2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/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 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
1490 * @page: page to read
1491 * @buf: buffer to store read page
1492 * @sectors: number of sectors to read
1494 * Read log page using READ_LOG_EXT command.
1497 * Kernel thread context (may sleep).
1500 * 0 on success, AC_ERR_* mask otherwise.
1502 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1503 u8 page
, void *buf
, unsigned int sectors
)
1505 struct ata_taskfile tf
;
1506 unsigned int err_mask
;
1508 DPRINTK("read log page - page %d\n", page
);
1510 ata_tf_init(dev
, &tf
);
1511 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1514 tf
.hob_nsect
= sectors
>> 8;
1515 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1516 tf
.protocol
= ATA_PROT_PIO
;
1518 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1519 buf
, sectors
* ATA_SECT_SIZE
, 0);
1521 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1526 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1527 * @dev: Device to read log page 10h from
1528 * @tag: Resulting tag of the failed command
1529 * @tf: Resulting taskfile registers of the failed command
1531 * Read log page 10h to obtain NCQ error details and clear error
1535 * Kernel thread context (may sleep).
1538 * 0 on success, -errno otherwise.
1540 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1541 int *tag
, struct ata_taskfile
*tf
)
1543 u8
*buf
= dev
->link
->ap
->sector_buf
;
1544 unsigned int err_mask
;
1548 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1553 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1556 ata_dev_warn(dev
, "invalid checksum 0x%x on log page 10h\n",
1562 *tag
= buf
[0] & 0x1f;
1564 tf
->command
= buf
[2];
1565 tf
->feature
= buf
[3];
1569 tf
->device
= buf
[7];
1570 tf
->hob_lbal
= buf
[8];
1571 tf
->hob_lbam
= buf
[9];
1572 tf
->hob_lbah
= buf
[10];
1573 tf
->nsect
= buf
[12];
1574 tf
->hob_nsect
= buf
[13];
1580 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1581 * @dev: target ATAPI device
1582 * @r_sense_key: out parameter for sense_key
1584 * Perform ATAPI TEST_UNIT_READY.
1587 * EH context (may sleep).
1590 * 0 on success, AC_ERR_* mask on failure.
1592 static unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1594 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1595 struct ata_taskfile tf
;
1596 unsigned int err_mask
;
1598 ata_tf_init(dev
, &tf
);
1600 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1601 tf
.command
= ATA_CMD_PACKET
;
1602 tf
.protocol
= ATAPI_PROT_NODATA
;
1604 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1605 if (err_mask
== AC_ERR_DEV
)
1606 *r_sense_key
= tf
.feature
>> 4;
1611 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1612 * @dev: device to perform REQUEST_SENSE to
1613 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1614 * @dfl_sense_key: default sense key to use
1616 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1617 * SENSE. This function is EH helper.
1620 * Kernel thread context (may sleep).
1623 * 0 on success, AC_ERR_* mask on failure
1625 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1626 u8
*sense_buf
, u8 dfl_sense_key
)
1628 u8 cdb
[ATAPI_CDB_LEN
] =
1629 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1630 struct ata_port
*ap
= dev
->link
->ap
;
1631 struct ata_taskfile tf
;
1633 DPRINTK("ATAPI request sense\n");
1635 /* FIXME: is this needed? */
1636 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1638 /* initialize sense_buf with the error register,
1639 * for the case where they are -not- overwritten
1641 sense_buf
[0] = 0x70;
1642 sense_buf
[2] = dfl_sense_key
;
1644 /* some devices time out if garbage left in tf */
1645 ata_tf_init(dev
, &tf
);
1647 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1648 tf
.command
= ATA_CMD_PACKET
;
1650 /* is it pointless to prefer PIO for "safety reasons"? */
1651 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1652 tf
.protocol
= ATAPI_PROT_DMA
;
1653 tf
.feature
|= ATAPI_PKT_DMA
;
1655 tf
.protocol
= ATAPI_PROT_PIO
;
1656 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1660 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1661 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1665 * ata_eh_analyze_serror - analyze SError for a failed port
1666 * @link: ATA link to analyze SError for
1668 * Analyze SError if available and further determine cause of
1674 static void ata_eh_analyze_serror(struct ata_link
*link
)
1676 struct ata_eh_context
*ehc
= &link
->eh_context
;
1677 u32 serror
= ehc
->i
.serror
;
1678 unsigned int err_mask
= 0, action
= 0;
1681 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1682 err_mask
|= AC_ERR_ATA_BUS
;
1683 action
|= ATA_EH_RESET
;
1685 if (serror
& SERR_PROTOCOL
) {
1686 err_mask
|= AC_ERR_HSM
;
1687 action
|= ATA_EH_RESET
;
1689 if (serror
& SERR_INTERNAL
) {
1690 err_mask
|= AC_ERR_SYSTEM
;
1691 action
|= ATA_EH_RESET
;
1694 /* Determine whether a hotplug event has occurred. Both
1695 * SError.N/X are considered hotplug events for enabled or
1696 * host links. For disabled PMP links, only N bit is
1697 * considered as X bit is left at 1 for link plugging.
1699 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
)
1700 hotplug_mask
= 0; /* hotplug doesn't work w/ LPM */
1701 else if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1702 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1704 hotplug_mask
= SERR_PHYRDY_CHG
;
1706 if (serror
& hotplug_mask
)
1707 ata_ehi_hotplugged(&ehc
->i
);
1709 ehc
->i
.err_mask
|= err_mask
;
1710 ehc
->i
.action
|= action
;
1714 * ata_eh_analyze_ncq_error - analyze NCQ error
1715 * @link: ATA link to analyze NCQ error for
1717 * Read log page 10h, determine the offending qc and acquire
1718 * error status TF. For NCQ device errors, all LLDDs have to do
1719 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1723 * Kernel thread context (may sleep).
1725 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1727 struct ata_port
*ap
= link
->ap
;
1728 struct ata_eh_context
*ehc
= &link
->eh_context
;
1729 struct ata_device
*dev
= link
->device
;
1730 struct ata_queued_cmd
*qc
;
1731 struct ata_taskfile tf
;
1734 /* if frozen, we can't do much */
1735 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1738 /* is it NCQ device error? */
1739 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1742 /* has LLDD analyzed already? */
1743 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1744 qc
= __ata_qc_from_tag(ap
, tag
);
1746 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1753 /* okay, this error is ours */
1754 memset(&tf
, 0, sizeof(tf
));
1755 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1757 ata_link_err(link
, "failed to read log page 10h (errno=%d)\n",
1762 if (!(link
->sactive
& (1 << tag
))) {
1763 ata_link_err(link
, "log page 10h reported inactive tag %d\n",
1768 /* we've got the perpetrator, condemn it */
1769 qc
= __ata_qc_from_tag(ap
, tag
);
1770 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1771 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1772 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1773 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1777 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1778 * @qc: qc to analyze
1779 * @tf: Taskfile registers to analyze
1781 * Analyze taskfile of @qc and further determine cause of
1782 * failure. This function also requests ATAPI sense data if
1786 * Kernel thread context (may sleep).
1789 * Determined recovery action
1791 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1792 const struct ata_taskfile
*tf
)
1794 unsigned int tmp
, action
= 0;
1795 u8 stat
= tf
->command
, err
= tf
->feature
;
1797 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1798 qc
->err_mask
|= AC_ERR_HSM
;
1799 return ATA_EH_RESET
;
1802 if (stat
& (ATA_ERR
| ATA_DF
))
1803 qc
->err_mask
|= AC_ERR_DEV
;
1807 switch (qc
->dev
->class) {
1810 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1812 qc
->err_mask
|= AC_ERR_MEDIA
;
1814 qc
->err_mask
|= AC_ERR_INVALID
;
1818 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1819 tmp
= atapi_eh_request_sense(qc
->dev
,
1820 qc
->scsicmd
->sense_buffer
,
1821 qc
->result_tf
.feature
>> 4);
1823 /* ATA_QCFLAG_SENSE_VALID is used to
1824 * tell atapi_qc_complete() that sense
1825 * data is already valid.
1827 * TODO: interpret sense data and set
1828 * appropriate err_mask.
1830 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1832 qc
->err_mask
|= tmp
;
1836 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1837 action
|= ATA_EH_RESET
;
1842 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1847 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1851 base
= ATA_ECAT_DUBIOUS_NONE
;
1853 if (err_mask
& AC_ERR_ATA_BUS
)
1854 return base
+ ATA_ECAT_ATA_BUS
;
1856 if (err_mask
& AC_ERR_TIMEOUT
)
1857 return base
+ ATA_ECAT_TOUT_HSM
;
1859 if (eflags
& ATA_EFLAG_IS_IO
) {
1860 if (err_mask
& AC_ERR_HSM
)
1861 return base
+ ATA_ECAT_TOUT_HSM
;
1863 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1864 return base
+ ATA_ECAT_UNK_DEV
;
1870 struct speed_down_verdict_arg
{
1873 int nr_errors
[ATA_ECAT_NR
];
1876 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1878 struct speed_down_verdict_arg
*arg
= void_arg
;
1881 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) || (ent
->timestamp
< arg
->since
))
1884 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1886 arg
->nr_errors
[cat
]++;
1892 * ata_eh_speed_down_verdict - Determine speed down verdict
1893 * @dev: Device of interest
1895 * This function examines error ring of @dev and determines
1896 * whether NCQ needs to be turned off, transfer speed should be
1897 * stepped down, or falling back to PIO is necessary.
1899 * ECAT_ATA_BUS : ATA_BUS error for any command
1901 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1904 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1906 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1907 * data transfer hasn't been verified.
1911 * NCQ_OFF : Turn off NCQ.
1913 * SPEED_DOWN : Speed down transfer speed but don't fall back
1916 * FALLBACK_TO_PIO : Fall back to PIO.
1918 * Even if multiple verdicts are returned, only one action is
1919 * taken per error. An action triggered by non-DUBIOUS errors
1920 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1921 * This is to expedite speed down decisions right after device is
1922 * initially configured.
1924 * The followings are speed down rules. #1 and #2 deal with
1927 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1928 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1930 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1931 * occurred during last 5 mins, NCQ_OFF.
1933 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1934 * occurred during last 5 mins, FALLBACK_TO_PIO
1936 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1937 * during last 10 mins, NCQ_OFF.
1939 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1940 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1943 * Inherited from caller.
1946 * OR of ATA_EH_SPDN_* flags.
1948 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1950 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1951 u64 j64
= get_jiffies_64();
1952 struct speed_down_verdict_arg arg
;
1953 unsigned int verdict
= 0;
1955 /* scan past 5 mins of error history */
1956 memset(&arg
, 0, sizeof(arg
));
1957 arg
.since
= j64
- min(j64
, j5mins
);
1958 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1960 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1961 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1962 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1963 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1965 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1966 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1967 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1969 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1970 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1971 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1972 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1974 /* scan past 10 mins of error history */
1975 memset(&arg
, 0, sizeof(arg
));
1976 arg
.since
= j64
- min(j64
, j10mins
);
1977 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1979 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1980 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1981 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1983 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1984 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1985 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1986 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1992 * ata_eh_speed_down - record error and speed down if necessary
1993 * @dev: Failed device
1994 * @eflags: mask of ATA_EFLAG_* flags
1995 * @err_mask: err_mask of the error
1997 * Record error and examine error history to determine whether
1998 * adjusting transmission speed is necessary. It also sets
1999 * transmission limits appropriately if such adjustment is
2003 * Kernel thread context (may sleep).
2006 * Determined recovery action.
2008 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
2009 unsigned int eflags
, unsigned int err_mask
)
2011 struct ata_link
*link
= ata_dev_phys_link(dev
);
2013 unsigned int verdict
;
2014 unsigned int action
= 0;
2016 /* don't bother if Cat-0 error */
2017 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
2020 /* record error and determine whether speed down is necessary */
2021 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
2022 verdict
= ata_eh_speed_down_verdict(dev
);
2025 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
2026 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
2027 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
2028 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
2029 ata_dev_warn(dev
, "NCQ disabled due to excessive errors\n");
2034 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
2035 /* speed down SATA link speed if possible */
2036 if (sata_down_spd_limit(link
, 0) == 0) {
2037 action
|= ATA_EH_RESET
;
2041 /* lower transfer mode */
2042 if (dev
->spdn_cnt
< 2) {
2043 static const int dma_dnxfer_sel
[] =
2044 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
2045 static const int pio_dnxfer_sel
[] =
2046 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
2049 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
2050 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
2052 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
2056 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
2057 action
|= ATA_EH_RESET
;
2063 /* Fall back to PIO? Slowing down to PIO is meaningless for
2064 * SATA ATA devices. Consider it only for PATA and SATAPI.
2066 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
2067 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
2068 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
2069 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
2071 action
|= ATA_EH_RESET
;
2078 /* device has been slowed down, blow error history */
2079 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
2080 ata_ering_clear(&dev
->ering
);
2085 * ata_eh_worth_retry - analyze error and decide whether to retry
2086 * @qc: qc to possibly retry
2088 * Look at the cause of the error and decide if a retry
2089 * might be useful or not. We don't want to retry media errors
2090 * because the drive itself has probably already taken 10-30 seconds
2091 * doing its own internal retries before reporting the failure.
2093 static inline int ata_eh_worth_retry(struct ata_queued_cmd
*qc
)
2095 if (qc
->flags
& AC_ERR_MEDIA
)
2096 return 0; /* don't retry media errors */
2097 if (qc
->flags
& ATA_QCFLAG_IO
)
2098 return 1; /* otherwise retry anything from fs stack */
2099 if (qc
->err_mask
& AC_ERR_INVALID
)
2100 return 0; /* don't retry these */
2101 return qc
->err_mask
!= AC_ERR_DEV
; /* retry if not dev error */
2105 * ata_eh_link_autopsy - analyze error and determine recovery action
2106 * @link: host link to perform autopsy on
2108 * Analyze why @link failed and determine which recovery actions
2109 * are needed. This function also sets more detailed AC_ERR_*
2110 * values and fills sense data for ATAPI CHECK SENSE.
2113 * Kernel thread context (may sleep).
2115 static void ata_eh_link_autopsy(struct ata_link
*link
)
2117 struct ata_port
*ap
= link
->ap
;
2118 struct ata_eh_context
*ehc
= &link
->eh_context
;
2119 struct ata_device
*dev
;
2120 unsigned int all_err_mask
= 0, eflags
= 0;
2127 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
2130 /* obtain and analyze SError */
2131 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
2133 ehc
->i
.serror
|= serror
;
2134 ata_eh_analyze_serror(link
);
2135 } else if (rc
!= -EOPNOTSUPP
) {
2136 /* SError read failed, force reset and probing */
2137 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
2138 ehc
->i
.action
|= ATA_EH_RESET
;
2139 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
2142 /* analyze NCQ failure */
2143 ata_eh_analyze_ncq_error(link
);
2145 /* any real error trumps AC_ERR_OTHER */
2146 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
2147 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
2149 all_err_mask
|= ehc
->i
.err_mask
;
2151 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2152 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2154 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2155 ata_dev_phys_link(qc
->dev
) != link
)
2158 /* inherit upper level err_mask */
2159 qc
->err_mask
|= ehc
->i
.err_mask
;
2162 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
2164 /* DEV errors are probably spurious in case of ATA_BUS error */
2165 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
2166 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
2169 /* any real error trumps unknown error */
2170 if (qc
->err_mask
& ~AC_ERR_OTHER
)
2171 qc
->err_mask
&= ~AC_ERR_OTHER
;
2173 /* SENSE_VALID trumps dev/unknown error and revalidation */
2174 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
2175 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
2177 /* determine whether the command is worth retrying */
2178 if (ata_eh_worth_retry(qc
))
2179 qc
->flags
|= ATA_QCFLAG_RETRY
;
2181 /* accumulate error info */
2182 ehc
->i
.dev
= qc
->dev
;
2183 all_err_mask
|= qc
->err_mask
;
2184 if (qc
->flags
& ATA_QCFLAG_IO
)
2185 eflags
|= ATA_EFLAG_IS_IO
;
2188 /* enforce default EH actions */
2189 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
2190 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
2191 ehc
->i
.action
|= ATA_EH_RESET
;
2192 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
2193 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
2194 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2196 /* If we have offending qcs and the associated failed device,
2197 * perform per-dev EH action only on the offending device.
2200 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
2201 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
2202 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2205 /* propagate timeout to host link */
2206 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2207 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2209 /* record error and consider speeding down */
2211 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2212 ata_dev_enabled(link
->device
))))
2216 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2217 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2218 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2225 * ata_eh_autopsy - analyze error and determine recovery action
2226 * @ap: host port to perform autopsy on
2228 * Analyze all links of @ap and determine why they failed and
2229 * which recovery actions are needed.
2232 * Kernel thread context (may sleep).
2234 void ata_eh_autopsy(struct ata_port
*ap
)
2236 struct ata_link
*link
;
2238 ata_for_each_link(link
, ap
, EDGE
)
2239 ata_eh_link_autopsy(link
);
2241 /* Handle the frigging slave link. Autopsy is done similarly
2242 * but actions and flags are transferred over to the master
2243 * link and handled from there.
2245 if (ap
->slave_link
) {
2246 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2247 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2249 /* transfer control flags from master to slave */
2250 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2252 /* perform autopsy on the slave link */
2253 ata_eh_link_autopsy(ap
->slave_link
);
2255 /* transfer actions from slave to master and clear slave */
2256 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2257 mehc
->i
.action
|= sehc
->i
.action
;
2258 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2259 mehc
->i
.flags
|= sehc
->i
.flags
;
2260 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2263 /* Autopsy of fanout ports can affect host link autopsy.
2264 * Perform host link autopsy last.
2266 if (sata_pmp_attached(ap
))
2267 ata_eh_link_autopsy(&ap
->link
);
2271 * ata_get_cmd_descript - get description for ATA command
2272 * @command: ATA command code to get description for
2274 * Return a textual description of the given command, or NULL if the
2275 * command is not known.
2280 const char *ata_get_cmd_descript(u8 command
)
2282 #ifdef CONFIG_ATA_VERBOSE_ERROR
2288 { ATA_CMD_DEV_RESET
, "DEVICE RESET" },
2289 { ATA_CMD_CHK_POWER
, "CHECK POWER MODE" },
2290 { ATA_CMD_STANDBY
, "STANDBY" },
2291 { ATA_CMD_IDLE
, "IDLE" },
2292 { ATA_CMD_EDD
, "EXECUTE DEVICE DIAGNOSTIC" },
2293 { ATA_CMD_DOWNLOAD_MICRO
, "DOWNLOAD MICROCODE" },
2294 { ATA_CMD_NOP
, "NOP" },
2295 { ATA_CMD_FLUSH
, "FLUSH CACHE" },
2296 { ATA_CMD_FLUSH_EXT
, "FLUSH CACHE EXT" },
2297 { ATA_CMD_ID_ATA
, "IDENTIFY DEVICE" },
2298 { ATA_CMD_ID_ATAPI
, "IDENTIFY PACKET DEVICE" },
2299 { ATA_CMD_SERVICE
, "SERVICE" },
2300 { ATA_CMD_READ
, "READ DMA" },
2301 { ATA_CMD_READ_EXT
, "READ DMA EXT" },
2302 { ATA_CMD_READ_QUEUED
, "READ DMA QUEUED" },
2303 { ATA_CMD_READ_STREAM_EXT
, "READ STREAM EXT" },
2304 { ATA_CMD_READ_STREAM_DMA_EXT
, "READ STREAM DMA EXT" },
2305 { ATA_CMD_WRITE
, "WRITE DMA" },
2306 { ATA_CMD_WRITE_EXT
, "WRITE DMA EXT" },
2307 { ATA_CMD_WRITE_QUEUED
, "WRITE DMA QUEUED EXT" },
2308 { ATA_CMD_WRITE_STREAM_EXT
, "WRITE STREAM EXT" },
2309 { ATA_CMD_WRITE_STREAM_DMA_EXT
, "WRITE STREAM DMA EXT" },
2310 { ATA_CMD_WRITE_FUA_EXT
, "WRITE DMA FUA EXT" },
2311 { ATA_CMD_WRITE_QUEUED_FUA_EXT
, "WRITE DMA QUEUED FUA EXT" },
2312 { ATA_CMD_FPDMA_READ
, "READ FPDMA QUEUED" },
2313 { ATA_CMD_FPDMA_WRITE
, "WRITE FPDMA QUEUED" },
2314 { ATA_CMD_PIO_READ
, "READ SECTOR(S)" },
2315 { ATA_CMD_PIO_READ_EXT
, "READ SECTOR(S) EXT" },
2316 { ATA_CMD_PIO_WRITE
, "WRITE SECTOR(S)" },
2317 { ATA_CMD_PIO_WRITE_EXT
, "WRITE SECTOR(S) EXT" },
2318 { ATA_CMD_READ_MULTI
, "READ MULTIPLE" },
2319 { ATA_CMD_READ_MULTI_EXT
, "READ MULTIPLE EXT" },
2320 { ATA_CMD_WRITE_MULTI
, "WRITE MULTIPLE" },
2321 { ATA_CMD_WRITE_MULTI_EXT
, "WRITE MULTIPLE EXT" },
2322 { ATA_CMD_WRITE_MULTI_FUA_EXT
, "WRITE MULTIPLE FUA EXT" },
2323 { ATA_CMD_SET_FEATURES
, "SET FEATURES" },
2324 { ATA_CMD_SET_MULTI
, "SET MULTIPLE MODE" },
2325 { ATA_CMD_VERIFY
, "READ VERIFY SECTOR(S)" },
2326 { ATA_CMD_VERIFY_EXT
, "READ VERIFY SECTOR(S) EXT" },
2327 { ATA_CMD_WRITE_UNCORR_EXT
, "WRITE UNCORRECTABLE EXT" },
2328 { ATA_CMD_STANDBYNOW1
, "STANDBY IMMEDIATE" },
2329 { ATA_CMD_IDLEIMMEDIATE
, "IDLE IMMEDIATE" },
2330 { ATA_CMD_SLEEP
, "SLEEP" },
2331 { ATA_CMD_INIT_DEV_PARAMS
, "INITIALIZE DEVICE PARAMETERS" },
2332 { ATA_CMD_READ_NATIVE_MAX
, "READ NATIVE MAX ADDRESS" },
2333 { ATA_CMD_READ_NATIVE_MAX_EXT
, "READ NATIVE MAX ADDRESS EXT" },
2334 { ATA_CMD_SET_MAX
, "SET MAX ADDRESS" },
2335 { ATA_CMD_SET_MAX_EXT
, "SET MAX ADDRESS EXT" },
2336 { ATA_CMD_READ_LOG_EXT
, "READ LOG EXT" },
2337 { ATA_CMD_WRITE_LOG_EXT
, "WRITE LOG EXT" },
2338 { ATA_CMD_READ_LOG_DMA_EXT
, "READ LOG DMA EXT" },
2339 { ATA_CMD_WRITE_LOG_DMA_EXT
, "WRITE LOG DMA EXT" },
2340 { ATA_CMD_TRUSTED_RCV
, "TRUSTED RECEIVE" },
2341 { ATA_CMD_TRUSTED_RCV_DMA
, "TRUSTED RECEIVE DMA" },
2342 { ATA_CMD_TRUSTED_SND
, "TRUSTED SEND" },
2343 { ATA_CMD_TRUSTED_SND_DMA
, "TRUSTED SEND DMA" },
2344 { ATA_CMD_PMP_READ
, "READ BUFFER" },
2345 { ATA_CMD_PMP_WRITE
, "WRITE BUFFER" },
2346 { ATA_CMD_CONF_OVERLAY
, "DEVICE CONFIGURATION OVERLAY" },
2347 { ATA_CMD_SEC_SET_PASS
, "SECURITY SET PASSWORD" },
2348 { ATA_CMD_SEC_UNLOCK
, "SECURITY UNLOCK" },
2349 { ATA_CMD_SEC_ERASE_PREP
, "SECURITY ERASE PREPARE" },
2350 { ATA_CMD_SEC_ERASE_UNIT
, "SECURITY ERASE UNIT" },
2351 { ATA_CMD_SEC_FREEZE_LOCK
, "SECURITY FREEZE LOCK" },
2352 { ATA_CMD_SEC_DISABLE_PASS
, "SECURITY DISABLE PASSWORD" },
2353 { ATA_CMD_CONFIG_STREAM
, "CONFIGURE STREAM" },
2354 { ATA_CMD_SMART
, "SMART" },
2355 { ATA_CMD_MEDIA_LOCK
, "DOOR LOCK" },
2356 { ATA_CMD_MEDIA_UNLOCK
, "DOOR UNLOCK" },
2357 { ATA_CMD_DSM
, "DATA SET MANAGEMENT" },
2358 { ATA_CMD_CHK_MED_CRD_TYP
, "CHECK MEDIA CARD TYPE" },
2359 { ATA_CMD_CFA_REQ_EXT_ERR
, "CFA REQUEST EXTENDED ERROR" },
2360 { ATA_CMD_CFA_WRITE_NE
, "CFA WRITE SECTORS WITHOUT ERASE" },
2361 { ATA_CMD_CFA_TRANS_SECT
, "CFA TRANSLATE SECTOR" },
2362 { ATA_CMD_CFA_ERASE
, "CFA ERASE SECTORS" },
2363 { ATA_CMD_CFA_WRITE_MULT_NE
, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2364 { ATA_CMD_READ_LONG
, "READ LONG (with retries)" },
2365 { ATA_CMD_READ_LONG_ONCE
, "READ LONG (without retries)" },
2366 { ATA_CMD_WRITE_LONG
, "WRITE LONG (with retries)" },
2367 { ATA_CMD_WRITE_LONG_ONCE
, "WRITE LONG (without retries)" },
2368 { ATA_CMD_RESTORE
, "RECALIBRATE" },
2369 { 0, NULL
} /* terminate list */
2373 for (i
= 0; cmd_descr
[i
].text
; i
++)
2374 if (cmd_descr
[i
].command
== command
)
2375 return cmd_descr
[i
].text
;
2382 * ata_eh_link_report - report error handling to user
2383 * @link: ATA link EH is going on
2385 * Report EH to user.
2390 static void ata_eh_link_report(struct ata_link
*link
)
2392 struct ata_port
*ap
= link
->ap
;
2393 struct ata_eh_context
*ehc
= &link
->eh_context
;
2394 const char *frozen
, *desc
;
2396 int tag
, nr_failed
= 0;
2398 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2402 if (ehc
->i
.desc
[0] != '\0')
2405 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2406 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2408 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2409 ata_dev_phys_link(qc
->dev
) != link
||
2410 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2411 qc
->err_mask
== AC_ERR_DEV
))
2413 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2419 if (!nr_failed
&& !ehc
->i
.err_mask
)
2423 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2426 memset(tries_buf
, 0, sizeof(tries_buf
));
2427 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2428 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
2432 ata_dev_err(ehc
->i
.dev
, "exception Emask 0x%x "
2433 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2434 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2435 ehc
->i
.action
, frozen
, tries_buf
);
2437 ata_dev_err(ehc
->i
.dev
, "%s\n", desc
);
2439 ata_link_err(link
, "exception Emask 0x%x "
2440 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2441 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2442 ehc
->i
.action
, frozen
, tries_buf
);
2444 ata_link_err(link
, "%s\n", desc
);
2447 #ifdef CONFIG_ATA_VERBOSE_ERROR
2450 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2451 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2452 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2453 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2454 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2455 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2456 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2457 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2458 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2459 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2460 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2461 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2462 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2463 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2464 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2465 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2466 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2467 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2470 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2471 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2472 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2473 const u8
*cdb
= qc
->cdb
;
2474 char data_buf
[20] = "";
2475 char cdb_buf
[70] = "";
2477 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2478 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2481 if (qc
->dma_dir
!= DMA_NONE
) {
2482 static const char *dma_str
[] = {
2483 [DMA_BIDIRECTIONAL
] = "bidi",
2484 [DMA_TO_DEVICE
] = "out",
2485 [DMA_FROM_DEVICE
] = "in",
2487 static const char *prot_str
[] = {
2488 [ATA_PROT_PIO
] = "pio",
2489 [ATA_PROT_DMA
] = "dma",
2490 [ATA_PROT_NCQ
] = "ncq",
2491 [ATAPI_PROT_PIO
] = "pio",
2492 [ATAPI_PROT_DMA
] = "dma",
2495 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2496 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2497 dma_str
[qc
->dma_dir
]);
2500 if (ata_is_atapi(qc
->tf
.protocol
)) {
2502 scsi_print_command(qc
->scsicmd
);
2504 snprintf(cdb_buf
, sizeof(cdb_buf
),
2505 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2506 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2507 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2508 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2509 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2510 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2512 const char *descr
= ata_get_cmd_descript(cmd
->command
);
2514 ata_dev_err(qc
->dev
, "failed command: %s\n",
2518 ata_dev_err(qc
->dev
,
2519 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2521 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2522 "Emask 0x%x (%s)%s\n",
2523 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2524 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2525 cmd
->hob_feature
, cmd
->hob_nsect
,
2526 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2527 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2528 res
->command
, res
->feature
, res
->nsect
,
2529 res
->lbal
, res
->lbam
, res
->lbah
,
2530 res
->hob_feature
, res
->hob_nsect
,
2531 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2532 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2533 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2535 #ifdef CONFIG_ATA_VERBOSE_ERROR
2536 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2538 if (res
->command
& ATA_BUSY
)
2539 ata_dev_err(qc
->dev
, "status: { Busy }\n");
2541 ata_dev_err(qc
->dev
, "status: { %s%s%s%s}\n",
2542 res
->command
& ATA_DRDY
? "DRDY " : "",
2543 res
->command
& ATA_DF
? "DF " : "",
2544 res
->command
& ATA_DRQ
? "DRQ " : "",
2545 res
->command
& ATA_ERR
? "ERR " : "");
2548 if (cmd
->command
!= ATA_CMD_PACKET
&&
2549 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2551 ata_dev_err(qc
->dev
, "error: { %s%s%s%s}\n",
2552 res
->feature
& ATA_ICRC
? "ICRC " : "",
2553 res
->feature
& ATA_UNC
? "UNC " : "",
2554 res
->feature
& ATA_IDNF
? "IDNF " : "",
2555 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2561 * ata_eh_report - report error handling to user
2562 * @ap: ATA port to report EH about
2564 * Report EH to user.
2569 void ata_eh_report(struct ata_port
*ap
)
2571 struct ata_link
*link
;
2573 ata_for_each_link(link
, ap
, HOST_FIRST
)
2574 ata_eh_link_report(link
);
2577 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2578 unsigned int *classes
, unsigned long deadline
,
2581 struct ata_device
*dev
;
2584 ata_for_each_dev(dev
, link
, ALL
)
2585 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2587 return reset(link
, classes
, deadline
);
2590 static int ata_eh_followup_srst_needed(struct ata_link
*link
, int rc
)
2592 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2596 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2601 int ata_eh_reset(struct ata_link
*link
, int classify
,
2602 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2603 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2605 struct ata_port
*ap
= link
->ap
;
2606 struct ata_link
*slave
= ap
->slave_link
;
2607 struct ata_eh_context
*ehc
= &link
->eh_context
;
2608 struct ata_eh_context
*sehc
= slave
? &slave
->eh_context
: NULL
;
2609 unsigned int *classes
= ehc
->classes
;
2610 unsigned int lflags
= link
->flags
;
2611 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2612 int max_tries
= 0, try = 0;
2613 struct ata_link
*failed_link
;
2614 struct ata_device
*dev
;
2615 unsigned long deadline
, now
;
2616 ata_reset_fn_t reset
;
2617 unsigned long flags
;
2624 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2626 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2628 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2631 /* make sure each reset attempt is at least COOL_DOWN apart */
2632 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2634 WARN_ON(time_after(ehc
->last_reset
, now
));
2635 deadline
= ata_deadline(ehc
->last_reset
,
2636 ATA_EH_RESET_COOL_DOWN
);
2637 if (time_before(now
, deadline
))
2638 schedule_timeout_uninterruptible(deadline
- now
);
2641 spin_lock_irqsave(ap
->lock
, flags
);
2642 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2643 spin_unlock_irqrestore(ap
->lock
, flags
);
2645 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2647 ata_for_each_dev(dev
, link
, ALL
) {
2648 /* If we issue an SRST then an ATA drive (not ATAPI)
2649 * may change configuration and be in PIO0 timing. If
2650 * we do a hard reset (or are coming from power on)
2651 * this is true for ATA or ATAPI. Until we've set a
2652 * suitable controller mode we should not touch the
2653 * bus as we may be talking too fast.
2655 dev
->pio_mode
= XFER_PIO_0
;
2657 /* If the controller has a pio mode setup function
2658 * then use it to set the chipset to rights. Don't
2659 * touch the DMA setup as that will be dealt with when
2660 * configuring devices.
2662 if (ap
->ops
->set_piomode
)
2663 ap
->ops
->set_piomode(ap
, dev
);
2666 /* prefer hardreset */
2668 ehc
->i
.action
&= ~ATA_EH_RESET
;
2671 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2672 } else if (softreset
) {
2674 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2678 unsigned long deadline
= ata_deadline(jiffies
,
2679 ATA_EH_PRERESET_TIMEOUT
);
2682 sehc
->i
.action
&= ~ATA_EH_RESET
;
2683 sehc
->i
.action
|= ehc
->i
.action
;
2686 rc
= prereset(link
, deadline
);
2688 /* If present, do prereset on slave link too. Reset
2689 * is skipped iff both master and slave links report
2690 * -ENOENT or clear ATA_EH_RESET.
2692 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2695 tmp
= prereset(slave
, deadline
);
2699 ehc
->i
.action
|= sehc
->i
.action
;
2703 if (rc
== -ENOENT
) {
2704 ata_link_dbg(link
, "port disabled--ignoring\n");
2705 ehc
->i
.action
&= ~ATA_EH_RESET
;
2707 ata_for_each_dev(dev
, link
, ALL
)
2708 classes
[dev
->devno
] = ATA_DEV_NONE
;
2713 "prereset failed (errno=%d)\n",
2718 /* prereset() might have cleared ATA_EH_RESET. If so,
2719 * bang classes, thaw and return.
2721 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2722 ata_for_each_dev(dev
, link
, ALL
)
2723 classes
[dev
->devno
] = ATA_DEV_NONE
;
2724 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) &&
2725 ata_is_host_link(link
))
2726 ata_eh_thaw_port(ap
);
2736 if (ata_is_host_link(link
))
2737 ata_eh_freeze_port(ap
);
2739 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2743 ata_link_info(link
, "%s resetting link\n",
2744 reset
== softreset
? "soft" : "hard");
2746 /* mark that this EH session started with reset */
2747 ehc
->last_reset
= jiffies
;
2748 if (reset
== hardreset
)
2749 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2751 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2753 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2754 if (rc
&& rc
!= -EAGAIN
) {
2759 /* hardreset slave link if existent */
2760 if (slave
&& reset
== hardreset
) {
2764 ata_link_info(slave
, "hard resetting link\n");
2766 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2767 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2775 failed_link
= slave
;
2781 /* perform follow-up SRST if necessary */
2782 if (reset
== hardreset
&&
2783 ata_eh_followup_srst_needed(link
, rc
)) {
2788 "follow-up softreset required but no softreset available\n");
2794 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2795 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2804 "no reset method available, skipping reset\n");
2805 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2806 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2810 * Post-reset processing
2812 ata_for_each_dev(dev
, link
, ALL
) {
2813 /* After the reset, the device state is PIO 0 and the
2814 * controller state is undefined. Reset also wakes up
2815 * drives from sleeping mode.
2817 dev
->pio_mode
= XFER_PIO_0
;
2818 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2820 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2823 /* apply class override */
2824 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2825 classes
[dev
->devno
] = ATA_DEV_ATA
;
2826 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2827 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2830 /* record current link speed */
2831 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2832 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2833 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2834 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2837 if (ata_is_host_link(link
))
2838 ata_eh_thaw_port(ap
);
2840 /* postreset() should clear hardware SError. Although SError
2841 * is cleared during link resume, clearing SError here is
2842 * necessary as some PHYs raise hotplug events after SRST.
2843 * This introduces race condition where hotplug occurs between
2844 * reset and here. This race is mediated by cross checking
2845 * link onlineness and classification result later.
2848 postreset(link
, classes
);
2850 postreset(slave
, classes
);
2854 * Some controllers can't be frozen very well and may set spurious
2855 * error conditions during reset. Clear accumulated error
2856 * information and re-thaw the port if frozen. As reset is the
2857 * final recovery action and we cross check link onlineness against
2858 * device classification later, no hotplug event is lost by this.
2860 spin_lock_irqsave(link
->ap
->lock
, flags
);
2861 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
2863 memset(&slave
->eh_info
, 0, sizeof(link
->eh_info
));
2864 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
2865 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2867 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2868 ata_eh_thaw_port(ap
);
2871 * Make sure onlineness and classification result correspond.
2872 * Hotplug could have happened during reset and some
2873 * controllers fail to wait while a drive is spinning up after
2874 * being hotplugged causing misdetection. By cross checking
2875 * link on/offlineness and classification result, those
2876 * conditions can be reliably detected and retried.
2879 ata_for_each_dev(dev
, link
, ALL
) {
2880 if (ata_phys_link_online(ata_dev_phys_link(dev
))) {
2881 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2882 ata_dev_dbg(dev
, "link online but device misclassified\n");
2883 classes
[dev
->devno
] = ATA_DEV_NONE
;
2886 } else if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2887 if (ata_class_enabled(classes
[dev
->devno
]))
2889 "link offline, clearing class %d to NONE\n",
2890 classes
[dev
->devno
]);
2891 classes
[dev
->devno
] = ATA_DEV_NONE
;
2892 } else if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2894 "link status unknown, clearing UNKNOWN to NONE\n");
2895 classes
[dev
->devno
] = ATA_DEV_NONE
;
2899 if (classify
&& nr_unknown
) {
2900 if (try < max_tries
) {
2902 "link online but %d devices misclassified, retrying\n",
2909 "link online but %d devices misclassified, "
2910 "device detection might fail\n", nr_unknown
);
2913 /* reset successful, schedule revalidation */
2914 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2916 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2917 ehc
->last_reset
= jiffies
; /* update to completion time */
2918 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2919 link
->lpm_policy
= ATA_LPM_UNKNOWN
; /* reset LPM state */
2923 /* clear hotplug flag */
2924 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2926 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2928 spin_lock_irqsave(ap
->lock
, flags
);
2929 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2930 spin_unlock_irqrestore(ap
->lock
, flags
);
2935 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2936 if (!ata_is_host_link(link
) &&
2937 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2940 if (try >= max_tries
) {
2942 * Thaw host port even if reset failed, so that the port
2943 * can be retried on the next phy event. This risks
2944 * repeated EH runs but seems to be a better tradeoff than
2945 * shutting down a port after a botched hotplug attempt.
2947 if (ata_is_host_link(link
))
2948 ata_eh_thaw_port(ap
);
2953 if (time_before(now
, deadline
)) {
2954 unsigned long delta
= deadline
- now
;
2956 ata_link_warn(failed_link
,
2957 "reset failed (errno=%d), retrying in %u secs\n",
2958 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2962 delta
= schedule_timeout_uninterruptible(delta
);
2967 * While disks spinup behind PMP, some controllers fail sending SRST.
2968 * They need to be reset - as well as the PMP - before retrying.
2970 if (rc
== -ERESTART
) {
2971 if (ata_is_host_link(link
))
2972 ata_eh_thaw_port(ap
);
2976 if (try == max_tries
- 1) {
2977 sata_down_spd_limit(link
, 0);
2979 sata_down_spd_limit(slave
, 0);
2980 } else if (rc
== -EPIPE
)
2981 sata_down_spd_limit(failed_link
, 0);
2988 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
2990 struct ata_link
*link
;
2991 struct ata_device
*dev
;
2992 unsigned long flags
;
2995 * This function can be thought of as an extended version of
2996 * ata_eh_about_to_do() specially crafted to accommodate the
2997 * requirements of ATA_EH_PARK handling. Since the EH thread
2998 * does not leave the do {} while () loop in ata_eh_recover as
2999 * long as the timeout for a park request to *one* device on
3000 * the port has not expired, and since we still want to pick
3001 * up park requests to other devices on the same port or
3002 * timeout updates for the same device, we have to pull
3003 * ATA_EH_PARK actions from eh_info into eh_context.i
3004 * ourselves at the beginning of each pass over the loop.
3006 * Additionally, all write accesses to &ap->park_req_pending
3007 * through INIT_COMPLETION() (see below) or complete_all()
3008 * (see ata_scsi_park_store()) are protected by the host lock.
3009 * As a result we have that park_req_pending.done is zero on
3010 * exit from this function, i.e. when ATA_EH_PARK actions for
3011 * *all* devices on port ap have been pulled into the
3012 * respective eh_context structs. If, and only if,
3013 * park_req_pending.done is non-zero by the time we reach
3014 * wait_for_completion_timeout(), another ATA_EH_PARK action
3015 * has been scheduled for at least one of the devices on port
3016 * ap and we have to cycle over the do {} while () loop in
3017 * ata_eh_recover() again.
3020 spin_lock_irqsave(ap
->lock
, flags
);
3021 INIT_COMPLETION(ap
->park_req_pending
);
3022 ata_for_each_link(link
, ap
, EDGE
) {
3023 ata_for_each_dev(dev
, link
, ALL
) {
3024 struct ata_eh_info
*ehi
= &link
->eh_info
;
3026 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
3027 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
3028 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
3031 spin_unlock_irqrestore(ap
->lock
, flags
);
3034 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
3036 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3037 struct ata_taskfile tf
;
3038 unsigned int err_mask
;
3040 ata_tf_init(dev
, &tf
);
3042 ehc
->unloaded_mask
|= 1 << dev
->devno
;
3043 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
3049 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3050 tf
.command
= ATA_CMD_CHK_POWER
;
3053 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
3054 tf
.protocol
|= ATA_PROT_NODATA
;
3055 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3056 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
3057 ata_dev_err(dev
, "head unload failed!\n");
3058 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3062 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
3063 struct ata_device
**r_failed_dev
)
3065 struct ata_port
*ap
= link
->ap
;
3066 struct ata_eh_context
*ehc
= &link
->eh_context
;
3067 struct ata_device
*dev
;
3068 unsigned int new_mask
= 0;
3069 unsigned long flags
;
3074 /* For PATA drive side cable detection to work, IDENTIFY must
3075 * be done backwards such that PDIAG- is released by the slave
3076 * device before the master device is identified.
3078 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
3079 unsigned int action
= ata_eh_dev_action(dev
);
3080 unsigned int readid_flags
= 0;
3082 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
3083 readid_flags
|= ATA_READID_POSTRESET
;
3085 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
3086 WARN_ON(dev
->class == ATA_DEV_PMP
);
3088 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
3093 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
3094 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
3099 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
3101 /* Configuration may have changed, reconfigure
3104 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3106 /* schedule the scsi_rescan_device() here */
3107 schedule_work(&(ap
->scsi_rescan_task
));
3108 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
3109 ehc
->tries
[dev
->devno
] &&
3110 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
3111 /* Temporarily set dev->class, it will be
3112 * permanently set once all configurations are
3113 * complete. This is necessary because new
3114 * device configuration is done in two
3117 dev
->class = ehc
->classes
[dev
->devno
];
3119 if (dev
->class == ATA_DEV_PMP
)
3120 rc
= sata_pmp_attach(dev
);
3122 rc
= ata_dev_read_id(dev
, &dev
->class,
3123 readid_flags
, dev
->id
);
3125 /* read_id might have changed class, store and reset */
3126 ehc
->classes
[dev
->devno
] = dev
->class;
3127 dev
->class = ATA_DEV_UNKNOWN
;
3131 /* clear error info accumulated during probe */
3132 ata_ering_clear(&dev
->ering
);
3133 new_mask
|= 1 << dev
->devno
;
3136 /* IDENTIFY was issued to non-existent
3137 * device. No need to reset. Just
3138 * thaw and ignore the device.
3140 ata_eh_thaw_port(ap
);
3148 /* PDIAG- should have been released, ask cable type if post-reset */
3149 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
3150 if (ap
->ops
->cable_detect
)
3151 ap
->cbl
= ap
->ops
->cable_detect(ap
);
3155 /* Configure new devices forward such that user doesn't see
3156 * device detection messages backwards.
3158 ata_for_each_dev(dev
, link
, ALL
) {
3159 if (!(new_mask
& (1 << dev
->devno
)))
3162 dev
->class = ehc
->classes
[dev
->devno
];
3164 if (dev
->class == ATA_DEV_PMP
)
3167 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
3168 rc
= ata_dev_configure(dev
);
3169 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
3171 dev
->class = ATA_DEV_UNKNOWN
;
3175 spin_lock_irqsave(ap
->lock
, flags
);
3176 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
3177 spin_unlock_irqrestore(ap
->lock
, flags
);
3179 /* new device discovered, configure xfermode */
3180 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3186 *r_failed_dev
= dev
;
3187 DPRINTK("EXIT rc=%d\n", rc
);
3192 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3193 * @link: link on which timings will be programmed
3194 * @r_failed_dev: out parameter for failed device
3196 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3197 * ata_set_mode() fails, pointer to the failing device is
3198 * returned in @r_failed_dev.
3201 * PCI/etc. bus probe sem.
3204 * 0 on success, negative errno otherwise
3206 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
3208 struct ata_port
*ap
= link
->ap
;
3209 struct ata_device
*dev
;
3212 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3213 ata_for_each_dev(dev
, link
, ENABLED
) {
3214 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
3215 struct ata_ering_entry
*ent
;
3217 ent
= ata_ering_top(&dev
->ering
);
3219 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
3223 /* has private set_mode? */
3224 if (ap
->ops
->set_mode
)
3225 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
3227 rc
= ata_do_set_mode(link
, r_failed_dev
);
3229 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3230 ata_for_each_dev(dev
, link
, ENABLED
) {
3231 struct ata_eh_context
*ehc
= &link
->eh_context
;
3232 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
3233 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
3235 if (dev
->xfer_mode
!= saved_xfer_mode
||
3236 ata_ncq_enabled(dev
) != saved_ncq
)
3237 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
3244 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3245 * @dev: ATAPI device to clear UA for
3247 * Resets and other operations can make an ATAPI device raise
3248 * UNIT ATTENTION which causes the next operation to fail. This
3249 * function clears UA.
3252 * EH context (may sleep).
3255 * 0 on success, -errno on failure.
3257 static int atapi_eh_clear_ua(struct ata_device
*dev
)
3261 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
3262 u8
*sense_buffer
= dev
->link
->ap
->sector_buf
;
3264 unsigned int err_mask
;
3266 err_mask
= atapi_eh_tur(dev
, &sense_key
);
3267 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
3269 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3274 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
3277 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
3279 ata_dev_warn(dev
, "failed to clear "
3280 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
3285 ata_dev_warn(dev
, "UNIT ATTENTION persists after %d tries\n",
3292 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3293 * @dev: ATA device which may need FLUSH retry
3295 * If @dev failed FLUSH, it needs to be reported upper layer
3296 * immediately as it means that @dev failed to remap and already
3297 * lost at least a sector and further FLUSH retrials won't make
3298 * any difference to the lost sector. However, if FLUSH failed
3299 * for other reasons, for example transmission error, FLUSH needs
3302 * This function determines whether FLUSH failure retry is
3303 * necessary and performs it if so.
3306 * 0 if EH can continue, -errno if EH needs to be repeated.
3308 static int ata_eh_maybe_retry_flush(struct ata_device
*dev
)
3310 struct ata_link
*link
= dev
->link
;
3311 struct ata_port
*ap
= link
->ap
;
3312 struct ata_queued_cmd
*qc
;
3313 struct ata_taskfile tf
;
3314 unsigned int err_mask
;
3317 /* did flush fail for this device? */
3318 if (!ata_tag_valid(link
->active_tag
))
3321 qc
= __ata_qc_from_tag(ap
, link
->active_tag
);
3322 if (qc
->dev
!= dev
|| (qc
->tf
.command
!= ATA_CMD_FLUSH_EXT
&&
3323 qc
->tf
.command
!= ATA_CMD_FLUSH
))
3326 /* if the device failed it, it should be reported to upper layers */
3327 if (qc
->err_mask
& AC_ERR_DEV
)
3330 /* flush failed for some other reason, give it another shot */
3331 ata_tf_init(dev
, &tf
);
3333 tf
.command
= qc
->tf
.command
;
3334 tf
.flags
|= ATA_TFLAG_DEVICE
;
3335 tf
.protocol
= ATA_PROT_NODATA
;
3337 ata_dev_warn(dev
, "retrying FLUSH 0x%x Emask 0x%x\n",
3338 tf
.command
, qc
->err_mask
);
3340 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3343 * FLUSH is complete but there's no way to
3344 * successfully complete a failed command from EH.
3345 * Making sure retry is allowed at least once and
3346 * retrying it should do the trick - whatever was in
3347 * the cache is already on the platter and this won't
3348 * cause infinite loop.
3350 qc
->scsicmd
->allowed
= max(qc
->scsicmd
->allowed
, 1);
3352 ata_dev_warn(dev
, "FLUSH failed Emask 0x%x\n",
3356 /* if device failed it, report it to upper layers */
3357 if (err_mask
& AC_ERR_DEV
) {
3358 qc
->err_mask
|= AC_ERR_DEV
;
3360 if (!(ap
->pflags
& ATA_PFLAG_FROZEN
))
3368 * ata_eh_set_lpm - configure SATA interface power management
3369 * @link: link to configure power management
3370 * @policy: the link power management policy
3371 * @r_failed_dev: out parameter for failed device
3373 * Enable SATA Interface power management. This will enable
3374 * Device Interface Power Management (DIPM) for min_power
3375 * policy, and then call driver specific callbacks for
3376 * enabling Host Initiated Power management.
3382 * 0 on success, -errno on failure.
3384 static int ata_eh_set_lpm(struct ata_link
*link
, enum ata_lpm_policy policy
,
3385 struct ata_device
**r_failed_dev
)
3387 struct ata_port
*ap
= ata_is_host_link(link
) ? link
->ap
: NULL
;
3388 struct ata_eh_context
*ehc
= &link
->eh_context
;
3389 struct ata_device
*dev
, *link_dev
= NULL
, *lpm_dev
= NULL
;
3390 enum ata_lpm_policy old_policy
= link
->lpm_policy
;
3391 bool no_dipm
= link
->ap
->flags
& ATA_FLAG_NO_DIPM
;
3392 unsigned int hints
= ATA_LPM_EMPTY
| ATA_LPM_HIPM
;
3393 unsigned int err_mask
;
3396 /* if the link or host doesn't do LPM, noop */
3397 if ((link
->flags
& ATA_LFLAG_NO_LPM
) || (ap
&& !ap
->ops
->set_lpm
))
3401 * DIPM is enabled only for MIN_POWER as some devices
3402 * misbehave when the host NACKs transition to SLUMBER. Order
3403 * device and link configurations such that the host always
3404 * allows DIPM requests.
3406 ata_for_each_dev(dev
, link
, ENABLED
) {
3407 bool hipm
= ata_id_has_hipm(dev
->id
);
3408 bool dipm
= ata_id_has_dipm(dev
->id
) && !no_dipm
;
3410 /* find the first enabled and LPM enabled devices */
3414 if (!lpm_dev
&& (hipm
|| dipm
))
3417 hints
&= ~ATA_LPM_EMPTY
;
3419 hints
&= ~ATA_LPM_HIPM
;
3421 /* disable DIPM before changing link config */
3422 if (policy
!= ATA_LPM_MIN_POWER
&& dipm
) {
3423 err_mask
= ata_dev_set_feature(dev
,
3424 SETFEATURES_SATA_DISABLE
, SATA_DIPM
);
3425 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3427 "failed to disable DIPM, Emask 0x%x\n",
3436 rc
= ap
->ops
->set_lpm(link
, policy
, hints
);
3437 if (!rc
&& ap
->slave_link
)
3438 rc
= ap
->ops
->set_lpm(ap
->slave_link
, policy
, hints
);
3440 rc
= sata_pmp_set_lpm(link
, policy
, hints
);
3443 * Attribute link config failure to the first (LPM) enabled
3444 * device on the link.
3447 if (rc
== -EOPNOTSUPP
) {
3448 link
->flags
|= ATA_LFLAG_NO_LPM
;
3451 dev
= lpm_dev
? lpm_dev
: link_dev
;
3456 * Low level driver acked the transition. Issue DIPM command
3457 * with the new policy set.
3459 link
->lpm_policy
= policy
;
3460 if (ap
&& ap
->slave_link
)
3461 ap
->slave_link
->lpm_policy
= policy
;
3463 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3464 ata_for_each_dev(dev
, link
, ENABLED
) {
3465 if (policy
== ATA_LPM_MIN_POWER
&& !no_dipm
&&
3466 ata_id_has_dipm(dev
->id
)) {
3467 err_mask
= ata_dev_set_feature(dev
,
3468 SETFEATURES_SATA_ENABLE
, SATA_DIPM
);
3469 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3471 "failed to enable DIPM, Emask 0x%x\n",
3482 /* restore the old policy */
3483 link
->lpm_policy
= old_policy
;
3484 if (ap
&& ap
->slave_link
)
3485 ap
->slave_link
->lpm_policy
= old_policy
;
3487 /* if no device or only one more chance is left, disable LPM */
3488 if (!dev
|| ehc
->tries
[dev
->devno
] <= 2) {
3489 ata_link_warn(link
, "disabling LPM on the link\n");
3490 link
->flags
|= ATA_LFLAG_NO_LPM
;
3493 *r_failed_dev
= dev
;
3497 int ata_link_nr_enabled(struct ata_link
*link
)
3499 struct ata_device
*dev
;
3502 ata_for_each_dev(dev
, link
, ENABLED
)
3507 static int ata_link_nr_vacant(struct ata_link
*link
)
3509 struct ata_device
*dev
;
3512 ata_for_each_dev(dev
, link
, ALL
)
3513 if (dev
->class == ATA_DEV_UNKNOWN
)
3518 static int ata_eh_skip_recovery(struct ata_link
*link
)
3520 struct ata_port
*ap
= link
->ap
;
3521 struct ata_eh_context
*ehc
= &link
->eh_context
;
3522 struct ata_device
*dev
;
3524 /* skip disabled links */
3525 if (link
->flags
& ATA_LFLAG_DISABLED
)
3528 /* skip if explicitly requested */
3529 if (ehc
->i
.flags
& ATA_EHI_NO_RECOVERY
)
3532 /* thaw frozen port and recover failed devices */
3533 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
3536 /* reset at least once if reset is requested */
3537 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
3538 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
3541 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3542 ata_for_each_dev(dev
, link
, ALL
) {
3543 if (dev
->class == ATA_DEV_UNKNOWN
&&
3544 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
3551 static int ata_count_probe_trials_cb(struct ata_ering_entry
*ent
, void *void_arg
)
3553 u64 interval
= msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL
);
3554 u64 now
= get_jiffies_64();
3555 int *trials
= void_arg
;
3557 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) ||
3558 (ent
->timestamp
< now
- min(now
, interval
)))
3565 static int ata_eh_schedule_probe(struct ata_device
*dev
)
3567 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3568 struct ata_link
*link
= ata_dev_phys_link(dev
);
3571 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
3572 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
3575 ata_eh_detach_dev(dev
);
3577 ehc
->did_probe_mask
|= (1 << dev
->devno
);
3578 ehc
->i
.action
|= ATA_EH_RESET
;
3579 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
3580 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
3582 /* the link maybe in a deep sleep, wake it up */
3583 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
) {
3584 if (ata_is_host_link(link
))
3585 link
->ap
->ops
->set_lpm(link
, ATA_LPM_MAX_POWER
,
3588 sata_pmp_set_lpm(link
, ATA_LPM_MAX_POWER
,
3592 /* Record and count probe trials on the ering. The specific
3593 * error mask used is irrelevant. Because a successful device
3594 * detection clears the ering, this count accumulates only if
3595 * there are consecutive failed probes.
3597 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3598 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3599 * forced to 1.5Gbps.
3601 * This is to work around cases where failed link speed
3602 * negotiation results in device misdetection leading to
3603 * infinite DEVXCHG or PHRDY CHG events.
3605 ata_ering_record(&dev
->ering
, 0, AC_ERR_OTHER
);
3606 ata_ering_map(&dev
->ering
, ata_count_probe_trials_cb
, &trials
);
3608 if (trials
> ATA_EH_PROBE_TRIALS
)
3609 sata_down_spd_limit(link
, 1);
3614 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
3616 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3618 /* -EAGAIN from EH routine indicates retry without prejudice.
3619 * The requester is responsible for ensuring forward progress.
3622 ehc
->tries
[dev
->devno
]--;
3626 /* device missing or wrong IDENTIFY data, schedule probing */
3627 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
3629 /* give it just one more chance */
3630 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
3632 if (ehc
->tries
[dev
->devno
] == 1) {
3633 /* This is the last chance, better to slow
3634 * down than lose it.
3636 sata_down_spd_limit(ata_dev_phys_link(dev
), 0);
3637 if (dev
->pio_mode
> XFER_PIO_0
)
3638 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
3642 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
3643 /* disable device if it has used up all its chances */
3644 ata_dev_disable(dev
);
3646 /* detach if offline */
3647 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
3648 ata_eh_detach_dev(dev
);
3650 /* schedule probe if necessary */
3651 if (ata_eh_schedule_probe(dev
)) {
3652 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3653 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3654 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3659 ehc
->i
.action
|= ATA_EH_RESET
;
3665 * ata_eh_recover - recover host port after error
3666 * @ap: host port to recover
3667 * @prereset: prereset method (can be NULL)
3668 * @softreset: softreset method (can be NULL)
3669 * @hardreset: hardreset method (can be NULL)
3670 * @postreset: postreset method (can be NULL)
3671 * @r_failed_link: out parameter for failed link
3673 * This is the alpha and omega, eum and yang, heart and soul of
3674 * libata exception handling. On entry, actions required to
3675 * recover each link and hotplug requests are recorded in the
3676 * link's eh_context. This function executes all the operations
3677 * with appropriate retrials and fallbacks to resurrect failed
3678 * devices, detach goners and greet newcomers.
3681 * Kernel thread context (may sleep).
3684 * 0 on success, -errno on failure.
3686 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3687 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3688 ata_postreset_fn_t postreset
,
3689 struct ata_link
**r_failed_link
)
3691 struct ata_link
*link
;
3692 struct ata_device
*dev
;
3694 unsigned long flags
, deadline
;
3698 /* prep for recovery */
3699 ata_for_each_link(link
, ap
, EDGE
) {
3700 struct ata_eh_context
*ehc
= &link
->eh_context
;
3702 /* re-enable link? */
3703 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3704 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3705 spin_lock_irqsave(ap
->lock
, flags
);
3706 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3707 spin_unlock_irqrestore(ap
->lock
, flags
);
3708 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3711 ata_for_each_dev(dev
, link
, ALL
) {
3712 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3713 ehc
->tries
[dev
->devno
] = 1;
3715 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3717 /* collect port action mask recorded in dev actions */
3718 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3719 ~ATA_EH_PERDEV_MASK
;
3720 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3722 /* process hotplug request */
3723 if (dev
->flags
& ATA_DFLAG_DETACH
)
3724 ata_eh_detach_dev(dev
);
3726 /* schedule probe if necessary */
3727 if (!ata_dev_enabled(dev
))
3728 ata_eh_schedule_probe(dev
);
3735 /* if UNLOADING, finish immediately */
3736 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3740 ata_for_each_link(link
, ap
, EDGE
) {
3741 struct ata_eh_context
*ehc
= &link
->eh_context
;
3743 /* skip EH if possible. */
3744 if (ata_eh_skip_recovery(link
))
3747 ata_for_each_dev(dev
, link
, ALL
)
3748 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3752 ata_for_each_link(link
, ap
, EDGE
) {
3753 struct ata_eh_context
*ehc
= &link
->eh_context
;
3755 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3758 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3759 prereset
, softreset
, hardreset
, postreset
);
3761 ata_link_err(link
, "reset failed, giving up\n");
3770 * clears ATA_EH_PARK in eh_info and resets
3771 * ap->park_req_pending
3773 ata_eh_pull_park_action(ap
);
3776 ata_for_each_link(link
, ap
, EDGE
) {
3777 ata_for_each_dev(dev
, link
, ALL
) {
3778 struct ata_eh_context
*ehc
= &link
->eh_context
;
3781 if (dev
->class != ATA_DEV_ATA
)
3783 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3786 tmp
= dev
->unpark_deadline
;
3787 if (time_before(deadline
, tmp
))
3789 else if (time_before_eq(tmp
, jiffies
))
3791 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3794 ata_eh_park_issue_cmd(dev
, 1);
3799 if (time_before_eq(deadline
, now
))
3803 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3807 ata_for_each_link(link
, ap
, EDGE
) {
3808 ata_for_each_dev(dev
, link
, ALL
) {
3809 if (!(link
->eh_context
.unloaded_mask
&
3813 ata_eh_park_issue_cmd(dev
, 0);
3814 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3820 ata_for_each_link(link
, ap
, PMP_FIRST
) {
3821 struct ata_eh_context
*ehc
= &link
->eh_context
;
3823 if (sata_pmp_attached(ap
) && ata_is_host_link(link
))
3826 /* revalidate existing devices and attach new ones */
3827 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3831 /* if PMP got attached, return, pmp EH will take care of it */
3832 if (link
->device
->class == ATA_DEV_PMP
) {
3837 /* configure transfer mode if necessary */
3838 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3839 rc
= ata_set_mode(link
, &dev
);
3842 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3845 /* If reset has been issued, clear UA to avoid
3846 * disrupting the current users of the device.
3848 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3849 ata_for_each_dev(dev
, link
, ALL
) {
3850 if (dev
->class != ATA_DEV_ATAPI
)
3852 rc
= atapi_eh_clear_ua(dev
);
3858 /* retry flush if necessary */
3859 ata_for_each_dev(dev
, link
, ALL
) {
3860 if (dev
->class != ATA_DEV_ATA
)
3862 rc
= ata_eh_maybe_retry_flush(dev
);
3868 /* configure link power saving */
3869 if (link
->lpm_policy
!= ap
->target_lpm_policy
) {
3870 rc
= ata_eh_set_lpm(link
, ap
->target_lpm_policy
, &dev
);
3875 /* this link is okay now */
3882 ata_eh_handle_dev_fail(dev
, rc
);
3884 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3885 /* PMP reset requires working host port.
3886 * Can't retry if it's frozen.
3888 if (sata_pmp_attached(ap
))
3898 if (rc
&& r_failed_link
)
3899 *r_failed_link
= link
;
3901 DPRINTK("EXIT, rc=%d\n", rc
);
3906 * ata_eh_finish - finish up EH
3907 * @ap: host port to finish EH for
3909 * Recovery is complete. Clean up EH states and retry or finish
3915 void ata_eh_finish(struct ata_port
*ap
)
3919 /* retry or finish qcs */
3920 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3921 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3923 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3927 /* FIXME: Once EH migration is complete,
3928 * generate sense data in this function,
3929 * considering both err_mask and tf.
3931 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3932 ata_eh_qc_retry(qc
);
3934 ata_eh_qc_complete(qc
);
3936 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3937 ata_eh_qc_complete(qc
);
3939 /* feed zero TF to sense generation */
3940 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3941 ata_eh_qc_retry(qc
);
3946 /* make sure nr_active_links is zero after EH */
3947 WARN_ON(ap
->nr_active_links
);
3948 ap
->nr_active_links
= 0;
3952 * ata_do_eh - do standard error handling
3953 * @ap: host port to handle error for
3955 * @prereset: prereset method (can be NULL)
3956 * @softreset: softreset method (can be NULL)
3957 * @hardreset: hardreset method (can be NULL)
3958 * @postreset: postreset method (can be NULL)
3960 * Perform standard error handling sequence.
3963 * Kernel thread context (may sleep).
3965 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3966 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3967 ata_postreset_fn_t postreset
)
3969 struct ata_device
*dev
;
3975 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3978 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3979 ata_dev_disable(dev
);
3986 * ata_std_error_handler - standard error handler
3987 * @ap: host port to handle error for
3989 * Standard error handler
3992 * Kernel thread context (may sleep).
3994 void ata_std_error_handler(struct ata_port
*ap
)
3996 struct ata_port_operations
*ops
= ap
->ops
;
3997 ata_reset_fn_t hardreset
= ops
->hardreset
;
3999 /* ignore built-in hardreset if SCR access is not available */
4000 if (hardreset
== sata_std_hardreset
&& !sata_scr_valid(&ap
->link
))
4003 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
4008 * ata_eh_handle_port_suspend - perform port suspend operation
4009 * @ap: port to suspend
4014 * Kernel thread context (may sleep).
4016 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
4018 unsigned long flags
;
4021 /* are we suspending? */
4022 spin_lock_irqsave(ap
->lock
, flags
);
4023 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4024 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
4025 spin_unlock_irqrestore(ap
->lock
, flags
);
4028 spin_unlock_irqrestore(ap
->lock
, flags
);
4030 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
4032 /* tell ACPI we're suspending */
4033 rc
= ata_acpi_on_suspend(ap
);
4038 ata_eh_freeze_port(ap
);
4040 if (ap
->ops
->port_suspend
)
4041 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
4043 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
4046 spin_lock_irqsave(ap
->lock
, flags
);
4048 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
4050 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
4051 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
4052 ata_port_schedule_eh(ap
);
4054 if (ap
->pm_result
) {
4055 *ap
->pm_result
= rc
;
4056 ap
->pm_result
= NULL
;
4059 spin_unlock_irqrestore(ap
->lock
, flags
);
4065 * ata_eh_handle_port_resume - perform port resume operation
4066 * @ap: port to resume
4071 * Kernel thread context (may sleep).
4073 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
4075 struct ata_link
*link
;
4076 struct ata_device
*dev
;
4077 unsigned long flags
;
4080 /* are we resuming? */
4081 spin_lock_irqsave(ap
->lock
, flags
);
4082 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4083 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
4084 spin_unlock_irqrestore(ap
->lock
, flags
);
4087 spin_unlock_irqrestore(ap
->lock
, flags
);
4089 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
4092 * Error timestamps are in jiffies which doesn't run while
4093 * suspended and PHY events during resume isn't too uncommon.
4094 * When the two are combined, it can lead to unnecessary speed
4095 * downs if the machine is suspended and resumed repeatedly.
4096 * Clear error history.
4098 ata_for_each_link(link
, ap
, HOST_FIRST
)
4099 ata_for_each_dev(dev
, link
, ALL
)
4100 ata_ering_clear(&dev
->ering
);
4102 ata_acpi_set_state(ap
, PMSG_ON
);
4104 if (ap
->ops
->port_resume
)
4105 rc
= ap
->ops
->port_resume(ap
);
4107 /* tell ACPI that we're resuming */
4108 ata_acpi_on_resume(ap
);
4111 spin_lock_irqsave(ap
->lock
, flags
);
4112 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
4113 if (ap
->pm_result
) {
4114 *ap
->pm_result
= rc
;
4115 ap
->pm_result
= NULL
;
4117 spin_unlock_irqrestore(ap
->lock
, flags
);
4119 #endif /* CONFIG_PM */