2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/pci.h>
38 #include <scsi/scsi.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi_eh.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_cmnd.h>
43 #include "../scsi/scsi_transport_api.h"
45 #include <linux/libata.h>
50 /* speed down verdicts */
51 ATA_EH_SPDN_NCQ_OFF
= (1 << 0),
52 ATA_EH_SPDN_SPEED_DOWN
= (1 << 1),
53 ATA_EH_SPDN_FALLBACK_TO_PIO
= (1 << 2),
54 ATA_EH_SPDN_KEEP_ERRORS
= (1 << 3),
57 ATA_EFLAG_IS_IO
= (1 << 0),
58 ATA_EFLAG_DUBIOUS_XFER
= (1 << 1),
60 /* error categories */
63 ATA_ECAT_TOUT_HSM
= 2,
65 ATA_ECAT_DUBIOUS_NONE
= 4,
66 ATA_ECAT_DUBIOUS_ATA_BUS
= 5,
67 ATA_ECAT_DUBIOUS_TOUT_HSM
= 6,
68 ATA_ECAT_DUBIOUS_UNK_DEV
= 7,
71 ATA_EH_CMD_DFL_TIMEOUT
= 5000,
73 /* always put at least this amount of time between resets */
74 ATA_EH_RESET_COOL_DOWN
= 5000,
76 /* Waiting in ->prereset can never be reliable. It's
77 * sometimes nice to wait there but it can't be depended upon;
78 * otherwise, we wouldn't be resetting. Just give it enough
79 * time for most drives to spin up.
81 ATA_EH_PRERESET_TIMEOUT
= 10000,
82 ATA_EH_FASTDRAIN_INTERVAL
= 3000,
87 /* The following table determines how we sequence resets. Each entry
88 * represents timeout for that try. The first try can be soft or
89 * hardreset. All others are hardreset if available. In most cases
90 * the first reset w/ 10sec timeout should succeed. Following entries
91 * are mostly for error handling, hotplug and retarded devices.
93 static const unsigned long ata_eh_reset_timeouts
[] = {
94 10000, /* most drives spin up by 10sec */
95 10000, /* > 99% working drives spin up before 20sec */
96 35000, /* give > 30 secs of idleness for retarded devices */
97 5000, /* and sweet one last chance */
98 ULONG_MAX
, /* > 1 min has elapsed, give up */
101 static const unsigned long ata_eh_identify_timeouts
[] = {
102 5000, /* covers > 99% of successes and not too boring on failures */
103 10000, /* combined time till here is enough even for media access */
104 30000, /* for true idiots */
108 static const unsigned long ata_eh_other_timeouts
[] = {
109 5000, /* same rationale as identify timeout */
111 /* but no merciful 30sec for other commands, it just isn't worth it */
115 struct ata_eh_cmd_timeout_ent
{
117 const unsigned long *timeouts
;
120 /* The following table determines timeouts to use for EH internal
121 * commands. Each table entry is a command class and matches the
122 * commands the entry applies to and the timeout table to use.
124 * On the retry after a command timed out, the next timeout value from
125 * the table is used. If the table doesn't contain further entries,
126 * the last value is used.
128 * ehc->cmd_timeout_idx keeps track of which timeout to use per
129 * command class, so if SET_FEATURES times out on the first try, the
130 * next try will use the second timeout value only for that class.
132 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
133 static const struct ata_eh_cmd_timeout_ent
134 ata_eh_cmd_timeout_table
[ATA_EH_CMD_TIMEOUT_TABLE_SIZE
] = {
135 { .commands
= CMDS(ATA_CMD_ID_ATA
, ATA_CMD_ID_ATAPI
),
136 .timeouts
= ata_eh_identify_timeouts
, },
137 { .commands
= CMDS(ATA_CMD_READ_NATIVE_MAX
, ATA_CMD_READ_NATIVE_MAX_EXT
),
138 .timeouts
= ata_eh_other_timeouts
, },
139 { .commands
= CMDS(ATA_CMD_SET_MAX
, ATA_CMD_SET_MAX_EXT
),
140 .timeouts
= ata_eh_other_timeouts
, },
141 { .commands
= CMDS(ATA_CMD_SET_FEATURES
),
142 .timeouts
= ata_eh_other_timeouts
, },
143 { .commands
= CMDS(ATA_CMD_INIT_DEV_PARAMS
),
144 .timeouts
= ata_eh_other_timeouts
, },
148 static void __ata_port_freeze(struct ata_port
*ap
);
150 static void ata_eh_handle_port_suspend(struct ata_port
*ap
);
151 static void ata_eh_handle_port_resume(struct ata_port
*ap
);
152 #else /* CONFIG_PM */
153 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
156 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
158 #endif /* CONFIG_PM */
160 static void __ata_ehi_pushv_desc(struct ata_eh_info
*ehi
, const char *fmt
,
163 ehi
->desc_len
+= vscnprintf(ehi
->desc
+ ehi
->desc_len
,
164 ATA_EH_DESC_LEN
- ehi
->desc_len
,
169 * __ata_ehi_push_desc - push error description without adding separator
171 * @fmt: printf format string
173 * Format string according to @fmt and append it to @ehi->desc.
176 * spin_lock_irqsave(host lock)
178 void __ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
183 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
188 * ata_ehi_push_desc - push error description with separator
190 * @fmt: printf format string
192 * Format string according to @fmt and append it to @ehi->desc.
193 * If @ehi->desc is not empty, ", " is added in-between.
196 * spin_lock_irqsave(host lock)
198 void ata_ehi_push_desc(struct ata_eh_info
*ehi
, const char *fmt
, ...)
203 __ata_ehi_push_desc(ehi
, ", ");
206 __ata_ehi_pushv_desc(ehi
, fmt
, args
);
211 * ata_ehi_clear_desc - clean error description
217 * spin_lock_irqsave(host lock)
219 void ata_ehi_clear_desc(struct ata_eh_info
*ehi
)
226 * ata_port_desc - append port description
227 * @ap: target ATA port
228 * @fmt: printf format string
230 * Format string according to @fmt and append it to port
231 * description. If port description is not empty, " " is added
232 * in-between. This function is to be used while initializing
233 * ata_host. The description is printed on host registration.
238 void ata_port_desc(struct ata_port
*ap
, const char *fmt
, ...)
242 WARN_ON(!(ap
->pflags
& ATA_PFLAG_INITIALIZING
));
244 if (ap
->link
.eh_info
.desc_len
)
245 __ata_ehi_push_desc(&ap
->link
.eh_info
, " ");
248 __ata_ehi_pushv_desc(&ap
->link
.eh_info
, fmt
, args
);
255 * ata_port_pbar_desc - append PCI BAR description
256 * @ap: target ATA port
257 * @bar: target PCI BAR
258 * @offset: offset into PCI BAR
259 * @name: name of the area
261 * If @offset is negative, this function formats a string which
262 * contains the name, address, size and type of the BAR and
263 * appends it to the port description. If @offset is zero or
264 * positive, only name and offsetted address is appended.
269 void ata_port_pbar_desc(struct ata_port
*ap
, int bar
, ssize_t offset
,
272 struct pci_dev
*pdev
= to_pci_dev(ap
->host
->dev
);
274 unsigned long long start
, len
;
276 if (pci_resource_flags(pdev
, bar
) & IORESOURCE_MEM
)
278 else if (pci_resource_flags(pdev
, bar
) & IORESOURCE_IO
)
281 start
= (unsigned long long)pci_resource_start(pdev
, bar
);
282 len
= (unsigned long long)pci_resource_len(pdev
, bar
);
285 ata_port_desc(ap
, "%s %s%llu@0x%llx", name
, type
, len
, start
);
287 ata_port_desc(ap
, "%s 0x%llx", name
,
288 start
+ (unsigned long long)offset
);
291 #endif /* CONFIG_PCI */
293 static int ata_lookup_timeout_table(u8 cmd
)
297 for (i
= 0; i
< ATA_EH_CMD_TIMEOUT_TABLE_SIZE
; i
++) {
300 for (cur
= ata_eh_cmd_timeout_table
[i
].commands
; *cur
; cur
++)
309 * ata_internal_cmd_timeout - determine timeout for an internal command
310 * @dev: target device
311 * @cmd: internal command to be issued
313 * Determine timeout for internal command @cmd for @dev.
319 * Determined timeout.
321 unsigned long ata_internal_cmd_timeout(struct ata_device
*dev
, u8 cmd
)
323 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
324 int ent
= ata_lookup_timeout_table(cmd
);
328 return ATA_EH_CMD_DFL_TIMEOUT
;
330 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
331 return ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
];
335 * ata_internal_cmd_timed_out - notification for internal command timeout
336 * @dev: target device
337 * @cmd: internal command which timed out
339 * Notify EH that internal command @cmd for @dev timed out. This
340 * function should be called only for commands whose timeouts are
341 * determined using ata_internal_cmd_timeout().
346 void ata_internal_cmd_timed_out(struct ata_device
*dev
, u8 cmd
)
348 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
349 int ent
= ata_lookup_timeout_table(cmd
);
355 idx
= ehc
->cmd_timeout_idx
[dev
->devno
][ent
];
356 if (ata_eh_cmd_timeout_table
[ent
].timeouts
[idx
+ 1] != ULONG_MAX
)
357 ehc
->cmd_timeout_idx
[dev
->devno
][ent
]++;
360 static void ata_ering_record(struct ata_ering
*ering
, unsigned int eflags
,
361 unsigned int err_mask
)
363 struct ata_ering_entry
*ent
;
368 ering
->cursor
%= ATA_ERING_SIZE
;
370 ent
= &ering
->ring
[ering
->cursor
];
371 ent
->eflags
= eflags
;
372 ent
->err_mask
= err_mask
;
373 ent
->timestamp
= get_jiffies_64();
376 static struct ata_ering_entry
*ata_ering_top(struct ata_ering
*ering
)
378 struct ata_ering_entry
*ent
= &ering
->ring
[ering
->cursor
];
385 static void ata_ering_clear(struct ata_ering
*ering
)
387 memset(ering
, 0, sizeof(*ering
));
390 static int ata_ering_map(struct ata_ering
*ering
,
391 int (*map_fn
)(struct ata_ering_entry
*, void *),
395 struct ata_ering_entry
*ent
;
399 ent
= &ering
->ring
[idx
];
402 rc
= map_fn(ent
, arg
);
405 idx
= (idx
- 1 + ATA_ERING_SIZE
) % ATA_ERING_SIZE
;
406 } while (idx
!= ering
->cursor
);
411 static unsigned int ata_eh_dev_action(struct ata_device
*dev
)
413 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
415 return ehc
->i
.action
| ehc
->i
.dev_action
[dev
->devno
];
418 static void ata_eh_clear_action(struct ata_link
*link
, struct ata_device
*dev
,
419 struct ata_eh_info
*ehi
, unsigned int action
)
421 struct ata_device
*tdev
;
424 ehi
->action
&= ~action
;
425 ata_for_each_dev(tdev
, link
, ALL
)
426 ehi
->dev_action
[tdev
->devno
] &= ~action
;
428 /* doesn't make sense for port-wide EH actions */
429 WARN_ON(!(action
& ATA_EH_PERDEV_MASK
));
431 /* break ehi->action into ehi->dev_action */
432 if (ehi
->action
& action
) {
433 ata_for_each_dev(tdev
, link
, ALL
)
434 ehi
->dev_action
[tdev
->devno
] |=
435 ehi
->action
& action
;
436 ehi
->action
&= ~action
;
439 /* turn off the specified per-dev action */
440 ehi
->dev_action
[dev
->devno
] &= ~action
;
445 * ata_scsi_timed_out - SCSI layer time out callback
446 * @cmd: timed out SCSI command
448 * Handles SCSI layer timeout. We race with normal completion of
449 * the qc for @cmd. If the qc is already gone, we lose and let
450 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
451 * timed out and EH should be invoked. Prevent ata_qc_complete()
452 * from finishing it by setting EH_SCHEDULED and return
455 * TODO: kill this function once old EH is gone.
458 * Called from timer context
461 * EH_HANDLED or EH_NOT_HANDLED
463 enum blk_eh_timer_return
ata_scsi_timed_out(struct scsi_cmnd
*cmd
)
465 struct Scsi_Host
*host
= cmd
->device
->host
;
466 struct ata_port
*ap
= ata_shost_to_port(host
);
468 struct ata_queued_cmd
*qc
;
469 enum blk_eh_timer_return ret
;
473 if (ap
->ops
->error_handler
) {
474 ret
= BLK_EH_NOT_HANDLED
;
478 ret
= BLK_EH_HANDLED
;
479 spin_lock_irqsave(ap
->lock
, flags
);
480 qc
= ata_qc_from_tag(ap
, ap
->link
.active_tag
);
482 WARN_ON(qc
->scsicmd
!= cmd
);
483 qc
->flags
|= ATA_QCFLAG_EH_SCHEDULED
;
484 qc
->err_mask
|= AC_ERR_TIMEOUT
;
485 ret
= BLK_EH_NOT_HANDLED
;
487 spin_unlock_irqrestore(ap
->lock
, flags
);
490 DPRINTK("EXIT, ret=%d\n", ret
);
494 static void ata_eh_unload(struct ata_port
*ap
)
496 struct ata_link
*link
;
497 struct ata_device
*dev
;
500 /* Restore SControl IPM and SPD for the next driver and
501 * disable attached devices.
503 ata_for_each_link(link
, ap
, PMP_FIRST
) {
504 sata_scr_write(link
, SCR_CONTROL
, link
->saved_scontrol
& 0xff0);
505 ata_for_each_dev(dev
, link
, ALL
)
506 ata_dev_disable(dev
);
509 /* freeze and set UNLOADED */
510 spin_lock_irqsave(ap
->lock
, flags
);
512 ata_port_freeze(ap
); /* won't be thawed */
513 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
; /* clear pending from freeze */
514 ap
->pflags
|= ATA_PFLAG_UNLOADED
;
516 spin_unlock_irqrestore(ap
->lock
, flags
);
520 * ata_scsi_error - SCSI layer error handler callback
521 * @host: SCSI host on which error occurred
523 * Handles SCSI-layer-thrown error events.
526 * Inherited from SCSI layer (none, can sleep)
531 void ata_scsi_error(struct Scsi_Host
*host
)
533 struct ata_port
*ap
= ata_shost_to_port(host
);
539 /* synchronize with port task */
540 ata_port_flush_task(ap
);
542 /* synchronize with host lock and sort out timeouts */
544 /* For new EH, all qcs are finished in one of three ways -
545 * normal completion, error completion, and SCSI timeout.
546 * Both cmpletions can race against SCSI timeout. When normal
547 * completion wins, the qc never reaches EH. When error
548 * completion wins, the qc has ATA_QCFLAG_FAILED set.
550 * When SCSI timeout wins, things are a bit more complex.
551 * Normal or error completion can occur after the timeout but
552 * before this point. In such cases, both types of
553 * completions are honored. A scmd is determined to have
554 * timed out iff its associated qc is active and not failed.
556 if (ap
->ops
->error_handler
) {
557 struct scsi_cmnd
*scmd
, *tmp
;
560 spin_lock_irqsave(ap
->lock
, flags
);
562 list_for_each_entry_safe(scmd
, tmp
, &host
->eh_cmd_q
, eh_entry
) {
563 struct ata_queued_cmd
*qc
;
565 for (i
= 0; i
< ATA_MAX_QUEUE
; i
++) {
566 qc
= __ata_qc_from_tag(ap
, i
);
567 if (qc
->flags
& ATA_QCFLAG_ACTIVE
&&
572 if (i
< ATA_MAX_QUEUE
) {
573 /* the scmd has an associated qc */
574 if (!(qc
->flags
& ATA_QCFLAG_FAILED
)) {
575 /* which hasn't failed yet, timeout */
576 qc
->err_mask
|= AC_ERR_TIMEOUT
;
577 qc
->flags
|= ATA_QCFLAG_FAILED
;
581 /* Normal completion occurred after
582 * SCSI timeout but before this point.
583 * Successfully complete it.
585 scmd
->retries
= scmd
->allowed
;
586 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
590 /* If we have timed out qcs. They belong to EH from
591 * this point but the state of the controller is
592 * unknown. Freeze the port to make sure the IRQ
593 * handler doesn't diddle with those qcs. This must
594 * be done atomically w.r.t. setting QCFLAG_FAILED.
597 __ata_port_freeze(ap
);
599 spin_unlock_irqrestore(ap
->lock
, flags
);
601 /* initialize eh_tries */
602 ap
->eh_tries
= ATA_EH_MAX_TRIES
;
604 spin_unlock_wait(ap
->lock
);
607 /* invoke error handler */
608 if (ap
->ops
->error_handler
) {
609 struct ata_link
*link
;
611 /* kill fast drain timer */
612 del_timer_sync(&ap
->fastdrain_timer
);
614 /* process port resume request */
615 ata_eh_handle_port_resume(ap
);
617 /* fetch & clear EH info */
618 spin_lock_irqsave(ap
->lock
, flags
);
620 ata_for_each_link(link
, ap
, HOST_FIRST
) {
621 struct ata_eh_context
*ehc
= &link
->eh_context
;
622 struct ata_device
*dev
;
624 memset(&link
->eh_context
, 0, sizeof(link
->eh_context
));
625 link
->eh_context
.i
= link
->eh_info
;
626 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
628 ata_for_each_dev(dev
, link
, ENABLED
) {
629 int devno
= dev
->devno
;
631 ehc
->saved_xfer_mode
[devno
] = dev
->xfer_mode
;
632 if (ata_ncq_enabled(dev
))
633 ehc
->saved_ncq_enabled
|= 1 << devno
;
637 ap
->pflags
|= ATA_PFLAG_EH_IN_PROGRESS
;
638 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
639 ap
->excl_link
= NULL
; /* don't maintain exclusion over EH */
641 spin_unlock_irqrestore(ap
->lock
, flags
);
643 /* invoke EH, skip if unloading or suspended */
644 if (!(ap
->pflags
& (ATA_PFLAG_UNLOADING
| ATA_PFLAG_SUSPENDED
)))
645 ap
->ops
->error_handler(ap
);
647 /* if unloading, commence suicide */
648 if ((ap
->pflags
& ATA_PFLAG_UNLOADING
) &&
649 !(ap
->pflags
& ATA_PFLAG_UNLOADED
))
654 /* process port suspend request */
655 ata_eh_handle_port_suspend(ap
);
657 /* Exception might have happend after ->error_handler
658 * recovered the port but before this point. Repeat
661 spin_lock_irqsave(ap
->lock
, flags
);
663 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
) {
664 if (--ap
->eh_tries
) {
665 spin_unlock_irqrestore(ap
->lock
, flags
);
668 ata_port_printk(ap
, KERN_ERR
, "EH pending after %d "
669 "tries, giving up\n", ATA_EH_MAX_TRIES
);
670 ap
->pflags
&= ~ATA_PFLAG_EH_PENDING
;
673 /* this run is complete, make sure EH info is clear */
674 ata_for_each_link(link
, ap
, HOST_FIRST
)
675 memset(&link
->eh_info
, 0, sizeof(link
->eh_info
));
677 /* Clear host_eh_scheduled while holding ap->lock such
678 * that if exception occurs after this point but
679 * before EH completion, SCSI midlayer will
682 host
->host_eh_scheduled
= 0;
684 spin_unlock_irqrestore(ap
->lock
, flags
);
686 WARN_ON(ata_qc_from_tag(ap
, ap
->link
.active_tag
) == NULL
);
687 ap
->ops
->eng_timeout(ap
);
690 /* finish or retry handled scmd's and clean up */
691 WARN_ON(host
->host_failed
|| !list_empty(&host
->eh_cmd_q
));
693 scsi_eh_flush_done_q(&ap
->eh_done_q
);
696 spin_lock_irqsave(ap
->lock
, flags
);
698 if (ap
->pflags
& ATA_PFLAG_LOADING
)
699 ap
->pflags
&= ~ATA_PFLAG_LOADING
;
700 else if (ap
->pflags
& ATA_PFLAG_SCSI_HOTPLUG
)
701 queue_delayed_work(ata_aux_wq
, &ap
->hotplug_task
, 0);
703 if (ap
->pflags
& ATA_PFLAG_RECOVERED
)
704 ata_port_printk(ap
, KERN_INFO
, "EH complete\n");
706 ap
->pflags
&= ~(ATA_PFLAG_SCSI_HOTPLUG
| ATA_PFLAG_RECOVERED
);
708 /* tell wait_eh that we're done */
709 ap
->pflags
&= ~ATA_PFLAG_EH_IN_PROGRESS
;
710 wake_up_all(&ap
->eh_wait_q
);
712 spin_unlock_irqrestore(ap
->lock
, flags
);
718 * ata_port_wait_eh - Wait for the currently pending EH to complete
719 * @ap: Port to wait EH for
721 * Wait until the currently pending EH is complete.
724 * Kernel thread context (may sleep).
726 void ata_port_wait_eh(struct ata_port
*ap
)
732 spin_lock_irqsave(ap
->lock
, flags
);
734 while (ap
->pflags
& (ATA_PFLAG_EH_PENDING
| ATA_PFLAG_EH_IN_PROGRESS
)) {
735 prepare_to_wait(&ap
->eh_wait_q
, &wait
, TASK_UNINTERRUPTIBLE
);
736 spin_unlock_irqrestore(ap
->lock
, flags
);
738 spin_lock_irqsave(ap
->lock
, flags
);
740 finish_wait(&ap
->eh_wait_q
, &wait
);
742 spin_unlock_irqrestore(ap
->lock
, flags
);
744 /* make sure SCSI EH is complete */
745 if (scsi_host_in_recovery(ap
->scsi_host
)) {
751 static int ata_eh_nr_in_flight(struct ata_port
*ap
)
756 /* count only non-internal commands */
757 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++)
758 if (ata_qc_from_tag(ap
, tag
))
764 void ata_eh_fastdrain_timerfn(unsigned long arg
)
766 struct ata_port
*ap
= (void *)arg
;
770 spin_lock_irqsave(ap
->lock
, flags
);
772 cnt
= ata_eh_nr_in_flight(ap
);
778 if (cnt
== ap
->fastdrain_cnt
) {
781 /* No progress during the last interval, tag all
782 * in-flight qcs as timed out and freeze the port.
784 for (tag
= 0; tag
< ATA_MAX_QUEUE
- 1; tag
++) {
785 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
787 qc
->err_mask
|= AC_ERR_TIMEOUT
;
792 /* some qcs have finished, give it another chance */
793 ap
->fastdrain_cnt
= cnt
;
794 ap
->fastdrain_timer
.expires
=
795 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
796 add_timer(&ap
->fastdrain_timer
);
800 spin_unlock_irqrestore(ap
->lock
, flags
);
804 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
805 * @ap: target ATA port
806 * @fastdrain: activate fast drain
808 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
809 * is non-zero and EH wasn't pending before. Fast drain ensures
810 * that EH kicks in in timely manner.
813 * spin_lock_irqsave(host lock)
815 static void ata_eh_set_pending(struct ata_port
*ap
, int fastdrain
)
819 /* already scheduled? */
820 if (ap
->pflags
& ATA_PFLAG_EH_PENDING
)
823 ap
->pflags
|= ATA_PFLAG_EH_PENDING
;
828 /* do we have in-flight qcs? */
829 cnt
= ata_eh_nr_in_flight(ap
);
833 /* activate fast drain */
834 ap
->fastdrain_cnt
= cnt
;
835 ap
->fastdrain_timer
.expires
=
836 ata_deadline(jiffies
, ATA_EH_FASTDRAIN_INTERVAL
);
837 add_timer(&ap
->fastdrain_timer
);
841 * ata_qc_schedule_eh - schedule qc for error handling
842 * @qc: command to schedule error handling for
844 * Schedule error handling for @qc. EH will kick in as soon as
845 * other commands are drained.
848 * spin_lock_irqsave(host lock)
850 void ata_qc_schedule_eh(struct ata_queued_cmd
*qc
)
852 struct ata_port
*ap
= qc
->ap
;
854 WARN_ON(!ap
->ops
->error_handler
);
856 qc
->flags
|= ATA_QCFLAG_FAILED
;
857 ata_eh_set_pending(ap
, 1);
859 /* The following will fail if timeout has already expired.
860 * ata_scsi_error() takes care of such scmds on EH entry.
861 * Note that ATA_QCFLAG_FAILED is unconditionally set after
862 * this function completes.
864 blk_abort_request(qc
->scsicmd
->request
);
868 * ata_port_schedule_eh - schedule error handling without a qc
869 * @ap: ATA port to schedule EH for
871 * Schedule error handling for @ap. EH will kick in as soon as
872 * all commands are drained.
875 * spin_lock_irqsave(host lock)
877 void ata_port_schedule_eh(struct ata_port
*ap
)
879 WARN_ON(!ap
->ops
->error_handler
);
881 if (ap
->pflags
& ATA_PFLAG_INITIALIZING
)
884 ata_eh_set_pending(ap
, 1);
885 scsi_schedule_eh(ap
->scsi_host
);
887 DPRINTK("port EH scheduled\n");
890 static int ata_do_link_abort(struct ata_port
*ap
, struct ata_link
*link
)
892 int tag
, nr_aborted
= 0;
894 WARN_ON(!ap
->ops
->error_handler
);
896 /* we're gonna abort all commands, no need for fast drain */
897 ata_eh_set_pending(ap
, 0);
899 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
900 struct ata_queued_cmd
*qc
= ata_qc_from_tag(ap
, tag
);
902 if (qc
&& (!link
|| qc
->dev
->link
== link
)) {
903 qc
->flags
|= ATA_QCFLAG_FAILED
;
910 ata_port_schedule_eh(ap
);
916 * ata_link_abort - abort all qc's on the link
917 * @link: ATA link to abort qc's for
919 * Abort all active qc's active on @link and schedule EH.
922 * spin_lock_irqsave(host lock)
925 * Number of aborted qc's.
927 int ata_link_abort(struct ata_link
*link
)
929 return ata_do_link_abort(link
->ap
, link
);
933 * ata_port_abort - abort all qc's on the port
934 * @ap: ATA port to abort qc's for
936 * Abort all active qc's of @ap and schedule EH.
939 * spin_lock_irqsave(host_set lock)
942 * Number of aborted qc's.
944 int ata_port_abort(struct ata_port
*ap
)
946 return ata_do_link_abort(ap
, NULL
);
950 * __ata_port_freeze - freeze port
951 * @ap: ATA port to freeze
953 * This function is called when HSM violation or some other
954 * condition disrupts normal operation of the port. Frozen port
955 * is not allowed to perform any operation until the port is
956 * thawed, which usually follows a successful reset.
958 * ap->ops->freeze() callback can be used for freezing the port
959 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
960 * port cannot be frozen hardware-wise, the interrupt handler
961 * must ack and clear interrupts unconditionally while the port
965 * spin_lock_irqsave(host lock)
967 static void __ata_port_freeze(struct ata_port
*ap
)
969 WARN_ON(!ap
->ops
->error_handler
);
974 ap
->pflags
|= ATA_PFLAG_FROZEN
;
976 DPRINTK("ata%u port frozen\n", ap
->print_id
);
980 * ata_port_freeze - abort & freeze port
981 * @ap: ATA port to freeze
983 * Abort and freeze @ap.
986 * spin_lock_irqsave(host lock)
989 * Number of aborted commands.
991 int ata_port_freeze(struct ata_port
*ap
)
995 WARN_ON(!ap
->ops
->error_handler
);
997 nr_aborted
= ata_port_abort(ap
);
998 __ata_port_freeze(ap
);
1004 * sata_async_notification - SATA async notification handler
1005 * @ap: ATA port where async notification is received
1007 * Handler to be called when async notification via SDB FIS is
1008 * received. This function schedules EH if necessary.
1011 * spin_lock_irqsave(host lock)
1014 * 1 if EH is scheduled, 0 otherwise.
1016 int sata_async_notification(struct ata_port
*ap
)
1021 if (!(ap
->flags
& ATA_FLAG_AN
))
1024 rc
= sata_scr_read(&ap
->link
, SCR_NOTIFICATION
, &sntf
);
1026 sata_scr_write(&ap
->link
, SCR_NOTIFICATION
, sntf
);
1028 if (!sata_pmp_attached(ap
) || rc
) {
1029 /* PMP is not attached or SNTF is not available */
1030 if (!sata_pmp_attached(ap
)) {
1031 /* PMP is not attached. Check whether ATAPI
1032 * AN is configured. If so, notify media
1035 struct ata_device
*dev
= ap
->link
.device
;
1037 if ((dev
->class == ATA_DEV_ATAPI
) &&
1038 (dev
->flags
& ATA_DFLAG_AN
))
1039 ata_scsi_media_change_notify(dev
);
1042 /* PMP is attached but SNTF is not available.
1043 * ATAPI async media change notification is
1044 * not used. The PMP must be reporting PHY
1045 * status change, schedule EH.
1047 ata_port_schedule_eh(ap
);
1051 /* PMP is attached and SNTF is available */
1052 struct ata_link
*link
;
1054 /* check and notify ATAPI AN */
1055 ata_for_each_link(link
, ap
, EDGE
) {
1056 if (!(sntf
& (1 << link
->pmp
)))
1059 if ((link
->device
->class == ATA_DEV_ATAPI
) &&
1060 (link
->device
->flags
& ATA_DFLAG_AN
))
1061 ata_scsi_media_change_notify(link
->device
);
1064 /* If PMP is reporting that PHY status of some
1065 * downstream ports has changed, schedule EH.
1067 if (sntf
& (1 << SATA_PMP_CTRL_PORT
)) {
1068 ata_port_schedule_eh(ap
);
1077 * ata_eh_freeze_port - EH helper to freeze port
1078 * @ap: ATA port to freeze
1085 void ata_eh_freeze_port(struct ata_port
*ap
)
1087 unsigned long flags
;
1089 if (!ap
->ops
->error_handler
)
1092 spin_lock_irqsave(ap
->lock
, flags
);
1093 __ata_port_freeze(ap
);
1094 spin_unlock_irqrestore(ap
->lock
, flags
);
1098 * ata_port_thaw_port - EH helper to thaw port
1099 * @ap: ATA port to thaw
1101 * Thaw frozen port @ap.
1106 void ata_eh_thaw_port(struct ata_port
*ap
)
1108 unsigned long flags
;
1110 if (!ap
->ops
->error_handler
)
1113 spin_lock_irqsave(ap
->lock
, flags
);
1115 ap
->pflags
&= ~ATA_PFLAG_FROZEN
;
1120 spin_unlock_irqrestore(ap
->lock
, flags
);
1122 DPRINTK("ata%u port thawed\n", ap
->print_id
);
1125 static void ata_eh_scsidone(struct scsi_cmnd
*scmd
)
1130 static void __ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1132 struct ata_port
*ap
= qc
->ap
;
1133 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1134 unsigned long flags
;
1136 spin_lock_irqsave(ap
->lock
, flags
);
1137 qc
->scsidone
= ata_eh_scsidone
;
1138 __ata_qc_complete(qc
);
1139 WARN_ON(ata_tag_valid(qc
->tag
));
1140 spin_unlock_irqrestore(ap
->lock
, flags
);
1142 scsi_eh_finish_cmd(scmd
, &ap
->eh_done_q
);
1146 * ata_eh_qc_complete - Complete an active ATA command from EH
1147 * @qc: Command to complete
1149 * Indicate to the mid and upper layers that an ATA command has
1150 * completed. To be used from EH.
1152 void ata_eh_qc_complete(struct ata_queued_cmd
*qc
)
1154 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1155 scmd
->retries
= scmd
->allowed
;
1156 __ata_eh_qc_complete(qc
);
1160 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1161 * @qc: Command to retry
1163 * Indicate to the mid and upper layers that an ATA command
1164 * should be retried. To be used from EH.
1166 * SCSI midlayer limits the number of retries to scmd->allowed.
1167 * scmd->retries is decremented for commands which get retried
1168 * due to unrelated failures (qc->err_mask is zero).
1170 void ata_eh_qc_retry(struct ata_queued_cmd
*qc
)
1172 struct scsi_cmnd
*scmd
= qc
->scsicmd
;
1173 if (!qc
->err_mask
&& scmd
->retries
)
1175 __ata_eh_qc_complete(qc
);
1179 * ata_dev_disable - disable ATA device
1180 * @dev: ATA device to disable
1187 void ata_dev_disable(struct ata_device
*dev
)
1189 if (!ata_dev_enabled(dev
))
1192 if (ata_msg_drv(dev
->link
->ap
))
1193 ata_dev_printk(dev
, KERN_WARNING
, "disabled\n");
1194 ata_acpi_on_disable(dev
);
1195 ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO0
| ATA_DNXFER_QUIET
);
1198 /* From now till the next successful probe, ering is used to
1199 * track probe failures. Clear accumulated device error info.
1201 ata_ering_clear(&dev
->ering
);
1205 * ata_eh_detach_dev - detach ATA device
1206 * @dev: ATA device to detach
1213 void ata_eh_detach_dev(struct ata_device
*dev
)
1215 struct ata_link
*link
= dev
->link
;
1216 struct ata_port
*ap
= link
->ap
;
1217 struct ata_eh_context
*ehc
= &link
->eh_context
;
1218 unsigned long flags
;
1220 ata_dev_disable(dev
);
1222 spin_lock_irqsave(ap
->lock
, flags
);
1224 dev
->flags
&= ~ATA_DFLAG_DETACH
;
1226 if (ata_scsi_offline_dev(dev
)) {
1227 dev
->flags
|= ATA_DFLAG_DETACHED
;
1228 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
1231 /* clear per-dev EH info */
1232 ata_eh_clear_action(link
, dev
, &link
->eh_info
, ATA_EH_PERDEV_MASK
);
1233 ata_eh_clear_action(link
, dev
, &link
->eh_context
.i
, ATA_EH_PERDEV_MASK
);
1234 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
1235 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
1237 spin_unlock_irqrestore(ap
->lock
, flags
);
1241 * ata_eh_about_to_do - about to perform eh_action
1242 * @link: target ATA link
1243 * @dev: target ATA dev for per-dev action (can be NULL)
1244 * @action: action about to be performed
1246 * Called just before performing EH actions to clear related bits
1247 * in @link->eh_info such that eh actions are not unnecessarily
1253 void ata_eh_about_to_do(struct ata_link
*link
, struct ata_device
*dev
,
1254 unsigned int action
)
1256 struct ata_port
*ap
= link
->ap
;
1257 struct ata_eh_info
*ehi
= &link
->eh_info
;
1258 struct ata_eh_context
*ehc
= &link
->eh_context
;
1259 unsigned long flags
;
1261 spin_lock_irqsave(ap
->lock
, flags
);
1263 ata_eh_clear_action(link
, dev
, ehi
, action
);
1265 /* About to take EH action, set RECOVERED. Ignore actions on
1266 * slave links as master will do them again.
1268 if (!(ehc
->i
.flags
& ATA_EHI_QUIET
) && link
!= ap
->slave_link
)
1269 ap
->pflags
|= ATA_PFLAG_RECOVERED
;
1271 spin_unlock_irqrestore(ap
->lock
, flags
);
1275 * ata_eh_done - EH action complete
1276 * @ap: target ATA port
1277 * @dev: target ATA dev for per-dev action (can be NULL)
1278 * @action: action just completed
1280 * Called right after performing EH actions to clear related bits
1281 * in @link->eh_context.
1286 void ata_eh_done(struct ata_link
*link
, struct ata_device
*dev
,
1287 unsigned int action
)
1289 struct ata_eh_context
*ehc
= &link
->eh_context
;
1291 ata_eh_clear_action(link
, dev
, &ehc
->i
, action
);
1295 * ata_err_string - convert err_mask to descriptive string
1296 * @err_mask: error mask to convert to string
1298 * Convert @err_mask to descriptive string. Errors are
1299 * prioritized according to severity and only the most severe
1300 * error is reported.
1306 * Descriptive string for @err_mask
1308 static const char *ata_err_string(unsigned int err_mask
)
1310 if (err_mask
& AC_ERR_HOST_BUS
)
1311 return "host bus error";
1312 if (err_mask
& AC_ERR_ATA_BUS
)
1313 return "ATA bus error";
1314 if (err_mask
& AC_ERR_TIMEOUT
)
1316 if (err_mask
& AC_ERR_HSM
)
1317 return "HSM violation";
1318 if (err_mask
& AC_ERR_SYSTEM
)
1319 return "internal error";
1320 if (err_mask
& AC_ERR_MEDIA
)
1321 return "media error";
1322 if (err_mask
& AC_ERR_INVALID
)
1323 return "invalid argument";
1324 if (err_mask
& AC_ERR_DEV
)
1325 return "device error";
1326 return "unknown error";
1330 * ata_read_log_page - read a specific log page
1331 * @dev: target device
1332 * @page: page to read
1333 * @buf: buffer to store read page
1334 * @sectors: number of sectors to read
1336 * Read log page using READ_LOG_EXT command.
1339 * Kernel thread context (may sleep).
1342 * 0 on success, AC_ERR_* mask otherwise.
1344 static unsigned int ata_read_log_page(struct ata_device
*dev
,
1345 u8 page
, void *buf
, unsigned int sectors
)
1347 struct ata_taskfile tf
;
1348 unsigned int err_mask
;
1350 DPRINTK("read log page - page %d\n", page
);
1352 ata_tf_init(dev
, &tf
);
1353 tf
.command
= ATA_CMD_READ_LOG_EXT
;
1356 tf
.hob_nsect
= sectors
>> 8;
1357 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA48
| ATA_TFLAG_DEVICE
;
1358 tf
.protocol
= ATA_PROT_PIO
;
1360 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_FROM_DEVICE
,
1361 buf
, sectors
* ATA_SECT_SIZE
, 0);
1363 DPRINTK("EXIT, err_mask=%x\n", err_mask
);
1368 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1369 * @dev: Device to read log page 10h from
1370 * @tag: Resulting tag of the failed command
1371 * @tf: Resulting taskfile registers of the failed command
1373 * Read log page 10h to obtain NCQ error details and clear error
1377 * Kernel thread context (may sleep).
1380 * 0 on success, -errno otherwise.
1382 static int ata_eh_read_log_10h(struct ata_device
*dev
,
1383 int *tag
, struct ata_taskfile
*tf
)
1385 u8
*buf
= dev
->link
->ap
->sector_buf
;
1386 unsigned int err_mask
;
1390 err_mask
= ata_read_log_page(dev
, ATA_LOG_SATA_NCQ
, buf
, 1);
1395 for (i
= 0; i
< ATA_SECT_SIZE
; i
++)
1398 ata_dev_printk(dev
, KERN_WARNING
,
1399 "invalid checksum 0x%x on log page 10h\n", csum
);
1404 *tag
= buf
[0] & 0x1f;
1406 tf
->command
= buf
[2];
1407 tf
->feature
= buf
[3];
1411 tf
->device
= buf
[7];
1412 tf
->hob_lbal
= buf
[8];
1413 tf
->hob_lbam
= buf
[9];
1414 tf
->hob_lbah
= buf
[10];
1415 tf
->nsect
= buf
[12];
1416 tf
->hob_nsect
= buf
[13];
1422 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1423 * @dev: target ATAPI device
1424 * @r_sense_key: out parameter for sense_key
1426 * Perform ATAPI TEST_UNIT_READY.
1429 * EH context (may sleep).
1432 * 0 on success, AC_ERR_* mask on failure.
1434 static unsigned int atapi_eh_tur(struct ata_device
*dev
, u8
*r_sense_key
)
1436 u8 cdb
[ATAPI_CDB_LEN
] = { TEST_UNIT_READY
, 0, 0, 0, 0, 0 };
1437 struct ata_taskfile tf
;
1438 unsigned int err_mask
;
1440 ata_tf_init(dev
, &tf
);
1442 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1443 tf
.command
= ATA_CMD_PACKET
;
1444 tf
.protocol
= ATAPI_PROT_NODATA
;
1446 err_mask
= ata_exec_internal(dev
, &tf
, cdb
, DMA_NONE
, NULL
, 0, 0);
1447 if (err_mask
== AC_ERR_DEV
)
1448 *r_sense_key
= tf
.feature
>> 4;
1453 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1454 * @dev: device to perform REQUEST_SENSE to
1455 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1456 * @dfl_sense_key: default sense key to use
1458 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1459 * SENSE. This function is EH helper.
1462 * Kernel thread context (may sleep).
1465 * 0 on success, AC_ERR_* mask on failure
1467 static unsigned int atapi_eh_request_sense(struct ata_device
*dev
,
1468 u8
*sense_buf
, u8 dfl_sense_key
)
1470 u8 cdb
[ATAPI_CDB_LEN
] =
1471 { REQUEST_SENSE
, 0, 0, 0, SCSI_SENSE_BUFFERSIZE
, 0 };
1472 struct ata_port
*ap
= dev
->link
->ap
;
1473 struct ata_taskfile tf
;
1475 DPRINTK("ATAPI request sense\n");
1477 /* FIXME: is this needed? */
1478 memset(sense_buf
, 0, SCSI_SENSE_BUFFERSIZE
);
1480 /* initialize sense_buf with the error register,
1481 * for the case where they are -not- overwritten
1483 sense_buf
[0] = 0x70;
1484 sense_buf
[2] = dfl_sense_key
;
1486 /* some devices time out if garbage left in tf */
1487 ata_tf_init(dev
, &tf
);
1489 tf
.flags
|= ATA_TFLAG_ISADDR
| ATA_TFLAG_DEVICE
;
1490 tf
.command
= ATA_CMD_PACKET
;
1492 /* is it pointless to prefer PIO for "safety reasons"? */
1493 if (ap
->flags
& ATA_FLAG_PIO_DMA
) {
1494 tf
.protocol
= ATAPI_PROT_DMA
;
1495 tf
.feature
|= ATAPI_PKT_DMA
;
1497 tf
.protocol
= ATAPI_PROT_PIO
;
1498 tf
.lbam
= SCSI_SENSE_BUFFERSIZE
;
1502 return ata_exec_internal(dev
, &tf
, cdb
, DMA_FROM_DEVICE
,
1503 sense_buf
, SCSI_SENSE_BUFFERSIZE
, 0);
1507 * ata_eh_analyze_serror - analyze SError for a failed port
1508 * @link: ATA link to analyze SError for
1510 * Analyze SError if available and further determine cause of
1516 static void ata_eh_analyze_serror(struct ata_link
*link
)
1518 struct ata_eh_context
*ehc
= &link
->eh_context
;
1519 u32 serror
= ehc
->i
.serror
;
1520 unsigned int err_mask
= 0, action
= 0;
1523 if (serror
& (SERR_PERSISTENT
| SERR_DATA
)) {
1524 err_mask
|= AC_ERR_ATA_BUS
;
1525 action
|= ATA_EH_RESET
;
1527 if (serror
& SERR_PROTOCOL
) {
1528 err_mask
|= AC_ERR_HSM
;
1529 action
|= ATA_EH_RESET
;
1531 if (serror
& SERR_INTERNAL
) {
1532 err_mask
|= AC_ERR_SYSTEM
;
1533 action
|= ATA_EH_RESET
;
1536 /* Determine whether a hotplug event has occurred. Both
1537 * SError.N/X are considered hotplug events for enabled or
1538 * host links. For disabled PMP links, only N bit is
1539 * considered as X bit is left at 1 for link plugging.
1543 if (!(link
->flags
& ATA_LFLAG_DISABLED
) || ata_is_host_link(link
))
1544 hotplug_mask
= SERR_PHYRDY_CHG
| SERR_DEV_XCHG
;
1546 hotplug_mask
= SERR_PHYRDY_CHG
;
1548 if (serror
& hotplug_mask
)
1549 ata_ehi_hotplugged(&ehc
->i
);
1551 ehc
->i
.err_mask
|= err_mask
;
1552 ehc
->i
.action
|= action
;
1556 * ata_eh_analyze_ncq_error - analyze NCQ error
1557 * @link: ATA link to analyze NCQ error for
1559 * Read log page 10h, determine the offending qc and acquire
1560 * error status TF. For NCQ device errors, all LLDDs have to do
1561 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1565 * Kernel thread context (may sleep).
1567 void ata_eh_analyze_ncq_error(struct ata_link
*link
)
1569 struct ata_port
*ap
= link
->ap
;
1570 struct ata_eh_context
*ehc
= &link
->eh_context
;
1571 struct ata_device
*dev
= link
->device
;
1572 struct ata_queued_cmd
*qc
;
1573 struct ata_taskfile tf
;
1576 /* if frozen, we can't do much */
1577 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
1580 /* is it NCQ device error? */
1581 if (!link
->sactive
|| !(ehc
->i
.err_mask
& AC_ERR_DEV
))
1584 /* has LLDD analyzed already? */
1585 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1586 qc
= __ata_qc_from_tag(ap
, tag
);
1588 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
1595 /* okay, this error is ours */
1596 rc
= ata_eh_read_log_10h(dev
, &tag
, &tf
);
1598 ata_link_printk(link
, KERN_ERR
, "failed to read log page 10h "
1599 "(errno=%d)\n", rc
);
1603 if (!(link
->sactive
& (1 << tag
))) {
1604 ata_link_printk(link
, KERN_ERR
, "log page 10h reported "
1605 "inactive tag %d\n", tag
);
1609 /* we've got the perpetrator, condemn it */
1610 qc
= __ata_qc_from_tag(ap
, tag
);
1611 memcpy(&qc
->result_tf
, &tf
, sizeof(tf
));
1612 qc
->result_tf
.flags
= ATA_TFLAG_ISADDR
| ATA_TFLAG_LBA
| ATA_TFLAG_LBA48
;
1613 qc
->err_mask
|= AC_ERR_DEV
| AC_ERR_NCQ
;
1614 ehc
->i
.err_mask
&= ~AC_ERR_DEV
;
1618 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1619 * @qc: qc to analyze
1620 * @tf: Taskfile registers to analyze
1622 * Analyze taskfile of @qc and further determine cause of
1623 * failure. This function also requests ATAPI sense data if
1627 * Kernel thread context (may sleep).
1630 * Determined recovery action
1632 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd
*qc
,
1633 const struct ata_taskfile
*tf
)
1635 unsigned int tmp
, action
= 0;
1636 u8 stat
= tf
->command
, err
= tf
->feature
;
1638 if ((stat
& (ATA_BUSY
| ATA_DRQ
| ATA_DRDY
)) != ATA_DRDY
) {
1639 qc
->err_mask
|= AC_ERR_HSM
;
1640 return ATA_EH_RESET
;
1643 if (stat
& (ATA_ERR
| ATA_DF
))
1644 qc
->err_mask
|= AC_ERR_DEV
;
1648 switch (qc
->dev
->class) {
1651 qc
->err_mask
|= AC_ERR_ATA_BUS
;
1653 qc
->err_mask
|= AC_ERR_MEDIA
;
1655 qc
->err_mask
|= AC_ERR_INVALID
;
1659 if (!(qc
->ap
->pflags
& ATA_PFLAG_FROZEN
)) {
1660 tmp
= atapi_eh_request_sense(qc
->dev
,
1661 qc
->scsicmd
->sense_buffer
,
1662 qc
->result_tf
.feature
>> 4);
1664 /* ATA_QCFLAG_SENSE_VALID is used to
1665 * tell atapi_qc_complete() that sense
1666 * data is already valid.
1668 * TODO: interpret sense data and set
1669 * appropriate err_mask.
1671 qc
->flags
|= ATA_QCFLAG_SENSE_VALID
;
1673 qc
->err_mask
|= tmp
;
1677 if (qc
->err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
| AC_ERR_ATA_BUS
))
1678 action
|= ATA_EH_RESET
;
1683 static int ata_eh_categorize_error(unsigned int eflags
, unsigned int err_mask
,
1688 if (!(eflags
& ATA_EFLAG_DUBIOUS_XFER
))
1692 base
= ATA_ECAT_DUBIOUS_NONE
;
1694 if (err_mask
& AC_ERR_ATA_BUS
)
1695 return base
+ ATA_ECAT_ATA_BUS
;
1697 if (err_mask
& AC_ERR_TIMEOUT
)
1698 return base
+ ATA_ECAT_TOUT_HSM
;
1700 if (eflags
& ATA_EFLAG_IS_IO
) {
1701 if (err_mask
& AC_ERR_HSM
)
1702 return base
+ ATA_ECAT_TOUT_HSM
;
1704 (AC_ERR_DEV
|AC_ERR_MEDIA
|AC_ERR_INVALID
)) == AC_ERR_DEV
)
1705 return base
+ ATA_ECAT_UNK_DEV
;
1711 struct speed_down_verdict_arg
{
1714 int nr_errors
[ATA_ECAT_NR
];
1717 static int speed_down_verdict_cb(struct ata_ering_entry
*ent
, void *void_arg
)
1719 struct speed_down_verdict_arg
*arg
= void_arg
;
1722 if (ent
->timestamp
< arg
->since
)
1725 cat
= ata_eh_categorize_error(ent
->eflags
, ent
->err_mask
,
1727 arg
->nr_errors
[cat
]++;
1733 * ata_eh_speed_down_verdict - Determine speed down verdict
1734 * @dev: Device of interest
1736 * This function examines error ring of @dev and determines
1737 * whether NCQ needs to be turned off, transfer speed should be
1738 * stepped down, or falling back to PIO is necessary.
1740 * ECAT_ATA_BUS : ATA_BUS error for any command
1742 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1745 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1747 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1748 * data transfer hasn't been verified.
1752 * NCQ_OFF : Turn off NCQ.
1754 * SPEED_DOWN : Speed down transfer speed but don't fall back
1757 * FALLBACK_TO_PIO : Fall back to PIO.
1759 * Even if multiple verdicts are returned, only one action is
1760 * taken per error. An action triggered by non-DUBIOUS errors
1761 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1762 * This is to expedite speed down decisions right after device is
1763 * initially configured.
1765 * The followings are speed down rules. #1 and #2 deal with
1768 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1769 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1771 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1772 * occurred during last 5 mins, NCQ_OFF.
1774 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1775 * ocurred during last 5 mins, FALLBACK_TO_PIO
1777 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1778 * during last 10 mins, NCQ_OFF.
1780 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1781 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1784 * Inherited from caller.
1787 * OR of ATA_EH_SPDN_* flags.
1789 static unsigned int ata_eh_speed_down_verdict(struct ata_device
*dev
)
1791 const u64 j5mins
= 5LLU * 60 * HZ
, j10mins
= 10LLU * 60 * HZ
;
1792 u64 j64
= get_jiffies_64();
1793 struct speed_down_verdict_arg arg
;
1794 unsigned int verdict
= 0;
1796 /* scan past 5 mins of error history */
1797 memset(&arg
, 0, sizeof(arg
));
1798 arg
.since
= j64
- min(j64
, j5mins
);
1799 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1801 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_ATA_BUS
] +
1802 arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] > 1)
1803 verdict
|= ATA_EH_SPDN_SPEED_DOWN
|
1804 ATA_EH_SPDN_FALLBACK_TO_PIO
| ATA_EH_SPDN_KEEP_ERRORS
;
1806 if (arg
.nr_errors
[ATA_ECAT_DUBIOUS_TOUT_HSM
] +
1807 arg
.nr_errors
[ATA_ECAT_DUBIOUS_UNK_DEV
] > 1)
1808 verdict
|= ATA_EH_SPDN_NCQ_OFF
| ATA_EH_SPDN_KEEP_ERRORS
;
1810 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1811 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1812 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1813 verdict
|= ATA_EH_SPDN_FALLBACK_TO_PIO
;
1815 /* scan past 10 mins of error history */
1816 memset(&arg
, 0, sizeof(arg
));
1817 arg
.since
= j64
- min(j64
, j10mins
);
1818 ata_ering_map(&dev
->ering
, speed_down_verdict_cb
, &arg
);
1820 if (arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] +
1821 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 3)
1822 verdict
|= ATA_EH_SPDN_NCQ_OFF
;
1824 if (arg
.nr_errors
[ATA_ECAT_ATA_BUS
] +
1825 arg
.nr_errors
[ATA_ECAT_TOUT_HSM
] > 3 ||
1826 arg
.nr_errors
[ATA_ECAT_UNK_DEV
] > 6)
1827 verdict
|= ATA_EH_SPDN_SPEED_DOWN
;
1833 * ata_eh_speed_down - record error and speed down if necessary
1834 * @dev: Failed device
1835 * @eflags: mask of ATA_EFLAG_* flags
1836 * @err_mask: err_mask of the error
1838 * Record error and examine error history to determine whether
1839 * adjusting transmission speed is necessary. It also sets
1840 * transmission limits appropriately if such adjustment is
1844 * Kernel thread context (may sleep).
1847 * Determined recovery action.
1849 static unsigned int ata_eh_speed_down(struct ata_device
*dev
,
1850 unsigned int eflags
, unsigned int err_mask
)
1852 struct ata_link
*link
= ata_dev_phys_link(dev
);
1854 unsigned int verdict
;
1855 unsigned int action
= 0;
1857 /* don't bother if Cat-0 error */
1858 if (ata_eh_categorize_error(eflags
, err_mask
, &xfer_ok
) == 0)
1861 /* record error and determine whether speed down is necessary */
1862 ata_ering_record(&dev
->ering
, eflags
, err_mask
);
1863 verdict
= ata_eh_speed_down_verdict(dev
);
1866 if ((verdict
& ATA_EH_SPDN_NCQ_OFF
) &&
1867 (dev
->flags
& (ATA_DFLAG_PIO
| ATA_DFLAG_NCQ
|
1868 ATA_DFLAG_NCQ_OFF
)) == ATA_DFLAG_NCQ
) {
1869 dev
->flags
|= ATA_DFLAG_NCQ_OFF
;
1870 ata_dev_printk(dev
, KERN_WARNING
,
1871 "NCQ disabled due to excessive errors\n");
1876 if (verdict
& ATA_EH_SPDN_SPEED_DOWN
) {
1877 /* speed down SATA link speed if possible */
1878 if (sata_down_spd_limit(link
) == 0) {
1879 action
|= ATA_EH_RESET
;
1883 /* lower transfer mode */
1884 if (dev
->spdn_cnt
< 2) {
1885 static const int dma_dnxfer_sel
[] =
1886 { ATA_DNXFER_DMA
, ATA_DNXFER_40C
};
1887 static const int pio_dnxfer_sel
[] =
1888 { ATA_DNXFER_PIO
, ATA_DNXFER_FORCE_PIO0
};
1891 if (dev
->xfer_shift
!= ATA_SHIFT_PIO
)
1892 sel
= dma_dnxfer_sel
[dev
->spdn_cnt
];
1894 sel
= pio_dnxfer_sel
[dev
->spdn_cnt
];
1898 if (ata_down_xfermask_limit(dev
, sel
) == 0) {
1899 action
|= ATA_EH_RESET
;
1905 /* Fall back to PIO? Slowing down to PIO is meaningless for
1906 * SATA ATA devices. Consider it only for PATA and SATAPI.
1908 if ((verdict
& ATA_EH_SPDN_FALLBACK_TO_PIO
) && (dev
->spdn_cnt
>= 2) &&
1909 (link
->ap
->cbl
!= ATA_CBL_SATA
|| dev
->class == ATA_DEV_ATAPI
) &&
1910 (dev
->xfer_shift
!= ATA_SHIFT_PIO
)) {
1911 if (ata_down_xfermask_limit(dev
, ATA_DNXFER_FORCE_PIO
) == 0) {
1913 action
|= ATA_EH_RESET
;
1920 /* device has been slowed down, blow error history */
1921 if (!(verdict
& ATA_EH_SPDN_KEEP_ERRORS
))
1922 ata_ering_clear(&dev
->ering
);
1927 * ata_eh_link_autopsy - analyze error and determine recovery action
1928 * @link: host link to perform autopsy on
1930 * Analyze why @link failed and determine which recovery actions
1931 * are needed. This function also sets more detailed AC_ERR_*
1932 * values and fills sense data for ATAPI CHECK SENSE.
1935 * Kernel thread context (may sleep).
1937 static void ata_eh_link_autopsy(struct ata_link
*link
)
1939 struct ata_port
*ap
= link
->ap
;
1940 struct ata_eh_context
*ehc
= &link
->eh_context
;
1941 struct ata_device
*dev
;
1942 unsigned int all_err_mask
= 0, eflags
= 0;
1949 if (ehc
->i
.flags
& ATA_EHI_NO_AUTOPSY
)
1952 /* obtain and analyze SError */
1953 rc
= sata_scr_read(link
, SCR_ERROR
, &serror
);
1955 ehc
->i
.serror
|= serror
;
1956 ata_eh_analyze_serror(link
);
1957 } else if (rc
!= -EOPNOTSUPP
) {
1958 /* SError read failed, force reset and probing */
1959 ehc
->i
.probe_mask
|= ATA_ALL_DEVICES
;
1960 ehc
->i
.action
|= ATA_EH_RESET
;
1961 ehc
->i
.err_mask
|= AC_ERR_OTHER
;
1964 /* analyze NCQ failure */
1965 ata_eh_analyze_ncq_error(link
);
1967 /* any real error trumps AC_ERR_OTHER */
1968 if (ehc
->i
.err_mask
& ~AC_ERR_OTHER
)
1969 ehc
->i
.err_mask
&= ~AC_ERR_OTHER
;
1971 all_err_mask
|= ehc
->i
.err_mask
;
1973 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
1974 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
1976 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
1977 ata_dev_phys_link(qc
->dev
) != link
)
1980 /* inherit upper level err_mask */
1981 qc
->err_mask
|= ehc
->i
.err_mask
;
1984 ehc
->i
.action
|= ata_eh_analyze_tf(qc
, &qc
->result_tf
);
1986 /* DEV errors are probably spurious in case of ATA_BUS error */
1987 if (qc
->err_mask
& AC_ERR_ATA_BUS
)
1988 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_MEDIA
|
1991 /* any real error trumps unknown error */
1992 if (qc
->err_mask
& ~AC_ERR_OTHER
)
1993 qc
->err_mask
&= ~AC_ERR_OTHER
;
1995 /* SENSE_VALID trumps dev/unknown error and revalidation */
1996 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
)
1997 qc
->err_mask
&= ~(AC_ERR_DEV
| AC_ERR_OTHER
);
1999 /* determine whether the command is worth retrying */
2000 if (!(qc
->err_mask
& AC_ERR_INVALID
) &&
2001 ((qc
->flags
& ATA_QCFLAG_IO
) || qc
->err_mask
!= AC_ERR_DEV
))
2002 qc
->flags
|= ATA_QCFLAG_RETRY
;
2004 /* accumulate error info */
2005 ehc
->i
.dev
= qc
->dev
;
2006 all_err_mask
|= qc
->err_mask
;
2007 if (qc
->flags
& ATA_QCFLAG_IO
)
2008 eflags
|= ATA_EFLAG_IS_IO
;
2011 /* enforce default EH actions */
2012 if (ap
->pflags
& ATA_PFLAG_FROZEN
||
2013 all_err_mask
& (AC_ERR_HSM
| AC_ERR_TIMEOUT
))
2014 ehc
->i
.action
|= ATA_EH_RESET
;
2015 else if (((eflags
& ATA_EFLAG_IS_IO
) && all_err_mask
) ||
2016 (!(eflags
& ATA_EFLAG_IS_IO
) && (all_err_mask
& ~AC_ERR_DEV
)))
2017 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2019 /* If we have offending qcs and the associated failed device,
2020 * perform per-dev EH action only on the offending device.
2023 ehc
->i
.dev_action
[ehc
->i
.dev
->devno
] |=
2024 ehc
->i
.action
& ATA_EH_PERDEV_MASK
;
2025 ehc
->i
.action
&= ~ATA_EH_PERDEV_MASK
;
2028 /* propagate timeout to host link */
2029 if ((all_err_mask
& AC_ERR_TIMEOUT
) && !ata_is_host_link(link
))
2030 ap
->link
.eh_context
.i
.err_mask
|= AC_ERR_TIMEOUT
;
2032 /* record error and consider speeding down */
2034 if (!dev
&& ((ata_link_max_devices(link
) == 1 &&
2035 ata_dev_enabled(link
->device
))))
2039 if (dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)
2040 eflags
|= ATA_EFLAG_DUBIOUS_XFER
;
2041 ehc
->i
.action
|= ata_eh_speed_down(dev
, eflags
, all_err_mask
);
2048 * ata_eh_autopsy - analyze error and determine recovery action
2049 * @ap: host port to perform autopsy on
2051 * Analyze all links of @ap and determine why they failed and
2052 * which recovery actions are needed.
2055 * Kernel thread context (may sleep).
2057 void ata_eh_autopsy(struct ata_port
*ap
)
2059 struct ata_link
*link
;
2061 ata_for_each_link(link
, ap
, EDGE
)
2062 ata_eh_link_autopsy(link
);
2064 /* Handle the frigging slave link. Autopsy is done similarly
2065 * but actions and flags are transferred over to the master
2066 * link and handled from there.
2068 if (ap
->slave_link
) {
2069 struct ata_eh_context
*mehc
= &ap
->link
.eh_context
;
2070 struct ata_eh_context
*sehc
= &ap
->slave_link
->eh_context
;
2072 /* transfer control flags from master to slave */
2073 sehc
->i
.flags
|= mehc
->i
.flags
& ATA_EHI_TO_SLAVE_MASK
;
2075 /* perform autopsy on the slave link */
2076 ata_eh_link_autopsy(ap
->slave_link
);
2078 /* transfer actions from slave to master and clear slave */
2079 ata_eh_about_to_do(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2080 mehc
->i
.action
|= sehc
->i
.action
;
2081 mehc
->i
.dev_action
[1] |= sehc
->i
.dev_action
[1];
2082 mehc
->i
.flags
|= sehc
->i
.flags
;
2083 ata_eh_done(ap
->slave_link
, NULL
, ATA_EH_ALL_ACTIONS
);
2086 /* Autopsy of fanout ports can affect host link autopsy.
2087 * Perform host link autopsy last.
2089 if (sata_pmp_attached(ap
))
2090 ata_eh_link_autopsy(&ap
->link
);
2094 * ata_eh_link_report - report error handling to user
2095 * @link: ATA link EH is going on
2097 * Report EH to user.
2102 static void ata_eh_link_report(struct ata_link
*link
)
2104 struct ata_port
*ap
= link
->ap
;
2105 struct ata_eh_context
*ehc
= &link
->eh_context
;
2106 const char *frozen
, *desc
;
2108 int tag
, nr_failed
= 0;
2110 if (ehc
->i
.flags
& ATA_EHI_QUIET
)
2114 if (ehc
->i
.desc
[0] != '\0')
2117 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2118 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2120 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2121 ata_dev_phys_link(qc
->dev
) != link
||
2122 ((qc
->flags
& ATA_QCFLAG_QUIET
) &&
2123 qc
->err_mask
== AC_ERR_DEV
))
2125 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
&& !qc
->err_mask
)
2131 if (!nr_failed
&& !ehc
->i
.err_mask
)
2135 if (ap
->pflags
& ATA_PFLAG_FROZEN
)
2138 memset(tries_buf
, 0, sizeof(tries_buf
));
2139 if (ap
->eh_tries
< ATA_EH_MAX_TRIES
)
2140 snprintf(tries_buf
, sizeof(tries_buf
) - 1, " t%d",
2144 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "exception Emask 0x%x "
2145 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2146 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2147 ehc
->i
.action
, frozen
, tries_buf
);
2149 ata_dev_printk(ehc
->i
.dev
, KERN_ERR
, "%s\n", desc
);
2151 ata_link_printk(link
, KERN_ERR
, "exception Emask 0x%x "
2152 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2153 ehc
->i
.err_mask
, link
->sactive
, ehc
->i
.serror
,
2154 ehc
->i
.action
, frozen
, tries_buf
);
2156 ata_link_printk(link
, KERN_ERR
, "%s\n", desc
);
2160 ata_link_printk(link
, KERN_ERR
,
2161 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2162 ehc
->i
.serror
& SERR_DATA_RECOVERED
? "RecovData " : "",
2163 ehc
->i
.serror
& SERR_COMM_RECOVERED
? "RecovComm " : "",
2164 ehc
->i
.serror
& SERR_DATA
? "UnrecovData " : "",
2165 ehc
->i
.serror
& SERR_PERSISTENT
? "Persist " : "",
2166 ehc
->i
.serror
& SERR_PROTOCOL
? "Proto " : "",
2167 ehc
->i
.serror
& SERR_INTERNAL
? "HostInt " : "",
2168 ehc
->i
.serror
& SERR_PHYRDY_CHG
? "PHYRdyChg " : "",
2169 ehc
->i
.serror
& SERR_PHY_INT_ERR
? "PHYInt " : "",
2170 ehc
->i
.serror
& SERR_COMM_WAKE
? "CommWake " : "",
2171 ehc
->i
.serror
& SERR_10B_8B_ERR
? "10B8B " : "",
2172 ehc
->i
.serror
& SERR_DISPARITY
? "Dispar " : "",
2173 ehc
->i
.serror
& SERR_CRC
? "BadCRC " : "",
2174 ehc
->i
.serror
& SERR_HANDSHAKE
? "Handshk " : "",
2175 ehc
->i
.serror
& SERR_LINK_SEQ_ERR
? "LinkSeq " : "",
2176 ehc
->i
.serror
& SERR_TRANS_ST_ERROR
? "TrStaTrns " : "",
2177 ehc
->i
.serror
& SERR_UNRECOG_FIS
? "UnrecFIS " : "",
2178 ehc
->i
.serror
& SERR_DEV_XCHG
? "DevExch " : "");
2180 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
2181 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
2182 struct ata_taskfile
*cmd
= &qc
->tf
, *res
= &qc
->result_tf
;
2183 const u8
*cdb
= qc
->cdb
;
2184 char data_buf
[20] = "";
2185 char cdb_buf
[70] = "";
2187 if (!(qc
->flags
& ATA_QCFLAG_FAILED
) ||
2188 ata_dev_phys_link(qc
->dev
) != link
|| !qc
->err_mask
)
2191 if (qc
->dma_dir
!= DMA_NONE
) {
2192 static const char *dma_str
[] = {
2193 [DMA_BIDIRECTIONAL
] = "bidi",
2194 [DMA_TO_DEVICE
] = "out",
2195 [DMA_FROM_DEVICE
] = "in",
2197 static const char *prot_str
[] = {
2198 [ATA_PROT_PIO
] = "pio",
2199 [ATA_PROT_DMA
] = "dma",
2200 [ATA_PROT_NCQ
] = "ncq",
2201 [ATAPI_PROT_PIO
] = "pio",
2202 [ATAPI_PROT_DMA
] = "dma",
2205 snprintf(data_buf
, sizeof(data_buf
), " %s %u %s",
2206 prot_str
[qc
->tf
.protocol
], qc
->nbytes
,
2207 dma_str
[qc
->dma_dir
]);
2210 if (ata_is_atapi(qc
->tf
.protocol
))
2211 snprintf(cdb_buf
, sizeof(cdb_buf
),
2212 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2213 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2214 cdb
[0], cdb
[1], cdb
[2], cdb
[3],
2215 cdb
[4], cdb
[5], cdb
[6], cdb
[7],
2216 cdb
[8], cdb
[9], cdb
[10], cdb
[11],
2217 cdb
[12], cdb
[13], cdb
[14], cdb
[15]);
2219 ata_dev_printk(qc
->dev
, KERN_ERR
,
2220 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2222 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2223 "Emask 0x%x (%s)%s\n",
2224 cmd
->command
, cmd
->feature
, cmd
->nsect
,
2225 cmd
->lbal
, cmd
->lbam
, cmd
->lbah
,
2226 cmd
->hob_feature
, cmd
->hob_nsect
,
2227 cmd
->hob_lbal
, cmd
->hob_lbam
, cmd
->hob_lbah
,
2228 cmd
->device
, qc
->tag
, data_buf
, cdb_buf
,
2229 res
->command
, res
->feature
, res
->nsect
,
2230 res
->lbal
, res
->lbam
, res
->lbah
,
2231 res
->hob_feature
, res
->hob_nsect
,
2232 res
->hob_lbal
, res
->hob_lbam
, res
->hob_lbah
,
2233 res
->device
, qc
->err_mask
, ata_err_string(qc
->err_mask
),
2234 qc
->err_mask
& AC_ERR_NCQ
? " <F>" : "");
2236 if (res
->command
& (ATA_BUSY
| ATA_DRDY
| ATA_DF
| ATA_DRQ
|
2238 if (res
->command
& ATA_BUSY
)
2239 ata_dev_printk(qc
->dev
, KERN_ERR
,
2240 "status: { Busy }\n");
2242 ata_dev_printk(qc
->dev
, KERN_ERR
,
2243 "status: { %s%s%s%s}\n",
2244 res
->command
& ATA_DRDY
? "DRDY " : "",
2245 res
->command
& ATA_DF
? "DF " : "",
2246 res
->command
& ATA_DRQ
? "DRQ " : "",
2247 res
->command
& ATA_ERR
? "ERR " : "");
2250 if (cmd
->command
!= ATA_CMD_PACKET
&&
2251 (res
->feature
& (ATA_ICRC
| ATA_UNC
| ATA_IDNF
|
2253 ata_dev_printk(qc
->dev
, KERN_ERR
,
2254 "error: { %s%s%s%s}\n",
2255 res
->feature
& ATA_ICRC
? "ICRC " : "",
2256 res
->feature
& ATA_UNC
? "UNC " : "",
2257 res
->feature
& ATA_IDNF
? "IDNF " : "",
2258 res
->feature
& ATA_ABORTED
? "ABRT " : "");
2263 * ata_eh_report - report error handling to user
2264 * @ap: ATA port to report EH about
2266 * Report EH to user.
2271 void ata_eh_report(struct ata_port
*ap
)
2273 struct ata_link
*link
;
2275 ata_for_each_link(link
, ap
, HOST_FIRST
)
2276 ata_eh_link_report(link
);
2279 static int ata_do_reset(struct ata_link
*link
, ata_reset_fn_t reset
,
2280 unsigned int *classes
, unsigned long deadline
,
2283 struct ata_device
*dev
;
2286 ata_for_each_dev(dev
, link
, ALL
)
2287 classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
2289 return reset(link
, classes
, deadline
);
2292 static int ata_eh_followup_srst_needed(struct ata_link
*link
,
2293 int rc
, const unsigned int *classes
)
2295 if ((link
->flags
& ATA_LFLAG_NO_SRST
) || ata_link_offline(link
))
2299 if (sata_pmp_supported(link
->ap
) && ata_is_host_link(link
))
2304 int ata_eh_reset(struct ata_link
*link
, int classify
,
2305 ata_prereset_fn_t prereset
, ata_reset_fn_t softreset
,
2306 ata_reset_fn_t hardreset
, ata_postreset_fn_t postreset
)
2308 struct ata_port
*ap
= link
->ap
;
2309 struct ata_link
*slave
= ap
->slave_link
;
2310 struct ata_eh_context
*ehc
= &link
->eh_context
;
2311 struct ata_eh_context
*sehc
= &slave
->eh_context
;
2312 unsigned int *classes
= ehc
->classes
;
2313 unsigned int lflags
= link
->flags
;
2314 int verbose
= !(ehc
->i
.flags
& ATA_EHI_QUIET
);
2315 int max_tries
= 0, try = 0;
2316 struct ata_link
*failed_link
;
2317 struct ata_device
*dev
;
2318 unsigned long deadline
, now
;
2319 ata_reset_fn_t reset
;
2320 unsigned long flags
;
2327 while (ata_eh_reset_timeouts
[max_tries
] != ULONG_MAX
)
2329 if (link
->flags
& ATA_LFLAG_NO_HRST
)
2331 if (link
->flags
& ATA_LFLAG_NO_SRST
)
2334 /* make sure each reset attemp is at least COOL_DOWN apart */
2335 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
2337 WARN_ON(time_after(ehc
->last_reset
, now
));
2338 deadline
= ata_deadline(ehc
->last_reset
,
2339 ATA_EH_RESET_COOL_DOWN
);
2340 if (time_before(now
, deadline
))
2341 schedule_timeout_uninterruptible(deadline
- now
);
2344 spin_lock_irqsave(ap
->lock
, flags
);
2345 ap
->pflags
|= ATA_PFLAG_RESETTING
;
2346 spin_unlock_irqrestore(ap
->lock
, flags
);
2348 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2350 ata_for_each_dev(dev
, link
, ALL
) {
2351 /* If we issue an SRST then an ATA drive (not ATAPI)
2352 * may change configuration and be in PIO0 timing. If
2353 * we do a hard reset (or are coming from power on)
2354 * this is true for ATA or ATAPI. Until we've set a
2355 * suitable controller mode we should not touch the
2356 * bus as we may be talking too fast.
2358 dev
->pio_mode
= XFER_PIO_0
;
2360 /* If the controller has a pio mode setup function
2361 * then use it to set the chipset to rights. Don't
2362 * touch the DMA setup as that will be dealt with when
2363 * configuring devices.
2365 if (ap
->ops
->set_piomode
)
2366 ap
->ops
->set_piomode(ap
, dev
);
2369 /* prefer hardreset */
2371 ehc
->i
.action
&= ~ATA_EH_RESET
;
2374 ehc
->i
.action
|= ATA_EH_HARDRESET
;
2375 } else if (softreset
) {
2377 ehc
->i
.action
|= ATA_EH_SOFTRESET
;
2381 unsigned long deadline
= ata_deadline(jiffies
,
2382 ATA_EH_PRERESET_TIMEOUT
);
2385 sehc
->i
.action
&= ~ATA_EH_RESET
;
2386 sehc
->i
.action
|= ehc
->i
.action
;
2389 rc
= prereset(link
, deadline
);
2391 /* If present, do prereset on slave link too. Reset
2392 * is skipped iff both master and slave links report
2393 * -ENOENT or clear ATA_EH_RESET.
2395 if (slave
&& (rc
== 0 || rc
== -ENOENT
)) {
2398 tmp
= prereset(slave
, deadline
);
2402 ehc
->i
.action
|= sehc
->i
.action
;
2406 if (rc
== -ENOENT
) {
2407 ata_link_printk(link
, KERN_DEBUG
,
2408 "port disabled. ignoring.\n");
2409 ehc
->i
.action
&= ~ATA_EH_RESET
;
2411 ata_for_each_dev(dev
, link
, ALL
)
2412 classes
[dev
->devno
] = ATA_DEV_NONE
;
2416 ata_link_printk(link
, KERN_ERR
,
2417 "prereset failed (errno=%d)\n", rc
);
2421 /* prereset() might have cleared ATA_EH_RESET. If so,
2422 * bang classes and return.
2424 if (reset
&& !(ehc
->i
.action
& ATA_EH_RESET
)) {
2425 ata_for_each_dev(dev
, link
, ALL
)
2426 classes
[dev
->devno
] = ATA_DEV_NONE
;
2436 if (ata_is_host_link(link
))
2437 ata_eh_freeze_port(ap
);
2439 deadline
= ata_deadline(jiffies
, ata_eh_reset_timeouts
[try++]);
2443 ata_link_printk(link
, KERN_INFO
, "%s resetting link\n",
2444 reset
== softreset
? "soft" : "hard");
2446 /* mark that this EH session started with reset */
2447 ehc
->last_reset
= jiffies
;
2448 if (reset
== hardreset
)
2449 ehc
->i
.flags
|= ATA_EHI_DID_HARDRESET
;
2451 ehc
->i
.flags
|= ATA_EHI_DID_SOFTRESET
;
2453 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2454 if (rc
&& rc
!= -EAGAIN
) {
2459 /* hardreset slave link if existent */
2460 if (slave
&& reset
== hardreset
) {
2464 ata_link_printk(slave
, KERN_INFO
,
2465 "hard resetting link\n");
2467 ata_eh_about_to_do(slave
, NULL
, ATA_EH_RESET
);
2468 tmp
= ata_do_reset(slave
, reset
, classes
, deadline
,
2476 failed_link
= slave
;
2482 /* perform follow-up SRST if necessary */
2483 if (reset
== hardreset
&&
2484 ata_eh_followup_srst_needed(link
, rc
, classes
)) {
2488 ata_link_printk(link
, KERN_ERR
,
2489 "follow-up softreset required "
2490 "but no softreset avaliable\n");
2496 ata_eh_about_to_do(link
, NULL
, ATA_EH_RESET
);
2497 rc
= ata_do_reset(link
, reset
, classes
, deadline
, true);
2501 ata_link_printk(link
, KERN_INFO
, "no reset method "
2502 "available, skipping reset\n");
2503 if (!(lflags
& ATA_LFLAG_ASSUME_CLASS
))
2504 lflags
|= ATA_LFLAG_ASSUME_ATA
;
2508 * Post-reset processing
2510 ata_for_each_dev(dev
, link
, ALL
) {
2511 /* After the reset, the device state is PIO 0 and the
2512 * controller state is undefined. Reset also wakes up
2513 * drives from sleeping mode.
2515 dev
->pio_mode
= XFER_PIO_0
;
2516 dev
->flags
&= ~ATA_DFLAG_SLEEPING
;
2518 if (!ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2519 /* apply class override */
2520 if (lflags
& ATA_LFLAG_ASSUME_ATA
)
2521 classes
[dev
->devno
] = ATA_DEV_ATA
;
2522 else if (lflags
& ATA_LFLAG_ASSUME_SEMB
)
2523 classes
[dev
->devno
] = ATA_DEV_SEMB_UNSUP
;
2525 classes
[dev
->devno
] = ATA_DEV_NONE
;
2528 /* record current link speed */
2529 if (sata_scr_read(link
, SCR_STATUS
, &sstatus
) == 0)
2530 link
->sata_spd
= (sstatus
>> 4) & 0xf;
2531 if (slave
&& sata_scr_read(slave
, SCR_STATUS
, &sstatus
) == 0)
2532 slave
->sata_spd
= (sstatus
>> 4) & 0xf;
2535 if (ata_is_host_link(link
))
2536 ata_eh_thaw_port(ap
);
2538 /* postreset() should clear hardware SError. Although SError
2539 * is cleared during link resume, clearing SError here is
2540 * necessary as some PHYs raise hotplug events after SRST.
2541 * This introduces race condition where hotplug occurs between
2542 * reset and here. This race is mediated by cross checking
2543 * link onlineness and classification result later.
2546 postreset(link
, classes
);
2548 postreset(slave
, classes
);
2551 /* clear cached SError */
2552 spin_lock_irqsave(link
->ap
->lock
, flags
);
2553 link
->eh_info
.serror
= 0;
2555 slave
->eh_info
.serror
= 0;
2556 spin_unlock_irqrestore(link
->ap
->lock
, flags
);
2558 /* Make sure onlineness and classification result correspond.
2559 * Hotplug could have happened during reset and some
2560 * controllers fail to wait while a drive is spinning up after
2561 * being hotplugged causing misdetection. By cross checking
2562 * link onlineness and classification result, those conditions
2563 * can be reliably detected and retried.
2566 ata_for_each_dev(dev
, link
, ALL
) {
2567 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2568 if (classes
[dev
->devno
] == ATA_DEV_UNKNOWN
) {
2569 classes
[dev
->devno
] = ATA_DEV_NONE
;
2570 if (ata_phys_link_online(ata_dev_phys_link(dev
)))
2575 if (classify
&& nr_unknown
) {
2576 if (try < max_tries
) {
2577 ata_link_printk(link
, KERN_WARNING
, "link online but "
2578 "device misclassified, retrying\n");
2583 ata_link_printk(link
, KERN_WARNING
,
2584 "link online but device misclassified, "
2585 "device detection might fail\n");
2588 /* reset successful, schedule revalidation */
2589 ata_eh_done(link
, NULL
, ATA_EH_RESET
);
2591 ata_eh_done(slave
, NULL
, ATA_EH_RESET
);
2592 ehc
->last_reset
= jiffies
; /* update to completion time */
2593 ehc
->i
.action
|= ATA_EH_REVALIDATE
;
2597 /* clear hotplug flag */
2598 ehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2600 sehc
->i
.flags
&= ~ATA_EHI_HOTPLUGGED
;
2602 spin_lock_irqsave(ap
->lock
, flags
);
2603 ap
->pflags
&= ~ATA_PFLAG_RESETTING
;
2604 spin_unlock_irqrestore(ap
->lock
, flags
);
2609 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2610 if (!ata_is_host_link(link
) &&
2611 sata_scr_read(link
, SCR_STATUS
, &sstatus
))
2614 if (rc
== -ERESTART
|| try >= max_tries
)
2618 if (time_before(now
, deadline
)) {
2619 unsigned long delta
= deadline
- now
;
2621 ata_link_printk(failed_link
, KERN_WARNING
,
2622 "reset failed (errno=%d), retrying in %u secs\n",
2623 rc
, DIV_ROUND_UP(jiffies_to_msecs(delta
), 1000));
2626 delta
= schedule_timeout_uninterruptible(delta
);
2629 if (try == max_tries
- 1) {
2630 sata_down_spd_limit(link
);
2632 sata_down_spd_limit(slave
);
2633 } else if (rc
== -EPIPE
)
2634 sata_down_spd_limit(failed_link
);
2641 static inline void ata_eh_pull_park_action(struct ata_port
*ap
)
2643 struct ata_link
*link
;
2644 struct ata_device
*dev
;
2645 unsigned long flags
;
2648 * This function can be thought of as an extended version of
2649 * ata_eh_about_to_do() specially crafted to accommodate the
2650 * requirements of ATA_EH_PARK handling. Since the EH thread
2651 * does not leave the do {} while () loop in ata_eh_recover as
2652 * long as the timeout for a park request to *one* device on
2653 * the port has not expired, and since we still want to pick
2654 * up park requests to other devices on the same port or
2655 * timeout updates for the same device, we have to pull
2656 * ATA_EH_PARK actions from eh_info into eh_context.i
2657 * ourselves at the beginning of each pass over the loop.
2659 * Additionally, all write accesses to &ap->park_req_pending
2660 * through INIT_COMPLETION() (see below) or complete_all()
2661 * (see ata_scsi_park_store()) are protected by the host lock.
2662 * As a result we have that park_req_pending.done is zero on
2663 * exit from this function, i.e. when ATA_EH_PARK actions for
2664 * *all* devices on port ap have been pulled into the
2665 * respective eh_context structs. If, and only if,
2666 * park_req_pending.done is non-zero by the time we reach
2667 * wait_for_completion_timeout(), another ATA_EH_PARK action
2668 * has been scheduled for at least one of the devices on port
2669 * ap and we have to cycle over the do {} while () loop in
2670 * ata_eh_recover() again.
2673 spin_lock_irqsave(ap
->lock
, flags
);
2674 INIT_COMPLETION(ap
->park_req_pending
);
2675 ata_for_each_link(link
, ap
, EDGE
) {
2676 ata_for_each_dev(dev
, link
, ALL
) {
2677 struct ata_eh_info
*ehi
= &link
->eh_info
;
2679 link
->eh_context
.i
.dev_action
[dev
->devno
] |=
2680 ehi
->dev_action
[dev
->devno
] & ATA_EH_PARK
;
2681 ata_eh_clear_action(link
, dev
, ehi
, ATA_EH_PARK
);
2684 spin_unlock_irqrestore(ap
->lock
, flags
);
2687 static void ata_eh_park_issue_cmd(struct ata_device
*dev
, int park
)
2689 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2690 struct ata_taskfile tf
;
2691 unsigned int err_mask
;
2693 ata_tf_init(dev
, &tf
);
2695 ehc
->unloaded_mask
|= 1 << dev
->devno
;
2696 tf
.command
= ATA_CMD_IDLEIMMEDIATE
;
2702 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2703 tf
.command
= ATA_CMD_CHK_POWER
;
2706 tf
.flags
|= ATA_TFLAG_DEVICE
| ATA_TFLAG_ISADDR
;
2707 tf
.protocol
|= ATA_PROT_NODATA
;
2708 err_mask
= ata_exec_internal(dev
, &tf
, NULL
, DMA_NONE
, NULL
, 0, 0);
2709 if (park
&& (err_mask
|| tf
.lbal
!= 0xc4)) {
2710 ata_dev_printk(dev
, KERN_ERR
, "head unload failed!\n");
2711 ehc
->unloaded_mask
&= ~(1 << dev
->devno
);
2715 static int ata_eh_revalidate_and_attach(struct ata_link
*link
,
2716 struct ata_device
**r_failed_dev
)
2718 struct ata_port
*ap
= link
->ap
;
2719 struct ata_eh_context
*ehc
= &link
->eh_context
;
2720 struct ata_device
*dev
;
2721 unsigned int new_mask
= 0;
2722 unsigned long flags
;
2727 /* For PATA drive side cable detection to work, IDENTIFY must
2728 * be done backwards such that PDIAG- is released by the slave
2729 * device before the master device is identified.
2731 ata_for_each_dev(dev
, link
, ALL_REVERSE
) {
2732 unsigned int action
= ata_eh_dev_action(dev
);
2733 unsigned int readid_flags
= 0;
2735 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
)
2736 readid_flags
|= ATA_READID_POSTRESET
;
2738 if ((action
& ATA_EH_REVALIDATE
) && ata_dev_enabled(dev
)) {
2739 WARN_ON(dev
->class == ATA_DEV_PMP
);
2741 if (ata_phys_link_offline(ata_dev_phys_link(dev
))) {
2746 ata_eh_about_to_do(link
, dev
, ATA_EH_REVALIDATE
);
2747 rc
= ata_dev_revalidate(dev
, ehc
->classes
[dev
->devno
],
2752 ata_eh_done(link
, dev
, ATA_EH_REVALIDATE
);
2754 /* Configuration may have changed, reconfigure
2757 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2759 /* schedule the scsi_rescan_device() here */
2760 queue_work(ata_aux_wq
, &(ap
->scsi_rescan_task
));
2761 } else if (dev
->class == ATA_DEV_UNKNOWN
&&
2762 ehc
->tries
[dev
->devno
] &&
2763 ata_class_enabled(ehc
->classes
[dev
->devno
])) {
2764 dev
->class = ehc
->classes
[dev
->devno
];
2766 if (dev
->class == ATA_DEV_PMP
)
2767 rc
= sata_pmp_attach(dev
);
2769 rc
= ata_dev_read_id(dev
, &dev
->class,
2770 readid_flags
, dev
->id
);
2773 /* clear error info accumulated during probe */
2774 ata_ering_clear(&dev
->ering
);
2775 new_mask
|= 1 << dev
->devno
;
2778 /* IDENTIFY was issued to non-existent
2779 * device. No need to reset. Just
2780 * thaw and kill the device.
2782 ata_eh_thaw_port(ap
);
2783 dev
->class = ATA_DEV_UNKNOWN
;
2786 dev
->class = ATA_DEV_UNKNOWN
;
2792 /* PDIAG- should have been released, ask cable type if post-reset */
2793 if ((ehc
->i
.flags
& ATA_EHI_DID_RESET
) && ata_is_host_link(link
)) {
2794 if (ap
->ops
->cable_detect
)
2795 ap
->cbl
= ap
->ops
->cable_detect(ap
);
2799 /* Configure new devices forward such that user doesn't see
2800 * device detection messages backwards.
2802 ata_for_each_dev(dev
, link
, ALL
) {
2803 if (!(new_mask
& (1 << dev
->devno
)) ||
2804 dev
->class == ATA_DEV_PMP
)
2807 ehc
->i
.flags
|= ATA_EHI_PRINTINFO
;
2808 rc
= ata_dev_configure(dev
);
2809 ehc
->i
.flags
&= ~ATA_EHI_PRINTINFO
;
2813 spin_lock_irqsave(ap
->lock
, flags
);
2814 ap
->pflags
|= ATA_PFLAG_SCSI_HOTPLUG
;
2815 spin_unlock_irqrestore(ap
->lock
, flags
);
2817 /* new device discovered, configure xfermode */
2818 ehc
->i
.flags
|= ATA_EHI_SETMODE
;
2824 *r_failed_dev
= dev
;
2825 DPRINTK("EXIT rc=%d\n", rc
);
2830 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2831 * @link: link on which timings will be programmed
2832 * @r_failed_dev: out paramter for failed device
2834 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2835 * ata_set_mode() fails, pointer to the failing device is
2836 * returned in @r_failed_dev.
2839 * PCI/etc. bus probe sem.
2842 * 0 on success, negative errno otherwise
2844 int ata_set_mode(struct ata_link
*link
, struct ata_device
**r_failed_dev
)
2846 struct ata_port
*ap
= link
->ap
;
2847 struct ata_device
*dev
;
2850 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2851 ata_for_each_dev(dev
, link
, ENABLED
) {
2852 if (!(dev
->flags
& ATA_DFLAG_DUBIOUS_XFER
)) {
2853 struct ata_ering_entry
*ent
;
2855 ent
= ata_ering_top(&dev
->ering
);
2857 ent
->eflags
&= ~ATA_EFLAG_DUBIOUS_XFER
;
2861 /* has private set_mode? */
2862 if (ap
->ops
->set_mode
)
2863 rc
= ap
->ops
->set_mode(link
, r_failed_dev
);
2865 rc
= ata_do_set_mode(link
, r_failed_dev
);
2867 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2868 ata_for_each_dev(dev
, link
, ENABLED
) {
2869 struct ata_eh_context
*ehc
= &link
->eh_context
;
2870 u8 saved_xfer_mode
= ehc
->saved_xfer_mode
[dev
->devno
];
2871 u8 saved_ncq
= !!(ehc
->saved_ncq_enabled
& (1 << dev
->devno
));
2873 if (dev
->xfer_mode
!= saved_xfer_mode
||
2874 ata_ncq_enabled(dev
) != saved_ncq
)
2875 dev
->flags
|= ATA_DFLAG_DUBIOUS_XFER
;
2882 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2883 * @dev: ATAPI device to clear UA for
2885 * Resets and other operations can make an ATAPI device raise
2886 * UNIT ATTENTION which causes the next operation to fail. This
2887 * function clears UA.
2890 * EH context (may sleep).
2893 * 0 on success, -errno on failure.
2895 static int atapi_eh_clear_ua(struct ata_device
*dev
)
2899 for (i
= 0; i
< ATA_EH_UA_TRIES
; i
++) {
2900 u8 sense_buffer
[SCSI_SENSE_BUFFERSIZE
];
2902 unsigned int err_mask
;
2904 err_mask
= atapi_eh_tur(dev
, &sense_key
);
2905 if (err_mask
!= 0 && err_mask
!= AC_ERR_DEV
) {
2906 ata_dev_printk(dev
, KERN_WARNING
, "TEST_UNIT_READY "
2907 "failed (err_mask=0x%x)\n", err_mask
);
2911 if (!err_mask
|| sense_key
!= UNIT_ATTENTION
)
2914 err_mask
= atapi_eh_request_sense(dev
, sense_buffer
, sense_key
);
2916 ata_dev_printk(dev
, KERN_WARNING
, "failed to clear "
2917 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask
);
2922 ata_dev_printk(dev
, KERN_WARNING
,
2923 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES
);
2928 static int ata_link_nr_enabled(struct ata_link
*link
)
2930 struct ata_device
*dev
;
2933 ata_for_each_dev(dev
, link
, ENABLED
)
2938 static int ata_link_nr_vacant(struct ata_link
*link
)
2940 struct ata_device
*dev
;
2943 ata_for_each_dev(dev
, link
, ALL
)
2944 if (dev
->class == ATA_DEV_UNKNOWN
)
2949 static int ata_eh_skip_recovery(struct ata_link
*link
)
2951 struct ata_port
*ap
= link
->ap
;
2952 struct ata_eh_context
*ehc
= &link
->eh_context
;
2953 struct ata_device
*dev
;
2955 /* skip disabled links */
2956 if (link
->flags
& ATA_LFLAG_DISABLED
)
2959 /* thaw frozen port and recover failed devices */
2960 if ((ap
->pflags
& ATA_PFLAG_FROZEN
) || ata_link_nr_enabled(link
))
2963 /* reset at least once if reset is requested */
2964 if ((ehc
->i
.action
& ATA_EH_RESET
) &&
2965 !(ehc
->i
.flags
& ATA_EHI_DID_RESET
))
2968 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2969 ata_for_each_dev(dev
, link
, ALL
) {
2970 if (dev
->class == ATA_DEV_UNKNOWN
&&
2971 ehc
->classes
[dev
->devno
] != ATA_DEV_NONE
)
2978 static int ata_eh_schedule_probe(struct ata_device
*dev
)
2980 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
2982 if (!(ehc
->i
.probe_mask
& (1 << dev
->devno
)) ||
2983 (ehc
->did_probe_mask
& (1 << dev
->devno
)))
2986 ata_eh_detach_dev(dev
);
2988 ehc
->did_probe_mask
|= (1 << dev
->devno
);
2989 ehc
->i
.action
|= ATA_EH_RESET
;
2990 ehc
->saved_xfer_mode
[dev
->devno
] = 0;
2991 ehc
->saved_ncq_enabled
&= ~(1 << dev
->devno
);
2996 static int ata_eh_handle_dev_fail(struct ata_device
*dev
, int err
)
2998 struct ata_eh_context
*ehc
= &dev
->link
->eh_context
;
3000 ehc
->tries
[dev
->devno
]--;
3004 /* device missing or wrong IDENTIFY data, schedule probing */
3005 ehc
->i
.probe_mask
|= (1 << dev
->devno
);
3007 /* give it just one more chance */
3008 ehc
->tries
[dev
->devno
] = min(ehc
->tries
[dev
->devno
], 1);
3010 if (ehc
->tries
[dev
->devno
] == 1) {
3011 /* This is the last chance, better to slow
3012 * down than lose it.
3014 sata_down_spd_limit(ata_dev_phys_link(dev
));
3015 if (dev
->pio_mode
> XFER_PIO_0
)
3016 ata_down_xfermask_limit(dev
, ATA_DNXFER_PIO
);
3020 if (ata_dev_enabled(dev
) && !ehc
->tries
[dev
->devno
]) {
3021 /* disable device if it has used up all its chances */
3022 ata_dev_disable(dev
);
3024 /* detach if offline */
3025 if (ata_phys_link_offline(ata_dev_phys_link(dev
)))
3026 ata_eh_detach_dev(dev
);
3028 /* schedule probe if necessary */
3029 if (ata_eh_schedule_probe(dev
)) {
3030 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3031 memset(ehc
->cmd_timeout_idx
[dev
->devno
], 0,
3032 sizeof(ehc
->cmd_timeout_idx
[dev
->devno
]));
3037 ehc
->i
.action
|= ATA_EH_RESET
;
3043 * ata_eh_recover - recover host port after error
3044 * @ap: host port to recover
3045 * @prereset: prereset method (can be NULL)
3046 * @softreset: softreset method (can be NULL)
3047 * @hardreset: hardreset method (can be NULL)
3048 * @postreset: postreset method (can be NULL)
3049 * @r_failed_link: out parameter for failed link
3051 * This is the alpha and omega, eum and yang, heart and soul of
3052 * libata exception handling. On entry, actions required to
3053 * recover each link and hotplug requests are recorded in the
3054 * link's eh_context. This function executes all the operations
3055 * with appropriate retrials and fallbacks to resurrect failed
3056 * devices, detach goners and greet newcomers.
3059 * Kernel thread context (may sleep).
3062 * 0 on success, -errno on failure.
3064 int ata_eh_recover(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3065 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3066 ata_postreset_fn_t postreset
,
3067 struct ata_link
**r_failed_link
)
3069 struct ata_link
*link
;
3070 struct ata_device
*dev
;
3073 unsigned long flags
, deadline
;
3077 /* prep for recovery */
3078 ata_for_each_link(link
, ap
, EDGE
) {
3079 struct ata_eh_context
*ehc
= &link
->eh_context
;
3081 /* re-enable link? */
3082 if (ehc
->i
.action
& ATA_EH_ENABLE_LINK
) {
3083 ata_eh_about_to_do(link
, NULL
, ATA_EH_ENABLE_LINK
);
3084 spin_lock_irqsave(ap
->lock
, flags
);
3085 link
->flags
&= ~ATA_LFLAG_DISABLED
;
3086 spin_unlock_irqrestore(ap
->lock
, flags
);
3087 ata_eh_done(link
, NULL
, ATA_EH_ENABLE_LINK
);
3090 ata_for_each_dev(dev
, link
, ALL
) {
3091 if (link
->flags
& ATA_LFLAG_NO_RETRY
)
3092 ehc
->tries
[dev
->devno
] = 1;
3094 ehc
->tries
[dev
->devno
] = ATA_EH_DEV_TRIES
;
3096 /* collect port action mask recorded in dev actions */
3097 ehc
->i
.action
|= ehc
->i
.dev_action
[dev
->devno
] &
3098 ~ATA_EH_PERDEV_MASK
;
3099 ehc
->i
.dev_action
[dev
->devno
] &= ATA_EH_PERDEV_MASK
;
3101 /* process hotplug request */
3102 if (dev
->flags
& ATA_DFLAG_DETACH
)
3103 ata_eh_detach_dev(dev
);
3105 /* schedule probe if necessary */
3106 if (!ata_dev_enabled(dev
))
3107 ata_eh_schedule_probe(dev
);
3115 /* if UNLOADING, finish immediately */
3116 if (ap
->pflags
& ATA_PFLAG_UNLOADING
)
3120 ata_for_each_link(link
, ap
, EDGE
) {
3121 struct ata_eh_context
*ehc
= &link
->eh_context
;
3123 /* skip EH if possible. */
3124 if (ata_eh_skip_recovery(link
))
3127 ata_for_each_dev(dev
, link
, ALL
)
3128 ehc
->classes
[dev
->devno
] = ATA_DEV_UNKNOWN
;
3132 ata_for_each_link(link
, ap
, EDGE
) {
3133 struct ata_eh_context
*ehc
= &link
->eh_context
;
3135 if (!(ehc
->i
.action
& ATA_EH_RESET
))
3138 rc
= ata_eh_reset(link
, ata_link_nr_vacant(link
),
3139 prereset
, softreset
, hardreset
, postreset
);
3141 ata_link_printk(link
, KERN_ERR
,
3142 "reset failed, giving up\n");
3151 * clears ATA_EH_PARK in eh_info and resets
3152 * ap->park_req_pending
3154 ata_eh_pull_park_action(ap
);
3157 ata_for_each_link(link
, ap
, EDGE
) {
3158 ata_for_each_dev(dev
, link
, ALL
) {
3159 struct ata_eh_context
*ehc
= &link
->eh_context
;
3162 if (dev
->class != ATA_DEV_ATA
)
3164 if (!(ehc
->i
.dev_action
[dev
->devno
] &
3167 tmp
= dev
->unpark_deadline
;
3168 if (time_before(deadline
, tmp
))
3170 else if (time_before_eq(tmp
, jiffies
))
3172 if (ehc
->unloaded_mask
& (1 << dev
->devno
))
3175 ata_eh_park_issue_cmd(dev
, 1);
3180 if (time_before_eq(deadline
, now
))
3183 deadline
= wait_for_completion_timeout(&ap
->park_req_pending
,
3186 ata_for_each_link(link
, ap
, EDGE
) {
3187 ata_for_each_dev(dev
, link
, ALL
) {
3188 if (!(link
->eh_context
.unloaded_mask
&
3192 ata_eh_park_issue_cmd(dev
, 0);
3193 ata_eh_done(link
, dev
, ATA_EH_PARK
);
3198 ata_for_each_link(link
, ap
, EDGE
) {
3199 struct ata_eh_context
*ehc
= &link
->eh_context
;
3201 /* revalidate existing devices and attach new ones */
3202 rc
= ata_eh_revalidate_and_attach(link
, &dev
);
3206 /* if PMP got attached, return, pmp EH will take care of it */
3207 if (link
->device
->class == ATA_DEV_PMP
) {
3212 /* configure transfer mode if necessary */
3213 if (ehc
->i
.flags
& ATA_EHI_SETMODE
) {
3214 rc
= ata_set_mode(link
, &dev
);
3217 ehc
->i
.flags
&= ~ATA_EHI_SETMODE
;
3220 /* If reset has been issued, clear UA to avoid
3221 * disrupting the current users of the device.
3223 if (ehc
->i
.flags
& ATA_EHI_DID_RESET
) {
3224 ata_for_each_dev(dev
, link
, ALL
) {
3225 if (dev
->class != ATA_DEV_ATAPI
)
3227 rc
= atapi_eh_clear_ua(dev
);
3233 /* configure link power saving */
3234 if (ehc
->i
.action
& ATA_EH_LPM
)
3235 ata_for_each_dev(dev
, link
, ALL
)
3236 ata_dev_enable_pm(dev
, ap
->pm_policy
);
3238 /* this link is okay now */
3244 ata_eh_handle_dev_fail(dev
, rc
);
3246 if (ap
->pflags
& ATA_PFLAG_FROZEN
) {
3247 /* PMP reset requires working host port.
3248 * Can't retry if it's frozen.
3250 if (sata_pmp_attached(ap
))
3260 if (rc
&& r_failed_link
)
3261 *r_failed_link
= link
;
3263 DPRINTK("EXIT, rc=%d\n", rc
);
3268 * ata_eh_finish - finish up EH
3269 * @ap: host port to finish EH for
3271 * Recovery is complete. Clean up EH states and retry or finish
3277 void ata_eh_finish(struct ata_port
*ap
)
3281 /* retry or finish qcs */
3282 for (tag
= 0; tag
< ATA_MAX_QUEUE
; tag
++) {
3283 struct ata_queued_cmd
*qc
= __ata_qc_from_tag(ap
, tag
);
3285 if (!(qc
->flags
& ATA_QCFLAG_FAILED
))
3289 /* FIXME: Once EH migration is complete,
3290 * generate sense data in this function,
3291 * considering both err_mask and tf.
3293 if (qc
->flags
& ATA_QCFLAG_RETRY
)
3294 ata_eh_qc_retry(qc
);
3296 ata_eh_qc_complete(qc
);
3298 if (qc
->flags
& ATA_QCFLAG_SENSE_VALID
) {
3299 ata_eh_qc_complete(qc
);
3301 /* feed zero TF to sense generation */
3302 memset(&qc
->result_tf
, 0, sizeof(qc
->result_tf
));
3303 ata_eh_qc_retry(qc
);
3308 /* make sure nr_active_links is zero after EH */
3309 WARN_ON(ap
->nr_active_links
);
3310 ap
->nr_active_links
= 0;
3314 * ata_do_eh - do standard error handling
3315 * @ap: host port to handle error for
3317 * @prereset: prereset method (can be NULL)
3318 * @softreset: softreset method (can be NULL)
3319 * @hardreset: hardreset method (can be NULL)
3320 * @postreset: postreset method (can be NULL)
3322 * Perform standard error handling sequence.
3325 * Kernel thread context (may sleep).
3327 void ata_do_eh(struct ata_port
*ap
, ata_prereset_fn_t prereset
,
3328 ata_reset_fn_t softreset
, ata_reset_fn_t hardreset
,
3329 ata_postreset_fn_t postreset
)
3331 struct ata_device
*dev
;
3337 rc
= ata_eh_recover(ap
, prereset
, softreset
, hardreset
, postreset
,
3340 ata_for_each_dev(dev
, &ap
->link
, ALL
)
3341 ata_dev_disable(dev
);
3348 * ata_std_error_handler - standard error handler
3349 * @ap: host port to handle error for
3351 * Standard error handler
3354 * Kernel thread context (may sleep).
3356 void ata_std_error_handler(struct ata_port
*ap
)
3358 struct ata_port_operations
*ops
= ap
->ops
;
3359 ata_reset_fn_t hardreset
= ops
->hardreset
;
3361 /* ignore built-in hardreset if SCR access is not available */
3362 if (ata_is_builtin_hardreset(hardreset
) && !sata_scr_valid(&ap
->link
))
3365 ata_do_eh(ap
, ops
->prereset
, ops
->softreset
, hardreset
, ops
->postreset
);
3370 * ata_eh_handle_port_suspend - perform port suspend operation
3371 * @ap: port to suspend
3376 * Kernel thread context (may sleep).
3378 static void ata_eh_handle_port_suspend(struct ata_port
*ap
)
3380 unsigned long flags
;
3383 /* are we suspending? */
3384 spin_lock_irqsave(ap
->lock
, flags
);
3385 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3386 ap
->pm_mesg
.event
== PM_EVENT_ON
) {
3387 spin_unlock_irqrestore(ap
->lock
, flags
);
3390 spin_unlock_irqrestore(ap
->lock
, flags
);
3392 WARN_ON(ap
->pflags
& ATA_PFLAG_SUSPENDED
);
3394 /* tell ACPI we're suspending */
3395 rc
= ata_acpi_on_suspend(ap
);
3400 ata_eh_freeze_port(ap
);
3402 if (ap
->ops
->port_suspend
)
3403 rc
= ap
->ops
->port_suspend(ap
, ap
->pm_mesg
);
3405 ata_acpi_set_state(ap
, PMSG_SUSPEND
);
3408 spin_lock_irqsave(ap
->lock
, flags
);
3410 ap
->pflags
&= ~ATA_PFLAG_PM_PENDING
;
3412 ap
->pflags
|= ATA_PFLAG_SUSPENDED
;
3413 else if (ap
->pflags
& ATA_PFLAG_FROZEN
)
3414 ata_port_schedule_eh(ap
);
3416 if (ap
->pm_result
) {
3417 *ap
->pm_result
= rc
;
3418 ap
->pm_result
= NULL
;
3421 spin_unlock_irqrestore(ap
->lock
, flags
);
3427 * ata_eh_handle_port_resume - perform port resume operation
3428 * @ap: port to resume
3433 * Kernel thread context (may sleep).
3435 static void ata_eh_handle_port_resume(struct ata_port
*ap
)
3437 unsigned long flags
;
3440 /* are we resuming? */
3441 spin_lock_irqsave(ap
->lock
, flags
);
3442 if (!(ap
->pflags
& ATA_PFLAG_PM_PENDING
) ||
3443 ap
->pm_mesg
.event
!= PM_EVENT_ON
) {
3444 spin_unlock_irqrestore(ap
->lock
, flags
);
3447 spin_unlock_irqrestore(ap
->lock
, flags
);
3449 WARN_ON(!(ap
->pflags
& ATA_PFLAG_SUSPENDED
));
3451 ata_acpi_set_state(ap
, PMSG_ON
);
3453 if (ap
->ops
->port_resume
)
3454 rc
= ap
->ops
->port_resume(ap
);
3456 /* tell ACPI that we're resuming */
3457 ata_acpi_on_resume(ap
);
3460 spin_lock_irqsave(ap
->lock
, flags
);
3461 ap
->pflags
&= ~(ATA_PFLAG_PM_PENDING
| ATA_PFLAG_SUSPENDED
);
3462 if (ap
->pm_result
) {
3463 *ap
->pm_result
= rc
;
3464 ap
->pm_result
= NULL
;
3466 spin_unlock_irqrestore(ap
->lock
, flags
);
3468 #endif /* CONFIG_PM */