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 (!ap
->nr_pmp_links
|| rc
) {
877 /* PMP is not attached or SNTF is not available */
878 if (!ap
->nr_pmp_links
) {
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 /* Reset is represented by combination of actions and EHI
1083 * flags. Suck in all related bits before clearing eh_info to
1084 * avoid losing requested action.
1086 if (action
& ATA_EH_RESET_MASK
) {
1087 ehc
->i
.action
|= ehi
->action
& ATA_EH_RESET_MASK
;
1088 ehc
->i
.flags
|= ehi
->flags
& ATA_EHI_RESET_MODIFIER_MASK
;
1090 /* make sure all reset actions are cleared & clear EHI flags */
1091 action
|= ATA_EH_RESET_MASK
;
1092 ehi
->flags
&= ~ATA_EHI_RESET_MODIFIER_MASK
;
1095 ata_eh_clear_action(link
, dev
, ehi
, action
);
1097 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
))
1098 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1100 spin_unlock_irqrestore(ap
->lock
, flags
);
1104 * ata_eh_done - EH action complete
1105 * @ap: target ATA port
1106 * @dev: target ATA dev for per-dev action (can be NULL)
1107 * @action: action just completed
1109 * Called right after performing EH actions to clear related bits
1110 * in @link->eh_context.
1115 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1116 unsigned int action
)
1118 struct ata_eh_context
*ehc
= &link
->eh_context
;
1120 /* if reset is complete, clear all reset actions & reset modifier */
1121 if (action
& ATA_EH_RESET_MASK
) {
1122 action
|= ATA_EH_RESET_MASK
;
1123 ehc
->i
.flags
&= ~ATA_EHI_RESET_MODIFIER_MASK
;
1126 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1130 * ata_err_string - convert err_mask to descriptive string
1131 * @err_mask: error mask to convert to string
1133 * Convert @err_mask to descriptive string. Errors are
1134 * prioritized according to severity and only the most severe
1135 * error is reported.
1141 * Descriptive string for @err_mask
1143 static const char *ata_err_string(unsigned int err_mask
)
1145 if (err_mask
& AC_ERR_HOST_BUS
)
1146 return "host bus error";
1147 if (err_mask
& AC_ERR_ATA_BUS
)
1148 return "ATA bus error";
1149 if (err_mask
& AC_ERR_TIMEOUT
)
1151 if (err_mask
& AC_ERR_HSM
)
1152 return "HSM violation";
1153 if (err_mask
& AC_ERR_SYSTEM
)
1154 return "internal error";
1155 if (err_mask
& AC_ERR_MEDIA
)
1156 return "media error";
1157 if (err_mask
& AC_ERR_INVALID
)
1158 return "invalid argument";
1159 if (err_mask
& AC_ERR_DEV
)
1160 return "device error";
1161 return "unknown error";
1165 * ata_read_log_page - read a specific log page
1166 * @dev: target device
1167 * @page: page to read
1168 * @buf: buffer to store read page
1169 * @sectors: number of sectors to read
1171 * Read log page using READ_LOG_EXT command.
1174 * Kernel thread context (may sleep).
1177 * 0 on success, AC_ERR_* mask otherwise.
1179 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1180 u8 page
, void *buf
, unsigned int sectors
)
1182 struct ata_taskfile tf
;
1183 unsigned int err_mask
;
1185 DPRINTK("read log page - page %d\n", page
);
1187 ata_tf_init(dev
, &tf
);
1188 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1191 tf
.hob_nsect
= sectors
>> 8;
1192 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1193 tf
.protocol
= ATA_PROT_PIO
;
1195 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1196 buf
, sectors
* ATA_SECT_SIZE
, 0);
1198 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1203 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1204 * @dev: Device to read log page 10h from
1205 * @tag: Resulting tag of the failed command
1206 * @tf: Resulting taskfile registers of the failed command
1208 * Read log page 10h to obtain NCQ error details and clear error
1212 * Kernel thread context (may sleep).
1215 * 0 on success, -errno otherwise.
1217 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1218 int *tag
, struct ata_taskfile
*tf
)
1220 u8
*buf
= dev
->link
->ap
->sector_buf
;
1221 unsigned int err_mask
;
1225 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1230 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1233 ata_dev_printk(dev
, KERN_WARNING
,
1234 "invalid checksum 0x%x on log page 10h\n", csum
);
1239 *tag
= buf
[0] & 0x1f;
1241 tf
->command
= buf
[2];
1242 tf
->feature
= buf
[3];
1246 tf
->device
= buf
[7];
1247 tf
->hob_lbal
= buf
[8];
1248 tf
->hob_lbam
= buf
[9];
1249 tf
->hob_lbah
= buf
[10];
1250 tf
->nsect
= buf
[12];
1251 tf
->hob_nsect
= buf
[13];
1257 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1258 * @dev: device to perform REQUEST_SENSE to
1259 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1261 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1262 * SENSE. This function is EH helper.
1265 * Kernel thread context (may sleep).
1268 * 0 on success, AC_ERR_* mask on failure
1270 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd
*qc
)
1272 struct ata_device
*dev
= qc
->dev
;
1273 unsigned char *sense_buf
= qc
->scsicmd
->sense_buffer
;
1274 struct ata_port
*ap
= dev
->link
->ap
;
1275 struct ata_taskfile tf
;
1276 u8 cdb
[ATAPI_CDB_LEN
];
1278 DPRINTK("ATAPI request sense\n");
1280 /* FIXME: is this needed? */
1281 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1283 /* initialize sense_buf with the error register,
1284 * for the case where they are -not- overwritten
1286 sense_buf
[0] = 0x70;
1287 sense_buf
[2] = qc
->result_tf
.feature
>> 4;
1289 /* some devices time out if garbage left in tf */
1290 ata_tf_init(dev
, &tf
);
1292 memset(cdb
, 0, ATAPI_CDB_LEN
);
1293 cdb
[0] = REQUEST_SENSE
;
1294 cdb
[4] = SCSI_SENSE_BUFFERSIZE
;
1296 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1297 tf
.command
= ATA_CMD_PACKET
;
1299 /* is it pointless to prefer PIO for "safety reasons"? */
1300 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1301 tf
.protocol
= ATAPI_PROT_DMA
;
1302 tf
.feature
|= ATAPI_PKT_DMA
;
1304 tf
.protocol
= ATAPI_PROT_PIO
;
1305 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1309 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1310 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1314 * ata_eh_analyze_serror - analyze SError for a failed port
1315 * @link: ATA link to analyze SError for
1317 * Analyze SError if available and further determine cause of
1323 static void ata_eh_analyze_serror(struct ata_link
*link
)
1325 struct ata_eh_context
*ehc
= &link
->eh_context
;
1326 u32 serror
= ehc
->i
.serror
;
1327 unsigned int err_mask
= 0, action
= 0;
1330 if (serror
& SERR_PERSISTENT
) {
1331 err_mask
|= AC_ERR_ATA_BUS
;
1332 action
|= ATA_EH_HARDRESET
;
1335 (SERR_DATA_RECOVERED
| SERR_COMM_RECOVERED
| SERR_DATA
)) {
1336 err_mask
|= AC_ERR_ATA_BUS
;
1337 action
|= ATA_EH_SOFTRESET
;
1339 if (serror
& SERR_PROTOCOL
) {
1340 err_mask
|= AC_ERR_HSM
;
1341 action
|= ATA_EH_SOFTRESET
;
1343 if (serror
& SERR_INTERNAL
) {
1344 err_mask
|= AC_ERR_SYSTEM
;
1345 action
|= ATA_EH_HARDRESET
;
1348 /* Determine whether a hotplug event has occurred. Both
1349 * SError.N/X are considered hotplug events for enabled or
1350 * host links. For disabled PMP links, only N bit is
1351 * considered as X bit is left at 1 for link plugging.
1355 if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1356 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1358 hotplug_mask
= SERR_PHYRDY_CHG
;
1360 if (serror
& hotplug_mask
)
1361 ata_ehi_hotplugged(&ehc
->i
);
1363 ehc
->i
.err_mask
|= err_mask
;
1364 ehc
->i
.action
|= action
;
1368 * ata_eh_analyze_ncq_error - analyze NCQ error
1369 * @link: ATA link to analyze NCQ error for
1371 * Read log page 10h, determine the offending qc and acquire
1372 * error status TF. For NCQ device errors, all LLDDs have to do
1373 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1377 * Kernel thread context (may sleep).
1379 static void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1381 struct ata_port
*ap
= link
->ap
;
1382 struct ata_eh_context
*ehc
= &link
->eh_context
;
1383 struct ata_device
*dev
= link
->device
;
1384 struct ata_queued_cmd
*qc
;
1385 struct ata_taskfile tf
;
1388 /* if frozen, we can't do much */
1389 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1392 /* is it NCQ device error? */
1393 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1396 /* has LLDD analyzed already? */
1397 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1398 qc
= __ata_qc_from_tag(ap
, tag
);
1400 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1407 /* okay, this error is ours */
1408 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1410 ata_link_printk(link
, KERN_ERR
, "failed to read log page 10h "
1411 "(errno=%d)\n", rc
);
1415 if (!(link
->sactive
& (1 << tag
))) {
1416 ata_link_printk(link
, KERN_ERR
, "log page 10h reported "
1417 "inactive tag %d\n", tag
);
1421 /* we've got the perpetrator, condemn it */
1422 qc
= __ata_qc_from_tag(ap
, tag
);
1423 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1424 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1425 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1429 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1430 * @qc: qc to analyze
1431 * @tf: Taskfile registers to analyze
1433 * Analyze taskfile of @qc and further determine cause of
1434 * failure. This function also requests ATAPI sense data if
1438 * Kernel thread context (may sleep).
1441 * Determined recovery action
1443 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1444 const struct ata_taskfile
*tf
)
1446 unsigned int tmp
, action
= 0;
1447 u8 stat
= tf
->command
, err
= tf
->feature
;
1449 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1450 qc
->err_mask
|= AC_ERR_HSM
;
1451 return ATA_EH_SOFTRESET
;
1454 if (stat
& (ATA_ERR
| ATA_DF
))
1455 qc
->err_mask
|= AC_ERR_DEV
;
1459 switch (qc
->dev
->class) {
1462 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1464 qc
->err_mask
|= AC_ERR_MEDIA
;
1466 qc
->err_mask
|= AC_ERR_INVALID
;
1470 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1471 tmp
= atapi_eh_request_sense(qc
);
1473 /* ATA_QCFLAG_SENSE_VALID is used to
1474 * tell atapi_qc_complete() that sense
1475 * data is already valid.
1477 * TODO: interpret sense data and set
1478 * appropriate err_mask.
1480 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1482 qc
->err_mask
|= tmp
;
1486 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1487 action
|= ATA_EH_SOFTRESET
;
1492 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1497 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1501 base
= ATA_ECAT_DUBIOUS_NONE
;
1503 if (err_mask
& AC_ERR_ATA_BUS
)
1504 return base
+ ATA_ECAT_ATA_BUS
;
1506 if (err_mask
& AC_ERR_TIMEOUT
)
1507 return base
+ ATA_ECAT_TOUT_HSM
;
1509 if (eflags
& ATA_EFLAG_IS_IO
) {
1510 if (err_mask
& AC_ERR_HSM
)
1511 return base
+ ATA_ECAT_TOUT_HSM
;
1513 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1514 return base
+ ATA_ECAT_UNK_DEV
;
1520 struct speed_down_verdict_arg
{
1523 int nr_errors
[ATA_ECAT_NR
];
1526 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1528 struct speed_down_verdict_arg
*arg
= void_arg
;
1531 if (ent
->timestamp
< arg
->since
)
1534 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1536 arg
->nr_errors
[cat
]++;
1542 * ata_eh_speed_down_verdict - Determine speed down verdict
1543 * @dev: Device of interest
1545 * This function examines error ring of @dev and determines
1546 * whether NCQ needs to be turned off, transfer speed should be
1547 * stepped down, or falling back to PIO is necessary.
1549 * ECAT_ATA_BUS : ATA_BUS error for any command
1551 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1554 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1556 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1557 * data transfer hasn't been verified.
1561 * NCQ_OFF : Turn off NCQ.
1563 * SPEED_DOWN : Speed down transfer speed but don't fall back
1566 * FALLBACK_TO_PIO : Fall back to PIO.
1568 * Even if multiple verdicts are returned, only one action is
1569 * taken per error. An action triggered by non-DUBIOUS errors
1570 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1571 * This is to expedite speed down decisions right after device is
1572 * initially configured.
1574 * The followings are speed down rules. #1 and #2 deal with
1577 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1578 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1580 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1581 * occurred during last 5 mins, NCQ_OFF.
1583 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1584 * ocurred during last 5 mins, FALLBACK_TO_PIO
1586 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1587 * during last 10 mins, NCQ_OFF.
1589 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1590 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1593 * Inherited from caller.
1596 * OR of ATA_EH_SPDN_* flags.
1598 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1600 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1601 u64 j64
= get_jiffies_64();
1602 struct speed_down_verdict_arg arg
;
1603 unsigned int verdict
= 0;
1605 /* scan past 5 mins of error history */
1606 memset(&arg
, 0, sizeof(arg
));
1607 arg
.since
= j64
- min(j64
, j5mins
);
1608 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1610 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1611 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1612 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1613 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1615 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1616 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1617 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1619 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1620 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1621 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1622 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1624 /* scan past 10 mins of error history */
1625 memset(&arg
, 0, sizeof(arg
));
1626 arg
.since
= j64
- min(j64
, j10mins
);
1627 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1629 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1630 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1631 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1633 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1634 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1635 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1636 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1642 * ata_eh_speed_down - record error and speed down if necessary
1643 * @dev: Failed device
1644 * @eflags: mask of ATA_EFLAG_* flags
1645 * @err_mask: err_mask of the error
1647 * Record error and examine error history to determine whether
1648 * adjusting transmission speed is necessary. It also sets
1649 * transmission limits appropriately if such adjustment is
1653 * Kernel thread context (may sleep).
1656 * Determined recovery action.
1658 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
1659 unsigned int eflags
, unsigned int err_mask
)
1661 struct ata_link
*link
= dev
->link
;
1663 unsigned int verdict
;
1664 unsigned int action
= 0;
1666 /* don't bother if Cat-0 error */
1667 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
1670 /* record error and determine whether speed down is necessary */
1671 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
1672 verdict
= ata_eh_speed_down_verdict(dev
);
1675 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1676 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1677 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1678 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1679 ata_dev_printk(dev
, KERN_WARNING
,
1680 "NCQ disabled due to excessive errors\n");
1685 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1686 /* speed down SATA link speed if possible */
1687 if (sata_down_spd_limit(link
) == 0) {
1688 action
|= ATA_EH_HARDRESET
;
1692 /* lower transfer mode */
1693 if (dev
->spdn_cnt
< 2) {
1694 static const int dma_dnxfer_sel
[] =
1695 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1696 static const int pio_dnxfer_sel
[] =
1697 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1700 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1701 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1703 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1707 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1708 action
|= ATA_EH_SOFTRESET
;
1714 /* Fall back to PIO? Slowing down to PIO is meaningless for
1715 * SATA ATA devices. Consider it only for PATA and SATAPI.
1717 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1718 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
1719 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1720 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1722 action
|= ATA_EH_SOFTRESET
;
1729 /* device has been slowed down, blow error history */
1730 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
1731 ata_ering_clear(&dev
->ering
);
1736 * ata_eh_link_autopsy - analyze error and determine recovery action
1737 * @link: host link to perform autopsy on
1739 * Analyze why @link failed and determine which recovery actions
1740 * are needed. This function also sets more detailed AC_ERR_*
1741 * values and fills sense data for ATAPI CHECK SENSE.
1744 * Kernel thread context (may sleep).
1746 static void ata_eh_link_autopsy(struct ata_link
*link
)
1748 struct ata_port
*ap
= link
->ap
;
1749 struct ata_eh_context
*ehc
= &link
->eh_context
;
1750 struct ata_device
*dev
;
1751 unsigned int all_err_mask
= 0, eflags
= 0;
1758 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1761 /* obtain and analyze SError */
1762 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
1764 ehc
->i
.serror
|= serror
;
1765 ata_eh_analyze_serror(link
);
1766 } else if (rc
!= -EOPNOTSUPP
) {
1767 /* SError read failed, force hardreset and probing */
1768 ata_ehi_schedule_probe(&ehc
->i
);
1769 ehc
->i
.action
|= ATA_EH_HARDRESET
;
1770 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
1773 /* analyze NCQ failure */
1774 ata_eh_analyze_ncq_error(link
);
1776 /* any real error trumps AC_ERR_OTHER */
1777 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1778 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1780 all_err_mask
|= ehc
->i
.err_mask
;
1782 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1783 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1785 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || qc
->dev
->link
!= link
)
1788 /* inherit upper level err_mask */
1789 qc
->err_mask
|= ehc
->i
.err_mask
;
1792 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1794 /* DEV errors are probably spurious in case of ATA_BUS error */
1795 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1796 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1799 /* any real error trumps unknown error */
1800 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1801 qc
->err_mask
&= ~AC_ERR_OTHER
;
1803 /* SENSE_VALID trumps dev/unknown error and revalidation */
1804 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
1805 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1807 /* accumulate error info */
1808 ehc
->i
.dev
= qc
->dev
;
1809 all_err_mask
|= qc
->err_mask
;
1810 if (qc
->flags
& ATA_QCFLAG_IO
)
1811 eflags
|= ATA_EFLAG_IS_IO
;
1814 /* enforce default EH actions */
1815 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
1816 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1817 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
1818 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
1819 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
1820 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1822 /* If we have offending qcs and the associated failed device,
1823 * perform per-dev EH action only on the offending device.
1826 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
1827 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
1828 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
1831 /* propagate timeout to host link */
1832 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
1833 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
1835 /* record error and consider speeding down */
1837 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
1838 ata_dev_enabled(link
->device
))))
1842 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
1843 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
1844 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
1851 * ata_eh_autopsy - analyze error and determine recovery action
1852 * @ap: host port to perform autopsy on
1854 * Analyze all links of @ap and determine why they failed and
1855 * which recovery actions are needed.
1858 * Kernel thread context (may sleep).
1860 void ata_eh_autopsy(struct ata_port
*ap
)
1862 struct ata_link
*link
;
1864 ata_port_for_each_link(link
, ap
)
1865 ata_eh_link_autopsy(link
);
1867 /* Autopsy of fanout ports can affect host link autopsy.
1868 * Perform host link autopsy last.
1870 if (ap
->nr_pmp_links
)
1871 ata_eh_link_autopsy(&ap
->link
);
1875 * ata_eh_link_report - report error handling to user
1876 * @link: ATA link EH is going on
1878 * Report EH to user.
1883 static void ata_eh_link_report(struct ata_link
*link
)
1885 struct ata_port
*ap
= link
->ap
;
1886 struct ata_eh_context
*ehc
= &link
->eh_context
;
1887 const char *frozen
, *desc
;
1889 int tag
, nr_failed
= 0;
1891 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
1895 if (ehc
->i
.desc
[0] != '\0')
1898 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1899 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1901 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) || qc
->dev
->link
!= link
||
1902 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
1903 qc
->err_mask
== AC_ERR_DEV
))
1905 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
1911 if (!nr_failed
&& !ehc
->i
.err_mask
)
1915 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1918 memset(tries_buf
, 0, sizeof(tries_buf
));
1919 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
1920 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
1924 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
1925 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1926 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
1927 ehc
->i
.action
, frozen
, tries_buf
);
1929 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "%s\n", desc
);
1931 ata_link_printk(link
, KERN_ERR
, "exception Emask 0x%x "
1932 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
1933 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
1934 ehc
->i
.action
, frozen
, tries_buf
);
1936 ata_link_printk(link
, KERN_ERR
, "%s\n", desc
);
1940 ata_port_printk(ap
, KERN_ERR
,
1941 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
1942 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
1943 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
1944 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
1945 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
1946 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
1947 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
1948 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
1949 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
1950 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
1951 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
1952 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
1953 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
1954 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
1955 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
1956 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
1957 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
1958 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
1960 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1961 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1962 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
1963 const u8
*cdb
= qc
->cdb
;
1964 char data_buf
[20] = "";
1965 char cdb_buf
[70] = "";
1967 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
1968 qc
->dev
->link
!= link
|| !qc
->err_mask
)
1971 if (qc
->dma_dir
!= DMA_NONE
) {
1972 static const char *dma_str
[] = {
1973 [DMA_BIDIRECTIONAL
] = "bidi",
1974 [DMA_TO_DEVICE
] = "out",
1975 [DMA_FROM_DEVICE
] = "in",
1977 static const char *prot_str
[] = {
1978 [ATA_PROT_PIO
] = "pio",
1979 [ATA_PROT_DMA
] = "dma",
1980 [ATA_PROT_NCQ
] = "ncq",
1981 [ATAPI_PROT_PIO
] = "pio",
1982 [ATAPI_PROT_DMA
] = "dma",
1985 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
1986 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
1987 dma_str
[qc
->dma_dir
]);
1990 if (ata_is_atapi(qc
->tf
.protocol
))
1991 snprintf(cdb_buf
, sizeof(cdb_buf
),
1992 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
1993 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
1994 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
1995 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
1996 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
1997 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
1999 ata_dev_printk(qc
->dev
, KERN_ERR
,
2000 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2002 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2003 "Emask 0x%x (%s)%s\n",
2004 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2005 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2006 cmd
->hob_feature
, cmd
->hob_nsect
,
2007 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2008 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2009 res
->command
, res
->feature
, res
->nsect
,
2010 res
->lbal
, res
->lbam
, res
->lbah
,
2011 res
->hob_feature
, res
->hob_nsect
,
2012 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2013 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2014 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2016 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2018 if (res
->command
& ATA_BUSY
)
2019 ata_dev_printk(qc
->dev
, KERN_ERR
,
2020 "status: { Busy }\n");
2022 ata_dev_printk(qc
->dev
, KERN_ERR
,
2023 "status: { %s%s%s%s}\n",
2024 res
->command
& ATA_DRDY
? "DRDY " : "",
2025 res
->command
& ATA_DF
? "DF " : "",
2026 res
->command
& ATA_DRQ
? "DRQ " : "",
2027 res
->command
& ATA_ERR
? "ERR " : "");
2030 if (cmd
->command
!= ATA_CMD_PACKET
&&
2031 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2033 ata_dev_printk(qc
->dev
, KERN_ERR
,
2034 "error: { %s%s%s%s}\n",
2035 res
->feature
& ATA_ICRC
? "ICRC " : "",
2036 res
->feature
& ATA_UNC
? "UNC " : "",
2037 res
->feature
& ATA_IDNF
? "IDNF " : "",
2038 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2043 * ata_eh_report - report error handling to user
2044 * @ap: ATA port to report EH about
2046 * Report EH to user.
2051 void ata_eh_report(struct ata_port
*ap
)
2053 struct ata_link
*link
;
2055 __ata_port_for_each_link(link
, ap
)
2056 ata_eh_link_report(link
);
2059 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2060 unsigned int *classes
, unsigned long deadline
)
2062 struct ata_device
*dev
;
2065 ata_link_for_each_dev(dev
, link
)
2066 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2068 rc
= reset(link
, classes
, deadline
);
2072 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
2073 * is complete and convert all ATA_DEV_UNKNOWN to
2076 ata_link_for_each_dev(dev
, link
)
2077 if (classes
[dev
->devno
] != ATA_DEV_UNKNOWN
)
2081 ata_link_for_each_dev(dev
, link
) {
2082 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
)
2083 classes
[dev
->devno
] = ATA_DEV_NONE
;
2090 static int ata_eh_followup_srst_needed(struct ata_link
*link
,
2091 int rc
, int classify
,
2092 const unsigned int *classes
)
2094 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2100 if ((link
->ap
->flags
& ATA_FLAG_PMP
) && ata_is_host_link(link
))
2102 if (classify
&& !(link
->flags
& ATA_LFLAG_ASSUME_CLASS
) &&
2103 classes
[0] == ATA_DEV_UNKNOWN
)
2108 int ata_eh_reset(struct ata_link
*link
, int classify
,
2109 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2110 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2112 const int max_tries
= ARRAY_SIZE(ata_eh_reset_timeouts
);
2113 struct ata_port
*ap
= link
->ap
;
2114 struct ata_eh_context
*ehc
= &link
->eh_context
;
2115 unsigned int *classes
= ehc
->classes
;
2116 unsigned int lflags
= link
->flags
;
2117 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2119 struct ata_device
*dev
;
2120 unsigned long deadline
, now
;
2121 unsigned int tmp_action
;
2122 ata_reset_fn_t reset
;
2123 unsigned long flags
;
2127 /* about to reset */
2128 spin_lock_irqsave(ap
->lock
, flags
);
2129 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2130 spin_unlock_irqrestore(ap
->lock
, flags
);
2132 ata_eh_about_to_do(link
, NULL
, ehc
->i
.action
& ATA_EH_RESET_MASK
);
2134 ata_link_for_each_dev(dev
, link
) {
2135 /* If we issue an SRST then an ATA drive (not ATAPI)
2136 * may change configuration and be in PIO0 timing. If
2137 * we do a hard reset (or are coming from power on)
2138 * this is true for ATA or ATAPI. Until we've set a
2139 * suitable controller mode we should not touch the
2140 * bus as we may be talking too fast.
2142 dev
->pio_mode
= XFER_PIO_0
;
2144 /* If the controller has a pio mode setup function
2145 * then use it to set the chipset to rights. Don't
2146 * touch the DMA setup as that will be dealt with when
2147 * configuring devices.
2149 if (ap
->ops
->set_piomode
)
2150 ap
->ops
->set_piomode(ap
, dev
);
2153 /* Determine which reset to use and record in ehc->i.action.
2154 * prereset() may examine and modify it.
2156 if (softreset
&& (!hardreset
|| (!(lflags
& ATA_LFLAG_NO_SRST
) &&
2157 !sata_set_spd_needed(link
) &&
2158 !(ehc
->i
.action
& ATA_EH_HARDRESET
))))
2159 tmp_action
= ATA_EH_SOFTRESET
;
2161 tmp_action
= ATA_EH_HARDRESET
;
2163 ehc
->i
.action
= (ehc
->i
.action
& ~ATA_EH_RESET_MASK
) | tmp_action
;
2166 rc
= prereset(link
, jiffies
+ ATA_EH_PRERESET_TIMEOUT
);
2168 if (rc
== -ENOENT
) {
2169 ata_link_printk(link
, KERN_DEBUG
,
2170 "port disabled. ignoring.\n");
2171 ehc
->i
.action
&= ~ATA_EH_RESET_MASK
;
2173 ata_link_for_each_dev(dev
, link
)
2174 classes
[dev
->devno
] = ATA_DEV_NONE
;
2178 ata_link_printk(link
, KERN_ERR
,
2179 "prereset failed (errno=%d)\n", rc
);
2184 /* prereset() might have modified ehc->i.action */
2185 if (ehc
->i
.action
& ATA_EH_HARDRESET
)
2187 else if (ehc
->i
.action
& ATA_EH_SOFTRESET
)
2190 /* prereset told us not to reset, bang classes and return */
2191 ata_link_for_each_dev(dev
, link
)
2192 classes
[dev
->devno
] = ATA_DEV_NONE
;
2197 /* did prereset() screw up? if so, fix up to avoid oopsing */
2206 deadline
= jiffies
+ ata_eh_reset_timeouts
[try++];
2208 /* shut up during boot probing */
2210 ata_link_printk(link
, KERN_INFO
, "%s resetting link\n",
2211 reset
== softreset
? "soft" : "hard");
2213 /* mark that this EH session started with reset */
2214 if (reset
== hardreset
)
2215 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2217 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2219 rc
= ata_do_reset(link
, reset
, classes
, deadline
);
2221 if (reset
== hardreset
&&
2222 ata_eh_followup_srst_needed(link
, rc
, classify
, classes
)) {
2223 /* okay, let's do follow-up softreset */
2227 ata_link_printk(link
, KERN_ERR
,
2228 "follow-up softreset required "
2229 "but no softreset avaliable\n");
2234 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET_MASK
);
2235 rc
= ata_do_reset(link
, reset
, classes
, deadline
);
2238 /* -EAGAIN can happen if we skipped followup SRST */
2239 if (rc
&& rc
!= -EAGAIN
)
2242 /* was classification successful? */
2243 if (classify
&& classes
[0] == ATA_DEV_UNKNOWN
&&
2244 !(lflags
& ATA_LFLAG_ASSUME_CLASS
)) {
2245 if (try < max_tries
) {
2246 ata_link_printk(link
, KERN_WARNING
,
2247 "classification failed\n");
2252 ata_link_printk(link
, KERN_WARNING
,
2253 "classfication failed, assuming ATA\n");
2254 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2257 ata_link_for_each_dev(dev
, link
) {
2258 /* After the reset, the device state is PIO 0 and the
2259 * controller state is undefined. Reset also wakes up
2260 * drives from sleeping mode.
2262 dev
->pio_mode
= XFER_PIO_0
;
2263 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2265 if (ata_link_offline(link
))
2268 /* apply class override */
2269 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2270 classes
[dev
->devno
] = ATA_DEV_ATA
;
2271 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2272 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
; /* not yet */
2275 /* record current link speed */
2276 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2277 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2280 postreset(link
, classes
);
2282 /* reset successful, schedule revalidation */
2283 ata_eh_done(link
, NULL
, ehc
->i
.action
& ATA_EH_RESET_MASK
);
2284 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2288 /* clear hotplug flag */
2289 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2291 spin_lock_irqsave(ap
->lock
, flags
);
2292 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2293 spin_unlock_irqrestore(ap
->lock
, flags
);
2298 if (rc
== -ERESTART
|| try >= max_tries
)
2302 if (time_before(now
, deadline
)) {
2303 unsigned long delta
= deadline
- now
;
2305 ata_link_printk(link
, KERN_WARNING
, "reset failed "
2306 "(errno=%d), retrying in %u secs\n",
2307 rc
, (jiffies_to_msecs(delta
) + 999) / 1000);
2310 delta
= schedule_timeout_uninterruptible(delta
);
2313 if (rc
== -EPIPE
|| try == max_tries
- 1)
2314 sata_down_spd_limit(link
);
2320 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
2321 struct ata_device
**r_failed_dev
)
2323 struct ata_port
*ap
= link
->ap
;
2324 struct ata_eh_context
*ehc
= &link
->eh_context
;
2325 struct ata_device
*dev
;
2326 unsigned int new_mask
= 0;
2327 unsigned long flags
;
2332 /* For PATA drive side cable detection to work, IDENTIFY must
2333 * be done backwards such that PDIAG- is released by the slave
2334 * device before the master device is identified.
2336 ata_link_for_each_dev_reverse(dev
, link
) {
2337 unsigned int action
= ata_eh_dev_action(dev
);
2338 unsigned int readid_flags
= 0;
2340 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2341 readid_flags
|= ATA_READID_POSTRESET
;
2343 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
2344 WARN_ON(dev
->class == ATA_DEV_PMP
);
2346 if (ata_link_offline(link
)) {
2351 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
2352 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
2357 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
2359 /* Configuration may have changed, reconfigure
2362 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2364 /* schedule the scsi_rescan_device() here */
2365 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
2366 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
2367 ehc
->tries
[dev
->devno
] &&
2368 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
2369 dev
->class = ehc
->classes
[dev
->devno
];
2371 if (dev
->class == ATA_DEV_PMP
)
2372 rc
= sata_pmp_attach(dev
);
2374 rc
= ata_dev_read_id(dev
, &dev
->class,
2375 readid_flags
, dev
->id
);
2378 new_mask
|= 1 << dev
->devno
;
2381 /* IDENTIFY was issued to non-existent
2382 * device. No need to reset. Just
2383 * thaw and kill the device.
2385 ata_eh_thaw_port(ap
);
2386 dev
->class = ATA_DEV_UNKNOWN
;
2389 dev
->class = ATA_DEV_UNKNOWN
;
2395 /* PDIAG- should have been released, ask cable type if post-reset */
2396 if (ata_is_host_link(link
) && ap
->ops
->cable_detect
&&
2397 (ehc
->i
.flags
& ATA_EHI_DID_RESET
))
2398 ap
->cbl
= ap
->ops
->cable_detect(ap
);
2400 /* Configure new devices forward such that user doesn't see
2401 * device detection messages backwards.
2403 ata_link_for_each_dev(dev
, link
) {
2404 if (!(new_mask
& (1 << dev
->devno
)) ||
2405 dev
->class == ATA_DEV_PMP
)
2408 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
2409 rc
= ata_dev_configure(dev
);
2410 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
2414 spin_lock_irqsave(ap
->lock
, flags
);
2415 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
2416 spin_unlock_irqrestore(ap
->lock
, flags
);
2418 /* new device discovered, configure xfermode */
2419 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2425 *r_failed_dev
= dev
;
2426 DPRINTK("EXIT rc=%d\n", rc
);
2431 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2432 * @link: link on which timings will be programmed
2433 * @r_failed_dev: out paramter for failed device
2435 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2436 * ata_set_mode() fails, pointer to the failing device is
2437 * returned in @r_failed_dev.
2440 * PCI/etc. bus probe sem.
2443 * 0 on success, negative errno otherwise
2445 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
2447 struct ata_port
*ap
= link
->ap
;
2448 struct ata_device
*dev
;
2451 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2452 ata_link_for_each_dev(dev
, link
) {
2453 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
2454 struct ata_ering_entry
*ent
;
2456 ent
= ata_ering_top(&dev
->ering
);
2458 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
2462 /* has private set_mode? */
2463 if (ap
->ops
->set_mode
)
2464 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
2466 rc
= ata_do_set_mode(link
, r_failed_dev
);
2468 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2469 ata_link_for_each_dev(dev
, link
) {
2470 struct ata_eh_context
*ehc
= &link
->eh_context
;
2471 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
2472 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
2474 if (dev
->xfer_mode
!= saved_xfer_mode
||
2475 ata_ncq_enabled(dev
) != saved_ncq
)
2476 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
2482 static int ata_link_nr_enabled(struct ata_link
*link
)
2484 struct ata_device
*dev
;
2487 ata_link_for_each_dev(dev
, link
)
2488 if (ata_dev_enabled(dev
))
2493 static int ata_link_nr_vacant(struct ata_link
*link
)
2495 struct ata_device
*dev
;
2498 ata_link_for_each_dev(dev
, link
)
2499 if (dev
->class == ATA_DEV_UNKNOWN
)
2504 static int ata_eh_skip_recovery(struct ata_link
*link
)
2506 struct ata_eh_context
*ehc
= &link
->eh_context
;
2507 struct ata_device
*dev
;
2509 /* skip disabled links */
2510 if (link
->flags
& ATA_LFLAG_DISABLED
)
2513 /* thaw frozen port, resume link and recover failed devices */
2514 if ((link
->ap
->pflags
& ATA_PFLAG_FROZEN
) ||
2515 (ehc
->i
.flags
& ATA_EHI_RESUME_LINK
) || ata_link_nr_enabled(link
))
2518 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2519 ata_link_for_each_dev(dev
, link
) {
2520 if (dev
->class == ATA_DEV_UNKNOWN
&&
2521 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
2528 static int ata_eh_schedule_probe(struct ata_device
*dev
)
2530 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2532 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
2533 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
2536 ata_eh_detach_dev(dev
);
2538 ehc
->did_probe_mask
|= (1 << dev
->devno
);
2539 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2540 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
2541 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
2546 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
2548 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2550 ehc
->tries
[dev
->devno
]--;
2554 /* device missing or wrong IDENTIFY data, schedule probing */
2555 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
2557 /* give it just one more chance */
2558 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
2560 if (ehc
->tries
[dev
->devno
] == 1 && dev
->pio_mode
> XFER_PIO_0
) {
2561 /* This is the last chance, better to slow
2562 * down than lose it.
2564 sata_down_spd_limit(dev
->link
);
2565 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
2569 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
2570 /* disable device if it has used up all its chances */
2571 ata_dev_disable(dev
);
2573 /* detach if offline */
2574 if (ata_link_offline(dev
->link
))
2575 ata_eh_detach_dev(dev
);
2577 /* schedule probe if necessary */
2578 if (ata_eh_schedule_probe(dev
))
2579 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
2583 /* soft didn't work? be haaaaard */
2584 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2585 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2587 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2594 * ata_eh_recover - recover host port after error
2595 * @ap: host port to recover
2596 * @prereset: prereset method (can be NULL)
2597 * @softreset: softreset method (can be NULL)
2598 * @hardreset: hardreset method (can be NULL)
2599 * @postreset: postreset method (can be NULL)
2600 * @r_failed_link: out parameter for failed link
2602 * This is the alpha and omega, eum and yang, heart and soul of
2603 * libata exception handling. On entry, actions required to
2604 * recover each link and hotplug requests are recorded in the
2605 * link's eh_context. This function executes all the operations
2606 * with appropriate retrials and fallbacks to resurrect failed
2607 * devices, detach goners and greet newcomers.
2610 * Kernel thread context (may sleep).
2613 * 0 on success, -errno on failure.
2615 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
2616 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
2617 ata_postreset_fn_t postreset
,
2618 struct ata_link
**r_failed_link
)
2620 struct ata_link
*link
;
2621 struct ata_device
*dev
;
2622 int nr_failed_devs
, nr_disabled_devs
;
2624 unsigned long flags
;
2628 /* prep for recovery */
2629 ata_port_for_each_link(link
, ap
) {
2630 struct ata_eh_context
*ehc
= &link
->eh_context
;
2632 /* re-enable link? */
2633 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
2634 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
2635 spin_lock_irqsave(ap
->lock
, flags
);
2636 link
->flags
&= ~ATA_LFLAG_DISABLED
;
2637 spin_unlock_irqrestore(ap
->lock
, flags
);
2638 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
2641 ata_link_for_each_dev(dev
, link
) {
2642 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
2643 ehc
->tries
[dev
->devno
] = 1;
2645 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
2647 /* collect port action mask recorded in dev actions */
2648 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
2649 ~ATA_EH_PERDEV_MASK
;
2650 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
2652 /* process hotplug request */
2653 if (dev
->flags
& ATA_DFLAG_DETACH
)
2654 ata_eh_detach_dev(dev
);
2656 /* schedule probe if necessary */
2657 if (!ata_dev_enabled(dev
))
2658 ata_eh_schedule_probe(dev
);
2665 nr_disabled_devs
= 0;
2668 /* if UNLOADING, finish immediately */
2669 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
2673 ata_port_for_each_link(link
, ap
) {
2674 struct ata_eh_context
*ehc
= &link
->eh_context
;
2676 /* skip EH if possible. */
2677 if (ata_eh_skip_recovery(link
))
2680 /* do we need to reset? */
2681 if (ehc
->i
.action
& ATA_EH_RESET_MASK
)
2684 ata_link_for_each_dev(dev
, link
)
2685 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2690 /* if PMP is attached, this function only deals with
2691 * downstream links, port should stay thawed.
2693 if (!ap
->nr_pmp_links
)
2694 ata_eh_freeze_port(ap
);
2696 ata_port_for_each_link(link
, ap
) {
2697 struct ata_eh_context
*ehc
= &link
->eh_context
;
2699 if (!(ehc
->i
.action
& ATA_EH_RESET_MASK
))
2702 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
2703 prereset
, softreset
, hardreset
,
2706 ata_link_printk(link
, KERN_ERR
,
2707 "reset failed, giving up\n");
2712 if (!ap
->nr_pmp_links
)
2713 ata_eh_thaw_port(ap
);
2717 ata_port_for_each_link(link
, ap
) {
2718 struct ata_eh_context
*ehc
= &link
->eh_context
;
2720 /* revalidate existing devices and attach new ones */
2721 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
2725 /* if PMP got attached, return, pmp EH will take care of it */
2726 if (link
->device
->class == ATA_DEV_PMP
) {
2731 /* configure transfer mode if necessary */
2732 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
2733 rc
= ata_set_mode(link
, &dev
);
2736 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
2739 if (ehc
->i
.action
& ATA_EHI_LPM
)
2740 ata_link_for_each_dev(dev
, link
)
2741 ata_dev_enable_pm(dev
, ap
->pm_policy
);
2743 /* this link is okay now */
2749 if (ata_eh_handle_dev_fail(dev
, rc
))
2752 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
2753 /* PMP reset requires working host port.
2754 * Can't retry if it's frozen.
2756 if (ap
->nr_pmp_links
)
2762 if (nr_failed_devs
) {
2763 if (nr_failed_devs
!= nr_disabled_devs
) {
2764 ata_port_printk(ap
, KERN_WARNING
, "failed to recover "
2765 "some devices, retrying in 5 secs\n");
2768 /* no device left to recover, repeat fast */
2776 if (rc
&& r_failed_link
)
2777 *r_failed_link
= link
;
2779 DPRINTK("EXIT, rc=%d\n", rc
);
2784 * ata_eh_finish - finish up EH
2785 * @ap: host port to finish EH for
2787 * Recovery is complete. Clean up EH states and retry or finish
2793 void ata_eh_finish(struct ata_port
*ap
)
2797 /* retry or finish qcs */
2798 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2799 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2801 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
2805 /* FIXME: Once EH migration is complete,
2806 * generate sense data in this function,
2807 * considering both err_mask and tf.
2809 * There's no point in retrying invalid
2810 * (detected by libata) and non-IO device
2811 * errors (rejected by device). Finish them
2814 if ((qc
->err_mask
& AC_ERR_INVALID
) ||
2815 (!(qc
->flags
& ATA_QCFLAG_IO
) &&
2816 qc
->err_mask
== AC_ERR_DEV
))
2817 ata_eh_qc_complete(qc
);
2819 ata_eh_qc_retry(qc
);
2821 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
2822 ata_eh_qc_complete(qc
);
2824 /* feed zero TF to sense generation */
2825 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
2826 ata_eh_qc_retry(qc
);
2831 /* make sure nr_active_links is zero after EH */
2832 WARN_ON(ap
->nr_active_links
);
2833 ap
->nr_active_links
= 0;
2837 * ata_do_eh - do standard error handling
2838 * @ap: host port to handle error for
2839 * @prereset: prereset method (can be NULL)
2840 * @softreset: softreset method (can be NULL)
2841 * @hardreset: hardreset method (can be NULL)
2842 * @postreset: postreset method (can be NULL)
2844 * Perform standard error handling sequence.
2847 * Kernel thread context (may sleep).
2849 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
2850 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
2851 ata_postreset_fn_t postreset
)
2853 struct ata_device
*dev
;
2859 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
2862 ata_link_for_each_dev(dev
, &ap
->link
)
2863 ata_dev_disable(dev
);
2871 * ata_eh_handle_port_suspend - perform port suspend operation
2872 * @ap: port to suspend
2877 * Kernel thread context (may sleep).
2879 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
2881 unsigned long flags
;
2884 /* are we suspending? */
2885 spin_lock_irqsave(ap
->lock
, flags
);
2886 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
2887 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
2888 spin_unlock_irqrestore(ap
->lock
, flags
);
2891 spin_unlock_irqrestore(ap
->lock
, flags
);
2893 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
2895 /* tell ACPI we're suspending */
2896 rc
= ata_acpi_on_suspend(ap
);
2901 ata_eh_freeze_port(ap
);
2903 if (ap
->ops
->port_suspend
)
2904 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
2906 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
2909 spin_lock_irqsave(ap
->lock
, flags
);
2911 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
2913 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
2914 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2915 ata_port_schedule_eh(ap
);
2917 if (ap
->pm_result
) {
2918 *ap
->pm_result
= rc
;
2919 ap
->pm_result
= NULL
;
2922 spin_unlock_irqrestore(ap
->lock
, flags
);
2928 * ata_eh_handle_port_resume - perform port resume operation
2929 * @ap: port to resume
2934 * Kernel thread context (may sleep).
2936 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
2938 unsigned long flags
;
2941 /* are we resuming? */
2942 spin_lock_irqsave(ap
->lock
, flags
);
2943 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
2944 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
2945 spin_unlock_irqrestore(ap
->lock
, flags
);
2948 spin_unlock_irqrestore(ap
->lock
, flags
);
2950 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
2952 ata_acpi_set_state(ap
, PMSG_ON
);
2954 if (ap
->ops
->port_resume
)
2955 rc
= ap
->ops
->port_resume(ap
);
2957 /* tell ACPI that we're resuming */
2958 ata_acpi_on_resume(ap
);
2961 spin_lock_irqsave(ap
->lock
, flags
);
2962 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
2963 if (ap
->pm_result
) {
2964 *ap
->pm_result
= rc
;
2965 ap
->pm_result
= NULL
;
2967 spin_unlock_irqrestore(ap
->lock
, flags
);
2969 #endif /* CONFIG_PM */