2 * libata-eh.c - libata error handling
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
47 #include <linux/libata.h>
52 /* speed down verdicts */
53 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
54 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
55 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
56 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
59 ATA_EFLAG_IS_IO
= (1 << 0),
60 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
61 ATA_EFLAG_OLD_ER
= (1 << 31),
63 /* error categories */
66 ATA_ECAT_TOUT_HSM
= 2,
68 ATA_ECAT_DUBIOUS_NONE
= 4,
69 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
70 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
71 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
74 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
76 /* always put at least this amount of time between resets */
77 ATA_EH_RESET_COOL_DOWN
= 5000,
79 /* Waiting in ->prereset can never be reliable. It's
80 * sometimes nice to wait there but it can't be depended upon;
81 * otherwise, we wouldn't be resetting. Just give it enough
82 * time for most drives to spin up.
84 ATA_EH_PRERESET_TIMEOUT
= 10000,
85 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
89 /* probe speed down parameters, see ata_eh_schedule_probe() */
90 ATA_EH_PROBE_TRIAL_INTERVAL
= 60000, /* 1 min */
91 ATA_EH_PROBE_TRIALS
= 2,
94 /* The following table determines how we sequence resets. Each entry
95 * represents timeout for that try. The first try can be soft or
96 * hardreset. All others are hardreset if available. In most cases
97 * the first reset w/ 10sec timeout should succeed. Following entries
98 * are mostly for error handling, hotplug and retarded devices.
100 static const unsigned long ata_eh_reset_timeouts
[] = {
101 10000, /* most drives spin up by 10sec */
102 10000, /* > 99% working drives spin up before 20sec */
103 35000, /* give > 30 secs of idleness for retarded devices */
104 5000, /* and sweet one last chance */
105 ULONG_MAX
, /* > 1 min has elapsed, give up */
108 static const unsigned long ata_eh_identify_timeouts
[] = {
109 5000, /* covers > 99% of successes and not too boring on failures */
110 10000, /* combined time till here is enough even for media access */
111 30000, /* for true idiots */
115 static const unsigned long ata_eh_flush_timeouts
[] = {
116 15000, /* be generous with flush */
118 30000, /* and even more generous */
122 static const unsigned long ata_eh_other_timeouts
[] = {
123 5000, /* same rationale as identify timeout */
125 /* but no merciful 30sec for other commands, it just isn't worth it */
129 struct ata_eh_cmd_timeout_ent
{
131 const unsigned long *timeouts
;
134 /* The following table determines timeouts to use for EH internal
135 * commands. Each table entry is a command class and matches the
136 * commands the entry applies to and the timeout table to use.
138 * On the retry after a command timed out, the next timeout value from
139 * the table is used. If the table doesn't contain further entries,
140 * the last value is used.
142 * ehc->cmd_timeout_idx keeps track of which timeout to use per
143 * command class, so if SET_FEATURES times out on the first try, the
144 * next try will use the second timeout value only for that class.
146 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
147 static const struct ata_eh_cmd_timeout_ent
148 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
149 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
150 .timeouts
= ata_eh_identify_timeouts
, },
151 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
152 .timeouts
= ata_eh_other_timeouts
, },
153 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
154 .timeouts
= ata_eh_other_timeouts
, },
155 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
156 .timeouts
= ata_eh_other_timeouts
, },
157 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
158 .timeouts
= ata_eh_other_timeouts
, },
159 { .commands
= CMDS(ATA_CMD_FLUSH
, ATA_CMD_FLUSH_EXT
),
160 .timeouts
= ata_eh_flush_timeouts
},
164 static void __ata_port_freeze(struct ata_port
*ap
);
166 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
167 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
168 #else /* CONFIG_PM */
169 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
172 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
174 #endif /* CONFIG_PM */
176 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
179 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
180 ATA_EH_DESC_LEN
- ehi
->desc_len
,
185 * __ata_ehi_push_desc - push error description without adding separator
187 * @fmt: printf format string
189 * Format string according to @fmt and append it to @ehi->desc.
192 * spin_lock_irqsave(host lock)
194 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
199 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
204 * ata_ehi_push_desc - push error description with separator
206 * @fmt: printf format string
208 * Format string according to @fmt and append it to @ehi->desc.
209 * If @ehi->desc is not empty, ", " is added in-between.
212 * spin_lock_irqsave(host lock)
214 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
219 __ata_ehi_push_desc(ehi
, ", ");
222 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
227 * ata_ehi_clear_desc - clean error description
233 * spin_lock_irqsave(host lock)
235 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
242 * ata_port_desc - append port description
243 * @ap: target ATA port
244 * @fmt: printf format string
246 * Format string according to @fmt and append it to port
247 * description. If port description is not empty, " " is added
248 * in-between. This function is to be used while initializing
249 * ata_host. The description is printed on host registration.
254 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
258 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
260 if (ap
->link
.eh_info
.desc_len
)
261 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
264 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
271 * ata_port_pbar_desc - append PCI BAR description
272 * @ap: target ATA port
273 * @bar: target PCI BAR
274 * @offset: offset into PCI BAR
275 * @name: name of the area
277 * If @offset is negative, this function formats a string which
278 * contains the name, address, size and type of the BAR and
279 * appends it to the port description. If @offset is zero or
280 * positive, only name and offsetted address is appended.
285 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
288 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
290 unsigned long long start
, len
;
292 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
294 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
297 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
298 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
301 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
303 ata_port_desc(ap
, "%s 0x%llx", name
,
304 start
+ (unsigned long long)offset
);
307 #endif /* CONFIG_PCI */
309 static int ata_lookup_timeout_table(u8 cmd
)
313 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
316 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
325 * ata_internal_cmd_timeout - determine timeout for an internal command
326 * @dev: target device
327 * @cmd: internal command to be issued
329 * Determine timeout for internal command @cmd for @dev.
335 * Determined timeout.
337 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
339 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
340 int ent
= ata_lookup_timeout_table(cmd
);
344 return ATA_EH_CMD_DFL_TIMEOUT
;
346 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
347 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
351 * ata_internal_cmd_timed_out - notification for internal command timeout
352 * @dev: target device
353 * @cmd: internal command which timed out
355 * Notify EH that internal command @cmd for @dev timed out. This
356 * function should be called only for commands whose timeouts are
357 * determined using ata_internal_cmd_timeout().
362 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
364 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
365 int ent
= ata_lookup_timeout_table(cmd
);
371 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
372 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
373 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
376 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
377 unsigned int err_mask
)
379 struct ata_ering_entry
*ent
;
384 ering
->cursor
%= ATA_ERING_SIZE
;
386 ent
= &ering
->ring
[ering
->cursor
];
387 ent
->eflags
= eflags
;
388 ent
->err_mask
= err_mask
;
389 ent
->timestamp
= get_jiffies_64();
392 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
394 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
401 int ata_ering_map(struct ata_ering
*ering
,
402 int (*map_fn
)(struct ata_ering_entry
*, void *),
406 struct ata_ering_entry
*ent
;
410 ent
= &ering
->ring
[idx
];
413 rc
= map_fn(ent
, arg
);
416 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
417 } while (idx
!= ering
->cursor
);
422 static int ata_ering_clear_cb(struct ata_ering_entry
*ent
, void *void_arg
)
424 ent
->eflags
|= ATA_EFLAG_OLD_ER
;
428 static void ata_ering_clear(struct ata_ering
*ering
)
430 ata_ering_map(ering
, ata_ering_clear_cb
, NULL
);
433 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
435 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
437 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
440 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
441 struct ata_eh_info
*ehi
, unsigned int action
)
443 struct ata_device
*tdev
;
446 ehi
->action
&= ~action
;
447 ata_for_each_dev(tdev
, link
, ALL
)
448 ehi
->dev_action
[tdev
->devno
] &= ~action
;
450 /* doesn't make sense for port-wide EH actions */
451 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
453 /* break ehi->action into ehi->dev_action */
454 if (ehi
->action
& action
) {
455 ata_for_each_dev(tdev
, link
, ALL
)
456 ehi
->dev_action
[tdev
->devno
] |=
457 ehi
->action
& action
;
458 ehi
->action
&= ~action
;
461 /* turn off the specified per-dev action */
462 ehi
->dev_action
[dev
->devno
] &= ~action
;
467 * ata_eh_acquire - acquire EH ownership
468 * @ap: ATA port to acquire EH ownership for
470 * Acquire EH ownership for @ap. This is the basic exclusion
471 * mechanism for ports sharing a host. Only one port hanging off
472 * the same host can claim the ownership of EH.
477 void ata_eh_acquire(struct ata_port
*ap
)
479 mutex_lock(&ap
->host
->eh_mutex
);
480 WARN_ON_ONCE(ap
->host
->eh_owner
);
481 ap
->host
->eh_owner
= current
;
485 * ata_eh_release - release EH ownership
486 * @ap: ATA port to release EH ownership for
488 * Release EH ownership for @ap if the caller. The caller must
489 * have acquired EH ownership using ata_eh_acquire() previously.
494 void ata_eh_release(struct ata_port
*ap
)
496 WARN_ON_ONCE(ap
->host
->eh_owner
!= current
);
497 ap
->host
->eh_owner
= NULL
;
498 mutex_unlock(&ap
->host
->eh_mutex
);
502 * ata_scsi_timed_out - SCSI layer time out callback
503 * @cmd: timed out SCSI command
505 * Handles SCSI layer timeout. We race with normal completion of
506 * the qc for @cmd. If the qc is already gone, we lose and let
507 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
508 * timed out and EH should be invoked. Prevent ata_qc_complete()
509 * from finishing it by setting EH_SCHEDULED and return
512 * TODO: kill this function once old EH is gone.
515 * Called from timer context
518 * EH_HANDLED or EH_NOT_HANDLED
520 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
522 struct Scsi_Host
*host
= cmd
->device
->host
;
523 struct ata_port
*ap
= ata_shost_to_port(host
);
525 struct ata_queued_cmd
*qc
;
526 enum blk_eh_timer_return ret
;
530 if (ap
->ops
->error_handler
) {
531 ret
= BLK_EH_NOT_HANDLED
;
535 ret
= BLK_EH_HANDLED
;
536 spin_lock_irqsave(ap
->lock
, flags
);
537 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
539 WARN_ON(qc
->scsicmd
!= cmd
);
540 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
541 qc
->err_mask
|= AC_ERR_TIMEOUT
;
542 ret
= BLK_EH_NOT_HANDLED
;
544 spin_unlock_irqrestore(ap
->lock
, flags
);
547 DPRINTK("EXIT, ret=%d\n", ret
);
551 static void ata_eh_unload(struct ata_port
*ap
)
553 struct ata_link
*link
;
554 struct ata_device
*dev
;
557 /* Restore SControl IPM and SPD for the next driver and
558 * disable attached devices.
560 ata_for_each_link(link
, ap
, PMP_FIRST
) {
561 sata_scr_write(link
, SCR_CONTROL
, link
->saved_scontrol
& 0xff0);
562 ata_for_each_dev(dev
, link
, ALL
)
563 ata_dev_disable(dev
);
566 /* freeze and set UNLOADED */
567 spin_lock_irqsave(ap
->lock
, flags
);
569 ata_port_freeze(ap
); /* won't be thawed */
570 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
; /* clear pending from freeze */
571 ap
->pflags
|= ATA_PFLAG_UNLOADED
;
573 spin_unlock_irqrestore(ap
->lock
, flags
);
577 * ata_scsi_error - SCSI layer error handler callback
578 * @host: SCSI host on which error occurred
580 * Handles SCSI-layer-thrown error events.
583 * Inherited from SCSI layer (none, can sleep)
588 void ata_scsi_error(struct Scsi_Host
*host
)
590 struct ata_port
*ap
= ata_shost_to_port(host
);
592 LIST_HEAD(eh_work_q
);
596 spin_lock_irqsave(host
->host_lock
, flags
);
597 list_splice_init(&host
->eh_cmd_q
, &eh_work_q
);
598 spin_unlock_irqrestore(host
->host_lock
, flags
);
600 ata_scsi_cmd_error_handler(host
, ap
, &eh_work_q
);
602 /* If we timed raced normal completion and there is nothing to
603 recover nr_timedout == 0 why exactly are we doing error recovery ? */
604 ata_scsi_port_error_handler(host
, ap
);
606 /* finish or retry handled scmd's and clean up */
607 WARN_ON(host
->host_failed
|| !list_empty(&eh_work_q
));
613 * ata_scsi_cmd_error_handler - error callback for a list of commands
614 * @host: scsi host containing the port
615 * @ap: ATA port within the host
616 * @eh_work_q: list of commands to process
618 * process the given list of commands and return those finished to the
619 * ap->eh_done_q. This function is the first part of the libata error
620 * handler which processes a given list of failed commands.
622 void ata_scsi_cmd_error_handler(struct Scsi_Host
*host
, struct ata_port
*ap
,
623 struct list_head
*eh_work_q
)
628 /* make sure sff pio task is not running */
629 ata_sff_flush_pio_task(ap
);
631 /* synchronize with host lock and sort out timeouts */
633 /* For new EH, all qcs are finished in one of three ways -
634 * normal completion, error completion, and SCSI timeout.
635 * Both completions can race against SCSI timeout. When normal
636 * completion wins, the qc never reaches EH. When error
637 * completion wins, the qc has ATA_QCFLAG_FAILED set.
639 * When SCSI timeout wins, things are a bit more complex.
640 * Normal or error completion can occur after the timeout but
641 * before this point. In such cases, both types of
642 * completions are honored. A scmd is determined to have
643 * timed out iff its associated qc is active and not failed.
645 if (ap
->ops
->error_handler
) {
646 struct scsi_cmnd
*scmd
, *tmp
;
649 spin_lock_irqsave(ap
->lock
, flags
);
651 /* This must occur under the ap->lock as we don't want
652 a polled recovery to race the real interrupt handler
654 The lost_interrupt handler checks for any completed but
655 non-notified command and completes much like an IRQ handler.
657 We then fall into the error recovery code which will treat
658 this as if normal completion won the race */
660 if (ap
->ops
->lost_interrupt
)
661 ap
->ops
->lost_interrupt(ap
);
663 list_for_each_entry_safe(scmd
, tmp
, eh_work_q
, eh_entry
) {
664 struct ata_queued_cmd
*qc
;
666 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
667 qc
= __ata_qc_from_tag(ap
, i
);
668 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
673 if (i
< ATA_MAX_QUEUE
) {
674 /* the scmd has an associated qc */
675 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
676 /* which hasn't failed yet, timeout */
677 qc
->err_mask
|= AC_ERR_TIMEOUT
;
678 qc
->flags
|= ATA_QCFLAG_FAILED
;
682 /* Normal completion occurred after
683 * SCSI timeout but before this point.
684 * Successfully complete it.
686 scmd
->retries
= scmd
->allowed
;
687 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
691 /* If we have timed out qcs. They belong to EH from
692 * this point but the state of the controller is
693 * unknown. Freeze the port to make sure the IRQ
694 * handler doesn't diddle with those qcs. This must
695 * be done atomically w.r.t. setting QCFLAG_FAILED.
698 __ata_port_freeze(ap
);
700 spin_unlock_irqrestore(ap
->lock
, flags
);
702 /* initialize eh_tries */
703 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
705 spin_unlock_wait(ap
->lock
);
708 EXPORT_SYMBOL(ata_scsi_cmd_error_handler
);
711 * ata_scsi_port_error_handler - recover the port after the commands
712 * @host: SCSI host containing the port
715 * Handle the recovery of the port @ap after all the commands
716 * have been recovered.
718 void ata_scsi_port_error_handler(struct Scsi_Host
*host
, struct ata_port
*ap
)
722 /* invoke error handler */
723 if (ap
->ops
->error_handler
) {
724 struct ata_link
*link
;
726 /* acquire EH ownership */
729 /* kill fast drain timer */
730 del_timer_sync(&ap
->fastdrain_timer
);
732 /* process port resume request */
733 ata_eh_handle_port_resume(ap
);
735 /* fetch & clear EH info */
736 spin_lock_irqsave(ap
->lock
, flags
);
738 ata_for_each_link(link
, ap
, HOST_FIRST
) {
739 struct ata_eh_context
*ehc
= &link
->eh_context
;
740 struct ata_device
*dev
;
742 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
743 link
->eh_context
.i
= link
->eh_info
;
744 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
746 ata_for_each_dev(dev
, link
, ENABLED
) {
747 int devno
= dev
->devno
;
749 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
750 if (ata_ncq_enabled(dev
))
751 ehc
->saved_ncq_enabled
|= 1 << devno
;
755 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
756 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
757 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
759 spin_unlock_irqrestore(ap
->lock
, flags
);
761 /* invoke EH, skip if unloading or suspended */
762 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
763 ap
->ops
->error_handler(ap
);
765 /* if unloading, commence suicide */
766 if ((ap
->pflags
& ATA_PFLAG_UNLOADING
) &&
767 !(ap
->pflags
& ATA_PFLAG_UNLOADED
))
772 /* process port suspend request */
773 ata_eh_handle_port_suspend(ap
);
775 /* Exception might have happened after ->error_handler
776 * recovered the port but before this point. Repeat
779 spin_lock_irqsave(ap
->lock
, flags
);
781 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
782 if (--ap
->eh_tries
) {
783 spin_unlock_irqrestore(ap
->lock
, flags
);
787 "EH pending after %d tries, giving up\n",
789 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
792 /* this run is complete, make sure EH info is clear */
793 ata_for_each_link(link
, ap
, HOST_FIRST
)
794 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
796 /* end eh (clear host_eh_scheduled) while holding
797 * ap->lock such that if exception occurs after this
798 * point but before EH completion, SCSI midlayer will
803 spin_unlock_irqrestore(ap
->lock
, flags
);
806 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
807 ap
->ops
->eng_timeout(ap
);
810 scsi_eh_flush_done_q(&ap
->eh_done_q
);
813 spin_lock_irqsave(ap
->lock
, flags
);
815 if (ap
->pflags
& ATA_PFLAG_LOADING
)
816 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
817 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
818 schedule_delayed_work(&ap
->hotplug_task
, 0);
820 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
821 ata_port_info(ap
, "EH complete\n");
823 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
825 /* tell wait_eh that we're done */
826 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
827 wake_up_all(&ap
->eh_wait_q
);
829 spin_unlock_irqrestore(ap
->lock
, flags
);
831 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler
);
834 * ata_port_wait_eh - Wait for the currently pending EH to complete
835 * @ap: Port to wait EH for
837 * Wait until the currently pending EH is complete.
840 * Kernel thread context (may sleep).
842 void ata_port_wait_eh(struct ata_port
*ap
)
848 spin_lock_irqsave(ap
->lock
, flags
);
850 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
851 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
852 spin_unlock_irqrestore(ap
->lock
, flags
);
854 spin_lock_irqsave(ap
->lock
, flags
);
856 finish_wait(&ap
->eh_wait_q
, &wait
);
858 spin_unlock_irqrestore(ap
->lock
, flags
);
860 /* make sure SCSI EH is complete */
861 if (scsi_host_in_recovery(ap
->scsi_host
)) {
866 EXPORT_SYMBOL_GPL(ata_port_wait_eh
);
868 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
873 /* count only non-internal commands */
874 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
875 if (ata_qc_from_tag(ap
, tag
))
881 void ata_eh_fastdrain_timerfn(unsigned long arg
)
883 struct ata_port
*ap
= (void *)arg
;
887 spin_lock_irqsave(ap
->lock
, flags
);
889 cnt
= ata_eh_nr_in_flight(ap
);
895 if (cnt
== ap
->fastdrain_cnt
) {
898 /* No progress during the last interval, tag all
899 * in-flight qcs as timed out and freeze the port.
901 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
902 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
904 qc
->err_mask
|= AC_ERR_TIMEOUT
;
909 /* some qcs have finished, give it another chance */
910 ap
->fastdrain_cnt
= cnt
;
911 ap
->fastdrain_timer
.expires
=
912 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
913 add_timer(&ap
->fastdrain_timer
);
917 spin_unlock_irqrestore(ap
->lock
, flags
);
921 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
922 * @ap: target ATA port
923 * @fastdrain: activate fast drain
925 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
926 * is non-zero and EH wasn't pending before. Fast drain ensures
927 * that EH kicks in in timely manner.
930 * spin_lock_irqsave(host lock)
932 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
936 /* already scheduled? */
937 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
940 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
945 /* do we have in-flight qcs? */
946 cnt
= ata_eh_nr_in_flight(ap
);
950 /* activate fast drain */
951 ap
->fastdrain_cnt
= cnt
;
952 ap
->fastdrain_timer
.expires
=
953 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
954 add_timer(&ap
->fastdrain_timer
);
958 * ata_qc_schedule_eh - schedule qc for error handling
959 * @qc: command to schedule error handling for
961 * Schedule error handling for @qc. EH will kick in as soon as
962 * other commands are drained.
965 * spin_lock_irqsave(host lock)
967 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
969 struct ata_port
*ap
= qc
->ap
;
970 struct request_queue
*q
= qc
->scsicmd
->device
->request_queue
;
973 WARN_ON(!ap
->ops
->error_handler
);
975 qc
->flags
|= ATA_QCFLAG_FAILED
;
976 ata_eh_set_pending(ap
, 1);
978 /* The following will fail if timeout has already expired.
979 * ata_scsi_error() takes care of such scmds on EH entry.
980 * Note that ATA_QCFLAG_FAILED is unconditionally set after
981 * this function completes.
983 spin_lock_irqsave(q
->queue_lock
, flags
);
984 blk_abort_request(qc
->scsicmd
->request
);
985 spin_unlock_irqrestore(q
->queue_lock
, flags
);
989 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
990 * @ap: ATA port to schedule EH for
992 * LOCKING: inherited from ata_port_schedule_eh
993 * spin_lock_irqsave(host lock)
995 void ata_std_sched_eh(struct ata_port
*ap
)
997 WARN_ON(!ap
->ops
->error_handler
);
999 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
1002 ata_eh_set_pending(ap
, 1);
1003 scsi_schedule_eh(ap
->scsi_host
);
1005 DPRINTK("port EH scheduled\n");
1007 EXPORT_SYMBOL_GPL(ata_std_sched_eh
);
1010 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1011 * @ap: ATA port to end EH for
1013 * In the libata object model there is a 1:1 mapping of ata_port to
1014 * shost, so host fields can be directly manipulated under ap->lock, in
1015 * the libsas case we need to hold a lock at the ha->level to coordinate
1019 * spin_lock_irqsave(host lock)
1021 void ata_std_end_eh(struct ata_port
*ap
)
1023 struct Scsi_Host
*host
= ap
->scsi_host
;
1025 host
->host_eh_scheduled
= 0;
1027 EXPORT_SYMBOL(ata_std_end_eh
);
1031 * ata_port_schedule_eh - schedule error handling without a qc
1032 * @ap: ATA port to schedule EH for
1034 * Schedule error handling for @ap. EH will kick in as soon as
1035 * all commands are drained.
1038 * spin_lock_irqsave(host lock)
1040 void ata_port_schedule_eh(struct ata_port
*ap
)
1042 /* see: ata_std_sched_eh, unless you know better */
1043 ap
->ops
->sched_eh(ap
);
1046 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
1048 int tag
, nr_aborted
= 0;
1050 WARN_ON(!ap
->ops
->error_handler
);
1052 /* we're gonna abort all commands, no need for fast drain */
1053 ata_eh_set_pending(ap
, 0);
1055 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1056 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
1058 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
1059 qc
->flags
|= ATA_QCFLAG_FAILED
;
1060 ata_qc_complete(qc
);
1066 ata_port_schedule_eh(ap
);
1072 * ata_link_abort - abort all qc's on the link
1073 * @link: ATA link to abort qc's for
1075 * Abort all active qc's active on @link and schedule EH.
1078 * spin_lock_irqsave(host lock)
1081 * Number of aborted qc's.
1083 int ata_link_abort(struct ata_link
*link
)
1085 return ata_do_link_abort(link
->ap
, link
);
1089 * ata_port_abort - abort all qc's on the port
1090 * @ap: ATA port to abort qc's for
1092 * Abort all active qc's of @ap and schedule EH.
1095 * spin_lock_irqsave(host_set lock)
1098 * Number of aborted qc's.
1100 int ata_port_abort(struct ata_port
*ap
)
1102 return ata_do_link_abort(ap
, NULL
);
1106 * __ata_port_freeze - freeze port
1107 * @ap: ATA port to freeze
1109 * This function is called when HSM violation or some other
1110 * condition disrupts normal operation of the port. Frozen port
1111 * is not allowed to perform any operation until the port is
1112 * thawed, which usually follows a successful reset.
1114 * ap->ops->freeze() callback can be used for freezing the port
1115 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1116 * port cannot be frozen hardware-wise, the interrupt handler
1117 * must ack and clear interrupts unconditionally while the port
1121 * spin_lock_irqsave(host lock)
1123 static void __ata_port_freeze(struct ata_port
*ap
)
1125 WARN_ON(!ap
->ops
->error_handler
);
1127 if (ap
->ops
->freeze
)
1128 ap
->ops
->freeze(ap
);
1130 ap
->pflags
|= ATA_PFLAG_FROZEN
;
1132 DPRINTK("ata%u port frozen\n", ap
->print_id
);
1136 * ata_port_freeze - abort & freeze port
1137 * @ap: ATA port to freeze
1139 * Abort and freeze @ap. The freeze operation must be called
1140 * first, because some hardware requires special operations
1141 * before the taskfile registers are accessible.
1144 * spin_lock_irqsave(host lock)
1147 * Number of aborted commands.
1149 int ata_port_freeze(struct ata_port
*ap
)
1153 WARN_ON(!ap
->ops
->error_handler
);
1155 __ata_port_freeze(ap
);
1156 nr_aborted
= ata_port_abort(ap
);
1162 * sata_async_notification - SATA async notification handler
1163 * @ap: ATA port where async notification is received
1165 * Handler to be called when async notification via SDB FIS is
1166 * received. This function schedules EH if necessary.
1169 * spin_lock_irqsave(host lock)
1172 * 1 if EH is scheduled, 0 otherwise.
1174 int sata_async_notification(struct ata_port
*ap
)
1179 if (!(ap
->flags
& ATA_FLAG_AN
))
1182 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
1184 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
1186 if (!sata_pmp_attached(ap
) || rc
) {
1187 /* PMP is not attached or SNTF is not available */
1188 if (!sata_pmp_attached(ap
)) {
1189 /* PMP is not attached. Check whether ATAPI
1190 * AN is configured. If so, notify media
1193 struct ata_device
*dev
= ap
->link
.device
;
1195 if ((dev
->class == ATA_DEV_ATAPI
) &&
1196 (dev
->flags
& ATA_DFLAG_AN
))
1197 ata_scsi_media_change_notify(dev
);
1200 /* PMP is attached but SNTF is not available.
1201 * ATAPI async media change notification is
1202 * not used. The PMP must be reporting PHY
1203 * status change, schedule EH.
1205 ata_port_schedule_eh(ap
);
1209 /* PMP is attached and SNTF is available */
1210 struct ata_link
*link
;
1212 /* check and notify ATAPI AN */
1213 ata_for_each_link(link
, ap
, EDGE
) {
1214 if (!(sntf
& (1 << link
->pmp
)))
1217 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1218 (link
->device
->flags
& ATA_DFLAG_AN
))
1219 ata_scsi_media_change_notify(link
->device
);
1222 /* If PMP is reporting that PHY status of some
1223 * downstream ports has changed, schedule EH.
1225 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1226 ata_port_schedule_eh(ap
);
1235 * ata_eh_freeze_port - EH helper to freeze port
1236 * @ap: ATA port to freeze
1243 void ata_eh_freeze_port(struct ata_port
*ap
)
1245 unsigned long flags
;
1247 if (!ap
->ops
->error_handler
)
1250 spin_lock_irqsave(ap
->lock
, flags
);
1251 __ata_port_freeze(ap
);
1252 spin_unlock_irqrestore(ap
->lock
, flags
);
1256 * ata_port_thaw_port - EH helper to thaw port
1257 * @ap: ATA port to thaw
1259 * Thaw frozen port @ap.
1264 void ata_eh_thaw_port(struct ata_port
*ap
)
1266 unsigned long flags
;
1268 if (!ap
->ops
->error_handler
)
1271 spin_lock_irqsave(ap
->lock
, flags
);
1273 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1278 spin_unlock_irqrestore(ap
->lock
, flags
);
1280 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1283 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1288 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1290 struct ata_port
*ap
= qc
->ap
;
1291 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1292 unsigned long flags
;
1294 spin_lock_irqsave(ap
->lock
, flags
);
1295 qc
->scsidone
= ata_eh_scsidone
;
1296 __ata_qc_complete(qc
);
1297 WARN_ON(ata_tag_valid(qc
->tag
));
1298 spin_unlock_irqrestore(ap
->lock
, flags
);
1300 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1304 * ata_eh_qc_complete - Complete an active ATA command from EH
1305 * @qc: Command to complete
1307 * Indicate to the mid and upper layers that an ATA command has
1308 * completed. To be used from EH.
1310 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1312 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1313 scmd
->retries
= scmd
->allowed
;
1314 __ata_eh_qc_complete(qc
);
1318 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1319 * @qc: Command to retry
1321 * Indicate to the mid and upper layers that an ATA command
1322 * should be retried. To be used from EH.
1324 * SCSI midlayer limits the number of retries to scmd->allowed.
1325 * scmd->retries is decremented for commands which get retried
1326 * due to unrelated failures (qc->err_mask is zero).
1328 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1330 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1331 if (!qc
->err_mask
&& scmd
->retries
)
1333 __ata_eh_qc_complete(qc
);
1337 * ata_dev_disable - disable ATA device
1338 * @dev: ATA device to disable
1345 void ata_dev_disable(struct ata_device
*dev
)
1347 if (!ata_dev_enabled(dev
))
1350 if (ata_msg_drv(dev
->link
->ap
))
1351 ata_dev_warn(dev
, "disabled\n");
1352 ata_acpi_on_disable(dev
);
1353 ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO0
| ATA_DNXFER_QUIET
);
1356 /* From now till the next successful probe, ering is used to
1357 * track probe failures. Clear accumulated device error info.
1359 ata_ering_clear(&dev
->ering
);
1363 * ata_eh_detach_dev - detach ATA device
1364 * @dev: ATA device to detach
1371 void ata_eh_detach_dev(struct ata_device
*dev
)
1373 struct ata_link
*link
= dev
->link
;
1374 struct ata_port
*ap
= link
->ap
;
1375 struct ata_eh_context
*ehc
= &link
->eh_context
;
1376 unsigned long flags
;
1378 ata_dev_disable(dev
);
1380 spin_lock_irqsave(ap
->lock
, flags
);
1382 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1384 if (ata_scsi_offline_dev(dev
)) {
1385 dev
->flags
|= ATA_DFLAG_DETACHED
;
1386 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1389 /* clear per-dev EH info */
1390 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1391 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1392 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1393 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1395 spin_unlock_irqrestore(ap
->lock
, flags
);
1399 * ata_eh_about_to_do - about to perform eh_action
1400 * @link: target ATA link
1401 * @dev: target ATA dev for per-dev action (can be NULL)
1402 * @action: action about to be performed
1404 * Called just before performing EH actions to clear related bits
1405 * in @link->eh_info such that eh actions are not unnecessarily
1411 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1412 unsigned int action
)
1414 struct ata_port
*ap
= link
->ap
;
1415 struct ata_eh_info
*ehi
= &link
->eh_info
;
1416 struct ata_eh_context
*ehc
= &link
->eh_context
;
1417 unsigned long flags
;
1419 spin_lock_irqsave(ap
->lock
, flags
);
1421 ata_eh_clear_action(link
, dev
, ehi
, action
);
1423 /* About to take EH action, set RECOVERED. Ignore actions on
1424 * slave links as master will do them again.
1426 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1427 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1429 spin_unlock_irqrestore(ap
->lock
, flags
);
1433 * ata_eh_done - EH action complete
1434 * @ap: target ATA port
1435 * @dev: target ATA dev for per-dev action (can be NULL)
1436 * @action: action just completed
1438 * Called right after performing EH actions to clear related bits
1439 * in @link->eh_context.
1444 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1445 unsigned int action
)
1447 struct ata_eh_context
*ehc
= &link
->eh_context
;
1449 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1453 * ata_err_string - convert err_mask to descriptive string
1454 * @err_mask: error mask to convert to string
1456 * Convert @err_mask to descriptive string. Errors are
1457 * prioritized according to severity and only the most severe
1458 * error is reported.
1464 * Descriptive string for @err_mask
1466 static const char *ata_err_string(unsigned int err_mask
)
1468 if (err_mask
& AC_ERR_HOST_BUS
)
1469 return "host bus error";
1470 if (err_mask
& AC_ERR_ATA_BUS
)
1471 return "ATA bus error";
1472 if (err_mask
& AC_ERR_TIMEOUT
)
1474 if (err_mask
& AC_ERR_HSM
)
1475 return "HSM violation";
1476 if (err_mask
& AC_ERR_SYSTEM
)
1477 return "internal error";
1478 if (err_mask
& AC_ERR_MEDIA
)
1479 return "media error";
1480 if (err_mask
& AC_ERR_INVALID
)
1481 return "invalid argument";
1482 if (err_mask
& AC_ERR_DEV
)
1483 return "device error";
1484 return "unknown error";
1488 * ata_read_log_page - read a specific log page
1489 * @dev: target device
1491 * @page: page to read
1492 * @buf: buffer to store read page
1493 * @sectors: number of sectors to read
1495 * Read log page using READ_LOG_EXT command.
1498 * Kernel thread context (may sleep).
1501 * 0 on success, AC_ERR_* mask otherwise.
1503 unsigned int ata_read_log_page(struct ata_device
*dev
, u8 log
,
1504 u8 page
, void *buf
, unsigned int sectors
)
1506 struct ata_taskfile tf
;
1507 unsigned int err_mask
;
1509 DPRINTK("read log page - log 0x%x, page 0x%x\n", log
, page
);
1511 ata_tf_init(dev
, &tf
);
1512 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1516 tf
.hob_nsect
= sectors
>> 8;
1517 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1518 tf
.protocol
= ATA_PROT_PIO
;
1520 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1521 buf
, sectors
* ATA_SECT_SIZE
, 0);
1523 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1528 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1529 * @dev: Device to read log page 10h from
1530 * @tag: Resulting tag of the failed command
1531 * @tf: Resulting taskfile registers of the failed command
1533 * Read log page 10h to obtain NCQ error details and clear error
1537 * Kernel thread context (may sleep).
1540 * 0 on success, -errno otherwise.
1542 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1543 int *tag
, struct ata_taskfile
*tf
)
1545 u8
*buf
= dev
->link
->ap
->sector_buf
;
1546 unsigned int err_mask
;
1550 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, 0, buf
, 1);
1555 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1558 ata_dev_warn(dev
, "invalid checksum 0x%x on log page 10h\n",
1564 *tag
= buf
[0] & 0x1f;
1566 tf
->command
= buf
[2];
1567 tf
->feature
= buf
[3];
1571 tf
->device
= buf
[7];
1572 tf
->hob_lbal
= buf
[8];
1573 tf
->hob_lbam
= buf
[9];
1574 tf
->hob_lbah
= buf
[10];
1575 tf
->nsect
= buf
[12];
1576 tf
->hob_nsect
= buf
[13];
1582 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1583 * @dev: target ATAPI device
1584 * @r_sense_key: out parameter for sense_key
1586 * Perform ATAPI TEST_UNIT_READY.
1589 * EH context (may sleep).
1592 * 0 on success, AC_ERR_* mask on failure.
1594 unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1596 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1597 struct ata_taskfile tf
;
1598 unsigned int err_mask
;
1600 ata_tf_init(dev
, &tf
);
1602 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1603 tf
.command
= ATA_CMD_PACKET
;
1604 tf
.protocol
= ATAPI_PROT_NODATA
;
1606 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1607 if (err_mask
== AC_ERR_DEV
)
1608 *r_sense_key
= tf
.feature
>> 4;
1613 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1614 * @dev: device to perform REQUEST_SENSE to
1615 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1616 * @dfl_sense_key: default sense key to use
1618 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1619 * SENSE. This function is EH helper.
1622 * Kernel thread context (may sleep).
1625 * 0 on success, AC_ERR_* mask on failure
1627 unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1628 u8
*sense_buf
, u8 dfl_sense_key
)
1630 u8 cdb
[ATAPI_CDB_LEN
] =
1631 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1632 struct ata_port
*ap
= dev
->link
->ap
;
1633 struct ata_taskfile tf
;
1635 DPRINTK("ATAPI request sense\n");
1637 /* FIXME: is this needed? */
1638 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1640 /* initialize sense_buf with the error register,
1641 * for the case where they are -not- overwritten
1643 sense_buf
[0] = 0x70;
1644 sense_buf
[2] = dfl_sense_key
;
1646 /* some devices time out if garbage left in tf */
1647 ata_tf_init(dev
, &tf
);
1649 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1650 tf
.command
= ATA_CMD_PACKET
;
1652 /* is it pointless to prefer PIO for "safety reasons"? */
1653 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1654 tf
.protocol
= ATAPI_PROT_DMA
;
1655 tf
.feature
|= ATAPI_PKT_DMA
;
1657 tf
.protocol
= ATAPI_PROT_PIO
;
1658 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1662 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1663 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1667 * ata_eh_analyze_serror - analyze SError for a failed port
1668 * @link: ATA link to analyze SError for
1670 * Analyze SError if available and further determine cause of
1676 static void ata_eh_analyze_serror(struct ata_link
*link
)
1678 struct ata_eh_context
*ehc
= &link
->eh_context
;
1679 u32 serror
= ehc
->i
.serror
;
1680 unsigned int err_mask
= 0, action
= 0;
1683 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1684 err_mask
|= AC_ERR_ATA_BUS
;
1685 action
|= ATA_EH_RESET
;
1687 if (serror
& SERR_PROTOCOL
) {
1688 err_mask
|= AC_ERR_HSM
;
1689 action
|= ATA_EH_RESET
;
1691 if (serror
& SERR_INTERNAL
) {
1692 err_mask
|= AC_ERR_SYSTEM
;
1693 action
|= ATA_EH_RESET
;
1696 /* Determine whether a hotplug event has occurred. Both
1697 * SError.N/X are considered hotplug events for enabled or
1698 * host links. For disabled PMP links, only N bit is
1699 * considered as X bit is left at 1 for link plugging.
1701 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
)
1702 hotplug_mask
= 0; /* hotplug doesn't work w/ LPM */
1703 else if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1704 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1706 hotplug_mask
= SERR_PHYRDY_CHG
;
1708 if (serror
& hotplug_mask
)
1709 ata_ehi_hotplugged(&ehc
->i
);
1711 ehc
->i
.err_mask
|= err_mask
;
1712 ehc
->i
.action
|= action
;
1716 * ata_eh_analyze_ncq_error - analyze NCQ error
1717 * @link: ATA link to analyze NCQ error for
1719 * Read log page 10h, determine the offending qc and acquire
1720 * error status TF. For NCQ device errors, all LLDDs have to do
1721 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1725 * Kernel thread context (may sleep).
1727 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1729 struct ata_port
*ap
= link
->ap
;
1730 struct ata_eh_context
*ehc
= &link
->eh_context
;
1731 struct ata_device
*dev
= link
->device
;
1732 struct ata_queued_cmd
*qc
;
1733 struct ata_taskfile tf
;
1736 /* if frozen, we can't do much */
1737 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1740 /* is it NCQ device error? */
1741 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1744 /* has LLDD analyzed already? */
1745 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1746 qc
= __ata_qc_from_tag(ap
, tag
);
1748 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1755 /* okay, this error is ours */
1756 memset(&tf
, 0, sizeof(tf
));
1757 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1759 ata_link_err(link
, "failed to read log page 10h (errno=%d)\n",
1764 if (!(link
->sactive
& (1 << tag
))) {
1765 ata_link_err(link
, "log page 10h reported inactive tag %d\n",
1770 /* we've got the perpetrator, condemn it */
1771 qc
= __ata_qc_from_tag(ap
, tag
);
1772 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1773 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1774 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1775 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1779 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1780 * @qc: qc to analyze
1781 * @tf: Taskfile registers to analyze
1783 * Analyze taskfile of @qc and further determine cause of
1784 * failure. This function also requests ATAPI sense data if
1788 * Kernel thread context (may sleep).
1791 * Determined recovery action
1793 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1794 const struct ata_taskfile
*tf
)
1796 unsigned int tmp
, action
= 0;
1797 u8 stat
= tf
->command
, err
= tf
->feature
;
1799 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1800 qc
->err_mask
|= AC_ERR_HSM
;
1801 return ATA_EH_RESET
;
1804 if (stat
& (ATA_ERR
| ATA_DF
))
1805 qc
->err_mask
|= AC_ERR_DEV
;
1809 switch (qc
->dev
->class) {
1812 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1814 qc
->err_mask
|= AC_ERR_MEDIA
;
1816 qc
->err_mask
|= AC_ERR_INVALID
;
1820 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1821 tmp
= atapi_eh_request_sense(qc
->dev
,
1822 qc
->scsicmd
->sense_buffer
,
1823 qc
->result_tf
.feature
>> 4);
1825 /* ATA_QCFLAG_SENSE_VALID is used to
1826 * tell atapi_qc_complete() that sense
1827 * data is already valid.
1829 * TODO: interpret sense data and set
1830 * appropriate err_mask.
1832 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1834 qc
->err_mask
|= tmp
;
1838 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1839 action
|= ATA_EH_RESET
;
1844 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1849 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1853 base
= ATA_ECAT_DUBIOUS_NONE
;
1855 if (err_mask
& AC_ERR_ATA_BUS
)
1856 return base
+ ATA_ECAT_ATA_BUS
;
1858 if (err_mask
& AC_ERR_TIMEOUT
)
1859 return base
+ ATA_ECAT_TOUT_HSM
;
1861 if (eflags
& ATA_EFLAG_IS_IO
) {
1862 if (err_mask
& AC_ERR_HSM
)
1863 return base
+ ATA_ECAT_TOUT_HSM
;
1865 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1866 return base
+ ATA_ECAT_UNK_DEV
;
1872 struct speed_down_verdict_arg
{
1875 int nr_errors
[ATA_ECAT_NR
];
1878 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1880 struct speed_down_verdict_arg
*arg
= void_arg
;
1883 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) || (ent
->timestamp
< arg
->since
))
1886 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1888 arg
->nr_errors
[cat
]++;
1894 * ata_eh_speed_down_verdict - Determine speed down verdict
1895 * @dev: Device of interest
1897 * This function examines error ring of @dev and determines
1898 * whether NCQ needs to be turned off, transfer speed should be
1899 * stepped down, or falling back to PIO is necessary.
1901 * ECAT_ATA_BUS : ATA_BUS error for any command
1903 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1906 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1908 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1909 * data transfer hasn't been verified.
1913 * NCQ_OFF : Turn off NCQ.
1915 * SPEED_DOWN : Speed down transfer speed but don't fall back
1918 * FALLBACK_TO_PIO : Fall back to PIO.
1920 * Even if multiple verdicts are returned, only one action is
1921 * taken per error. An action triggered by non-DUBIOUS errors
1922 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1923 * This is to expedite speed down decisions right after device is
1924 * initially configured.
1926 * The followings are speed down rules. #1 and #2 deal with
1929 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1930 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1932 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1933 * occurred during last 5 mins, NCQ_OFF.
1935 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1936 * occurred during last 5 mins, FALLBACK_TO_PIO
1938 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1939 * during last 10 mins, NCQ_OFF.
1941 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1942 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1945 * Inherited from caller.
1948 * OR of ATA_EH_SPDN_* flags.
1950 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1952 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1953 u64 j64
= get_jiffies_64();
1954 struct speed_down_verdict_arg arg
;
1955 unsigned int verdict
= 0;
1957 /* scan past 5 mins of error history */
1958 memset(&arg
, 0, sizeof(arg
));
1959 arg
.since
= j64
- min(j64
, j5mins
);
1960 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1962 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1963 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1964 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1965 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1967 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1968 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1969 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1971 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1972 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1973 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1974 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1976 /* scan past 10 mins of error history */
1977 memset(&arg
, 0, sizeof(arg
));
1978 arg
.since
= j64
- min(j64
, j10mins
);
1979 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1981 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1982 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1983 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1985 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1986 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1987 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1988 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1994 * ata_eh_speed_down - record error and speed down if necessary
1995 * @dev: Failed device
1996 * @eflags: mask of ATA_EFLAG_* flags
1997 * @err_mask: err_mask of the error
1999 * Record error and examine error history to determine whether
2000 * adjusting transmission speed is necessary. It also sets
2001 * transmission limits appropriately if such adjustment is
2005 * Kernel thread context (may sleep).
2008 * Determined recovery action.
2010 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
2011 unsigned int eflags
, unsigned int err_mask
)
2013 struct ata_link
*link
= ata_dev_phys_link(dev
);
2015 unsigned int verdict
;
2016 unsigned int action
= 0;
2018 /* don't bother if Cat-0 error */
2019 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
2022 /* record error and determine whether speed down is necessary */
2023 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
2024 verdict
= ata_eh_speed_down_verdict(dev
);
2027 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
2028 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
2029 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
2030 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
2031 ata_dev_warn(dev
, "NCQ disabled due to excessive errors\n");
2036 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
2037 /* speed down SATA link speed if possible */
2038 if (sata_down_spd_limit(link
, 0) == 0) {
2039 action
|= ATA_EH_RESET
;
2043 /* lower transfer mode */
2044 if (dev
->spdn_cnt
< 2) {
2045 static const int dma_dnxfer_sel
[] =
2046 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
2047 static const int pio_dnxfer_sel
[] =
2048 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
2051 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
2052 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
2054 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
2058 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
2059 action
|= ATA_EH_RESET
;
2065 /* Fall back to PIO? Slowing down to PIO is meaningless for
2066 * SATA ATA devices. Consider it only for PATA and SATAPI.
2068 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
2069 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
2070 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
2071 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
2073 action
|= ATA_EH_RESET
;
2080 /* device has been slowed down, blow error history */
2081 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
2082 ata_ering_clear(&dev
->ering
);
2087 * ata_eh_worth_retry - analyze error and decide whether to retry
2088 * @qc: qc to possibly retry
2090 * Look at the cause of the error and decide if a retry
2091 * might be useful or not. We don't want to retry media errors
2092 * because the drive itself has probably already taken 10-30 seconds
2093 * doing its own internal retries before reporting the failure.
2095 static inline int ata_eh_worth_retry(struct ata_queued_cmd
*qc
)
2097 if (qc
->err_mask
& AC_ERR_MEDIA
)
2098 return 0; /* don't retry media errors */
2099 if (qc
->flags
& ATA_QCFLAG_IO
)
2100 return 1; /* otherwise retry anything from fs stack */
2101 if (qc
->err_mask
& AC_ERR_INVALID
)
2102 return 0; /* don't retry these */
2103 return qc
->err_mask
!= AC_ERR_DEV
; /* retry if not dev error */
2107 * ata_eh_link_autopsy - analyze error and determine recovery action
2108 * @link: host link to perform autopsy on
2110 * Analyze why @link failed and determine which recovery actions
2111 * are needed. This function also sets more detailed AC_ERR_*
2112 * values and fills sense data for ATAPI CHECK SENSE.
2115 * Kernel thread context (may sleep).
2117 static void ata_eh_link_autopsy(struct ata_link
*link
)
2119 struct ata_port
*ap
= link
->ap
;
2120 struct ata_eh_context
*ehc
= &link
->eh_context
;
2121 struct ata_device
*dev
;
2122 unsigned int all_err_mask
= 0, eflags
= 0;
2129 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
2132 /* obtain and analyze SError */
2133 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
2135 ehc
->i
.serror
|= serror
;
2136 ata_eh_analyze_serror(link
);
2137 } else if (rc
!= -EOPNOTSUPP
) {
2138 /* SError read failed, force reset and probing */
2139 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
2140 ehc
->i
.action
|= ATA_EH_RESET
;
2141 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
2144 /* analyze NCQ failure */
2145 ata_eh_analyze_ncq_error(link
);
2147 /* any real error trumps AC_ERR_OTHER */
2148 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
2149 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
2151 all_err_mask
|= ehc
->i
.err_mask
;
2153 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2154 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2156 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2157 ata_dev_phys_link(qc
->dev
) != link
)
2160 /* inherit upper level err_mask */
2161 qc
->err_mask
|= ehc
->i
.err_mask
;
2164 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
2166 /* DEV errors are probably spurious in case of ATA_BUS error */
2167 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
2168 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
2171 /* any real error trumps unknown error */
2172 if (qc
->err_mask
& ~AC_ERR_OTHER
)
2173 qc
->err_mask
&= ~AC_ERR_OTHER
;
2175 /* SENSE_VALID trumps dev/unknown error and revalidation */
2176 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
2177 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
2179 /* determine whether the command is worth retrying */
2180 if (ata_eh_worth_retry(qc
))
2181 qc
->flags
|= ATA_QCFLAG_RETRY
;
2183 /* accumulate error info */
2184 ehc
->i
.dev
= qc
->dev
;
2185 all_err_mask
|= qc
->err_mask
;
2186 if (qc
->flags
& ATA_QCFLAG_IO
)
2187 eflags
|= ATA_EFLAG_IS_IO
;
2190 /* enforce default EH actions */
2191 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
2192 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
2193 ehc
->i
.action
|= ATA_EH_RESET
;
2194 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
2195 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
2196 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2198 /* If we have offending qcs and the associated failed device,
2199 * perform per-dev EH action only on the offending device.
2202 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
2203 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
2204 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2207 /* propagate timeout to host link */
2208 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2209 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2211 /* record error and consider speeding down */
2213 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2214 ata_dev_enabled(link
->device
))))
2218 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2219 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2220 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2227 * ata_eh_autopsy - analyze error and determine recovery action
2228 * @ap: host port to perform autopsy on
2230 * Analyze all links of @ap and determine why they failed and
2231 * which recovery actions are needed.
2234 * Kernel thread context (may sleep).
2236 void ata_eh_autopsy(struct ata_port
*ap
)
2238 struct ata_link
*link
;
2240 ata_for_each_link(link
, ap
, EDGE
)
2241 ata_eh_link_autopsy(link
);
2243 /* Handle the frigging slave link. Autopsy is done similarly
2244 * but actions and flags are transferred over to the master
2245 * link and handled from there.
2247 if (ap
->slave_link
) {
2248 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2249 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2251 /* transfer control flags from master to slave */
2252 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2254 /* perform autopsy on the slave link */
2255 ata_eh_link_autopsy(ap
->slave_link
);
2257 /* transfer actions from slave to master and clear slave */
2258 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2259 mehc
->i
.action
|= sehc
->i
.action
;
2260 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2261 mehc
->i
.flags
|= sehc
->i
.flags
;
2262 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2265 /* Autopsy of fanout ports can affect host link autopsy.
2266 * Perform host link autopsy last.
2268 if (sata_pmp_attached(ap
))
2269 ata_eh_link_autopsy(&ap
->link
);
2273 * ata_get_cmd_descript - get description for ATA command
2274 * @command: ATA command code to get description for
2276 * Return a textual description of the given command, or NULL if the
2277 * command is not known.
2282 const char *ata_get_cmd_descript(u8 command
)
2284 #ifdef CONFIG_ATA_VERBOSE_ERROR
2290 { ATA_CMD_DEV_RESET
, "DEVICE RESET" },
2291 { ATA_CMD_CHK_POWER
, "CHECK POWER MODE" },
2292 { ATA_CMD_STANDBY
, "STANDBY" },
2293 { ATA_CMD_IDLE
, "IDLE" },
2294 { ATA_CMD_EDD
, "EXECUTE DEVICE DIAGNOSTIC" },
2295 { ATA_CMD_DOWNLOAD_MICRO
, "DOWNLOAD MICROCODE" },
2296 { ATA_CMD_NOP
, "NOP" },
2297 { ATA_CMD_FLUSH
, "FLUSH CACHE" },
2298 { ATA_CMD_FLUSH_EXT
, "FLUSH CACHE EXT" },
2299 { ATA_CMD_ID_ATA
, "IDENTIFY DEVICE" },
2300 { ATA_CMD_ID_ATAPI
, "IDENTIFY PACKET DEVICE" },
2301 { ATA_CMD_SERVICE
, "SERVICE" },
2302 { ATA_CMD_READ
, "READ DMA" },
2303 { ATA_CMD_READ_EXT
, "READ DMA EXT" },
2304 { ATA_CMD_READ_QUEUED
, "READ DMA QUEUED" },
2305 { ATA_CMD_READ_STREAM_EXT
, "READ STREAM EXT" },
2306 { ATA_CMD_READ_STREAM_DMA_EXT
, "READ STREAM DMA EXT" },
2307 { ATA_CMD_WRITE
, "WRITE DMA" },
2308 { ATA_CMD_WRITE_EXT
, "WRITE DMA EXT" },
2309 { ATA_CMD_WRITE_QUEUED
, "WRITE DMA QUEUED EXT" },
2310 { ATA_CMD_WRITE_STREAM_EXT
, "WRITE STREAM EXT" },
2311 { ATA_CMD_WRITE_STREAM_DMA_EXT
, "WRITE STREAM DMA EXT" },
2312 { ATA_CMD_WRITE_FUA_EXT
, "WRITE DMA FUA EXT" },
2313 { ATA_CMD_WRITE_QUEUED_FUA_EXT
, "WRITE DMA QUEUED FUA EXT" },
2314 { ATA_CMD_FPDMA_READ
, "READ FPDMA QUEUED" },
2315 { ATA_CMD_FPDMA_WRITE
, "WRITE FPDMA QUEUED" },
2316 { ATA_CMD_PIO_READ
, "READ SECTOR(S)" },
2317 { ATA_CMD_PIO_READ_EXT
, "READ SECTOR(S) EXT" },
2318 { ATA_CMD_PIO_WRITE
, "WRITE SECTOR(S)" },
2319 { ATA_CMD_PIO_WRITE_EXT
, "WRITE SECTOR(S) EXT" },
2320 { ATA_CMD_READ_MULTI
, "READ MULTIPLE" },
2321 { ATA_CMD_READ_MULTI_EXT
, "READ MULTIPLE EXT" },
2322 { ATA_CMD_WRITE_MULTI
, "WRITE MULTIPLE" },
2323 { ATA_CMD_WRITE_MULTI_EXT
, "WRITE MULTIPLE EXT" },
2324 { ATA_CMD_WRITE_MULTI_FUA_EXT
, "WRITE MULTIPLE FUA EXT" },
2325 { ATA_CMD_SET_FEATURES
, "SET FEATURES" },
2326 { ATA_CMD_SET_MULTI
, "SET MULTIPLE MODE" },
2327 { ATA_CMD_VERIFY
, "READ VERIFY SECTOR(S)" },
2328 { ATA_CMD_VERIFY_EXT
, "READ VERIFY SECTOR(S) EXT" },
2329 { ATA_CMD_WRITE_UNCORR_EXT
, "WRITE UNCORRECTABLE EXT" },
2330 { ATA_CMD_STANDBYNOW1
, "STANDBY IMMEDIATE" },
2331 { ATA_CMD_IDLEIMMEDIATE
, "IDLE IMMEDIATE" },
2332 { ATA_CMD_SLEEP
, "SLEEP" },
2333 { ATA_CMD_INIT_DEV_PARAMS
, "INITIALIZE DEVICE PARAMETERS" },
2334 { ATA_CMD_READ_NATIVE_MAX
, "READ NATIVE MAX ADDRESS" },
2335 { ATA_CMD_READ_NATIVE_MAX_EXT
, "READ NATIVE MAX ADDRESS EXT" },
2336 { ATA_CMD_SET_MAX
, "SET MAX ADDRESS" },
2337 { ATA_CMD_SET_MAX_EXT
, "SET MAX ADDRESS EXT" },
2338 { ATA_CMD_READ_LOG_EXT
, "READ LOG EXT" },
2339 { ATA_CMD_WRITE_LOG_EXT
, "WRITE LOG EXT" },
2340 { ATA_CMD_READ_LOG_DMA_EXT
, "READ LOG DMA EXT" },
2341 { ATA_CMD_WRITE_LOG_DMA_EXT
, "WRITE LOG DMA EXT" },
2342 { ATA_CMD_TRUSTED_RCV
, "TRUSTED RECEIVE" },
2343 { ATA_CMD_TRUSTED_RCV_DMA
, "TRUSTED RECEIVE DMA" },
2344 { ATA_CMD_TRUSTED_SND
, "TRUSTED SEND" },
2345 { ATA_CMD_TRUSTED_SND_DMA
, "TRUSTED SEND DMA" },
2346 { ATA_CMD_PMP_READ
, "READ BUFFER" },
2347 { ATA_CMD_PMP_WRITE
, "WRITE BUFFER" },
2348 { ATA_CMD_CONF_OVERLAY
, "DEVICE CONFIGURATION OVERLAY" },
2349 { ATA_CMD_SEC_SET_PASS
, "SECURITY SET PASSWORD" },
2350 { ATA_CMD_SEC_UNLOCK
, "SECURITY UNLOCK" },
2351 { ATA_CMD_SEC_ERASE_PREP
, "SECURITY ERASE PREPARE" },
2352 { ATA_CMD_SEC_ERASE_UNIT
, "SECURITY ERASE UNIT" },
2353 { ATA_CMD_SEC_FREEZE_LOCK
, "SECURITY FREEZE LOCK" },
2354 { ATA_CMD_SEC_DISABLE_PASS
, "SECURITY DISABLE PASSWORD" },
2355 { ATA_CMD_CONFIG_STREAM
, "CONFIGURE STREAM" },
2356 { ATA_CMD_SMART
, "SMART" },
2357 { ATA_CMD_MEDIA_LOCK
, "DOOR LOCK" },
2358 { ATA_CMD_MEDIA_UNLOCK
, "DOOR UNLOCK" },
2359 { ATA_CMD_DSM
, "DATA SET MANAGEMENT" },
2360 { ATA_CMD_CHK_MED_CRD_TYP
, "CHECK MEDIA CARD TYPE" },
2361 { ATA_CMD_CFA_REQ_EXT_ERR
, "CFA REQUEST EXTENDED ERROR" },
2362 { ATA_CMD_CFA_WRITE_NE
, "CFA WRITE SECTORS WITHOUT ERASE" },
2363 { ATA_CMD_CFA_TRANS_SECT
, "CFA TRANSLATE SECTOR" },
2364 { ATA_CMD_CFA_ERASE
, "CFA ERASE SECTORS" },
2365 { ATA_CMD_CFA_WRITE_MULT_NE
, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2366 { ATA_CMD_READ_LONG
, "READ LONG (with retries)" },
2367 { ATA_CMD_READ_LONG_ONCE
, "READ LONG (without retries)" },
2368 { ATA_CMD_WRITE_LONG
, "WRITE LONG (with retries)" },
2369 { ATA_CMD_WRITE_LONG_ONCE
, "WRITE LONG (without retries)" },
2370 { ATA_CMD_RESTORE
, "RECALIBRATE" },
2371 { 0, NULL
} /* terminate list */
2375 for (i
= 0; cmd_descr
[i
].text
; i
++)
2376 if (cmd_descr
[i
].command
== command
)
2377 return cmd_descr
[i
].text
;
2384 * ata_eh_link_report - report error handling to user
2385 * @link: ATA link EH is going on
2387 * Report EH to user.
2392 static void ata_eh_link_report(struct ata_link
*link
)
2394 struct ata_port
*ap
= link
->ap
;
2395 struct ata_eh_context
*ehc
= &link
->eh_context
;
2396 const char *frozen
, *desc
;
2398 int tag
, nr_failed
= 0;
2400 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2404 if (ehc
->i
.desc
[0] != '\0')
2407 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2408 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2410 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2411 ata_dev_phys_link(qc
->dev
) != link
||
2412 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2413 qc
->err_mask
== AC_ERR_DEV
))
2415 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2421 if (!nr_failed
&& !ehc
->i
.err_mask
)
2425 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2428 memset(tries_buf
, 0, sizeof(tries_buf
));
2429 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2430 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
2434 ata_dev_err(ehc
->i
.dev
, "exception Emask 0x%x "
2435 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2436 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2437 ehc
->i
.action
, frozen
, tries_buf
);
2439 ata_dev_err(ehc
->i
.dev
, "%s\n", desc
);
2441 ata_link_err(link
, "exception Emask 0x%x "
2442 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2443 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2444 ehc
->i
.action
, frozen
, tries_buf
);
2446 ata_link_err(link
, "%s\n", desc
);
2449 #ifdef CONFIG_ATA_VERBOSE_ERROR
2452 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2453 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2454 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2455 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2456 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2457 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2458 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2459 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2460 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2461 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2462 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2463 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2464 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2465 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2466 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2467 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2468 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2469 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2472 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2473 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2474 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2475 const u8
*cdb
= qc
->cdb
;
2476 char data_buf
[20] = "";
2477 char cdb_buf
[70] = "";
2479 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2480 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2483 if (qc
->dma_dir
!= DMA_NONE
) {
2484 static const char *dma_str
[] = {
2485 [DMA_BIDIRECTIONAL
] = "bidi",
2486 [DMA_TO_DEVICE
] = "out",
2487 [DMA_FROM_DEVICE
] = "in",
2489 static const char *prot_str
[] = {
2490 [ATA_PROT_PIO
] = "pio",
2491 [ATA_PROT_DMA
] = "dma",
2492 [ATA_PROT_NCQ
] = "ncq",
2493 [ATAPI_PROT_PIO
] = "pio",
2494 [ATAPI_PROT_DMA
] = "dma",
2497 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2498 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2499 dma_str
[qc
->dma_dir
]);
2502 if (ata_is_atapi(qc
->tf
.protocol
)) {
2504 scsi_print_command(qc
->scsicmd
);
2506 snprintf(cdb_buf
, sizeof(cdb_buf
),
2507 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2508 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2509 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2510 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2511 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2512 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2514 const char *descr
= ata_get_cmd_descript(cmd
->command
);
2516 ata_dev_err(qc
->dev
, "failed command: %s\n",
2520 ata_dev_err(qc
->dev
,
2521 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2523 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2524 "Emask 0x%x (%s)%s\n",
2525 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2526 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2527 cmd
->hob_feature
, cmd
->hob_nsect
,
2528 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2529 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2530 res
->command
, res
->feature
, res
->nsect
,
2531 res
->lbal
, res
->lbam
, res
->lbah
,
2532 res
->hob_feature
, res
->hob_nsect
,
2533 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2534 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2535 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2537 #ifdef CONFIG_ATA_VERBOSE_ERROR
2538 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2540 if (res
->command
& ATA_BUSY
)
2541 ata_dev_err(qc
->dev
, "status: { Busy }\n");
2543 ata_dev_err(qc
->dev
, "status: { %s%s%s%s}\n",
2544 res
->command
& ATA_DRDY
? "DRDY " : "",
2545 res
->command
& ATA_DF
? "DF " : "",
2546 res
->command
& ATA_DRQ
? "DRQ " : "",
2547 res
->command
& ATA_ERR
? "ERR " : "");
2550 if (cmd
->command
!= ATA_CMD_PACKET
&&
2551 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2553 ata_dev_err(qc
->dev
, "error: { %s%s%s%s}\n",
2554 res
->feature
& ATA_ICRC
? "ICRC " : "",
2555 res
->feature
& ATA_UNC
? "UNC " : "",
2556 res
->feature
& ATA_IDNF
? "IDNF " : "",
2557 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2563 * ata_eh_report - report error handling to user
2564 * @ap: ATA port to report EH about
2566 * Report EH to user.
2571 void ata_eh_report(struct ata_port
*ap
)
2573 struct ata_link
*link
;
2575 ata_for_each_link(link
, ap
, HOST_FIRST
)
2576 ata_eh_link_report(link
);
2579 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2580 unsigned int *classes
, unsigned long deadline
,
2583 struct ata_device
*dev
;
2586 ata_for_each_dev(dev
, link
, ALL
)
2587 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2589 return reset(link
, classes
, deadline
);
2592 static int ata_eh_followup_srst_needed(struct ata_link
*link
, int rc
)
2594 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2598 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2603 int ata_eh_reset(struct ata_link
*link
, int classify
,
2604 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2605 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2607 struct ata_port
*ap
= link
->ap
;
2608 struct ata_link
*slave
= ap
->slave_link
;
2609 struct ata_eh_context
*ehc
= &link
->eh_context
;
2610 struct ata_eh_context
*sehc
= slave
? &slave
->eh_context
: NULL
;
2611 unsigned int *classes
= ehc
->classes
;
2612 unsigned int lflags
= link
->flags
;
2613 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2614 int max_tries
= 0, try = 0;
2615 struct ata_link
*failed_link
;
2616 struct ata_device
*dev
;
2617 unsigned long deadline
, now
;
2618 ata_reset_fn_t reset
;
2619 unsigned long flags
;
2626 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2628 if (link
->flags
& ATA_LFLAG_RST_ONCE
)
2630 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2632 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2635 /* make sure each reset attempt is at least COOL_DOWN apart */
2636 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2638 WARN_ON(time_after(ehc
->last_reset
, now
));
2639 deadline
= ata_deadline(ehc
->last_reset
,
2640 ATA_EH_RESET_COOL_DOWN
);
2641 if (time_before(now
, deadline
))
2642 schedule_timeout_uninterruptible(deadline
- now
);
2645 spin_lock_irqsave(ap
->lock
, flags
);
2646 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2647 spin_unlock_irqrestore(ap
->lock
, flags
);
2649 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2651 ata_for_each_dev(dev
, link
, ALL
) {
2652 /* If we issue an SRST then an ATA drive (not ATAPI)
2653 * may change configuration and be in PIO0 timing. If
2654 * we do a hard reset (or are coming from power on)
2655 * this is true for ATA or ATAPI. Until we've set a
2656 * suitable controller mode we should not touch the
2657 * bus as we may be talking too fast.
2659 dev
->pio_mode
= XFER_PIO_0
;
2660 dev
->dma_mode
= 0xff;
2662 /* If the controller has a pio mode setup function
2663 * then use it to set the chipset to rights. Don't
2664 * touch the DMA setup as that will be dealt with when
2665 * configuring devices.
2667 if (ap
->ops
->set_piomode
)
2668 ap
->ops
->set_piomode(ap
, dev
);
2671 /* prefer hardreset */
2673 ehc
->i
.action
&= ~ATA_EH_RESET
;
2676 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2677 } else if (softreset
) {
2679 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2683 unsigned long deadline
= ata_deadline(jiffies
,
2684 ATA_EH_PRERESET_TIMEOUT
);
2687 sehc
->i
.action
&= ~ATA_EH_RESET
;
2688 sehc
->i
.action
|= ehc
->i
.action
;
2691 rc
= prereset(link
, deadline
);
2693 /* If present, do prereset on slave link too. Reset
2694 * is skipped iff both master and slave links report
2695 * -ENOENT or clear ATA_EH_RESET.
2697 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2700 tmp
= prereset(slave
, deadline
);
2704 ehc
->i
.action
|= sehc
->i
.action
;
2708 if (rc
== -ENOENT
) {
2709 ata_link_dbg(link
, "port disabled--ignoring\n");
2710 ehc
->i
.action
&= ~ATA_EH_RESET
;
2712 ata_for_each_dev(dev
, link
, ALL
)
2713 classes
[dev
->devno
] = ATA_DEV_NONE
;
2718 "prereset failed (errno=%d)\n",
2723 /* prereset() might have cleared ATA_EH_RESET. If so,
2724 * bang classes, thaw and return.
2726 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2727 ata_for_each_dev(dev
, link
, ALL
)
2728 classes
[dev
->devno
] = ATA_DEV_NONE
;
2729 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) &&
2730 ata_is_host_link(link
))
2731 ata_eh_thaw_port(ap
);
2741 if (ata_is_host_link(link
))
2742 ata_eh_freeze_port(ap
);
2744 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2748 ata_link_info(link
, "%s resetting link\n",
2749 reset
== softreset
? "soft" : "hard");
2751 /* mark that this EH session started with reset */
2752 ehc
->last_reset
= jiffies
;
2753 if (reset
== hardreset
)
2754 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2756 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2758 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2759 if (rc
&& rc
!= -EAGAIN
) {
2764 /* hardreset slave link if existent */
2765 if (slave
&& reset
== hardreset
) {
2769 ata_link_info(slave
, "hard resetting link\n");
2771 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2772 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2780 failed_link
= slave
;
2786 /* perform follow-up SRST if necessary */
2787 if (reset
== hardreset
&&
2788 ata_eh_followup_srst_needed(link
, rc
)) {
2793 "follow-up softreset required but no softreset available\n");
2799 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2800 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2809 "no reset method available, skipping reset\n");
2810 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2811 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2815 * Post-reset processing
2817 ata_for_each_dev(dev
, link
, ALL
) {
2818 /* After the reset, the device state is PIO 0 and the
2819 * controller state is undefined. Reset also wakes up
2820 * drives from sleeping mode.
2822 dev
->pio_mode
= XFER_PIO_0
;
2823 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2825 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2828 /* apply class override */
2829 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2830 classes
[dev
->devno
] = ATA_DEV_ATA
;
2831 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2832 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2835 /* record current link speed */
2836 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2837 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2838 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2839 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2842 if (ata_is_host_link(link
))
2843 ata_eh_thaw_port(ap
);
2845 /* postreset() should clear hardware SError. Although SError
2846 * is cleared during link resume, clearing SError here is
2847 * necessary as some PHYs raise hotplug events after SRST.
2848 * This introduces race condition where hotplug occurs between
2849 * reset and here. This race is mediated by cross checking
2850 * link onlineness and classification result later.
2853 postreset(link
, classes
);
2855 postreset(slave
, classes
);
2859 * Some controllers can't be frozen very well and may set spurious
2860 * error conditions during reset. Clear accumulated error
2861 * information and re-thaw the port if frozen. As reset is the
2862 * final recovery action and we cross check link onlineness against
2863 * device classification later, no hotplug event is lost by this.
2865 spin_lock_irqsave(link
->ap
->lock
, flags
);
2866 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
2868 memset(&slave
->eh_info
, 0, sizeof(link
->eh_info
));
2869 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
2870 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2872 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2873 ata_eh_thaw_port(ap
);
2876 * Make sure onlineness and classification result correspond.
2877 * Hotplug could have happened during reset and some
2878 * controllers fail to wait while a drive is spinning up after
2879 * being hotplugged causing misdetection. By cross checking
2880 * link on/offlineness and classification result, those
2881 * conditions can be reliably detected and retried.
2884 ata_for_each_dev(dev
, link
, ALL
) {
2885 if (ata_phys_link_online(ata_dev_phys_link(dev
))) {
2886 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2887 ata_dev_dbg(dev
, "link online but device misclassified\n");
2888 classes
[dev
->devno
] = ATA_DEV_NONE
;
2891 } else if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2892 if (ata_class_enabled(classes
[dev
->devno
]))
2894 "link offline, clearing class %d to NONE\n",
2895 classes
[dev
->devno
]);
2896 classes
[dev
->devno
] = ATA_DEV_NONE
;
2897 } else if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2899 "link status unknown, clearing UNKNOWN to NONE\n");
2900 classes
[dev
->devno
] = ATA_DEV_NONE
;
2904 if (classify
&& nr_unknown
) {
2905 if (try < max_tries
) {
2907 "link online but %d devices misclassified, retrying\n",
2914 "link online but %d devices misclassified, "
2915 "device detection might fail\n", nr_unknown
);
2918 /* reset successful, schedule revalidation */
2919 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2921 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2922 ehc
->last_reset
= jiffies
; /* update to completion time */
2923 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2924 link
->lpm_policy
= ATA_LPM_UNKNOWN
; /* reset LPM state */
2928 /* clear hotplug flag */
2929 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2931 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2933 spin_lock_irqsave(ap
->lock
, flags
);
2934 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2935 spin_unlock_irqrestore(ap
->lock
, flags
);
2940 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2941 if (!ata_is_host_link(link
) &&
2942 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2945 if (try >= max_tries
) {
2947 * Thaw host port even if reset failed, so that the port
2948 * can be retried on the next phy event. This risks
2949 * repeated EH runs but seems to be a better tradeoff than
2950 * shutting down a port after a botched hotplug attempt.
2952 if (ata_is_host_link(link
))
2953 ata_eh_thaw_port(ap
);
2958 if (time_before(now
, deadline
)) {
2959 unsigned long delta
= deadline
- now
;
2961 ata_link_warn(failed_link
,
2962 "reset failed (errno=%d), retrying in %u secs\n",
2963 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2967 delta
= schedule_timeout_uninterruptible(delta
);
2972 * While disks spinup behind PMP, some controllers fail sending SRST.
2973 * They need to be reset - as well as the PMP - before retrying.
2975 if (rc
== -ERESTART
) {
2976 if (ata_is_host_link(link
))
2977 ata_eh_thaw_port(ap
);
2981 if (try == max_tries
- 1) {
2982 sata_down_spd_limit(link
, 0);
2984 sata_down_spd_limit(slave
, 0);
2985 } else if (rc
== -EPIPE
)
2986 sata_down_spd_limit(failed_link
, 0);
2993 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
2995 struct ata_link
*link
;
2996 struct ata_device
*dev
;
2997 unsigned long flags
;
3000 * This function can be thought of as an extended version of
3001 * ata_eh_about_to_do() specially crafted to accommodate the
3002 * requirements of ATA_EH_PARK handling. Since the EH thread
3003 * does not leave the do {} while () loop in ata_eh_recover as
3004 * long as the timeout for a park request to *one* device on
3005 * the port has not expired, and since we still want to pick
3006 * up park requests to other devices on the same port or
3007 * timeout updates for the same device, we have to pull
3008 * ATA_EH_PARK actions from eh_info into eh_context.i
3009 * ourselves at the beginning of each pass over the loop.
3011 * Additionally, all write accesses to &ap->park_req_pending
3012 * through INIT_COMPLETION() (see below) or complete_all()
3013 * (see ata_scsi_park_store()) are protected by the host lock.
3014 * As a result we have that park_req_pending.done is zero on
3015 * exit from this function, i.e. when ATA_EH_PARK actions for
3016 * *all* devices on port ap have been pulled into the
3017 * respective eh_context structs. If, and only if,
3018 * park_req_pending.done is non-zero by the time we reach
3019 * wait_for_completion_timeout(), another ATA_EH_PARK action
3020 * has been scheduled for at least one of the devices on port
3021 * ap and we have to cycle over the do {} while () loop in
3022 * ata_eh_recover() again.
3025 spin_lock_irqsave(ap
->lock
, flags
);
3026 INIT_COMPLETION(ap
->park_req_pending
);
3027 ata_for_each_link(link
, ap
, EDGE
) {
3028 ata_for_each_dev(dev
, link
, ALL
) {
3029 struct ata_eh_info
*ehi
= &link
->eh_info
;
3031 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
3032 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
3033 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
3036 spin_unlock_irqrestore(ap
->lock
, flags
);
3039 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
3041 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3042 struct ata_taskfile tf
;
3043 unsigned int err_mask
;
3045 ata_tf_init(dev
, &tf
);
3047 ehc
->unloaded_mask
|= 1 << dev
->devno
;
3048 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
3054 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3055 tf
.command
= ATA_CMD_CHK_POWER
;
3058 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
3059 tf
.protocol
|= ATA_PROT_NODATA
;
3060 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3061 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
3062 ata_dev_err(dev
, "head unload failed!\n");
3063 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3067 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
3068 struct ata_device
**r_failed_dev
)
3070 struct ata_port
*ap
= link
->ap
;
3071 struct ata_eh_context
*ehc
= &link
->eh_context
;
3072 struct ata_device
*dev
;
3073 unsigned int new_mask
= 0;
3074 unsigned long flags
;
3079 /* For PATA drive side cable detection to work, IDENTIFY must
3080 * be done backwards such that PDIAG- is released by the slave
3081 * device before the master device is identified.
3083 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
3084 unsigned int action
= ata_eh_dev_action(dev
);
3085 unsigned int readid_flags
= 0;
3087 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
3088 readid_flags
|= ATA_READID_POSTRESET
;
3090 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
3091 WARN_ON(dev
->class == ATA_DEV_PMP
);
3093 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
3098 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
3099 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
3104 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
3106 /* Configuration may have changed, reconfigure
3109 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3111 /* schedule the scsi_rescan_device() here */
3112 schedule_work(&(ap
->scsi_rescan_task
));
3113 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
3114 ehc
->tries
[dev
->devno
] &&
3115 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
3116 /* Temporarily set dev->class, it will be
3117 * permanently set once all configurations are
3118 * complete. This is necessary because new
3119 * device configuration is done in two
3122 dev
->class = ehc
->classes
[dev
->devno
];
3124 if (dev
->class == ATA_DEV_PMP
)
3125 rc
= sata_pmp_attach(dev
);
3127 rc
= ata_dev_read_id(dev
, &dev
->class,
3128 readid_flags
, dev
->id
);
3130 /* read_id might have changed class, store and reset */
3131 ehc
->classes
[dev
->devno
] = dev
->class;
3132 dev
->class = ATA_DEV_UNKNOWN
;
3136 /* clear error info accumulated during probe */
3137 ata_ering_clear(&dev
->ering
);
3138 new_mask
|= 1 << dev
->devno
;
3141 /* IDENTIFY was issued to non-existent
3142 * device. No need to reset. Just
3143 * thaw and ignore the device.
3145 ata_eh_thaw_port(ap
);
3153 /* PDIAG- should have been released, ask cable type if post-reset */
3154 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
3155 if (ap
->ops
->cable_detect
)
3156 ap
->cbl
= ap
->ops
->cable_detect(ap
);
3160 /* Configure new devices forward such that user doesn't see
3161 * device detection messages backwards.
3163 ata_for_each_dev(dev
, link
, ALL
) {
3164 if (!(new_mask
& (1 << dev
->devno
)))
3167 dev
->class = ehc
->classes
[dev
->devno
];
3169 if (dev
->class == ATA_DEV_PMP
)
3172 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
3173 rc
= ata_dev_configure(dev
);
3174 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
3176 dev
->class = ATA_DEV_UNKNOWN
;
3180 spin_lock_irqsave(ap
->lock
, flags
);
3181 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
3182 spin_unlock_irqrestore(ap
->lock
, flags
);
3184 /* new device discovered, configure xfermode */
3185 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3191 *r_failed_dev
= dev
;
3192 DPRINTK("EXIT rc=%d\n", rc
);
3197 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3198 * @link: link on which timings will be programmed
3199 * @r_failed_dev: out parameter for failed device
3201 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3202 * ata_set_mode() fails, pointer to the failing device is
3203 * returned in @r_failed_dev.
3206 * PCI/etc. bus probe sem.
3209 * 0 on success, negative errno otherwise
3211 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
3213 struct ata_port
*ap
= link
->ap
;
3214 struct ata_device
*dev
;
3217 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3218 ata_for_each_dev(dev
, link
, ENABLED
) {
3219 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
3220 struct ata_ering_entry
*ent
;
3222 ent
= ata_ering_top(&dev
->ering
);
3224 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
3228 /* has private set_mode? */
3229 if (ap
->ops
->set_mode
)
3230 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
3232 rc
= ata_do_set_mode(link
, r_failed_dev
);
3234 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3235 ata_for_each_dev(dev
, link
, ENABLED
) {
3236 struct ata_eh_context
*ehc
= &link
->eh_context
;
3237 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
3238 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
3240 if (dev
->xfer_mode
!= saved_xfer_mode
||
3241 ata_ncq_enabled(dev
) != saved_ncq
)
3242 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
3249 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3250 * @dev: ATAPI device to clear UA for
3252 * Resets and other operations can make an ATAPI device raise
3253 * UNIT ATTENTION which causes the next operation to fail. This
3254 * function clears UA.
3257 * EH context (may sleep).
3260 * 0 on success, -errno on failure.
3262 static int atapi_eh_clear_ua(struct ata_device
*dev
)
3266 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
3267 u8
*sense_buffer
= dev
->link
->ap
->sector_buf
;
3269 unsigned int err_mask
;
3271 err_mask
= atapi_eh_tur(dev
, &sense_key
);
3272 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
3274 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3279 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
3282 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
3284 ata_dev_warn(dev
, "failed to clear "
3285 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
3290 ata_dev_warn(dev
, "UNIT ATTENTION persists after %d tries\n",
3297 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3298 * @dev: ATA device which may need FLUSH retry
3300 * If @dev failed FLUSH, it needs to be reported upper layer
3301 * immediately as it means that @dev failed to remap and already
3302 * lost at least a sector and further FLUSH retrials won't make
3303 * any difference to the lost sector. However, if FLUSH failed
3304 * for other reasons, for example transmission error, FLUSH needs
3307 * This function determines whether FLUSH failure retry is
3308 * necessary and performs it if so.
3311 * 0 if EH can continue, -errno if EH needs to be repeated.
3313 static int ata_eh_maybe_retry_flush(struct ata_device
*dev
)
3315 struct ata_link
*link
= dev
->link
;
3316 struct ata_port
*ap
= link
->ap
;
3317 struct ata_queued_cmd
*qc
;
3318 struct ata_taskfile tf
;
3319 unsigned int err_mask
;
3322 /* did flush fail for this device? */
3323 if (!ata_tag_valid(link
->active_tag
))
3326 qc
= __ata_qc_from_tag(ap
, link
->active_tag
);
3327 if (qc
->dev
!= dev
|| (qc
->tf
.command
!= ATA_CMD_FLUSH_EXT
&&
3328 qc
->tf
.command
!= ATA_CMD_FLUSH
))
3331 /* if the device failed it, it should be reported to upper layers */
3332 if (qc
->err_mask
& AC_ERR_DEV
)
3335 /* flush failed for some other reason, give it another shot */
3336 ata_tf_init(dev
, &tf
);
3338 tf
.command
= qc
->tf
.command
;
3339 tf
.flags
|= ATA_TFLAG_DEVICE
;
3340 tf
.protocol
= ATA_PROT_NODATA
;
3342 ata_dev_warn(dev
, "retrying FLUSH 0x%x Emask 0x%x\n",
3343 tf
.command
, qc
->err_mask
);
3345 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3348 * FLUSH is complete but there's no way to
3349 * successfully complete a failed command from EH.
3350 * Making sure retry is allowed at least once and
3351 * retrying it should do the trick - whatever was in
3352 * the cache is already on the platter and this won't
3353 * cause infinite loop.
3355 qc
->scsicmd
->allowed
= max(qc
->scsicmd
->allowed
, 1);
3357 ata_dev_warn(dev
, "FLUSH failed Emask 0x%x\n",
3361 /* if device failed it, report it to upper layers */
3362 if (err_mask
& AC_ERR_DEV
) {
3363 qc
->err_mask
|= AC_ERR_DEV
;
3365 if (!(ap
->pflags
& ATA_PFLAG_FROZEN
))
3373 * ata_eh_set_lpm - configure SATA interface power management
3374 * @link: link to configure power management
3375 * @policy: the link power management policy
3376 * @r_failed_dev: out parameter for failed device
3378 * Enable SATA Interface power management. This will enable
3379 * Device Interface Power Management (DIPM) for min_power
3380 * policy, and then call driver specific callbacks for
3381 * enabling Host Initiated Power management.
3387 * 0 on success, -errno on failure.
3389 static int ata_eh_set_lpm(struct ata_link
*link
, enum ata_lpm_policy policy
,
3390 struct ata_device
**r_failed_dev
)
3392 struct ata_port
*ap
= ata_is_host_link(link
) ? link
->ap
: NULL
;
3393 struct ata_eh_context
*ehc
= &link
->eh_context
;
3394 struct ata_device
*dev
, *link_dev
= NULL
, *lpm_dev
= NULL
;
3395 enum ata_lpm_policy old_policy
= link
->lpm_policy
;
3396 bool no_dipm
= link
->ap
->flags
& ATA_FLAG_NO_DIPM
;
3397 unsigned int hints
= ATA_LPM_EMPTY
| ATA_LPM_HIPM
;
3398 unsigned int err_mask
;
3401 /* if the link or host doesn't do LPM, noop */
3402 if ((link
->flags
& ATA_LFLAG_NO_LPM
) || (ap
&& !ap
->ops
->set_lpm
))
3406 * DIPM is enabled only for MIN_POWER as some devices
3407 * misbehave when the host NACKs transition to SLUMBER. Order
3408 * device and link configurations such that the host always
3409 * allows DIPM requests.
3411 ata_for_each_dev(dev
, link
, ENABLED
) {
3412 bool hipm
= ata_id_has_hipm(dev
->id
);
3413 bool dipm
= ata_id_has_dipm(dev
->id
) && !no_dipm
;
3415 /* find the first enabled and LPM enabled devices */
3419 if (!lpm_dev
&& (hipm
|| dipm
))
3422 hints
&= ~ATA_LPM_EMPTY
;
3424 hints
&= ~ATA_LPM_HIPM
;
3426 /* disable DIPM before changing link config */
3427 if (policy
!= ATA_LPM_MIN_POWER
&& dipm
) {
3428 err_mask
= ata_dev_set_feature(dev
,
3429 SETFEATURES_SATA_DISABLE
, SATA_DIPM
);
3430 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3432 "failed to disable DIPM, Emask 0x%x\n",
3441 rc
= ap
->ops
->set_lpm(link
, policy
, hints
);
3442 if (!rc
&& ap
->slave_link
)
3443 rc
= ap
->ops
->set_lpm(ap
->slave_link
, policy
, hints
);
3445 rc
= sata_pmp_set_lpm(link
, policy
, hints
);
3448 * Attribute link config failure to the first (LPM) enabled
3449 * device on the link.
3452 if (rc
== -EOPNOTSUPP
) {
3453 link
->flags
|= ATA_LFLAG_NO_LPM
;
3456 dev
= lpm_dev
? lpm_dev
: link_dev
;
3461 * Low level driver acked the transition. Issue DIPM command
3462 * with the new policy set.
3464 link
->lpm_policy
= policy
;
3465 if (ap
&& ap
->slave_link
)
3466 ap
->slave_link
->lpm_policy
= policy
;
3468 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3469 ata_for_each_dev(dev
, link
, ENABLED
) {
3470 if (policy
== ATA_LPM_MIN_POWER
&& !no_dipm
&&
3471 ata_id_has_dipm(dev
->id
)) {
3472 err_mask
= ata_dev_set_feature(dev
,
3473 SETFEATURES_SATA_ENABLE
, SATA_DIPM
);
3474 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3476 "failed to enable DIPM, Emask 0x%x\n",
3487 /* restore the old policy */
3488 link
->lpm_policy
= old_policy
;
3489 if (ap
&& ap
->slave_link
)
3490 ap
->slave_link
->lpm_policy
= old_policy
;
3492 /* if no device or only one more chance is left, disable LPM */
3493 if (!dev
|| ehc
->tries
[dev
->devno
] <= 2) {
3494 ata_link_warn(link
, "disabling LPM on the link\n");
3495 link
->flags
|= ATA_LFLAG_NO_LPM
;
3498 *r_failed_dev
= dev
;
3502 int ata_link_nr_enabled(struct ata_link
*link
)
3504 struct ata_device
*dev
;
3507 ata_for_each_dev(dev
, link
, ENABLED
)
3512 static int ata_link_nr_vacant(struct ata_link
*link
)
3514 struct ata_device
*dev
;
3517 ata_for_each_dev(dev
, link
, ALL
)
3518 if (dev
->class == ATA_DEV_UNKNOWN
)
3523 static int ata_eh_skip_recovery(struct ata_link
*link
)
3525 struct ata_port
*ap
= link
->ap
;
3526 struct ata_eh_context
*ehc
= &link
->eh_context
;
3527 struct ata_device
*dev
;
3529 /* skip disabled links */
3530 if (link
->flags
& ATA_LFLAG_DISABLED
)
3533 /* skip if explicitly requested */
3534 if (ehc
->i
.flags
& ATA_EHI_NO_RECOVERY
)
3537 /* thaw frozen port and recover failed devices */
3538 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
3541 /* reset at least once if reset is requested */
3542 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
3543 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
3546 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3547 ata_for_each_dev(dev
, link
, ALL
) {
3548 if (dev
->class == ATA_DEV_UNKNOWN
&&
3549 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
3556 static int ata_count_probe_trials_cb(struct ata_ering_entry
*ent
, void *void_arg
)
3558 u64 interval
= msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL
);
3559 u64 now
= get_jiffies_64();
3560 int *trials
= void_arg
;
3562 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) ||
3563 (ent
->timestamp
< now
- min(now
, interval
)))
3570 static int ata_eh_schedule_probe(struct ata_device
*dev
)
3572 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3573 struct ata_link
*link
= ata_dev_phys_link(dev
);
3576 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
3577 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
3580 ata_eh_detach_dev(dev
);
3582 ehc
->did_probe_mask
|= (1 << dev
->devno
);
3583 ehc
->i
.action
|= ATA_EH_RESET
;
3584 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
3585 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
3587 /* the link maybe in a deep sleep, wake it up */
3588 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
) {
3589 if (ata_is_host_link(link
))
3590 link
->ap
->ops
->set_lpm(link
, ATA_LPM_MAX_POWER
,
3593 sata_pmp_set_lpm(link
, ATA_LPM_MAX_POWER
,
3597 /* Record and count probe trials on the ering. The specific
3598 * error mask used is irrelevant. Because a successful device
3599 * detection clears the ering, this count accumulates only if
3600 * there are consecutive failed probes.
3602 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3603 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3604 * forced to 1.5Gbps.
3606 * This is to work around cases where failed link speed
3607 * negotiation results in device misdetection leading to
3608 * infinite DEVXCHG or PHRDY CHG events.
3610 ata_ering_record(&dev
->ering
, 0, AC_ERR_OTHER
);
3611 ata_ering_map(&dev
->ering
, ata_count_probe_trials_cb
, &trials
);
3613 if (trials
> ATA_EH_PROBE_TRIALS
)
3614 sata_down_spd_limit(link
, 1);
3619 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
3621 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3623 /* -EAGAIN from EH routine indicates retry without prejudice.
3624 * The requester is responsible for ensuring forward progress.
3627 ehc
->tries
[dev
->devno
]--;
3631 /* device missing or wrong IDENTIFY data, schedule probing */
3632 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
3634 /* give it just one more chance */
3635 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
3637 if (ehc
->tries
[dev
->devno
] == 1) {
3638 /* This is the last chance, better to slow
3639 * down than lose it.
3641 sata_down_spd_limit(ata_dev_phys_link(dev
), 0);
3642 if (dev
->pio_mode
> XFER_PIO_0
)
3643 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
3647 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
3648 /* disable device if it has used up all its chances */
3649 ata_dev_disable(dev
);
3651 /* detach if offline */
3652 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
3653 ata_eh_detach_dev(dev
);
3655 /* schedule probe if necessary */
3656 if (ata_eh_schedule_probe(dev
)) {
3657 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3658 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3659 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3664 ehc
->i
.action
|= ATA_EH_RESET
;
3670 * ata_eh_recover - recover host port after error
3671 * @ap: host port to recover
3672 * @prereset: prereset method (can be NULL)
3673 * @softreset: softreset method (can be NULL)
3674 * @hardreset: hardreset method (can be NULL)
3675 * @postreset: postreset method (can be NULL)
3676 * @r_failed_link: out parameter for failed link
3678 * This is the alpha and omega, eum and yang, heart and soul of
3679 * libata exception handling. On entry, actions required to
3680 * recover each link and hotplug requests are recorded in the
3681 * link's eh_context. This function executes all the operations
3682 * with appropriate retrials and fallbacks to resurrect failed
3683 * devices, detach goners and greet newcomers.
3686 * Kernel thread context (may sleep).
3689 * 0 on success, -errno on failure.
3691 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3692 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3693 ata_postreset_fn_t postreset
,
3694 struct ata_link
**r_failed_link
)
3696 struct ata_link
*link
;
3697 struct ata_device
*dev
;
3699 unsigned long flags
, deadline
;
3703 /* prep for recovery */
3704 ata_for_each_link(link
, ap
, EDGE
) {
3705 struct ata_eh_context
*ehc
= &link
->eh_context
;
3707 /* re-enable link? */
3708 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3709 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3710 spin_lock_irqsave(ap
->lock
, flags
);
3711 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3712 spin_unlock_irqrestore(ap
->lock
, flags
);
3713 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3716 ata_for_each_dev(dev
, link
, ALL
) {
3717 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3718 ehc
->tries
[dev
->devno
] = 1;
3720 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3722 /* collect port action mask recorded in dev actions */
3723 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3724 ~ATA_EH_PERDEV_MASK
;
3725 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3727 /* process hotplug request */
3728 if (dev
->flags
& ATA_DFLAG_DETACH
)
3729 ata_eh_detach_dev(dev
);
3731 /* schedule probe if necessary */
3732 if (!ata_dev_enabled(dev
))
3733 ata_eh_schedule_probe(dev
);
3740 /* if UNLOADING, finish immediately */
3741 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3745 ata_for_each_link(link
, ap
, EDGE
) {
3746 struct ata_eh_context
*ehc
= &link
->eh_context
;
3748 /* skip EH if possible. */
3749 if (ata_eh_skip_recovery(link
))
3752 ata_for_each_dev(dev
, link
, ALL
)
3753 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3757 ata_for_each_link(link
, ap
, EDGE
) {
3758 struct ata_eh_context
*ehc
= &link
->eh_context
;
3760 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3763 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3764 prereset
, softreset
, hardreset
, postreset
);
3766 ata_link_err(link
, "reset failed, giving up\n");
3775 * clears ATA_EH_PARK in eh_info and resets
3776 * ap->park_req_pending
3778 ata_eh_pull_park_action(ap
);
3781 ata_for_each_link(link
, ap
, EDGE
) {
3782 ata_for_each_dev(dev
, link
, ALL
) {
3783 struct ata_eh_context
*ehc
= &link
->eh_context
;
3786 if (dev
->class != ATA_DEV_ATA
)
3788 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3791 tmp
= dev
->unpark_deadline
;
3792 if (time_before(deadline
, tmp
))
3794 else if (time_before_eq(tmp
, jiffies
))
3796 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3799 ata_eh_park_issue_cmd(dev
, 1);
3804 if (time_before_eq(deadline
, now
))
3808 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3812 ata_for_each_link(link
, ap
, EDGE
) {
3813 ata_for_each_dev(dev
, link
, ALL
) {
3814 if (!(link
->eh_context
.unloaded_mask
&
3818 ata_eh_park_issue_cmd(dev
, 0);
3819 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3825 ata_for_each_link(link
, ap
, PMP_FIRST
) {
3826 struct ata_eh_context
*ehc
= &link
->eh_context
;
3828 if (sata_pmp_attached(ap
) && ata_is_host_link(link
))
3831 /* revalidate existing devices and attach new ones */
3832 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3836 /* if PMP got attached, return, pmp EH will take care of it */
3837 if (link
->device
->class == ATA_DEV_PMP
) {
3842 /* configure transfer mode if necessary */
3843 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3844 rc
= ata_set_mode(link
, &dev
);
3847 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3850 /* If reset has been issued, clear UA to avoid
3851 * disrupting the current users of the device.
3853 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3854 ata_for_each_dev(dev
, link
, ALL
) {
3855 if (dev
->class != ATA_DEV_ATAPI
)
3857 rc
= atapi_eh_clear_ua(dev
);
3860 if (zpodd_dev_enabled(dev
))
3861 zpodd_post_poweron(dev
);
3865 /* retry flush if necessary */
3866 ata_for_each_dev(dev
, link
, ALL
) {
3867 if (dev
->class != ATA_DEV_ATA
)
3869 rc
= ata_eh_maybe_retry_flush(dev
);
3875 /* configure link power saving */
3876 if (link
->lpm_policy
!= ap
->target_lpm_policy
) {
3877 rc
= ata_eh_set_lpm(link
, ap
->target_lpm_policy
, &dev
);
3882 /* this link is okay now */
3889 ata_eh_handle_dev_fail(dev
, rc
);
3891 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3892 /* PMP reset requires working host port.
3893 * Can't retry if it's frozen.
3895 if (sata_pmp_attached(ap
))
3905 if (rc
&& r_failed_link
)
3906 *r_failed_link
= link
;
3908 DPRINTK("EXIT, rc=%d\n", rc
);
3913 * ata_eh_finish - finish up EH
3914 * @ap: host port to finish EH for
3916 * Recovery is complete. Clean up EH states and retry or finish
3922 void ata_eh_finish(struct ata_port
*ap
)
3926 /* retry or finish qcs */
3927 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3928 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3930 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3934 /* FIXME: Once EH migration is complete,
3935 * generate sense data in this function,
3936 * considering both err_mask and tf.
3938 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3939 ata_eh_qc_retry(qc
);
3941 ata_eh_qc_complete(qc
);
3943 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3944 ata_eh_qc_complete(qc
);
3946 /* feed zero TF to sense generation */
3947 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3948 ata_eh_qc_retry(qc
);
3953 /* make sure nr_active_links is zero after EH */
3954 WARN_ON(ap
->nr_active_links
);
3955 ap
->nr_active_links
= 0;
3959 * ata_do_eh - do standard error handling
3960 * @ap: host port to handle error for
3962 * @prereset: prereset method (can be NULL)
3963 * @softreset: softreset method (can be NULL)
3964 * @hardreset: hardreset method (can be NULL)
3965 * @postreset: postreset method (can be NULL)
3967 * Perform standard error handling sequence.
3970 * Kernel thread context (may sleep).
3972 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3973 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3974 ata_postreset_fn_t postreset
)
3976 struct ata_device
*dev
;
3982 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3985 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3986 ata_dev_disable(dev
);
3993 * ata_std_error_handler - standard error handler
3994 * @ap: host port to handle error for
3996 * Standard error handler
3999 * Kernel thread context (may sleep).
4001 void ata_std_error_handler(struct ata_port
*ap
)
4003 struct ata_port_operations
*ops
= ap
->ops
;
4004 ata_reset_fn_t hardreset
= ops
->hardreset
;
4006 /* ignore built-in hardreset if SCR access is not available */
4007 if (hardreset
== sata_std_hardreset
&& !sata_scr_valid(&ap
->link
))
4010 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
4015 * ata_eh_handle_port_suspend - perform port suspend operation
4016 * @ap: port to suspend
4021 * Kernel thread context (may sleep).
4023 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
4025 unsigned long flags
;
4027 struct ata_device
*dev
;
4029 /* are we suspending? */
4030 spin_lock_irqsave(ap
->lock
, flags
);
4031 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4032 ap
->pm_mesg
.event
& PM_EVENT_RESUME
) {
4033 spin_unlock_irqrestore(ap
->lock
, flags
);
4036 spin_unlock_irqrestore(ap
->lock
, flags
);
4038 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
4041 * If we have a ZPODD attached, check its zero
4042 * power ready status before the port is frozen.
4043 * Only needed for runtime suspend.
4045 if (PMSG_IS_AUTO(ap
->pm_mesg
)) {
4046 ata_for_each_dev(dev
, &ap
->link
, ENABLED
) {
4047 if (zpodd_dev_enabled(dev
))
4048 zpodd_on_suspend(dev
);
4052 /* tell ACPI we're suspending */
4053 rc
= ata_acpi_on_suspend(ap
);
4058 ata_eh_freeze_port(ap
);
4060 if (ap
->ops
->port_suspend
)
4061 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
4063 ata_acpi_set_state(ap
, ap
->pm_mesg
);
4066 spin_lock_irqsave(ap
->lock
, flags
);
4068 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
4070 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
4071 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
4072 ata_port_schedule_eh(ap
);
4074 if (ap
->pm_result
) {
4075 *ap
->pm_result
= rc
;
4076 ap
->pm_result
= NULL
;
4079 spin_unlock_irqrestore(ap
->lock
, flags
);
4085 * ata_eh_handle_port_resume - perform port resume operation
4086 * @ap: port to resume
4091 * Kernel thread context (may sleep).
4093 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
4095 struct ata_link
*link
;
4096 struct ata_device
*dev
;
4097 unsigned long flags
;
4100 /* are we resuming? */
4101 spin_lock_irqsave(ap
->lock
, flags
);
4102 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4103 !(ap
->pm_mesg
.event
& PM_EVENT_RESUME
)) {
4104 spin_unlock_irqrestore(ap
->lock
, flags
);
4107 spin_unlock_irqrestore(ap
->lock
, flags
);
4109 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
4112 * Error timestamps are in jiffies which doesn't run while
4113 * suspended and PHY events during resume isn't too uncommon.
4114 * When the two are combined, it can lead to unnecessary speed
4115 * downs if the machine is suspended and resumed repeatedly.
4116 * Clear error history.
4118 ata_for_each_link(link
, ap
, HOST_FIRST
)
4119 ata_for_each_dev(dev
, link
, ALL
)
4120 ata_ering_clear(&dev
->ering
);
4122 ata_acpi_set_state(ap
, ap
->pm_mesg
);
4124 if (ap
->ops
->port_resume
)
4125 rc
= ap
->ops
->port_resume(ap
);
4127 /* tell ACPI that we're resuming */
4128 ata_acpi_on_resume(ap
);
4131 spin_lock_irqsave(ap
->lock
, flags
);
4132 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
4133 if (ap
->pm_result
) {
4134 *ap
->pm_result
= rc
;
4135 ap
->pm_result
= NULL
;
4137 spin_unlock_irqrestore(ap
->lock
, flags
);
4139 #endif /* CONFIG_PM */