2 * libata-eh.c - libata error handling
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/driver-api/libata.rst
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>
49 #include <trace/events/libata.h>
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
60 ATA_EFLAG_IS_IO
= (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
62 ATA_EFLAG_OLD_ER
= (1 << 31),
64 /* error categories */
67 ATA_ECAT_TOUT_HSM
= 2,
69 ATA_ECAT_DUBIOUS_NONE
= 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
75 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN
= 5000,
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
85 ATA_EH_PRERESET_TIMEOUT
= 10000,
86 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL
= 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS
= 2,
95 /* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
102 static const unsigned long ata_eh_reset_timeouts
[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX
, /* > 1 min has elapsed, give up */
110 static const unsigned long ata_eh_identify_timeouts
[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
117 static const unsigned long ata_eh_flush_timeouts
[] = {
118 15000, /* be generous with flush */
120 30000, /* and even more generous */
124 static const unsigned long ata_eh_other_timeouts
[] = {
125 5000, /* same rationale as identify timeout */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
131 struct ata_eh_cmd_timeout_ent
{
133 const unsigned long *timeouts
;
136 /* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
148 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
149 static const struct ata_eh_cmd_timeout_ent
150 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
151 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
152 .timeouts
= ata_eh_identify_timeouts
, },
153 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
154 .timeouts
= ata_eh_other_timeouts
, },
155 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
156 .timeouts
= ata_eh_other_timeouts
, },
157 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
158 .timeouts
= ata_eh_other_timeouts
, },
159 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
160 .timeouts
= ata_eh_other_timeouts
, },
161 { .commands
= CMDS(ATA_CMD_FLUSH
, ATA_CMD_FLUSH_EXT
),
162 .timeouts
= ata_eh_flush_timeouts
},
166 static void __ata_port_freeze(struct ata_port
*ap
);
168 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
169 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
170 #else /* CONFIG_PM */
171 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
174 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
176 #endif /* CONFIG_PM */
178 static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
,
179 const char *fmt
, va_list args
)
181 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
182 ATA_EH_DESC_LEN
- ehi
->desc_len
,
187 * __ata_ehi_push_desc - push error description without adding separator
189 * @fmt: printf format string
191 * Format string according to @fmt and append it to @ehi->desc.
194 * spin_lock_irqsave(host lock)
196 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
201 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
206 * ata_ehi_push_desc - push error description with separator
208 * @fmt: printf format string
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
214 * spin_lock_irqsave(host lock)
216 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
221 __ata_ehi_push_desc(ehi
, ", ");
224 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
229 * ata_ehi_clear_desc - clean error description
235 * spin_lock_irqsave(host lock)
237 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
256 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
260 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
262 if (ap
->link
.eh_info
.desc_len
)
263 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
266 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
287 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
290 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
292 unsigned long long start
, len
;
294 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
296 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
299 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
300 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
303 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
305 ata_port_desc(ap
, "%s 0x%llx", name
,
306 start
+ (unsigned long long)offset
);
309 #endif /* CONFIG_PCI */
311 static int ata_lookup_timeout_table(u8 cmd
)
315 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
318 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
331 * Determine timeout for internal command @cmd for @dev.
337 * Determined timeout.
339 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
341 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
342 int ent
= ata_lookup_timeout_table(cmd
);
346 return ATA_EH_CMD_DFL_TIMEOUT
;
348 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
349 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
364 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
366 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
367 int ent
= ata_lookup_timeout_table(cmd
);
373 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
374 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
375 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
378 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
379 unsigned int err_mask
)
381 struct ata_ering_entry
*ent
;
386 ering
->cursor
%= ATA_ERING_SIZE
;
388 ent
= &ering
->ring
[ering
->cursor
];
389 ent
->eflags
= eflags
;
390 ent
->err_mask
= err_mask
;
391 ent
->timestamp
= get_jiffies_64();
394 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
396 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
403 int ata_ering_map(struct ata_ering
*ering
,
404 int (*map_fn
)(struct ata_ering_entry
*, void *),
408 struct ata_ering_entry
*ent
;
412 ent
= &ering
->ring
[idx
];
415 rc
= map_fn(ent
, arg
);
418 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
419 } while (idx
!= ering
->cursor
);
424 static int ata_ering_clear_cb(struct ata_ering_entry
*ent
, void *void_arg
)
426 ent
->eflags
|= ATA_EFLAG_OLD_ER
;
430 static void ata_ering_clear(struct ata_ering
*ering
)
432 ata_ering_map(ering
, ata_ering_clear_cb
, NULL
);
435 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
437 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
439 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
442 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
443 struct ata_eh_info
*ehi
, unsigned int action
)
445 struct ata_device
*tdev
;
448 ehi
->action
&= ~action
;
449 ata_for_each_dev(tdev
, link
, ALL
)
450 ehi
->dev_action
[tdev
->devno
] &= ~action
;
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
455 /* break ehi->action into ehi->dev_action */
456 if (ehi
->action
& action
) {
457 ata_for_each_dev(tdev
, link
, ALL
)
458 ehi
->dev_action
[tdev
->devno
] |=
459 ehi
->action
& action
;
460 ehi
->action
&= ~action
;
463 /* turn off the specified per-dev action */
464 ehi
->dev_action
[dev
->devno
] &= ~action
;
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
479 void ata_eh_acquire(struct ata_port
*ap
)
481 mutex_lock(&ap
->host
->eh_mutex
);
482 WARN_ON_ONCE(ap
->host
->eh_owner
);
483 ap
->host
->eh_owner
= current
;
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
496 void ata_eh_release(struct ata_port
*ap
)
498 WARN_ON_ONCE(ap
->host
->eh_owner
!= current
);
499 ap
->host
->eh_owner
= NULL
;
500 mutex_unlock(&ap
->host
->eh_mutex
);
504 * ata_scsi_timed_out - SCSI layer time out callback
505 * @cmd: timed out SCSI command
507 * Handles SCSI layer timeout. We race with normal completion of
508 * the qc for @cmd. If the qc is already gone, we lose and let
509 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
510 * timed out and EH should be invoked. Prevent ata_qc_complete()
511 * from finishing it by setting EH_SCHEDULED and return
514 * TODO: kill this function once old EH is gone.
517 * Called from timer context
520 * EH_HANDLED or EH_NOT_HANDLED
522 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
524 struct Scsi_Host
*host
= cmd
->device
->host
;
525 struct ata_port
*ap
= ata_shost_to_port(host
);
527 struct ata_queued_cmd
*qc
;
528 enum blk_eh_timer_return ret
;
532 if (ap
->ops
->error_handler
) {
533 ret
= BLK_EH_NOT_HANDLED
;
537 ret
= BLK_EH_HANDLED
;
538 spin_lock_irqsave(ap
->lock
, flags
);
539 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
541 WARN_ON(qc
->scsicmd
!= cmd
);
542 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
543 qc
->err_mask
|= AC_ERR_TIMEOUT
;
544 ret
= BLK_EH_NOT_HANDLED
;
546 spin_unlock_irqrestore(ap
->lock
, flags
);
549 DPRINTK("EXIT, ret=%d\n", ret
);
552 EXPORT_SYMBOL(ata_scsi_timed_out
);
554 static void ata_eh_unload(struct ata_port
*ap
)
556 struct ata_link
*link
;
557 struct ata_device
*dev
;
560 /* Restore SControl IPM and SPD for the next driver and
561 * disable attached devices.
563 ata_for_each_link(link
, ap
, PMP_FIRST
) {
564 sata_scr_write(link
, SCR_CONTROL
, link
->saved_scontrol
& 0xff0);
565 ata_for_each_dev(dev
, link
, ALL
)
566 ata_dev_disable(dev
);
569 /* freeze and set UNLOADED */
570 spin_lock_irqsave(ap
->lock
, flags
);
572 ata_port_freeze(ap
); /* won't be thawed */
573 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
; /* clear pending from freeze */
574 ap
->pflags
|= ATA_PFLAG_UNLOADED
;
576 spin_unlock_irqrestore(ap
->lock
, flags
);
580 * ata_scsi_error - SCSI layer error handler callback
581 * @host: SCSI host on which error occurred
583 * Handles SCSI-layer-thrown error events.
586 * Inherited from SCSI layer (none, can sleep)
591 void ata_scsi_error(struct Scsi_Host
*host
)
593 struct ata_port
*ap
= ata_shost_to_port(host
);
595 LIST_HEAD(eh_work_q
);
599 spin_lock_irqsave(host
->host_lock
, flags
);
600 list_splice_init(&host
->eh_cmd_q
, &eh_work_q
);
601 spin_unlock_irqrestore(host
->host_lock
, flags
);
603 ata_scsi_cmd_error_handler(host
, ap
, &eh_work_q
);
605 /* If we timed raced normal completion and there is nothing to
606 recover nr_timedout == 0 why exactly are we doing error recovery ? */
607 ata_scsi_port_error_handler(host
, ap
);
609 /* finish or retry handled scmd's and clean up */
610 WARN_ON(!list_empty(&eh_work_q
));
616 * ata_scsi_cmd_error_handler - error callback for a list of commands
617 * @host: scsi host containing the port
618 * @ap: ATA port within the host
619 * @eh_work_q: list of commands to process
621 * process the given list of commands and return those finished to the
622 * ap->eh_done_q. This function is the first part of the libata error
623 * handler which processes a given list of failed commands.
625 void ata_scsi_cmd_error_handler(struct Scsi_Host
*host
, struct ata_port
*ap
,
626 struct list_head
*eh_work_q
)
631 /* make sure sff pio task is not running */
632 ata_sff_flush_pio_task(ap
);
634 /* synchronize with host lock and sort out timeouts */
636 /* For new EH, all qcs are finished in one of three ways -
637 * normal completion, error completion, and SCSI timeout.
638 * Both completions can race against SCSI timeout. When normal
639 * completion wins, the qc never reaches EH. When error
640 * completion wins, the qc has ATA_QCFLAG_FAILED set.
642 * When SCSI timeout wins, things are a bit more complex.
643 * Normal or error completion can occur after the timeout but
644 * before this point. In such cases, both types of
645 * completions are honored. A scmd is determined to have
646 * timed out iff its associated qc is active and not failed.
648 spin_lock_irqsave(ap
->lock
, flags
);
649 if (ap
->ops
->error_handler
) {
650 struct scsi_cmnd
*scmd
, *tmp
;
653 /* This must occur under the ap->lock as we don't want
654 a polled recovery to race the real interrupt handler
656 The lost_interrupt handler checks for any completed but
657 non-notified command and completes much like an IRQ handler.
659 We then fall into the error recovery code which will treat
660 this as if normal completion won the race */
662 if (ap
->ops
->lost_interrupt
)
663 ap
->ops
->lost_interrupt(ap
);
665 list_for_each_entry_safe(scmd
, tmp
, eh_work_q
, eh_entry
) {
666 struct ata_queued_cmd
*qc
;
668 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
669 qc
= __ata_qc_from_tag(ap
, i
);
670 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
675 if (i
< ATA_MAX_QUEUE
) {
676 /* the scmd has an associated qc */
677 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
678 /* which hasn't failed yet, timeout */
679 qc
->err_mask
|= AC_ERR_TIMEOUT
;
680 qc
->flags
|= ATA_QCFLAG_FAILED
;
684 /* Normal completion occurred after
685 * SCSI timeout but before this point.
686 * Successfully complete it.
688 scmd
->retries
= scmd
->allowed
;
689 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
693 /* If we have timed out qcs. They belong to EH from
694 * this point but the state of the controller is
695 * unknown. Freeze the port to make sure the IRQ
696 * handler doesn't diddle with those qcs. This must
697 * be done atomically w.r.t. setting QCFLAG_FAILED.
700 __ata_port_freeze(ap
);
703 /* initialize eh_tries */
704 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
706 spin_unlock_irqrestore(ap
->lock
, flags
);
709 EXPORT_SYMBOL(ata_scsi_cmd_error_handler
);
712 * ata_scsi_port_error_handler - recover the port after the commands
713 * @host: SCSI host containing the port
716 * Handle the recovery of the port @ap after all the commands
717 * have been recovered.
719 void ata_scsi_port_error_handler(struct Scsi_Host
*host
, struct ata_port
*ap
)
723 /* invoke error handler */
724 if (ap
->ops
->error_handler
) {
725 struct ata_link
*link
;
727 /* acquire EH ownership */
730 /* kill fast drain timer */
731 del_timer_sync(&ap
->fastdrain_timer
);
733 /* process port resume request */
734 ata_eh_handle_port_resume(ap
);
736 /* fetch & clear EH info */
737 spin_lock_irqsave(ap
->lock
, flags
);
739 ata_for_each_link(link
, ap
, HOST_FIRST
) {
740 struct ata_eh_context
*ehc
= &link
->eh_context
;
741 struct ata_device
*dev
;
743 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
744 link
->eh_context
.i
= link
->eh_info
;
745 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
747 ata_for_each_dev(dev
, link
, ENABLED
) {
748 int devno
= dev
->devno
;
750 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
751 if (ata_ncq_enabled(dev
))
752 ehc
->saved_ncq_enabled
|= 1 << devno
;
756 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
757 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
758 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
760 spin_unlock_irqrestore(ap
->lock
, flags
);
762 /* invoke EH, skip if unloading or suspended */
763 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
764 ap
->ops
->error_handler(ap
);
766 /* if unloading, commence suicide */
767 if ((ap
->pflags
& ATA_PFLAG_UNLOADING
) &&
768 !(ap
->pflags
& ATA_PFLAG_UNLOADED
))
773 /* process port suspend request */
774 ata_eh_handle_port_suspend(ap
);
776 /* Exception might have happened after ->error_handler
777 * recovered the port but before this point. Repeat
780 spin_lock_irqsave(ap
->lock
, flags
);
782 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
783 if (--ap
->eh_tries
) {
784 spin_unlock_irqrestore(ap
->lock
, flags
);
788 "EH pending after %d tries, giving up\n",
790 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
793 /* this run is complete, make sure EH info is clear */
794 ata_for_each_link(link
, ap
, HOST_FIRST
)
795 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
797 /* end eh (clear host_eh_scheduled) while holding
798 * ap->lock such that if exception occurs after this
799 * point but before EH completion, SCSI midlayer will
804 spin_unlock_irqrestore(ap
->lock
, flags
);
807 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
808 ap
->ops
->eng_timeout(ap
);
811 scsi_eh_flush_done_q(&ap
->eh_done_q
);
814 spin_lock_irqsave(ap
->lock
, flags
);
816 if (ap
->pflags
& ATA_PFLAG_LOADING
)
817 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
818 else if ((ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
) &&
819 !(ap
->flags
& ATA_FLAG_SAS_HOST
))
820 schedule_delayed_work(&ap
->hotplug_task
, 0);
822 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
823 ata_port_info(ap
, "EH complete\n");
825 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
827 /* tell wait_eh that we're done */
828 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
829 wake_up_all(&ap
->eh_wait_q
);
831 spin_unlock_irqrestore(ap
->lock
, flags
);
833 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler
);
836 * ata_port_wait_eh - Wait for the currently pending EH to complete
837 * @ap: Port to wait EH for
839 * Wait until the currently pending EH is complete.
842 * Kernel thread context (may sleep).
844 void ata_port_wait_eh(struct ata_port
*ap
)
850 spin_lock_irqsave(ap
->lock
, flags
);
852 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
853 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
854 spin_unlock_irqrestore(ap
->lock
, flags
);
856 spin_lock_irqsave(ap
->lock
, flags
);
858 finish_wait(&ap
->eh_wait_q
, &wait
);
860 spin_unlock_irqrestore(ap
->lock
, flags
);
862 /* make sure SCSI EH is complete */
863 if (scsi_host_in_recovery(ap
->scsi_host
)) {
868 EXPORT_SYMBOL_GPL(ata_port_wait_eh
);
870 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
875 /* count only non-internal commands */
876 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
877 if (ata_qc_from_tag(ap
, tag
))
883 void ata_eh_fastdrain_timerfn(struct timer_list
*t
)
885 struct ata_port
*ap
= from_timer(ap
, t
, fastdrain_timer
);
889 spin_lock_irqsave(ap
->lock
, flags
);
891 cnt
= ata_eh_nr_in_flight(ap
);
897 if (cnt
== ap
->fastdrain_cnt
) {
900 /* No progress during the last interval, tag all
901 * in-flight qcs as timed out and freeze the port.
903 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
904 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
906 qc
->err_mask
|= AC_ERR_TIMEOUT
;
911 /* some qcs have finished, give it another chance */
912 ap
->fastdrain_cnt
= cnt
;
913 ap
->fastdrain_timer
.expires
=
914 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
915 add_timer(&ap
->fastdrain_timer
);
919 spin_unlock_irqrestore(ap
->lock
, flags
);
923 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
924 * @ap: target ATA port
925 * @fastdrain: activate fast drain
927 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
928 * is non-zero and EH wasn't pending before. Fast drain ensures
929 * that EH kicks in in timely manner.
932 * spin_lock_irqsave(host lock)
934 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
938 /* already scheduled? */
939 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
942 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
947 /* do we have in-flight qcs? */
948 cnt
= ata_eh_nr_in_flight(ap
);
952 /* activate fast drain */
953 ap
->fastdrain_cnt
= cnt
;
954 ap
->fastdrain_timer
.expires
=
955 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
956 add_timer(&ap
->fastdrain_timer
);
960 * ata_qc_schedule_eh - schedule qc for error handling
961 * @qc: command to schedule error handling for
963 * Schedule error handling for @qc. EH will kick in as soon as
964 * other commands are drained.
967 * spin_lock_irqsave(host lock)
969 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
971 struct ata_port
*ap
= qc
->ap
;
972 struct request_queue
*q
= qc
->scsicmd
->device
->request_queue
;
975 WARN_ON(!ap
->ops
->error_handler
);
977 qc
->flags
|= ATA_QCFLAG_FAILED
;
978 ata_eh_set_pending(ap
, 1);
980 /* The following will fail if timeout has already expired.
981 * ata_scsi_error() takes care of such scmds on EH entry.
982 * Note that ATA_QCFLAG_FAILED is unconditionally set after
983 * this function completes.
985 spin_lock_irqsave(q
->queue_lock
, flags
);
986 blk_abort_request(qc
->scsicmd
->request
);
987 spin_unlock_irqrestore(q
->queue_lock
, flags
);
991 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
992 * @ap: ATA port to schedule EH for
994 * LOCKING: inherited from ata_port_schedule_eh
995 * spin_lock_irqsave(host lock)
997 void ata_std_sched_eh(struct ata_port
*ap
)
999 WARN_ON(!ap
->ops
->error_handler
);
1001 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
1004 ata_eh_set_pending(ap
, 1);
1005 scsi_schedule_eh(ap
->scsi_host
);
1007 DPRINTK("port EH scheduled\n");
1009 EXPORT_SYMBOL_GPL(ata_std_sched_eh
);
1012 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1013 * @ap: ATA port to end EH for
1015 * In the libata object model there is a 1:1 mapping of ata_port to
1016 * shost, so host fields can be directly manipulated under ap->lock, in
1017 * the libsas case we need to hold a lock at the ha->level to coordinate
1021 * spin_lock_irqsave(host lock)
1023 void ata_std_end_eh(struct ata_port
*ap
)
1025 struct Scsi_Host
*host
= ap
->scsi_host
;
1027 host
->host_eh_scheduled
= 0;
1029 EXPORT_SYMBOL(ata_std_end_eh
);
1033 * ata_port_schedule_eh - schedule error handling without a qc
1034 * @ap: ATA port to schedule EH for
1036 * Schedule error handling for @ap. EH will kick in as soon as
1037 * all commands are drained.
1040 * spin_lock_irqsave(host lock)
1042 void ata_port_schedule_eh(struct ata_port
*ap
)
1044 /* see: ata_std_sched_eh, unless you know better */
1045 ap
->ops
->sched_eh(ap
);
1048 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
1050 int tag
, nr_aborted
= 0;
1052 WARN_ON(!ap
->ops
->error_handler
);
1054 /* we're gonna abort all commands, no need for fast drain */
1055 ata_eh_set_pending(ap
, 0);
1057 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1058 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
1060 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
1061 qc
->flags
|= ATA_QCFLAG_FAILED
;
1062 ata_qc_complete(qc
);
1068 ata_port_schedule_eh(ap
);
1074 * ata_link_abort - abort all qc's on the link
1075 * @link: ATA link to abort qc's for
1077 * Abort all active qc's active on @link and schedule EH.
1080 * spin_lock_irqsave(host lock)
1083 * Number of aborted qc's.
1085 int ata_link_abort(struct ata_link
*link
)
1087 return ata_do_link_abort(link
->ap
, link
);
1091 * ata_port_abort - abort all qc's on the port
1092 * @ap: ATA port to abort qc's for
1094 * Abort all active qc's of @ap and schedule EH.
1097 * spin_lock_irqsave(host_set lock)
1100 * Number of aborted qc's.
1102 int ata_port_abort(struct ata_port
*ap
)
1104 return ata_do_link_abort(ap
, NULL
);
1108 * __ata_port_freeze - freeze port
1109 * @ap: ATA port to freeze
1111 * This function is called when HSM violation or some other
1112 * condition disrupts normal operation of the port. Frozen port
1113 * is not allowed to perform any operation until the port is
1114 * thawed, which usually follows a successful reset.
1116 * ap->ops->freeze() callback can be used for freezing the port
1117 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1118 * port cannot be frozen hardware-wise, the interrupt handler
1119 * must ack and clear interrupts unconditionally while the port
1123 * spin_lock_irqsave(host lock)
1125 static void __ata_port_freeze(struct ata_port
*ap
)
1127 WARN_ON(!ap
->ops
->error_handler
);
1129 if (ap
->ops
->freeze
)
1130 ap
->ops
->freeze(ap
);
1132 ap
->pflags
|= ATA_PFLAG_FROZEN
;
1134 DPRINTK("ata%u port frozen\n", ap
->print_id
);
1138 * ata_port_freeze - abort & freeze port
1139 * @ap: ATA port to freeze
1141 * Abort and freeze @ap. The freeze operation must be called
1142 * first, because some hardware requires special operations
1143 * before the taskfile registers are accessible.
1146 * spin_lock_irqsave(host lock)
1149 * Number of aborted commands.
1151 int ata_port_freeze(struct ata_port
*ap
)
1155 WARN_ON(!ap
->ops
->error_handler
);
1157 __ata_port_freeze(ap
);
1158 nr_aborted
= ata_port_abort(ap
);
1164 * sata_async_notification - SATA async notification handler
1165 * @ap: ATA port where async notification is received
1167 * Handler to be called when async notification via SDB FIS is
1168 * received. This function schedules EH if necessary.
1171 * spin_lock_irqsave(host lock)
1174 * 1 if EH is scheduled, 0 otherwise.
1176 int sata_async_notification(struct ata_port
*ap
)
1181 if (!(ap
->flags
& ATA_FLAG_AN
))
1184 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
1186 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
1188 if (!sata_pmp_attached(ap
) || rc
) {
1189 /* PMP is not attached or SNTF is not available */
1190 if (!sata_pmp_attached(ap
)) {
1191 /* PMP is not attached. Check whether ATAPI
1192 * AN is configured. If so, notify media
1195 struct ata_device
*dev
= ap
->link
.device
;
1197 if ((dev
->class == ATA_DEV_ATAPI
) &&
1198 (dev
->flags
& ATA_DFLAG_AN
))
1199 ata_scsi_media_change_notify(dev
);
1202 /* PMP is attached but SNTF is not available.
1203 * ATAPI async media change notification is
1204 * not used. The PMP must be reporting PHY
1205 * status change, schedule EH.
1207 ata_port_schedule_eh(ap
);
1211 /* PMP is attached and SNTF is available */
1212 struct ata_link
*link
;
1214 /* check and notify ATAPI AN */
1215 ata_for_each_link(link
, ap
, EDGE
) {
1216 if (!(sntf
& (1 << link
->pmp
)))
1219 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1220 (link
->device
->flags
& ATA_DFLAG_AN
))
1221 ata_scsi_media_change_notify(link
->device
);
1224 /* If PMP is reporting that PHY status of some
1225 * downstream ports has changed, schedule EH.
1227 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1228 ata_port_schedule_eh(ap
);
1237 * ata_eh_freeze_port - EH helper to freeze port
1238 * @ap: ATA port to freeze
1245 void ata_eh_freeze_port(struct ata_port
*ap
)
1247 unsigned long flags
;
1249 if (!ap
->ops
->error_handler
)
1252 spin_lock_irqsave(ap
->lock
, flags
);
1253 __ata_port_freeze(ap
);
1254 spin_unlock_irqrestore(ap
->lock
, flags
);
1258 * ata_port_thaw_port - EH helper to thaw port
1259 * @ap: ATA port to thaw
1261 * Thaw frozen port @ap.
1266 void ata_eh_thaw_port(struct ata_port
*ap
)
1268 unsigned long flags
;
1270 if (!ap
->ops
->error_handler
)
1273 spin_lock_irqsave(ap
->lock
, flags
);
1275 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1280 spin_unlock_irqrestore(ap
->lock
, flags
);
1282 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1285 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1290 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1292 struct ata_port
*ap
= qc
->ap
;
1293 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1294 unsigned long flags
;
1296 spin_lock_irqsave(ap
->lock
, flags
);
1297 qc
->scsidone
= ata_eh_scsidone
;
1298 __ata_qc_complete(qc
);
1299 WARN_ON(ata_tag_valid(qc
->tag
));
1300 spin_unlock_irqrestore(ap
->lock
, flags
);
1302 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1306 * ata_eh_qc_complete - Complete an active ATA command from EH
1307 * @qc: Command to complete
1309 * Indicate to the mid and upper layers that an ATA command has
1310 * completed. To be used from EH.
1312 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1314 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1315 scmd
->retries
= scmd
->allowed
;
1316 __ata_eh_qc_complete(qc
);
1320 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1321 * @qc: Command to retry
1323 * Indicate to the mid and upper layers that an ATA command
1324 * should be retried. To be used from EH.
1326 * SCSI midlayer limits the number of retries to scmd->allowed.
1327 * scmd->allowed is incremented for commands which get retried
1328 * due to unrelated failures (qc->err_mask is zero).
1330 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1332 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1335 __ata_eh_qc_complete(qc
);
1339 * ata_dev_disable - disable ATA device
1340 * @dev: ATA device to disable
1347 void ata_dev_disable(struct ata_device
*dev
)
1349 if (!ata_dev_enabled(dev
))
1352 if (ata_msg_drv(dev
->link
->ap
))
1353 ata_dev_warn(dev
, "disabled\n");
1354 ata_acpi_on_disable(dev
);
1355 ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO0
| ATA_DNXFER_QUIET
);
1358 /* From now till the next successful probe, ering is used to
1359 * track probe failures. Clear accumulated device error info.
1361 ata_ering_clear(&dev
->ering
);
1365 * ata_eh_detach_dev - detach ATA device
1366 * @dev: ATA device to detach
1373 void ata_eh_detach_dev(struct ata_device
*dev
)
1375 struct ata_link
*link
= dev
->link
;
1376 struct ata_port
*ap
= link
->ap
;
1377 struct ata_eh_context
*ehc
= &link
->eh_context
;
1378 unsigned long flags
;
1380 ata_dev_disable(dev
);
1382 spin_lock_irqsave(ap
->lock
, flags
);
1384 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1386 if (ata_scsi_offline_dev(dev
)) {
1387 dev
->flags
|= ATA_DFLAG_DETACHED
;
1388 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1391 /* clear per-dev EH info */
1392 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1393 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1394 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1395 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1397 spin_unlock_irqrestore(ap
->lock
, flags
);
1401 * ata_eh_about_to_do - about to perform eh_action
1402 * @link: target ATA link
1403 * @dev: target ATA dev for per-dev action (can be NULL)
1404 * @action: action about to be performed
1406 * Called just before performing EH actions to clear related bits
1407 * in @link->eh_info such that eh actions are not unnecessarily
1413 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1414 unsigned int action
)
1416 struct ata_port
*ap
= link
->ap
;
1417 struct ata_eh_info
*ehi
= &link
->eh_info
;
1418 struct ata_eh_context
*ehc
= &link
->eh_context
;
1419 unsigned long flags
;
1421 spin_lock_irqsave(ap
->lock
, flags
);
1423 ata_eh_clear_action(link
, dev
, ehi
, action
);
1425 /* About to take EH action, set RECOVERED. Ignore actions on
1426 * slave links as master will do them again.
1428 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1429 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1431 spin_unlock_irqrestore(ap
->lock
, flags
);
1435 * ata_eh_done - EH action complete
1436 * @link: ATA link for which EH actions are complete
1437 * @dev: target ATA dev for per-dev action (can be NULL)
1438 * @action: action just completed
1440 * Called right after performing EH actions to clear related bits
1441 * in @link->eh_context.
1446 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1447 unsigned int action
)
1449 struct ata_eh_context
*ehc
= &link
->eh_context
;
1451 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1455 * ata_err_string - convert err_mask to descriptive string
1456 * @err_mask: error mask to convert to string
1458 * Convert @err_mask to descriptive string. Errors are
1459 * prioritized according to severity and only the most severe
1460 * error is reported.
1466 * Descriptive string for @err_mask
1468 static const char *ata_err_string(unsigned int err_mask
)
1470 if (err_mask
& AC_ERR_HOST_BUS
)
1471 return "host bus error";
1472 if (err_mask
& AC_ERR_ATA_BUS
)
1473 return "ATA bus error";
1474 if (err_mask
& AC_ERR_TIMEOUT
)
1476 if (err_mask
& AC_ERR_HSM
)
1477 return "HSM violation";
1478 if (err_mask
& AC_ERR_SYSTEM
)
1479 return "internal error";
1480 if (err_mask
& AC_ERR_MEDIA
)
1481 return "media error";
1482 if (err_mask
& AC_ERR_INVALID
)
1483 return "invalid argument";
1484 if (err_mask
& AC_ERR_DEV
)
1485 return "device error";
1486 return "unknown error";
1490 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1491 * @dev: Device to read log page 10h from
1492 * @tag: Resulting tag of the failed command
1493 * @tf: Resulting taskfile registers of the failed command
1495 * Read log page 10h to obtain NCQ error details and clear error
1499 * Kernel thread context (may sleep).
1502 * 0 on success, -errno otherwise.
1504 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1505 int *tag
, struct ata_taskfile
*tf
)
1507 u8
*buf
= dev
->link
->ap
->sector_buf
;
1508 unsigned int err_mask
;
1512 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, 0, buf
, 1);
1517 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1520 ata_dev_warn(dev
, "invalid checksum 0x%x on log page 10h\n",
1526 *tag
= buf
[0] & 0x1f;
1528 tf
->command
= buf
[2];
1529 tf
->feature
= buf
[3];
1533 tf
->device
= buf
[7];
1534 tf
->hob_lbal
= buf
[8];
1535 tf
->hob_lbam
= buf
[9];
1536 tf
->hob_lbah
= buf
[10];
1537 tf
->nsect
= buf
[12];
1538 tf
->hob_nsect
= buf
[13];
1539 if (ata_id_has_ncq_autosense(dev
->id
))
1540 tf
->auxiliary
= buf
[14] << 16 | buf
[15] << 8 | buf
[16];
1546 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1547 * @dev: target ATAPI device
1548 * @r_sense_key: out parameter for sense_key
1550 * Perform ATAPI TEST_UNIT_READY.
1553 * EH context (may sleep).
1556 * 0 on success, AC_ERR_* mask on failure.
1558 unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1560 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1561 struct ata_taskfile tf
;
1562 unsigned int err_mask
;
1564 ata_tf_init(dev
, &tf
);
1566 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1567 tf
.command
= ATA_CMD_PACKET
;
1568 tf
.protocol
= ATAPI_PROT_NODATA
;
1570 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1571 if (err_mask
== AC_ERR_DEV
)
1572 *r_sense_key
= tf
.feature
>> 4;
1577 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1578 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1579 * @cmd: scsi command for which the sense code should be set
1581 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1582 * SENSE. This function is an EH helper.
1585 * Kernel thread context (may sleep).
1587 static void ata_eh_request_sense(struct ata_queued_cmd
*qc
,
1588 struct scsi_cmnd
*cmd
)
1590 struct ata_device
*dev
= qc
->dev
;
1591 struct ata_taskfile tf
;
1592 unsigned int err_mask
;
1594 if (qc
->ap
->pflags
& ATA_PFLAG_FROZEN
) {
1595 ata_dev_warn(dev
, "sense data available but port frozen\n");
1599 if (!cmd
|| qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
1602 if (!ata_id_sense_reporting_enabled(dev
->id
)) {
1603 ata_dev_warn(qc
->dev
, "sense data reporting disabled\n");
1607 DPRINTK("ATA request sense\n");
1609 ata_tf_init(dev
, &tf
);
1610 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1611 tf
.flags
|= ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1612 tf
.command
= ATA_CMD_REQ_SENSE_DATA
;
1613 tf
.protocol
= ATA_PROT_NODATA
;
1615 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
1616 /* Ignore err_mask; ATA_ERR might be set */
1617 if (tf
.command
& ATA_SENSE
) {
1618 ata_scsi_set_sense(dev
, cmd
, tf
.lbah
, tf
.lbam
, tf
.lbal
);
1619 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1621 ata_dev_warn(dev
, "request sense failed stat %02x emask %x\n",
1622 tf
.command
, err_mask
);
1627 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1628 * @dev: device to perform REQUEST_SENSE to
1629 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1630 * @dfl_sense_key: default sense key to use
1632 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1633 * SENSE. This function is EH helper.
1636 * Kernel thread context (may sleep).
1639 * 0 on success, AC_ERR_* mask on failure
1641 unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1642 u8
*sense_buf
, u8 dfl_sense_key
)
1644 u8 cdb
[ATAPI_CDB_LEN
] =
1645 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1646 struct ata_port
*ap
= dev
->link
->ap
;
1647 struct ata_taskfile tf
;
1649 DPRINTK("ATAPI request sense\n");
1651 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1653 /* initialize sense_buf with the error register,
1654 * for the case where they are -not- overwritten
1656 sense_buf
[0] = 0x70;
1657 sense_buf
[2] = dfl_sense_key
;
1659 /* some devices time out if garbage left in tf */
1660 ata_tf_init(dev
, &tf
);
1662 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1663 tf
.command
= ATA_CMD_PACKET
;
1665 /* is it pointless to prefer PIO for "safety reasons"? */
1666 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1667 tf
.protocol
= ATAPI_PROT_DMA
;
1668 tf
.feature
|= ATAPI_PKT_DMA
;
1670 tf
.protocol
= ATAPI_PROT_PIO
;
1671 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1675 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1676 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1680 * ata_eh_analyze_serror - analyze SError for a failed port
1681 * @link: ATA link to analyze SError for
1683 * Analyze SError if available and further determine cause of
1689 static void ata_eh_analyze_serror(struct ata_link
*link
)
1691 struct ata_eh_context
*ehc
= &link
->eh_context
;
1692 u32 serror
= ehc
->i
.serror
;
1693 unsigned int err_mask
= 0, action
= 0;
1696 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1697 err_mask
|= AC_ERR_ATA_BUS
;
1698 action
|= ATA_EH_RESET
;
1700 if (serror
& SERR_PROTOCOL
) {
1701 err_mask
|= AC_ERR_HSM
;
1702 action
|= ATA_EH_RESET
;
1704 if (serror
& SERR_INTERNAL
) {
1705 err_mask
|= AC_ERR_SYSTEM
;
1706 action
|= ATA_EH_RESET
;
1709 /* Determine whether a hotplug event has occurred. Both
1710 * SError.N/X are considered hotplug events for enabled or
1711 * host links. For disabled PMP links, only N bit is
1712 * considered as X bit is left at 1 for link plugging.
1714 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
)
1715 hotplug_mask
= 0; /* hotplug doesn't work w/ LPM */
1716 else if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1717 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1719 hotplug_mask
= SERR_PHYRDY_CHG
;
1721 if (serror
& hotplug_mask
)
1722 ata_ehi_hotplugged(&ehc
->i
);
1724 ehc
->i
.err_mask
|= err_mask
;
1725 ehc
->i
.action
|= action
;
1729 * ata_eh_analyze_ncq_error - analyze NCQ error
1730 * @link: ATA link to analyze NCQ error for
1732 * Read log page 10h, determine the offending qc and acquire
1733 * error status TF. For NCQ device errors, all LLDDs have to do
1734 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1738 * Kernel thread context (may sleep).
1740 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1742 struct ata_port
*ap
= link
->ap
;
1743 struct ata_eh_context
*ehc
= &link
->eh_context
;
1744 struct ata_device
*dev
= link
->device
;
1745 struct ata_queued_cmd
*qc
;
1746 struct ata_taskfile tf
;
1749 /* if frozen, we can't do much */
1750 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1753 /* is it NCQ device error? */
1754 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1757 /* has LLDD analyzed already? */
1758 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1759 qc
= __ata_qc_from_tag(ap
, tag
);
1761 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1768 /* okay, this error is ours */
1769 memset(&tf
, 0, sizeof(tf
));
1770 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1772 ata_link_err(link
, "failed to read log page 10h (errno=%d)\n",
1777 if (!(link
->sactive
& (1 << tag
))) {
1778 ata_link_err(link
, "log page 10h reported inactive tag %d\n",
1783 /* we've got the perpetrator, condemn it */
1784 qc
= __ata_qc_from_tag(ap
, tag
);
1785 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1786 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1787 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1788 if ((qc
->result_tf
.command
& ATA_SENSE
) || qc
->result_tf
.auxiliary
) {
1789 char sense_key
, asc
, ascq
;
1791 sense_key
= (qc
->result_tf
.auxiliary
>> 16) & 0xff;
1792 asc
= (qc
->result_tf
.auxiliary
>> 8) & 0xff;
1793 ascq
= qc
->result_tf
.auxiliary
& 0xff;
1794 ata_scsi_set_sense(dev
, qc
->scsicmd
, sense_key
, asc
, ascq
);
1795 ata_scsi_set_sense_information(dev
, qc
->scsicmd
,
1797 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1800 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1804 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1805 * @qc: qc to analyze
1806 * @tf: Taskfile registers to analyze
1808 * Analyze taskfile of @qc and further determine cause of
1809 * failure. This function also requests ATAPI sense data if
1813 * Kernel thread context (may sleep).
1816 * Determined recovery action
1818 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1819 const struct ata_taskfile
*tf
)
1821 unsigned int tmp
, action
= 0;
1822 u8 stat
= tf
->command
, err
= tf
->feature
;
1824 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1825 qc
->err_mask
|= AC_ERR_HSM
;
1826 return ATA_EH_RESET
;
1829 if (stat
& (ATA_ERR
| ATA_DF
)) {
1830 qc
->err_mask
|= AC_ERR_DEV
;
1832 * Sense data reporting does not work if the
1833 * device fault bit is set.
1841 switch (qc
->dev
->class) {
1844 if (stat
& ATA_SENSE
)
1845 ata_eh_request_sense(qc
, qc
->scsicmd
);
1847 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1848 if (err
& (ATA_UNC
| ATA_AMNF
))
1849 qc
->err_mask
|= AC_ERR_MEDIA
;
1851 qc
->err_mask
|= AC_ERR_INVALID
;
1855 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1856 tmp
= atapi_eh_request_sense(qc
->dev
,
1857 qc
->scsicmd
->sense_buffer
,
1858 qc
->result_tf
.feature
>> 4);
1860 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1862 qc
->err_mask
|= tmp
;
1866 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
1867 int ret
= scsi_check_sense(qc
->scsicmd
);
1869 * SUCCESS here means that the sense code could
1870 * evaluated and should be passed to the upper layers
1871 * for correct evaluation.
1872 * FAILED means the sense code could not interpreted
1873 * and the device would need to be reset.
1874 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1875 * command would need to be retried.
1877 if (ret
== NEEDS_RETRY
|| ret
== ADD_TO_MLQUEUE
) {
1878 qc
->flags
|= ATA_QCFLAG_RETRY
;
1879 qc
->err_mask
|= AC_ERR_OTHER
;
1880 } else if (ret
!= SUCCESS
) {
1881 qc
->err_mask
|= AC_ERR_HSM
;
1884 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1885 action
|= ATA_EH_RESET
;
1890 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1895 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1899 base
= ATA_ECAT_DUBIOUS_NONE
;
1901 if (err_mask
& AC_ERR_ATA_BUS
)
1902 return base
+ ATA_ECAT_ATA_BUS
;
1904 if (err_mask
& AC_ERR_TIMEOUT
)
1905 return base
+ ATA_ECAT_TOUT_HSM
;
1907 if (eflags
& ATA_EFLAG_IS_IO
) {
1908 if (err_mask
& AC_ERR_HSM
)
1909 return base
+ ATA_ECAT_TOUT_HSM
;
1911 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1912 return base
+ ATA_ECAT_UNK_DEV
;
1918 struct speed_down_verdict_arg
{
1921 int nr_errors
[ATA_ECAT_NR
];
1924 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1926 struct speed_down_verdict_arg
*arg
= void_arg
;
1929 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) || (ent
->timestamp
< arg
->since
))
1932 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1934 arg
->nr_errors
[cat
]++;
1940 * ata_eh_speed_down_verdict - Determine speed down verdict
1941 * @dev: Device of interest
1943 * This function examines error ring of @dev and determines
1944 * whether NCQ needs to be turned off, transfer speed should be
1945 * stepped down, or falling back to PIO is necessary.
1947 * ECAT_ATA_BUS : ATA_BUS error for any command
1949 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1952 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1954 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1955 * data transfer hasn't been verified.
1959 * NCQ_OFF : Turn off NCQ.
1961 * SPEED_DOWN : Speed down transfer speed but don't fall back
1964 * FALLBACK_TO_PIO : Fall back to PIO.
1966 * Even if multiple verdicts are returned, only one action is
1967 * taken per error. An action triggered by non-DUBIOUS errors
1968 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1969 * This is to expedite speed down decisions right after device is
1970 * initially configured.
1972 * The following are speed down rules. #1 and #2 deal with
1975 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1976 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1978 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1979 * occurred during last 5 mins, NCQ_OFF.
1981 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1982 * occurred during last 5 mins, FALLBACK_TO_PIO
1984 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1985 * during last 10 mins, NCQ_OFF.
1987 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1988 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1991 * Inherited from caller.
1994 * OR of ATA_EH_SPDN_* flags.
1996 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1998 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1999 u64 j64
= get_jiffies_64();
2000 struct speed_down_verdict_arg arg
;
2001 unsigned int verdict
= 0;
2003 /* scan past 5 mins of error history */
2004 memset(&arg
, 0, sizeof(arg
));
2005 arg
.since
= j64
- min(j64
, j5mins
);
2006 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
2008 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
2009 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
2010 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
2011 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
2013 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
2014 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
2015 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
2017 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
2018 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
2019 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
2020 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
2022 /* scan past 10 mins of error history */
2023 memset(&arg
, 0, sizeof(arg
));
2024 arg
.since
= j64
- min(j64
, j10mins
);
2025 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
2027 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
2028 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
2029 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
2031 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
2032 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
2033 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
2034 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
2040 * ata_eh_speed_down - record error and speed down if necessary
2041 * @dev: Failed device
2042 * @eflags: mask of ATA_EFLAG_* flags
2043 * @err_mask: err_mask of the error
2045 * Record error and examine error history to determine whether
2046 * adjusting transmission speed is necessary. It also sets
2047 * transmission limits appropriately if such adjustment is
2051 * Kernel thread context (may sleep).
2054 * Determined recovery action.
2056 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
2057 unsigned int eflags
, unsigned int err_mask
)
2059 struct ata_link
*link
= ata_dev_phys_link(dev
);
2061 unsigned int verdict
;
2062 unsigned int action
= 0;
2064 /* don't bother if Cat-0 error */
2065 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
2068 /* record error and determine whether speed down is necessary */
2069 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
2070 verdict
= ata_eh_speed_down_verdict(dev
);
2073 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
2074 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
2075 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
2076 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
2077 ata_dev_warn(dev
, "NCQ disabled due to excessive errors\n");
2082 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
2083 /* speed down SATA link speed if possible */
2084 if (sata_down_spd_limit(link
, 0) == 0) {
2085 action
|= ATA_EH_RESET
;
2089 /* lower transfer mode */
2090 if (dev
->spdn_cnt
< 2) {
2091 static const int dma_dnxfer_sel
[] =
2092 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
2093 static const int pio_dnxfer_sel
[] =
2094 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
2097 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
2098 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
2100 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
2104 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
2105 action
|= ATA_EH_RESET
;
2111 /* Fall back to PIO? Slowing down to PIO is meaningless for
2112 * SATA ATA devices. Consider it only for PATA and SATAPI.
2114 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
2115 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
2116 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
2117 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
2119 action
|= ATA_EH_RESET
;
2126 /* device has been slowed down, blow error history */
2127 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
2128 ata_ering_clear(&dev
->ering
);
2133 * ata_eh_worth_retry - analyze error and decide whether to retry
2134 * @qc: qc to possibly retry
2136 * Look at the cause of the error and decide if a retry
2137 * might be useful or not. We don't want to retry media errors
2138 * because the drive itself has probably already taken 10-30 seconds
2139 * doing its own internal retries before reporting the failure.
2141 static inline int ata_eh_worth_retry(struct ata_queued_cmd
*qc
)
2143 if (qc
->err_mask
& AC_ERR_MEDIA
)
2144 return 0; /* don't retry media errors */
2145 if (qc
->flags
& ATA_QCFLAG_IO
)
2146 return 1; /* otherwise retry anything from fs stack */
2147 if (qc
->err_mask
& AC_ERR_INVALID
)
2148 return 0; /* don't retry these */
2149 return qc
->err_mask
!= AC_ERR_DEV
; /* retry if not dev error */
2153 * ata_eh_link_autopsy - analyze error and determine recovery action
2154 * @link: host link to perform autopsy on
2156 * Analyze why @link failed and determine which recovery actions
2157 * are needed. This function also sets more detailed AC_ERR_*
2158 * values and fills sense data for ATAPI CHECK SENSE.
2161 * Kernel thread context (may sleep).
2163 static void ata_eh_link_autopsy(struct ata_link
*link
)
2165 struct ata_port
*ap
= link
->ap
;
2166 struct ata_eh_context
*ehc
= &link
->eh_context
;
2167 struct ata_device
*dev
;
2168 unsigned int all_err_mask
= 0, eflags
= 0;
2175 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
2178 /* obtain and analyze SError */
2179 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
2181 ehc
->i
.serror
|= serror
;
2182 ata_eh_analyze_serror(link
);
2183 } else if (rc
!= -EOPNOTSUPP
) {
2184 /* SError read failed, force reset and probing */
2185 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
2186 ehc
->i
.action
|= ATA_EH_RESET
;
2187 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
2190 /* analyze NCQ failure */
2191 ata_eh_analyze_ncq_error(link
);
2193 /* any real error trumps AC_ERR_OTHER */
2194 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
2195 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
2197 all_err_mask
|= ehc
->i
.err_mask
;
2199 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2200 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2202 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2203 ata_dev_phys_link(qc
->dev
) != link
)
2206 /* inherit upper level err_mask */
2207 qc
->err_mask
|= ehc
->i
.err_mask
;
2210 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
2212 /* DEV errors are probably spurious in case of ATA_BUS error */
2213 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
2214 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
2217 /* any real error trumps unknown error */
2218 if (qc
->err_mask
& ~AC_ERR_OTHER
)
2219 qc
->err_mask
&= ~AC_ERR_OTHER
;
2222 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2223 * layers will determine whether the command is worth retrying
2224 * based on the sense data and device class/type. Otherwise,
2225 * determine directly if the command is worth retrying using its
2226 * error mask and flags.
2228 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
2229 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
2230 else if (ata_eh_worth_retry(qc
))
2231 qc
->flags
|= ATA_QCFLAG_RETRY
;
2233 /* accumulate error info */
2234 ehc
->i
.dev
= qc
->dev
;
2235 all_err_mask
|= qc
->err_mask
;
2236 if (qc
->flags
& ATA_QCFLAG_IO
)
2237 eflags
|= ATA_EFLAG_IS_IO
;
2238 trace_ata_eh_link_autopsy_qc(qc
);
2241 /* enforce default EH actions */
2242 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
2243 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
2244 ehc
->i
.action
|= ATA_EH_RESET
;
2245 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
2246 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
2247 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2249 /* If we have offending qcs and the associated failed device,
2250 * perform per-dev EH action only on the offending device.
2253 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
2254 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
2255 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2258 /* propagate timeout to host link */
2259 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2260 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2262 /* record error and consider speeding down */
2264 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2265 ata_dev_enabled(link
->device
))))
2269 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2270 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2271 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2272 trace_ata_eh_link_autopsy(dev
, ehc
->i
.action
, all_err_mask
);
2278 * ata_eh_autopsy - analyze error and determine recovery action
2279 * @ap: host port to perform autopsy on
2281 * Analyze all links of @ap and determine why they failed and
2282 * which recovery actions are needed.
2285 * Kernel thread context (may sleep).
2287 void ata_eh_autopsy(struct ata_port
*ap
)
2289 struct ata_link
*link
;
2291 ata_for_each_link(link
, ap
, EDGE
)
2292 ata_eh_link_autopsy(link
);
2294 /* Handle the frigging slave link. Autopsy is done similarly
2295 * but actions and flags are transferred over to the master
2296 * link and handled from there.
2298 if (ap
->slave_link
) {
2299 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2300 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2302 /* transfer control flags from master to slave */
2303 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2305 /* perform autopsy on the slave link */
2306 ata_eh_link_autopsy(ap
->slave_link
);
2308 /* transfer actions from slave to master and clear slave */
2309 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2310 mehc
->i
.action
|= sehc
->i
.action
;
2311 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2312 mehc
->i
.flags
|= sehc
->i
.flags
;
2313 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2316 /* Autopsy of fanout ports can affect host link autopsy.
2317 * Perform host link autopsy last.
2319 if (sata_pmp_attached(ap
))
2320 ata_eh_link_autopsy(&ap
->link
);
2324 * ata_get_cmd_descript - get description for ATA command
2325 * @command: ATA command code to get description for
2327 * Return a textual description of the given command, or NULL if the
2328 * command is not known.
2333 const char *ata_get_cmd_descript(u8 command
)
2335 #ifdef CONFIG_ATA_VERBOSE_ERROR
2341 { ATA_CMD_DEV_RESET
, "DEVICE RESET" },
2342 { ATA_CMD_CHK_POWER
, "CHECK POWER MODE" },
2343 { ATA_CMD_STANDBY
, "STANDBY" },
2344 { ATA_CMD_IDLE
, "IDLE" },
2345 { ATA_CMD_EDD
, "EXECUTE DEVICE DIAGNOSTIC" },
2346 { ATA_CMD_DOWNLOAD_MICRO
, "DOWNLOAD MICROCODE" },
2347 { ATA_CMD_DOWNLOAD_MICRO_DMA
, "DOWNLOAD MICROCODE DMA" },
2348 { ATA_CMD_NOP
, "NOP" },
2349 { ATA_CMD_FLUSH
, "FLUSH CACHE" },
2350 { ATA_CMD_FLUSH_EXT
, "FLUSH CACHE EXT" },
2351 { ATA_CMD_ID_ATA
, "IDENTIFY DEVICE" },
2352 { ATA_CMD_ID_ATAPI
, "IDENTIFY PACKET DEVICE" },
2353 { ATA_CMD_SERVICE
, "SERVICE" },
2354 { ATA_CMD_READ
, "READ DMA" },
2355 { ATA_CMD_READ_EXT
, "READ DMA EXT" },
2356 { ATA_CMD_READ_QUEUED
, "READ DMA QUEUED" },
2357 { ATA_CMD_READ_STREAM_EXT
, "READ STREAM EXT" },
2358 { ATA_CMD_READ_STREAM_DMA_EXT
, "READ STREAM DMA EXT" },
2359 { ATA_CMD_WRITE
, "WRITE DMA" },
2360 { ATA_CMD_WRITE_EXT
, "WRITE DMA EXT" },
2361 { ATA_CMD_WRITE_QUEUED
, "WRITE DMA QUEUED EXT" },
2362 { ATA_CMD_WRITE_STREAM_EXT
, "WRITE STREAM EXT" },
2363 { ATA_CMD_WRITE_STREAM_DMA_EXT
, "WRITE STREAM DMA EXT" },
2364 { ATA_CMD_WRITE_FUA_EXT
, "WRITE DMA FUA EXT" },
2365 { ATA_CMD_WRITE_QUEUED_FUA_EXT
, "WRITE DMA QUEUED FUA EXT" },
2366 { ATA_CMD_FPDMA_READ
, "READ FPDMA QUEUED" },
2367 { ATA_CMD_FPDMA_WRITE
, "WRITE FPDMA QUEUED" },
2368 { ATA_CMD_FPDMA_SEND
, "SEND FPDMA QUEUED" },
2369 { ATA_CMD_FPDMA_RECV
, "RECEIVE FPDMA QUEUED" },
2370 { ATA_CMD_PIO_READ
, "READ SECTOR(S)" },
2371 { ATA_CMD_PIO_READ_EXT
, "READ SECTOR(S) EXT" },
2372 { ATA_CMD_PIO_WRITE
, "WRITE SECTOR(S)" },
2373 { ATA_CMD_PIO_WRITE_EXT
, "WRITE SECTOR(S) EXT" },
2374 { ATA_CMD_READ_MULTI
, "READ MULTIPLE" },
2375 { ATA_CMD_READ_MULTI_EXT
, "READ MULTIPLE EXT" },
2376 { ATA_CMD_WRITE_MULTI
, "WRITE MULTIPLE" },
2377 { ATA_CMD_WRITE_MULTI_EXT
, "WRITE MULTIPLE EXT" },
2378 { ATA_CMD_WRITE_MULTI_FUA_EXT
, "WRITE MULTIPLE FUA EXT" },
2379 { ATA_CMD_SET_FEATURES
, "SET FEATURES" },
2380 { ATA_CMD_SET_MULTI
, "SET MULTIPLE MODE" },
2381 { ATA_CMD_VERIFY
, "READ VERIFY SECTOR(S)" },
2382 { ATA_CMD_VERIFY_EXT
, "READ VERIFY SECTOR(S) EXT" },
2383 { ATA_CMD_WRITE_UNCORR_EXT
, "WRITE UNCORRECTABLE EXT" },
2384 { ATA_CMD_STANDBYNOW1
, "STANDBY IMMEDIATE" },
2385 { ATA_CMD_IDLEIMMEDIATE
, "IDLE IMMEDIATE" },
2386 { ATA_CMD_SLEEP
, "SLEEP" },
2387 { ATA_CMD_INIT_DEV_PARAMS
, "INITIALIZE DEVICE PARAMETERS" },
2388 { ATA_CMD_READ_NATIVE_MAX
, "READ NATIVE MAX ADDRESS" },
2389 { ATA_CMD_READ_NATIVE_MAX_EXT
, "READ NATIVE MAX ADDRESS EXT" },
2390 { ATA_CMD_SET_MAX
, "SET MAX ADDRESS" },
2391 { ATA_CMD_SET_MAX_EXT
, "SET MAX ADDRESS EXT" },
2392 { ATA_CMD_READ_LOG_EXT
, "READ LOG EXT" },
2393 { ATA_CMD_WRITE_LOG_EXT
, "WRITE LOG EXT" },
2394 { ATA_CMD_READ_LOG_DMA_EXT
, "READ LOG DMA EXT" },
2395 { ATA_CMD_WRITE_LOG_DMA_EXT
, "WRITE LOG DMA EXT" },
2396 { ATA_CMD_TRUSTED_NONDATA
, "TRUSTED NON-DATA" },
2397 { ATA_CMD_TRUSTED_RCV
, "TRUSTED RECEIVE" },
2398 { ATA_CMD_TRUSTED_RCV_DMA
, "TRUSTED RECEIVE DMA" },
2399 { ATA_CMD_TRUSTED_SND
, "TRUSTED SEND" },
2400 { ATA_CMD_TRUSTED_SND_DMA
, "TRUSTED SEND DMA" },
2401 { ATA_CMD_PMP_READ
, "READ BUFFER" },
2402 { ATA_CMD_PMP_READ_DMA
, "READ BUFFER DMA" },
2403 { ATA_CMD_PMP_WRITE
, "WRITE BUFFER" },
2404 { ATA_CMD_PMP_WRITE_DMA
, "WRITE BUFFER DMA" },
2405 { ATA_CMD_CONF_OVERLAY
, "DEVICE CONFIGURATION OVERLAY" },
2406 { ATA_CMD_SEC_SET_PASS
, "SECURITY SET PASSWORD" },
2407 { ATA_CMD_SEC_UNLOCK
, "SECURITY UNLOCK" },
2408 { ATA_CMD_SEC_ERASE_PREP
, "SECURITY ERASE PREPARE" },
2409 { ATA_CMD_SEC_ERASE_UNIT
, "SECURITY ERASE UNIT" },
2410 { ATA_CMD_SEC_FREEZE_LOCK
, "SECURITY FREEZE LOCK" },
2411 { ATA_CMD_SEC_DISABLE_PASS
, "SECURITY DISABLE PASSWORD" },
2412 { ATA_CMD_CONFIG_STREAM
, "CONFIGURE STREAM" },
2413 { ATA_CMD_SMART
, "SMART" },
2414 { ATA_CMD_MEDIA_LOCK
, "DOOR LOCK" },
2415 { ATA_CMD_MEDIA_UNLOCK
, "DOOR UNLOCK" },
2416 { ATA_CMD_DSM
, "DATA SET MANAGEMENT" },
2417 { ATA_CMD_CHK_MED_CRD_TYP
, "CHECK MEDIA CARD TYPE" },
2418 { ATA_CMD_CFA_REQ_EXT_ERR
, "CFA REQUEST EXTENDED ERROR" },
2419 { ATA_CMD_CFA_WRITE_NE
, "CFA WRITE SECTORS WITHOUT ERASE" },
2420 { ATA_CMD_CFA_TRANS_SECT
, "CFA TRANSLATE SECTOR" },
2421 { ATA_CMD_CFA_ERASE
, "CFA ERASE SECTORS" },
2422 { ATA_CMD_CFA_WRITE_MULT_NE
, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2423 { ATA_CMD_REQ_SENSE_DATA
, "REQUEST SENSE DATA EXT" },
2424 { ATA_CMD_SANITIZE_DEVICE
, "SANITIZE DEVICE" },
2425 { ATA_CMD_ZAC_MGMT_IN
, "ZAC MANAGEMENT IN" },
2426 { ATA_CMD_ZAC_MGMT_OUT
, "ZAC MANAGEMENT OUT" },
2427 { ATA_CMD_READ_LONG
, "READ LONG (with retries)" },
2428 { ATA_CMD_READ_LONG_ONCE
, "READ LONG (without retries)" },
2429 { ATA_CMD_WRITE_LONG
, "WRITE LONG (with retries)" },
2430 { ATA_CMD_WRITE_LONG_ONCE
, "WRITE LONG (without retries)" },
2431 { ATA_CMD_RESTORE
, "RECALIBRATE" },
2432 { 0, NULL
} /* terminate list */
2436 for (i
= 0; cmd_descr
[i
].text
; i
++)
2437 if (cmd_descr
[i
].command
== command
)
2438 return cmd_descr
[i
].text
;
2443 EXPORT_SYMBOL_GPL(ata_get_cmd_descript
);
2446 * ata_eh_link_report - report error handling to user
2447 * @link: ATA link EH is going on
2449 * Report EH to user.
2454 static void ata_eh_link_report(struct ata_link
*link
)
2456 struct ata_port
*ap
= link
->ap
;
2457 struct ata_eh_context
*ehc
= &link
->eh_context
;
2458 const char *frozen
, *desc
;
2459 char tries_buf
[6] = "";
2460 int tag
, nr_failed
= 0;
2462 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2466 if (ehc
->i
.desc
[0] != '\0')
2469 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2470 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2472 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2473 ata_dev_phys_link(qc
->dev
) != link
||
2474 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2475 qc
->err_mask
== AC_ERR_DEV
))
2477 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2483 if (!nr_failed
&& !ehc
->i
.err_mask
)
2487 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2490 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2491 snprintf(tries_buf
, sizeof(tries_buf
), " t%d",
2495 ata_dev_err(ehc
->i
.dev
, "exception Emask 0x%x "
2496 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2497 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2498 ehc
->i
.action
, frozen
, tries_buf
);
2500 ata_dev_err(ehc
->i
.dev
, "%s\n", desc
);
2502 ata_link_err(link
, "exception Emask 0x%x "
2503 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2504 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2505 ehc
->i
.action
, frozen
, tries_buf
);
2507 ata_link_err(link
, "%s\n", desc
);
2510 #ifdef CONFIG_ATA_VERBOSE_ERROR
2513 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2514 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2515 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2516 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2517 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2518 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2519 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2520 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2521 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2522 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2523 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2524 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2525 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2526 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2527 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2528 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2529 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2530 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2533 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2534 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2535 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2536 char data_buf
[20] = "";
2537 char cdb_buf
[70] = "";
2539 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2540 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2543 if (qc
->dma_dir
!= DMA_NONE
) {
2544 static const char *dma_str
[] = {
2545 [DMA_BIDIRECTIONAL
] = "bidi",
2546 [DMA_TO_DEVICE
] = "out",
2547 [DMA_FROM_DEVICE
] = "in",
2549 const char *prot_str
= NULL
;
2551 switch (qc
->tf
.protocol
) {
2552 case ATA_PROT_UNKNOWN
:
2553 prot_str
= "unknown";
2555 case ATA_PROT_NODATA
:
2556 prot_str
= "nodata";
2565 prot_str
= "ncq dma";
2567 case ATA_PROT_NCQ_NODATA
:
2568 prot_str
= "ncq nodata";
2570 case ATAPI_PROT_NODATA
:
2571 prot_str
= "nodata";
2573 case ATAPI_PROT_PIO
:
2576 case ATAPI_PROT_DMA
:
2580 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2581 prot_str
, qc
->nbytes
, dma_str
[qc
->dma_dir
]);
2584 if (ata_is_atapi(qc
->tf
.protocol
)) {
2585 const u8
*cdb
= qc
->cdb
;
2586 size_t cdb_len
= qc
->dev
->cdb_len
;
2589 cdb
= qc
->scsicmd
->cmnd
;
2590 cdb_len
= qc
->scsicmd
->cmd_len
;
2592 __scsi_format_command(cdb_buf
, sizeof(cdb_buf
),
2595 const char *descr
= ata_get_cmd_descript(cmd
->command
);
2597 ata_dev_err(qc
->dev
, "failed command: %s\n",
2601 ata_dev_err(qc
->dev
,
2602 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2604 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2605 "Emask 0x%x (%s)%s\n",
2606 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2607 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2608 cmd
->hob_feature
, cmd
->hob_nsect
,
2609 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2610 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2611 res
->command
, res
->feature
, res
->nsect
,
2612 res
->lbal
, res
->lbam
, res
->lbah
,
2613 res
->hob_feature
, res
->hob_nsect
,
2614 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2615 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2616 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2618 #ifdef CONFIG_ATA_VERBOSE_ERROR
2619 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2620 ATA_SENSE
| ATA_ERR
)) {
2621 if (res
->command
& ATA_BUSY
)
2622 ata_dev_err(qc
->dev
, "status: { Busy }\n");
2624 ata_dev_err(qc
->dev
, "status: { %s%s%s%s%s}\n",
2625 res
->command
& ATA_DRDY
? "DRDY " : "",
2626 res
->command
& ATA_DF
? "DF " : "",
2627 res
->command
& ATA_DRQ
? "DRQ " : "",
2628 res
->command
& ATA_SENSE
? "SENSE " : "",
2629 res
->command
& ATA_ERR
? "ERR " : "");
2632 if (cmd
->command
!= ATA_CMD_PACKET
&&
2633 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_AMNF
|
2634 ATA_IDNF
| ATA_ABORTED
)))
2635 ata_dev_err(qc
->dev
, "error: { %s%s%s%s%s}\n",
2636 res
->feature
& ATA_ICRC
? "ICRC " : "",
2637 res
->feature
& ATA_UNC
? "UNC " : "",
2638 res
->feature
& ATA_AMNF
? "AMNF " : "",
2639 res
->feature
& ATA_IDNF
? "IDNF " : "",
2640 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2646 * ata_eh_report - report error handling to user
2647 * @ap: ATA port to report EH about
2649 * Report EH to user.
2654 void ata_eh_report(struct ata_port
*ap
)
2656 struct ata_link
*link
;
2658 ata_for_each_link(link
, ap
, HOST_FIRST
)
2659 ata_eh_link_report(link
);
2662 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2663 unsigned int *classes
, unsigned long deadline
,
2666 struct ata_device
*dev
;
2669 ata_for_each_dev(dev
, link
, ALL
)
2670 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2672 return reset(link
, classes
, deadline
);
2675 static int ata_eh_followup_srst_needed(struct ata_link
*link
, int rc
)
2677 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2681 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2686 int ata_eh_reset(struct ata_link
*link
, int classify
,
2687 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2688 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2690 struct ata_port
*ap
= link
->ap
;
2691 struct ata_link
*slave
= ap
->slave_link
;
2692 struct ata_eh_context
*ehc
= &link
->eh_context
;
2693 struct ata_eh_context
*sehc
= slave
? &slave
->eh_context
: NULL
;
2694 unsigned int *classes
= ehc
->classes
;
2695 unsigned int lflags
= link
->flags
;
2696 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2697 int max_tries
= 0, try = 0;
2698 struct ata_link
*failed_link
;
2699 struct ata_device
*dev
;
2700 unsigned long deadline
, now
;
2701 ata_reset_fn_t reset
;
2702 unsigned long flags
;
2709 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2711 if (link
->flags
& ATA_LFLAG_RST_ONCE
)
2713 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2715 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2718 /* make sure each reset attempt is at least COOL_DOWN apart */
2719 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2721 WARN_ON(time_after(ehc
->last_reset
, now
));
2722 deadline
= ata_deadline(ehc
->last_reset
,
2723 ATA_EH_RESET_COOL_DOWN
);
2724 if (time_before(now
, deadline
))
2725 schedule_timeout_uninterruptible(deadline
- now
);
2728 spin_lock_irqsave(ap
->lock
, flags
);
2729 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2730 spin_unlock_irqrestore(ap
->lock
, flags
);
2732 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2734 ata_for_each_dev(dev
, link
, ALL
) {
2735 /* If we issue an SRST then an ATA drive (not ATAPI)
2736 * may change configuration and be in PIO0 timing. If
2737 * we do a hard reset (or are coming from power on)
2738 * this is true for ATA or ATAPI. Until we've set a
2739 * suitable controller mode we should not touch the
2740 * bus as we may be talking too fast.
2742 dev
->pio_mode
= XFER_PIO_0
;
2743 dev
->dma_mode
= 0xff;
2745 /* If the controller has a pio mode setup function
2746 * then use it to set the chipset to rights. Don't
2747 * touch the DMA setup as that will be dealt with when
2748 * configuring devices.
2750 if (ap
->ops
->set_piomode
)
2751 ap
->ops
->set_piomode(ap
, dev
);
2754 /* prefer hardreset */
2756 ehc
->i
.action
&= ~ATA_EH_RESET
;
2759 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2760 } else if (softreset
) {
2762 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2766 unsigned long deadline
= ata_deadline(jiffies
,
2767 ATA_EH_PRERESET_TIMEOUT
);
2770 sehc
->i
.action
&= ~ATA_EH_RESET
;
2771 sehc
->i
.action
|= ehc
->i
.action
;
2774 rc
= prereset(link
, deadline
);
2776 /* If present, do prereset on slave link too. Reset
2777 * is skipped iff both master and slave links report
2778 * -ENOENT or clear ATA_EH_RESET.
2780 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2783 tmp
= prereset(slave
, deadline
);
2787 ehc
->i
.action
|= sehc
->i
.action
;
2791 if (rc
== -ENOENT
) {
2792 ata_link_dbg(link
, "port disabled--ignoring\n");
2793 ehc
->i
.action
&= ~ATA_EH_RESET
;
2795 ata_for_each_dev(dev
, link
, ALL
)
2796 classes
[dev
->devno
] = ATA_DEV_NONE
;
2801 "prereset failed (errno=%d)\n",
2806 /* prereset() might have cleared ATA_EH_RESET. If so,
2807 * bang classes, thaw and return.
2809 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2810 ata_for_each_dev(dev
, link
, ALL
)
2811 classes
[dev
->devno
] = ATA_DEV_NONE
;
2812 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) &&
2813 ata_is_host_link(link
))
2814 ata_eh_thaw_port(ap
);
2824 if (ata_is_host_link(link
))
2825 ata_eh_freeze_port(ap
);
2827 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2831 ata_link_info(link
, "%s resetting link\n",
2832 reset
== softreset
? "soft" : "hard");
2834 /* mark that this EH session started with reset */
2835 ehc
->last_reset
= jiffies
;
2836 if (reset
== hardreset
)
2837 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2839 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2841 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2842 if (rc
&& rc
!= -EAGAIN
) {
2847 /* hardreset slave link if existent */
2848 if (slave
&& reset
== hardreset
) {
2852 ata_link_info(slave
, "hard resetting link\n");
2854 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2855 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2863 failed_link
= slave
;
2869 /* perform follow-up SRST if necessary */
2870 if (reset
== hardreset
&&
2871 ata_eh_followup_srst_needed(link
, rc
)) {
2876 "follow-up softreset required but no softreset available\n");
2882 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2883 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2892 "no reset method available, skipping reset\n");
2893 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2894 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2898 * Post-reset processing
2900 ata_for_each_dev(dev
, link
, ALL
) {
2901 /* After the reset, the device state is PIO 0 and the
2902 * controller state is undefined. Reset also wakes up
2903 * drives from sleeping mode.
2905 dev
->pio_mode
= XFER_PIO_0
;
2906 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2908 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2911 /* apply class override */
2912 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2913 classes
[dev
->devno
] = ATA_DEV_ATA
;
2914 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2915 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2918 /* record current link speed */
2919 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2920 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2921 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2922 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2925 if (ata_is_host_link(link
))
2926 ata_eh_thaw_port(ap
);
2928 /* postreset() should clear hardware SError. Although SError
2929 * is cleared during link resume, clearing SError here is
2930 * necessary as some PHYs raise hotplug events after SRST.
2931 * This introduces race condition where hotplug occurs between
2932 * reset and here. This race is mediated by cross checking
2933 * link onlineness and classification result later.
2936 postreset(link
, classes
);
2938 postreset(slave
, classes
);
2942 * Some controllers can't be frozen very well and may set spurious
2943 * error conditions during reset. Clear accumulated error
2944 * information and re-thaw the port if frozen. As reset is the
2945 * final recovery action and we cross check link onlineness against
2946 * device classification later, no hotplug event is lost by this.
2948 spin_lock_irqsave(link
->ap
->lock
, flags
);
2949 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
2951 memset(&slave
->eh_info
, 0, sizeof(link
->eh_info
));
2952 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
2953 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2955 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2956 ata_eh_thaw_port(ap
);
2959 * Make sure onlineness and classification result correspond.
2960 * Hotplug could have happened during reset and some
2961 * controllers fail to wait while a drive is spinning up after
2962 * being hotplugged causing misdetection. By cross checking
2963 * link on/offlineness and classification result, those
2964 * conditions can be reliably detected and retried.
2967 ata_for_each_dev(dev
, link
, ALL
) {
2968 if (ata_phys_link_online(ata_dev_phys_link(dev
))) {
2969 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2970 ata_dev_dbg(dev
, "link online but device misclassified\n");
2971 classes
[dev
->devno
] = ATA_DEV_NONE
;
2974 } else if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2975 if (ata_class_enabled(classes
[dev
->devno
]))
2977 "link offline, clearing class %d to NONE\n",
2978 classes
[dev
->devno
]);
2979 classes
[dev
->devno
] = ATA_DEV_NONE
;
2980 } else if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2982 "link status unknown, clearing UNKNOWN to NONE\n");
2983 classes
[dev
->devno
] = ATA_DEV_NONE
;
2987 if (classify
&& nr_unknown
) {
2988 if (try < max_tries
) {
2990 "link online but %d devices misclassified, retrying\n",
2997 "link online but %d devices misclassified, "
2998 "device detection might fail\n", nr_unknown
);
3001 /* reset successful, schedule revalidation */
3002 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
3004 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
3005 ehc
->last_reset
= jiffies
; /* update to completion time */
3006 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
3007 link
->lpm_policy
= ATA_LPM_UNKNOWN
; /* reset LPM state */
3011 /* clear hotplug flag */
3012 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
3014 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
3016 spin_lock_irqsave(ap
->lock
, flags
);
3017 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
3018 spin_unlock_irqrestore(ap
->lock
, flags
);
3023 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
3024 if (!ata_is_host_link(link
) &&
3025 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
3028 if (try >= max_tries
) {
3030 * Thaw host port even if reset failed, so that the port
3031 * can be retried on the next phy event. This risks
3032 * repeated EH runs but seems to be a better tradeoff than
3033 * shutting down a port after a botched hotplug attempt.
3035 if (ata_is_host_link(link
))
3036 ata_eh_thaw_port(ap
);
3041 if (time_before(now
, deadline
)) {
3042 unsigned long delta
= deadline
- now
;
3044 ata_link_warn(failed_link
,
3045 "reset failed (errno=%d), retrying in %u secs\n",
3046 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
3050 delta
= schedule_timeout_uninterruptible(delta
);
3055 * While disks spinup behind PMP, some controllers fail sending SRST.
3056 * They need to be reset - as well as the PMP - before retrying.
3058 if (rc
== -ERESTART
) {
3059 if (ata_is_host_link(link
))
3060 ata_eh_thaw_port(ap
);
3064 if (try == max_tries
- 1) {
3065 sata_down_spd_limit(link
, 0);
3067 sata_down_spd_limit(slave
, 0);
3068 } else if (rc
== -EPIPE
)
3069 sata_down_spd_limit(failed_link
, 0);
3076 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
3078 struct ata_link
*link
;
3079 struct ata_device
*dev
;
3080 unsigned long flags
;
3083 * This function can be thought of as an extended version of
3084 * ata_eh_about_to_do() specially crafted to accommodate the
3085 * requirements of ATA_EH_PARK handling. Since the EH thread
3086 * does not leave the do {} while () loop in ata_eh_recover as
3087 * long as the timeout for a park request to *one* device on
3088 * the port has not expired, and since we still want to pick
3089 * up park requests to other devices on the same port or
3090 * timeout updates for the same device, we have to pull
3091 * ATA_EH_PARK actions from eh_info into eh_context.i
3092 * ourselves at the beginning of each pass over the loop.
3094 * Additionally, all write accesses to &ap->park_req_pending
3095 * through reinit_completion() (see below) or complete_all()
3096 * (see ata_scsi_park_store()) are protected by the host lock.
3097 * As a result we have that park_req_pending.done is zero on
3098 * exit from this function, i.e. when ATA_EH_PARK actions for
3099 * *all* devices on port ap have been pulled into the
3100 * respective eh_context structs. If, and only if,
3101 * park_req_pending.done is non-zero by the time we reach
3102 * wait_for_completion_timeout(), another ATA_EH_PARK action
3103 * has been scheduled for at least one of the devices on port
3104 * ap and we have to cycle over the do {} while () loop in
3105 * ata_eh_recover() again.
3108 spin_lock_irqsave(ap
->lock
, flags
);
3109 reinit_completion(&ap
->park_req_pending
);
3110 ata_for_each_link(link
, ap
, EDGE
) {
3111 ata_for_each_dev(dev
, link
, ALL
) {
3112 struct ata_eh_info
*ehi
= &link
->eh_info
;
3114 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
3115 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
3116 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
3119 spin_unlock_irqrestore(ap
->lock
, flags
);
3122 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
3124 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3125 struct ata_taskfile tf
;
3126 unsigned int err_mask
;
3128 ata_tf_init(dev
, &tf
);
3130 ehc
->unloaded_mask
|= 1 << dev
->devno
;
3131 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
3137 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3138 tf
.command
= ATA_CMD_CHK_POWER
;
3141 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
3142 tf
.protocol
= ATA_PROT_NODATA
;
3143 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3144 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
3145 ata_dev_err(dev
, "head unload failed!\n");
3146 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3150 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
3151 struct ata_device
**r_failed_dev
)
3153 struct ata_port
*ap
= link
->ap
;
3154 struct ata_eh_context
*ehc
= &link
->eh_context
;
3155 struct ata_device
*dev
;
3156 unsigned int new_mask
= 0;
3157 unsigned long flags
;
3162 /* For PATA drive side cable detection to work, IDENTIFY must
3163 * be done backwards such that PDIAG- is released by the slave
3164 * device before the master device is identified.
3166 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
3167 unsigned int action
= ata_eh_dev_action(dev
);
3168 unsigned int readid_flags
= 0;
3170 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
3171 readid_flags
|= ATA_READID_POSTRESET
;
3173 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
3174 WARN_ON(dev
->class == ATA_DEV_PMP
);
3176 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
3181 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
3182 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
3187 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
3189 /* Configuration may have changed, reconfigure
3192 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3194 /* schedule the scsi_rescan_device() here */
3195 schedule_work(&(ap
->scsi_rescan_task
));
3196 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
3197 ehc
->tries
[dev
->devno
] &&
3198 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
3199 /* Temporarily set dev->class, it will be
3200 * permanently set once all configurations are
3201 * complete. This is necessary because new
3202 * device configuration is done in two
3205 dev
->class = ehc
->classes
[dev
->devno
];
3207 if (dev
->class == ATA_DEV_PMP
)
3208 rc
= sata_pmp_attach(dev
);
3210 rc
= ata_dev_read_id(dev
, &dev
->class,
3211 readid_flags
, dev
->id
);
3213 /* read_id might have changed class, store and reset */
3214 ehc
->classes
[dev
->devno
] = dev
->class;
3215 dev
->class = ATA_DEV_UNKNOWN
;
3219 /* clear error info accumulated during probe */
3220 ata_ering_clear(&dev
->ering
);
3221 new_mask
|= 1 << dev
->devno
;
3224 /* IDENTIFY was issued to non-existent
3225 * device. No need to reset. Just
3226 * thaw and ignore the device.
3228 ata_eh_thaw_port(ap
);
3236 /* PDIAG- should have been released, ask cable type if post-reset */
3237 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
3238 if (ap
->ops
->cable_detect
)
3239 ap
->cbl
= ap
->ops
->cable_detect(ap
);
3243 /* Configure new devices forward such that user doesn't see
3244 * device detection messages backwards.
3246 ata_for_each_dev(dev
, link
, ALL
) {
3247 if (!(new_mask
& (1 << dev
->devno
)))
3250 dev
->class = ehc
->classes
[dev
->devno
];
3252 if (dev
->class == ATA_DEV_PMP
)
3255 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
3256 rc
= ata_dev_configure(dev
);
3257 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
3259 dev
->class = ATA_DEV_UNKNOWN
;
3263 spin_lock_irqsave(ap
->lock
, flags
);
3264 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
3265 spin_unlock_irqrestore(ap
->lock
, flags
);
3267 /* new device discovered, configure xfermode */
3268 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3274 *r_failed_dev
= dev
;
3275 DPRINTK("EXIT rc=%d\n", rc
);
3280 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3281 * @link: link on which timings will be programmed
3282 * @r_failed_dev: out parameter for failed device
3284 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3285 * ata_set_mode() fails, pointer to the failing device is
3286 * returned in @r_failed_dev.
3289 * PCI/etc. bus probe sem.
3292 * 0 on success, negative errno otherwise
3294 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
3296 struct ata_port
*ap
= link
->ap
;
3297 struct ata_device
*dev
;
3300 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3301 ata_for_each_dev(dev
, link
, ENABLED
) {
3302 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
3303 struct ata_ering_entry
*ent
;
3305 ent
= ata_ering_top(&dev
->ering
);
3307 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
3311 /* has private set_mode? */
3312 if (ap
->ops
->set_mode
)
3313 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
3315 rc
= ata_do_set_mode(link
, r_failed_dev
);
3317 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3318 ata_for_each_dev(dev
, link
, ENABLED
) {
3319 struct ata_eh_context
*ehc
= &link
->eh_context
;
3320 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
3321 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
3323 if (dev
->xfer_mode
!= saved_xfer_mode
||
3324 ata_ncq_enabled(dev
) != saved_ncq
)
3325 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
3332 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3333 * @dev: ATAPI device to clear UA for
3335 * Resets and other operations can make an ATAPI device raise
3336 * UNIT ATTENTION which causes the next operation to fail. This
3337 * function clears UA.
3340 * EH context (may sleep).
3343 * 0 on success, -errno on failure.
3345 static int atapi_eh_clear_ua(struct ata_device
*dev
)
3349 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
3350 u8
*sense_buffer
= dev
->link
->ap
->sector_buf
;
3352 unsigned int err_mask
;
3354 err_mask
= atapi_eh_tur(dev
, &sense_key
);
3355 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
3357 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3362 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
3365 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
3367 ata_dev_warn(dev
, "failed to clear "
3368 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
3373 ata_dev_warn(dev
, "UNIT ATTENTION persists after %d tries\n",
3380 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3381 * @dev: ATA device which may need FLUSH retry
3383 * If @dev failed FLUSH, it needs to be reported upper layer
3384 * immediately as it means that @dev failed to remap and already
3385 * lost at least a sector and further FLUSH retrials won't make
3386 * any difference to the lost sector. However, if FLUSH failed
3387 * for other reasons, for example transmission error, FLUSH needs
3390 * This function determines whether FLUSH failure retry is
3391 * necessary and performs it if so.
3394 * 0 if EH can continue, -errno if EH needs to be repeated.
3396 static int ata_eh_maybe_retry_flush(struct ata_device
*dev
)
3398 struct ata_link
*link
= dev
->link
;
3399 struct ata_port
*ap
= link
->ap
;
3400 struct ata_queued_cmd
*qc
;
3401 struct ata_taskfile tf
;
3402 unsigned int err_mask
;
3405 /* did flush fail for this device? */
3406 if (!ata_tag_valid(link
->active_tag
))
3409 qc
= __ata_qc_from_tag(ap
, link
->active_tag
);
3410 if (qc
->dev
!= dev
|| (qc
->tf
.command
!= ATA_CMD_FLUSH_EXT
&&
3411 qc
->tf
.command
!= ATA_CMD_FLUSH
))
3414 /* if the device failed it, it should be reported to upper layers */
3415 if (qc
->err_mask
& AC_ERR_DEV
)
3418 /* flush failed for some other reason, give it another shot */
3419 ata_tf_init(dev
, &tf
);
3421 tf
.command
= qc
->tf
.command
;
3422 tf
.flags
|= ATA_TFLAG_DEVICE
;
3423 tf
.protocol
= ATA_PROT_NODATA
;
3425 ata_dev_warn(dev
, "retrying FLUSH 0x%x Emask 0x%x\n",
3426 tf
.command
, qc
->err_mask
);
3428 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3431 * FLUSH is complete but there's no way to
3432 * successfully complete a failed command from EH.
3433 * Making sure retry is allowed at least once and
3434 * retrying it should do the trick - whatever was in
3435 * the cache is already on the platter and this won't
3436 * cause infinite loop.
3438 qc
->scsicmd
->allowed
= max(qc
->scsicmd
->allowed
, 1);
3440 ata_dev_warn(dev
, "FLUSH failed Emask 0x%x\n",
3444 /* if device failed it, report it to upper layers */
3445 if (err_mask
& AC_ERR_DEV
) {
3446 qc
->err_mask
|= AC_ERR_DEV
;
3448 if (!(ap
->pflags
& ATA_PFLAG_FROZEN
))
3456 * ata_eh_set_lpm - configure SATA interface power management
3457 * @link: link to configure power management
3458 * @policy: the link power management policy
3459 * @r_failed_dev: out parameter for failed device
3461 * Enable SATA Interface power management. This will enable
3462 * Device Interface Power Management (DIPM) for min_power and
3463 * medium_power_with_dipm policies, and then call driver specific
3464 * callbacks for enabling Host Initiated Power management.
3470 * 0 on success, -errno on failure.
3472 static int ata_eh_set_lpm(struct ata_link
*link
, enum ata_lpm_policy policy
,
3473 struct ata_device
**r_failed_dev
)
3475 struct ata_port
*ap
= ata_is_host_link(link
) ? link
->ap
: NULL
;
3476 struct ata_eh_context
*ehc
= &link
->eh_context
;
3477 struct ata_device
*dev
, *link_dev
= NULL
, *lpm_dev
= NULL
;
3478 enum ata_lpm_policy old_policy
= link
->lpm_policy
;
3479 bool no_dipm
= link
->ap
->flags
& ATA_FLAG_NO_DIPM
;
3480 unsigned int hints
= ATA_LPM_EMPTY
| ATA_LPM_HIPM
;
3481 unsigned int err_mask
;
3484 /* if the link or host doesn't do LPM, noop */
3485 if ((link
->flags
& ATA_LFLAG_NO_LPM
) || (ap
&& !ap
->ops
->set_lpm
))
3489 * DIPM is enabled only for MIN_POWER as some devices
3490 * misbehave when the host NACKs transition to SLUMBER. Order
3491 * device and link configurations such that the host always
3492 * allows DIPM requests.
3494 ata_for_each_dev(dev
, link
, ENABLED
) {
3495 bool hipm
= ata_id_has_hipm(dev
->id
);
3496 bool dipm
= ata_id_has_dipm(dev
->id
) && !no_dipm
;
3498 /* find the first enabled and LPM enabled devices */
3502 if (!lpm_dev
&& (hipm
|| dipm
))
3505 hints
&= ~ATA_LPM_EMPTY
;
3507 hints
&= ~ATA_LPM_HIPM
;
3509 /* disable DIPM before changing link config */
3510 if (policy
< ATA_LPM_MED_POWER_WITH_DIPM
&& dipm
) {
3511 err_mask
= ata_dev_set_feature(dev
,
3512 SETFEATURES_SATA_DISABLE
, SATA_DIPM
);
3513 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3515 "failed to disable DIPM, Emask 0x%x\n",
3524 rc
= ap
->ops
->set_lpm(link
, policy
, hints
);
3525 if (!rc
&& ap
->slave_link
)
3526 rc
= ap
->ops
->set_lpm(ap
->slave_link
, policy
, hints
);
3528 rc
= sata_pmp_set_lpm(link
, policy
, hints
);
3531 * Attribute link config failure to the first (LPM) enabled
3532 * device on the link.
3535 if (rc
== -EOPNOTSUPP
) {
3536 link
->flags
|= ATA_LFLAG_NO_LPM
;
3539 dev
= lpm_dev
? lpm_dev
: link_dev
;
3544 * Low level driver acked the transition. Issue DIPM command
3545 * with the new policy set.
3547 link
->lpm_policy
= policy
;
3548 if (ap
&& ap
->slave_link
)
3549 ap
->slave_link
->lpm_policy
= policy
;
3551 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3552 ata_for_each_dev(dev
, link
, ENABLED
) {
3553 if (policy
>= ATA_LPM_MED_POWER_WITH_DIPM
&& !no_dipm
&&
3554 ata_id_has_dipm(dev
->id
)) {
3555 err_mask
= ata_dev_set_feature(dev
,
3556 SETFEATURES_SATA_ENABLE
, SATA_DIPM
);
3557 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3559 "failed to enable DIPM, Emask 0x%x\n",
3567 link
->last_lpm_change
= jiffies
;
3568 link
->flags
|= ATA_LFLAG_CHANGED
;
3573 /* restore the old policy */
3574 link
->lpm_policy
= old_policy
;
3575 if (ap
&& ap
->slave_link
)
3576 ap
->slave_link
->lpm_policy
= old_policy
;
3578 /* if no device or only one more chance is left, disable LPM */
3579 if (!dev
|| ehc
->tries
[dev
->devno
] <= 2) {
3580 ata_link_warn(link
, "disabling LPM on the link\n");
3581 link
->flags
|= ATA_LFLAG_NO_LPM
;
3584 *r_failed_dev
= dev
;
3588 int ata_link_nr_enabled(struct ata_link
*link
)
3590 struct ata_device
*dev
;
3593 ata_for_each_dev(dev
, link
, ENABLED
)
3598 static int ata_link_nr_vacant(struct ata_link
*link
)
3600 struct ata_device
*dev
;
3603 ata_for_each_dev(dev
, link
, ALL
)
3604 if (dev
->class == ATA_DEV_UNKNOWN
)
3609 static int ata_eh_skip_recovery(struct ata_link
*link
)
3611 struct ata_port
*ap
= link
->ap
;
3612 struct ata_eh_context
*ehc
= &link
->eh_context
;
3613 struct ata_device
*dev
;
3615 /* skip disabled links */
3616 if (link
->flags
& ATA_LFLAG_DISABLED
)
3619 /* skip if explicitly requested */
3620 if (ehc
->i
.flags
& ATA_EHI_NO_RECOVERY
)
3623 /* thaw frozen port and recover failed devices */
3624 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
3627 /* reset at least once if reset is requested */
3628 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
3629 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
3632 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3633 ata_for_each_dev(dev
, link
, ALL
) {
3634 if (dev
->class == ATA_DEV_UNKNOWN
&&
3635 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
3642 static int ata_count_probe_trials_cb(struct ata_ering_entry
*ent
, void *void_arg
)
3644 u64 interval
= msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL
);
3645 u64 now
= get_jiffies_64();
3646 int *trials
= void_arg
;
3648 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) ||
3649 (ent
->timestamp
< now
- min(now
, interval
)))
3656 static int ata_eh_schedule_probe(struct ata_device
*dev
)
3658 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3659 struct ata_link
*link
= ata_dev_phys_link(dev
);
3662 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
3663 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
3666 ata_eh_detach_dev(dev
);
3668 ehc
->did_probe_mask
|= (1 << dev
->devno
);
3669 ehc
->i
.action
|= ATA_EH_RESET
;
3670 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
3671 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
3673 /* the link maybe in a deep sleep, wake it up */
3674 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
) {
3675 if (ata_is_host_link(link
))
3676 link
->ap
->ops
->set_lpm(link
, ATA_LPM_MAX_POWER
,
3679 sata_pmp_set_lpm(link
, ATA_LPM_MAX_POWER
,
3683 /* Record and count probe trials on the ering. The specific
3684 * error mask used is irrelevant. Because a successful device
3685 * detection clears the ering, this count accumulates only if
3686 * there are consecutive failed probes.
3688 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3689 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3690 * forced to 1.5Gbps.
3692 * This is to work around cases where failed link speed
3693 * negotiation results in device misdetection leading to
3694 * infinite DEVXCHG or PHRDY CHG events.
3696 ata_ering_record(&dev
->ering
, 0, AC_ERR_OTHER
);
3697 ata_ering_map(&dev
->ering
, ata_count_probe_trials_cb
, &trials
);
3699 if (trials
> ATA_EH_PROBE_TRIALS
)
3700 sata_down_spd_limit(link
, 1);
3705 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
3707 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3709 /* -EAGAIN from EH routine indicates retry without prejudice.
3710 * The requester is responsible for ensuring forward progress.
3713 ehc
->tries
[dev
->devno
]--;
3717 /* device missing or wrong IDENTIFY data, schedule probing */
3718 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
3721 /* give it just one more chance */
3722 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
3725 if (ehc
->tries
[dev
->devno
] == 1) {
3726 /* This is the last chance, better to slow
3727 * down than lose it.
3729 sata_down_spd_limit(ata_dev_phys_link(dev
), 0);
3730 if (dev
->pio_mode
> XFER_PIO_0
)
3731 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
3735 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
3736 /* disable device if it has used up all its chances */
3737 ata_dev_disable(dev
);
3739 /* detach if offline */
3740 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
3741 ata_eh_detach_dev(dev
);
3743 /* schedule probe if necessary */
3744 if (ata_eh_schedule_probe(dev
)) {
3745 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3746 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3747 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3752 ehc
->i
.action
|= ATA_EH_RESET
;
3758 * ata_eh_recover - recover host port after error
3759 * @ap: host port to recover
3760 * @prereset: prereset method (can be NULL)
3761 * @softreset: softreset method (can be NULL)
3762 * @hardreset: hardreset method (can be NULL)
3763 * @postreset: postreset method (can be NULL)
3764 * @r_failed_link: out parameter for failed link
3766 * This is the alpha and omega, eum and yang, heart and soul of
3767 * libata exception handling. On entry, actions required to
3768 * recover each link and hotplug requests are recorded in the
3769 * link's eh_context. This function executes all the operations
3770 * with appropriate retrials and fallbacks to resurrect failed
3771 * devices, detach goners and greet newcomers.
3774 * Kernel thread context (may sleep).
3777 * 0 on success, -errno on failure.
3779 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3780 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3781 ata_postreset_fn_t postreset
,
3782 struct ata_link
**r_failed_link
)
3784 struct ata_link
*link
;
3785 struct ata_device
*dev
;
3787 unsigned long flags
, deadline
;
3791 /* prep for recovery */
3792 ata_for_each_link(link
, ap
, EDGE
) {
3793 struct ata_eh_context
*ehc
= &link
->eh_context
;
3795 /* re-enable link? */
3796 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3797 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3798 spin_lock_irqsave(ap
->lock
, flags
);
3799 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3800 spin_unlock_irqrestore(ap
->lock
, flags
);
3801 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3804 ata_for_each_dev(dev
, link
, ALL
) {
3805 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3806 ehc
->tries
[dev
->devno
] = 1;
3808 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3810 /* collect port action mask recorded in dev actions */
3811 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3812 ~ATA_EH_PERDEV_MASK
;
3813 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3815 /* process hotplug request */
3816 if (dev
->flags
& ATA_DFLAG_DETACH
)
3817 ata_eh_detach_dev(dev
);
3819 /* schedule probe if necessary */
3820 if (!ata_dev_enabled(dev
))
3821 ata_eh_schedule_probe(dev
);
3828 /* if UNLOADING, finish immediately */
3829 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3833 ata_for_each_link(link
, ap
, EDGE
) {
3834 struct ata_eh_context
*ehc
= &link
->eh_context
;
3836 /* skip EH if possible. */
3837 if (ata_eh_skip_recovery(link
))
3840 ata_for_each_dev(dev
, link
, ALL
)
3841 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3845 ata_for_each_link(link
, ap
, EDGE
) {
3846 struct ata_eh_context
*ehc
= &link
->eh_context
;
3848 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3851 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3852 prereset
, softreset
, hardreset
, postreset
);
3854 ata_link_err(link
, "reset failed, giving up\n");
3863 * clears ATA_EH_PARK in eh_info and resets
3864 * ap->park_req_pending
3866 ata_eh_pull_park_action(ap
);
3869 ata_for_each_link(link
, ap
, EDGE
) {
3870 ata_for_each_dev(dev
, link
, ALL
) {
3871 struct ata_eh_context
*ehc
= &link
->eh_context
;
3874 if (dev
->class != ATA_DEV_ATA
&&
3875 dev
->class != ATA_DEV_ZAC
)
3877 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3880 tmp
= dev
->unpark_deadline
;
3881 if (time_before(deadline
, tmp
))
3883 else if (time_before_eq(tmp
, jiffies
))
3885 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3888 ata_eh_park_issue_cmd(dev
, 1);
3893 if (time_before_eq(deadline
, now
))
3897 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3901 ata_for_each_link(link
, ap
, EDGE
) {
3902 ata_for_each_dev(dev
, link
, ALL
) {
3903 if (!(link
->eh_context
.unloaded_mask
&
3907 ata_eh_park_issue_cmd(dev
, 0);
3908 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3914 ata_for_each_link(link
, ap
, PMP_FIRST
) {
3915 struct ata_eh_context
*ehc
= &link
->eh_context
;
3917 if (sata_pmp_attached(ap
) && ata_is_host_link(link
))
3920 /* revalidate existing devices and attach new ones */
3921 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3925 /* if PMP got attached, return, pmp EH will take care of it */
3926 if (link
->device
->class == ATA_DEV_PMP
) {
3931 /* configure transfer mode if necessary */
3932 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3933 rc
= ata_set_mode(link
, &dev
);
3936 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3939 /* If reset has been issued, clear UA to avoid
3940 * disrupting the current users of the device.
3942 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3943 ata_for_each_dev(dev
, link
, ALL
) {
3944 if (dev
->class != ATA_DEV_ATAPI
)
3946 rc
= atapi_eh_clear_ua(dev
);
3949 if (zpodd_dev_enabled(dev
))
3950 zpodd_post_poweron(dev
);
3954 /* retry flush if necessary */
3955 ata_for_each_dev(dev
, link
, ALL
) {
3956 if (dev
->class != ATA_DEV_ATA
&&
3957 dev
->class != ATA_DEV_ZAC
)
3959 rc
= ata_eh_maybe_retry_flush(dev
);
3965 /* configure link power saving */
3966 if (link
->lpm_policy
!= ap
->target_lpm_policy
) {
3967 rc
= ata_eh_set_lpm(link
, ap
->target_lpm_policy
, &dev
);
3972 /* this link is okay now */
3979 ata_eh_handle_dev_fail(dev
, rc
);
3981 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3982 /* PMP reset requires working host port.
3983 * Can't retry if it's frozen.
3985 if (sata_pmp_attached(ap
))
3995 if (rc
&& r_failed_link
)
3996 *r_failed_link
= link
;
3998 DPRINTK("EXIT, rc=%d\n", rc
);
4003 * ata_eh_finish - finish up EH
4004 * @ap: host port to finish EH for
4006 * Recovery is complete. Clean up EH states and retry or finish
4012 void ata_eh_finish(struct ata_port
*ap
)
4016 /* retry or finish qcs */
4017 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
4018 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
4020 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
4024 /* FIXME: Once EH migration is complete,
4025 * generate sense data in this function,
4026 * considering both err_mask and tf.
4028 if (qc
->flags
& ATA_QCFLAG_RETRY
)
4029 ata_eh_qc_retry(qc
);
4031 ata_eh_qc_complete(qc
);
4033 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
4034 ata_eh_qc_complete(qc
);
4036 /* feed zero TF to sense generation */
4037 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
4038 ata_eh_qc_retry(qc
);
4043 /* make sure nr_active_links is zero after EH */
4044 WARN_ON(ap
->nr_active_links
);
4045 ap
->nr_active_links
= 0;
4049 * ata_do_eh - do standard error handling
4050 * @ap: host port to handle error for
4052 * @prereset: prereset method (can be NULL)
4053 * @softreset: softreset method (can be NULL)
4054 * @hardreset: hardreset method (can be NULL)
4055 * @postreset: postreset method (can be NULL)
4057 * Perform standard error handling sequence.
4060 * Kernel thread context (may sleep).
4062 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
4063 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
4064 ata_postreset_fn_t postreset
)
4066 struct ata_device
*dev
;
4072 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
4075 ata_for_each_dev(dev
, &ap
->link
, ALL
)
4076 ata_dev_disable(dev
);
4083 * ata_std_error_handler - standard error handler
4084 * @ap: host port to handle error for
4086 * Standard error handler
4089 * Kernel thread context (may sleep).
4091 void ata_std_error_handler(struct ata_port
*ap
)
4093 struct ata_port_operations
*ops
= ap
->ops
;
4094 ata_reset_fn_t hardreset
= ops
->hardreset
;
4096 /* ignore built-in hardreset if SCR access is not available */
4097 if (hardreset
== sata_std_hardreset
&& !sata_scr_valid(&ap
->link
))
4100 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
4105 * ata_eh_handle_port_suspend - perform port suspend operation
4106 * @ap: port to suspend
4111 * Kernel thread context (may sleep).
4113 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
4115 unsigned long flags
;
4117 struct ata_device
*dev
;
4119 /* are we suspending? */
4120 spin_lock_irqsave(ap
->lock
, flags
);
4121 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4122 ap
->pm_mesg
.event
& PM_EVENT_RESUME
) {
4123 spin_unlock_irqrestore(ap
->lock
, flags
);
4126 spin_unlock_irqrestore(ap
->lock
, flags
);
4128 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
4131 * If we have a ZPODD attached, check its zero
4132 * power ready status before the port is frozen.
4133 * Only needed for runtime suspend.
4135 if (PMSG_IS_AUTO(ap
->pm_mesg
)) {
4136 ata_for_each_dev(dev
, &ap
->link
, ENABLED
) {
4137 if (zpodd_dev_enabled(dev
))
4138 zpodd_on_suspend(dev
);
4142 /* tell ACPI we're suspending */
4143 rc
= ata_acpi_on_suspend(ap
);
4148 ata_eh_freeze_port(ap
);
4150 if (ap
->ops
->port_suspend
)
4151 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
4153 ata_acpi_set_state(ap
, ap
->pm_mesg
);
4155 /* update the flags */
4156 spin_lock_irqsave(ap
->lock
, flags
);
4158 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
4160 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
4161 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
4162 ata_port_schedule_eh(ap
);
4164 spin_unlock_irqrestore(ap
->lock
, flags
);
4170 * ata_eh_handle_port_resume - perform port resume operation
4171 * @ap: port to resume
4176 * Kernel thread context (may sleep).
4178 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
4180 struct ata_link
*link
;
4181 struct ata_device
*dev
;
4182 unsigned long flags
;
4184 /* are we resuming? */
4185 spin_lock_irqsave(ap
->lock
, flags
);
4186 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4187 !(ap
->pm_mesg
.event
& PM_EVENT_RESUME
)) {
4188 spin_unlock_irqrestore(ap
->lock
, flags
);
4191 spin_unlock_irqrestore(ap
->lock
, flags
);
4193 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
4196 * Error timestamps are in jiffies which doesn't run while
4197 * suspended and PHY events during resume isn't too uncommon.
4198 * When the two are combined, it can lead to unnecessary speed
4199 * downs if the machine is suspended and resumed repeatedly.
4200 * Clear error history.
4202 ata_for_each_link(link
, ap
, HOST_FIRST
)
4203 ata_for_each_dev(dev
, link
, ALL
)
4204 ata_ering_clear(&dev
->ering
);
4206 ata_acpi_set_state(ap
, ap
->pm_mesg
);
4208 if (ap
->ops
->port_resume
)
4209 ap
->ops
->port_resume(ap
);
4211 /* tell ACPI that we're resuming */
4212 ata_acpi_on_resume(ap
);
4214 /* update the flags */
4215 spin_lock_irqsave(ap
->lock
, flags
);
4216 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
4217 spin_unlock_irqrestore(ap
->lock
, flags
);
4219 #endif /* CONFIG_PM */