2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include "../scsi/scsi_transport_api.h"
45 #include <linux/libata.h>
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
57 ATA_EFLAG_IS_IO
= (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
60 /* error categories */
63 ATA_ECAT_TOUT_HSM
= 2,
65 ATA_ECAT_DUBIOUS_NONE
= 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
71 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN
= 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT
= 10000,
82 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
87 /* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
93 static const unsigned long ata_eh_reset_timeouts
[] = {
94 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
98 ULONG_MAX
, /* > 1 min has elapsed, give up */
101 static const unsigned long ata_eh_identify_timeouts
[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
108 static const unsigned long ata_eh_other_timeouts
[] = {
109 5000, /* same rationale as identify timeout */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
115 struct ata_eh_cmd_timeout_ent
{
117 const unsigned long *timeouts
;
120 /* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
132 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
133 static const struct ata_eh_cmd_timeout_ent
134 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
135 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
136 .timeouts
= ata_eh_identify_timeouts
, },
137 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
138 .timeouts
= ata_eh_other_timeouts
, },
139 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
140 .timeouts
= ata_eh_other_timeouts
, },
141 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
142 .timeouts
= ata_eh_other_timeouts
, },
143 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
144 .timeouts
= ata_eh_other_timeouts
, },
148 static void __ata_port_freeze(struct ata_port
*ap
);
150 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
151 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
152 #else /* CONFIG_PM */
153 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
156 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
158 #endif /* CONFIG_PM */
160 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
163 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
164 ATA_EH_DESC_LEN
- ehi
->desc_len
,
169 * __ata_ehi_push_desc - push error description without adding separator
171 * @fmt: printf format string
173 * Format string according to @fmt and append it to @ehi->desc.
176 * spin_lock_irqsave(host lock)
178 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
183 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
188 * ata_ehi_push_desc - push error description with separator
190 * @fmt: printf format string
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
196 * spin_lock_irqsave(host lock)
198 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
203 __ata_ehi_push_desc(ehi
, ", ");
206 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
211 * ata_ehi_clear_desc - clean error description
217 * spin_lock_irqsave(host lock)
219 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
238 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
242 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
244 if (ap
->link
.eh_info
.desc_len
)
245 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
248 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
269 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
272 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
274 unsigned long long start
, len
;
276 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
278 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
281 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
282 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
285 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
287 ata_port_desc(ap
, "%s 0x%llx", name
,
288 start
+ (unsigned long long)offset
);
291 #endif /* CONFIG_PCI */
293 static int ata_lookup_timeout_table(u8 cmd
)
297 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
300 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
313 * Determine timeout for internal command @cmd for @dev.
319 * Determined timeout.
321 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
323 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
324 int ent
= ata_lookup_timeout_table(cmd
);
328 return ATA_EH_CMD_DFL_TIMEOUT
;
330 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
331 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
346 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
348 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
349 int ent
= ata_lookup_timeout_table(cmd
);
355 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
356 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
357 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
360 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
361 unsigned int err_mask
)
363 struct ata_ering_entry
*ent
;
368 ering
->cursor
%= ATA_ERING_SIZE
;
370 ent
= &ering
->ring
[ering
->cursor
];
371 ent
->eflags
= eflags
;
372 ent
->err_mask
= err_mask
;
373 ent
->timestamp
= get_jiffies_64();
376 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
378 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
385 static void ata_ering_clear(struct ata_ering
*ering
)
387 memset(ering
, 0, sizeof(*ering
));
390 static int ata_ering_map(struct ata_ering
*ering
,
391 int (*map_fn
)(struct ata_ering_entry
*, void *),
395 struct ata_ering_entry
*ent
;
399 ent
= &ering
->ring
[idx
];
402 rc
= map_fn(ent
, arg
);
405 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
406 } while (idx
!= ering
->cursor
);
411 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
413 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
415 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
418 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
419 struct ata_eh_info
*ehi
, unsigned int action
)
421 struct ata_device
*tdev
;
424 ehi
->action
&= ~action
;
425 ata_for_each_dev(tdev
, link
, ALL
)
426 ehi
->dev_action
[tdev
->devno
] &= ~action
;
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
431 /* break ehi->action into ehi->dev_action */
432 if (ehi
->action
& action
) {
433 ata_for_each_dev(tdev
, link
, ALL
)
434 ehi
->dev_action
[tdev
->devno
] |=
435 ehi
->action
& action
;
436 ehi
->action
&= ~action
;
439 /* turn off the specified per-dev action */
440 ehi
->dev_action
[dev
->devno
] &= ~action
;
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
455 * TODO: kill this function once old EH is gone.
458 * Called from timer context
461 * EH_HANDLED or EH_NOT_HANDLED
463 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
465 struct Scsi_Host
*host
= cmd
->device
->host
;
466 struct ata_port
*ap
= ata_shost_to_port(host
);
468 struct ata_queued_cmd
*qc
;
469 enum blk_eh_timer_return ret
;
473 if (ap
->ops
->error_handler
) {
474 ret
= BLK_EH_NOT_HANDLED
;
478 ret
= BLK_EH_HANDLED
;
479 spin_lock_irqsave(ap
->lock
, flags
);
480 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
482 WARN_ON(qc
->scsicmd
!= cmd
);
483 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
484 qc
->err_mask
|= AC_ERR_TIMEOUT
;
485 ret
= BLK_EH_NOT_HANDLED
;
487 spin_unlock_irqrestore(ap
->lock
, flags
);
490 DPRINTK("EXIT, ret=%d\n", ret
);
495 * ata_scsi_error - SCSI layer error handler callback
496 * @host: SCSI host on which error occurred
498 * Handles SCSI-layer-thrown error events.
501 * Inherited from SCSI layer (none, can sleep)
506 void ata_scsi_error(struct Scsi_Host
*host
)
508 struct ata_port
*ap
= ata_shost_to_port(host
);
514 /* synchronize with port task */
515 ata_port_flush_task(ap
);
517 /* synchronize with host lock and sort out timeouts */
519 /* For new EH, all qcs are finished in one of three ways -
520 * normal completion, error completion, and SCSI timeout.
521 * Both cmpletions can race against SCSI timeout. When normal
522 * completion wins, the qc never reaches EH. When error
523 * completion wins, the qc has ATA_QCFLAG_FAILED set.
525 * When SCSI timeout wins, things are a bit more complex.
526 * Normal or error completion can occur after the timeout but
527 * before this point. In such cases, both types of
528 * completions are honored. A scmd is determined to have
529 * timed out iff its associated qc is active and not failed.
531 if (ap
->ops
->error_handler
) {
532 struct scsi_cmnd
*scmd
, *tmp
;
535 spin_lock_irqsave(ap
->lock
, flags
);
537 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
538 struct ata_queued_cmd
*qc
;
540 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
541 qc
= __ata_qc_from_tag(ap
, i
);
542 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
547 if (i
< ATA_MAX_QUEUE
) {
548 /* the scmd has an associated qc */
549 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
550 /* which hasn't failed yet, timeout */
551 qc
->err_mask
|= AC_ERR_TIMEOUT
;
552 qc
->flags
|= ATA_QCFLAG_FAILED
;
556 /* Normal completion occurred after
557 * SCSI timeout but before this point.
558 * Successfully complete it.
560 scmd
->retries
= scmd
->allowed
;
561 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
565 /* If we have timed out qcs. They belong to EH from
566 * this point but the state of the controller is
567 * unknown. Freeze the port to make sure the IRQ
568 * handler doesn't diddle with those qcs. This must
569 * be done atomically w.r.t. setting QCFLAG_FAILED.
572 __ata_port_freeze(ap
);
574 spin_unlock_irqrestore(ap
->lock
, flags
);
576 /* initialize eh_tries */
577 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
579 spin_unlock_wait(ap
->lock
);
582 /* invoke error handler */
583 if (ap
->ops
->error_handler
) {
584 struct ata_link
*link
;
586 /* kill fast drain timer */
587 del_timer_sync(&ap
->fastdrain_timer
);
589 /* process port resume request */
590 ata_eh_handle_port_resume(ap
);
592 /* fetch & clear EH info */
593 spin_lock_irqsave(ap
->lock
, flags
);
595 ata_for_each_link(link
, ap
, HOST_FIRST
) {
596 struct ata_eh_context
*ehc
= &link
->eh_context
;
597 struct ata_device
*dev
;
599 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
600 link
->eh_context
.i
= link
->eh_info
;
601 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
603 ata_for_each_dev(dev
, link
, ENABLED
) {
604 int devno
= dev
->devno
;
606 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
607 if (ata_ncq_enabled(dev
))
608 ehc
->saved_ncq_enabled
|= 1 << devno
;
612 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
613 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
614 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
616 spin_unlock_irqrestore(ap
->lock
, flags
);
618 /* invoke EH, skip if unloading or suspended */
619 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
620 ap
->ops
->error_handler(ap
);
624 /* process port suspend request */
625 ata_eh_handle_port_suspend(ap
);
627 /* Exception might have happend after ->error_handler
628 * recovered the port but before this point. Repeat
631 spin_lock_irqsave(ap
->lock
, flags
);
633 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
634 if (--ap
->eh_tries
) {
635 spin_unlock_irqrestore(ap
->lock
, flags
);
638 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
639 "tries, giving up\n", ATA_EH_MAX_TRIES
);
640 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
643 /* this run is complete, make sure EH info is clear */
644 ata_for_each_link(link
, ap
, HOST_FIRST
)
645 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
647 /* Clear host_eh_scheduled while holding ap->lock such
648 * that if exception occurs after this point but
649 * before EH completion, SCSI midlayer will
652 host
->host_eh_scheduled
= 0;
654 spin_unlock_irqrestore(ap
->lock
, flags
);
656 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
657 ap
->ops
->eng_timeout(ap
);
660 /* finish or retry handled scmd's and clean up */
661 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
663 scsi_eh_flush_done_q(&ap
->eh_done_q
);
666 spin_lock_irqsave(ap
->lock
, flags
);
668 if (ap
->pflags
& ATA_PFLAG_LOADING
)
669 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
670 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
671 queue_delayed_work(ata_aux_wq
, &ap
->hotplug_task
, 0);
673 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
674 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
676 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
678 /* tell wait_eh that we're done */
679 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
680 wake_up_all(&ap
->eh_wait_q
);
682 spin_unlock_irqrestore(ap
->lock
, flags
);
688 * ata_port_wait_eh - Wait for the currently pending EH to complete
689 * @ap: Port to wait EH for
691 * Wait until the currently pending EH is complete.
694 * Kernel thread context (may sleep).
696 void ata_port_wait_eh(struct ata_port
*ap
)
702 spin_lock_irqsave(ap
->lock
, flags
);
704 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
705 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
706 spin_unlock_irqrestore(ap
->lock
, flags
);
708 spin_lock_irqsave(ap
->lock
, flags
);
710 finish_wait(&ap
->eh_wait_q
, &wait
);
712 spin_unlock_irqrestore(ap
->lock
, flags
);
714 /* make sure SCSI EH is complete */
715 if (scsi_host_in_recovery(ap
->scsi_host
)) {
721 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
726 /* count only non-internal commands */
727 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
728 if (ata_qc_from_tag(ap
, tag
))
734 void ata_eh_fastdrain_timerfn(unsigned long arg
)
736 struct ata_port
*ap
= (void *)arg
;
740 spin_lock_irqsave(ap
->lock
, flags
);
742 cnt
= ata_eh_nr_in_flight(ap
);
748 if (cnt
== ap
->fastdrain_cnt
) {
751 /* No progress during the last interval, tag all
752 * in-flight qcs as timed out and freeze the port.
754 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
755 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
757 qc
->err_mask
|= AC_ERR_TIMEOUT
;
762 /* some qcs have finished, give it another chance */
763 ap
->fastdrain_cnt
= cnt
;
764 ap
->fastdrain_timer
.expires
=
765 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
766 add_timer(&ap
->fastdrain_timer
);
770 spin_unlock_irqrestore(ap
->lock
, flags
);
774 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
775 * @ap: target ATA port
776 * @fastdrain: activate fast drain
778 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
779 * is non-zero and EH wasn't pending before. Fast drain ensures
780 * that EH kicks in in timely manner.
783 * spin_lock_irqsave(host lock)
785 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
789 /* already scheduled? */
790 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
793 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
798 /* do we have in-flight qcs? */
799 cnt
= ata_eh_nr_in_flight(ap
);
803 /* activate fast drain */
804 ap
->fastdrain_cnt
= cnt
;
805 ap
->fastdrain_timer
.expires
=
806 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
807 add_timer(&ap
->fastdrain_timer
);
811 * ata_qc_schedule_eh - schedule qc for error handling
812 * @qc: command to schedule error handling for
814 * Schedule error handling for @qc. EH will kick in as soon as
815 * other commands are drained.
818 * spin_lock_irqsave(host lock)
820 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
822 struct ata_port
*ap
= qc
->ap
;
824 WARN_ON(!ap
->ops
->error_handler
);
826 qc
->flags
|= ATA_QCFLAG_FAILED
;
827 ata_eh_set_pending(ap
, 1);
829 /* The following will fail if timeout has already expired.
830 * ata_scsi_error() takes care of such scmds on EH entry.
831 * Note that ATA_QCFLAG_FAILED is unconditionally set after
832 * this function completes.
834 blk_abort_request(qc
->scsicmd
->request
);
838 * ata_port_schedule_eh - schedule error handling without a qc
839 * @ap: ATA port to schedule EH for
841 * Schedule error handling for @ap. EH will kick in as soon as
842 * all commands are drained.
845 * spin_lock_irqsave(host lock)
847 void ata_port_schedule_eh(struct ata_port
*ap
)
849 WARN_ON(!ap
->ops
->error_handler
);
851 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
854 ata_eh_set_pending(ap
, 1);
855 scsi_schedule_eh(ap
->scsi_host
);
857 DPRINTK("port EH scheduled\n");
860 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
862 int tag
, nr_aborted
= 0;
864 WARN_ON(!ap
->ops
->error_handler
);
866 /* we're gonna abort all commands, no need for fast drain */
867 ata_eh_set_pending(ap
, 0);
869 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
870 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
872 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
873 qc
->flags
|= ATA_QCFLAG_FAILED
;
880 ata_port_schedule_eh(ap
);
886 * ata_link_abort - abort all qc's on the link
887 * @link: ATA link to abort qc's for
889 * Abort all active qc's active on @link and schedule EH.
892 * spin_lock_irqsave(host lock)
895 * Number of aborted qc's.
897 int ata_link_abort(struct ata_link
*link
)
899 return ata_do_link_abort(link
->ap
, link
);
903 * ata_port_abort - abort all qc's on the port
904 * @ap: ATA port to abort qc's for
906 * Abort all active qc's of @ap and schedule EH.
909 * spin_lock_irqsave(host_set lock)
912 * Number of aborted qc's.
914 int ata_port_abort(struct ata_port
*ap
)
916 return ata_do_link_abort(ap
, NULL
);
920 * __ata_port_freeze - freeze port
921 * @ap: ATA port to freeze
923 * This function is called when HSM violation or some other
924 * condition disrupts normal operation of the port. Frozen port
925 * is not allowed to perform any operation until the port is
926 * thawed, which usually follows a successful reset.
928 * ap->ops->freeze() callback can be used for freezing the port
929 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
930 * port cannot be frozen hardware-wise, the interrupt handler
931 * must ack and clear interrupts unconditionally while the port
935 * spin_lock_irqsave(host lock)
937 static void __ata_port_freeze(struct ata_port
*ap
)
939 WARN_ON(!ap
->ops
->error_handler
);
944 ap
->pflags
|= ATA_PFLAG_FROZEN
;
946 DPRINTK("ata%u port frozen\n", ap
->print_id
);
950 * ata_port_freeze - abort & freeze port
951 * @ap: ATA port to freeze
953 * Abort and freeze @ap.
956 * spin_lock_irqsave(host lock)
959 * Number of aborted commands.
961 int ata_port_freeze(struct ata_port
*ap
)
965 WARN_ON(!ap
->ops
->error_handler
);
967 nr_aborted
= ata_port_abort(ap
);
968 __ata_port_freeze(ap
);
974 * sata_async_notification - SATA async notification handler
975 * @ap: ATA port where async notification is received
977 * Handler to be called when async notification via SDB FIS is
978 * received. This function schedules EH if necessary.
981 * spin_lock_irqsave(host lock)
984 * 1 if EH is scheduled, 0 otherwise.
986 int sata_async_notification(struct ata_port
*ap
)
991 if (!(ap
->flags
& ATA_FLAG_AN
))
994 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
996 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
998 if (!sata_pmp_attached(ap
) || rc
) {
999 /* PMP is not attached or SNTF is not available */
1000 if (!sata_pmp_attached(ap
)) {
1001 /* PMP is not attached. Check whether ATAPI
1002 * AN is configured. If so, notify media
1005 struct ata_device
*dev
= ap
->link
.device
;
1007 if ((dev
->class == ATA_DEV_ATAPI
) &&
1008 (dev
->flags
& ATA_DFLAG_AN
))
1009 ata_scsi_media_change_notify(dev
);
1012 /* PMP is attached but SNTF is not available.
1013 * ATAPI async media change notification is
1014 * not used. The PMP must be reporting PHY
1015 * status change, schedule EH.
1017 ata_port_schedule_eh(ap
);
1021 /* PMP is attached and SNTF is available */
1022 struct ata_link
*link
;
1024 /* check and notify ATAPI AN */
1025 ata_for_each_link(link
, ap
, EDGE
) {
1026 if (!(sntf
& (1 << link
->pmp
)))
1029 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1030 (link
->device
->flags
& ATA_DFLAG_AN
))
1031 ata_scsi_media_change_notify(link
->device
);
1034 /* If PMP is reporting that PHY status of some
1035 * downstream ports has changed, schedule EH.
1037 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1038 ata_port_schedule_eh(ap
);
1047 * ata_eh_freeze_port - EH helper to freeze port
1048 * @ap: ATA port to freeze
1055 void ata_eh_freeze_port(struct ata_port
*ap
)
1057 unsigned long flags
;
1059 if (!ap
->ops
->error_handler
)
1062 spin_lock_irqsave(ap
->lock
, flags
);
1063 __ata_port_freeze(ap
);
1064 spin_unlock_irqrestore(ap
->lock
, flags
);
1068 * ata_port_thaw_port - EH helper to thaw port
1069 * @ap: ATA port to thaw
1071 * Thaw frozen port @ap.
1076 void ata_eh_thaw_port(struct ata_port
*ap
)
1078 unsigned long flags
;
1080 if (!ap
->ops
->error_handler
)
1083 spin_lock_irqsave(ap
->lock
, flags
);
1085 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1090 spin_unlock_irqrestore(ap
->lock
, flags
);
1092 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1095 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1100 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1102 struct ata_port
*ap
= qc
->ap
;
1103 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1104 unsigned long flags
;
1106 spin_lock_irqsave(ap
->lock
, flags
);
1107 qc
->scsidone
= ata_eh_scsidone
;
1108 __ata_qc_complete(qc
);
1109 WARN_ON(ata_tag_valid(qc
->tag
));
1110 spin_unlock_irqrestore(ap
->lock
, flags
);
1112 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1116 * ata_eh_qc_complete - Complete an active ATA command from EH
1117 * @qc: Command to complete
1119 * Indicate to the mid and upper layers that an ATA command has
1120 * completed. To be used from EH.
1122 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1124 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1125 scmd
->retries
= scmd
->allowed
;
1126 __ata_eh_qc_complete(qc
);
1130 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1131 * @qc: Command to retry
1133 * Indicate to the mid and upper layers that an ATA command
1134 * should be retried. To be used from EH.
1136 * SCSI midlayer limits the number of retries to scmd->allowed.
1137 * scmd->retries is decremented for commands which get retried
1138 * due to unrelated failures (qc->err_mask is zero).
1140 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1142 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1143 if (!qc
->err_mask
&& scmd
->retries
)
1145 __ata_eh_qc_complete(qc
);
1149 * ata_eh_detach_dev - detach ATA device
1150 * @dev: ATA device to detach
1157 void ata_eh_detach_dev(struct ata_device
*dev
)
1159 struct ata_link
*link
= dev
->link
;
1160 struct ata_port
*ap
= link
->ap
;
1161 struct ata_eh_context
*ehc
= &link
->eh_context
;
1162 unsigned long flags
;
1164 ata_dev_disable(dev
);
1166 spin_lock_irqsave(ap
->lock
, flags
);
1168 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1170 if (ata_scsi_offline_dev(dev
)) {
1171 dev
->flags
|= ATA_DFLAG_DETACHED
;
1172 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1175 /* clear per-dev EH info */
1176 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1177 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1178 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1179 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1181 spin_unlock_irqrestore(ap
->lock
, flags
);
1185 * ata_eh_about_to_do - about to perform eh_action
1186 * @link: target ATA link
1187 * @dev: target ATA dev for per-dev action (can be NULL)
1188 * @action: action about to be performed
1190 * Called just before performing EH actions to clear related bits
1191 * in @link->eh_info such that eh actions are not unnecessarily
1197 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1198 unsigned int action
)
1200 struct ata_port
*ap
= link
->ap
;
1201 struct ata_eh_info
*ehi
= &link
->eh_info
;
1202 struct ata_eh_context
*ehc
= &link
->eh_context
;
1203 unsigned long flags
;
1205 spin_lock_irqsave(ap
->lock
, flags
);
1207 ata_eh_clear_action(link
, dev
, ehi
, action
);
1209 /* About to take EH action, set RECOVERED. Ignore actions on
1210 * slave links as master will do them again.
1212 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1213 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1215 spin_unlock_irqrestore(ap
->lock
, flags
);
1219 * ata_eh_done - EH action complete
1220 * @ap: target ATA port
1221 * @dev: target ATA dev for per-dev action (can be NULL)
1222 * @action: action just completed
1224 * Called right after performing EH actions to clear related bits
1225 * in @link->eh_context.
1230 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1231 unsigned int action
)
1233 struct ata_eh_context
*ehc
= &link
->eh_context
;
1235 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1239 * ata_err_string - convert err_mask to descriptive string
1240 * @err_mask: error mask to convert to string
1242 * Convert @err_mask to descriptive string. Errors are
1243 * prioritized according to severity and only the most severe
1244 * error is reported.
1250 * Descriptive string for @err_mask
1252 static const char *ata_err_string(unsigned int err_mask
)
1254 if (err_mask
& AC_ERR_HOST_BUS
)
1255 return "host bus error";
1256 if (err_mask
& AC_ERR_ATA_BUS
)
1257 return "ATA bus error";
1258 if (err_mask
& AC_ERR_TIMEOUT
)
1260 if (err_mask
& AC_ERR_HSM
)
1261 return "HSM violation";
1262 if (err_mask
& AC_ERR_SYSTEM
)
1263 return "internal error";
1264 if (err_mask
& AC_ERR_MEDIA
)
1265 return "media error";
1266 if (err_mask
& AC_ERR_INVALID
)
1267 return "invalid argument";
1268 if (err_mask
& AC_ERR_DEV
)
1269 return "device error";
1270 return "unknown error";
1274 * ata_read_log_page - read a specific log page
1275 * @dev: target device
1276 * @page: page to read
1277 * @buf: buffer to store read page
1278 * @sectors: number of sectors to read
1280 * Read log page using READ_LOG_EXT command.
1283 * Kernel thread context (may sleep).
1286 * 0 on success, AC_ERR_* mask otherwise.
1288 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1289 u8 page
, void *buf
, unsigned int sectors
)
1291 struct ata_taskfile tf
;
1292 unsigned int err_mask
;
1294 DPRINTK("read log page - page %d\n", page
);
1296 ata_tf_init(dev
, &tf
);
1297 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1300 tf
.hob_nsect
= sectors
>> 8;
1301 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1302 tf
.protocol
= ATA_PROT_PIO
;
1304 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1305 buf
, sectors
* ATA_SECT_SIZE
, 0);
1307 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1312 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1313 * @dev: Device to read log page 10h from
1314 * @tag: Resulting tag of the failed command
1315 * @tf: Resulting taskfile registers of the failed command
1317 * Read log page 10h to obtain NCQ error details and clear error
1321 * Kernel thread context (may sleep).
1324 * 0 on success, -errno otherwise.
1326 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1327 int *tag
, struct ata_taskfile
*tf
)
1329 u8
*buf
= dev
->link
->ap
->sector_buf
;
1330 unsigned int err_mask
;
1334 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1339 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1342 ata_dev_printk(dev
, KERN_WARNING
,
1343 "invalid checksum 0x%x on log page 10h\n", csum
);
1348 *tag
= buf
[0] & 0x1f;
1350 tf
->command
= buf
[2];
1351 tf
->feature
= buf
[3];
1355 tf
->device
= buf
[7];
1356 tf
->hob_lbal
= buf
[8];
1357 tf
->hob_lbam
= buf
[9];
1358 tf
->hob_lbah
= buf
[10];
1359 tf
->nsect
= buf
[12];
1360 tf
->hob_nsect
= buf
[13];
1366 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1367 * @dev: target ATAPI device
1368 * @r_sense_key: out parameter for sense_key
1370 * Perform ATAPI TEST_UNIT_READY.
1373 * EH context (may sleep).
1376 * 0 on success, AC_ERR_* mask on failure.
1378 static unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1380 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1381 struct ata_taskfile tf
;
1382 unsigned int err_mask
;
1384 ata_tf_init(dev
, &tf
);
1386 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1387 tf
.command
= ATA_CMD_PACKET
;
1388 tf
.protocol
= ATAPI_PROT_NODATA
;
1390 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1391 if (err_mask
== AC_ERR_DEV
)
1392 *r_sense_key
= tf
.feature
>> 4;
1397 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1398 * @dev: device to perform REQUEST_SENSE to
1399 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1400 * @dfl_sense_key: default sense key to use
1402 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1403 * SENSE. This function is EH helper.
1406 * Kernel thread context (may sleep).
1409 * 0 on success, AC_ERR_* mask on failure
1411 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1412 u8
*sense_buf
, u8 dfl_sense_key
)
1414 u8 cdb
[ATAPI_CDB_LEN
] =
1415 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1416 struct ata_port
*ap
= dev
->link
->ap
;
1417 struct ata_taskfile tf
;
1419 DPRINTK("ATAPI request sense\n");
1421 /* FIXME: is this needed? */
1422 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1424 /* initialize sense_buf with the error register,
1425 * for the case where they are -not- overwritten
1427 sense_buf
[0] = 0x70;
1428 sense_buf
[2] = dfl_sense_key
;
1430 /* some devices time out if garbage left in tf */
1431 ata_tf_init(dev
, &tf
);
1433 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1434 tf
.command
= ATA_CMD_PACKET
;
1436 /* is it pointless to prefer PIO for "safety reasons"? */
1437 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1438 tf
.protocol
= ATAPI_PROT_DMA
;
1439 tf
.feature
|= ATAPI_PKT_DMA
;
1441 tf
.protocol
= ATAPI_PROT_PIO
;
1442 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1446 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1447 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1451 * ata_eh_analyze_serror - analyze SError for a failed port
1452 * @link: ATA link to analyze SError for
1454 * Analyze SError if available and further determine cause of
1460 static void ata_eh_analyze_serror(struct ata_link
*link
)
1462 struct ata_eh_context
*ehc
= &link
->eh_context
;
1463 u32 serror
= ehc
->i
.serror
;
1464 unsigned int err_mask
= 0, action
= 0;
1467 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1468 err_mask
|= AC_ERR_ATA_BUS
;
1469 action
|= ATA_EH_RESET
;
1471 if (serror
& SERR_PROTOCOL
) {
1472 err_mask
|= AC_ERR_HSM
;
1473 action
|= ATA_EH_RESET
;
1475 if (serror
& SERR_INTERNAL
) {
1476 err_mask
|= AC_ERR_SYSTEM
;
1477 action
|= ATA_EH_RESET
;
1480 /* Determine whether a hotplug event has occurred. Both
1481 * SError.N/X are considered hotplug events for enabled or
1482 * host links. For disabled PMP links, only N bit is
1483 * considered as X bit is left at 1 for link plugging.
1487 if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1488 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1490 hotplug_mask
= SERR_PHYRDY_CHG
;
1492 if (serror
& hotplug_mask
)
1493 ata_ehi_hotplugged(&ehc
->i
);
1495 ehc
->i
.err_mask
|= err_mask
;
1496 ehc
->i
.action
|= action
;
1500 * ata_eh_analyze_ncq_error - analyze NCQ error
1501 * @link: ATA link to analyze NCQ error for
1503 * Read log page 10h, determine the offending qc and acquire
1504 * error status TF. For NCQ device errors, all LLDDs have to do
1505 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1509 * Kernel thread context (may sleep).
1511 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1513 struct ata_port
*ap
= link
->ap
;
1514 struct ata_eh_context
*ehc
= &link
->eh_context
;
1515 struct ata_device
*dev
= link
->device
;
1516 struct ata_queued_cmd
*qc
;
1517 struct ata_taskfile tf
;
1520 /* if frozen, we can't do much */
1521 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1524 /* is it NCQ device error? */
1525 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1528 /* has LLDD analyzed already? */
1529 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1530 qc
= __ata_qc_from_tag(ap
, tag
);
1532 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1539 /* okay, this error is ours */
1540 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1542 ata_link_printk(link
, KERN_ERR
, "failed to read log page 10h "
1543 "(errno=%d)\n", rc
);
1547 if (!(link
->sactive
& (1 << tag
))) {
1548 ata_link_printk(link
, KERN_ERR
, "log page 10h reported "
1549 "inactive tag %d\n", tag
);
1553 /* we've got the perpetrator, condemn it */
1554 qc
= __ata_qc_from_tag(ap
, tag
);
1555 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1556 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1557 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1558 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1562 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1563 * @qc: qc to analyze
1564 * @tf: Taskfile registers to analyze
1566 * Analyze taskfile of @qc and further determine cause of
1567 * failure. This function also requests ATAPI sense data if
1571 * Kernel thread context (may sleep).
1574 * Determined recovery action
1576 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1577 const struct ata_taskfile
*tf
)
1579 unsigned int tmp
, action
= 0;
1580 u8 stat
= tf
->command
, err
= tf
->feature
;
1582 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1583 qc
->err_mask
|= AC_ERR_HSM
;
1584 return ATA_EH_RESET
;
1587 if (stat
& (ATA_ERR
| ATA_DF
))
1588 qc
->err_mask
|= AC_ERR_DEV
;
1592 switch (qc
->dev
->class) {
1595 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1597 qc
->err_mask
|= AC_ERR_MEDIA
;
1599 qc
->err_mask
|= AC_ERR_INVALID
;
1603 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1604 tmp
= atapi_eh_request_sense(qc
->dev
,
1605 qc
->scsicmd
->sense_buffer
,
1606 qc
->result_tf
.feature
>> 4);
1608 /* ATA_QCFLAG_SENSE_VALID is used to
1609 * tell atapi_qc_complete() that sense
1610 * data is already valid.
1612 * TODO: interpret sense data and set
1613 * appropriate err_mask.
1615 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1617 qc
->err_mask
|= tmp
;
1621 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1622 action
|= ATA_EH_RESET
;
1627 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1632 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1636 base
= ATA_ECAT_DUBIOUS_NONE
;
1638 if (err_mask
& AC_ERR_ATA_BUS
)
1639 return base
+ ATA_ECAT_ATA_BUS
;
1641 if (err_mask
& AC_ERR_TIMEOUT
)
1642 return base
+ ATA_ECAT_TOUT_HSM
;
1644 if (eflags
& ATA_EFLAG_IS_IO
) {
1645 if (err_mask
& AC_ERR_HSM
)
1646 return base
+ ATA_ECAT_TOUT_HSM
;
1648 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1649 return base
+ ATA_ECAT_UNK_DEV
;
1655 struct speed_down_verdict_arg
{
1658 int nr_errors
[ATA_ECAT_NR
];
1661 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1663 struct speed_down_verdict_arg
*arg
= void_arg
;
1666 if (ent
->timestamp
< arg
->since
)
1669 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1671 arg
->nr_errors
[cat
]++;
1677 * ata_eh_speed_down_verdict - Determine speed down verdict
1678 * @dev: Device of interest
1680 * This function examines error ring of @dev and determines
1681 * whether NCQ needs to be turned off, transfer speed should be
1682 * stepped down, or falling back to PIO is necessary.
1684 * ECAT_ATA_BUS : ATA_BUS error for any command
1686 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1689 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1691 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1692 * data transfer hasn't been verified.
1696 * NCQ_OFF : Turn off NCQ.
1698 * SPEED_DOWN : Speed down transfer speed but don't fall back
1701 * FALLBACK_TO_PIO : Fall back to PIO.
1703 * Even if multiple verdicts are returned, only one action is
1704 * taken per error. An action triggered by non-DUBIOUS errors
1705 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1706 * This is to expedite speed down decisions right after device is
1707 * initially configured.
1709 * The followings are speed down rules. #1 and #2 deal with
1712 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1713 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1715 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1716 * occurred during last 5 mins, NCQ_OFF.
1718 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1719 * ocurred during last 5 mins, FALLBACK_TO_PIO
1721 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1722 * during last 10 mins, NCQ_OFF.
1724 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1725 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1728 * Inherited from caller.
1731 * OR of ATA_EH_SPDN_* flags.
1733 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1735 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1736 u64 j64
= get_jiffies_64();
1737 struct speed_down_verdict_arg arg
;
1738 unsigned int verdict
= 0;
1740 /* scan past 5 mins of error history */
1741 memset(&arg
, 0, sizeof(arg
));
1742 arg
.since
= j64
- min(j64
, j5mins
);
1743 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1745 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1746 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1747 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1748 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1750 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1751 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1752 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1754 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1755 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1756 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1757 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1759 /* scan past 10 mins of error history */
1760 memset(&arg
, 0, sizeof(arg
));
1761 arg
.since
= j64
- min(j64
, j10mins
);
1762 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1764 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1765 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1766 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1768 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1769 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1770 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1771 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1777 * ata_eh_speed_down - record error and speed down if necessary
1778 * @dev: Failed device
1779 * @eflags: mask of ATA_EFLAG_* flags
1780 * @err_mask: err_mask of the error
1782 * Record error and examine error history to determine whether
1783 * adjusting transmission speed is necessary. It also sets
1784 * transmission limits appropriately if such adjustment is
1788 * Kernel thread context (may sleep).
1791 * Determined recovery action.
1793 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
1794 unsigned int eflags
, unsigned int err_mask
)
1796 struct ata_link
*link
= ata_dev_phys_link(dev
);
1798 unsigned int verdict
;
1799 unsigned int action
= 0;
1801 /* don't bother if Cat-0 error */
1802 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
1805 /* record error and determine whether speed down is necessary */
1806 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
1807 verdict
= ata_eh_speed_down_verdict(dev
);
1810 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1811 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1812 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1813 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1814 ata_dev_printk(dev
, KERN_WARNING
,
1815 "NCQ disabled due to excessive errors\n");
1820 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1821 /* speed down SATA link speed if possible */
1822 if (sata_down_spd_limit(link
) == 0) {
1823 action
|= ATA_EH_RESET
;
1827 /* lower transfer mode */
1828 if (dev
->spdn_cnt
< 2) {
1829 static const int dma_dnxfer_sel
[] =
1830 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1831 static const int pio_dnxfer_sel
[] =
1832 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1835 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1836 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1838 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1842 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1843 action
|= ATA_EH_RESET
;
1849 /* Fall back to PIO? Slowing down to PIO is meaningless for
1850 * SATA ATA devices. Consider it only for PATA and SATAPI.
1852 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1853 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
1854 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1855 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1857 action
|= ATA_EH_RESET
;
1864 /* device has been slowed down, blow error history */
1865 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
1866 ata_ering_clear(&dev
->ering
);
1871 * ata_eh_link_autopsy - analyze error and determine recovery action
1872 * @link: host link to perform autopsy on
1874 * Analyze why @link failed and determine which recovery actions
1875 * are needed. This function also sets more detailed AC_ERR_*
1876 * values and fills sense data for ATAPI CHECK SENSE.
1879 * Kernel thread context (may sleep).
1881 static void ata_eh_link_autopsy(struct ata_link
*link
)
1883 struct ata_port
*ap
= link
->ap
;
1884 struct ata_eh_context
*ehc
= &link
->eh_context
;
1885 struct ata_device
*dev
;
1886 unsigned int all_err_mask
= 0, eflags
= 0;
1893 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1896 /* obtain and analyze SError */
1897 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
1899 ehc
->i
.serror
|= serror
;
1900 ata_eh_analyze_serror(link
);
1901 } else if (rc
!= -EOPNOTSUPP
) {
1902 /* SError read failed, force reset and probing */
1903 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
1904 ehc
->i
.action
|= ATA_EH_RESET
;
1905 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
1908 /* analyze NCQ failure */
1909 ata_eh_analyze_ncq_error(link
);
1911 /* any real error trumps AC_ERR_OTHER */
1912 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1913 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1915 all_err_mask
|= ehc
->i
.err_mask
;
1917 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1918 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1920 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
1921 ata_dev_phys_link(qc
->dev
) != link
)
1924 /* inherit upper level err_mask */
1925 qc
->err_mask
|= ehc
->i
.err_mask
;
1928 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1930 /* DEV errors are probably spurious in case of ATA_BUS error */
1931 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1932 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1935 /* any real error trumps unknown error */
1936 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1937 qc
->err_mask
&= ~AC_ERR_OTHER
;
1939 /* SENSE_VALID trumps dev/unknown error and revalidation */
1940 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
1941 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1943 /* determine whether the command is worth retrying */
1944 if (!(qc
->err_mask
& AC_ERR_INVALID
) &&
1945 ((qc
->flags
& ATA_QCFLAG_IO
) || qc
->err_mask
!= AC_ERR_DEV
))
1946 qc
->flags
|= ATA_QCFLAG_RETRY
;
1948 /* accumulate error info */
1949 ehc
->i
.dev
= qc
->dev
;
1950 all_err_mask
|= qc
->err_mask
;
1951 if (qc
->flags
& ATA_QCFLAG_IO
)
1952 eflags
|= ATA_EFLAG_IS_IO
;
1955 /* enforce default EH actions */
1956 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
1957 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
1958 ehc
->i
.action
|= ATA_EH_RESET
;
1959 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
1960 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
1961 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
1963 /* If we have offending qcs and the associated failed device,
1964 * perform per-dev EH action only on the offending device.
1967 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
1968 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
1969 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
1972 /* propagate timeout to host link */
1973 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
1974 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
1976 /* record error and consider speeding down */
1978 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
1979 ata_dev_enabled(link
->device
))))
1983 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
1984 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
1985 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
1992 * ata_eh_autopsy - analyze error and determine recovery action
1993 * @ap: host port to perform autopsy on
1995 * Analyze all links of @ap and determine why they failed and
1996 * which recovery actions are needed.
1999 * Kernel thread context (may sleep).
2001 void ata_eh_autopsy(struct ata_port
*ap
)
2003 struct ata_link
*link
;
2005 ata_for_each_link(link
, ap
, EDGE
)
2006 ata_eh_link_autopsy(link
);
2008 /* Handle the frigging slave link. Autopsy is done similarly
2009 * but actions and flags are transferred over to the master
2010 * link and handled from there.
2012 if (ap
->slave_link
) {
2013 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2014 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2016 /* transfer control flags from master to slave */
2017 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2019 /* perform autopsy on the slave link */
2020 ata_eh_link_autopsy(ap
->slave_link
);
2022 /* transfer actions from slave to master and clear slave */
2023 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2024 mehc
->i
.action
|= sehc
->i
.action
;
2025 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2026 mehc
->i
.flags
|= sehc
->i
.flags
;
2027 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2030 /* Autopsy of fanout ports can affect host link autopsy.
2031 * Perform host link autopsy last.
2033 if (sata_pmp_attached(ap
))
2034 ata_eh_link_autopsy(&ap
->link
);
2038 * ata_eh_link_report - report error handling to user
2039 * @link: ATA link EH is going on
2041 * Report EH to user.
2046 static void ata_eh_link_report(struct ata_link
*link
)
2048 struct ata_port
*ap
= link
->ap
;
2049 struct ata_eh_context
*ehc
= &link
->eh_context
;
2050 const char *frozen
, *desc
;
2052 int tag
, nr_failed
= 0;
2054 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2058 if (ehc
->i
.desc
[0] != '\0')
2061 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2062 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2064 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2065 ata_dev_phys_link(qc
->dev
) != link
||
2066 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2067 qc
->err_mask
== AC_ERR_DEV
))
2069 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2075 if (!nr_failed
&& !ehc
->i
.err_mask
)
2079 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2082 memset(tries_buf
, 0, sizeof(tries_buf
));
2083 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2084 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
2088 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
2089 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2090 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2091 ehc
->i
.action
, frozen
, tries_buf
);
2093 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "%s\n", desc
);
2095 ata_link_printk(link
, KERN_ERR
, "exception Emask 0x%x "
2096 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2097 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2098 ehc
->i
.action
, frozen
, tries_buf
);
2100 ata_link_printk(link
, KERN_ERR
, "%s\n", desc
);
2104 ata_link_printk(link
, KERN_ERR
,
2105 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2106 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2107 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2108 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2109 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2110 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2111 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2112 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2113 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2114 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2115 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2116 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2117 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2118 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2119 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2120 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2121 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2122 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2124 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2125 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2126 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2127 const u8
*cdb
= qc
->cdb
;
2128 char data_buf
[20] = "";
2129 char cdb_buf
[70] = "";
2131 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2132 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2135 if (qc
->dma_dir
!= DMA_NONE
) {
2136 static const char *dma_str
[] = {
2137 [DMA_BIDIRECTIONAL
] = "bidi",
2138 [DMA_TO_DEVICE
] = "out",
2139 [DMA_FROM_DEVICE
] = "in",
2141 static const char *prot_str
[] = {
2142 [ATA_PROT_PIO
] = "pio",
2143 [ATA_PROT_DMA
] = "dma",
2144 [ATA_PROT_NCQ
] = "ncq",
2145 [ATAPI_PROT_PIO
] = "pio",
2146 [ATAPI_PROT_DMA
] = "dma",
2149 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2150 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2151 dma_str
[qc
->dma_dir
]);
2154 if (ata_is_atapi(qc
->tf
.protocol
))
2155 snprintf(cdb_buf
, sizeof(cdb_buf
),
2156 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2157 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2158 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2159 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2160 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2161 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2163 ata_dev_printk(qc
->dev
, KERN_ERR
,
2164 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2166 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2167 "Emask 0x%x (%s)%s\n",
2168 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2169 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2170 cmd
->hob_feature
, cmd
->hob_nsect
,
2171 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2172 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2173 res
->command
, res
->feature
, res
->nsect
,
2174 res
->lbal
, res
->lbam
, res
->lbah
,
2175 res
->hob_feature
, res
->hob_nsect
,
2176 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2177 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2178 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2180 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2182 if (res
->command
& ATA_BUSY
)
2183 ata_dev_printk(qc
->dev
, KERN_ERR
,
2184 "status: { Busy }\n");
2186 ata_dev_printk(qc
->dev
, KERN_ERR
,
2187 "status: { %s%s%s%s}\n",
2188 res
->command
& ATA_DRDY
? "DRDY " : "",
2189 res
->command
& ATA_DF
? "DF " : "",
2190 res
->command
& ATA_DRQ
? "DRQ " : "",
2191 res
->command
& ATA_ERR
? "ERR " : "");
2194 if (cmd
->command
!= ATA_CMD_PACKET
&&
2195 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2197 ata_dev_printk(qc
->dev
, KERN_ERR
,
2198 "error: { %s%s%s%s}\n",
2199 res
->feature
& ATA_ICRC
? "ICRC " : "",
2200 res
->feature
& ATA_UNC
? "UNC " : "",
2201 res
->feature
& ATA_IDNF
? "IDNF " : "",
2202 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2207 * ata_eh_report - report error handling to user
2208 * @ap: ATA port to report EH about
2210 * Report EH to user.
2215 void ata_eh_report(struct ata_port
*ap
)
2217 struct ata_link
*link
;
2219 ata_for_each_link(link
, ap
, HOST_FIRST
)
2220 ata_eh_link_report(link
);
2223 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2224 unsigned int *classes
, unsigned long deadline
,
2227 struct ata_device
*dev
;
2230 ata_for_each_dev(dev
, link
, ALL
)
2231 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2233 return reset(link
, classes
, deadline
);
2236 static int ata_eh_followup_srst_needed(struct ata_link
*link
,
2237 int rc
, const unsigned int *classes
)
2239 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2243 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2248 int ata_eh_reset(struct ata_link
*link
, int classify
,
2249 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2250 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2252 struct ata_port
*ap
= link
->ap
;
2253 struct ata_link
*slave
= ap
->slave_link
;
2254 struct ata_eh_context
*ehc
= &link
->eh_context
;
2255 struct ata_eh_context
*sehc
= &slave
->eh_context
;
2256 unsigned int *classes
= ehc
->classes
;
2257 unsigned int lflags
= link
->flags
;
2258 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2259 int max_tries
= 0, try = 0;
2260 struct ata_link
*failed_link
;
2261 struct ata_device
*dev
;
2262 unsigned long deadline
, now
;
2263 ata_reset_fn_t reset
;
2264 unsigned long flags
;
2271 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2273 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2275 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2278 /* make sure each reset attemp is at least COOL_DOWN apart */
2279 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2281 WARN_ON(time_after(ehc
->last_reset
, now
));
2282 deadline
= ata_deadline(ehc
->last_reset
,
2283 ATA_EH_RESET_COOL_DOWN
);
2284 if (time_before(now
, deadline
))
2285 schedule_timeout_uninterruptible(deadline
- now
);
2288 spin_lock_irqsave(ap
->lock
, flags
);
2289 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2290 spin_unlock_irqrestore(ap
->lock
, flags
);
2292 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2294 ata_for_each_dev(dev
, link
, ALL
) {
2295 /* If we issue an SRST then an ATA drive (not ATAPI)
2296 * may change configuration and be in PIO0 timing. If
2297 * we do a hard reset (or are coming from power on)
2298 * this is true for ATA or ATAPI. Until we've set a
2299 * suitable controller mode we should not touch the
2300 * bus as we may be talking too fast.
2302 dev
->pio_mode
= XFER_PIO_0
;
2304 /* If the controller has a pio mode setup function
2305 * then use it to set the chipset to rights. Don't
2306 * touch the DMA setup as that will be dealt with when
2307 * configuring devices.
2309 if (ap
->ops
->set_piomode
)
2310 ap
->ops
->set_piomode(ap
, dev
);
2313 /* prefer hardreset */
2315 ehc
->i
.action
&= ~ATA_EH_RESET
;
2318 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2319 } else if (softreset
) {
2321 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2325 unsigned long deadline
= ata_deadline(jiffies
,
2326 ATA_EH_PRERESET_TIMEOUT
);
2329 sehc
->i
.action
&= ~ATA_EH_RESET
;
2330 sehc
->i
.action
|= ehc
->i
.action
;
2333 rc
= prereset(link
, deadline
);
2335 /* If present, do prereset on slave link too. Reset
2336 * is skipped iff both master and slave links report
2337 * -ENOENT or clear ATA_EH_RESET.
2339 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2342 tmp
= prereset(slave
, deadline
);
2346 ehc
->i
.action
|= sehc
->i
.action
;
2350 if (rc
== -ENOENT
) {
2351 ata_link_printk(link
, KERN_DEBUG
,
2352 "port disabled. ignoring.\n");
2353 ehc
->i
.action
&= ~ATA_EH_RESET
;
2355 ata_for_each_dev(dev
, link
, ALL
)
2356 classes
[dev
->devno
] = ATA_DEV_NONE
;
2360 ata_link_printk(link
, KERN_ERR
,
2361 "prereset failed (errno=%d)\n", rc
);
2365 /* prereset() might have cleared ATA_EH_RESET. If so,
2366 * bang classes and return.
2368 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2369 ata_for_each_dev(dev
, link
, ALL
)
2370 classes
[dev
->devno
] = ATA_DEV_NONE
;
2380 if (ata_is_host_link(link
))
2381 ata_eh_freeze_port(ap
);
2383 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2387 ata_link_printk(link
, KERN_INFO
, "%s resetting link\n",
2388 reset
== softreset
? "soft" : "hard");
2390 /* mark that this EH session started with reset */
2391 ehc
->last_reset
= jiffies
;
2392 if (reset
== hardreset
)
2393 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2395 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2397 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2398 if (rc
&& rc
!= -EAGAIN
) {
2403 /* hardreset slave link if existent */
2404 if (slave
&& reset
== hardreset
) {
2408 ata_link_printk(slave
, KERN_INFO
,
2409 "hard resetting link\n");
2411 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2412 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2420 failed_link
= slave
;
2426 /* perform follow-up SRST if necessary */
2427 if (reset
== hardreset
&&
2428 ata_eh_followup_srst_needed(link
, rc
, classes
)) {
2432 ata_link_printk(link
, KERN_ERR
,
2433 "follow-up softreset required "
2434 "but no softreset avaliable\n");
2440 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2441 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2445 ata_link_printk(link
, KERN_INFO
, "no reset method "
2446 "available, skipping reset\n");
2447 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2448 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2452 * Post-reset processing
2454 ata_for_each_dev(dev
, link
, ALL
) {
2455 /* After the reset, the device state is PIO 0 and the
2456 * controller state is undefined. Reset also wakes up
2457 * drives from sleeping mode.
2459 dev
->pio_mode
= XFER_PIO_0
;
2460 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2462 if (!ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2463 /* apply class override */
2464 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2465 classes
[dev
->devno
] = ATA_DEV_ATA
;
2466 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2467 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2469 classes
[dev
->devno
] = ATA_DEV_NONE
;
2472 /* record current link speed */
2473 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2474 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2475 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2476 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2479 if (ata_is_host_link(link
))
2480 ata_eh_thaw_port(ap
);
2482 /* postreset() should clear hardware SError. Although SError
2483 * is cleared during link resume, clearing SError here is
2484 * necessary as some PHYs raise hotplug events after SRST.
2485 * This introduces race condition where hotplug occurs between
2486 * reset and here. This race is mediated by cross checking
2487 * link onlineness and classification result later.
2490 postreset(link
, classes
);
2492 postreset(slave
, classes
);
2495 /* clear cached SError */
2496 spin_lock_irqsave(link
->ap
->lock
, flags
);
2497 link
->eh_info
.serror
= 0;
2499 slave
->eh_info
.serror
= 0;
2500 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2502 /* Make sure onlineness and classification result correspond.
2503 * Hotplug could have happened during reset and some
2504 * controllers fail to wait while a drive is spinning up after
2505 * being hotplugged causing misdetection. By cross checking
2506 * link onlineness and classification result, those conditions
2507 * can be reliably detected and retried.
2510 ata_for_each_dev(dev
, link
, ALL
) {
2511 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2512 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2513 classes
[dev
->devno
] = ATA_DEV_NONE
;
2514 if (ata_phys_link_online(ata_dev_phys_link(dev
)))
2519 if (classify
&& nr_unknown
) {
2520 if (try < max_tries
) {
2521 ata_link_printk(link
, KERN_WARNING
, "link online but "
2522 "device misclassified, retrying\n");
2527 ata_link_printk(link
, KERN_WARNING
,
2528 "link online but device misclassified, "
2529 "device detection might fail\n");
2532 /* reset successful, schedule revalidation */
2533 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2535 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2536 ehc
->last_reset
= jiffies
; /* update to completion time */
2537 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2541 /* clear hotplug flag */
2542 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2544 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2546 spin_lock_irqsave(ap
->lock
, flags
);
2547 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2548 spin_unlock_irqrestore(ap
->lock
, flags
);
2553 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2554 if (!ata_is_host_link(link
) &&
2555 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2558 if (rc
== -ERESTART
|| try >= max_tries
)
2562 if (time_before(now
, deadline
)) {
2563 unsigned long delta
= deadline
- now
;
2565 ata_link_printk(failed_link
, KERN_WARNING
,
2566 "reset failed (errno=%d), retrying in %u secs\n",
2567 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2570 delta
= schedule_timeout_uninterruptible(delta
);
2573 if (try == max_tries
- 1) {
2574 sata_down_spd_limit(link
);
2576 sata_down_spd_limit(slave
);
2577 } else if (rc
== -EPIPE
)
2578 sata_down_spd_limit(failed_link
);
2585 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
2587 struct ata_link
*link
;
2588 struct ata_device
*dev
;
2589 unsigned long flags
;
2592 * This function can be thought of as an extended version of
2593 * ata_eh_about_to_do() specially crafted to accommodate the
2594 * requirements of ATA_EH_PARK handling. Since the EH thread
2595 * does not leave the do {} while () loop in ata_eh_recover as
2596 * long as the timeout for a park request to *one* device on
2597 * the port has not expired, and since we still want to pick
2598 * up park requests to other devices on the same port or
2599 * timeout updates for the same device, we have to pull
2600 * ATA_EH_PARK actions from eh_info into eh_context.i
2601 * ourselves at the beginning of each pass over the loop.
2603 * Additionally, all write accesses to &ap->park_req_pending
2604 * through INIT_COMPLETION() (see below) or complete_all()
2605 * (see ata_scsi_park_store()) are protected by the host lock.
2606 * As a result we have that park_req_pending.done is zero on
2607 * exit from this function, i.e. when ATA_EH_PARK actions for
2608 * *all* devices on port ap have been pulled into the
2609 * respective eh_context structs. If, and only if,
2610 * park_req_pending.done is non-zero by the time we reach
2611 * wait_for_completion_timeout(), another ATA_EH_PARK action
2612 * has been scheduled for at least one of the devices on port
2613 * ap and we have to cycle over the do {} while () loop in
2614 * ata_eh_recover() again.
2617 spin_lock_irqsave(ap
->lock
, flags
);
2618 INIT_COMPLETION(ap
->park_req_pending
);
2619 ata_for_each_link(link
, ap
, EDGE
) {
2620 ata_for_each_dev(dev
, link
, ALL
) {
2621 struct ata_eh_info
*ehi
= &link
->eh_info
;
2623 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
2624 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
2625 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
2628 spin_unlock_irqrestore(ap
->lock
, flags
);
2631 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
2633 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2634 struct ata_taskfile tf
;
2635 unsigned int err_mask
;
2637 ata_tf_init(dev
, &tf
);
2639 ehc
->unloaded_mask
|= 1 << dev
->devno
;
2640 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
2646 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2647 tf
.command
= ATA_CMD_CHK_POWER
;
2650 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
2651 tf
.protocol
|= ATA_PROT_NODATA
;
2652 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
2653 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
2654 ata_dev_printk(dev
, KERN_ERR
, "head unload failed!\n");
2655 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2659 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
2660 struct ata_device
**r_failed_dev
)
2662 struct ata_port
*ap
= link
->ap
;
2663 struct ata_eh_context
*ehc
= &link
->eh_context
;
2664 struct ata_device
*dev
;
2665 unsigned int new_mask
= 0;
2666 unsigned long flags
;
2671 /* For PATA drive side cable detection to work, IDENTIFY must
2672 * be done backwards such that PDIAG- is released by the slave
2673 * device before the master device is identified.
2675 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
2676 unsigned int action
= ata_eh_dev_action(dev
);
2677 unsigned int readid_flags
= 0;
2679 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2680 readid_flags
|= ATA_READID_POSTRESET
;
2682 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
2683 WARN_ON(dev
->class == ATA_DEV_PMP
);
2685 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2690 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
2691 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
2696 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
2698 /* Configuration may have changed, reconfigure
2701 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2703 /* schedule the scsi_rescan_device() here */
2704 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
2705 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
2706 ehc
->tries
[dev
->devno
] &&
2707 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
2708 dev
->class = ehc
->classes
[dev
->devno
];
2710 if (dev
->class == ATA_DEV_PMP
)
2711 rc
= sata_pmp_attach(dev
);
2713 rc
= ata_dev_read_id(dev
, &dev
->class,
2714 readid_flags
, dev
->id
);
2717 new_mask
|= 1 << dev
->devno
;
2720 /* IDENTIFY was issued to non-existent
2721 * device. No need to reset. Just
2722 * thaw and kill the device.
2724 ata_eh_thaw_port(ap
);
2725 dev
->class = ATA_DEV_UNKNOWN
;
2728 dev
->class = ATA_DEV_UNKNOWN
;
2734 /* PDIAG- should have been released, ask cable type if post-reset */
2735 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
2736 if (ap
->ops
->cable_detect
)
2737 ap
->cbl
= ap
->ops
->cable_detect(ap
);
2741 /* Configure new devices forward such that user doesn't see
2742 * device detection messages backwards.
2744 ata_for_each_dev(dev
, link
, ALL
) {
2745 if (!(new_mask
& (1 << dev
->devno
)) ||
2746 dev
->class == ATA_DEV_PMP
)
2749 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
2750 rc
= ata_dev_configure(dev
);
2751 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
2755 spin_lock_irqsave(ap
->lock
, flags
);
2756 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
2757 spin_unlock_irqrestore(ap
->lock
, flags
);
2759 /* new device discovered, configure xfermode */
2760 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2766 *r_failed_dev
= dev
;
2767 DPRINTK("EXIT rc=%d\n", rc
);
2772 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2773 * @link: link on which timings will be programmed
2774 * @r_failed_dev: out paramter for failed device
2776 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2777 * ata_set_mode() fails, pointer to the failing device is
2778 * returned in @r_failed_dev.
2781 * PCI/etc. bus probe sem.
2784 * 0 on success, negative errno otherwise
2786 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
2788 struct ata_port
*ap
= link
->ap
;
2789 struct ata_device
*dev
;
2792 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2793 ata_for_each_dev(dev
, link
, ENABLED
) {
2794 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
2795 struct ata_ering_entry
*ent
;
2797 ent
= ata_ering_top(&dev
->ering
);
2799 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
2803 /* has private set_mode? */
2804 if (ap
->ops
->set_mode
)
2805 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
2807 rc
= ata_do_set_mode(link
, r_failed_dev
);
2809 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2810 ata_for_each_dev(dev
, link
, ENABLED
) {
2811 struct ata_eh_context
*ehc
= &link
->eh_context
;
2812 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
2813 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
2815 if (dev
->xfer_mode
!= saved_xfer_mode
||
2816 ata_ncq_enabled(dev
) != saved_ncq
)
2817 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
2824 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2825 * @dev: ATAPI device to clear UA for
2827 * Resets and other operations can make an ATAPI device raise
2828 * UNIT ATTENTION which causes the next operation to fail. This
2829 * function clears UA.
2832 * EH context (may sleep).
2835 * 0 on success, -errno on failure.
2837 static int atapi_eh_clear_ua(struct ata_device
*dev
)
2841 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
2842 u8 sense_buffer
[SCSI_SENSE_BUFFERSIZE
];
2844 unsigned int err_mask
;
2846 err_mask
= atapi_eh_tur(dev
, &sense_key
);
2847 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
2848 ata_dev_printk(dev
, KERN_WARNING
, "TEST_UNIT_READY "
2849 "failed (err_mask=0x%x)\n", err_mask
);
2853 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
2856 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
2858 ata_dev_printk(dev
, KERN_WARNING
, "failed to clear "
2859 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
2864 ata_dev_printk(dev
, KERN_WARNING
,
2865 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES
);
2870 static int ata_link_nr_enabled(struct ata_link
*link
)
2872 struct ata_device
*dev
;
2875 ata_for_each_dev(dev
, link
, ENABLED
)
2880 static int ata_link_nr_vacant(struct ata_link
*link
)
2882 struct ata_device
*dev
;
2885 ata_for_each_dev(dev
, link
, ALL
)
2886 if (dev
->class == ATA_DEV_UNKNOWN
)
2891 static int ata_eh_skip_recovery(struct ata_link
*link
)
2893 struct ata_port
*ap
= link
->ap
;
2894 struct ata_eh_context
*ehc
= &link
->eh_context
;
2895 struct ata_device
*dev
;
2897 /* skip disabled links */
2898 if (link
->flags
& ATA_LFLAG_DISABLED
)
2901 /* thaw frozen port and recover failed devices */
2902 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
2905 /* reset at least once if reset is requested */
2906 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
2907 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
2910 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2911 ata_for_each_dev(dev
, link
, ALL
) {
2912 if (dev
->class == ATA_DEV_UNKNOWN
&&
2913 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
2920 static int ata_eh_schedule_probe(struct ata_device
*dev
)
2922 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2924 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
2925 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
2928 ata_eh_detach_dev(dev
);
2930 ehc
->did_probe_mask
|= (1 << dev
->devno
);
2931 ehc
->i
.action
|= ATA_EH_RESET
;
2932 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
2933 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
2938 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
2940 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2942 ehc
->tries
[dev
->devno
]--;
2946 /* device missing or wrong IDENTIFY data, schedule probing */
2947 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
2949 /* give it just one more chance */
2950 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
2952 if (ehc
->tries
[dev
->devno
] == 1 && dev
->pio_mode
> XFER_PIO_0
) {
2953 /* This is the last chance, better to slow
2954 * down than lose it.
2956 sata_down_spd_limit(ata_dev_phys_link(dev
));
2957 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
2961 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
2962 /* disable device if it has used up all its chances */
2963 ata_dev_disable(dev
);
2965 /* detach if offline */
2966 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2967 ata_eh_detach_dev(dev
);
2969 /* schedule probe if necessary */
2970 if (ata_eh_schedule_probe(dev
)) {
2971 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
2972 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
2973 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
2978 ehc
->i
.action
|= ATA_EH_RESET
;
2984 * ata_eh_recover - recover host port after error
2985 * @ap: host port to recover
2986 * @prereset: prereset method (can be NULL)
2987 * @softreset: softreset method (can be NULL)
2988 * @hardreset: hardreset method (can be NULL)
2989 * @postreset: postreset method (can be NULL)
2990 * @r_failed_link: out parameter for failed link
2992 * This is the alpha and omega, eum and yang, heart and soul of
2993 * libata exception handling. On entry, actions required to
2994 * recover each link and hotplug requests are recorded in the
2995 * link's eh_context. This function executes all the operations
2996 * with appropriate retrials and fallbacks to resurrect failed
2997 * devices, detach goners and greet newcomers.
3000 * Kernel thread context (may sleep).
3003 * 0 on success, -errno on failure.
3005 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3006 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3007 ata_postreset_fn_t postreset
,
3008 struct ata_link
**r_failed_link
)
3010 struct ata_link
*link
;
3011 struct ata_device
*dev
;
3014 unsigned long flags
, deadline
;
3018 /* prep for recovery */
3019 ata_for_each_link(link
, ap
, EDGE
) {
3020 struct ata_eh_context
*ehc
= &link
->eh_context
;
3022 /* re-enable link? */
3023 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3024 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3025 spin_lock_irqsave(ap
->lock
, flags
);
3026 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3027 spin_unlock_irqrestore(ap
->lock
, flags
);
3028 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3031 ata_for_each_dev(dev
, link
, ALL
) {
3032 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3033 ehc
->tries
[dev
->devno
] = 1;
3035 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3037 /* collect port action mask recorded in dev actions */
3038 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3039 ~ATA_EH_PERDEV_MASK
;
3040 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3042 /* process hotplug request */
3043 if (dev
->flags
& ATA_DFLAG_DETACH
)
3044 ata_eh_detach_dev(dev
);
3046 /* schedule probe if necessary */
3047 if (!ata_dev_enabled(dev
))
3048 ata_eh_schedule_probe(dev
);
3056 /* if UNLOADING, finish immediately */
3057 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3061 ata_for_each_link(link
, ap
, EDGE
) {
3062 struct ata_eh_context
*ehc
= &link
->eh_context
;
3064 /* skip EH if possible. */
3065 if (ata_eh_skip_recovery(link
))
3068 ata_for_each_dev(dev
, link
, ALL
)
3069 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3073 ata_for_each_link(link
, ap
, EDGE
) {
3074 struct ata_eh_context
*ehc
= &link
->eh_context
;
3076 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3079 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3080 prereset
, softreset
, hardreset
, postreset
);
3082 ata_link_printk(link
, KERN_ERR
,
3083 "reset failed, giving up\n");
3092 * clears ATA_EH_PARK in eh_info and resets
3093 * ap->park_req_pending
3095 ata_eh_pull_park_action(ap
);
3098 ata_for_each_link(link
, ap
, EDGE
) {
3099 ata_for_each_dev(dev
, link
, ALL
) {
3100 struct ata_eh_context
*ehc
= &link
->eh_context
;
3103 if (dev
->class != ATA_DEV_ATA
)
3105 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3108 tmp
= dev
->unpark_deadline
;
3109 if (time_before(deadline
, tmp
))
3111 else if (time_before_eq(tmp
, jiffies
))
3113 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3116 ata_eh_park_issue_cmd(dev
, 1);
3121 if (time_before_eq(deadline
, now
))
3124 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3127 ata_for_each_link(link
, ap
, EDGE
) {
3128 ata_for_each_dev(dev
, link
, ALL
) {
3129 if (!(link
->eh_context
.unloaded_mask
&
3133 ata_eh_park_issue_cmd(dev
, 0);
3134 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3139 ata_for_each_link(link
, ap
, EDGE
) {
3140 struct ata_eh_context
*ehc
= &link
->eh_context
;
3142 /* revalidate existing devices and attach new ones */
3143 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3147 /* if PMP got attached, return, pmp EH will take care of it */
3148 if (link
->device
->class == ATA_DEV_PMP
) {
3153 /* configure transfer mode if necessary */
3154 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3155 rc
= ata_set_mode(link
, &dev
);
3158 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3161 /* If reset has been issued, clear UA to avoid
3162 * disrupting the current users of the device.
3164 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3165 ata_for_each_dev(dev
, link
, ALL
) {
3166 if (dev
->class != ATA_DEV_ATAPI
)
3168 rc
= atapi_eh_clear_ua(dev
);
3174 /* configure link power saving */
3175 if (ehc
->i
.action
& ATA_EH_LPM
)
3176 ata_for_each_dev(dev
, link
, ALL
)
3177 ata_dev_enable_pm(dev
, ap
->pm_policy
);
3179 /* this link is okay now */
3185 ata_eh_handle_dev_fail(dev
, rc
);
3187 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3188 /* PMP reset requires working host port.
3189 * Can't retry if it's frozen.
3191 if (sata_pmp_attached(ap
))
3201 if (rc
&& r_failed_link
)
3202 *r_failed_link
= link
;
3204 DPRINTK("EXIT, rc=%d\n", rc
);
3209 * ata_eh_finish - finish up EH
3210 * @ap: host port to finish EH for
3212 * Recovery is complete. Clean up EH states and retry or finish
3218 void ata_eh_finish(struct ata_port
*ap
)
3222 /* retry or finish qcs */
3223 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3224 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3226 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3230 /* FIXME: Once EH migration is complete,
3231 * generate sense data in this function,
3232 * considering both err_mask and tf.
3234 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3235 ata_eh_qc_retry(qc
);
3237 ata_eh_qc_complete(qc
);
3239 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3240 ata_eh_qc_complete(qc
);
3242 /* feed zero TF to sense generation */
3243 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3244 ata_eh_qc_retry(qc
);
3249 /* make sure nr_active_links is zero after EH */
3250 WARN_ON(ap
->nr_active_links
);
3251 ap
->nr_active_links
= 0;
3255 * ata_do_eh - do standard error handling
3256 * @ap: host port to handle error for
3258 * @prereset: prereset method (can be NULL)
3259 * @softreset: softreset method (can be NULL)
3260 * @hardreset: hardreset method (can be NULL)
3261 * @postreset: postreset method (can be NULL)
3263 * Perform standard error handling sequence.
3266 * Kernel thread context (may sleep).
3268 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3269 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3270 ata_postreset_fn_t postreset
)
3272 struct ata_device
*dev
;
3278 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3281 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3282 ata_dev_disable(dev
);
3289 * ata_std_error_handler - standard error handler
3290 * @ap: host port to handle error for
3292 * Standard error handler
3295 * Kernel thread context (may sleep).
3297 void ata_std_error_handler(struct ata_port
*ap
)
3299 struct ata_port_operations
*ops
= ap
->ops
;
3300 ata_reset_fn_t hardreset
= ops
->hardreset
;
3302 /* ignore built-in hardreset if SCR access is not available */
3303 if (ata_is_builtin_hardreset(hardreset
) && !sata_scr_valid(&ap
->link
))
3306 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
3311 * ata_eh_handle_port_suspend - perform port suspend operation
3312 * @ap: port to suspend
3317 * Kernel thread context (may sleep).
3319 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
3321 unsigned long flags
;
3324 /* are we suspending? */
3325 spin_lock_irqsave(ap
->lock
, flags
);
3326 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3327 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
3328 spin_unlock_irqrestore(ap
->lock
, flags
);
3331 spin_unlock_irqrestore(ap
->lock
, flags
);
3333 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
3335 /* tell ACPI we're suspending */
3336 rc
= ata_acpi_on_suspend(ap
);
3341 ata_eh_freeze_port(ap
);
3343 if (ap
->ops
->port_suspend
)
3344 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
3346 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
3349 spin_lock_irqsave(ap
->lock
, flags
);
3351 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
3353 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
3354 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
3355 ata_port_schedule_eh(ap
);
3357 if (ap
->pm_result
) {
3358 *ap
->pm_result
= rc
;
3359 ap
->pm_result
= NULL
;
3362 spin_unlock_irqrestore(ap
->lock
, flags
);
3368 * ata_eh_handle_port_resume - perform port resume operation
3369 * @ap: port to resume
3374 * Kernel thread context (may sleep).
3376 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
3378 unsigned long flags
;
3381 /* are we resuming? */
3382 spin_lock_irqsave(ap
->lock
, flags
);
3383 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3384 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
3385 spin_unlock_irqrestore(ap
->lock
, flags
);
3388 spin_unlock_irqrestore(ap
->lock
, flags
);
3390 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
3392 ata_acpi_set_state(ap
, PMSG_ON
);
3394 if (ap
->ops
->port_resume
)
3395 rc
= ap
->ops
->port_resume(ap
);
3397 /* tell ACPI that we're resuming */
3398 ata_acpi_on_resume(ap
);
3401 spin_lock_irqsave(ap
->lock
, flags
);
3402 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
3403 if (ap
->pm_result
) {
3404 *ap
->pm_result
= rc
;
3405 ap
->pm_result
= NULL
;
3407 spin_unlock_irqrestore(ap
->lock
, flags
);
3409 #endif /* CONFIG_PM */