2 * libata-eh.c - libata error handling
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
47 #include <linux/libata.h>
52 /* speed down verdicts */
53 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
54 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
55 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
56 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
59 ATA_EFLAG_IS_IO
= (1 << 0),
60 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
61 ATA_EFLAG_OLD_ER
= (1 << 31),
63 /* error categories */
66 ATA_ECAT_TOUT_HSM
= 2,
68 ATA_ECAT_DUBIOUS_NONE
= 4,
69 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
70 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
71 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
74 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
76 /* always put at least this amount of time between resets */
77 ATA_EH_RESET_COOL_DOWN
= 5000,
79 /* Waiting in ->prereset can never be reliable. It's
80 * sometimes nice to wait there but it can't be depended upon;
81 * otherwise, we wouldn't be resetting. Just give it enough
82 * time for most drives to spin up.
84 ATA_EH_PRERESET_TIMEOUT
= 10000,
85 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
89 /* probe speed down parameters, see ata_eh_schedule_probe() */
90 ATA_EH_PROBE_TRIAL_INTERVAL
= 60000, /* 1 min */
91 ATA_EH_PROBE_TRIALS
= 2,
94 /* The following table determines how we sequence resets. Each entry
95 * represents timeout for that try. The first try can be soft or
96 * hardreset. All others are hardreset if available. In most cases
97 * the first reset w/ 10sec timeout should succeed. Following entries
98 * are mostly for error handling, hotplug and those outlier devices that
99 * take an exceptionally long time to recover from reset.
101 static const unsigned long ata_eh_reset_timeouts
[] = {
102 10000, /* most drives spin up by 10sec */
103 10000, /* > 99% working drives spin up before 20sec */
104 35000, /* give > 30 secs of idleness for outlier devices */
105 5000, /* and sweet one last chance */
106 ULONG_MAX
, /* > 1 min has elapsed, give up */
109 static const unsigned long ata_eh_identify_timeouts
[] = {
110 5000, /* covers > 99% of successes and not too boring on failures */
111 10000, /* combined time till here is enough even for media access */
112 30000, /* for true idiots */
116 static const unsigned long ata_eh_flush_timeouts
[] = {
117 15000, /* be generous with flush */
119 30000, /* and even more generous */
123 static const unsigned long ata_eh_other_timeouts
[] = {
124 5000, /* same rationale as identify timeout */
126 /* but no merciful 30sec for other commands, it just isn't worth it */
130 struct ata_eh_cmd_timeout_ent
{
132 const unsigned long *timeouts
;
135 /* The following table determines timeouts to use for EH internal
136 * commands. Each table entry is a command class and matches the
137 * commands the entry applies to and the timeout table to use.
139 * On the retry after a command timed out, the next timeout value from
140 * the table is used. If the table doesn't contain further entries,
141 * the last value is used.
143 * ehc->cmd_timeout_idx keeps track of which timeout to use per
144 * command class, so if SET_FEATURES times out on the first try, the
145 * next try will use the second timeout value only for that class.
147 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
148 static const struct ata_eh_cmd_timeout_ent
149 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
150 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
151 .timeouts
= ata_eh_identify_timeouts
, },
152 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
153 .timeouts
= ata_eh_other_timeouts
, },
154 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
155 .timeouts
= ata_eh_other_timeouts
, },
156 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
157 .timeouts
= ata_eh_other_timeouts
, },
158 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
159 .timeouts
= ata_eh_other_timeouts
, },
160 { .commands
= CMDS(ATA_CMD_FLUSH
, ATA_CMD_FLUSH_EXT
),
161 .timeouts
= ata_eh_flush_timeouts
},
165 static void __ata_port_freeze(struct ata_port
*ap
);
167 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
168 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
169 #else /* CONFIG_PM */
170 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
173 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
175 #endif /* CONFIG_PM */
177 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
180 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
181 ATA_EH_DESC_LEN
- ehi
->desc_len
,
186 * __ata_ehi_push_desc - push error description without adding separator
188 * @fmt: printf format string
190 * Format string according to @fmt and append it to @ehi->desc.
193 * spin_lock_irqsave(host lock)
195 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
200 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
205 * ata_ehi_push_desc - push error description with separator
207 * @fmt: printf format string
209 * Format string according to @fmt and append it to @ehi->desc.
210 * If @ehi->desc is not empty, ", " is added in-between.
213 * spin_lock_irqsave(host lock)
215 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
220 __ata_ehi_push_desc(ehi
, ", ");
223 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
228 * ata_ehi_clear_desc - clean error description
234 * spin_lock_irqsave(host lock)
236 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
243 * ata_port_desc - append port description
244 * @ap: target ATA port
245 * @fmt: printf format string
247 * Format string according to @fmt and append it to port
248 * description. If port description is not empty, " " is added
249 * in-between. This function is to be used while initializing
250 * ata_host. The description is printed on host registration.
255 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
259 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
261 if (ap
->link
.eh_info
.desc_len
)
262 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
265 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
272 * ata_port_pbar_desc - append PCI BAR description
273 * @ap: target ATA port
274 * @bar: target PCI BAR
275 * @offset: offset into PCI BAR
276 * @name: name of the area
278 * If @offset is negative, this function formats a string which
279 * contains the name, address, size and type of the BAR and
280 * appends it to the port description. If @offset is zero or
281 * positive, only name and offsetted address is appended.
286 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
289 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
291 unsigned long long start
, len
;
293 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
295 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
298 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
299 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
302 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
304 ata_port_desc(ap
, "%s 0x%llx", name
,
305 start
+ (unsigned long long)offset
);
308 #endif /* CONFIG_PCI */
310 static int ata_lookup_timeout_table(u8 cmd
)
314 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
317 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
326 * ata_internal_cmd_timeout - determine timeout for an internal command
327 * @dev: target device
328 * @cmd: internal command to be issued
330 * Determine timeout for internal command @cmd for @dev.
336 * Determined timeout.
338 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
340 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
341 int ent
= ata_lookup_timeout_table(cmd
);
345 return ATA_EH_CMD_DFL_TIMEOUT
;
347 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
348 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
352 * ata_internal_cmd_timed_out - notification for internal command timeout
353 * @dev: target device
354 * @cmd: internal command which timed out
356 * Notify EH that internal command @cmd for @dev timed out. This
357 * function should be called only for commands whose timeouts are
358 * determined using ata_internal_cmd_timeout().
363 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
365 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
366 int ent
= ata_lookup_timeout_table(cmd
);
372 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
373 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
374 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
377 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
378 unsigned int err_mask
)
380 struct ata_ering_entry
*ent
;
385 ering
->cursor
%= ATA_ERING_SIZE
;
387 ent
= &ering
->ring
[ering
->cursor
];
388 ent
->eflags
= eflags
;
389 ent
->err_mask
= err_mask
;
390 ent
->timestamp
= get_jiffies_64();
393 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
395 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
402 int ata_ering_map(struct ata_ering
*ering
,
403 int (*map_fn
)(struct ata_ering_entry
*, void *),
407 struct ata_ering_entry
*ent
;
411 ent
= &ering
->ring
[idx
];
414 rc
= map_fn(ent
, arg
);
417 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
418 } while (idx
!= ering
->cursor
);
423 static int ata_ering_clear_cb(struct ata_ering_entry
*ent
, void *void_arg
)
425 ent
->eflags
|= ATA_EFLAG_OLD_ER
;
429 static void ata_ering_clear(struct ata_ering
*ering
)
431 ata_ering_map(ering
, ata_ering_clear_cb
, NULL
);
434 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
436 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
438 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
441 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
442 struct ata_eh_info
*ehi
, unsigned int action
)
444 struct ata_device
*tdev
;
447 ehi
->action
&= ~action
;
448 ata_for_each_dev(tdev
, link
, ALL
)
449 ehi
->dev_action
[tdev
->devno
] &= ~action
;
451 /* doesn't make sense for port-wide EH actions */
452 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
454 /* break ehi->action into ehi->dev_action */
455 if (ehi
->action
& action
) {
456 ata_for_each_dev(tdev
, link
, ALL
)
457 ehi
->dev_action
[tdev
->devno
] |=
458 ehi
->action
& action
;
459 ehi
->action
&= ~action
;
462 /* turn off the specified per-dev action */
463 ehi
->dev_action
[dev
->devno
] &= ~action
;
468 * ata_eh_acquire - acquire EH ownership
469 * @ap: ATA port to acquire EH ownership for
471 * Acquire EH ownership for @ap. This is the basic exclusion
472 * mechanism for ports sharing a host. Only one port hanging off
473 * the same host can claim the ownership of EH.
478 void ata_eh_acquire(struct ata_port
*ap
)
480 mutex_lock(&ap
->host
->eh_mutex
);
481 WARN_ON_ONCE(ap
->host
->eh_owner
);
482 ap
->host
->eh_owner
= current
;
486 * ata_eh_release - release EH ownership
487 * @ap: ATA port to release EH ownership for
489 * Release EH ownership for @ap if the caller. The caller must
490 * have acquired EH ownership using ata_eh_acquire() previously.
495 void ata_eh_release(struct ata_port
*ap
)
497 WARN_ON_ONCE(ap
->host
->eh_owner
!= current
);
498 ap
->host
->eh_owner
= NULL
;
499 mutex_unlock(&ap
->host
->eh_mutex
);
503 * ata_scsi_timed_out - SCSI layer time out callback
504 * @cmd: timed out SCSI command
506 * Handles SCSI layer timeout. We race with normal completion of
507 * the qc for @cmd. If the qc is already gone, we lose and let
508 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
509 * timed out and EH should be invoked. Prevent ata_qc_complete()
510 * from finishing it by setting EH_SCHEDULED and return
513 * TODO: kill this function once old EH is gone.
516 * Called from timer context
519 * EH_HANDLED or EH_NOT_HANDLED
521 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
523 struct Scsi_Host
*host
= cmd
->device
->host
;
524 struct ata_port
*ap
= ata_shost_to_port(host
);
526 struct ata_queued_cmd
*qc
;
527 enum blk_eh_timer_return ret
;
531 if (ap
->ops
->error_handler
) {
532 ret
= BLK_EH_NOT_HANDLED
;
536 ret
= BLK_EH_HANDLED
;
537 spin_lock_irqsave(ap
->lock
, flags
);
538 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
540 WARN_ON(qc
->scsicmd
!= cmd
);
541 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
542 qc
->err_mask
|= AC_ERR_TIMEOUT
;
543 ret
= BLK_EH_NOT_HANDLED
;
545 spin_unlock_irqrestore(ap
->lock
, flags
);
548 DPRINTK("EXIT, ret=%d\n", ret
);
552 static void ata_eh_unload(struct ata_port
*ap
)
554 struct ata_link
*link
;
555 struct ata_device
*dev
;
558 /* Restore SControl IPM and SPD for the next driver and
559 * disable attached devices.
561 ata_for_each_link(link
, ap
, PMP_FIRST
) {
562 sata_scr_write(link
, SCR_CONTROL
, link
->saved_scontrol
& 0xff0);
563 ata_for_each_dev(dev
, link
, ALL
)
564 ata_dev_disable(dev
);
567 /* freeze and set UNLOADED */
568 spin_lock_irqsave(ap
->lock
, flags
);
570 ata_port_freeze(ap
); /* won't be thawed */
571 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
; /* clear pending from freeze */
572 ap
->pflags
|= ATA_PFLAG_UNLOADED
;
574 spin_unlock_irqrestore(ap
->lock
, flags
);
578 * ata_scsi_error - SCSI layer error handler callback
579 * @host: SCSI host on which error occurred
581 * Handles SCSI-layer-thrown error events.
584 * Inherited from SCSI layer (none, can sleep)
589 void ata_scsi_error(struct Scsi_Host
*host
)
591 struct ata_port
*ap
= ata_shost_to_port(host
);
593 LIST_HEAD(eh_work_q
);
597 spin_lock_irqsave(host
->host_lock
, flags
);
598 list_splice_init(&host
->eh_cmd_q
, &eh_work_q
);
599 spin_unlock_irqrestore(host
->host_lock
, flags
);
601 ata_scsi_cmd_error_handler(host
, ap
, &eh_work_q
);
603 /* If we timed raced normal completion and there is nothing to
604 recover nr_timedout == 0 why exactly are we doing error recovery ? */
605 ata_scsi_port_error_handler(host
, ap
);
607 /* finish or retry handled scmd's and clean up */
608 WARN_ON(host
->host_failed
|| !list_empty(&eh_work_q
));
614 * ata_scsi_cmd_error_handler - error callback for a list of commands
615 * @host: scsi host containing the port
616 * @ap: ATA port within the host
617 * @eh_work_q: list of commands to process
619 * process the given list of commands and return those finished to the
620 * ap->eh_done_q. This function is the first part of the libata error
621 * handler which processes a given list of failed commands.
623 void ata_scsi_cmd_error_handler(struct Scsi_Host
*host
, struct ata_port
*ap
,
624 struct list_head
*eh_work_q
)
629 /* make sure sff pio task is not running */
630 ata_sff_flush_pio_task(ap
);
632 /* synchronize with host lock and sort out timeouts */
634 /* For new EH, all qcs are finished in one of three ways -
635 * normal completion, error completion, and SCSI timeout.
636 * Both completions can race against SCSI timeout. When normal
637 * completion wins, the qc never reaches EH. When error
638 * completion wins, the qc has ATA_QCFLAG_FAILED set.
640 * When SCSI timeout wins, things are a bit more complex.
641 * Normal or error completion can occur after the timeout but
642 * before this point. In such cases, both types of
643 * completions are honored. A scmd is determined to have
644 * timed out iff its associated qc is active and not failed.
646 if (ap
->ops
->error_handler
) {
647 struct scsi_cmnd
*scmd
, *tmp
;
650 spin_lock_irqsave(ap
->lock
, flags
);
652 /* This must occur under the ap->lock as we don't want
653 a polled recovery to race the real interrupt handler
655 The lost_interrupt handler checks for any completed but
656 non-notified command and completes much like an IRQ handler.
658 We then fall into the error recovery code which will treat
659 this as if normal completion won the race */
661 if (ap
->ops
->lost_interrupt
)
662 ap
->ops
->lost_interrupt(ap
);
664 list_for_each_entry_safe(scmd
, tmp
, eh_work_q
, eh_entry
) {
665 struct ata_queued_cmd
*qc
;
667 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
668 qc
= __ata_qc_from_tag(ap
, i
);
669 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
674 if (i
< ATA_MAX_QUEUE
) {
675 /* the scmd has an associated qc */
676 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
677 /* which hasn't failed yet, timeout */
678 qc
->err_mask
|= AC_ERR_TIMEOUT
;
679 qc
->flags
|= ATA_QCFLAG_FAILED
;
683 /* Normal completion occurred after
684 * SCSI timeout but before this point.
685 * Successfully complete it.
687 scmd
->retries
= scmd
->allowed
;
688 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
692 /* If we have timed out qcs. They belong to EH from
693 * this point but the state of the controller is
694 * unknown. Freeze the port to make sure the IRQ
695 * handler doesn't diddle with those qcs. This must
696 * be done atomically w.r.t. setting QCFLAG_FAILED.
699 __ata_port_freeze(ap
);
701 spin_unlock_irqrestore(ap
->lock
, flags
);
703 /* initialize eh_tries */
704 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
706 spin_unlock_wait(ap
->lock
);
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 schedule_delayed_work(&ap
->hotplug_task
, 0);
821 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
822 ata_port_info(ap
, "EH complete\n");
824 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
826 /* tell wait_eh that we're done */
827 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
828 wake_up_all(&ap
->eh_wait_q
);
830 spin_unlock_irqrestore(ap
->lock
, flags
);
832 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler
);
835 * ata_port_wait_eh - Wait for the currently pending EH to complete
836 * @ap: Port to wait EH for
838 * Wait until the currently pending EH is complete.
841 * Kernel thread context (may sleep).
843 void ata_port_wait_eh(struct ata_port
*ap
)
849 spin_lock_irqsave(ap
->lock
, flags
);
851 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
852 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
853 spin_unlock_irqrestore(ap
->lock
, flags
);
855 spin_lock_irqsave(ap
->lock
, flags
);
857 finish_wait(&ap
->eh_wait_q
, &wait
);
859 spin_unlock_irqrestore(ap
->lock
, flags
);
861 /* make sure SCSI EH is complete */
862 if (scsi_host_in_recovery(ap
->scsi_host
)) {
867 EXPORT_SYMBOL_GPL(ata_port_wait_eh
);
869 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
874 /* count only non-internal commands */
875 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
876 if (ata_qc_from_tag(ap
, tag
))
882 void ata_eh_fastdrain_timerfn(unsigned long arg
)
884 struct ata_port
*ap
= (void *)arg
;
888 spin_lock_irqsave(ap
->lock
, flags
);
890 cnt
= ata_eh_nr_in_flight(ap
);
896 if (cnt
== ap
->fastdrain_cnt
) {
899 /* No progress during the last interval, tag all
900 * in-flight qcs as timed out and freeze the port.
902 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
903 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
905 qc
->err_mask
|= AC_ERR_TIMEOUT
;
910 /* some qcs have finished, give it another chance */
911 ap
->fastdrain_cnt
= cnt
;
912 ap
->fastdrain_timer
.expires
=
913 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
914 add_timer(&ap
->fastdrain_timer
);
918 spin_unlock_irqrestore(ap
->lock
, flags
);
922 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
923 * @ap: target ATA port
924 * @fastdrain: activate fast drain
926 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
927 * is non-zero and EH wasn't pending before. Fast drain ensures
928 * that EH kicks in in timely manner.
931 * spin_lock_irqsave(host lock)
933 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
937 /* already scheduled? */
938 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
941 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
946 /* do we have in-flight qcs? */
947 cnt
= ata_eh_nr_in_flight(ap
);
951 /* activate fast drain */
952 ap
->fastdrain_cnt
= cnt
;
953 ap
->fastdrain_timer
.expires
=
954 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
955 add_timer(&ap
->fastdrain_timer
);
959 * ata_qc_schedule_eh - schedule qc for error handling
960 * @qc: command to schedule error handling for
962 * Schedule error handling for @qc. EH will kick in as soon as
963 * other commands are drained.
966 * spin_lock_irqsave(host lock)
968 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
970 struct ata_port
*ap
= qc
->ap
;
971 struct request_queue
*q
= qc
->scsicmd
->device
->request_queue
;
974 WARN_ON(!ap
->ops
->error_handler
);
976 qc
->flags
|= ATA_QCFLAG_FAILED
;
977 ata_eh_set_pending(ap
, 1);
979 /* The following will fail if timeout has already expired.
980 * ata_scsi_error() takes care of such scmds on EH entry.
981 * Note that ATA_QCFLAG_FAILED is unconditionally set after
982 * this function completes.
984 spin_lock_irqsave(q
->queue_lock
, flags
);
985 blk_abort_request(qc
->scsicmd
->request
);
986 spin_unlock_irqrestore(q
->queue_lock
, flags
);
990 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
991 * @ap: ATA port to schedule EH for
993 * LOCKING: inherited from ata_port_schedule_eh
994 * spin_lock_irqsave(host lock)
996 void ata_std_sched_eh(struct ata_port
*ap
)
998 WARN_ON(!ap
->ops
->error_handler
);
1000 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
1003 ata_eh_set_pending(ap
, 1);
1004 scsi_schedule_eh(ap
->scsi_host
);
1006 DPRINTK("port EH scheduled\n");
1008 EXPORT_SYMBOL_GPL(ata_std_sched_eh
);
1011 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1012 * @ap: ATA port to end EH for
1014 * In the libata object model there is a 1:1 mapping of ata_port to
1015 * shost, so host fields can be directly manipulated under ap->lock, in
1016 * the libsas case we need to hold a lock at the ha->level to coordinate
1020 * spin_lock_irqsave(host lock)
1022 void ata_std_end_eh(struct ata_port
*ap
)
1024 struct Scsi_Host
*host
= ap
->scsi_host
;
1026 host
->host_eh_scheduled
= 0;
1028 EXPORT_SYMBOL(ata_std_end_eh
);
1032 * ata_port_schedule_eh - schedule error handling without a qc
1033 * @ap: ATA port to schedule EH for
1035 * Schedule error handling for @ap. EH will kick in as soon as
1036 * all commands are drained.
1039 * spin_lock_irqsave(host lock)
1041 void ata_port_schedule_eh(struct ata_port
*ap
)
1043 /* see: ata_std_sched_eh, unless you know better */
1044 ap
->ops
->sched_eh(ap
);
1047 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
1049 int tag
, nr_aborted
= 0;
1051 WARN_ON(!ap
->ops
->error_handler
);
1053 /* we're gonna abort all commands, no need for fast drain */
1054 ata_eh_set_pending(ap
, 0);
1056 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1057 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
1059 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
1060 qc
->flags
|= ATA_QCFLAG_FAILED
;
1061 ata_qc_complete(qc
);
1067 ata_port_schedule_eh(ap
);
1073 * ata_link_abort - abort all qc's on the link
1074 * @link: ATA link to abort qc's for
1076 * Abort all active qc's active on @link and schedule EH.
1079 * spin_lock_irqsave(host lock)
1082 * Number of aborted qc's.
1084 int ata_link_abort(struct ata_link
*link
)
1086 return ata_do_link_abort(link
->ap
, link
);
1090 * ata_port_abort - abort all qc's on the port
1091 * @ap: ATA port to abort qc's for
1093 * Abort all active qc's of @ap and schedule EH.
1096 * spin_lock_irqsave(host_set lock)
1099 * Number of aborted qc's.
1101 int ata_port_abort(struct ata_port
*ap
)
1103 return ata_do_link_abort(ap
, NULL
);
1107 * __ata_port_freeze - freeze port
1108 * @ap: ATA port to freeze
1110 * This function is called when HSM violation or some other
1111 * condition disrupts normal operation of the port. Frozen port
1112 * is not allowed to perform any operation until the port is
1113 * thawed, which usually follows a successful reset.
1115 * ap->ops->freeze() callback can be used for freezing the port
1116 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1117 * port cannot be frozen hardware-wise, the interrupt handler
1118 * must ack and clear interrupts unconditionally while the port
1122 * spin_lock_irqsave(host lock)
1124 static void __ata_port_freeze(struct ata_port
*ap
)
1126 WARN_ON(!ap
->ops
->error_handler
);
1128 if (ap
->ops
->freeze
)
1129 ap
->ops
->freeze(ap
);
1131 ap
->pflags
|= ATA_PFLAG_FROZEN
;
1133 DPRINTK("ata%u port frozen\n", ap
->print_id
);
1137 * ata_port_freeze - abort & freeze port
1138 * @ap: ATA port to freeze
1140 * Abort and freeze @ap. The freeze operation must be called
1141 * first, because some hardware requires special operations
1142 * before the taskfile registers are accessible.
1145 * spin_lock_irqsave(host lock)
1148 * Number of aborted commands.
1150 int ata_port_freeze(struct ata_port
*ap
)
1154 WARN_ON(!ap
->ops
->error_handler
);
1156 __ata_port_freeze(ap
);
1157 nr_aborted
= ata_port_abort(ap
);
1163 * sata_async_notification - SATA async notification handler
1164 * @ap: ATA port where async notification is received
1166 * Handler to be called when async notification via SDB FIS is
1167 * received. This function schedules EH if necessary.
1170 * spin_lock_irqsave(host lock)
1173 * 1 if EH is scheduled, 0 otherwise.
1175 int sata_async_notification(struct ata_port
*ap
)
1180 if (!(ap
->flags
& ATA_FLAG_AN
))
1183 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
1185 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
1187 if (!sata_pmp_attached(ap
) || rc
) {
1188 /* PMP is not attached or SNTF is not available */
1189 if (!sata_pmp_attached(ap
)) {
1190 /* PMP is not attached. Check whether ATAPI
1191 * AN is configured. If so, notify media
1194 struct ata_device
*dev
= ap
->link
.device
;
1196 if ((dev
->class == ATA_DEV_ATAPI
) &&
1197 (dev
->flags
& ATA_DFLAG_AN
))
1198 ata_scsi_media_change_notify(dev
);
1201 /* PMP is attached but SNTF is not available.
1202 * ATAPI async media change notification is
1203 * not used. The PMP must be reporting PHY
1204 * status change, schedule EH.
1206 ata_port_schedule_eh(ap
);
1210 /* PMP is attached and SNTF is available */
1211 struct ata_link
*link
;
1213 /* check and notify ATAPI AN */
1214 ata_for_each_link(link
, ap
, EDGE
) {
1215 if (!(sntf
& (1 << link
->pmp
)))
1218 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1219 (link
->device
->flags
& ATA_DFLAG_AN
))
1220 ata_scsi_media_change_notify(link
->device
);
1223 /* If PMP is reporting that PHY status of some
1224 * downstream ports has changed, schedule EH.
1226 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1227 ata_port_schedule_eh(ap
);
1236 * ata_eh_freeze_port - EH helper to freeze port
1237 * @ap: ATA port to freeze
1244 void ata_eh_freeze_port(struct ata_port
*ap
)
1246 unsigned long flags
;
1248 if (!ap
->ops
->error_handler
)
1251 spin_lock_irqsave(ap
->lock
, flags
);
1252 __ata_port_freeze(ap
);
1253 spin_unlock_irqrestore(ap
->lock
, flags
);
1257 * ata_port_thaw_port - EH helper to thaw port
1258 * @ap: ATA port to thaw
1260 * Thaw frozen port @ap.
1265 void ata_eh_thaw_port(struct ata_port
*ap
)
1267 unsigned long flags
;
1269 if (!ap
->ops
->error_handler
)
1272 spin_lock_irqsave(ap
->lock
, flags
);
1274 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1279 spin_unlock_irqrestore(ap
->lock
, flags
);
1281 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1284 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1289 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1291 struct ata_port
*ap
= qc
->ap
;
1292 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1293 unsigned long flags
;
1295 spin_lock_irqsave(ap
->lock
, flags
);
1296 qc
->scsidone
= ata_eh_scsidone
;
1297 __ata_qc_complete(qc
);
1298 WARN_ON(ata_tag_valid(qc
->tag
));
1299 spin_unlock_irqrestore(ap
->lock
, flags
);
1301 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1305 * ata_eh_qc_complete - Complete an active ATA command from EH
1306 * @qc: Command to complete
1308 * Indicate to the mid and upper layers that an ATA command has
1309 * completed. To be used from EH.
1311 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1313 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1314 scmd
->retries
= scmd
->allowed
;
1315 __ata_eh_qc_complete(qc
);
1319 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1320 * @qc: Command to retry
1322 * Indicate to the mid and upper layers that an ATA command
1323 * should be retried. To be used from EH.
1325 * SCSI midlayer limits the number of retries to scmd->allowed.
1326 * scmd->allowed is incremented for commands which get retried
1327 * due to unrelated failures (qc->err_mask is zero).
1329 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1331 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1334 __ata_eh_qc_complete(qc
);
1338 * ata_dev_disable - disable ATA device
1339 * @dev: ATA device to disable
1346 void ata_dev_disable(struct ata_device
*dev
)
1348 if (!ata_dev_enabled(dev
))
1351 if (ata_msg_drv(dev
->link
->ap
))
1352 ata_dev_warn(dev
, "disabled\n");
1353 ata_acpi_on_disable(dev
);
1354 ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO0
| ATA_DNXFER_QUIET
);
1357 /* From now till the next successful probe, ering is used to
1358 * track probe failures. Clear accumulated device error info.
1360 ata_ering_clear(&dev
->ering
);
1364 * ata_eh_detach_dev - detach ATA device
1365 * @dev: ATA device to detach
1372 void ata_eh_detach_dev(struct ata_device
*dev
)
1374 struct ata_link
*link
= dev
->link
;
1375 struct ata_port
*ap
= link
->ap
;
1376 struct ata_eh_context
*ehc
= &link
->eh_context
;
1377 unsigned long flags
;
1379 ata_dev_disable(dev
);
1381 spin_lock_irqsave(ap
->lock
, flags
);
1383 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1385 if (ata_scsi_offline_dev(dev
)) {
1386 dev
->flags
|= ATA_DFLAG_DETACHED
;
1387 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1390 /* clear per-dev EH info */
1391 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1392 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1393 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1394 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1396 spin_unlock_irqrestore(ap
->lock
, flags
);
1400 * ata_eh_about_to_do - about to perform eh_action
1401 * @link: target ATA link
1402 * @dev: target ATA dev for per-dev action (can be NULL)
1403 * @action: action about to be performed
1405 * Called just before performing EH actions to clear related bits
1406 * in @link->eh_info such that eh actions are not unnecessarily
1412 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1413 unsigned int action
)
1415 struct ata_port
*ap
= link
->ap
;
1416 struct ata_eh_info
*ehi
= &link
->eh_info
;
1417 struct ata_eh_context
*ehc
= &link
->eh_context
;
1418 unsigned long flags
;
1420 spin_lock_irqsave(ap
->lock
, flags
);
1422 ata_eh_clear_action(link
, dev
, ehi
, action
);
1424 /* About to take EH action, set RECOVERED. Ignore actions on
1425 * slave links as master will do them again.
1427 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1428 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1430 spin_unlock_irqrestore(ap
->lock
, flags
);
1434 * ata_eh_done - EH action complete
1435 * @ap: target ATA port
1436 * @dev: target ATA dev for per-dev action (can be NULL)
1437 * @action: action just completed
1439 * Called right after performing EH actions to clear related bits
1440 * in @link->eh_context.
1445 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1446 unsigned int action
)
1448 struct ata_eh_context
*ehc
= &link
->eh_context
;
1450 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1454 * ata_err_string - convert err_mask to descriptive string
1455 * @err_mask: error mask to convert to string
1457 * Convert @err_mask to descriptive string. Errors are
1458 * prioritized according to severity and only the most severe
1459 * error is reported.
1465 * Descriptive string for @err_mask
1467 static const char *ata_err_string(unsigned int err_mask
)
1469 if (err_mask
& AC_ERR_HOST_BUS
)
1470 return "host bus error";
1471 if (err_mask
& AC_ERR_ATA_BUS
)
1472 return "ATA bus error";
1473 if (err_mask
& AC_ERR_TIMEOUT
)
1475 if (err_mask
& AC_ERR_HSM
)
1476 return "HSM violation";
1477 if (err_mask
& AC_ERR_SYSTEM
)
1478 return "internal error";
1479 if (err_mask
& AC_ERR_MEDIA
)
1480 return "media error";
1481 if (err_mask
& AC_ERR_INVALID
)
1482 return "invalid argument";
1483 if (err_mask
& AC_ERR_DEV
)
1484 return "device error";
1485 return "unknown error";
1489 * ata_read_log_page - read a specific log page
1490 * @dev: target device
1492 * @page: page to read
1493 * @buf: buffer to store read page
1494 * @sectors: number of sectors to read
1496 * Read log page using READ_LOG_EXT command.
1499 * Kernel thread context (may sleep).
1502 * 0 on success, AC_ERR_* mask otherwise.
1504 unsigned int ata_read_log_page(struct ata_device
*dev
, u8 log
,
1505 u8 page
, void *buf
, unsigned int sectors
)
1507 struct ata_taskfile tf
;
1508 unsigned int err_mask
;
1510 DPRINTK("read log page - log 0x%x, page 0x%x\n", log
, page
);
1512 ata_tf_init(dev
, &tf
);
1513 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1517 tf
.hob_nsect
= sectors
>> 8;
1518 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1519 tf
.protocol
= ATA_PROT_PIO
;
1521 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1522 buf
, sectors
* ATA_SECT_SIZE
, 0);
1524 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1529 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1530 * @dev: Device to read log page 10h from
1531 * @tag: Resulting tag of the failed command
1532 * @tf: Resulting taskfile registers of the failed command
1534 * Read log page 10h to obtain NCQ error details and clear error
1538 * Kernel thread context (may sleep).
1541 * 0 on success, -errno otherwise.
1543 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1544 int *tag
, struct ata_taskfile
*tf
)
1546 u8
*buf
= dev
->link
->ap
->sector_buf
;
1547 unsigned int err_mask
;
1551 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, 0, buf
, 1);
1556 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1559 ata_dev_warn(dev
, "invalid checksum 0x%x on log page 10h\n",
1565 *tag
= buf
[0] & 0x1f;
1567 tf
->command
= buf
[2];
1568 tf
->feature
= buf
[3];
1572 tf
->device
= buf
[7];
1573 tf
->hob_lbal
= buf
[8];
1574 tf
->hob_lbam
= buf
[9];
1575 tf
->hob_lbah
= buf
[10];
1576 tf
->nsect
= buf
[12];
1577 tf
->hob_nsect
= buf
[13];
1583 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1584 * @dev: target ATAPI device
1585 * @r_sense_key: out parameter for sense_key
1587 * Perform ATAPI TEST_UNIT_READY.
1590 * EH context (may sleep).
1593 * 0 on success, AC_ERR_* mask on failure.
1595 unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1597 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1598 struct ata_taskfile tf
;
1599 unsigned int err_mask
;
1601 ata_tf_init(dev
, &tf
);
1603 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1604 tf
.command
= ATA_CMD_PACKET
;
1605 tf
.protocol
= ATAPI_PROT_NODATA
;
1607 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1608 if (err_mask
== AC_ERR_DEV
)
1609 *r_sense_key
= tf
.feature
>> 4;
1614 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1615 * @dev: device to perform REQUEST_SENSE to
1616 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1617 * @dfl_sense_key: default sense key to use
1619 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1620 * SENSE. This function is EH helper.
1623 * Kernel thread context (may sleep).
1626 * 0 on success, AC_ERR_* mask on failure
1628 unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1629 u8
*sense_buf
, u8 dfl_sense_key
)
1631 u8 cdb
[ATAPI_CDB_LEN
] =
1632 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1633 struct ata_port
*ap
= dev
->link
->ap
;
1634 struct ata_taskfile tf
;
1636 DPRINTK("ATAPI request sense\n");
1638 /* FIXME: is this needed? */
1639 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1641 /* initialize sense_buf with the error register,
1642 * for the case where they are -not- overwritten
1644 sense_buf
[0] = 0x70;
1645 sense_buf
[2] = dfl_sense_key
;
1647 /* some devices time out if garbage left in tf */
1648 ata_tf_init(dev
, &tf
);
1650 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1651 tf
.command
= ATA_CMD_PACKET
;
1653 /* is it pointless to prefer PIO for "safety reasons"? */
1654 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1655 tf
.protocol
= ATAPI_PROT_DMA
;
1656 tf
.feature
|= ATAPI_PKT_DMA
;
1658 tf
.protocol
= ATAPI_PROT_PIO
;
1659 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1663 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1664 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1668 * ata_eh_analyze_serror - analyze SError for a failed port
1669 * @link: ATA link to analyze SError for
1671 * Analyze SError if available and further determine cause of
1677 static void ata_eh_analyze_serror(struct ata_link
*link
)
1679 struct ata_eh_context
*ehc
= &link
->eh_context
;
1680 u32 serror
= ehc
->i
.serror
;
1681 unsigned int err_mask
= 0, action
= 0;
1684 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1685 err_mask
|= AC_ERR_ATA_BUS
;
1686 action
|= ATA_EH_RESET
;
1688 if (serror
& SERR_PROTOCOL
) {
1689 err_mask
|= AC_ERR_HSM
;
1690 action
|= ATA_EH_RESET
;
1692 if (serror
& SERR_INTERNAL
) {
1693 err_mask
|= AC_ERR_SYSTEM
;
1694 action
|= ATA_EH_RESET
;
1697 /* Determine whether a hotplug event has occurred. Both
1698 * SError.N/X are considered hotplug events for enabled or
1699 * host links. For disabled PMP links, only N bit is
1700 * considered as X bit is left at 1 for link plugging.
1702 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
)
1703 hotplug_mask
= 0; /* hotplug doesn't work w/ LPM */
1704 else if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1705 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1707 hotplug_mask
= SERR_PHYRDY_CHG
;
1709 if (serror
& hotplug_mask
)
1710 ata_ehi_hotplugged(&ehc
->i
);
1712 ehc
->i
.err_mask
|= err_mask
;
1713 ehc
->i
.action
|= action
;
1717 * ata_eh_analyze_ncq_error - analyze NCQ error
1718 * @link: ATA link to analyze NCQ error for
1720 * Read log page 10h, determine the offending qc and acquire
1721 * error status TF. For NCQ device errors, all LLDDs have to do
1722 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1726 * Kernel thread context (may sleep).
1728 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1730 struct ata_port
*ap
= link
->ap
;
1731 struct ata_eh_context
*ehc
= &link
->eh_context
;
1732 struct ata_device
*dev
= link
->device
;
1733 struct ata_queued_cmd
*qc
;
1734 struct ata_taskfile tf
;
1737 /* if frozen, we can't do much */
1738 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1741 /* is it NCQ device error? */
1742 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1745 /* has LLDD analyzed already? */
1746 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1747 qc
= __ata_qc_from_tag(ap
, tag
);
1749 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1756 /* okay, this error is ours */
1757 memset(&tf
, 0, sizeof(tf
));
1758 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1760 ata_link_err(link
, "failed to read log page 10h (errno=%d)\n",
1765 if (!(link
->sactive
& (1 << tag
))) {
1766 ata_link_err(link
, "log page 10h reported inactive tag %d\n",
1771 /* we've got the perpetrator, condemn it */
1772 qc
= __ata_qc_from_tag(ap
, tag
);
1773 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1774 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1775 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1776 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1780 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1781 * @qc: qc to analyze
1782 * @tf: Taskfile registers to analyze
1784 * Analyze taskfile of @qc and further determine cause of
1785 * failure. This function also requests ATAPI sense data if
1789 * Kernel thread context (may sleep).
1792 * Determined recovery action
1794 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1795 const struct ata_taskfile
*tf
)
1797 unsigned int tmp
, action
= 0;
1798 u8 stat
= tf
->command
, err
= tf
->feature
;
1800 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1801 qc
->err_mask
|= AC_ERR_HSM
;
1802 return ATA_EH_RESET
;
1805 if (stat
& (ATA_ERR
| ATA_DF
))
1806 qc
->err_mask
|= AC_ERR_DEV
;
1810 switch (qc
->dev
->class) {
1813 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1815 qc
->err_mask
|= AC_ERR_MEDIA
;
1817 qc
->err_mask
|= AC_ERR_INVALID
;
1821 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1822 tmp
= atapi_eh_request_sense(qc
->dev
,
1823 qc
->scsicmd
->sense_buffer
,
1824 qc
->result_tf
.feature
>> 4);
1826 /* ATA_QCFLAG_SENSE_VALID is used to
1827 * tell atapi_qc_complete() that sense
1828 * data is already valid.
1830 * TODO: interpret sense data and set
1831 * appropriate err_mask.
1833 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1835 qc
->err_mask
|= tmp
;
1839 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1840 action
|= ATA_EH_RESET
;
1845 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1850 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1854 base
= ATA_ECAT_DUBIOUS_NONE
;
1856 if (err_mask
& AC_ERR_ATA_BUS
)
1857 return base
+ ATA_ECAT_ATA_BUS
;
1859 if (err_mask
& AC_ERR_TIMEOUT
)
1860 return base
+ ATA_ECAT_TOUT_HSM
;
1862 if (eflags
& ATA_EFLAG_IS_IO
) {
1863 if (err_mask
& AC_ERR_HSM
)
1864 return base
+ ATA_ECAT_TOUT_HSM
;
1866 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1867 return base
+ ATA_ECAT_UNK_DEV
;
1873 struct speed_down_verdict_arg
{
1876 int nr_errors
[ATA_ECAT_NR
];
1879 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1881 struct speed_down_verdict_arg
*arg
= void_arg
;
1884 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) || (ent
->timestamp
< arg
->since
))
1887 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1889 arg
->nr_errors
[cat
]++;
1895 * ata_eh_speed_down_verdict - Determine speed down verdict
1896 * @dev: Device of interest
1898 * This function examines error ring of @dev and determines
1899 * whether NCQ needs to be turned off, transfer speed should be
1900 * stepped down, or falling back to PIO is necessary.
1902 * ECAT_ATA_BUS : ATA_BUS error for any command
1904 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1907 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1909 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1910 * data transfer hasn't been verified.
1914 * NCQ_OFF : Turn off NCQ.
1916 * SPEED_DOWN : Speed down transfer speed but don't fall back
1919 * FALLBACK_TO_PIO : Fall back to PIO.
1921 * Even if multiple verdicts are returned, only one action is
1922 * taken per error. An action triggered by non-DUBIOUS errors
1923 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1924 * This is to expedite speed down decisions right after device is
1925 * initially configured.
1927 * The followings are speed down rules. #1 and #2 deal with
1930 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1931 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1933 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1934 * occurred during last 5 mins, NCQ_OFF.
1936 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1937 * occurred during last 5 mins, FALLBACK_TO_PIO
1939 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1940 * during last 10 mins, NCQ_OFF.
1942 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1943 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1946 * Inherited from caller.
1949 * OR of ATA_EH_SPDN_* flags.
1951 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1953 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1954 u64 j64
= get_jiffies_64();
1955 struct speed_down_verdict_arg arg
;
1956 unsigned int verdict
= 0;
1958 /* scan past 5 mins of error history */
1959 memset(&arg
, 0, sizeof(arg
));
1960 arg
.since
= j64
- min(j64
, j5mins
);
1961 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1963 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1964 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1965 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1966 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1968 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1969 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1970 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1972 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1973 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1974 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1975 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1977 /* scan past 10 mins of error history */
1978 memset(&arg
, 0, sizeof(arg
));
1979 arg
.since
= j64
- min(j64
, j10mins
);
1980 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1982 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1983 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1984 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1986 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1987 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1988 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1989 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1995 * ata_eh_speed_down - record error and speed down if necessary
1996 * @dev: Failed device
1997 * @eflags: mask of ATA_EFLAG_* flags
1998 * @err_mask: err_mask of the error
2000 * Record error and examine error history to determine whether
2001 * adjusting transmission speed is necessary. It also sets
2002 * transmission limits appropriately if such adjustment is
2006 * Kernel thread context (may sleep).
2009 * Determined recovery action.
2011 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
2012 unsigned int eflags
, unsigned int err_mask
)
2014 struct ata_link
*link
= ata_dev_phys_link(dev
);
2016 unsigned int verdict
;
2017 unsigned int action
= 0;
2019 /* don't bother if Cat-0 error */
2020 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
2023 /* record error and determine whether speed down is necessary */
2024 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
2025 verdict
= ata_eh_speed_down_verdict(dev
);
2028 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
2029 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
2030 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
2031 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
2032 ata_dev_warn(dev
, "NCQ disabled due to excessive errors\n");
2037 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
2038 /* speed down SATA link speed if possible */
2039 if (sata_down_spd_limit(link
, 0) == 0) {
2040 action
|= ATA_EH_RESET
;
2044 /* lower transfer mode */
2045 if (dev
->spdn_cnt
< 2) {
2046 static const int dma_dnxfer_sel
[] =
2047 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
2048 static const int pio_dnxfer_sel
[] =
2049 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
2052 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
2053 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
2055 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
2059 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
2060 action
|= ATA_EH_RESET
;
2066 /* Fall back to PIO? Slowing down to PIO is meaningless for
2067 * SATA ATA devices. Consider it only for PATA and SATAPI.
2069 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
2070 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
2071 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
2072 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
2074 action
|= ATA_EH_RESET
;
2081 /* device has been slowed down, blow error history */
2082 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
2083 ata_ering_clear(&dev
->ering
);
2088 * ata_eh_worth_retry - analyze error and decide whether to retry
2089 * @qc: qc to possibly retry
2091 * Look at the cause of the error and decide if a retry
2092 * might be useful or not. We don't want to retry media errors
2093 * because the drive itself has probably already taken 10-30 seconds
2094 * doing its own internal retries before reporting the failure.
2096 static inline int ata_eh_worth_retry(struct ata_queued_cmd
*qc
)
2098 if (qc
->err_mask
& AC_ERR_MEDIA
)
2099 return 0; /* don't retry media errors */
2100 if (qc
->flags
& ATA_QCFLAG_IO
)
2101 return 1; /* otherwise retry anything from fs stack */
2102 if (qc
->err_mask
& AC_ERR_INVALID
)
2103 return 0; /* don't retry these */
2104 return qc
->err_mask
!= AC_ERR_DEV
; /* retry if not dev error */
2108 * ata_eh_link_autopsy - analyze error and determine recovery action
2109 * @link: host link to perform autopsy on
2111 * Analyze why @link failed and determine which recovery actions
2112 * are needed. This function also sets more detailed AC_ERR_*
2113 * values and fills sense data for ATAPI CHECK SENSE.
2116 * Kernel thread context (may sleep).
2118 static void ata_eh_link_autopsy(struct ata_link
*link
)
2120 struct ata_port
*ap
= link
->ap
;
2121 struct ata_eh_context
*ehc
= &link
->eh_context
;
2122 struct ata_device
*dev
;
2123 unsigned int all_err_mask
= 0, eflags
= 0;
2130 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
2133 /* obtain and analyze SError */
2134 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
2136 ehc
->i
.serror
|= serror
;
2137 ata_eh_analyze_serror(link
);
2138 } else if (rc
!= -EOPNOTSUPP
) {
2139 /* SError read failed, force reset and probing */
2140 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
2141 ehc
->i
.action
|= ATA_EH_RESET
;
2142 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
2145 /* analyze NCQ failure */
2146 ata_eh_analyze_ncq_error(link
);
2148 /* any real error trumps AC_ERR_OTHER */
2149 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
2150 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
2152 all_err_mask
|= ehc
->i
.err_mask
;
2154 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2155 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2157 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2158 ata_dev_phys_link(qc
->dev
) != link
)
2161 /* inherit upper level err_mask */
2162 qc
->err_mask
|= ehc
->i
.err_mask
;
2165 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
2167 /* DEV errors are probably spurious in case of ATA_BUS error */
2168 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
2169 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
2172 /* any real error trumps unknown error */
2173 if (qc
->err_mask
& ~AC_ERR_OTHER
)
2174 qc
->err_mask
&= ~AC_ERR_OTHER
;
2176 /* SENSE_VALID trumps dev/unknown error and revalidation */
2177 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
2178 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
2180 /* determine whether the command is worth retrying */
2181 if (ata_eh_worth_retry(qc
))
2182 qc
->flags
|= ATA_QCFLAG_RETRY
;
2184 /* accumulate error info */
2185 ehc
->i
.dev
= qc
->dev
;
2186 all_err_mask
|= qc
->err_mask
;
2187 if (qc
->flags
& ATA_QCFLAG_IO
)
2188 eflags
|= ATA_EFLAG_IS_IO
;
2191 /* enforce default EH actions */
2192 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
2193 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
2194 ehc
->i
.action
|= ATA_EH_RESET
;
2195 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
2196 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
2197 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2199 /* If we have offending qcs and the associated failed device,
2200 * perform per-dev EH action only on the offending device.
2203 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
2204 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
2205 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2208 /* propagate timeout to host link */
2209 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2210 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2212 /* record error and consider speeding down */
2214 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2215 ata_dev_enabled(link
->device
))))
2219 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2220 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2221 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2228 * ata_eh_autopsy - analyze error and determine recovery action
2229 * @ap: host port to perform autopsy on
2231 * Analyze all links of @ap and determine why they failed and
2232 * which recovery actions are needed.
2235 * Kernel thread context (may sleep).
2237 void ata_eh_autopsy(struct ata_port
*ap
)
2239 struct ata_link
*link
;
2241 ata_for_each_link(link
, ap
, EDGE
)
2242 ata_eh_link_autopsy(link
);
2244 /* Handle the frigging slave link. Autopsy is done similarly
2245 * but actions and flags are transferred over to the master
2246 * link and handled from there.
2248 if (ap
->slave_link
) {
2249 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2250 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2252 /* transfer control flags from master to slave */
2253 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2255 /* perform autopsy on the slave link */
2256 ata_eh_link_autopsy(ap
->slave_link
);
2258 /* transfer actions from slave to master and clear slave */
2259 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2260 mehc
->i
.action
|= sehc
->i
.action
;
2261 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2262 mehc
->i
.flags
|= sehc
->i
.flags
;
2263 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2266 /* Autopsy of fanout ports can affect host link autopsy.
2267 * Perform host link autopsy last.
2269 if (sata_pmp_attached(ap
))
2270 ata_eh_link_autopsy(&ap
->link
);
2274 * ata_get_cmd_descript - get description for ATA command
2275 * @command: ATA command code to get description for
2277 * Return a textual description of the given command, or NULL if the
2278 * command is not known.
2283 const char *ata_get_cmd_descript(u8 command
)
2285 #ifdef CONFIG_ATA_VERBOSE_ERROR
2291 { ATA_CMD_DEV_RESET
, "DEVICE RESET" },
2292 { ATA_CMD_CHK_POWER
, "CHECK POWER MODE" },
2293 { ATA_CMD_STANDBY
, "STANDBY" },
2294 { ATA_CMD_IDLE
, "IDLE" },
2295 { ATA_CMD_EDD
, "EXECUTE DEVICE DIAGNOSTIC" },
2296 { ATA_CMD_DOWNLOAD_MICRO
, "DOWNLOAD MICROCODE" },
2297 { ATA_CMD_DOWNLOAD_MICRO_DMA
, "DOWNLOAD MICROCODE DMA" },
2298 { ATA_CMD_NOP
, "NOP" },
2299 { ATA_CMD_FLUSH
, "FLUSH CACHE" },
2300 { ATA_CMD_FLUSH_EXT
, "FLUSH CACHE EXT" },
2301 { ATA_CMD_ID_ATA
, "IDENTIFY DEVICE" },
2302 { ATA_CMD_ID_ATAPI
, "IDENTIFY PACKET DEVICE" },
2303 { ATA_CMD_SERVICE
, "SERVICE" },
2304 { ATA_CMD_READ
, "READ DMA" },
2305 { ATA_CMD_READ_EXT
, "READ DMA EXT" },
2306 { ATA_CMD_READ_QUEUED
, "READ DMA QUEUED" },
2307 { ATA_CMD_READ_STREAM_EXT
, "READ STREAM EXT" },
2308 { ATA_CMD_READ_STREAM_DMA_EXT
, "READ STREAM DMA EXT" },
2309 { ATA_CMD_WRITE
, "WRITE DMA" },
2310 { ATA_CMD_WRITE_EXT
, "WRITE DMA EXT" },
2311 { ATA_CMD_WRITE_QUEUED
, "WRITE DMA QUEUED EXT" },
2312 { ATA_CMD_WRITE_STREAM_EXT
, "WRITE STREAM EXT" },
2313 { ATA_CMD_WRITE_STREAM_DMA_EXT
, "WRITE STREAM DMA EXT" },
2314 { ATA_CMD_WRITE_FUA_EXT
, "WRITE DMA FUA EXT" },
2315 { ATA_CMD_WRITE_QUEUED_FUA_EXT
, "WRITE DMA QUEUED FUA EXT" },
2316 { ATA_CMD_FPDMA_READ
, "READ FPDMA QUEUED" },
2317 { ATA_CMD_FPDMA_WRITE
, "WRITE FPDMA QUEUED" },
2318 { ATA_CMD_FPDMA_SEND
, "SEND FPDMA QUEUED" },
2319 { ATA_CMD_FPDMA_RECV
, "RECEIVE FPDMA QUEUED" },
2320 { ATA_CMD_PIO_READ
, "READ SECTOR(S)" },
2321 { ATA_CMD_PIO_READ_EXT
, "READ SECTOR(S) EXT" },
2322 { ATA_CMD_PIO_WRITE
, "WRITE SECTOR(S)" },
2323 { ATA_CMD_PIO_WRITE_EXT
, "WRITE SECTOR(S) EXT" },
2324 { ATA_CMD_READ_MULTI
, "READ MULTIPLE" },
2325 { ATA_CMD_READ_MULTI_EXT
, "READ MULTIPLE EXT" },
2326 { ATA_CMD_WRITE_MULTI
, "WRITE MULTIPLE" },
2327 { ATA_CMD_WRITE_MULTI_EXT
, "WRITE MULTIPLE EXT" },
2328 { ATA_CMD_WRITE_MULTI_FUA_EXT
, "WRITE MULTIPLE FUA EXT" },
2329 { ATA_CMD_SET_FEATURES
, "SET FEATURES" },
2330 { ATA_CMD_SET_MULTI
, "SET MULTIPLE MODE" },
2331 { ATA_CMD_VERIFY
, "READ VERIFY SECTOR(S)" },
2332 { ATA_CMD_VERIFY_EXT
, "READ VERIFY SECTOR(S) EXT" },
2333 { ATA_CMD_WRITE_UNCORR_EXT
, "WRITE UNCORRECTABLE EXT" },
2334 { ATA_CMD_STANDBYNOW1
, "STANDBY IMMEDIATE" },
2335 { ATA_CMD_IDLEIMMEDIATE
, "IDLE IMMEDIATE" },
2336 { ATA_CMD_SLEEP
, "SLEEP" },
2337 { ATA_CMD_INIT_DEV_PARAMS
, "INITIALIZE DEVICE PARAMETERS" },
2338 { ATA_CMD_READ_NATIVE_MAX
, "READ NATIVE MAX ADDRESS" },
2339 { ATA_CMD_READ_NATIVE_MAX_EXT
, "READ NATIVE MAX ADDRESS EXT" },
2340 { ATA_CMD_SET_MAX
, "SET MAX ADDRESS" },
2341 { ATA_CMD_SET_MAX_EXT
, "SET MAX ADDRESS EXT" },
2342 { ATA_CMD_READ_LOG_EXT
, "READ LOG EXT" },
2343 { ATA_CMD_WRITE_LOG_EXT
, "WRITE LOG EXT" },
2344 { ATA_CMD_READ_LOG_DMA_EXT
, "READ LOG DMA EXT" },
2345 { ATA_CMD_WRITE_LOG_DMA_EXT
, "WRITE LOG DMA EXT" },
2346 { ATA_CMD_TRUSTED_NONDATA
, "TRUSTED NON-DATA" },
2347 { ATA_CMD_TRUSTED_RCV
, "TRUSTED RECEIVE" },
2348 { ATA_CMD_TRUSTED_RCV_DMA
, "TRUSTED RECEIVE DMA" },
2349 { ATA_CMD_TRUSTED_SND
, "TRUSTED SEND" },
2350 { ATA_CMD_TRUSTED_SND_DMA
, "TRUSTED SEND DMA" },
2351 { ATA_CMD_PMP_READ
, "READ BUFFER" },
2352 { ATA_CMD_PMP_READ_DMA
, "READ BUFFER DMA" },
2353 { ATA_CMD_PMP_WRITE
, "WRITE BUFFER" },
2354 { ATA_CMD_PMP_WRITE_DMA
, "WRITE BUFFER DMA" },
2355 { ATA_CMD_CONF_OVERLAY
, "DEVICE CONFIGURATION OVERLAY" },
2356 { ATA_CMD_SEC_SET_PASS
, "SECURITY SET PASSWORD" },
2357 { ATA_CMD_SEC_UNLOCK
, "SECURITY UNLOCK" },
2358 { ATA_CMD_SEC_ERASE_PREP
, "SECURITY ERASE PREPARE" },
2359 { ATA_CMD_SEC_ERASE_UNIT
, "SECURITY ERASE UNIT" },
2360 { ATA_CMD_SEC_FREEZE_LOCK
, "SECURITY FREEZE LOCK" },
2361 { ATA_CMD_SEC_DISABLE_PASS
, "SECURITY DISABLE PASSWORD" },
2362 { ATA_CMD_CONFIG_STREAM
, "CONFIGURE STREAM" },
2363 { ATA_CMD_SMART
, "SMART" },
2364 { ATA_CMD_MEDIA_LOCK
, "DOOR LOCK" },
2365 { ATA_CMD_MEDIA_UNLOCK
, "DOOR UNLOCK" },
2366 { ATA_CMD_DSM
, "DATA SET MANAGEMENT" },
2367 { ATA_CMD_CHK_MED_CRD_TYP
, "CHECK MEDIA CARD TYPE" },
2368 { ATA_CMD_CFA_REQ_EXT_ERR
, "CFA REQUEST EXTENDED ERROR" },
2369 { ATA_CMD_CFA_WRITE_NE
, "CFA WRITE SECTORS WITHOUT ERASE" },
2370 { ATA_CMD_CFA_TRANS_SECT
, "CFA TRANSLATE SECTOR" },
2371 { ATA_CMD_CFA_ERASE
, "CFA ERASE SECTORS" },
2372 { ATA_CMD_CFA_WRITE_MULT_NE
, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2373 { ATA_CMD_REQ_SENSE_DATA
, "REQUEST SENSE DATA EXT" },
2374 { ATA_CMD_SANITIZE_DEVICE
, "SANITIZE DEVICE" },
2375 { ATA_CMD_READ_LONG
, "READ LONG (with retries)" },
2376 { ATA_CMD_READ_LONG_ONCE
, "READ LONG (without retries)" },
2377 { ATA_CMD_WRITE_LONG
, "WRITE LONG (with retries)" },
2378 { ATA_CMD_WRITE_LONG_ONCE
, "WRITE LONG (without retries)" },
2379 { ATA_CMD_RESTORE
, "RECALIBRATE" },
2380 { 0, NULL
} /* terminate list */
2384 for (i
= 0; cmd_descr
[i
].text
; i
++)
2385 if (cmd_descr
[i
].command
== command
)
2386 return cmd_descr
[i
].text
;
2393 * ata_eh_link_report - report error handling to user
2394 * @link: ATA link EH is going on
2396 * Report EH to user.
2401 static void ata_eh_link_report(struct ata_link
*link
)
2403 struct ata_port
*ap
= link
->ap
;
2404 struct ata_eh_context
*ehc
= &link
->eh_context
;
2405 const char *frozen
, *desc
;
2406 char tries_buf
[6] = "";
2407 int tag
, nr_failed
= 0;
2409 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2413 if (ehc
->i
.desc
[0] != '\0')
2416 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2417 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2419 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2420 ata_dev_phys_link(qc
->dev
) != link
||
2421 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2422 qc
->err_mask
== AC_ERR_DEV
))
2424 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2430 if (!nr_failed
&& !ehc
->i
.err_mask
)
2434 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2437 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2438 snprintf(tries_buf
, sizeof(tries_buf
), " t%d",
2442 ata_dev_err(ehc
->i
.dev
, "exception Emask 0x%x "
2443 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2444 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2445 ehc
->i
.action
, frozen
, tries_buf
);
2447 ata_dev_err(ehc
->i
.dev
, "%s\n", desc
);
2449 ata_link_err(link
, "exception Emask 0x%x "
2450 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2451 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2452 ehc
->i
.action
, frozen
, tries_buf
);
2454 ata_link_err(link
, "%s\n", desc
);
2457 #ifdef CONFIG_ATA_VERBOSE_ERROR
2460 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2461 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2462 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2463 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2464 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2465 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2466 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2467 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2468 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2469 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2470 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2471 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2472 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2473 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2474 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2475 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2476 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2477 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2480 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2481 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2482 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2483 const u8
*cdb
= qc
->cdb
;
2484 char data_buf
[20] = "";
2485 char cdb_buf
[70] = "";
2487 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2488 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2491 if (qc
->dma_dir
!= DMA_NONE
) {
2492 static const char *dma_str
[] = {
2493 [DMA_BIDIRECTIONAL
] = "bidi",
2494 [DMA_TO_DEVICE
] = "out",
2495 [DMA_FROM_DEVICE
] = "in",
2497 static const char *prot_str
[] = {
2498 [ATA_PROT_PIO
] = "pio",
2499 [ATA_PROT_DMA
] = "dma",
2500 [ATA_PROT_NCQ
] = "ncq",
2501 [ATAPI_PROT_PIO
] = "pio",
2502 [ATAPI_PROT_DMA
] = "dma",
2505 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2506 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2507 dma_str
[qc
->dma_dir
]);
2510 if (ata_is_atapi(qc
->tf
.protocol
)) {
2512 scsi_print_command(qc
->scsicmd
);
2514 snprintf(cdb_buf
, sizeof(cdb_buf
),
2515 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2516 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2517 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2518 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2519 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2520 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2522 const char *descr
= ata_get_cmd_descript(cmd
->command
);
2524 ata_dev_err(qc
->dev
, "failed command: %s\n",
2528 ata_dev_err(qc
->dev
,
2529 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2531 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2532 "Emask 0x%x (%s)%s\n",
2533 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2534 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2535 cmd
->hob_feature
, cmd
->hob_nsect
,
2536 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2537 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2538 res
->command
, res
->feature
, res
->nsect
,
2539 res
->lbal
, res
->lbam
, res
->lbah
,
2540 res
->hob_feature
, res
->hob_nsect
,
2541 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2542 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2543 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2545 #ifdef CONFIG_ATA_VERBOSE_ERROR
2546 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2548 if (res
->command
& ATA_BUSY
)
2549 ata_dev_err(qc
->dev
, "status: { Busy }\n");
2551 ata_dev_err(qc
->dev
, "status: { %s%s%s%s}\n",
2552 res
->command
& ATA_DRDY
? "DRDY " : "",
2553 res
->command
& ATA_DF
? "DF " : "",
2554 res
->command
& ATA_DRQ
? "DRQ " : "",
2555 res
->command
& ATA_ERR
? "ERR " : "");
2558 if (cmd
->command
!= ATA_CMD_PACKET
&&
2559 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2561 ata_dev_err(qc
->dev
, "error: { %s%s%s%s}\n",
2562 res
->feature
& ATA_ICRC
? "ICRC " : "",
2563 res
->feature
& ATA_UNC
? "UNC " : "",
2564 res
->feature
& ATA_IDNF
? "IDNF " : "",
2565 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2571 * ata_eh_report - report error handling to user
2572 * @ap: ATA port to report EH about
2574 * Report EH to user.
2579 void ata_eh_report(struct ata_port
*ap
)
2581 struct ata_link
*link
;
2583 ata_for_each_link(link
, ap
, HOST_FIRST
)
2584 ata_eh_link_report(link
);
2587 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2588 unsigned int *classes
, unsigned long deadline
,
2591 struct ata_device
*dev
;
2594 ata_for_each_dev(dev
, link
, ALL
)
2595 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2597 return reset(link
, classes
, deadline
);
2600 static int ata_eh_followup_srst_needed(struct ata_link
*link
, int rc
)
2602 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2606 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2611 int ata_eh_reset(struct ata_link
*link
, int classify
,
2612 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2613 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2615 struct ata_port
*ap
= link
->ap
;
2616 struct ata_link
*slave
= ap
->slave_link
;
2617 struct ata_eh_context
*ehc
= &link
->eh_context
;
2618 struct ata_eh_context
*sehc
= slave
? &slave
->eh_context
: NULL
;
2619 unsigned int *classes
= ehc
->classes
;
2620 unsigned int lflags
= link
->flags
;
2621 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2622 int max_tries
= 0, try = 0;
2623 struct ata_link
*failed_link
;
2624 struct ata_device
*dev
;
2625 unsigned long deadline
, now
;
2626 ata_reset_fn_t reset
;
2627 unsigned long flags
;
2634 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2636 if (link
->flags
& ATA_LFLAG_RST_ONCE
)
2638 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2640 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2643 /* make sure each reset attempt is at least COOL_DOWN apart */
2644 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2646 WARN_ON(time_after(ehc
->last_reset
, now
));
2647 deadline
= ata_deadline(ehc
->last_reset
,
2648 ATA_EH_RESET_COOL_DOWN
);
2649 if (time_before(now
, deadline
))
2650 schedule_timeout_uninterruptible(deadline
- now
);
2653 spin_lock_irqsave(ap
->lock
, flags
);
2654 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2655 spin_unlock_irqrestore(ap
->lock
, flags
);
2657 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2659 ata_for_each_dev(dev
, link
, ALL
) {
2660 /* If we issue an SRST then an ATA drive (not ATAPI)
2661 * may change configuration and be in PIO0 timing. If
2662 * we do a hard reset (or are coming from power on)
2663 * this is true for ATA or ATAPI. Until we've set a
2664 * suitable controller mode we should not touch the
2665 * bus as we may be talking too fast.
2667 dev
->pio_mode
= XFER_PIO_0
;
2668 dev
->dma_mode
= 0xff;
2670 /* If the controller has a pio mode setup function
2671 * then use it to set the chipset to rights. Don't
2672 * touch the DMA setup as that will be dealt with when
2673 * configuring devices.
2675 if (ap
->ops
->set_piomode
)
2676 ap
->ops
->set_piomode(ap
, dev
);
2679 /* prefer hardreset */
2681 ehc
->i
.action
&= ~ATA_EH_RESET
;
2684 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2685 } else if (softreset
) {
2687 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2691 unsigned long deadline
= ata_deadline(jiffies
,
2692 ATA_EH_PRERESET_TIMEOUT
);
2695 sehc
->i
.action
&= ~ATA_EH_RESET
;
2696 sehc
->i
.action
|= ehc
->i
.action
;
2699 rc
= prereset(link
, deadline
);
2701 /* If present, do prereset on slave link too. Reset
2702 * is skipped iff both master and slave links report
2703 * -ENOENT or clear ATA_EH_RESET.
2705 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2708 tmp
= prereset(slave
, deadline
);
2712 ehc
->i
.action
|= sehc
->i
.action
;
2716 if (rc
== -ENOENT
) {
2717 ata_link_dbg(link
, "port disabled--ignoring\n");
2718 ehc
->i
.action
&= ~ATA_EH_RESET
;
2720 ata_for_each_dev(dev
, link
, ALL
)
2721 classes
[dev
->devno
] = ATA_DEV_NONE
;
2726 "prereset failed (errno=%d)\n",
2731 /* prereset() might have cleared ATA_EH_RESET. If so,
2732 * bang classes, thaw and return.
2734 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2735 ata_for_each_dev(dev
, link
, ALL
)
2736 classes
[dev
->devno
] = ATA_DEV_NONE
;
2737 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) &&
2738 ata_is_host_link(link
))
2739 ata_eh_thaw_port(ap
);
2749 if (ata_is_host_link(link
))
2750 ata_eh_freeze_port(ap
);
2752 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2756 ata_link_info(link
, "%s resetting link\n",
2757 reset
== softreset
? "soft" : "hard");
2759 /* mark that this EH session started with reset */
2760 ehc
->last_reset
= jiffies
;
2761 if (reset
== hardreset
)
2762 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2764 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2766 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2767 if (rc
&& rc
!= -EAGAIN
) {
2772 /* hardreset slave link if existent */
2773 if (slave
&& reset
== hardreset
) {
2777 ata_link_info(slave
, "hard resetting link\n");
2779 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2780 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2788 failed_link
= slave
;
2794 /* perform follow-up SRST if necessary */
2795 if (reset
== hardreset
&&
2796 ata_eh_followup_srst_needed(link
, rc
)) {
2801 "follow-up softreset required but no softreset available\n");
2807 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2808 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2817 "no reset method available, skipping reset\n");
2818 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2819 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2823 * Post-reset processing
2825 ata_for_each_dev(dev
, link
, ALL
) {
2826 /* After the reset, the device state is PIO 0 and the
2827 * controller state is undefined. Reset also wakes up
2828 * drives from sleeping mode.
2830 dev
->pio_mode
= XFER_PIO_0
;
2831 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2833 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
2836 /* apply class override */
2837 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2838 classes
[dev
->devno
] = ATA_DEV_ATA
;
2839 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2840 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2843 /* record current link speed */
2844 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2845 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2846 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2847 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2850 if (ata_is_host_link(link
))
2851 ata_eh_thaw_port(ap
);
2853 /* postreset() should clear hardware SError. Although SError
2854 * is cleared during link resume, clearing SError here is
2855 * necessary as some PHYs raise hotplug events after SRST.
2856 * This introduces race condition where hotplug occurs between
2857 * reset and here. This race is mediated by cross checking
2858 * link onlineness and classification result later.
2861 postreset(link
, classes
);
2863 postreset(slave
, classes
);
2867 * Some controllers can't be frozen very well and may set spurious
2868 * error conditions during reset. Clear accumulated error
2869 * information and re-thaw the port if frozen. As reset is the
2870 * final recovery action and we cross check link onlineness against
2871 * device classification later, no hotplug event is lost by this.
2873 spin_lock_irqsave(link
->ap
->lock
, flags
);
2874 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
2876 memset(&slave
->eh_info
, 0, sizeof(link
->eh_info
));
2877 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
2878 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2880 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2881 ata_eh_thaw_port(ap
);
2884 * Make sure onlineness and classification result correspond.
2885 * Hotplug could have happened during reset and some
2886 * controllers fail to wait while a drive is spinning up after
2887 * being hotplugged causing misdetection. By cross checking
2888 * link on/offlineness and classification result, those
2889 * conditions can be reliably detected and retried.
2892 ata_for_each_dev(dev
, link
, ALL
) {
2893 if (ata_phys_link_online(ata_dev_phys_link(dev
))) {
2894 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2895 ata_dev_dbg(dev
, "link online but device misclassified\n");
2896 classes
[dev
->devno
] = ATA_DEV_NONE
;
2899 } else if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2900 if (ata_class_enabled(classes
[dev
->devno
]))
2902 "link offline, clearing class %d to NONE\n",
2903 classes
[dev
->devno
]);
2904 classes
[dev
->devno
] = ATA_DEV_NONE
;
2905 } else if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2907 "link status unknown, clearing UNKNOWN to NONE\n");
2908 classes
[dev
->devno
] = ATA_DEV_NONE
;
2912 if (classify
&& nr_unknown
) {
2913 if (try < max_tries
) {
2915 "link online but %d devices misclassified, retrying\n",
2922 "link online but %d devices misclassified, "
2923 "device detection might fail\n", nr_unknown
);
2926 /* reset successful, schedule revalidation */
2927 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2929 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2930 ehc
->last_reset
= jiffies
; /* update to completion time */
2931 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2932 link
->lpm_policy
= ATA_LPM_UNKNOWN
; /* reset LPM state */
2936 /* clear hotplug flag */
2937 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2939 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2941 spin_lock_irqsave(ap
->lock
, flags
);
2942 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2943 spin_unlock_irqrestore(ap
->lock
, flags
);
2948 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2949 if (!ata_is_host_link(link
) &&
2950 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2953 if (try >= max_tries
) {
2955 * Thaw host port even if reset failed, so that the port
2956 * can be retried on the next phy event. This risks
2957 * repeated EH runs but seems to be a better tradeoff than
2958 * shutting down a port after a botched hotplug attempt.
2960 if (ata_is_host_link(link
))
2961 ata_eh_thaw_port(ap
);
2966 if (time_before(now
, deadline
)) {
2967 unsigned long delta
= deadline
- now
;
2969 ata_link_warn(failed_link
,
2970 "reset failed (errno=%d), retrying in %u secs\n",
2971 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2975 delta
= schedule_timeout_uninterruptible(delta
);
2980 * While disks spinup behind PMP, some controllers fail sending SRST.
2981 * They need to be reset - as well as the PMP - before retrying.
2983 if (rc
== -ERESTART
) {
2984 if (ata_is_host_link(link
))
2985 ata_eh_thaw_port(ap
);
2989 if (try == max_tries
- 1) {
2990 sata_down_spd_limit(link
, 0);
2992 sata_down_spd_limit(slave
, 0);
2993 } else if (rc
== -EPIPE
)
2994 sata_down_spd_limit(failed_link
, 0);
3001 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
3003 struct ata_link
*link
;
3004 struct ata_device
*dev
;
3005 unsigned long flags
;
3008 * This function can be thought of as an extended version of
3009 * ata_eh_about_to_do() specially crafted to accommodate the
3010 * requirements of ATA_EH_PARK handling. Since the EH thread
3011 * does not leave the do {} while () loop in ata_eh_recover as
3012 * long as the timeout for a park request to *one* device on
3013 * the port has not expired, and since we still want to pick
3014 * up park requests to other devices on the same port or
3015 * timeout updates for the same device, we have to pull
3016 * ATA_EH_PARK actions from eh_info into eh_context.i
3017 * ourselves at the beginning of each pass over the loop.
3019 * Additionally, all write accesses to &ap->park_req_pending
3020 * through reinit_completion() (see below) or complete_all()
3021 * (see ata_scsi_park_store()) are protected by the host lock.
3022 * As a result we have that park_req_pending.done is zero on
3023 * exit from this function, i.e. when ATA_EH_PARK actions for
3024 * *all* devices on port ap have been pulled into the
3025 * respective eh_context structs. If, and only if,
3026 * park_req_pending.done is non-zero by the time we reach
3027 * wait_for_completion_timeout(), another ATA_EH_PARK action
3028 * has been scheduled for at least one of the devices on port
3029 * ap and we have to cycle over the do {} while () loop in
3030 * ata_eh_recover() again.
3033 spin_lock_irqsave(ap
->lock
, flags
);
3034 reinit_completion(&ap
->park_req_pending
);
3035 ata_for_each_link(link
, ap
, EDGE
) {
3036 ata_for_each_dev(dev
, link
, ALL
) {
3037 struct ata_eh_info
*ehi
= &link
->eh_info
;
3039 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
3040 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
3041 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
3044 spin_unlock_irqrestore(ap
->lock
, flags
);
3047 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
3049 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3050 struct ata_taskfile tf
;
3051 unsigned int err_mask
;
3053 ata_tf_init(dev
, &tf
);
3055 ehc
->unloaded_mask
|= 1 << dev
->devno
;
3056 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
3062 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3063 tf
.command
= ATA_CMD_CHK_POWER
;
3066 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
3067 tf
.protocol
|= ATA_PROT_NODATA
;
3068 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3069 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
3070 ata_dev_err(dev
, "head unload failed!\n");
3071 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
3075 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
3076 struct ata_device
**r_failed_dev
)
3078 struct ata_port
*ap
= link
->ap
;
3079 struct ata_eh_context
*ehc
= &link
->eh_context
;
3080 struct ata_device
*dev
;
3081 unsigned int new_mask
= 0;
3082 unsigned long flags
;
3087 /* For PATA drive side cable detection to work, IDENTIFY must
3088 * be done backwards such that PDIAG- is released by the slave
3089 * device before the master device is identified.
3091 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
3092 unsigned int action
= ata_eh_dev_action(dev
);
3093 unsigned int readid_flags
= 0;
3095 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
3096 readid_flags
|= ATA_READID_POSTRESET
;
3098 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
3099 WARN_ON(dev
->class == ATA_DEV_PMP
);
3101 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
3106 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
3107 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
3112 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
3114 /* Configuration may have changed, reconfigure
3117 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3119 /* schedule the scsi_rescan_device() here */
3120 schedule_work(&(ap
->scsi_rescan_task
));
3121 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
3122 ehc
->tries
[dev
->devno
] &&
3123 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
3124 /* Temporarily set dev->class, it will be
3125 * permanently set once all configurations are
3126 * complete. This is necessary because new
3127 * device configuration is done in two
3130 dev
->class = ehc
->classes
[dev
->devno
];
3132 if (dev
->class == ATA_DEV_PMP
)
3133 rc
= sata_pmp_attach(dev
);
3135 rc
= ata_dev_read_id(dev
, &dev
->class,
3136 readid_flags
, dev
->id
);
3138 /* read_id might have changed class, store and reset */
3139 ehc
->classes
[dev
->devno
] = dev
->class;
3140 dev
->class = ATA_DEV_UNKNOWN
;
3144 /* clear error info accumulated during probe */
3145 ata_ering_clear(&dev
->ering
);
3146 new_mask
|= 1 << dev
->devno
;
3149 /* IDENTIFY was issued to non-existent
3150 * device. No need to reset. Just
3151 * thaw and ignore the device.
3153 ata_eh_thaw_port(ap
);
3161 /* PDIAG- should have been released, ask cable type if post-reset */
3162 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
3163 if (ap
->ops
->cable_detect
)
3164 ap
->cbl
= ap
->ops
->cable_detect(ap
);
3168 /* Configure new devices forward such that user doesn't see
3169 * device detection messages backwards.
3171 ata_for_each_dev(dev
, link
, ALL
) {
3172 if (!(new_mask
& (1 << dev
->devno
)))
3175 dev
->class = ehc
->classes
[dev
->devno
];
3177 if (dev
->class == ATA_DEV_PMP
)
3180 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
3181 rc
= ata_dev_configure(dev
);
3182 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
3184 dev
->class = ATA_DEV_UNKNOWN
;
3188 spin_lock_irqsave(ap
->lock
, flags
);
3189 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
3190 spin_unlock_irqrestore(ap
->lock
, flags
);
3192 /* new device discovered, configure xfermode */
3193 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
3199 *r_failed_dev
= dev
;
3200 DPRINTK("EXIT rc=%d\n", rc
);
3205 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3206 * @link: link on which timings will be programmed
3207 * @r_failed_dev: out parameter for failed device
3209 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3210 * ata_set_mode() fails, pointer to the failing device is
3211 * returned in @r_failed_dev.
3214 * PCI/etc. bus probe sem.
3217 * 0 on success, negative errno otherwise
3219 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
3221 struct ata_port
*ap
= link
->ap
;
3222 struct ata_device
*dev
;
3225 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3226 ata_for_each_dev(dev
, link
, ENABLED
) {
3227 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
3228 struct ata_ering_entry
*ent
;
3230 ent
= ata_ering_top(&dev
->ering
);
3232 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
3236 /* has private set_mode? */
3237 if (ap
->ops
->set_mode
)
3238 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
3240 rc
= ata_do_set_mode(link
, r_failed_dev
);
3242 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3243 ata_for_each_dev(dev
, link
, ENABLED
) {
3244 struct ata_eh_context
*ehc
= &link
->eh_context
;
3245 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
3246 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
3248 if (dev
->xfer_mode
!= saved_xfer_mode
||
3249 ata_ncq_enabled(dev
) != saved_ncq
)
3250 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
3257 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3258 * @dev: ATAPI device to clear UA for
3260 * Resets and other operations can make an ATAPI device raise
3261 * UNIT ATTENTION which causes the next operation to fail. This
3262 * function clears UA.
3265 * EH context (may sleep).
3268 * 0 on success, -errno on failure.
3270 static int atapi_eh_clear_ua(struct ata_device
*dev
)
3274 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
3275 u8
*sense_buffer
= dev
->link
->ap
->sector_buf
;
3277 unsigned int err_mask
;
3279 err_mask
= atapi_eh_tur(dev
, &sense_key
);
3280 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
3282 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3287 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
3290 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
3292 ata_dev_warn(dev
, "failed to clear "
3293 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
3298 ata_dev_warn(dev
, "UNIT ATTENTION persists after %d tries\n",
3305 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3306 * @dev: ATA device which may need FLUSH retry
3308 * If @dev failed FLUSH, it needs to be reported upper layer
3309 * immediately as it means that @dev failed to remap and already
3310 * lost at least a sector and further FLUSH retrials won't make
3311 * any difference to the lost sector. However, if FLUSH failed
3312 * for other reasons, for example transmission error, FLUSH needs
3315 * This function determines whether FLUSH failure retry is
3316 * necessary and performs it if so.
3319 * 0 if EH can continue, -errno if EH needs to be repeated.
3321 static int ata_eh_maybe_retry_flush(struct ata_device
*dev
)
3323 struct ata_link
*link
= dev
->link
;
3324 struct ata_port
*ap
= link
->ap
;
3325 struct ata_queued_cmd
*qc
;
3326 struct ata_taskfile tf
;
3327 unsigned int err_mask
;
3330 /* did flush fail for this device? */
3331 if (!ata_tag_valid(link
->active_tag
))
3334 qc
= __ata_qc_from_tag(ap
, link
->active_tag
);
3335 if (qc
->dev
!= dev
|| (qc
->tf
.command
!= ATA_CMD_FLUSH_EXT
&&
3336 qc
->tf
.command
!= ATA_CMD_FLUSH
))
3339 /* if the device failed it, it should be reported to upper layers */
3340 if (qc
->err_mask
& AC_ERR_DEV
)
3343 /* flush failed for some other reason, give it another shot */
3344 ata_tf_init(dev
, &tf
);
3346 tf
.command
= qc
->tf
.command
;
3347 tf
.flags
|= ATA_TFLAG_DEVICE
;
3348 tf
.protocol
= ATA_PROT_NODATA
;
3350 ata_dev_warn(dev
, "retrying FLUSH 0x%x Emask 0x%x\n",
3351 tf
.command
, qc
->err_mask
);
3353 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
3356 * FLUSH is complete but there's no way to
3357 * successfully complete a failed command from EH.
3358 * Making sure retry is allowed at least once and
3359 * retrying it should do the trick - whatever was in
3360 * the cache is already on the platter and this won't
3361 * cause infinite loop.
3363 qc
->scsicmd
->allowed
= max(qc
->scsicmd
->allowed
, 1);
3365 ata_dev_warn(dev
, "FLUSH failed Emask 0x%x\n",
3369 /* if device failed it, report it to upper layers */
3370 if (err_mask
& AC_ERR_DEV
) {
3371 qc
->err_mask
|= AC_ERR_DEV
;
3373 if (!(ap
->pflags
& ATA_PFLAG_FROZEN
))
3381 * ata_eh_set_lpm - configure SATA interface power management
3382 * @link: link to configure power management
3383 * @policy: the link power management policy
3384 * @r_failed_dev: out parameter for failed device
3386 * Enable SATA Interface power management. This will enable
3387 * Device Interface Power Management (DIPM) for min_power
3388 * policy, and then call driver specific callbacks for
3389 * enabling Host Initiated Power management.
3395 * 0 on success, -errno on failure.
3397 static int ata_eh_set_lpm(struct ata_link
*link
, enum ata_lpm_policy policy
,
3398 struct ata_device
**r_failed_dev
)
3400 struct ata_port
*ap
= ata_is_host_link(link
) ? link
->ap
: NULL
;
3401 struct ata_eh_context
*ehc
= &link
->eh_context
;
3402 struct ata_device
*dev
, *link_dev
= NULL
, *lpm_dev
= NULL
;
3403 enum ata_lpm_policy old_policy
= link
->lpm_policy
;
3404 bool no_dipm
= link
->ap
->flags
& ATA_FLAG_NO_DIPM
;
3405 unsigned int hints
= ATA_LPM_EMPTY
| ATA_LPM_HIPM
;
3406 unsigned int err_mask
;
3409 /* if the link or host doesn't do LPM, noop */
3410 if ((link
->flags
& ATA_LFLAG_NO_LPM
) || (ap
&& !ap
->ops
->set_lpm
))
3414 * DIPM is enabled only for MIN_POWER as some devices
3415 * misbehave when the host NACKs transition to SLUMBER. Order
3416 * device and link configurations such that the host always
3417 * allows DIPM requests.
3419 ata_for_each_dev(dev
, link
, ENABLED
) {
3420 bool hipm
= ata_id_has_hipm(dev
->id
);
3421 bool dipm
= ata_id_has_dipm(dev
->id
) && !no_dipm
;
3423 /* find the first enabled and LPM enabled devices */
3427 if (!lpm_dev
&& (hipm
|| dipm
))
3430 hints
&= ~ATA_LPM_EMPTY
;
3432 hints
&= ~ATA_LPM_HIPM
;
3434 /* disable DIPM before changing link config */
3435 if (policy
!= ATA_LPM_MIN_POWER
&& dipm
) {
3436 err_mask
= ata_dev_set_feature(dev
,
3437 SETFEATURES_SATA_DISABLE
, SATA_DIPM
);
3438 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3440 "failed to disable DIPM, Emask 0x%x\n",
3449 rc
= ap
->ops
->set_lpm(link
, policy
, hints
);
3450 if (!rc
&& ap
->slave_link
)
3451 rc
= ap
->ops
->set_lpm(ap
->slave_link
, policy
, hints
);
3453 rc
= sata_pmp_set_lpm(link
, policy
, hints
);
3456 * Attribute link config failure to the first (LPM) enabled
3457 * device on the link.
3460 if (rc
== -EOPNOTSUPP
) {
3461 link
->flags
|= ATA_LFLAG_NO_LPM
;
3464 dev
= lpm_dev
? lpm_dev
: link_dev
;
3469 * Low level driver acked the transition. Issue DIPM command
3470 * with the new policy set.
3472 link
->lpm_policy
= policy
;
3473 if (ap
&& ap
->slave_link
)
3474 ap
->slave_link
->lpm_policy
= policy
;
3476 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3477 ata_for_each_dev(dev
, link
, ENABLED
) {
3478 if (policy
== ATA_LPM_MIN_POWER
&& !no_dipm
&&
3479 ata_id_has_dipm(dev
->id
)) {
3480 err_mask
= ata_dev_set_feature(dev
,
3481 SETFEATURES_SATA_ENABLE
, SATA_DIPM
);
3482 if (err_mask
&& err_mask
!= AC_ERR_DEV
) {
3484 "failed to enable DIPM, Emask 0x%x\n",
3495 /* restore the old policy */
3496 link
->lpm_policy
= old_policy
;
3497 if (ap
&& ap
->slave_link
)
3498 ap
->slave_link
->lpm_policy
= old_policy
;
3500 /* if no device or only one more chance is left, disable LPM */
3501 if (!dev
|| ehc
->tries
[dev
->devno
] <= 2) {
3502 ata_link_warn(link
, "disabling LPM on the link\n");
3503 link
->flags
|= ATA_LFLAG_NO_LPM
;
3506 *r_failed_dev
= dev
;
3510 int ata_link_nr_enabled(struct ata_link
*link
)
3512 struct ata_device
*dev
;
3515 ata_for_each_dev(dev
, link
, ENABLED
)
3520 static int ata_link_nr_vacant(struct ata_link
*link
)
3522 struct ata_device
*dev
;
3525 ata_for_each_dev(dev
, link
, ALL
)
3526 if (dev
->class == ATA_DEV_UNKNOWN
)
3531 static int ata_eh_skip_recovery(struct ata_link
*link
)
3533 struct ata_port
*ap
= link
->ap
;
3534 struct ata_eh_context
*ehc
= &link
->eh_context
;
3535 struct ata_device
*dev
;
3537 /* skip disabled links */
3538 if (link
->flags
& ATA_LFLAG_DISABLED
)
3541 /* skip if explicitly requested */
3542 if (ehc
->i
.flags
& ATA_EHI_NO_RECOVERY
)
3545 /* thaw frozen port and recover failed devices */
3546 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
3549 /* reset at least once if reset is requested */
3550 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
3551 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
3554 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3555 ata_for_each_dev(dev
, link
, ALL
) {
3556 if (dev
->class == ATA_DEV_UNKNOWN
&&
3557 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
3564 static int ata_count_probe_trials_cb(struct ata_ering_entry
*ent
, void *void_arg
)
3566 u64 interval
= msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL
);
3567 u64 now
= get_jiffies_64();
3568 int *trials
= void_arg
;
3570 if ((ent
->eflags
& ATA_EFLAG_OLD_ER
) ||
3571 (ent
->timestamp
< now
- min(now
, interval
)))
3578 static int ata_eh_schedule_probe(struct ata_device
*dev
)
3580 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3581 struct ata_link
*link
= ata_dev_phys_link(dev
);
3584 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
3585 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
3588 ata_eh_detach_dev(dev
);
3590 ehc
->did_probe_mask
|= (1 << dev
->devno
);
3591 ehc
->i
.action
|= ATA_EH_RESET
;
3592 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
3593 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
3595 /* the link maybe in a deep sleep, wake it up */
3596 if (link
->lpm_policy
> ATA_LPM_MAX_POWER
) {
3597 if (ata_is_host_link(link
))
3598 link
->ap
->ops
->set_lpm(link
, ATA_LPM_MAX_POWER
,
3601 sata_pmp_set_lpm(link
, ATA_LPM_MAX_POWER
,
3605 /* Record and count probe trials on the ering. The specific
3606 * error mask used is irrelevant. Because a successful device
3607 * detection clears the ering, this count accumulates only if
3608 * there are consecutive failed probes.
3610 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3611 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3612 * forced to 1.5Gbps.
3614 * This is to work around cases where failed link speed
3615 * negotiation results in device misdetection leading to
3616 * infinite DEVXCHG or PHRDY CHG events.
3618 ata_ering_record(&dev
->ering
, 0, AC_ERR_OTHER
);
3619 ata_ering_map(&dev
->ering
, ata_count_probe_trials_cb
, &trials
);
3621 if (trials
> ATA_EH_PROBE_TRIALS
)
3622 sata_down_spd_limit(link
, 1);
3627 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
3629 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3631 /* -EAGAIN from EH routine indicates retry without prejudice.
3632 * The requester is responsible for ensuring forward progress.
3635 ehc
->tries
[dev
->devno
]--;
3639 /* device missing or wrong IDENTIFY data, schedule probing */
3640 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
3642 /* give it just one more chance */
3643 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
3645 if (ehc
->tries
[dev
->devno
] == 1) {
3646 /* This is the last chance, better to slow
3647 * down than lose it.
3649 sata_down_spd_limit(ata_dev_phys_link(dev
), 0);
3650 if (dev
->pio_mode
> XFER_PIO_0
)
3651 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
3655 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
3656 /* disable device if it has used up all its chances */
3657 ata_dev_disable(dev
);
3659 /* detach if offline */
3660 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
3661 ata_eh_detach_dev(dev
);
3663 /* schedule probe if necessary */
3664 if (ata_eh_schedule_probe(dev
)) {
3665 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3666 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3667 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3672 ehc
->i
.action
|= ATA_EH_RESET
;
3678 * ata_eh_recover - recover host port after error
3679 * @ap: host port to recover
3680 * @prereset: prereset method (can be NULL)
3681 * @softreset: softreset method (can be NULL)
3682 * @hardreset: hardreset method (can be NULL)
3683 * @postreset: postreset method (can be NULL)
3684 * @r_failed_link: out parameter for failed link
3686 * This is the alpha and omega, eum and yang, heart and soul of
3687 * libata exception handling. On entry, actions required to
3688 * recover each link and hotplug requests are recorded in the
3689 * link's eh_context. This function executes all the operations
3690 * with appropriate retrials and fallbacks to resurrect failed
3691 * devices, detach goners and greet newcomers.
3694 * Kernel thread context (may sleep).
3697 * 0 on success, -errno on failure.
3699 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3700 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3701 ata_postreset_fn_t postreset
,
3702 struct ata_link
**r_failed_link
)
3704 struct ata_link
*link
;
3705 struct ata_device
*dev
;
3707 unsigned long flags
, deadline
;
3711 /* prep for recovery */
3712 ata_for_each_link(link
, ap
, EDGE
) {
3713 struct ata_eh_context
*ehc
= &link
->eh_context
;
3715 /* re-enable link? */
3716 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3717 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3718 spin_lock_irqsave(ap
->lock
, flags
);
3719 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3720 spin_unlock_irqrestore(ap
->lock
, flags
);
3721 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3724 ata_for_each_dev(dev
, link
, ALL
) {
3725 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3726 ehc
->tries
[dev
->devno
] = 1;
3728 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3730 /* collect port action mask recorded in dev actions */
3731 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3732 ~ATA_EH_PERDEV_MASK
;
3733 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3735 /* process hotplug request */
3736 if (dev
->flags
& ATA_DFLAG_DETACH
)
3737 ata_eh_detach_dev(dev
);
3739 /* schedule probe if necessary */
3740 if (!ata_dev_enabled(dev
))
3741 ata_eh_schedule_probe(dev
);
3748 /* if UNLOADING, finish immediately */
3749 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3753 ata_for_each_link(link
, ap
, EDGE
) {
3754 struct ata_eh_context
*ehc
= &link
->eh_context
;
3756 /* skip EH if possible. */
3757 if (ata_eh_skip_recovery(link
))
3760 ata_for_each_dev(dev
, link
, ALL
)
3761 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3765 ata_for_each_link(link
, ap
, EDGE
) {
3766 struct ata_eh_context
*ehc
= &link
->eh_context
;
3768 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3771 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3772 prereset
, softreset
, hardreset
, postreset
);
3774 ata_link_err(link
, "reset failed, giving up\n");
3783 * clears ATA_EH_PARK in eh_info and resets
3784 * ap->park_req_pending
3786 ata_eh_pull_park_action(ap
);
3789 ata_for_each_link(link
, ap
, EDGE
) {
3790 ata_for_each_dev(dev
, link
, ALL
) {
3791 struct ata_eh_context
*ehc
= &link
->eh_context
;
3794 if (dev
->class != ATA_DEV_ATA
)
3796 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3799 tmp
= dev
->unpark_deadline
;
3800 if (time_before(deadline
, tmp
))
3802 else if (time_before_eq(tmp
, jiffies
))
3804 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3807 ata_eh_park_issue_cmd(dev
, 1);
3812 if (time_before_eq(deadline
, now
))
3816 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3820 ata_for_each_link(link
, ap
, EDGE
) {
3821 ata_for_each_dev(dev
, link
, ALL
) {
3822 if (!(link
->eh_context
.unloaded_mask
&
3826 ata_eh_park_issue_cmd(dev
, 0);
3827 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3833 ata_for_each_link(link
, ap
, PMP_FIRST
) {
3834 struct ata_eh_context
*ehc
= &link
->eh_context
;
3836 if (sata_pmp_attached(ap
) && ata_is_host_link(link
))
3839 /* revalidate existing devices and attach new ones */
3840 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3844 /* if PMP got attached, return, pmp EH will take care of it */
3845 if (link
->device
->class == ATA_DEV_PMP
) {
3850 /* configure transfer mode if necessary */
3851 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3852 rc
= ata_set_mode(link
, &dev
);
3855 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3858 /* If reset has been issued, clear UA to avoid
3859 * disrupting the current users of the device.
3861 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3862 ata_for_each_dev(dev
, link
, ALL
) {
3863 if (dev
->class != ATA_DEV_ATAPI
)
3865 rc
= atapi_eh_clear_ua(dev
);
3868 if (zpodd_dev_enabled(dev
))
3869 zpodd_post_poweron(dev
);
3873 /* retry flush if necessary */
3874 ata_for_each_dev(dev
, link
, ALL
) {
3875 if (dev
->class != ATA_DEV_ATA
)
3877 rc
= ata_eh_maybe_retry_flush(dev
);
3883 /* configure link power saving */
3884 if (link
->lpm_policy
!= ap
->target_lpm_policy
) {
3885 rc
= ata_eh_set_lpm(link
, ap
->target_lpm_policy
, &dev
);
3890 /* this link is okay now */
3897 ata_eh_handle_dev_fail(dev
, rc
);
3899 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3900 /* PMP reset requires working host port.
3901 * Can't retry if it's frozen.
3903 if (sata_pmp_attached(ap
))
3913 if (rc
&& r_failed_link
)
3914 *r_failed_link
= link
;
3916 DPRINTK("EXIT, rc=%d\n", rc
);
3921 * ata_eh_finish - finish up EH
3922 * @ap: host port to finish EH for
3924 * Recovery is complete. Clean up EH states and retry or finish
3930 void ata_eh_finish(struct ata_port
*ap
)
3934 /* retry or finish qcs */
3935 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3936 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3938 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3942 /* FIXME: Once EH migration is complete,
3943 * generate sense data in this function,
3944 * considering both err_mask and tf.
3946 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3947 ata_eh_qc_retry(qc
);
3949 ata_eh_qc_complete(qc
);
3951 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3952 ata_eh_qc_complete(qc
);
3954 /* feed zero TF to sense generation */
3955 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3956 ata_eh_qc_retry(qc
);
3961 /* make sure nr_active_links is zero after EH */
3962 WARN_ON(ap
->nr_active_links
);
3963 ap
->nr_active_links
= 0;
3967 * ata_do_eh - do standard error handling
3968 * @ap: host port to handle error for
3970 * @prereset: prereset method (can be NULL)
3971 * @softreset: softreset method (can be NULL)
3972 * @hardreset: hardreset method (can be NULL)
3973 * @postreset: postreset method (can be NULL)
3975 * Perform standard error handling sequence.
3978 * Kernel thread context (may sleep).
3980 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3981 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3982 ata_postreset_fn_t postreset
)
3984 struct ata_device
*dev
;
3990 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3993 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3994 ata_dev_disable(dev
);
4001 * ata_std_error_handler - standard error handler
4002 * @ap: host port to handle error for
4004 * Standard error handler
4007 * Kernel thread context (may sleep).
4009 void ata_std_error_handler(struct ata_port
*ap
)
4011 struct ata_port_operations
*ops
= ap
->ops
;
4012 ata_reset_fn_t hardreset
= ops
->hardreset
;
4014 /* ignore built-in hardreset if SCR access is not available */
4015 if (hardreset
== sata_std_hardreset
&& !sata_scr_valid(&ap
->link
))
4018 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
4023 * ata_eh_handle_port_suspend - perform port suspend operation
4024 * @ap: port to suspend
4029 * Kernel thread context (may sleep).
4031 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
4033 unsigned long flags
;
4035 struct ata_device
*dev
;
4037 /* are we suspending? */
4038 spin_lock_irqsave(ap
->lock
, flags
);
4039 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4040 ap
->pm_mesg
.event
& PM_EVENT_RESUME
) {
4041 spin_unlock_irqrestore(ap
->lock
, flags
);
4044 spin_unlock_irqrestore(ap
->lock
, flags
);
4046 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
4049 * If we have a ZPODD attached, check its zero
4050 * power ready status before the port is frozen.
4051 * Only needed for runtime suspend.
4053 if (PMSG_IS_AUTO(ap
->pm_mesg
)) {
4054 ata_for_each_dev(dev
, &ap
->link
, ENABLED
) {
4055 if (zpodd_dev_enabled(dev
))
4056 zpodd_on_suspend(dev
);
4060 /* tell ACPI we're suspending */
4061 rc
= ata_acpi_on_suspend(ap
);
4066 ata_eh_freeze_port(ap
);
4068 if (ap
->ops
->port_suspend
)
4069 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
4071 ata_acpi_set_state(ap
, ap
->pm_mesg
);
4073 /* update the flags */
4074 spin_lock_irqsave(ap
->lock
, flags
);
4076 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
4078 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
4079 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
4080 ata_port_schedule_eh(ap
);
4082 spin_unlock_irqrestore(ap
->lock
, flags
);
4088 * ata_eh_handle_port_resume - perform port resume operation
4089 * @ap: port to resume
4094 * Kernel thread context (may sleep).
4096 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
4098 struct ata_link
*link
;
4099 struct ata_device
*dev
;
4100 unsigned long flags
;
4103 /* are we resuming? */
4104 spin_lock_irqsave(ap
->lock
, flags
);
4105 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
4106 !(ap
->pm_mesg
.event
& PM_EVENT_RESUME
)) {
4107 spin_unlock_irqrestore(ap
->lock
, flags
);
4110 spin_unlock_irqrestore(ap
->lock
, flags
);
4112 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
4115 * Error timestamps are in jiffies which doesn't run while
4116 * suspended and PHY events during resume isn't too uncommon.
4117 * When the two are combined, it can lead to unnecessary speed
4118 * downs if the machine is suspended and resumed repeatedly.
4119 * Clear error history.
4121 ata_for_each_link(link
, ap
, HOST_FIRST
)
4122 ata_for_each_dev(dev
, link
, ALL
)
4123 ata_ering_clear(&dev
->ering
);
4125 ata_acpi_set_state(ap
, ap
->pm_mesg
);
4127 if (ap
->ops
->port_resume
)
4128 rc
= ap
->ops
->port_resume(ap
);
4130 /* tell ACPI that we're resuming */
4131 ata_acpi_on_resume(ap
);
4133 /* update the flags */
4134 spin_lock_irqsave(ap
->lock
, flags
);
4135 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
4136 spin_unlock_irqrestore(ap
->lock
, flags
);
4138 #endif /* CONFIG_PM */