libata-eh: clear UNIT ATTENTION after reset
[linux-2.6/verdex.git] / drivers / ata / libata-eh.c
blobf2dd99122bd6fc43652f9f9c948d61a33745e5bc
1 /*
2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
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,
24 * USA.
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/pci.h>
37 #include <scsi/scsi.h>
38 #include <scsi/scsi_host.h>
39 #include <scsi/scsi_eh.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_cmnd.h>
42 #include "../scsi/scsi_transport_api.h"
44 #include <linux/libata.h>
46 #include "libata.h"
48 enum {
49 /* speed down verdicts */
50 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
51 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
52 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
53 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
55 /* error flags */
56 ATA_EFLAG_IS_IO = (1 << 0),
57 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
59 /* error categories */
60 ATA_ECAT_NONE = 0,
61 ATA_ECAT_ATA_BUS = 1,
62 ATA_ECAT_TOUT_HSM = 2,
63 ATA_ECAT_UNK_DEV = 3,
64 ATA_ECAT_DUBIOUS_NONE = 4,
65 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
66 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
67 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
68 ATA_ECAT_NR = 8,
70 ATA_EH_CMD_DFL_TIMEOUT = 5000,
72 /* always put at least this amount of time between resets */
73 ATA_EH_RESET_COOL_DOWN = 5000,
75 /* Waiting in ->prereset can never be reliable. It's
76 * sometimes nice to wait there but it can't be depended upon;
77 * otherwise, we wouldn't be resetting. Just give it enough
78 * time for most drives to spin up.
80 ATA_EH_PRERESET_TIMEOUT = 10000,
81 ATA_EH_FASTDRAIN_INTERVAL = 3000,
83 ATA_EH_UA_TRIES = 5,
86 /* The following table determines how we sequence resets. Each entry
87 * represents timeout for that try. The first try can be soft or
88 * hardreset. All others are hardreset if available. In most cases
89 * the first reset w/ 10sec timeout should succeed. Following entries
90 * are mostly for error handling, hotplug and retarded devices.
92 static const unsigned long ata_eh_reset_timeouts[] = {
93 10000, /* most drives spin up by 10sec */
94 10000, /* > 99% working drives spin up before 20sec */
95 35000, /* give > 30 secs of idleness for retarded devices */
96 5000, /* and sweet one last chance */
97 ULONG_MAX, /* > 1 min has elapsed, give up */
100 static const unsigned long ata_eh_identify_timeouts[] = {
101 5000, /* covers > 99% of successes and not too boring on failures */
102 10000, /* combined time till here is enough even for media access */
103 30000, /* for true idiots */
104 ULONG_MAX,
107 static const unsigned long ata_eh_other_timeouts[] = {
108 5000, /* same rationale as identify timeout */
109 10000, /* ditto */
110 /* but no merciful 30sec for other commands, it just isn't worth it */
111 ULONG_MAX,
114 struct ata_eh_cmd_timeout_ent {
115 const u8 *commands;
116 const unsigned long *timeouts;
119 /* The following table determines timeouts to use for EH internal
120 * commands. Each table entry is a command class and matches the
121 * commands the entry applies to and the timeout table to use.
123 * On the retry after a command timed out, the next timeout value from
124 * the table is used. If the table doesn't contain further entries,
125 * the last value is used.
127 * ehc->cmd_timeout_idx keeps track of which timeout to use per
128 * command class, so if SET_FEATURES times out on the first try, the
129 * next try will use the second timeout value only for that class.
131 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
132 static const struct ata_eh_cmd_timeout_ent
133 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
134 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
135 .timeouts = ata_eh_identify_timeouts, },
136 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
137 .timeouts = ata_eh_other_timeouts, },
138 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
139 .timeouts = ata_eh_other_timeouts, },
140 { .commands = CMDS(ATA_CMD_SET_FEATURES),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
143 .timeouts = ata_eh_other_timeouts, },
145 #undef CMDS
147 static void __ata_port_freeze(struct ata_port *ap);
148 #ifdef CONFIG_PM
149 static void ata_eh_handle_port_suspend(struct ata_port *ap);
150 static void ata_eh_handle_port_resume(struct ata_port *ap);
151 #else /* CONFIG_PM */
152 static void ata_eh_handle_port_suspend(struct ata_port *ap)
155 static void ata_eh_handle_port_resume(struct ata_port *ap)
157 #endif /* CONFIG_PM */
159 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
160 va_list args)
162 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
163 ATA_EH_DESC_LEN - ehi->desc_len,
164 fmt, args);
168 * __ata_ehi_push_desc - push error description without adding separator
169 * @ehi: target EHI
170 * @fmt: printf format string
172 * Format string according to @fmt and append it to @ehi->desc.
174 * LOCKING:
175 * spin_lock_irqsave(host lock)
177 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
179 va_list args;
181 va_start(args, fmt);
182 __ata_ehi_pushv_desc(ehi, fmt, args);
183 va_end(args);
187 * ata_ehi_push_desc - push error description with separator
188 * @ehi: target EHI
189 * @fmt: printf format string
191 * Format string according to @fmt and append it to @ehi->desc.
192 * If @ehi->desc is not empty, ", " is added in-between.
194 * LOCKING:
195 * spin_lock_irqsave(host lock)
197 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
199 va_list args;
201 if (ehi->desc_len)
202 __ata_ehi_push_desc(ehi, ", ");
204 va_start(args, fmt);
205 __ata_ehi_pushv_desc(ehi, fmt, args);
206 va_end(args);
210 * ata_ehi_clear_desc - clean error description
211 * @ehi: target EHI
213 * Clear @ehi->desc.
215 * LOCKING:
216 * spin_lock_irqsave(host lock)
218 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
220 ehi->desc[0] = '\0';
221 ehi->desc_len = 0;
225 * ata_port_desc - append port description
226 * @ap: target ATA port
227 * @fmt: printf format string
229 * Format string according to @fmt and append it to port
230 * description. If port description is not empty, " " is added
231 * in-between. This function is to be used while initializing
232 * ata_host. The description is printed on host registration.
234 * LOCKING:
235 * None.
237 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
239 va_list args;
241 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
243 if (ap->link.eh_info.desc_len)
244 __ata_ehi_push_desc(&ap->link.eh_info, " ");
246 va_start(args, fmt);
247 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
248 va_end(args);
251 #ifdef CONFIG_PCI
254 * ata_port_pbar_desc - append PCI BAR description
255 * @ap: target ATA port
256 * @bar: target PCI BAR
257 * @offset: offset into PCI BAR
258 * @name: name of the area
260 * If @offset is negative, this function formats a string which
261 * contains the name, address, size and type of the BAR and
262 * appends it to the port description. If @offset is zero or
263 * positive, only name and offsetted address is appended.
265 * LOCKING:
266 * None.
268 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
269 const char *name)
271 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
272 char *type = "";
273 unsigned long long start, len;
275 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
276 type = "m";
277 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
278 type = "i";
280 start = (unsigned long long)pci_resource_start(pdev, bar);
281 len = (unsigned long long)pci_resource_len(pdev, bar);
283 if (offset < 0)
284 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
285 else
286 ata_port_desc(ap, "%s 0x%llx", name,
287 start + (unsigned long long)offset);
290 #endif /* CONFIG_PCI */
292 static int ata_lookup_timeout_table(u8 cmd)
294 int i;
296 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
297 const u8 *cur;
299 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
300 if (*cur == cmd)
301 return i;
304 return -1;
308 * ata_internal_cmd_timeout - determine timeout for an internal command
309 * @dev: target device
310 * @cmd: internal command to be issued
312 * Determine timeout for internal command @cmd for @dev.
314 * LOCKING:
315 * EH context.
317 * RETURNS:
318 * Determined timeout.
320 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
322 struct ata_eh_context *ehc = &dev->link->eh_context;
323 int ent = ata_lookup_timeout_table(cmd);
324 int idx;
326 if (ent < 0)
327 return ATA_EH_CMD_DFL_TIMEOUT;
329 idx = ehc->cmd_timeout_idx[dev->devno][ent];
330 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
334 * ata_internal_cmd_timed_out - notification for internal command timeout
335 * @dev: target device
336 * @cmd: internal command which timed out
338 * Notify EH that internal command @cmd for @dev timed out. This
339 * function should be called only for commands whose timeouts are
340 * determined using ata_internal_cmd_timeout().
342 * LOCKING:
343 * EH context.
345 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
347 struct ata_eh_context *ehc = &dev->link->eh_context;
348 int ent = ata_lookup_timeout_table(cmd);
349 int idx;
351 if (ent < 0)
352 return;
354 idx = ehc->cmd_timeout_idx[dev->devno][ent];
355 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
356 ehc->cmd_timeout_idx[dev->devno][ent]++;
359 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
360 unsigned int err_mask)
362 struct ata_ering_entry *ent;
364 WARN_ON(!err_mask);
366 ering->cursor++;
367 ering->cursor %= ATA_ERING_SIZE;
369 ent = &ering->ring[ering->cursor];
370 ent->eflags = eflags;
371 ent->err_mask = err_mask;
372 ent->timestamp = get_jiffies_64();
375 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
377 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
379 if (ent->err_mask)
380 return ent;
381 return NULL;
384 static void ata_ering_clear(struct ata_ering *ering)
386 memset(ering, 0, sizeof(*ering));
389 static int ata_ering_map(struct ata_ering *ering,
390 int (*map_fn)(struct ata_ering_entry *, void *),
391 void *arg)
393 int idx, rc = 0;
394 struct ata_ering_entry *ent;
396 idx = ering->cursor;
397 do {
398 ent = &ering->ring[idx];
399 if (!ent->err_mask)
400 break;
401 rc = map_fn(ent, arg);
402 if (rc)
403 break;
404 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
405 } while (idx != ering->cursor);
407 return rc;
410 static unsigned int ata_eh_dev_action(struct ata_device *dev)
412 struct ata_eh_context *ehc = &dev->link->eh_context;
414 return ehc->i.action | ehc->i.dev_action[dev->devno];
417 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
418 struct ata_eh_info *ehi, unsigned int action)
420 struct ata_device *tdev;
422 if (!dev) {
423 ehi->action &= ~action;
424 ata_link_for_each_dev(tdev, link)
425 ehi->dev_action[tdev->devno] &= ~action;
426 } else {
427 /* doesn't make sense for port-wide EH actions */
428 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
430 /* break ehi->action into ehi->dev_action */
431 if (ehi->action & action) {
432 ata_link_for_each_dev(tdev, link)
433 ehi->dev_action[tdev->devno] |=
434 ehi->action & action;
435 ehi->action &= ~action;
438 /* turn off the specified per-dev action */
439 ehi->dev_action[dev->devno] &= ~action;
444 * ata_scsi_timed_out - SCSI layer time out callback
445 * @cmd: timed out SCSI command
447 * Handles SCSI layer timeout. We race with normal completion of
448 * the qc for @cmd. If the qc is already gone, we lose and let
449 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
450 * timed out and EH should be invoked. Prevent ata_qc_complete()
451 * from finishing it by setting EH_SCHEDULED and return
452 * EH_NOT_HANDLED.
454 * TODO: kill this function once old EH is gone.
456 * LOCKING:
457 * Called from timer context
459 * RETURNS:
460 * EH_HANDLED or EH_NOT_HANDLED
462 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
464 struct Scsi_Host *host = cmd->device->host;
465 struct ata_port *ap = ata_shost_to_port(host);
466 unsigned long flags;
467 struct ata_queued_cmd *qc;
468 enum scsi_eh_timer_return ret;
470 DPRINTK("ENTER\n");
472 if (ap->ops->error_handler) {
473 ret = EH_NOT_HANDLED;
474 goto out;
477 ret = EH_HANDLED;
478 spin_lock_irqsave(ap->lock, flags);
479 qc = ata_qc_from_tag(ap, ap->link.active_tag);
480 if (qc) {
481 WARN_ON(qc->scsicmd != cmd);
482 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
483 qc->err_mask |= AC_ERR_TIMEOUT;
484 ret = EH_NOT_HANDLED;
486 spin_unlock_irqrestore(ap->lock, flags);
488 out:
489 DPRINTK("EXIT, ret=%d\n", ret);
490 return ret;
494 * ata_scsi_error - SCSI layer error handler callback
495 * @host: SCSI host on which error occurred
497 * Handles SCSI-layer-thrown error events.
499 * LOCKING:
500 * Inherited from SCSI layer (none, can sleep)
502 * RETURNS:
503 * Zero.
505 void ata_scsi_error(struct Scsi_Host *host)
507 struct ata_port *ap = ata_shost_to_port(host);
508 int i;
509 unsigned long flags;
511 DPRINTK("ENTER\n");
513 /* synchronize with port task */
514 ata_port_flush_task(ap);
516 /* synchronize with host lock and sort out timeouts */
518 /* For new EH, all qcs are finished in one of three ways -
519 * normal completion, error completion, and SCSI timeout.
520 * Both cmpletions can race against SCSI timeout. When normal
521 * completion wins, the qc never reaches EH. When error
522 * completion wins, the qc has ATA_QCFLAG_FAILED set.
524 * When SCSI timeout wins, things are a bit more complex.
525 * Normal or error completion can occur after the timeout but
526 * before this point. In such cases, both types of
527 * completions are honored. A scmd is determined to have
528 * timed out iff its associated qc is active and not failed.
530 if (ap->ops->error_handler) {
531 struct scsi_cmnd *scmd, *tmp;
532 int nr_timedout = 0;
534 spin_lock_irqsave(ap->lock, flags);
536 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
537 struct ata_queued_cmd *qc;
539 for (i = 0; i < ATA_MAX_QUEUE; i++) {
540 qc = __ata_qc_from_tag(ap, i);
541 if (qc->flags & ATA_QCFLAG_ACTIVE &&
542 qc->scsicmd == scmd)
543 break;
546 if (i < ATA_MAX_QUEUE) {
547 /* the scmd has an associated qc */
548 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
549 /* which hasn't failed yet, timeout */
550 qc->err_mask |= AC_ERR_TIMEOUT;
551 qc->flags |= ATA_QCFLAG_FAILED;
552 nr_timedout++;
554 } else {
555 /* Normal completion occurred after
556 * SCSI timeout but before this point.
557 * Successfully complete it.
559 scmd->retries = scmd->allowed;
560 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
564 /* If we have timed out qcs. They belong to EH from
565 * this point but the state of the controller is
566 * unknown. Freeze the port to make sure the IRQ
567 * handler doesn't diddle with those qcs. This must
568 * be done atomically w.r.t. setting QCFLAG_FAILED.
570 if (nr_timedout)
571 __ata_port_freeze(ap);
573 spin_unlock_irqrestore(ap->lock, flags);
575 /* initialize eh_tries */
576 ap->eh_tries = ATA_EH_MAX_TRIES;
577 } else
578 spin_unlock_wait(ap->lock);
580 repeat:
581 /* invoke error handler */
582 if (ap->ops->error_handler) {
583 struct ata_link *link;
585 /* kill fast drain timer */
586 del_timer_sync(&ap->fastdrain_timer);
588 /* process port resume request */
589 ata_eh_handle_port_resume(ap);
591 /* fetch & clear EH info */
592 spin_lock_irqsave(ap->lock, flags);
594 __ata_port_for_each_link(link, ap) {
595 struct ata_eh_context *ehc = &link->eh_context;
596 struct ata_device *dev;
598 memset(&link->eh_context, 0, sizeof(link->eh_context));
599 link->eh_context.i = link->eh_info;
600 memset(&link->eh_info, 0, sizeof(link->eh_info));
602 ata_link_for_each_dev(dev, link) {
603 int devno = dev->devno;
605 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
606 if (ata_ncq_enabled(dev))
607 ehc->saved_ncq_enabled |= 1 << devno;
610 /* set last reset timestamp to some time in the past */
611 ehc->last_reset = jiffies - 60 * HZ;
614 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
615 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
616 ap->excl_link = NULL; /* don't maintain exclusion over EH */
618 spin_unlock_irqrestore(ap->lock, flags);
620 /* invoke EH, skip if unloading or suspended */
621 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
622 ap->ops->error_handler(ap);
623 else
624 ata_eh_finish(ap);
626 /* process port suspend request */
627 ata_eh_handle_port_suspend(ap);
629 /* Exception might have happend after ->error_handler
630 * recovered the port but before this point. Repeat
631 * EH in such case.
633 spin_lock_irqsave(ap->lock, flags);
635 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
636 if (--ap->eh_tries) {
637 spin_unlock_irqrestore(ap->lock, flags);
638 goto repeat;
640 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
641 "tries, giving up\n", ATA_EH_MAX_TRIES);
642 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
645 /* this run is complete, make sure EH info is clear */
646 __ata_port_for_each_link(link, ap)
647 memset(&link->eh_info, 0, sizeof(link->eh_info));
649 /* Clear host_eh_scheduled while holding ap->lock such
650 * that if exception occurs after this point but
651 * before EH completion, SCSI midlayer will
652 * re-initiate EH.
654 host->host_eh_scheduled = 0;
656 spin_unlock_irqrestore(ap->lock, flags);
657 } else {
658 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
659 ap->ops->eng_timeout(ap);
662 /* finish or retry handled scmd's and clean up */
663 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
665 scsi_eh_flush_done_q(&ap->eh_done_q);
667 /* clean up */
668 spin_lock_irqsave(ap->lock, flags);
670 if (ap->pflags & ATA_PFLAG_LOADING)
671 ap->pflags &= ~ATA_PFLAG_LOADING;
672 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
673 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
675 if (ap->pflags & ATA_PFLAG_RECOVERED)
676 ata_port_printk(ap, KERN_INFO, "EH complete\n");
678 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
680 /* tell wait_eh that we're done */
681 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
682 wake_up_all(&ap->eh_wait_q);
684 spin_unlock_irqrestore(ap->lock, flags);
686 DPRINTK("EXIT\n");
690 * ata_port_wait_eh - Wait for the currently pending EH to complete
691 * @ap: Port to wait EH for
693 * Wait until the currently pending EH is complete.
695 * LOCKING:
696 * Kernel thread context (may sleep).
698 void ata_port_wait_eh(struct ata_port *ap)
700 unsigned long flags;
701 DEFINE_WAIT(wait);
703 retry:
704 spin_lock_irqsave(ap->lock, flags);
706 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
707 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
708 spin_unlock_irqrestore(ap->lock, flags);
709 schedule();
710 spin_lock_irqsave(ap->lock, flags);
712 finish_wait(&ap->eh_wait_q, &wait);
714 spin_unlock_irqrestore(ap->lock, flags);
716 /* make sure SCSI EH is complete */
717 if (scsi_host_in_recovery(ap->scsi_host)) {
718 msleep(10);
719 goto retry;
723 static int ata_eh_nr_in_flight(struct ata_port *ap)
725 unsigned int tag;
726 int nr = 0;
728 /* count only non-internal commands */
729 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
730 if (ata_qc_from_tag(ap, tag))
731 nr++;
733 return nr;
736 void ata_eh_fastdrain_timerfn(unsigned long arg)
738 struct ata_port *ap = (void *)arg;
739 unsigned long flags;
740 int cnt;
742 spin_lock_irqsave(ap->lock, flags);
744 cnt = ata_eh_nr_in_flight(ap);
746 /* are we done? */
747 if (!cnt)
748 goto out_unlock;
750 if (cnt == ap->fastdrain_cnt) {
751 unsigned int tag;
753 /* No progress during the last interval, tag all
754 * in-flight qcs as timed out and freeze the port.
756 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
757 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
758 if (qc)
759 qc->err_mask |= AC_ERR_TIMEOUT;
762 ata_port_freeze(ap);
763 } else {
764 /* some qcs have finished, give it another chance */
765 ap->fastdrain_cnt = cnt;
766 ap->fastdrain_timer.expires =
767 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
768 add_timer(&ap->fastdrain_timer);
771 out_unlock:
772 spin_unlock_irqrestore(ap->lock, flags);
776 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
777 * @ap: target ATA port
778 * @fastdrain: activate fast drain
780 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
781 * is non-zero and EH wasn't pending before. Fast drain ensures
782 * that EH kicks in in timely manner.
784 * LOCKING:
785 * spin_lock_irqsave(host lock)
787 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
789 int cnt;
791 /* already scheduled? */
792 if (ap->pflags & ATA_PFLAG_EH_PENDING)
793 return;
795 ap->pflags |= ATA_PFLAG_EH_PENDING;
797 if (!fastdrain)
798 return;
800 /* do we have in-flight qcs? */
801 cnt = ata_eh_nr_in_flight(ap);
802 if (!cnt)
803 return;
805 /* activate fast drain */
806 ap->fastdrain_cnt = cnt;
807 ap->fastdrain_timer.expires =
808 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
809 add_timer(&ap->fastdrain_timer);
813 * ata_qc_schedule_eh - schedule qc for error handling
814 * @qc: command to schedule error handling for
816 * Schedule error handling for @qc. EH will kick in as soon as
817 * other commands are drained.
819 * LOCKING:
820 * spin_lock_irqsave(host lock)
822 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
824 struct ata_port *ap = qc->ap;
826 WARN_ON(!ap->ops->error_handler);
828 qc->flags |= ATA_QCFLAG_FAILED;
829 ata_eh_set_pending(ap, 1);
831 /* The following will fail if timeout has already expired.
832 * ata_scsi_error() takes care of such scmds on EH entry.
833 * Note that ATA_QCFLAG_FAILED is unconditionally set after
834 * this function completes.
836 scsi_req_abort_cmd(qc->scsicmd);
840 * ata_port_schedule_eh - schedule error handling without a qc
841 * @ap: ATA port to schedule EH for
843 * Schedule error handling for @ap. EH will kick in as soon as
844 * all commands are drained.
846 * LOCKING:
847 * spin_lock_irqsave(host lock)
849 void ata_port_schedule_eh(struct ata_port *ap)
851 WARN_ON(!ap->ops->error_handler);
853 if (ap->pflags & ATA_PFLAG_INITIALIZING)
854 return;
856 ata_eh_set_pending(ap, 1);
857 scsi_schedule_eh(ap->scsi_host);
859 DPRINTK("port EH scheduled\n");
862 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
864 int tag, nr_aborted = 0;
866 WARN_ON(!ap->ops->error_handler);
868 /* we're gonna abort all commands, no need for fast drain */
869 ata_eh_set_pending(ap, 0);
871 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
872 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
874 if (qc && (!link || qc->dev->link == link)) {
875 qc->flags |= ATA_QCFLAG_FAILED;
876 ata_qc_complete(qc);
877 nr_aborted++;
881 if (!nr_aborted)
882 ata_port_schedule_eh(ap);
884 return nr_aborted;
888 * ata_link_abort - abort all qc's on the link
889 * @link: ATA link to abort qc's for
891 * Abort all active qc's active on @link and schedule EH.
893 * LOCKING:
894 * spin_lock_irqsave(host lock)
896 * RETURNS:
897 * Number of aborted qc's.
899 int ata_link_abort(struct ata_link *link)
901 return ata_do_link_abort(link->ap, link);
905 * ata_port_abort - abort all qc's on the port
906 * @ap: ATA port to abort qc's for
908 * Abort all active qc's of @ap and schedule EH.
910 * LOCKING:
911 * spin_lock_irqsave(host_set lock)
913 * RETURNS:
914 * Number of aborted qc's.
916 int ata_port_abort(struct ata_port *ap)
918 return ata_do_link_abort(ap, NULL);
922 * __ata_port_freeze - freeze port
923 * @ap: ATA port to freeze
925 * This function is called when HSM violation or some other
926 * condition disrupts normal operation of the port. Frozen port
927 * is not allowed to perform any operation until the port is
928 * thawed, which usually follows a successful reset.
930 * ap->ops->freeze() callback can be used for freezing the port
931 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
932 * port cannot be frozen hardware-wise, the interrupt handler
933 * must ack and clear interrupts unconditionally while the port
934 * is frozen.
936 * LOCKING:
937 * spin_lock_irqsave(host lock)
939 static void __ata_port_freeze(struct ata_port *ap)
941 WARN_ON(!ap->ops->error_handler);
943 if (ap->ops->freeze)
944 ap->ops->freeze(ap);
946 ap->pflags |= ATA_PFLAG_FROZEN;
948 DPRINTK("ata%u port frozen\n", ap->print_id);
952 * ata_port_freeze - abort & freeze port
953 * @ap: ATA port to freeze
955 * Abort and freeze @ap.
957 * LOCKING:
958 * spin_lock_irqsave(host lock)
960 * RETURNS:
961 * Number of aborted commands.
963 int ata_port_freeze(struct ata_port *ap)
965 int nr_aborted;
967 WARN_ON(!ap->ops->error_handler);
969 nr_aborted = ata_port_abort(ap);
970 __ata_port_freeze(ap);
972 return nr_aborted;
976 * sata_async_notification - SATA async notification handler
977 * @ap: ATA port where async notification is received
979 * Handler to be called when async notification via SDB FIS is
980 * received. This function schedules EH if necessary.
982 * LOCKING:
983 * spin_lock_irqsave(host lock)
985 * RETURNS:
986 * 1 if EH is scheduled, 0 otherwise.
988 int sata_async_notification(struct ata_port *ap)
990 u32 sntf;
991 int rc;
993 if (!(ap->flags & ATA_FLAG_AN))
994 return 0;
996 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
997 if (rc == 0)
998 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1000 if (!sata_pmp_attached(ap) || rc) {
1001 /* PMP is not attached or SNTF is not available */
1002 if (!sata_pmp_attached(ap)) {
1003 /* PMP is not attached. Check whether ATAPI
1004 * AN is configured. If so, notify media
1005 * change.
1007 struct ata_device *dev = ap->link.device;
1009 if ((dev->class == ATA_DEV_ATAPI) &&
1010 (dev->flags & ATA_DFLAG_AN))
1011 ata_scsi_media_change_notify(dev);
1012 return 0;
1013 } else {
1014 /* PMP is attached but SNTF is not available.
1015 * ATAPI async media change notification is
1016 * not used. The PMP must be reporting PHY
1017 * status change, schedule EH.
1019 ata_port_schedule_eh(ap);
1020 return 1;
1022 } else {
1023 /* PMP is attached and SNTF is available */
1024 struct ata_link *link;
1026 /* check and notify ATAPI AN */
1027 ata_port_for_each_link(link, ap) {
1028 if (!(sntf & (1 << link->pmp)))
1029 continue;
1031 if ((link->device->class == ATA_DEV_ATAPI) &&
1032 (link->device->flags & ATA_DFLAG_AN))
1033 ata_scsi_media_change_notify(link->device);
1036 /* If PMP is reporting that PHY status of some
1037 * downstream ports has changed, schedule EH.
1039 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1040 ata_port_schedule_eh(ap);
1041 return 1;
1044 return 0;
1049 * ata_eh_freeze_port - EH helper to freeze port
1050 * @ap: ATA port to freeze
1052 * Freeze @ap.
1054 * LOCKING:
1055 * None.
1057 void ata_eh_freeze_port(struct ata_port *ap)
1059 unsigned long flags;
1061 if (!ap->ops->error_handler)
1062 return;
1064 spin_lock_irqsave(ap->lock, flags);
1065 __ata_port_freeze(ap);
1066 spin_unlock_irqrestore(ap->lock, flags);
1070 * ata_port_thaw_port - EH helper to thaw port
1071 * @ap: ATA port to thaw
1073 * Thaw frozen port @ap.
1075 * LOCKING:
1076 * None.
1078 void ata_eh_thaw_port(struct ata_port *ap)
1080 unsigned long flags;
1082 if (!ap->ops->error_handler)
1083 return;
1085 spin_lock_irqsave(ap->lock, flags);
1087 ap->pflags &= ~ATA_PFLAG_FROZEN;
1089 if (ap->ops->thaw)
1090 ap->ops->thaw(ap);
1092 spin_unlock_irqrestore(ap->lock, flags);
1094 DPRINTK("ata%u port thawed\n", ap->print_id);
1097 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1099 /* nada */
1102 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1104 struct ata_port *ap = qc->ap;
1105 struct scsi_cmnd *scmd = qc->scsicmd;
1106 unsigned long flags;
1108 spin_lock_irqsave(ap->lock, flags);
1109 qc->scsidone = ata_eh_scsidone;
1110 __ata_qc_complete(qc);
1111 WARN_ON(ata_tag_valid(qc->tag));
1112 spin_unlock_irqrestore(ap->lock, flags);
1114 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1118 * ata_eh_qc_complete - Complete an active ATA command from EH
1119 * @qc: Command to complete
1121 * Indicate to the mid and upper layers that an ATA command has
1122 * completed. To be used from EH.
1124 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1126 struct scsi_cmnd *scmd = qc->scsicmd;
1127 scmd->retries = scmd->allowed;
1128 __ata_eh_qc_complete(qc);
1132 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1133 * @qc: Command to retry
1135 * Indicate to the mid and upper layers that an ATA command
1136 * should be retried. To be used from EH.
1138 * SCSI midlayer limits the number of retries to scmd->allowed.
1139 * scmd->retries is decremented for commands which get retried
1140 * due to unrelated failures (qc->err_mask is zero).
1142 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1144 struct scsi_cmnd *scmd = qc->scsicmd;
1145 if (!qc->err_mask && scmd->retries)
1146 scmd->retries--;
1147 __ata_eh_qc_complete(qc);
1151 * ata_eh_detach_dev - detach ATA device
1152 * @dev: ATA device to detach
1154 * Detach @dev.
1156 * LOCKING:
1157 * None.
1159 void ata_eh_detach_dev(struct ata_device *dev)
1161 struct ata_link *link = dev->link;
1162 struct ata_port *ap = link->ap;
1163 unsigned long flags;
1165 ata_dev_disable(dev);
1167 spin_lock_irqsave(ap->lock, flags);
1169 dev->flags &= ~ATA_DFLAG_DETACH;
1171 if (ata_scsi_offline_dev(dev)) {
1172 dev->flags |= ATA_DFLAG_DETACHED;
1173 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1176 /* clear per-dev EH actions */
1177 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1178 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1180 spin_unlock_irqrestore(ap->lock, flags);
1184 * ata_eh_about_to_do - about to perform eh_action
1185 * @link: target ATA link
1186 * @dev: target ATA dev for per-dev action (can be NULL)
1187 * @action: action about to be performed
1189 * Called just before performing EH actions to clear related bits
1190 * in @link->eh_info such that eh actions are not unnecessarily
1191 * repeated.
1193 * LOCKING:
1194 * None.
1196 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1197 unsigned int action)
1199 struct ata_port *ap = link->ap;
1200 struct ata_eh_info *ehi = &link->eh_info;
1201 struct ata_eh_context *ehc = &link->eh_context;
1202 unsigned long flags;
1204 spin_lock_irqsave(ap->lock, flags);
1206 ata_eh_clear_action(link, dev, ehi, action);
1208 if (!(ehc->i.flags & ATA_EHI_QUIET))
1209 ap->pflags |= ATA_PFLAG_RECOVERED;
1211 spin_unlock_irqrestore(ap->lock, flags);
1215 * ata_eh_done - EH action complete
1216 * @ap: target ATA port
1217 * @dev: target ATA dev for per-dev action (can be NULL)
1218 * @action: action just completed
1220 * Called right after performing EH actions to clear related bits
1221 * in @link->eh_context.
1223 * LOCKING:
1224 * None.
1226 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1227 unsigned int action)
1229 struct ata_eh_context *ehc = &link->eh_context;
1231 ata_eh_clear_action(link, dev, &ehc->i, action);
1235 * ata_err_string - convert err_mask to descriptive string
1236 * @err_mask: error mask to convert to string
1238 * Convert @err_mask to descriptive string. Errors are
1239 * prioritized according to severity and only the most severe
1240 * error is reported.
1242 * LOCKING:
1243 * None.
1245 * RETURNS:
1246 * Descriptive string for @err_mask
1248 static const char *ata_err_string(unsigned int err_mask)
1250 if (err_mask & AC_ERR_HOST_BUS)
1251 return "host bus error";
1252 if (err_mask & AC_ERR_ATA_BUS)
1253 return "ATA bus error";
1254 if (err_mask & AC_ERR_TIMEOUT)
1255 return "timeout";
1256 if (err_mask & AC_ERR_HSM)
1257 return "HSM violation";
1258 if (err_mask & AC_ERR_SYSTEM)
1259 return "internal error";
1260 if (err_mask & AC_ERR_MEDIA)
1261 return "media error";
1262 if (err_mask & AC_ERR_INVALID)
1263 return "invalid argument";
1264 if (err_mask & AC_ERR_DEV)
1265 return "device error";
1266 return "unknown error";
1270 * ata_read_log_page - read a specific log page
1271 * @dev: target device
1272 * @page: page to read
1273 * @buf: buffer to store read page
1274 * @sectors: number of sectors to read
1276 * Read log page using READ_LOG_EXT command.
1278 * LOCKING:
1279 * Kernel thread context (may sleep).
1281 * RETURNS:
1282 * 0 on success, AC_ERR_* mask otherwise.
1284 static unsigned int ata_read_log_page(struct ata_device *dev,
1285 u8 page, void *buf, unsigned int sectors)
1287 struct ata_taskfile tf;
1288 unsigned int err_mask;
1290 DPRINTK("read log page - page %d\n", page);
1292 ata_tf_init(dev, &tf);
1293 tf.command = ATA_CMD_READ_LOG_EXT;
1294 tf.lbal = page;
1295 tf.nsect = sectors;
1296 tf.hob_nsect = sectors >> 8;
1297 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1298 tf.protocol = ATA_PROT_PIO;
1300 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1301 buf, sectors * ATA_SECT_SIZE, 0);
1303 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1304 return err_mask;
1308 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1309 * @dev: Device to read log page 10h from
1310 * @tag: Resulting tag of the failed command
1311 * @tf: Resulting taskfile registers of the failed command
1313 * Read log page 10h to obtain NCQ error details and clear error
1314 * condition.
1316 * LOCKING:
1317 * Kernel thread context (may sleep).
1319 * RETURNS:
1320 * 0 on success, -errno otherwise.
1322 static int ata_eh_read_log_10h(struct ata_device *dev,
1323 int *tag, struct ata_taskfile *tf)
1325 u8 *buf = dev->link->ap->sector_buf;
1326 unsigned int err_mask;
1327 u8 csum;
1328 int i;
1330 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1331 if (err_mask)
1332 return -EIO;
1334 csum = 0;
1335 for (i = 0; i < ATA_SECT_SIZE; i++)
1336 csum += buf[i];
1337 if (csum)
1338 ata_dev_printk(dev, KERN_WARNING,
1339 "invalid checksum 0x%x on log page 10h\n", csum);
1341 if (buf[0] & 0x80)
1342 return -ENOENT;
1344 *tag = buf[0] & 0x1f;
1346 tf->command = buf[2];
1347 tf->feature = buf[3];
1348 tf->lbal = buf[4];
1349 tf->lbam = buf[5];
1350 tf->lbah = buf[6];
1351 tf->device = buf[7];
1352 tf->hob_lbal = buf[8];
1353 tf->hob_lbam = buf[9];
1354 tf->hob_lbah = buf[10];
1355 tf->nsect = buf[12];
1356 tf->hob_nsect = buf[13];
1358 return 0;
1362 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1363 * @dev: target ATAPI device
1364 * @r_sense_key: out parameter for sense_key
1366 * Perform ATAPI TEST_UNIT_READY.
1368 * LOCKING:
1369 * EH context (may sleep).
1371 * RETURNS:
1372 * 0 on success, AC_ERR_* mask on failure.
1374 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1376 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1377 struct ata_taskfile tf;
1378 unsigned int err_mask;
1380 ata_tf_init(dev, &tf);
1382 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1383 tf.command = ATA_CMD_PACKET;
1384 tf.protocol = ATAPI_PROT_NODATA;
1386 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1387 if (err_mask == AC_ERR_DEV)
1388 *r_sense_key = tf.feature >> 4;
1389 return err_mask;
1393 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1394 * @dev: device to perform REQUEST_SENSE to
1395 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1396 * @dfl_sense_key: default sense key to use
1398 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1399 * SENSE. This function is EH helper.
1401 * LOCKING:
1402 * Kernel thread context (may sleep).
1404 * RETURNS:
1405 * 0 on success, AC_ERR_* mask on failure
1407 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1408 u8 *sense_buf, u8 dfl_sense_key)
1410 u8 cdb[ATAPI_CDB_LEN] =
1411 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1412 struct ata_port *ap = dev->link->ap;
1413 struct ata_taskfile tf;
1415 DPRINTK("ATAPI request sense\n");
1417 /* FIXME: is this needed? */
1418 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1420 /* initialize sense_buf with the error register,
1421 * for the case where they are -not- overwritten
1423 sense_buf[0] = 0x70;
1424 sense_buf[2] = dfl_sense_key;
1426 /* some devices time out if garbage left in tf */
1427 ata_tf_init(dev, &tf);
1429 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1430 tf.command = ATA_CMD_PACKET;
1432 /* is it pointless to prefer PIO for "safety reasons"? */
1433 if (ap->flags & ATA_FLAG_PIO_DMA) {
1434 tf.protocol = ATAPI_PROT_DMA;
1435 tf.feature |= ATAPI_PKT_DMA;
1436 } else {
1437 tf.protocol = ATAPI_PROT_PIO;
1438 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1439 tf.lbah = 0;
1442 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1443 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1447 * ata_eh_analyze_serror - analyze SError for a failed port
1448 * @link: ATA link to analyze SError for
1450 * Analyze SError if available and further determine cause of
1451 * failure.
1453 * LOCKING:
1454 * None.
1456 static void ata_eh_analyze_serror(struct ata_link *link)
1458 struct ata_eh_context *ehc = &link->eh_context;
1459 u32 serror = ehc->i.serror;
1460 unsigned int err_mask = 0, action = 0;
1461 u32 hotplug_mask;
1463 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1464 err_mask |= AC_ERR_ATA_BUS;
1465 action |= ATA_EH_RESET;
1467 if (serror & SERR_PROTOCOL) {
1468 err_mask |= AC_ERR_HSM;
1469 action |= ATA_EH_RESET;
1471 if (serror & SERR_INTERNAL) {
1472 err_mask |= AC_ERR_SYSTEM;
1473 action |= ATA_EH_RESET;
1476 /* Determine whether a hotplug event has occurred. Both
1477 * SError.N/X are considered hotplug events for enabled or
1478 * host links. For disabled PMP links, only N bit is
1479 * considered as X bit is left at 1 for link plugging.
1481 hotplug_mask = 0;
1483 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1484 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1485 else
1486 hotplug_mask = SERR_PHYRDY_CHG;
1488 if (serror & hotplug_mask)
1489 ata_ehi_hotplugged(&ehc->i);
1491 ehc->i.err_mask |= err_mask;
1492 ehc->i.action |= action;
1496 * ata_eh_analyze_ncq_error - analyze NCQ error
1497 * @link: ATA link to analyze NCQ error for
1499 * Read log page 10h, determine the offending qc and acquire
1500 * error status TF. For NCQ device errors, all LLDDs have to do
1501 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1502 * care of the rest.
1504 * LOCKING:
1505 * Kernel thread context (may sleep).
1507 void ata_eh_analyze_ncq_error(struct ata_link *link)
1509 struct ata_port *ap = link->ap;
1510 struct ata_eh_context *ehc = &link->eh_context;
1511 struct ata_device *dev = link->device;
1512 struct ata_queued_cmd *qc;
1513 struct ata_taskfile tf;
1514 int tag, rc;
1516 /* if frozen, we can't do much */
1517 if (ap->pflags & ATA_PFLAG_FROZEN)
1518 return;
1520 /* is it NCQ device error? */
1521 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1522 return;
1524 /* has LLDD analyzed already? */
1525 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1526 qc = __ata_qc_from_tag(ap, tag);
1528 if (!(qc->flags & ATA_QCFLAG_FAILED))
1529 continue;
1531 if (qc->err_mask)
1532 return;
1535 /* okay, this error is ours */
1536 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1537 if (rc) {
1538 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1539 "(errno=%d)\n", rc);
1540 return;
1543 if (!(link->sactive & (1 << tag))) {
1544 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1545 "inactive tag %d\n", tag);
1546 return;
1549 /* we've got the perpetrator, condemn it */
1550 qc = __ata_qc_from_tag(ap, tag);
1551 memcpy(&qc->result_tf, &tf, sizeof(tf));
1552 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1553 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1554 ehc->i.err_mask &= ~AC_ERR_DEV;
1558 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1559 * @qc: qc to analyze
1560 * @tf: Taskfile registers to analyze
1562 * Analyze taskfile of @qc and further determine cause of
1563 * failure. This function also requests ATAPI sense data if
1564 * avaliable.
1566 * LOCKING:
1567 * Kernel thread context (may sleep).
1569 * RETURNS:
1570 * Determined recovery action
1572 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1573 const struct ata_taskfile *tf)
1575 unsigned int tmp, action = 0;
1576 u8 stat = tf->command, err = tf->feature;
1578 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1579 qc->err_mask |= AC_ERR_HSM;
1580 return ATA_EH_RESET;
1583 if (stat & (ATA_ERR | ATA_DF))
1584 qc->err_mask |= AC_ERR_DEV;
1585 else
1586 return 0;
1588 switch (qc->dev->class) {
1589 case ATA_DEV_ATA:
1590 if (err & ATA_ICRC)
1591 qc->err_mask |= AC_ERR_ATA_BUS;
1592 if (err & ATA_UNC)
1593 qc->err_mask |= AC_ERR_MEDIA;
1594 if (err & ATA_IDNF)
1595 qc->err_mask |= AC_ERR_INVALID;
1596 break;
1598 case ATA_DEV_ATAPI:
1599 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1600 tmp = atapi_eh_request_sense(qc->dev,
1601 qc->scsicmd->sense_buffer,
1602 qc->result_tf.feature >> 4);
1603 if (!tmp) {
1604 /* ATA_QCFLAG_SENSE_VALID is used to
1605 * tell atapi_qc_complete() that sense
1606 * data is already valid.
1608 * TODO: interpret sense data and set
1609 * appropriate err_mask.
1611 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1612 } else
1613 qc->err_mask |= tmp;
1617 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1618 action |= ATA_EH_RESET;
1620 return action;
1623 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1624 int *xfer_ok)
1626 int base = 0;
1628 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1629 *xfer_ok = 1;
1631 if (!*xfer_ok)
1632 base = ATA_ECAT_DUBIOUS_NONE;
1634 if (err_mask & AC_ERR_ATA_BUS)
1635 return base + ATA_ECAT_ATA_BUS;
1637 if (err_mask & AC_ERR_TIMEOUT)
1638 return base + ATA_ECAT_TOUT_HSM;
1640 if (eflags & ATA_EFLAG_IS_IO) {
1641 if (err_mask & AC_ERR_HSM)
1642 return base + ATA_ECAT_TOUT_HSM;
1643 if ((err_mask &
1644 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1645 return base + ATA_ECAT_UNK_DEV;
1648 return 0;
1651 struct speed_down_verdict_arg {
1652 u64 since;
1653 int xfer_ok;
1654 int nr_errors[ATA_ECAT_NR];
1657 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1659 struct speed_down_verdict_arg *arg = void_arg;
1660 int cat;
1662 if (ent->timestamp < arg->since)
1663 return -1;
1665 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1666 &arg->xfer_ok);
1667 arg->nr_errors[cat]++;
1669 return 0;
1673 * ata_eh_speed_down_verdict - Determine speed down verdict
1674 * @dev: Device of interest
1676 * This function examines error ring of @dev and determines
1677 * whether NCQ needs to be turned off, transfer speed should be
1678 * stepped down, or falling back to PIO is necessary.
1680 * ECAT_ATA_BUS : ATA_BUS error for any command
1682 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1683 * IO commands
1685 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1687 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1688 * data transfer hasn't been verified.
1690 * Verdicts are
1692 * NCQ_OFF : Turn off NCQ.
1694 * SPEED_DOWN : Speed down transfer speed but don't fall back
1695 * to PIO.
1697 * FALLBACK_TO_PIO : Fall back to PIO.
1699 * Even if multiple verdicts are returned, only one action is
1700 * taken per error. An action triggered by non-DUBIOUS errors
1701 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1702 * This is to expedite speed down decisions right after device is
1703 * initially configured.
1705 * The followings are speed down rules. #1 and #2 deal with
1706 * DUBIOUS errors.
1708 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1709 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1711 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1712 * occurred during last 5 mins, NCQ_OFF.
1714 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1715 * ocurred during last 5 mins, FALLBACK_TO_PIO
1717 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1718 * during last 10 mins, NCQ_OFF.
1720 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1721 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1723 * LOCKING:
1724 * Inherited from caller.
1726 * RETURNS:
1727 * OR of ATA_EH_SPDN_* flags.
1729 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1731 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1732 u64 j64 = get_jiffies_64();
1733 struct speed_down_verdict_arg arg;
1734 unsigned int verdict = 0;
1736 /* scan past 5 mins of error history */
1737 memset(&arg, 0, sizeof(arg));
1738 arg.since = j64 - min(j64, j5mins);
1739 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1741 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1742 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1743 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1744 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1746 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1747 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1748 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1750 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1751 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1752 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1753 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1755 /* scan past 10 mins of error history */
1756 memset(&arg, 0, sizeof(arg));
1757 arg.since = j64 - min(j64, j10mins);
1758 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1760 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1761 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1762 verdict |= ATA_EH_SPDN_NCQ_OFF;
1764 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1765 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1766 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1767 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1769 return verdict;
1773 * ata_eh_speed_down - record error and speed down if necessary
1774 * @dev: Failed device
1775 * @eflags: mask of ATA_EFLAG_* flags
1776 * @err_mask: err_mask of the error
1778 * Record error and examine error history to determine whether
1779 * adjusting transmission speed is necessary. It also sets
1780 * transmission limits appropriately if such adjustment is
1781 * necessary.
1783 * LOCKING:
1784 * Kernel thread context (may sleep).
1786 * RETURNS:
1787 * Determined recovery action.
1789 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1790 unsigned int eflags, unsigned int err_mask)
1792 struct ata_link *link = ata_dev_phys_link(dev);
1793 int xfer_ok = 0;
1794 unsigned int verdict;
1795 unsigned int action = 0;
1797 /* don't bother if Cat-0 error */
1798 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1799 return 0;
1801 /* record error and determine whether speed down is necessary */
1802 ata_ering_record(&dev->ering, eflags, err_mask);
1803 verdict = ata_eh_speed_down_verdict(dev);
1805 /* turn off NCQ? */
1806 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1807 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1808 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1809 dev->flags |= ATA_DFLAG_NCQ_OFF;
1810 ata_dev_printk(dev, KERN_WARNING,
1811 "NCQ disabled due to excessive errors\n");
1812 goto done;
1815 /* speed down? */
1816 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1817 /* speed down SATA link speed if possible */
1818 if (sata_down_spd_limit(link) == 0) {
1819 action |= ATA_EH_RESET;
1820 goto done;
1823 /* lower transfer mode */
1824 if (dev->spdn_cnt < 2) {
1825 static const int dma_dnxfer_sel[] =
1826 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1827 static const int pio_dnxfer_sel[] =
1828 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1829 int sel;
1831 if (dev->xfer_shift != ATA_SHIFT_PIO)
1832 sel = dma_dnxfer_sel[dev->spdn_cnt];
1833 else
1834 sel = pio_dnxfer_sel[dev->spdn_cnt];
1836 dev->spdn_cnt++;
1838 if (ata_down_xfermask_limit(dev, sel) == 0) {
1839 action |= ATA_EH_RESET;
1840 goto done;
1845 /* Fall back to PIO? Slowing down to PIO is meaningless for
1846 * SATA ATA devices. Consider it only for PATA and SATAPI.
1848 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1849 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1850 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1851 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1852 dev->spdn_cnt = 0;
1853 action |= ATA_EH_RESET;
1854 goto done;
1858 return 0;
1859 done:
1860 /* device has been slowed down, blow error history */
1861 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1862 ata_ering_clear(&dev->ering);
1863 return action;
1867 * ata_eh_link_autopsy - analyze error and determine recovery action
1868 * @link: host link to perform autopsy on
1870 * Analyze why @link failed and determine which recovery actions
1871 * are needed. This function also sets more detailed AC_ERR_*
1872 * values and fills sense data for ATAPI CHECK SENSE.
1874 * LOCKING:
1875 * Kernel thread context (may sleep).
1877 static void ata_eh_link_autopsy(struct ata_link *link)
1879 struct ata_port *ap = link->ap;
1880 struct ata_eh_context *ehc = &link->eh_context;
1881 struct ata_device *dev;
1882 unsigned int all_err_mask = 0, eflags = 0;
1883 int tag;
1884 u32 serror;
1885 int rc;
1887 DPRINTK("ENTER\n");
1889 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1890 return;
1892 /* obtain and analyze SError */
1893 rc = sata_scr_read(link, SCR_ERROR, &serror);
1894 if (rc == 0) {
1895 ehc->i.serror |= serror;
1896 ata_eh_analyze_serror(link);
1897 } else if (rc != -EOPNOTSUPP) {
1898 /* SError read failed, force reset and probing */
1899 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1900 ehc->i.action |= ATA_EH_RESET;
1901 ehc->i.err_mask |= AC_ERR_OTHER;
1904 /* analyze NCQ failure */
1905 ata_eh_analyze_ncq_error(link);
1907 /* any real error trumps AC_ERR_OTHER */
1908 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1909 ehc->i.err_mask &= ~AC_ERR_OTHER;
1911 all_err_mask |= ehc->i.err_mask;
1913 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1914 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1916 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1917 ata_dev_phys_link(qc->dev) != link)
1918 continue;
1920 /* inherit upper level err_mask */
1921 qc->err_mask |= ehc->i.err_mask;
1923 /* analyze TF */
1924 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1926 /* DEV errors are probably spurious in case of ATA_BUS error */
1927 if (qc->err_mask & AC_ERR_ATA_BUS)
1928 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1929 AC_ERR_INVALID);
1931 /* any real error trumps unknown error */
1932 if (qc->err_mask & ~AC_ERR_OTHER)
1933 qc->err_mask &= ~AC_ERR_OTHER;
1935 /* SENSE_VALID trumps dev/unknown error and revalidation */
1936 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1937 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1939 /* determine whether the command is worth retrying */
1940 if (!(qc->err_mask & AC_ERR_INVALID) &&
1941 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
1942 qc->flags |= ATA_QCFLAG_RETRY;
1944 /* accumulate error info */
1945 ehc->i.dev = qc->dev;
1946 all_err_mask |= qc->err_mask;
1947 if (qc->flags & ATA_QCFLAG_IO)
1948 eflags |= ATA_EFLAG_IS_IO;
1951 /* enforce default EH actions */
1952 if (ap->pflags & ATA_PFLAG_FROZEN ||
1953 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1954 ehc->i.action |= ATA_EH_RESET;
1955 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
1956 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
1957 ehc->i.action |= ATA_EH_REVALIDATE;
1959 /* If we have offending qcs and the associated failed device,
1960 * perform per-dev EH action only on the offending device.
1962 if (ehc->i.dev) {
1963 ehc->i.dev_action[ehc->i.dev->devno] |=
1964 ehc->i.action & ATA_EH_PERDEV_MASK;
1965 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1968 /* propagate timeout to host link */
1969 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
1970 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
1972 /* record error and consider speeding down */
1973 dev = ehc->i.dev;
1974 if (!dev && ((ata_link_max_devices(link) == 1 &&
1975 ata_dev_enabled(link->device))))
1976 dev = link->device;
1978 if (dev) {
1979 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
1980 eflags |= ATA_EFLAG_DUBIOUS_XFER;
1981 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
1984 DPRINTK("EXIT\n");
1988 * ata_eh_autopsy - analyze error and determine recovery action
1989 * @ap: host port to perform autopsy on
1991 * Analyze all links of @ap and determine why they failed and
1992 * which recovery actions are needed.
1994 * LOCKING:
1995 * Kernel thread context (may sleep).
1997 void ata_eh_autopsy(struct ata_port *ap)
1999 struct ata_link *link;
2001 ata_port_for_each_link(link, ap)
2002 ata_eh_link_autopsy(link);
2004 /* Handle the frigging slave link. Autopsy is done similarly
2005 * but actions and flags are transferred over to the master
2006 * link and handled from there.
2008 if (ap->slave_link) {
2009 struct ata_eh_context *mehc = &ap->link.eh_context;
2010 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2012 ata_eh_link_autopsy(ap->slave_link);
2014 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2015 mehc->i.action |= sehc->i.action;
2016 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2017 mehc->i.flags |= sehc->i.flags;
2018 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2021 /* Autopsy of fanout ports can affect host link autopsy.
2022 * Perform host link autopsy last.
2024 if (sata_pmp_attached(ap))
2025 ata_eh_link_autopsy(&ap->link);
2029 * ata_eh_link_report - report error handling to user
2030 * @link: ATA link EH is going on
2032 * Report EH to user.
2034 * LOCKING:
2035 * None.
2037 static void ata_eh_link_report(struct ata_link *link)
2039 struct ata_port *ap = link->ap;
2040 struct ata_eh_context *ehc = &link->eh_context;
2041 const char *frozen, *desc;
2042 char tries_buf[6];
2043 int tag, nr_failed = 0;
2045 if (ehc->i.flags & ATA_EHI_QUIET)
2046 return;
2048 desc = NULL;
2049 if (ehc->i.desc[0] != '\0')
2050 desc = ehc->i.desc;
2052 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2053 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2055 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2056 ata_dev_phys_link(qc->dev) != link ||
2057 ((qc->flags & ATA_QCFLAG_QUIET) &&
2058 qc->err_mask == AC_ERR_DEV))
2059 continue;
2060 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2061 continue;
2063 nr_failed++;
2066 if (!nr_failed && !ehc->i.err_mask)
2067 return;
2069 frozen = "";
2070 if (ap->pflags & ATA_PFLAG_FROZEN)
2071 frozen = " frozen";
2073 memset(tries_buf, 0, sizeof(tries_buf));
2074 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2075 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2076 ap->eh_tries);
2078 if (ehc->i.dev) {
2079 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2080 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2081 ehc->i.err_mask, link->sactive, ehc->i.serror,
2082 ehc->i.action, frozen, tries_buf);
2083 if (desc)
2084 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2085 } else {
2086 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2087 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2088 ehc->i.err_mask, link->sactive, ehc->i.serror,
2089 ehc->i.action, frozen, tries_buf);
2090 if (desc)
2091 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2094 if (ehc->i.serror)
2095 ata_link_printk(link, KERN_ERR,
2096 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2097 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2098 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2099 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2100 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2101 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2102 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2103 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2104 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2105 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2106 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2107 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2108 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2109 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2110 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2111 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2112 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2113 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2115 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2116 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2117 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2118 const u8 *cdb = qc->cdb;
2119 char data_buf[20] = "";
2120 char cdb_buf[70] = "";
2122 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2123 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2124 continue;
2126 if (qc->dma_dir != DMA_NONE) {
2127 static const char *dma_str[] = {
2128 [DMA_BIDIRECTIONAL] = "bidi",
2129 [DMA_TO_DEVICE] = "out",
2130 [DMA_FROM_DEVICE] = "in",
2132 static const char *prot_str[] = {
2133 [ATA_PROT_PIO] = "pio",
2134 [ATA_PROT_DMA] = "dma",
2135 [ATA_PROT_NCQ] = "ncq",
2136 [ATAPI_PROT_PIO] = "pio",
2137 [ATAPI_PROT_DMA] = "dma",
2140 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2141 prot_str[qc->tf.protocol], qc->nbytes,
2142 dma_str[qc->dma_dir]);
2145 if (ata_is_atapi(qc->tf.protocol))
2146 snprintf(cdb_buf, sizeof(cdb_buf),
2147 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2148 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2149 cdb[0], cdb[1], cdb[2], cdb[3],
2150 cdb[4], cdb[5], cdb[6], cdb[7],
2151 cdb[8], cdb[9], cdb[10], cdb[11],
2152 cdb[12], cdb[13], cdb[14], cdb[15]);
2154 ata_dev_printk(qc->dev, KERN_ERR,
2155 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2156 "tag %d%s\n %s"
2157 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2158 "Emask 0x%x (%s)%s\n",
2159 cmd->command, cmd->feature, cmd->nsect,
2160 cmd->lbal, cmd->lbam, cmd->lbah,
2161 cmd->hob_feature, cmd->hob_nsect,
2162 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2163 cmd->device, qc->tag, data_buf, cdb_buf,
2164 res->command, res->feature, res->nsect,
2165 res->lbal, res->lbam, res->lbah,
2166 res->hob_feature, res->hob_nsect,
2167 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2168 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2169 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2171 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2172 ATA_ERR)) {
2173 if (res->command & ATA_BUSY)
2174 ata_dev_printk(qc->dev, KERN_ERR,
2175 "status: { Busy }\n");
2176 else
2177 ata_dev_printk(qc->dev, KERN_ERR,
2178 "status: { %s%s%s%s}\n",
2179 res->command & ATA_DRDY ? "DRDY " : "",
2180 res->command & ATA_DF ? "DF " : "",
2181 res->command & ATA_DRQ ? "DRQ " : "",
2182 res->command & ATA_ERR ? "ERR " : "");
2185 if (cmd->command != ATA_CMD_PACKET &&
2186 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2187 ATA_ABORTED)))
2188 ata_dev_printk(qc->dev, KERN_ERR,
2189 "error: { %s%s%s%s}\n",
2190 res->feature & ATA_ICRC ? "ICRC " : "",
2191 res->feature & ATA_UNC ? "UNC " : "",
2192 res->feature & ATA_IDNF ? "IDNF " : "",
2193 res->feature & ATA_ABORTED ? "ABRT " : "");
2198 * ata_eh_report - report error handling to user
2199 * @ap: ATA port to report EH about
2201 * Report EH to user.
2203 * LOCKING:
2204 * None.
2206 void ata_eh_report(struct ata_port *ap)
2208 struct ata_link *link;
2210 __ata_port_for_each_link(link, ap)
2211 ata_eh_link_report(link);
2214 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2215 unsigned int *classes, unsigned long deadline,
2216 bool clear_classes)
2218 struct ata_device *dev;
2220 if (clear_classes)
2221 ata_link_for_each_dev(dev, link)
2222 classes[dev->devno] = ATA_DEV_UNKNOWN;
2224 return reset(link, classes, deadline);
2227 static int ata_eh_followup_srst_needed(struct ata_link *link,
2228 int rc, const unsigned int *classes)
2230 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2231 return 0;
2232 if (rc == -EAGAIN)
2233 return 1;
2234 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2235 return 1;
2236 return 0;
2239 int ata_eh_reset(struct ata_link *link, int classify,
2240 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2241 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2243 struct ata_port *ap = link->ap;
2244 struct ata_link *slave = ap->slave_link;
2245 struct ata_eh_context *ehc = &link->eh_context;
2246 struct ata_eh_context *sehc = &slave->eh_context;
2247 unsigned int *classes = ehc->classes;
2248 unsigned int lflags = link->flags;
2249 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2250 int max_tries = 0, try = 0;
2251 struct ata_link *failed_link;
2252 struct ata_device *dev;
2253 unsigned long deadline, now;
2254 ata_reset_fn_t reset;
2255 unsigned long flags;
2256 u32 sstatus;
2257 int nr_unknown, rc;
2260 * Prepare to reset
2262 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2263 max_tries++;
2264 if (link->flags & ATA_LFLAG_NO_HRST)
2265 hardreset = NULL;
2266 if (link->flags & ATA_LFLAG_NO_SRST)
2267 softreset = NULL;
2269 now = jiffies;
2270 deadline = ata_deadline(ehc->last_reset, ATA_EH_RESET_COOL_DOWN);
2271 if (time_before(now, deadline))
2272 schedule_timeout_uninterruptible(deadline - now);
2274 spin_lock_irqsave(ap->lock, flags);
2275 ap->pflags |= ATA_PFLAG_RESETTING;
2276 spin_unlock_irqrestore(ap->lock, flags);
2278 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2279 ehc->last_reset = jiffies;
2281 ata_link_for_each_dev(dev, link) {
2282 /* If we issue an SRST then an ATA drive (not ATAPI)
2283 * may change configuration and be in PIO0 timing. If
2284 * we do a hard reset (or are coming from power on)
2285 * this is true for ATA or ATAPI. Until we've set a
2286 * suitable controller mode we should not touch the
2287 * bus as we may be talking too fast.
2289 dev->pio_mode = XFER_PIO_0;
2291 /* If the controller has a pio mode setup function
2292 * then use it to set the chipset to rights. Don't
2293 * touch the DMA setup as that will be dealt with when
2294 * configuring devices.
2296 if (ap->ops->set_piomode)
2297 ap->ops->set_piomode(ap, dev);
2300 /* prefer hardreset */
2301 reset = NULL;
2302 ehc->i.action &= ~ATA_EH_RESET;
2303 if (hardreset) {
2304 reset = hardreset;
2305 ehc->i.action |= ATA_EH_HARDRESET;
2306 } else if (softreset) {
2307 reset = softreset;
2308 ehc->i.action |= ATA_EH_SOFTRESET;
2311 if (prereset) {
2312 unsigned long deadline = ata_deadline(jiffies,
2313 ATA_EH_PRERESET_TIMEOUT);
2315 if (slave) {
2316 sehc->i.action &= ~ATA_EH_RESET;
2317 sehc->i.action |= ehc->i.action;
2320 rc = prereset(link, deadline);
2322 /* If present, do prereset on slave link too. Reset
2323 * is skipped iff both master and slave links report
2324 * -ENOENT or clear ATA_EH_RESET.
2326 if (slave && (rc == 0 || rc == -ENOENT)) {
2327 int tmp;
2329 tmp = prereset(slave, deadline);
2330 if (tmp != -ENOENT)
2331 rc = tmp;
2333 ehc->i.action |= sehc->i.action;
2336 if (rc) {
2337 if (rc == -ENOENT) {
2338 ata_link_printk(link, KERN_DEBUG,
2339 "port disabled. ignoring.\n");
2340 ehc->i.action &= ~ATA_EH_RESET;
2342 ata_link_for_each_dev(dev, link)
2343 classes[dev->devno] = ATA_DEV_NONE;
2345 rc = 0;
2346 } else
2347 ata_link_printk(link, KERN_ERR,
2348 "prereset failed (errno=%d)\n", rc);
2349 goto out;
2352 /* prereset() might have cleared ATA_EH_RESET. If so,
2353 * bang classes and return.
2355 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2356 ata_link_for_each_dev(dev, link)
2357 classes[dev->devno] = ATA_DEV_NONE;
2358 rc = 0;
2359 goto out;
2363 retry:
2365 * Perform reset
2367 ehc->last_reset = jiffies;
2368 if (ata_is_host_link(link))
2369 ata_eh_freeze_port(ap);
2371 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2373 if (reset) {
2374 if (verbose)
2375 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2376 reset == softreset ? "soft" : "hard");
2378 /* mark that this EH session started with reset */
2379 if (reset == hardreset)
2380 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2381 else
2382 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2384 rc = ata_do_reset(link, reset, classes, deadline, true);
2385 if (rc && rc != -EAGAIN) {
2386 failed_link = link;
2387 goto fail;
2390 /* hardreset slave link if existent */
2391 if (slave && reset == hardreset) {
2392 int tmp;
2394 if (verbose)
2395 ata_link_printk(slave, KERN_INFO,
2396 "hard resetting link\n");
2398 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2399 tmp = ata_do_reset(slave, reset, classes, deadline,
2400 false);
2401 switch (tmp) {
2402 case -EAGAIN:
2403 rc = -EAGAIN;
2404 case 0:
2405 break;
2406 default:
2407 failed_link = slave;
2408 rc = tmp;
2409 goto fail;
2413 /* perform follow-up SRST if necessary */
2414 if (reset == hardreset &&
2415 ata_eh_followup_srst_needed(link, rc, classes)) {
2416 reset = softreset;
2418 if (!reset) {
2419 ata_link_printk(link, KERN_ERR,
2420 "follow-up softreset required "
2421 "but no softreset avaliable\n");
2422 failed_link = link;
2423 rc = -EINVAL;
2424 goto fail;
2427 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2428 rc = ata_do_reset(link, reset, classes, deadline, true);
2430 } else {
2431 if (verbose)
2432 ata_link_printk(link, KERN_INFO, "no reset method "
2433 "available, skipping reset\n");
2434 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2435 lflags |= ATA_LFLAG_ASSUME_ATA;
2439 * Post-reset processing
2441 ata_link_for_each_dev(dev, link) {
2442 /* After the reset, the device state is PIO 0 and the
2443 * controller state is undefined. Reset also wakes up
2444 * drives from sleeping mode.
2446 dev->pio_mode = XFER_PIO_0;
2447 dev->flags &= ~ATA_DFLAG_SLEEPING;
2449 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2450 continue;
2452 /* apply class override */
2453 if (lflags & ATA_LFLAG_ASSUME_ATA)
2454 classes[dev->devno] = ATA_DEV_ATA;
2455 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2456 classes[dev->devno] = ATA_DEV_SEMB_UNSUP; /* not yet */
2459 /* record current link speed */
2460 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2461 link->sata_spd = (sstatus >> 4) & 0xf;
2462 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2463 slave->sata_spd = (sstatus >> 4) & 0xf;
2465 /* thaw the port */
2466 if (ata_is_host_link(link))
2467 ata_eh_thaw_port(ap);
2469 /* postreset() should clear hardware SError. Although SError
2470 * is cleared during link resume, clearing SError here is
2471 * necessary as some PHYs raise hotplug events after SRST.
2472 * This introduces race condition where hotplug occurs between
2473 * reset and here. This race is mediated by cross checking
2474 * link onlineness and classification result later.
2476 if (postreset) {
2477 postreset(link, classes);
2478 if (slave)
2479 postreset(slave, classes);
2482 /* clear cached SError */
2483 spin_lock_irqsave(link->ap->lock, flags);
2484 link->eh_info.serror = 0;
2485 if (slave)
2486 slave->eh_info.serror = 0;
2487 spin_unlock_irqrestore(link->ap->lock, flags);
2489 /* Make sure onlineness and classification result correspond.
2490 * Hotplug could have happened during reset and some
2491 * controllers fail to wait while a drive is spinning up after
2492 * being hotplugged causing misdetection. By cross checking
2493 * link onlineness and classification result, those conditions
2494 * can be reliably detected and retried.
2496 nr_unknown = 0;
2497 ata_link_for_each_dev(dev, link) {
2498 /* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
2499 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2500 classes[dev->devno] = ATA_DEV_NONE;
2501 if (ata_phys_link_online(ata_dev_phys_link(dev)))
2502 nr_unknown++;
2506 if (classify && nr_unknown) {
2507 if (try < max_tries) {
2508 ata_link_printk(link, KERN_WARNING, "link online but "
2509 "device misclassified, retrying\n");
2510 failed_link = link;
2511 rc = -EAGAIN;
2512 goto fail;
2514 ata_link_printk(link, KERN_WARNING,
2515 "link online but device misclassified, "
2516 "device detection might fail\n");
2519 /* reset successful, schedule revalidation */
2520 ata_eh_done(link, NULL, ATA_EH_RESET);
2521 if (slave)
2522 ata_eh_done(slave, NULL, ATA_EH_RESET);
2523 ehc->last_reset = jiffies;
2524 ehc->i.action |= ATA_EH_REVALIDATE;
2526 rc = 0;
2527 out:
2528 /* clear hotplug flag */
2529 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2530 if (slave)
2531 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2533 spin_lock_irqsave(ap->lock, flags);
2534 ap->pflags &= ~ATA_PFLAG_RESETTING;
2535 spin_unlock_irqrestore(ap->lock, flags);
2537 return rc;
2539 fail:
2540 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2541 if (!ata_is_host_link(link) &&
2542 sata_scr_read(link, SCR_STATUS, &sstatus))
2543 rc = -ERESTART;
2545 if (rc == -ERESTART || try >= max_tries)
2546 goto out;
2548 now = jiffies;
2549 if (time_before(now, deadline)) {
2550 unsigned long delta = deadline - now;
2552 ata_link_printk(failed_link, KERN_WARNING,
2553 "reset failed (errno=%d), retrying in %u secs\n",
2554 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2556 while (delta)
2557 delta = schedule_timeout_uninterruptible(delta);
2560 if (try == max_tries - 1) {
2561 sata_down_spd_limit(link);
2562 if (slave)
2563 sata_down_spd_limit(slave);
2564 } else if (rc == -EPIPE)
2565 sata_down_spd_limit(failed_link);
2567 if (hardreset)
2568 reset = hardreset;
2569 goto retry;
2572 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2574 struct ata_link *link;
2575 struct ata_device *dev;
2576 unsigned long flags;
2579 * This function can be thought of as an extended version of
2580 * ata_eh_about_to_do() specially crafted to accommodate the
2581 * requirements of ATA_EH_PARK handling. Since the EH thread
2582 * does not leave the do {} while () loop in ata_eh_recover as
2583 * long as the timeout for a park request to *one* device on
2584 * the port has not expired, and since we still want to pick
2585 * up park requests to other devices on the same port or
2586 * timeout updates for the same device, we have to pull
2587 * ATA_EH_PARK actions from eh_info into eh_context.i
2588 * ourselves at the beginning of each pass over the loop.
2590 * Additionally, all write accesses to &ap->park_req_pending
2591 * through INIT_COMPLETION() (see below) or complete_all()
2592 * (see ata_scsi_park_store()) are protected by the host lock.
2593 * As a result we have that park_req_pending.done is zero on
2594 * exit from this function, i.e. when ATA_EH_PARK actions for
2595 * *all* devices on port ap have been pulled into the
2596 * respective eh_context structs. If, and only if,
2597 * park_req_pending.done is non-zero by the time we reach
2598 * wait_for_completion_timeout(), another ATA_EH_PARK action
2599 * has been scheduled for at least one of the devices on port
2600 * ap and we have to cycle over the do {} while () loop in
2601 * ata_eh_recover() again.
2604 spin_lock_irqsave(ap->lock, flags);
2605 INIT_COMPLETION(ap->park_req_pending);
2606 ata_port_for_each_link(link, ap) {
2607 ata_link_for_each_dev(dev, link) {
2608 struct ata_eh_info *ehi = &link->eh_info;
2610 link->eh_context.i.dev_action[dev->devno] |=
2611 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2612 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2615 spin_unlock_irqrestore(ap->lock, flags);
2618 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2620 struct ata_eh_context *ehc = &dev->link->eh_context;
2621 struct ata_taskfile tf;
2622 unsigned int err_mask;
2624 ata_tf_init(dev, &tf);
2625 if (park) {
2626 ehc->unloaded_mask |= 1 << dev->devno;
2627 tf.command = ATA_CMD_IDLEIMMEDIATE;
2628 tf.feature = 0x44;
2629 tf.lbal = 0x4c;
2630 tf.lbam = 0x4e;
2631 tf.lbah = 0x55;
2632 } else {
2633 ehc->unloaded_mask &= ~(1 << dev->devno);
2634 tf.command = ATA_CMD_CHK_POWER;
2637 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2638 tf.protocol |= ATA_PROT_NODATA;
2639 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2640 if (park && (err_mask || tf.lbal != 0xc4)) {
2641 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2642 ehc->unloaded_mask &= ~(1 << dev->devno);
2646 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2647 struct ata_device **r_failed_dev)
2649 struct ata_port *ap = link->ap;
2650 struct ata_eh_context *ehc = &link->eh_context;
2651 struct ata_device *dev;
2652 unsigned int new_mask = 0;
2653 unsigned long flags;
2654 int rc = 0;
2656 DPRINTK("ENTER\n");
2658 /* For PATA drive side cable detection to work, IDENTIFY must
2659 * be done backwards such that PDIAG- is released by the slave
2660 * device before the master device is identified.
2662 ata_link_for_each_dev_reverse(dev, link) {
2663 unsigned int action = ata_eh_dev_action(dev);
2664 unsigned int readid_flags = 0;
2666 if (ehc->i.flags & ATA_EHI_DID_RESET)
2667 readid_flags |= ATA_READID_POSTRESET;
2669 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2670 WARN_ON(dev->class == ATA_DEV_PMP);
2672 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2673 rc = -EIO;
2674 goto err;
2677 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2678 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2679 readid_flags);
2680 if (rc)
2681 goto err;
2683 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2685 /* Configuration may have changed, reconfigure
2686 * transfer mode.
2688 ehc->i.flags |= ATA_EHI_SETMODE;
2690 /* schedule the scsi_rescan_device() here */
2691 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2692 } else if (dev->class == ATA_DEV_UNKNOWN &&
2693 ehc->tries[dev->devno] &&
2694 ata_class_enabled(ehc->classes[dev->devno])) {
2695 dev->class = ehc->classes[dev->devno];
2697 if (dev->class == ATA_DEV_PMP)
2698 rc = sata_pmp_attach(dev);
2699 else
2700 rc = ata_dev_read_id(dev, &dev->class,
2701 readid_flags, dev->id);
2702 switch (rc) {
2703 case 0:
2704 new_mask |= 1 << dev->devno;
2705 break;
2706 case -ENOENT:
2707 /* IDENTIFY was issued to non-existent
2708 * device. No need to reset. Just
2709 * thaw and kill the device.
2711 ata_eh_thaw_port(ap);
2712 dev->class = ATA_DEV_UNKNOWN;
2713 break;
2714 default:
2715 dev->class = ATA_DEV_UNKNOWN;
2716 goto err;
2721 /* PDIAG- should have been released, ask cable type if post-reset */
2722 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2723 if (ap->ops->cable_detect)
2724 ap->cbl = ap->ops->cable_detect(ap);
2725 ata_force_cbl(ap);
2728 /* Configure new devices forward such that user doesn't see
2729 * device detection messages backwards.
2731 ata_link_for_each_dev(dev, link) {
2732 if (!(new_mask & (1 << dev->devno)) ||
2733 dev->class == ATA_DEV_PMP)
2734 continue;
2736 ehc->i.flags |= ATA_EHI_PRINTINFO;
2737 rc = ata_dev_configure(dev);
2738 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
2739 if (rc)
2740 goto err;
2742 spin_lock_irqsave(ap->lock, flags);
2743 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
2744 spin_unlock_irqrestore(ap->lock, flags);
2746 /* new device discovered, configure xfermode */
2747 ehc->i.flags |= ATA_EHI_SETMODE;
2750 return 0;
2752 err:
2753 *r_failed_dev = dev;
2754 DPRINTK("EXIT rc=%d\n", rc);
2755 return rc;
2759 * ata_set_mode - Program timings and issue SET FEATURES - XFER
2760 * @link: link on which timings will be programmed
2761 * @r_failed_dev: out paramter for failed device
2763 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
2764 * ata_set_mode() fails, pointer to the failing device is
2765 * returned in @r_failed_dev.
2767 * LOCKING:
2768 * PCI/etc. bus probe sem.
2770 * RETURNS:
2771 * 0 on success, negative errno otherwise
2773 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
2775 struct ata_port *ap = link->ap;
2776 struct ata_device *dev;
2777 int rc;
2779 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
2780 ata_link_for_each_dev(dev, link) {
2781 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
2782 struct ata_ering_entry *ent;
2784 ent = ata_ering_top(&dev->ering);
2785 if (ent)
2786 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
2790 /* has private set_mode? */
2791 if (ap->ops->set_mode)
2792 rc = ap->ops->set_mode(link, r_failed_dev);
2793 else
2794 rc = ata_do_set_mode(link, r_failed_dev);
2796 /* if transfer mode has changed, set DUBIOUS_XFER on device */
2797 ata_link_for_each_dev(dev, link) {
2798 struct ata_eh_context *ehc = &link->eh_context;
2799 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
2800 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
2802 if (dev->xfer_mode != saved_xfer_mode ||
2803 ata_ncq_enabled(dev) != saved_ncq)
2804 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
2807 return rc;
2811 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
2812 * @dev: ATAPI device to clear UA for
2814 * Resets and other operations can make an ATAPI device raise
2815 * UNIT ATTENTION which causes the next operation to fail. This
2816 * function clears UA.
2818 * LOCKING:
2819 * EH context (may sleep).
2821 * RETURNS:
2822 * 0 on success, -errno on failure.
2824 static int atapi_eh_clear_ua(struct ata_device *dev)
2826 int i;
2828 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
2829 u8 sense_buffer[SCSI_SENSE_BUFFERSIZE];
2830 u8 sense_key = 0;
2831 unsigned int err_mask;
2833 err_mask = atapi_eh_tur(dev, &sense_key);
2834 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
2835 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
2836 "failed (err_mask=0x%x)\n", err_mask);
2837 return -EIO;
2840 if (!err_mask || sense_key != UNIT_ATTENTION)
2841 return 0;
2843 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
2844 if (err_mask) {
2845 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
2846 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
2847 return -EIO;
2851 ata_dev_printk(dev, KERN_WARNING,
2852 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
2854 return 0;
2857 static int ata_link_nr_enabled(struct ata_link *link)
2859 struct ata_device *dev;
2860 int cnt = 0;
2862 ata_link_for_each_dev(dev, link)
2863 if (ata_dev_enabled(dev))
2864 cnt++;
2865 return cnt;
2868 static int ata_link_nr_vacant(struct ata_link *link)
2870 struct ata_device *dev;
2871 int cnt = 0;
2873 ata_link_for_each_dev(dev, link)
2874 if (dev->class == ATA_DEV_UNKNOWN)
2875 cnt++;
2876 return cnt;
2879 static int ata_eh_skip_recovery(struct ata_link *link)
2881 struct ata_port *ap = link->ap;
2882 struct ata_eh_context *ehc = &link->eh_context;
2883 struct ata_device *dev;
2885 /* skip disabled links */
2886 if (link->flags & ATA_LFLAG_DISABLED)
2887 return 1;
2889 /* thaw frozen port and recover failed devices */
2890 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
2891 return 0;
2893 /* reset at least once if reset is requested */
2894 if ((ehc->i.action & ATA_EH_RESET) &&
2895 !(ehc->i.flags & ATA_EHI_DID_RESET))
2896 return 0;
2898 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
2899 ata_link_for_each_dev(dev, link) {
2900 if (dev->class == ATA_DEV_UNKNOWN &&
2901 ehc->classes[dev->devno] != ATA_DEV_NONE)
2902 return 0;
2905 return 1;
2908 static int ata_eh_schedule_probe(struct ata_device *dev)
2910 struct ata_eh_context *ehc = &dev->link->eh_context;
2912 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
2913 (ehc->did_probe_mask & (1 << dev->devno)))
2914 return 0;
2916 ata_eh_detach_dev(dev);
2917 ata_dev_init(dev);
2918 ehc->did_probe_mask |= (1 << dev->devno);
2919 ehc->i.action |= ATA_EH_RESET;
2920 ehc->saved_xfer_mode[dev->devno] = 0;
2921 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
2923 return 1;
2926 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
2928 struct ata_eh_context *ehc = &dev->link->eh_context;
2930 ehc->tries[dev->devno]--;
2932 switch (err) {
2933 case -ENODEV:
2934 /* device missing or wrong IDENTIFY data, schedule probing */
2935 ehc->i.probe_mask |= (1 << dev->devno);
2936 case -EINVAL:
2937 /* give it just one more chance */
2938 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2939 case -EIO:
2940 if (ehc->tries[dev->devno] == 1 && dev->pio_mode > XFER_PIO_0) {
2941 /* This is the last chance, better to slow
2942 * down than lose it.
2944 sata_down_spd_limit(ata_dev_phys_link(dev));
2945 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2949 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2950 /* disable device if it has used up all its chances */
2951 ata_dev_disable(dev);
2953 /* detach if offline */
2954 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2955 ata_eh_detach_dev(dev);
2957 /* schedule probe if necessary */
2958 if (ata_eh_schedule_probe(dev)) {
2959 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2960 memset(ehc->cmd_timeout_idx[dev->devno], 0,
2961 sizeof(ehc->cmd_timeout_idx[dev->devno]));
2964 return 1;
2965 } else {
2966 ehc->i.action |= ATA_EH_RESET;
2967 return 0;
2972 * ata_eh_recover - recover host port after error
2973 * @ap: host port to recover
2974 * @prereset: prereset method (can be NULL)
2975 * @softreset: softreset method (can be NULL)
2976 * @hardreset: hardreset method (can be NULL)
2977 * @postreset: postreset method (can be NULL)
2978 * @r_failed_link: out parameter for failed link
2980 * This is the alpha and omega, eum and yang, heart and soul of
2981 * libata exception handling. On entry, actions required to
2982 * recover each link and hotplug requests are recorded in the
2983 * link's eh_context. This function executes all the operations
2984 * with appropriate retrials and fallbacks to resurrect failed
2985 * devices, detach goners and greet newcomers.
2987 * LOCKING:
2988 * Kernel thread context (may sleep).
2990 * RETURNS:
2991 * 0 on success, -errno on failure.
2993 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2994 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2995 ata_postreset_fn_t postreset,
2996 struct ata_link **r_failed_link)
2998 struct ata_link *link;
2999 struct ata_device *dev;
3000 int nr_failed_devs;
3001 int rc;
3002 unsigned long flags, deadline;
3004 DPRINTK("ENTER\n");
3006 /* prep for recovery */
3007 ata_port_for_each_link(link, ap) {
3008 struct ata_eh_context *ehc = &link->eh_context;
3010 /* re-enable link? */
3011 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3012 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3013 spin_lock_irqsave(ap->lock, flags);
3014 link->flags &= ~ATA_LFLAG_DISABLED;
3015 spin_unlock_irqrestore(ap->lock, flags);
3016 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3019 ata_link_for_each_dev(dev, link) {
3020 if (link->flags & ATA_LFLAG_NO_RETRY)
3021 ehc->tries[dev->devno] = 1;
3022 else
3023 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3025 /* collect port action mask recorded in dev actions */
3026 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3027 ~ATA_EH_PERDEV_MASK;
3028 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3030 /* process hotplug request */
3031 if (dev->flags & ATA_DFLAG_DETACH)
3032 ata_eh_detach_dev(dev);
3034 /* schedule probe if necessary */
3035 if (!ata_dev_enabled(dev))
3036 ata_eh_schedule_probe(dev);
3040 retry:
3041 rc = 0;
3042 nr_failed_devs = 0;
3044 /* if UNLOADING, finish immediately */
3045 if (ap->pflags & ATA_PFLAG_UNLOADING)
3046 goto out;
3048 /* prep for EH */
3049 ata_port_for_each_link(link, ap) {
3050 struct ata_eh_context *ehc = &link->eh_context;
3052 /* skip EH if possible. */
3053 if (ata_eh_skip_recovery(link))
3054 ehc->i.action = 0;
3056 ata_link_for_each_dev(dev, link)
3057 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3060 /* reset */
3061 ata_port_for_each_link(link, ap) {
3062 struct ata_eh_context *ehc = &link->eh_context;
3064 if (!(ehc->i.action & ATA_EH_RESET))
3065 continue;
3067 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3068 prereset, softreset, hardreset, postreset);
3069 if (rc) {
3070 ata_link_printk(link, KERN_ERR,
3071 "reset failed, giving up\n");
3072 goto out;
3076 do {
3077 unsigned long now;
3080 * clears ATA_EH_PARK in eh_info and resets
3081 * ap->park_req_pending
3083 ata_eh_pull_park_action(ap);
3085 deadline = jiffies;
3086 ata_port_for_each_link(link, ap) {
3087 ata_link_for_each_dev(dev, link) {
3088 struct ata_eh_context *ehc = &link->eh_context;
3089 unsigned long tmp;
3091 if (dev->class != ATA_DEV_ATA)
3092 continue;
3093 if (!(ehc->i.dev_action[dev->devno] &
3094 ATA_EH_PARK))
3095 continue;
3096 tmp = dev->unpark_deadline;
3097 if (time_before(deadline, tmp))
3098 deadline = tmp;
3099 else if (time_before_eq(tmp, jiffies))
3100 continue;
3101 if (ehc->unloaded_mask & (1 << dev->devno))
3102 continue;
3104 ata_eh_park_issue_cmd(dev, 1);
3108 now = jiffies;
3109 if (time_before_eq(deadline, now))
3110 break;
3112 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3113 deadline - now);
3114 } while (deadline);
3115 ata_port_for_each_link(link, ap) {
3116 ata_link_for_each_dev(dev, link) {
3117 if (!(link->eh_context.unloaded_mask &
3118 (1 << dev->devno)))
3119 continue;
3121 ata_eh_park_issue_cmd(dev, 0);
3122 ata_eh_done(link, dev, ATA_EH_PARK);
3126 /* the rest */
3127 ata_port_for_each_link(link, ap) {
3128 struct ata_eh_context *ehc = &link->eh_context;
3130 /* revalidate existing devices and attach new ones */
3131 rc = ata_eh_revalidate_and_attach(link, &dev);
3132 if (rc)
3133 goto dev_fail;
3135 /* if PMP got attached, return, pmp EH will take care of it */
3136 if (link->device->class == ATA_DEV_PMP) {
3137 ehc->i.action = 0;
3138 return 0;
3141 /* configure transfer mode if necessary */
3142 if (ehc->i.flags & ATA_EHI_SETMODE) {
3143 rc = ata_set_mode(link, &dev);
3144 if (rc)
3145 goto dev_fail;
3146 ehc->i.flags &= ~ATA_EHI_SETMODE;
3149 /* If reset has been issued, clear UA to avoid
3150 * disrupting the current users of the device.
3152 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3153 ata_link_for_each_dev(dev, link) {
3154 if (dev->class != ATA_DEV_ATAPI)
3155 continue;
3156 rc = atapi_eh_clear_ua(dev);
3157 if (rc)
3158 goto dev_fail;
3162 /* configure link power saving */
3163 if (ehc->i.action & ATA_EH_LPM)
3164 ata_link_for_each_dev(dev, link)
3165 ata_dev_enable_pm(dev, ap->pm_policy);
3167 /* this link is okay now */
3168 ehc->i.flags = 0;
3169 continue;
3171 dev_fail:
3172 nr_failed_devs++;
3173 ata_eh_handle_dev_fail(dev, rc);
3175 if (ap->pflags & ATA_PFLAG_FROZEN) {
3176 /* PMP reset requires working host port.
3177 * Can't retry if it's frozen.
3179 if (sata_pmp_attached(ap))
3180 goto out;
3181 break;
3185 if (nr_failed_devs)
3186 goto retry;
3188 out:
3189 if (rc && r_failed_link)
3190 *r_failed_link = link;
3192 DPRINTK("EXIT, rc=%d\n", rc);
3193 return rc;
3197 * ata_eh_finish - finish up EH
3198 * @ap: host port to finish EH for
3200 * Recovery is complete. Clean up EH states and retry or finish
3201 * failed qcs.
3203 * LOCKING:
3204 * None.
3206 void ata_eh_finish(struct ata_port *ap)
3208 int tag;
3210 /* retry or finish qcs */
3211 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3212 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3214 if (!(qc->flags & ATA_QCFLAG_FAILED))
3215 continue;
3217 if (qc->err_mask) {
3218 /* FIXME: Once EH migration is complete,
3219 * generate sense data in this function,
3220 * considering both err_mask and tf.
3222 if (qc->flags & ATA_QCFLAG_RETRY)
3223 ata_eh_qc_retry(qc);
3224 else
3225 ata_eh_qc_complete(qc);
3226 } else {
3227 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3228 ata_eh_qc_complete(qc);
3229 } else {
3230 /* feed zero TF to sense generation */
3231 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3232 ata_eh_qc_retry(qc);
3237 /* make sure nr_active_links is zero after EH */
3238 WARN_ON(ap->nr_active_links);
3239 ap->nr_active_links = 0;
3243 * ata_do_eh - do standard error handling
3244 * @ap: host port to handle error for
3246 * @prereset: prereset method (can be NULL)
3247 * @softreset: softreset method (can be NULL)
3248 * @hardreset: hardreset method (can be NULL)
3249 * @postreset: postreset method (can be NULL)
3251 * Perform standard error handling sequence.
3253 * LOCKING:
3254 * Kernel thread context (may sleep).
3256 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3257 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3258 ata_postreset_fn_t postreset)
3260 struct ata_device *dev;
3261 int rc;
3263 ata_eh_autopsy(ap);
3264 ata_eh_report(ap);
3266 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3267 NULL);
3268 if (rc) {
3269 ata_link_for_each_dev(dev, &ap->link)
3270 ata_dev_disable(dev);
3273 ata_eh_finish(ap);
3277 * ata_std_error_handler - standard error handler
3278 * @ap: host port to handle error for
3280 * Standard error handler
3282 * LOCKING:
3283 * Kernel thread context (may sleep).
3285 void ata_std_error_handler(struct ata_port *ap)
3287 struct ata_port_operations *ops = ap->ops;
3288 ata_reset_fn_t hardreset = ops->hardreset;
3290 /* ignore built-in hardreset if SCR access is not available */
3291 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3292 hardreset = NULL;
3294 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3297 #ifdef CONFIG_PM
3299 * ata_eh_handle_port_suspend - perform port suspend operation
3300 * @ap: port to suspend
3302 * Suspend @ap.
3304 * LOCKING:
3305 * Kernel thread context (may sleep).
3307 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3309 unsigned long flags;
3310 int rc = 0;
3312 /* are we suspending? */
3313 spin_lock_irqsave(ap->lock, flags);
3314 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3315 ap->pm_mesg.event == PM_EVENT_ON) {
3316 spin_unlock_irqrestore(ap->lock, flags);
3317 return;
3319 spin_unlock_irqrestore(ap->lock, flags);
3321 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3323 /* tell ACPI we're suspending */
3324 rc = ata_acpi_on_suspend(ap);
3325 if (rc)
3326 goto out;
3328 /* suspend */
3329 ata_eh_freeze_port(ap);
3331 if (ap->ops->port_suspend)
3332 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3334 ata_acpi_set_state(ap, PMSG_SUSPEND);
3335 out:
3336 /* report result */
3337 spin_lock_irqsave(ap->lock, flags);
3339 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3340 if (rc == 0)
3341 ap->pflags |= ATA_PFLAG_SUSPENDED;
3342 else if (ap->pflags & ATA_PFLAG_FROZEN)
3343 ata_port_schedule_eh(ap);
3345 if (ap->pm_result) {
3346 *ap->pm_result = rc;
3347 ap->pm_result = NULL;
3350 spin_unlock_irqrestore(ap->lock, flags);
3352 return;
3356 * ata_eh_handle_port_resume - perform port resume operation
3357 * @ap: port to resume
3359 * Resume @ap.
3361 * LOCKING:
3362 * Kernel thread context (may sleep).
3364 static void ata_eh_handle_port_resume(struct ata_port *ap)
3366 unsigned long flags;
3367 int rc = 0;
3369 /* are we resuming? */
3370 spin_lock_irqsave(ap->lock, flags);
3371 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3372 ap->pm_mesg.event != PM_EVENT_ON) {
3373 spin_unlock_irqrestore(ap->lock, flags);
3374 return;
3376 spin_unlock_irqrestore(ap->lock, flags);
3378 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3380 ata_acpi_set_state(ap, PMSG_ON);
3382 if (ap->ops->port_resume)
3383 rc = ap->ops->port_resume(ap);
3385 /* tell ACPI that we're resuming */
3386 ata_acpi_on_resume(ap);
3388 /* report result */
3389 spin_lock_irqsave(ap->lock, flags);
3390 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3391 if (ap->pm_result) {
3392 *ap->pm_result = rc;
3393 ap->pm_result = NULL;
3395 spin_unlock_irqrestore(ap->lock, flags);
3397 #endif /* CONFIG_PM */