh_tries = ATA_EH_MAX_TRIES;
458 spin_unlock_wait(ap->lock);
461 /* invoke error handler */
462 if (ap->ops->error_handler) {
463 struct ata_link *link;
465 /* kill fast drain timer */
466 del_timer_sync(&ap->fastdrain_timer);
468 /* process port resume request */
469 ata_eh_handle_port_resume(ap);
471 /* fetch & clear EH info */
472 spin_lock_irqsave(ap->lock, flags);
474 __ata_port_for_each_link(link, ap) {
475 struct ata_eh_context *ehc = &link->eh_context;
476 struct ata_device *dev;
478 memset(&link->eh_context, 0, sizeof(link->eh_context));
479 link->eh_context.i = link->eh_info;
480 memset(&link->eh_info, 0, sizeof(link->eh_info));
482 ata_link_for_each_dev(dev, link) {
483 int devno = dev->devno;
485 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
486 if (ata_ncq_enabled(dev))
487 ehc->saved_ncq_enabled |= 1 << devno;
491 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
492 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
493 ap->excl_link = NULL; /* don't maintain exclusion over EH */
495 spin_unlock_irqrestore(ap->lock, flags);
497 /* invoke EH, skip if unloading or suspended */
498 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
499 ap->ops->error_handler(ap);
503 /* process port suspend request */
504 ata_eh_handle_port_suspend(ap);
506 /* Exception might have happend after ->error_handler
507 * recovered the port but before this point. Repeat
510 spin_lock_irqsave(ap->lock, flags);
512 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
513 if (--ap->eh_tries) {
514 spin_unlock_irqrestore(ap->lock, flags);
517 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
518 "tries, giving up\n", ATA_EH_MAX_TRIES);
519 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
522 /* this run is complete, make sure EH info is clear */
523 __ata_port_for_each_link(link, ap)
524 memset(&link->eh_info, 0, sizeof(link->eh_info));
526 /* Clear host_eh_scheduled while holding ap->lock such
527 * that if exception occurs after this point but
528 * before EH completion, SCSI midlayer will
531 host->host_eh_scheduled = 0;
533 spin_unlock_irqrestore(ap->lock, flags);
535 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
536 ap->ops->eng_timeout(ap);
539 /* finish or retry handled scmd's and clean up */
540 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
542 scsi_eh_flush_done_q(&ap->eh_done_q);
545 spin_lock_irqsave(ap->lock, flags);
547 if (ap->pflags & ATA_PFLAG_LOADING)
548 ap->pflags &= ~ATA_PFLAG_LOADING;
549 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
550 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
552 if (ap->pflags & ATA_PFLAG_RECOVERED)
553 ata_port_printk(ap, KERN_INFO, "EH complete\n");
555 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
557 /* tell wait_eh that we're done */
558 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
559 wake_up_all(&ap->eh_wait_q);
561 spin_unlock_irqrestore(ap->lock, flags);
567 * ata_port_wait_eh - Wait for the currently pending EH to complete
568 * @ap: Port to wait EH for
570 * Wait until the currently pending EH is complete.
573 * Kernel thread context (may sleep).
575 void ata_port_wait_eh(struct ata_port *ap)
581 spin_lock_irqsave(ap->lock, flags);
583 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
584 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
585 spin_unlock_irqrestore(ap->lock, flags);
587 spin_lock_irqsave(ap->lock, flags);
589 finish_wait(&ap->eh_wait_q, &wait);
591 spin_unlock_irqrestore(ap->lock, flags);
593 /* make sure SCSI EH is complete */
594 if (scsi_host_in_recovery(ap->scsi_host)) {
600 static int ata_eh_nr_in_flight(struct ata_port *ap)
605 /* count only non-internal commands */
606 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
607 if (ata_qc_from_tag(ap, tag))
613 void ata_eh_fastdrain_timerfn(unsigned long arg)
615 struct ata_port *ap = (void *)arg;
619 spin_lock_irqsave(ap->lock, flags);
621 cnt = ata_eh_nr_in_flight(ap);
627 if (cnt == ap->fastdrain_cnt) {
630 /* No progress during the last interval, tag all
631 * in-flight qcs as timed out and freeze the port.
633 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
634 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
636 qc->err_mask |= AC_ERR_TIMEOUT;
641 /* some qcs have finished, give it another chance */
642 ap->fastdrain_cnt = cnt;
643 ap->fastdrain_timer.expires =
644 jiffies + ATA_EH_FASTDRAIN_INTERVAL;
645 add_timer(&ap->fastdrain_timer);
649 spin_unlock_irqrestore(ap->lock, flags);
653 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
654 * @ap: target ATA port
655 * @fastdrain: activate fast drain
657 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
658 * is non-zero and EH wasn't pending before. Fast drain ensures
659 * that EH kicks in in timely manner.
662 * spin_lock_irqsave(host lock)
664 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
668 /* already scheduled? */
669 if (ap->pflags & ATA_PFLAG_EH_PENDING)
672 ap->pflags |= ATA_PFLAG_EH_PENDING;
677 /* do we have in-flight qcs? */
678 cnt = ata_eh_nr_in_flight(ap);
682 /* activate fast drain */
683 ap->fastdrain_cnt = cnt;
684 ap->fastdrain_timer.expires = jiffies + ATA_EH_FASTDRAIN_INTERVAL;
685 add_timer(&ap->fastdrain_timer);
689 * ata_qc_schedule_eh - schedule qc for error handling
690 * @qc: command to schedule error handling for
692 * Schedule error handling for @qc. EH will kick in as soon as
693 * other commands are drained.
696 * spin_lock_irqsave(host lock)
698 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
700 struct ata_port *ap = qc->ap;
702 WARN_ON(!ap->ops->error_handler);
704 qc->flags |= ATA_QCFLAG_FAILED;
705 ata_eh_set_pending(ap, 1);
707 /* The following will fail if timeout has already expired.
708 * ata_scsi_error() takes care of such scmds on EH entry.
709 * Note that ATA_QCFLAG_FAILED is unconditionally set after
710 * this function completes.
712 scsi_req_abort_cmd(qc->scsicmd);
716 * ata_port_schedule_eh - schedule error handling without a qc
717 * @ap: ATA port to schedule EH for
719 * Schedule error handling for @ap. EH will kick in as soon as
720 * all commands are drained.
723 * spin_lock_irqsave(host lock)
725 void ata_port_schedule_eh(struct ata_port *ap)
727 WARN_ON(!ap->ops->error_handler);
729 if (ap->pflags & ATA_PFLAG_INITIALIZING)
732 ata_eh_set_pending(ap, 1);
733 scsi_schedule_eh(ap->scsi_host);
735 DPRINTK("port EH scheduled\n");
738 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
740 int tag, nr_aborted = 0;
742 WARN_ON(!ap->ops->error_handler);
744 /* we're gonna abort all commands, no need for fast drain */
745 ata_eh_set_pending(ap, 0);
747 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
748 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
750 if (qc && (!link || qc->dev->link == link)) {
751 qc->flags |= ATA_QCFLAG_FAILED;
758 ata_port_schedule_eh(ap);
764 * ata_link_abort - abort all qc's on the link
765 * @link: ATA link to abort qc's for
767 * Abort all active qc's active on @link and schedule EH.
770 * spin_lock_irqsave(host lock)
773 * Number of aborted qc's.
775 int ata_link_abort(struct ata_link *link)
777 return ata_do_link_abort(link->ap, link);
781 * ata_port_abort - abort all qc's on the port
782 * @ap: ATA port to abort qc's for
784 * Abort all active qc's of @ap and schedule EH.
787 * spin_lock_irqsave(host_set lock)
790 * Number of aborted qc's.
792 int ata_port_abort(struct ata_port *ap)
794 return ata_do_link_abort(ap, NULL);
798 * __ata_port_freeze - freeze port
799 * @ap: ATA port to freeze
801 * This function is called when HSM violation or some other
802 * condition disrupts normal operation of the port. Frozen port
803 * is not allowed to perform any operation until the port is
804 * thawed, which usually follows a successful reset.
806 * ap->ops->freeze() callback can be used for freezing the port
807 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
808 * port cannot be frozen hardware-wise, the interrupt handler
809 * must ack and clear interrupts unconditionally while the port
813 * spin_lock_irqsave(host lock)
815 static void __ata_port_freeze(struct ata_port *ap)
817 WARN_ON(!ap->ops->error_handler);
822 ap->pflags |= ATA_PFLAG_FROZEN;
824 DPRINTK("ata%u port frozen\n", ap->print_id);
828 * ata_port_freeze - abort & freeze port
829 * @ap: ATA port to freeze
831 * Abort and freeze @ap.
834 * spin_lock_irqsave(host lock)
837 * Number of aborted commands.
839 int ata_port_freeze(struct ata_port *ap)
843 WARN_ON(!ap->ops->error_handler);
845 nr_aborted = ata_port_abort(ap);
846 __ata_port_freeze(ap);
852 * sata_async_notification - SATA async notification handler
853 * @ap: ATA port where async notification is received
855 * Handler to be called when async notification via SDB FIS is
856 * received. This function schedules EH if necessary.
859 * spin_lock_irqsave(host lock)
862 * 1 if EH is scheduled, 0 otherwise.
864 int sata_async_notification(struct ata_port *ap)
869 if (!(ap->flags & ATA_FLAG_AN))
872 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
874 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
876 if (!sata_pmp_attached(ap) || rc) {
877 /* PMP is not attached or SNTF is not available */
878 if (!sata_pmp_attached(ap)) {
879 /* PMP is not attached. Check whether ATAPI
880 * AN is configured. If so, notify media
883 struct ata_device *dev = ap->link.device;
885 if ((dev->class == ATA_DEV_ATAPI) &&
886 (dev->flags & ATA_DFLAG_AN))
887 ata_scsi_media_change_notify(dev);
890 /* PMP is attached but SNTF is not available.
891 * ATAPI async media change notification is
892 * not used. The PMP must be reporting PHY
893 * status change, schedule EH.
895 ata_port_schedule_eh(ap);
899 /* PMP is attached and SNTF is available */
900 struct ata_link *link;
902 /* check and notify ATAPI AN */
903 ata_port_for_each_link(link, ap) {
904 if (!(sntf & (1 << link->pmp)))
907 if ((link->device->class == ATA_DEV_ATAPI) &&
908 (link->device->flags & ATA_DFLAG_AN))
909 ata_scsi_media_change_notify(link->device);
912 /* If PMP is reporting that PHY status of some
913 * downstream ports has changed, schedule EH.
915 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
916 ata_port_schedule_eh(ap);
925 * ata_eh_freeze_port - EH helper to freeze port
926 * @ap: ATA port to freeze
933 void ata_eh_freeze_port(struct ata_port *ap)
937 if (!ap->ops->error_handler)
940 spin_lock_irqsave(ap->lock, flags);
941 __ata_port_freeze(ap);
942 spin_unlock_irqrestore(ap->lock, flags);
946 * ata_port_thaw_port - EH helper to thaw port
947 * @ap: ATA port to thaw
949 * Thaw frozen port @ap.
954 void ata_eh_thaw_port(struct ata_port *ap)
958 if (!ap->ops->error_handler)
961 spin_lock_irqsave(ap->lock, flags);
963 ap->pflags &= ~ATA_PFLAG_FROZEN;
968 spin_unlock_irqrestore(ap->lock, flags);
970 DPRINTK("ata%u port thawed\n", ap->print_id);
973 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
978 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
980 struct ata_port *ap = qc->ap;
981 struct scsi_cmnd *scmd = qc->scsicmd;
984 spin_lock_irqsave(ap->lock, flags);
985 qc->scsidone = ata_eh_scsidone;
986 __ata_qc_complete(qc);
987 WARN_ON(ata_tag_valid(qc->tag));
988 spin_unlock_irqrestore(ap->lock, flags);
990 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
994 * ata_eh_qc_complete - Complete an active ATA command from EH
995 * @qc: Command to complete
997 * Indicate to the mid and upper layers that an ATA command has
998 * completed. To be used from EH.
1000 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1002 struct scsi_cmnd *scmd = qc->scsicmd;
1003 scmd->retries = scmd->allowed;
1004 __ata_eh_qc_complete(qc);
1008 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1009 * @qc: Command to retry
1011 * Indicate to the mid and upper layers that an ATA command
1012 * should be retried. To be used from EH.
1014 * SCSI midlayer limits the number of retries to scmd->allowed.
1015 * scmd->retries is decremented for commands which get retried
1016 * due to unrelated failures (qc->err_mask is zero).
1018 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1020 struct scsi_cmnd *scmd = qc->scsicmd;
1021 if (!qc->err_mask && scmd->retries)
1023 __ata_eh_qc_complete(qc);
1027 * ata_eh_detach_dev - detach ATA device
1028 * @dev: ATA device to detach
1035 void ata_eh_detach_dev(struct ata_device *dev)
1037 struct ata_link *link = dev->link;
1038 struct ata_port *ap = link->ap;
1039 unsigned long flags;
1041 ata_dev_disable(dev);
1043 spin_lock_irqsave(ap->lock, flags);
1045 dev->flags &= ~ATA_DFLAG_DETACH;
1047 if (ata_scsi_offline_dev(dev)) {
1048 dev->flags |= ATA_DFLAG_DETACHED;
1049 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1052 /* clear per-dev EH actions */
1053 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1054 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1056 spin_unlock_irqrestore(ap->lock, flags);
1060 * ata_eh_about_to_do - about to perform eh_action
1061 * @link: target ATA link
1062 * @dev: target ATA dev for per-dev action (can be NULL)
1063 * @action: action about to be performed
1065 * Called just before performing EH actions to clear related bits
1066 * in @link->eh_info such that eh actions are not unnecessarily
1072 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1073 unsigned int action)
1075 struct ata_port *ap = link->ap;
1076 struct ata_eh_info *ehi = &link->eh_info;
1077 struct ata_eh_context *ehc = &link->eh_context;
1078 unsigned long flags;
1080 spin_lock_irqsave(ap->lock, flags);
1082 ata_eh_clear_action(link, dev, ehi, action);
1084 if (!(ehc->i.flags & ATA_EHI_QUIET))
1085 ap->pflags |= ATA_PFLAG_RECOVERED;
1087 spin_unlock_irqrestore(ap->lock, flags);
1091 * ata_eh_done - EH action complete
1092 * @ap: target ATA port
1093 * @dev: target ATA dev for per-dev action (can be NULL)
1094 * @action: action just completed
1096 * Called right after performing EH actions to clear related bits
1097 * in @link->eh_context.
1102 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1103 unsigned int action)
1105 struct ata_eh_context *ehc = &link->eh_context;
1107 ata_eh_clear_action(link, dev, &ehc->i, action);
1111 * ata_err_string - convert err_mask to descriptive string
1112 * @err_mask: error mask to convert to string
1114 * Convert @err_mask to descriptive string. Errors are
1115 * prioritized according to severity and only the most severe
1116 * error is reported.
1122 * Descriptive string for @err_mask
1124 static const char *ata_err_string(unsigned int err_mask)
1126 if (err_mask & AC_ERR_HOST_BUS)
1127 return "host bus error";
1128 if (err_mask & AC_ERR_ATA_BUS)
1129 return "ATA bus error";
1130 if (err_mask & AC_ERR_TIMEOUT)
1132 if (err_mask & AC_ERR_HSM)
1133 return "HSM violation";
1134 if (err_mask & AC_ERR_SYSTEM)
1135 return "internal error";
1136 if (err_mask & AC_ERR_MEDIA)
1137 return "media error";
1138 if (err_mask & AC_ERR_INVALID)
1139 return "invalid argument";
1140 if (err_mask & AC_ERR_DEV)
1141 return "device error";
1142 return "unknown error";
1146 * ata_read_log_page - read a specific log page
1147 * @dev: target device
1148 * @page: page to read
1149 * @buf: buffer to store read page
1150 * @sectors: number of sectors to read
1152 * Read log page using READ_LOG_EXT command.
1155 * Kernel thread context (may sleep).
1158 * 0 on success, AC_ERR_* mask otherwise.
1160 static unsigned int ata_read_log_page(struct ata_device *dev,
1161 u8 page, void *buf, unsigned int sectors)
1163 struct ata_taskfile tf;
1164 unsigned int err_mask;
1166 DPRINTK("read log page - page %d\n", page);
1168 ata_tf_init(dev, &tf);
1169 tf.command = ATA_CMD_READ_LOG_EXT;
1172 tf.hob_nsect = sectors >> 8;
1173 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1174 tf.protocol = ATA_PROT_PIO;
1176 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1177 buf, sectors * ATA_SECT_SIZE, 0);
1179 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1184 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1185 * @dev: Device to read log page 10h from
1186 * @tag: Resulting tag of the failed command
1187 * @tf: Resulting taskfile registers of the failed command
1189 * Read log page 10h to obtain NCQ error details and clear error
1193 * Kernel thread context (may sleep).
1196 * 0 on success, -errno otherwise.
1198 static int ata_eh_read_log_10h(struct ata_device *dev,
1199 int *tag, struct ata_taskfile *tf)
1201 u8 *buf = dev->link->ap->sector_buf;
1202 unsigned int err_mask;
1206 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1211 for (i = 0; i < ATA_SECT_SIZE; i++)
1214 ata_dev_printk(dev, KERN_WARNING,
1215 "invalid checksum 0x%x on log page 10h\n", csum);
1220 *tag = buf[0] & 0x1f;
1222 tf->command = buf[2];
1223 tf->feature = buf[3];
1227 tf->device = buf[7];
1228 tf->hob_lbal = buf[8];
1229 tf->hob_lbam = buf[9];
1230 tf->hob_lbah = buf[10];
1231 tf->nsect = buf[12];
1232 tf->hob_nsect = buf[13];
1238 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1239 * @dev: device to perform REQUEST_SENSE to
1240 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1242 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1243 * SENSE. This function is EH helper.
1246 * Kernel thread context (may sleep).
1249 * 0 on success, AC_ERR_* mask on failure
1251 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
1253 struct ata_device *dev = qc->dev;
1254 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
1255 struct ata_port *ap = dev->link->ap;
1256 struct ata_taskfile tf;
1257 u8 cdb[ATAPI_CDB_LEN];
1259 DPRINTK("ATAPI request sense\n");
1261 /* FIXME: is this needed? */
1262 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1264 /* initialize sense_buf with the error register,
1265 * for the case where they are -not- overwritten
1267 sense_buf[0] = 0x70;
1268 sense_buf[2] = qc->result_tf.feature >> 4;
1270 /* some devices time out if garbage left in tf */
1271 ata_tf_init(dev, &tf);
1273 memset(cdb, 0, ATAPI_CDB_LEN);
1274 cdb[0] = REQUEST_SENSE;
1275 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1277 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1278 tf.command = ATA_CMD_PACKET;
1280 /* is it pointless to prefer PIO for "safety reasons"? */
1281 if (ap->flags & ATA_FLAG_PIO_DMA) {
1282 tf.protocol = ATAPI_PROT_DMA;
1283 tf.feature |= ATAPI_PKT_DMA;
1285 tf.protocol = ATAPI_PROT_PIO;
1286 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1290 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1291 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1295 * ata_eh_analyze_serror - analyze SError for a failed port
1296 * @link: ATA link to analyze SError for
1298 * Analyze SError if available and further determine cause of
1304 static void ata_eh_analyze_serror(struct ata_link *link)
1306 struct ata_eh_context *ehc = &link->eh_context;
1307 u32 serror = ehc->i.serror;
1308 unsigned int err_mask = 0, action = 0;
1311 if (serror & SERR_PERSISTENT) {
1312 err_mask |= AC_ERR_ATA_BUS;
1313 action |= ATA_EH_RESET;
1316 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1317 err_mask |= AC_ERR_ATA_BUS;
1318 action |= ATA_EH_RESET;
1320 if (serror & SERR_PROTOCOL) {
1321 err_mask |= AC_ERR_HSM;
1322 action |= ATA_EH_RESET;
1324 if (serror & SERR_INTERNAL) {
1325 err_mask |= AC_ERR_SYSTEM;
1326 action |= ATA_EH_RESET;
1329 /* Determine whether a hotplug event has occurred. Both
1330 * SError.N/X are considered hotplug events for enabled or
1331 * host links. For disabled PMP links, only N bit is
1332 * considered as X bit is left at 1 for link plugging.
1336 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1337 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1339 hotplug_mask = SERR_PHYRDY_CHG;
1341 if (serror & hotplug_mask)
1342 ata_ehi_hotplugged(&ehc->i);
1344 ehc->i.err_mask |= err_mask;
1345 ehc->i.action |= action;
1349 * ata_eh_analyze_ncq_error - analyze NCQ error
1350 * @link: ATA link to analyze NCQ error for
1352 * Read log page 10h, determine the offending qc and acquire
1353 * error status TF. For NCQ device errors, all LLDDs have to do
1354 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1358 * Kernel thread context (may sleep).
1360 static void ata_eh_analyze_ncq_error(struct ata_link *link)
1362 struct ata_port *ap = link->ap;
1363 struct ata_eh_context *ehc = &link->eh_context;
1364 struct ata_device *dev = link->device;
1365 struct ata_queued_cmd *qc;
1366 struct ata_taskfile tf;
1369 /* if frozen, we can't do much */
1370 if (ap->pflags & ATA_PFLAG_FROZEN)
1373 /* is it NCQ device error? */
1374 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1377 /* has LLDD analyzed already? */
1378 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1379 qc = __ata_qc_from_tag(ap, tag);
1381 if (!(qc->flags & ATA_QCFLAG_FAILED))
1388 /* okay, this error is ours */
1389 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1391 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1392 "(errno=%d)\n", rc);
1396 if (!(link->sactive & (1 << tag))) {
1397 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1398 "inactive tag %d\n", tag);
1402 /* we've got the perpetrator, condemn it */
1403 qc = __ata_qc_from_tag(ap, tag);
1404 memcpy(&qc->result_tf, &tf, sizeof(tf));
1405 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1406 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1407 ehc->i.err_mask &= ~AC_ERR_DEV;
1411 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1412 * @qc: qc to analyze
1413 * @tf: Taskfile registers to analyze
1415 * Analyze taskfile of @qc and further determine cause of
1416 * failure. This function also requests ATAPI sense data if
1420 * Kernel thread context (may sleep).
1423 * Determined recovery action
1425 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1426 const struct ata_taskfile *tf)
1428 unsigned int tmp, action = 0;
1429 u8 stat = tf->command, err = tf->feature;
1431 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1432 qc->err_mask |= AC_ERR_HSM;
1433 return ATA_EH_RESET;
1436 if (stat & (ATA_ERR | ATA_DF))
1437 qc->err_mask |= AC_ERR_DEV;
1441 switch (qc->dev->class) {
1444 qc->err_mask |= AC_ERR_ATA_BUS;
1446 qc->err_mask |= AC_ERR_MEDIA;
1448 qc->err_mask |= AC_ERR_INVALID;
1452 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1453 tmp = atapi_eh_request_sense(qc);
1455 /* ATA_QCFLAG_SENSE_VALID is used to
1456 * tell atapi_qc_complete() that sense
1457 * data is already valid.
1459 * TODO: interpret sense data and set
1460 * appropriate err_mask.
1462 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1464 qc->err_mask |= tmp;
1468 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1469 action |= ATA_EH_RESET;
1474 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1479 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1483 base = ATA_ECAT_DUBIOUS_NONE;
1485 if (err_mask & AC_ERR_ATA_BUS)
1486 return base + ATA_ECAT_ATA_BUS;
1488 if (err_mask & AC_ERR_TIMEOUT)
1489 return base + ATA_ECAT_TOUT_HSM;
1491 if (eflags & ATA_EFLAG_IS_IO) {
1492 if (err_mask & AC_ERR_HSM)
1493 return base + ATA_ECAT_TOUT_HSM;
1495 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1496 return base + ATA_ECAT_UNK_DEV;
1502 struct speed_down_verdict_arg {
1505 int nr_errors[ATA_ECAT_NR];
1508 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1510 struct speed_down_verdict_arg *arg = void_arg;
1513 if (ent->timestamp < arg->since)
1516 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1518 arg->nr_errors[cat]++;
1524 * ata_eh_speed_down_verdict - Determine speed down verdict
1525 * @dev: Device of interest
1527 * This function examines error ring of @dev and determines
1528 * whether NCQ needs to be turned off, transfer speed should be
1529 * stepped down, or falling back to PIO is necessary.
1531 * ECAT_ATA_BUS : ATA_BUS error for any command
1533 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1536 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1538 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1539 * data transfer hasn't been verified.
1543 * NCQ_OFF : Turn off NCQ.
1545 * SPEED_DOWN : Speed down transfer speed but don't fall back
1548 * FALLBACK_TO_PIO : Fall back to PIO.
1550 * Even if multiple verdicts are returned, only one action is
1551 * taken per error. An action triggered by non-DUBIOUS errors
1552 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1553 * This is to expedite speed down decisions right after device is
1554 * initially configured.
1556 * The followings are speed down rules. #1 and #2 deal with
1559 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1560 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1562 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1563 * occurred during last 5 mins, NCQ_OFF.
1565 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1566 * ocurred during last 5 mins, FALLBACK_TO_PIO
1568 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1569 * during last 10 mins, NCQ_OFF.
1571 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1572 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1575 * Inherited from caller.
1578 * OR of ATA_EH_SPDN_* flags.
1580 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1582 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1583 u64 j64 = get_jiffies_64();
1584 struct speed_down_verdict_arg arg;
1585 unsigned int verdict = 0;
1587 /* scan past 5 mins of error history */
1588 memset(&arg, 0, sizeof(arg));
1589 arg.since = j64 - min(j64, j5mins);
1590 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1592 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1593 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1594 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1595 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1597 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1598 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1599 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1601 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1602 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1603 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1604 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1606 /* scan past 10 mins of error history */
1607 memset(&arg, 0, sizeof(arg));
1608 arg.since = j64 - min(j64, j10mins);
1609 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1611 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1612 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1613 verdict |= ATA_EH_SPDN_NCQ_OFF;
1615 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1616 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1617 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1618 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1624 * ata_eh_speed_down - record error and speed down if necessary
1625 * @dev: Failed device
1626 * @eflags: mask of ATA_EFLAG_* flags
1627 * @err_mask: err_mask of the error
1629 * Record error and examine error history to determine whether
1630 * adjusting transmission speed is necessary. It also sets
1631 * transmission limits appropriately if such adjustment is
1635 * Kernel thread context (may sleep).
1638 * Determined recovery action.
1640 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1641 unsigned int eflags, unsigned int err_mask)
1643 struct ata_link *link = dev->link;
1645 unsigned int verdict;
1646 unsigned int action = 0;
1648 /* don't bother if Cat-0 error */
1649 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1652 /* record error and determine whether speed down is necessary */
1653 ata_ering_record(&dev->ering, eflags, err_mask);
1654 verdict = ata_eh_speed_down_verdict(dev);
1657 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1658 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1659 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1660 dev->flags |= ATA_DFLAG_NCQ_OFF;
1661 ata_dev_printk(dev, KERN_WARNING,
1662 "NCQ disabled due to excessive errors\n");
1667 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1668 /* speed down SATA link speed if possible */
1669 if (sata_down_spd_limit(link) == 0) {
1670 action |= ATA_EH_RESET;
1674 /* lower transfer mode */
1675 if (dev->spdn_cnt < 2) {
1676 static const int dma_dnxfer_sel[] =
1677 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1678 static const int pio_dnxfer_sel[] =
1679 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1682 if (dev->xfer_shift != ATA_SHIFT_PIO)
1683 sel = dma_dnxfer_sel[dev->spdn_cnt];
1685 sel = pio_dnxfer_sel[dev->spdn_cnt];
1689 if (ata_down_xfermask_limit(dev, sel) == 0) {
1690 action |= ATA_EH_RESET;
1696 /* Fall back to PIO? Slowing down to PIO is meaningless for
1697 * SATA ATA devices. Consider it only for PATA and SATAPI.
1699 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1700 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1701 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1702 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1704 action |= ATA_EH_RESET;
1711 /* device has been slowed down, blow error history */
1712 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1713 ata_ering_clear(&dev->ering);
1718 * ata_eh_link_autopsy - analyze error and determine recovery action
1719 * @link: host link to perform autopsy on
1721 * Analyze why @link failed and determine which recovery actions
1722 * are needed. This function also sets more detailed AC_ERR_*
1723 * values and fills sense data for ATAPI CHECK SENSE.
1726 * Kernel thread context (may sleep).
1728 static void ata_eh_link_autopsy(struct ata_link *link)
1730 struct ata_port *ap = link->ap;
1731 struct ata_eh_context *ehc = &link->eh_context;
1732 struct ata_device *dev;
1733 unsigned int all_err_mask = 0, eflags = 0;
1740 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1743 /* obtain and analyze SError */
1744 rc = sata_scr_read(link, SCR_ERROR, &serror);
1746 ehc->i.serror |= serror;
1747 ata_eh_analyze_serror(link);
1748 } else if (rc != -EOPNOTSUPP) {
1749 /* SError read failed, force reset and probing */
1750 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1751 ehc->i.action |= ATA_EH_RESET;
1752 ehc->i.err_mask |= AC_ERR_OTHER;
1755 /* analyze NCQ failure */
1756 ata_eh_analyze_ncq_error(link);
1758 /* any real error trumps AC_ERR_OTHER */
1759 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1760 ehc->i.err_mask &= ~AC_ERR_OTHER;
1762 all_err_mask |= ehc->i.err_mask;
1764 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1765 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1767 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link)
1770 /* inherit upper level err_mask */
1771 qc->err_mask |= ehc->i.err_mask;
1774 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1776 /* DEV errors are probably spurious in case of ATA_BUS error */
1777 if (qc->err_mask & AC_ERR_ATA_BUS)
1778 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1781 /* any real error trumps unknown error */
1782 if (qc->err_mask & ~AC_ERR_OTHER)
1783 qc->err_mask &= ~AC_ERR_OTHER;
1785 /* SENSE_VALID trumps dev/unknown error and revalidation */
1786 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1787 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1789 /* determine whether the command is worth retrying */
1790 if (!(qc->err_mask & AC_ERR_INVALID) &&
1791 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1792 qc->flags |= ATA_QCFLAG_RETRY;
1794 /* accumulate error info */
1795 ehc->i.dev = qc->dev;
1796 all_err_mask |= qc->err_mask;
1797 if (qc->flags & ATA_QCFLAG_IO)
1798 eflags |= ATA_EFLAG_IS_IO;
1801 /* enforce default EH actions */
1802 if (ap->pflags & ATA_PFLAG_FROZEN ||
1803 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1804 ehc->i.action |= ATA_EH_RESET;
1805 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1806 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
1807 ehc->i.action |= ATA_EH_REVALIDATE;
1809 /* If we have offending qcs and the associated failed device,
1810 * perform per-dev EH action only on the offending device.
1813 ehc->i.dev_action[ehc->i.dev->devno] |=
1814 ehc->i.action & ATA_EH_PERDEV_MASK;
1815 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1818 /* propagate timeout to host link */
1819 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1820 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1822 /* record error and consider speeding down */
1824 if (!dev && ((ata_link_max_devices(link) == 1 &&
1825 ata_dev_enabled(link->device))))
1829 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1830 eflags |= ATA_EFLAG_DUBIOUS_XFER;
1831 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
1838 * ata_eh_autopsy - analyze error and determine recovery action
1839 * @ap: host port to perform autopsy on
1841 * Analyze all links of @ap and determine why they failed and
1842 * which recovery actions are needed.
1845 * Kernel thread context (may sleep).
1847 void ata_eh_autopsy(struct ata_port *ap)
1849 struct ata_link *link;
1851 ata_port_for_each_link(link, ap)
1852 ata_eh_link_autopsy(link);
1854 /* Autopsy of fanout ports can affect host link autopsy.
1855 * Perform host link autopsy last.
1857 if (sata_pmp_attached(ap))
1858 ata_eh_link_autopsy(&ap->link);
1862 * ata_eh_link_report - report error handling to user
1863 * @link: ATA link EH is going on
1865 * Report EH to user.
1870 static void ata_eh_link_report(struct ata_link *link)
1872 struct ata_port *ap = link->ap;
1873 struct ata_eh_context *ehc = &link->eh_context;
1874 const char *frozen, *desc;
1876 int tag, nr_failed = 0;
1878 if (ehc->i.flags & ATA_EHI_QUIET)
1882 if (ehc->i.desc[0] != '\0')
1885 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1886 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1888 if (!(qc->flags & ATA_QCFLAG_FAILED) || qc->dev->link != link ||
1889 ((qc->flags & ATA_QCFLAG_QUIET) &&
1890 qc->err_mask == AC_ERR_DEV))
1892 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1898 if (!nr_failed && !ehc->i.err_mask)
1902 if (ap->pflags & ATA_PFLAG_FROZEN)
1905 memset(tries_buf, 0, sizeof(tries_buf));
1906 if (ap->eh_tries < ATA_EH_MAX_TRIES)
1907 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
1911 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1912 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1913 ehc->i.err_mask, link->sactive, ehc->i.serror,
1914 ehc->i.action, frozen, tries_buf);
1916 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
1918 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
1919 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1920 ehc->i.err_mask, link->sactive, ehc->i.serror,
1921 ehc->i.action, frozen, tries_buf);
1923 ata_link_printk(link, KERN_ERR, "%s\n", desc);
1927 ata_port_printk(ap, KERN_ERR,
1928 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1929 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
1930 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
1931 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
1932 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
1933 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
1934 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
1935 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
1936 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
1937 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
1938 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
1939 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
1940 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
1941 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
1942 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
1943 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
1944 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
1945 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
1947 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1948 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1949 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1950 const u8 *cdb = qc->cdb;
1951 char data_buf[20] = "";
1952 char cdb_buf[70] = "";
1954 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1955 qc->dev->link != link || !qc->err_mask)
1958 if (qc->dma_dir != DMA_NONE) {
1959 static const char *dma_str[] = {
1960 [DMA_BIDIRECTIONAL] = "bidi",
1961 [DMA_TO_DEVICE] = "out",
1962 [DMA_FROM_DEVICE] = "in",
1964 static const char *prot_str[] = {
1965 [ATA_PROT_PIO] = "pio",
1966 [ATA_PROT_DMA] = "dma",
1967 [ATA_PROT_NCQ] = "ncq",
1968 [ATAPI_PROT_PIO] = "pio",
1969 [ATAPI_PROT_DMA] = "dma",
1972 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
1973 prot_str[qc->tf.protocol], qc->nbytes,
1974 dma_str[qc->dma_dir]);
1977 if (ata_is_atapi(qc->tf.protocol))
1978 snprintf(cdb_buf, sizeof(cdb_buf),
1979 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
1980 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
1981 cdb[0], cdb[1], cdb[2], cdb[3],
1982 cdb[4], cdb[5], cdb[6], cdb[7],
1983 cdb[8], cdb[9], cdb[10], cdb[11],
1984 cdb[12], cdb[13], cdb[14], cdb[15]);
1986 ata_dev_printk(qc->dev, KERN_ERR,
1987 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1989 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1990 "Emask 0x%x (%s)%s\n",
1991 cmd->command, cmd->feature, cmd->nsect,
1992 cmd->lbal, cmd->lbam, cmd->lbah,
1993 cmd->hob_feature, cmd->hob_nsect,
1994 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1995 cmd->device, qc->tag, data_buf, cdb_buf,
1996 res->command, res->feature, res->nsect,
1997 res->lbal, res->lbam, res->lbah,
1998 res->hob_feature, res->hob_nsect,
1999 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2000 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2001 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2003 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2005 if (res->command & ATA_BUSY)
2006 ata_dev_printk(qc->dev, KERN_ERR,
2007 "status: { Busy }\n");
2009 ata_dev_printk(qc->dev, KERN_ERR,
2010 "status: { %s%s%s%s}\n",
2011 res->command & ATA_DRDY ? "DRDY " : "",
2012 res->command & ATA_DF ? "DF " : "",
2013 res->command & ATA_DRQ ? "DRQ " : "",
2014 res->command & ATA_ERR ? "ERR " : "");
2017 if (cmd->command != ATA_CMD_PACKET &&
2018 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2020 ata_dev_printk(qc->dev, KERN_ERR,
2021 "error: { %s%s%s%s}\n",
2022 res->feature & ATA_ICRC ? "ICRC " : "",
2023 res->feature & ATA_UNC ? "UNC " : "",
2024 res->feature & ATA_IDNF ? "IDNF " : "",
2025 res->feature & ATA_ABORTED ? "ABRT " : "");
2030 * ata_eh_report - report error handling to user
2031 * @ap: ATA port to report EH about
2033 * Report EH to user.
2038 void ata_eh_report(struct ata_port *ap)
2040 struct ata_link *link;
2042 __ata_port_for_each_link(link, ap)
2043 ata_eh_link_report(link);
2046 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2047 unsigned int *classes, unsigned long deadline)
2049 struct ata_device *dev;
2052 ata_link_for_each_dev(dev, link)
2053 classes[dev->devno] = ATA_DEV_UNKNOWN;
2055 rc = reset(link, classes, deadline);
2057 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2058 ata_link_for_each_dev(dev, link)
2059 if (classes[dev->devno] == ATA_DEV_UNKNOWN)
2060 classes[dev->devno] = ATA_DEV_NONE;
2065 static int ata_eh_followup_srst_needed(struct ata_link *link,
2066 int rc, int classify,
2067 const unsigned int *classes)
2069 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2071 if (rc == -EAGAIN) {
2078 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2083 int ata_eh_reset(struct ata_link *link, int classify,
2084 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2085 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2087 const int max_tries = ARRAY_SIZE(ata_eh_reset_timeouts);
2088 struct ata_port *ap = link->ap;
2089 struct ata_eh_context *ehc = &link->eh_context;
2090 unsigned int *classes = ehc->classes;
2091 unsigned int lflags = link->flags;
2092 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2094 struct ata_device *dev;
2095 unsigned long deadline, now;
2096 ata_reset_fn_t reset;
2097 unsigned long flags;
2101 /* about to reset */
2102 spin_lock_irqsave(ap->lock, flags);
2103 ap->pflags |= ATA_PFLAG_RESETTING;
2104 spin_unlock_irqrestore(ap->lock, flags);
2106 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2108 ata_link_for_each_dev(dev, link) {
2109 /* If we issue an SRST then an ATA drive (not ATAPI)
2110 * may change configuration and be in PIO0 timing. If
2111 * we do a hard reset (or are coming from power on)
2112 * this is true for ATA or ATAPI. Until we've set a
2113 * suitable controller mode we should not touch the
2114 * bus as we may be talking too fast.
2116 dev->pio_mode = XFER_PIO_0;
2118 /* If the controller has a pio mode setup function
2119 * then use it to set the chipset to rights. Don't
2120 * touch the DMA setup as that will be dealt with when
2121 * configuring devices.
2123 if (ap->ops->set_piomode)
2124 ap->ops->set_piomode(ap, dev);
2127 if (!softreset && !hardreset) {
2129 ata_link_printk(link, KERN_INFO, "no reset method "
2130 "available, skipping reset\n");
2131 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2132 lflags |= ATA_LFLAG_ASSUME_ATA;
2136 /* prefer hardreset */
2137 ehc->i.action &= ~ATA_EH_RESET;
2140 ehc->i.action = ATA_EH_HARDRESET;
2141 } else if (softreset) {
2143 ehc->i.action = ATA_EH_SOFTRESET;
2145 ata_link_printk(link, KERN_ERR, "BUG: no reset method, "
2146 "please report to linux-ide@vger.kernel.org\n");
2152 rc = prereset(link, jiffies + ATA_EH_PRERESET_TIMEOUT);
2154 if (rc == -ENOENT) {
2155 ata_link_printk(link, KERN_DEBUG,
2156 "port disabled. ignoring.\n");
2157 ehc->i.action &= ~ATA_EH_RESET;
2159 ata_link_for_each_dev(dev, link)
2160 classes[dev->devno] = ATA_DEV_NONE;
2164 ata_link_printk(link, KERN_ERR,
2165 "prereset failed (errno=%d)\n", rc);
2170 /* prereset() might have cleared ATA_EH_RESET */
2171 if (!(ehc->i.action & ATA_EH_RESET)) {
2172 /* prereset told us not to reset, bang classes and return */
2173 ata_link_for_each_dev(dev, link)
2174 classes[dev->devno] = ATA_DEV_NONE;
2180 deadline = jiffies + ata_eh_reset_timeouts[try++];
2182 /* shut up during boot probing */
2184 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2185 reset == softreset ? "soft" : "hard");
2187 /* mark that this EH session started with reset */
2188 if (reset == hardreset)
2189 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2191 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2193 rc = ata_do_reset(link, reset, classes, deadline);
2195 if (reset == hardreset &&
2196 ata_eh_followup_srst_needed(link, rc, classify, classes)) {
2197 /* okay, let's do follow-up softreset */
2201 ata_link_printk(link, KERN_ERR,
2202 "follow-up softreset required "
2203 "but no softreset avaliable\n");
2208 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2209 rc = ata_do_reset(link, reset, classes, deadline);
2212 /* -EAGAIN can happen if we skipped followup SRST */
2213 if (rc && rc != -EAGAIN)
2217 ata_link_for_each_dev(dev, link) {
2218 /* After the reset, the device state is PIO 0 and the
2219 * controller state is undefined. Reset also wakes up
2220 * drives from sleeping mode.
2222 dev->pio_mode = XFER_PIO_0;
2223 dev->flags &= ~ATA_DFLAG_SLEEPING;
2225 if (ata_link_offline(link))
2228 /* apply class override */
2229 if (lflags & ATA_LFLAG_ASSUME_ATA)
2230 classes[dev->devno] = ATA_DEV_ATA;
2231 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2232 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
2235 /* record current link speed */
2236 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2237 link->sata_spd = (sstatus >> 4) & 0xf;
2240 postreset(link, classes);
2242 /* reset successful, schedule revalidation */
2243 ata_eh_done(link, NULL, ATA_EH_RESET);
2244 ehc->i.action |= ATA_EH_REVALIDATE;
2248 /* clear hotplug flag */
2249 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2251 spin_lock_irqsave(ap->lock, flags);
2252 ap->pflags &= ~ATA_PFLAG_RESETTING;
2253 spin_unlock_irqrestore(ap->lock, flags);
2258 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2259 if (!ata_is_host_link(link) &&
2260 sata_scr_read(link, SCR_STATUS, &sstatus))
2263 if (rc == -ERESTART || try >= max_tries)
2267 if (time_before(now, deadline)) {
2268 unsigned long delta = deadline - now;
2270 ata_link_printk(link, KERN_WARNING, "reset failed "
2271 "(errno=%d), retrying in %u secs\n",
2272 rc, (jiffies_to_msecs(delta) + 999) / 1000);
2275 delta = schedule_timeout_uninterruptible(delta);
2278 if (rc == -EPIPE || try == max_tries - 1)
2279 sata_down_spd_limit(link);
2285 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2286 struct ata_device **r_failed_dev)
2288 struct ata_port *ap = link->ap;
2289 struct ata_eh_context *ehc = &link->eh_context;
2290 struct ata_device *dev;
2291 unsigned int new_mask = 0;
2292 unsigned long flags;
2297 /* For PATA drive side cable detection to work, IDENTIFY must
2298 * be done backwards such that PDIAG- is released by the slave
2299 * device before the master device is identified.
2301 ata_link_for_each_dev_reverse(dev, link) {
2302 unsigned int action = ata_eh_dev_action(dev);
2303 unsigned int readid_flags = 0;
2305 if (ehc->i.flags & ATA_EHI_DID_RESET)
2306 readid_flags |= ATA_READID_POSTRESET;
2308 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2309 WARN_ON(dev->class == ATA_DEV_PMP);
2311 if (ata_link_offline(link)) {
2316 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2317 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2322 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2324 /* Configuration may have changed, reconfigure
2327 ehc->i.flags |= ATA_EHI_SETMODE;
2329 /* schedule the scsi_rescan_device() here */
2330 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2331 } else if (dev->class == ATA_DEV_UNKNOWN &&
2332 ehc->tries[dev->devno] &&
2333 ata_class_enabled(ehc->classes[dev->devno])) {
2334 dev->class = ehc->classes[dev->devno];
2336 if (dev->class == ATA_DEV_PMP)
2337 rc = sata_pmp_attach(dev);
2339 rc = ata_dev_read_id(dev, &dev->class,
2340 readid_flags, dev->id);
2343 new_mask |= 1 << dev->devno;
2346 /* IDENTIFY was issued to non-existent
2347 * device. No need to reset. Just
2348 * thaw and kill the device.
2350 ata_eh_thaw_port(ap);
2351 dev->class = ATA_DEV_UNKNOWN;
2354 dev->class = ATA_DEV_UNKNOWN;
2360 /* PDIAG- should have been released, ask cable type if post-reset */
2361 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2362 if (ap->ops->cable_detect)
2363 ap->cbl = ap->ops->cable_detect(ap);
2367 /* Configure new devices forward such that user doesn't see
2368 * device detection messages backwards.
2370 ata_link_for_each_dev(dev, link) {
2371 if (!(new_mask & (1 << dev->devno)) ||
2372 dev->class == ATA_DEV_PMP)
2375 ehc->i.flags |= ATA_EHI_PRINTINFO;
2376 rc = ata_dev_configure(dev);
2377 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2381 spin_lock_irqsave(ap->lock, flags);
2382 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2383 spin_unlock_irqrestore(ap->lock, flags);
2385 /* new device discovered, configure xfermode */
2386 ehc->i.flags |= ATA_EHI_SETMODE;
2392 *r_failed_dev = dev;
2393 DPRINTK("EXIT rc=%d\n", rc);
2398 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2399 * @link: link on which timings will be programmed
2400 * @r_failed_dev: out paramter for failed device
2402 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2403 * ata_set_mode() fails, pointer to the failing device is
2404 * returned in @r_failed_dev.
2407 * PCI/etc. bus probe sem.
2410 * 0 on success, negative errno otherwise
2412 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2414 struct ata_port *ap = link->ap;
2415 struct ata_device *dev;
2418 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2419 ata_link_for_each_dev(dev, link) {
2420 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2421 struct ata_ering_entry *ent;
2423 ent = ata_ering_top(&dev->ering);
2425 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2429 /* has private set_mode? */
2430 if (ap->ops->set_mode)
2431 rc = ap->ops->set_mode(link, r_failed_dev);
2433 rc = ata_do_set_mode(link, r_failed_dev);
2435 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2436 ata_link_for_each_dev(dev, link) {
2437 struct ata_eh_context *ehc = &link->eh_context;
2438 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2439 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2441 if (dev->xfer_mode != saved_xfer_mode ||
2442 ata_ncq_enabled(dev) != saved_ncq)
2443 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2449 static int ata_link_nr_enabled(struct ata_link *link)
2451 struct ata_device *dev;
2454 ata_link_for_each_dev(dev, link)
2455 if (ata_dev_enabled(dev))
2460 static int ata_link_nr_vacant(struct ata_link *link)
2462 struct ata_device *dev;
2465 ata_link_for_each_dev(dev, link)
2466 if (dev->class == ATA_DEV_UNKNOWN)
2471 static int ata_eh_skip_recovery(struct ata_link *link)
2473 struct ata_port *ap = link->ap;
2474 struct ata_eh_context *ehc = &link->eh_context;
2475 struct ata_device *dev;
2477 /* skip disabled links */
2478 if (link->flags & ATA_LFLAG_DISABLED)
2481 /* thaw frozen port and recover failed devices */
2482 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2485 /* reset at least once if reset is requested */
2486 if ((ehc->i.action & ATA_EH_RESET) &&
2487 !(ehc->i.flags & ATA_EHI_DID_RESET))
2490 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2491 ata_link_for_each_dev(dev, link) {
2492 if (dev->class == ATA_DEV_UNKNOWN &&
2493 ehc->classes[dev->devno] != ATA_DEV_NONE)
2500 static int ata_eh_schedule_probe(struct ata_device *dev)
2502 struct ata_eh_context *ehc = &dev->link->eh_context;
2504 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2505 (ehc->did_probe_mask & (1 << dev->devno)))
2508 ata_eh_detach_dev(dev);
2510 ehc->did_probe_mask |= (1 << dev->devno);
2511 ehc->i.action |= ATA_EH_RESET;
2512 ehc->saved_xfer_mode[dev->devno] = 0;
2513 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
2518 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2520 struct ata_eh_context *ehc = &dev->link->eh_context;
2522 ehc->tries[dev->devno]--;
2526 /* device missing or wrong IDENTIFY data, schedule probing */
2527 ehc->i.probe_mask |= (1 << dev->devno);
2529 /* give it just one more chance */
2530 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2532 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
2533 /* This is the last chance, better to slow
2534 * down than lose it.
2536 sata_down_spd_limit(dev->link);
2537 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2541 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2542 /* disable device if it has used up all its chances */
2543 ata_dev_disable(dev);
2545 /* detach if offline */
2546 if (ata_link_offline(dev->link))
2547 ata_eh_detach_dev(dev);
2549 /* schedule probe if necessary */
2550 if (ata_eh_schedule_probe(dev))
2551 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2555 ehc->i.action |= ATA_EH_RESET;
2561 * ata_eh_recover - recover host port after error
2562 * @ap: host port to recover
2563 * @prereset: prereset method (can be NULL)
2564 * @softreset: softreset method (can be NULL)
2565 * @hardreset: hardreset method (can be NULL)
2566 * @postreset: postreset method (can be NULL)
2567 * @r_failed_link: out parameter for failed link
2569 * This is the alpha and omega, eum and yang, heart and soul of
2570 * libata exception handling. On entry, actions required to
2571 * recover each link and hotplug requests are recorded in the
2572 * link's eh_context. This function executes all the operations
2573 * with appropriate retrials and fallbacks to resurrect failed
2574 * devices, detach goners and greet newcomers.
2577 * Kernel thread context (may sleep).
2580 * 0 on success, -errno on failure.
2582 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2583 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2584 ata_postreset_fn_t postreset,
2585 struct ata_link **r_failed_link)
2587 struct ata_link *link;
2588 struct ata_device *dev;
2589 int nr_failed_devs, nr_disabled_devs;
2591 unsigned long flags;
2595 /* prep for recovery */
2596 ata_port_for_each_link(link, ap) {
2597 struct ata_eh_context *ehc = &link->eh_context;
2599 /* re-enable link? */
2600 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
2601 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
2602 spin_lock_irqsave(ap->lock, flags);
2603 link->flags &= ~ATA_LFLAG_DISABLED;
2604 spin_unlock_irqrestore(ap->lock, flags);
2605 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
2608 ata_link_for_each_dev(dev, link) {
2609 if (link->flags & ATA_LFLAG_NO_RETRY)
2610 ehc->tries[dev->devno] = 1;
2612 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2614 /* collect port action mask recorded in dev actions */
2615 ehc->i.action |= ehc->i.dev_action[dev->devno] &
2616 ~ATA_EH_PERDEV_MASK;
2617 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
2619 /* process hotplug request */
2620 if (dev->flags & ATA_DFLAG_DETACH)
2621 ata_eh_detach_dev(dev);
2623 /* schedule probe if necessary */
2624 if (!ata_dev_enabled(dev))
2625 ata_eh_schedule_probe(dev);
2632 nr_disabled_devs = 0;
2635 /* if UNLOADING, finish immediately */
2636 if (ap->pflags & ATA_PFLAG_UNLOADING)
2640 ata_port_for_each_link(link, ap) {
2641 struct ata_eh_context *ehc = &link->eh_context;
2643 /* skip EH if possible. */
2644 if (ata_eh_skip_recovery(link))
2647 /* do we need to reset? */
2648 if (ehc->i.action & ATA_EH_RESET)
2651 ata_link_for_each_dev(dev, link)
2652 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
2657 /* if PMP is attached, this function only deals with
2658 * downstream links, port should stay thawed.
2660 if (!sata_pmp_attached(ap))
2661 ata_eh_freeze_port(ap);
2663 ata_port_for_each_link(link, ap) {
2664 struct ata_eh_context *ehc = &link->eh_context;
2666 if (!(ehc->i.action & ATA_EH_RESET))
2669 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
2670 prereset, softreset, hardreset,
2673 ata_link_printk(link, KERN_ERR,
2674 "reset failed, giving up\n");
2679 if (!sata_pmp_attached(ap))
2680 ata_eh_thaw_port(ap);
2684 ata_port_for_each_link(link, ap) {
2685 struct ata_eh_context *ehc = &link->eh_context;
2687 /* revalidate existing devices and attach new ones */
2688 rc = ata_eh_revalidate_and_attach(link, &dev);
2692 /* if PMP got attached, return, pmp EH will take care of it */
2693 if (link->device->class == ATA_DEV_PMP) {
2698 /* configure transfer mode if necessary */
2699 if (ehc->i.flags & ATA_EHI_SETMODE) {
2700 rc = ata_set_mode(link, &dev);
2703 ehc->i.flags &= ~ATA_EHI_SETMODE;
2706 if (ehc->i.action & ATA_EH_LPM)
2707 ata_link_for_each_dev(dev, link)
2708 ata_dev_enable_pm(dev, ap->pm_policy);
2710 /* this link is okay now */
2716 if (ata_eh_handle_dev_fail(dev, rc))
2719 if (ap->pflags & ATA_PFLAG_FROZEN) {
2720 /* PMP reset requires working host port.
2721 * Can't retry if it's frozen.
2723 if (sata_pmp_attached(ap))
2729 if (nr_failed_devs) {
2730 if (nr_failed_devs != nr_disabled_devs) {
2731 ata_port_printk(ap, KERN_WARNING, "failed to recover "
2732 "some devices, retrying in 5 secs\n");
2735 /* no device left to recover, repeat fast */
2743 if (rc && r_failed_link)
2744 *r_failed_link = link;
2746 DPRINTK("EXIT, rc=%d\n", rc);
2751 * ata_eh_finish - finish up EH
2752 * @ap: host port to finish EH for
2754 * Recovery is complete. Clean up EH states and retry or finish
2760 void ata_eh_finish(struct ata_port *ap)
2764 /* retry or finish qcs */
2765 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2766 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2768 if (!(qc->flags & ATA_QCFLAG_FAILED))
2772 /* FIXME: Once EH migration is complete,
2773 * generate sense data in this function,
2774 * considering both err_mask and tf.
2776 if (qc->flags & ATA_QCFLAG_RETRY)
2777 ata_eh_qc_retry(qc);
2779 ata_eh_qc_complete(qc);
2781 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2782 ata_eh_qc_complete(qc);
2784 /* feed zero TF to sense generation */
2785 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2786 ata_eh_qc_retry(qc);
2791 /* make sure nr_active_links is zero after EH */
2792 WARN_ON(ap->nr_active_links);
2793 ap->nr_active_links = 0;
2797 * ata_do_eh - do standard error handling
2798 * @ap: host port to handle error for
2800 * @prereset: prereset method (can be NULL)
2801 * @softreset: softreset method (can be NULL)
2802 * @hardreset: hardreset method (can be NULL)
2803 * @postreset: postreset method (can be NULL)
2805 * Perform standard error handling sequence.
2808 * Kernel thread context (may sleep).
2810 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2811 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2812 ata_postreset_fn_t postreset)
2814 struct ata_device *dev;
2820 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
2823 ata_link_for_each_dev(dev, &ap->link)
2824 ata_dev_disable(dev);
2831 * ata_std_error_handler - standard error handler
2832 * @ap: host port to handle error for
2834 * Standard error handler
2837 * Kernel thread context (may sleep).
2839 void ata_std_error_handler(struct ata_port *ap)
2841 struct ata_port_operations *ops = ap->ops;
2842 ata_reset_fn_t hardreset = ops->hardreset;
2844 /* ignore built-in hardreset if SCR access is not available */
2845 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
2848 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
2853 * ata_eh_handle_port_suspend - perform port suspend operation
2854 * @ap: port to suspend
2859 * Kernel thread context (may sleep).
2861 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2863 unsigned long flags;
2866 /* are we suspending? */
2867 spin_lock_irqsave(ap->lock, flags);
2868 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2869 ap->pm_mesg.event == PM_EVENT_ON) {
2870 spin_unlock_irqrestore(ap->lock, flags);
2873 spin_unlock_irqrestore(ap->lock, flags);
2875 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2877 /* tell ACPI we're suspending */
2878 rc = ata_acpi_on_suspend(ap);
2883 ata_eh_freeze_port(ap);
2885 if (ap->ops->port_suspend)
2886 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2888 ata_acpi_set_state(ap, PMSG_SUSPEND);
2891 spin_lock_irqsave(ap->lock, flags);
2893 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2895 ap->pflags |= ATA_PFLAG_SUSPENDED;
2896 else if (ap->pflags & ATA_PFLAG_FROZEN)
2897 ata_port_schedule_eh(ap);
2899 if (ap->pm_result) {
2900 *ap->pm_result = rc;
2901 ap->pm_result = NULL;
2904 spin_unlock_irqrestore(ap->lock, flags);
2910 * ata_eh_handle_port_resume - perform port resume operation
2911 * @ap: port to resume
2916 * Kernel thread context (may sleep).
2918 static void ata_eh_handle_port_resume(struct ata_port *ap)
2920 unsigned long flags;
2923 /* are we resuming? */
2924 spin_lock_irqsave(ap->lock, flags);
2925 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2926 ap->pm_mesg.event != PM_EVENT_ON) {
2927 spin_unlock_irqrestore(ap->lock, flags);
2930 spin_unlock_irqrestore(ap->lock, flags);
2932 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
2934 ata_acpi_set_state(ap, PMSG_ON);
2936 if (ap->ops->port_resume)
2937 rc = ap->ops->port_resume(ap);
2939 /* tell ACPI that we're resuming */
2940 ata_acpi_on_resume(ap);
2943 spin_lock_irqsave(ap->lock, flags);
2944 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2945 if (ap->pm_result) {
2946 *ap->pm_result = rc;
2947 ap->pm_result = NULL;
2949 spin_unlock_irqrestore(ap->lock, flags);
2951 #endif /* CONFIG_PM */