ARM: 5989/1: ARM: KGDB: add support for SMP platforms
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / ata / libata-eh.c
blob9f6cfac0f2cce6ac50bf8ca2262326cad5c21201
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/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_dbg.h>
44 #include "../scsi/scsi_transport_api.h"
46 #include <linux/libata.h>
48 #include "libata.h"
50 enum {
51 /* speed down verdicts */
52 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
55 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
59 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
61 /* error categories */
62 ATA_ECAT_NONE = 0,
63 ATA_ECAT_ATA_BUS = 1,
64 ATA_ECAT_TOUT_HSM = 2,
65 ATA_ECAT_UNK_DEV = 3,
66 ATA_ECAT_DUBIOUS_NONE = 4,
67 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
68 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
69 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
70 ATA_ECAT_NR = 8,
72 ATA_EH_CMD_DFL_TIMEOUT = 5000,
74 /* always put at least this amount of time between resets */
75 ATA_EH_RESET_COOL_DOWN = 5000,
77 /* Waiting in ->prereset can never be reliable. It's
78 * sometimes nice to wait there but it can't be depended upon;
79 * otherwise, we wouldn't be resetting. Just give it enough
80 * time for most drives to spin up.
82 ATA_EH_PRERESET_TIMEOUT = 10000,
83 ATA_EH_FASTDRAIN_INTERVAL = 3000,
85 ATA_EH_UA_TRIES = 5,
87 /* probe speed down parameters, see ata_eh_schedule_probe() */
88 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
89 ATA_EH_PROBE_TRIALS = 2,
92 /* The following table determines how we sequence resets. Each entry
93 * represents timeout for that try. The first try can be soft or
94 * hardreset. All others are hardreset if available. In most cases
95 * the first reset w/ 10sec timeout should succeed. Following entries
96 * are mostly for error handling, hotplug and retarded devices.
98 static const unsigned long ata_eh_reset_timeouts[] = {
99 10000, /* most drives spin up by 10sec */
100 10000, /* > 99% working drives spin up before 20sec */
101 35000, /* give > 30 secs of idleness for retarded devices */
102 5000, /* and sweet one last chance */
103 ULONG_MAX, /* > 1 min has elapsed, give up */
106 static const unsigned long ata_eh_identify_timeouts[] = {
107 5000, /* covers > 99% of successes and not too boring on failures */
108 10000, /* combined time till here is enough even for media access */
109 30000, /* for true idiots */
110 ULONG_MAX,
113 static const unsigned long ata_eh_flush_timeouts[] = {
114 15000, /* be generous with flush */
115 15000, /* ditto */
116 30000, /* and even more generous */
117 ULONG_MAX,
120 static const unsigned long ata_eh_other_timeouts[] = {
121 5000, /* same rationale as identify timeout */
122 10000, /* ditto */
123 /* but no merciful 30sec for other commands, it just isn't worth it */
124 ULONG_MAX,
127 struct ata_eh_cmd_timeout_ent {
128 const u8 *commands;
129 const unsigned long *timeouts;
132 /* The following table determines timeouts to use for EH internal
133 * commands. Each table entry is a command class and matches the
134 * commands the entry applies to and the timeout table to use.
136 * On the retry after a command timed out, the next timeout value from
137 * the table is used. If the table doesn't contain further entries,
138 * the last value is used.
140 * ehc->cmd_timeout_idx keeps track of which timeout to use per
141 * command class, so if SET_FEATURES times out on the first try, the
142 * next try will use the second timeout value only for that class.
144 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
145 static const struct ata_eh_cmd_timeout_ent
146 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
147 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
148 .timeouts = ata_eh_identify_timeouts, },
149 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
150 .timeouts = ata_eh_other_timeouts, },
151 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
152 .timeouts = ata_eh_other_timeouts, },
153 { .commands = CMDS(ATA_CMD_SET_FEATURES),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
158 .timeouts = ata_eh_flush_timeouts },
160 #undef CMDS
162 static void __ata_port_freeze(struct ata_port *ap);
163 #ifdef CONFIG_PM
164 static void ata_eh_handle_port_suspend(struct ata_port *ap);
165 static void ata_eh_handle_port_resume(struct ata_port *ap);
166 #else /* CONFIG_PM */
167 static void ata_eh_handle_port_suspend(struct ata_port *ap)
170 static void ata_eh_handle_port_resume(struct ata_port *ap)
172 #endif /* CONFIG_PM */
174 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
175 va_list args)
177 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
178 ATA_EH_DESC_LEN - ehi->desc_len,
179 fmt, args);
183 * __ata_ehi_push_desc - push error description without adding separator
184 * @ehi: target EHI
185 * @fmt: printf format string
187 * Format string according to @fmt and append it to @ehi->desc.
189 * LOCKING:
190 * spin_lock_irqsave(host lock)
192 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
194 va_list args;
196 va_start(args, fmt);
197 __ata_ehi_pushv_desc(ehi, fmt, args);
198 va_end(args);
202 * ata_ehi_push_desc - push error description with separator
203 * @ehi: target EHI
204 * @fmt: printf format string
206 * Format string according to @fmt and append it to @ehi->desc.
207 * If @ehi->desc is not empty, ", " is added in-between.
209 * LOCKING:
210 * spin_lock_irqsave(host lock)
212 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
214 va_list args;
216 if (ehi->desc_len)
217 __ata_ehi_push_desc(ehi, ", ");
219 va_start(args, fmt);
220 __ata_ehi_pushv_desc(ehi, fmt, args);
221 va_end(args);
225 * ata_ehi_clear_desc - clean error description
226 * @ehi: target EHI
228 * Clear @ehi->desc.
230 * LOCKING:
231 * spin_lock_irqsave(host lock)
233 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
235 ehi->desc[0] = '\0';
236 ehi->desc_len = 0;
240 * ata_port_desc - append port description
241 * @ap: target ATA port
242 * @fmt: printf format string
244 * Format string according to @fmt and append it to port
245 * description. If port description is not empty, " " is added
246 * in-between. This function is to be used while initializing
247 * ata_host. The description is printed on host registration.
249 * LOCKING:
250 * None.
252 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
254 va_list args;
256 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
258 if (ap->link.eh_info.desc_len)
259 __ata_ehi_push_desc(&ap->link.eh_info, " ");
261 va_start(args, fmt);
262 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
263 va_end(args);
266 #ifdef CONFIG_PCI
269 * ata_port_pbar_desc - append PCI BAR description
270 * @ap: target ATA port
271 * @bar: target PCI BAR
272 * @offset: offset into PCI BAR
273 * @name: name of the area
275 * If @offset is negative, this function formats a string which
276 * contains the name, address, size and type of the BAR and
277 * appends it to the port description. If @offset is zero or
278 * positive, only name and offsetted address is appended.
280 * LOCKING:
281 * None.
283 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
284 const char *name)
286 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
287 char *type = "";
288 unsigned long long start, len;
290 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
291 type = "m";
292 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
293 type = "i";
295 start = (unsigned long long)pci_resource_start(pdev, bar);
296 len = (unsigned long long)pci_resource_len(pdev, bar);
298 if (offset < 0)
299 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
300 else
301 ata_port_desc(ap, "%s 0x%llx", name,
302 start + (unsigned long long)offset);
305 #endif /* CONFIG_PCI */
307 static int ata_lookup_timeout_table(u8 cmd)
309 int i;
311 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
312 const u8 *cur;
314 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
315 if (*cur == cmd)
316 return i;
319 return -1;
323 * ata_internal_cmd_timeout - determine timeout for an internal command
324 * @dev: target device
325 * @cmd: internal command to be issued
327 * Determine timeout for internal command @cmd for @dev.
329 * LOCKING:
330 * EH context.
332 * RETURNS:
333 * Determined timeout.
335 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
337 struct ata_eh_context *ehc = &dev->link->eh_context;
338 int ent = ata_lookup_timeout_table(cmd);
339 int idx;
341 if (ent < 0)
342 return ATA_EH_CMD_DFL_TIMEOUT;
344 idx = ehc->cmd_timeout_idx[dev->devno][ent];
345 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
349 * ata_internal_cmd_timed_out - notification for internal command timeout
350 * @dev: target device
351 * @cmd: internal command which timed out
353 * Notify EH that internal command @cmd for @dev timed out. This
354 * function should be called only for commands whose timeouts are
355 * determined using ata_internal_cmd_timeout().
357 * LOCKING:
358 * EH context.
360 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
362 struct ata_eh_context *ehc = &dev->link->eh_context;
363 int ent = ata_lookup_timeout_table(cmd);
364 int idx;
366 if (ent < 0)
367 return;
369 idx = ehc->cmd_timeout_idx[dev->devno][ent];
370 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
371 ehc->cmd_timeout_idx[dev->devno][ent]++;
374 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
375 unsigned int err_mask)
377 struct ata_ering_entry *ent;
379 WARN_ON(!err_mask);
381 ering->cursor++;
382 ering->cursor %= ATA_ERING_SIZE;
384 ent = &ering->ring[ering->cursor];
385 ent->eflags = eflags;
386 ent->err_mask = err_mask;
387 ent->timestamp = get_jiffies_64();
390 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
392 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
394 if (ent->err_mask)
395 return ent;
396 return NULL;
399 static void ata_ering_clear(struct ata_ering *ering)
401 memset(ering, 0, sizeof(*ering));
404 static int ata_ering_map(struct ata_ering *ering,
405 int (*map_fn)(struct ata_ering_entry *, void *),
406 void *arg)
408 int idx, rc = 0;
409 struct ata_ering_entry *ent;
411 idx = ering->cursor;
412 do {
413 ent = &ering->ring[idx];
414 if (!ent->err_mask)
415 break;
416 rc = map_fn(ent, arg);
417 if (rc)
418 break;
419 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
420 } while (idx != ering->cursor);
422 return rc;
425 static unsigned int ata_eh_dev_action(struct ata_device *dev)
427 struct ata_eh_context *ehc = &dev->link->eh_context;
429 return ehc->i.action | ehc->i.dev_action[dev->devno];
432 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
433 struct ata_eh_info *ehi, unsigned int action)
435 struct ata_device *tdev;
437 if (!dev) {
438 ehi->action &= ~action;
439 ata_for_each_dev(tdev, link, ALL)
440 ehi->dev_action[tdev->devno] &= ~action;
441 } else {
442 /* doesn't make sense for port-wide EH actions */
443 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
445 /* break ehi->action into ehi->dev_action */
446 if (ehi->action & action) {
447 ata_for_each_dev(tdev, link, ALL)
448 ehi->dev_action[tdev->devno] |=
449 ehi->action & action;
450 ehi->action &= ~action;
453 /* turn off the specified per-dev action */
454 ehi->dev_action[dev->devno] &= ~action;
459 * ata_scsi_timed_out - SCSI layer time out callback
460 * @cmd: timed out SCSI command
462 * Handles SCSI layer timeout. We race with normal completion of
463 * the qc for @cmd. If the qc is already gone, we lose and let
464 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
465 * timed out and EH should be invoked. Prevent ata_qc_complete()
466 * from finishing it by setting EH_SCHEDULED and return
467 * EH_NOT_HANDLED.
469 * TODO: kill this function once old EH is gone.
471 * LOCKING:
472 * Called from timer context
474 * RETURNS:
475 * EH_HANDLED or EH_NOT_HANDLED
477 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
479 struct Scsi_Host *host = cmd->device->host;
480 struct ata_port *ap = ata_shost_to_port(host);
481 unsigned long flags;
482 struct ata_queued_cmd *qc;
483 enum blk_eh_timer_return ret;
485 DPRINTK("ENTER\n");
487 if (ap->ops->error_handler) {
488 ret = BLK_EH_NOT_HANDLED;
489 goto out;
492 ret = BLK_EH_HANDLED;
493 spin_lock_irqsave(ap->lock, flags);
494 qc = ata_qc_from_tag(ap, ap->link.active_tag);
495 if (qc) {
496 WARN_ON(qc->scsicmd != cmd);
497 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
498 qc->err_mask |= AC_ERR_TIMEOUT;
499 ret = BLK_EH_NOT_HANDLED;
501 spin_unlock_irqrestore(ap->lock, flags);
503 out:
504 DPRINTK("EXIT, ret=%d\n", ret);
505 return ret;
508 static void ata_eh_unload(struct ata_port *ap)
510 struct ata_link *link;
511 struct ata_device *dev;
512 unsigned long flags;
514 /* Restore SControl IPM and SPD for the next driver and
515 * disable attached devices.
517 ata_for_each_link(link, ap, PMP_FIRST) {
518 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
519 ata_for_each_dev(dev, link, ALL)
520 ata_dev_disable(dev);
523 /* freeze and set UNLOADED */
524 spin_lock_irqsave(ap->lock, flags);
526 ata_port_freeze(ap); /* won't be thawed */
527 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
528 ap->pflags |= ATA_PFLAG_UNLOADED;
530 spin_unlock_irqrestore(ap->lock, flags);
534 * ata_scsi_error - SCSI layer error handler callback
535 * @host: SCSI host on which error occurred
537 * Handles SCSI-layer-thrown error events.
539 * LOCKING:
540 * Inherited from SCSI layer (none, can sleep)
542 * RETURNS:
543 * Zero.
545 void ata_scsi_error(struct Scsi_Host *host)
547 struct ata_port *ap = ata_shost_to_port(host);
548 int i;
549 unsigned long flags;
551 DPRINTK("ENTER\n");
553 /* synchronize with port task */
554 ata_port_flush_task(ap);
556 /* synchronize with host lock and sort out timeouts */
558 /* For new EH, all qcs are finished in one of three ways -
559 * normal completion, error completion, and SCSI timeout.
560 * Both completions can race against SCSI timeout. When normal
561 * completion wins, the qc never reaches EH. When error
562 * completion wins, the qc has ATA_QCFLAG_FAILED set.
564 * When SCSI timeout wins, things are a bit more complex.
565 * Normal or error completion can occur after the timeout but
566 * before this point. In such cases, both types of
567 * completions are honored. A scmd is determined to have
568 * timed out iff its associated qc is active and not failed.
570 if (ap->ops->error_handler) {
571 struct scsi_cmnd *scmd, *tmp;
572 int nr_timedout = 0;
574 spin_lock_irqsave(ap->lock, flags);
576 /* This must occur under the ap->lock as we don't want
577 a polled recovery to race the real interrupt handler
579 The lost_interrupt handler checks for any completed but
580 non-notified command and completes much like an IRQ handler.
582 We then fall into the error recovery code which will treat
583 this as if normal completion won the race */
585 if (ap->ops->lost_interrupt)
586 ap->ops->lost_interrupt(ap);
588 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
589 struct ata_queued_cmd *qc;
591 for (i = 0; i < ATA_MAX_QUEUE; i++) {
592 qc = __ata_qc_from_tag(ap, i);
593 if (qc->flags & ATA_QCFLAG_ACTIVE &&
594 qc->scsicmd == scmd)
595 break;
598 if (i < ATA_MAX_QUEUE) {
599 /* the scmd has an associated qc */
600 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
601 /* which hasn't failed yet, timeout */
602 qc->err_mask |= AC_ERR_TIMEOUT;
603 qc->flags |= ATA_QCFLAG_FAILED;
604 nr_timedout++;
606 } else {
607 /* Normal completion occurred after
608 * SCSI timeout but before this point.
609 * Successfully complete it.
611 scmd->retries = scmd->allowed;
612 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
616 /* If we have timed out qcs. They belong to EH from
617 * this point but the state of the controller is
618 * unknown. Freeze the port to make sure the IRQ
619 * handler doesn't diddle with those qcs. This must
620 * be done atomically w.r.t. setting QCFLAG_FAILED.
622 if (nr_timedout)
623 __ata_port_freeze(ap);
625 spin_unlock_irqrestore(ap->lock, flags);
627 /* initialize eh_tries */
628 ap->eh_tries = ATA_EH_MAX_TRIES;
629 } else
630 spin_unlock_wait(ap->lock);
632 /* If we timed raced normal completion and there is nothing to
633 recover nr_timedout == 0 why exactly are we doing error recovery ? */
635 repeat:
636 /* invoke error handler */
637 if (ap->ops->error_handler) {
638 struct ata_link *link;
640 /* kill fast drain timer */
641 del_timer_sync(&ap->fastdrain_timer);
643 /* process port resume request */
644 ata_eh_handle_port_resume(ap);
646 /* fetch & clear EH info */
647 spin_lock_irqsave(ap->lock, flags);
649 ata_for_each_link(link, ap, HOST_FIRST) {
650 struct ata_eh_context *ehc = &link->eh_context;
651 struct ata_device *dev;
653 memset(&link->eh_context, 0, sizeof(link->eh_context));
654 link->eh_context.i = link->eh_info;
655 memset(&link->eh_info, 0, sizeof(link->eh_info));
657 ata_for_each_dev(dev, link, ENABLED) {
658 int devno = dev->devno;
660 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
661 if (ata_ncq_enabled(dev))
662 ehc->saved_ncq_enabled |= 1 << devno;
666 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
667 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
668 ap->excl_link = NULL; /* don't maintain exclusion over EH */
670 spin_unlock_irqrestore(ap->lock, flags);
672 /* invoke EH, skip if unloading or suspended */
673 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
674 ap->ops->error_handler(ap);
675 else {
676 /* if unloading, commence suicide */
677 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
678 !(ap->pflags & ATA_PFLAG_UNLOADED))
679 ata_eh_unload(ap);
680 ata_eh_finish(ap);
683 /* process port suspend request */
684 ata_eh_handle_port_suspend(ap);
686 /* Exception might have happend after ->error_handler
687 * recovered the port but before this point. Repeat
688 * EH in such case.
690 spin_lock_irqsave(ap->lock, flags);
692 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
693 if (--ap->eh_tries) {
694 spin_unlock_irqrestore(ap->lock, flags);
695 goto repeat;
697 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
698 "tries, giving up\n", ATA_EH_MAX_TRIES);
699 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
702 /* this run is complete, make sure EH info is clear */
703 ata_for_each_link(link, ap, HOST_FIRST)
704 memset(&link->eh_info, 0, sizeof(link->eh_info));
706 /* Clear host_eh_scheduled while holding ap->lock such
707 * that if exception occurs after this point but
708 * before EH completion, SCSI midlayer will
709 * re-initiate EH.
711 host->host_eh_scheduled = 0;
713 spin_unlock_irqrestore(ap->lock, flags);
714 } else {
715 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
716 ap->ops->eng_timeout(ap);
719 /* finish or retry handled scmd's and clean up */
720 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
722 scsi_eh_flush_done_q(&ap->eh_done_q);
724 /* clean up */
725 spin_lock_irqsave(ap->lock, flags);
727 if (ap->pflags & ATA_PFLAG_LOADING)
728 ap->pflags &= ~ATA_PFLAG_LOADING;
729 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
730 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
732 if (ap->pflags & ATA_PFLAG_RECOVERED)
733 ata_port_printk(ap, KERN_INFO, "EH complete\n");
735 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
737 /* tell wait_eh that we're done */
738 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
739 wake_up_all(&ap->eh_wait_q);
741 spin_unlock_irqrestore(ap->lock, flags);
743 DPRINTK("EXIT\n");
747 * ata_port_wait_eh - Wait for the currently pending EH to complete
748 * @ap: Port to wait EH for
750 * Wait until the currently pending EH is complete.
752 * LOCKING:
753 * Kernel thread context (may sleep).
755 void ata_port_wait_eh(struct ata_port *ap)
757 unsigned long flags;
758 DEFINE_WAIT(wait);
760 retry:
761 spin_lock_irqsave(ap->lock, flags);
763 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
764 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
765 spin_unlock_irqrestore(ap->lock, flags);
766 schedule();
767 spin_lock_irqsave(ap->lock, flags);
769 finish_wait(&ap->eh_wait_q, &wait);
771 spin_unlock_irqrestore(ap->lock, flags);
773 /* make sure SCSI EH is complete */
774 if (scsi_host_in_recovery(ap->scsi_host)) {
775 msleep(10);
776 goto retry;
780 static int ata_eh_nr_in_flight(struct ata_port *ap)
782 unsigned int tag;
783 int nr = 0;
785 /* count only non-internal commands */
786 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
787 if (ata_qc_from_tag(ap, tag))
788 nr++;
790 return nr;
793 void ata_eh_fastdrain_timerfn(unsigned long arg)
795 struct ata_port *ap = (void *)arg;
796 unsigned long flags;
797 int cnt;
799 spin_lock_irqsave(ap->lock, flags);
801 cnt = ata_eh_nr_in_flight(ap);
803 /* are we done? */
804 if (!cnt)
805 goto out_unlock;
807 if (cnt == ap->fastdrain_cnt) {
808 unsigned int tag;
810 /* No progress during the last interval, tag all
811 * in-flight qcs as timed out and freeze the port.
813 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
814 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
815 if (qc)
816 qc->err_mask |= AC_ERR_TIMEOUT;
819 ata_port_freeze(ap);
820 } else {
821 /* some qcs have finished, give it another chance */
822 ap->fastdrain_cnt = cnt;
823 ap->fastdrain_timer.expires =
824 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
825 add_timer(&ap->fastdrain_timer);
828 out_unlock:
829 spin_unlock_irqrestore(ap->lock, flags);
833 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
834 * @ap: target ATA port
835 * @fastdrain: activate fast drain
837 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
838 * is non-zero and EH wasn't pending before. Fast drain ensures
839 * that EH kicks in in timely manner.
841 * LOCKING:
842 * spin_lock_irqsave(host lock)
844 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
846 int cnt;
848 /* already scheduled? */
849 if (ap->pflags & ATA_PFLAG_EH_PENDING)
850 return;
852 ap->pflags |= ATA_PFLAG_EH_PENDING;
854 if (!fastdrain)
855 return;
857 /* do we have in-flight qcs? */
858 cnt = ata_eh_nr_in_flight(ap);
859 if (!cnt)
860 return;
862 /* activate fast drain */
863 ap->fastdrain_cnt = cnt;
864 ap->fastdrain_timer.expires =
865 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
866 add_timer(&ap->fastdrain_timer);
870 * ata_qc_schedule_eh - schedule qc for error handling
871 * @qc: command to schedule error handling for
873 * Schedule error handling for @qc. EH will kick in as soon as
874 * other commands are drained.
876 * LOCKING:
877 * spin_lock_irqsave(host lock)
879 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
881 struct ata_port *ap = qc->ap;
883 WARN_ON(!ap->ops->error_handler);
885 qc->flags |= ATA_QCFLAG_FAILED;
886 ata_eh_set_pending(ap, 1);
888 /* The following will fail if timeout has already expired.
889 * ata_scsi_error() takes care of such scmds on EH entry.
890 * Note that ATA_QCFLAG_FAILED is unconditionally set after
891 * this function completes.
893 blk_abort_request(qc->scsicmd->request);
897 * ata_port_schedule_eh - schedule error handling without a qc
898 * @ap: ATA port to schedule EH for
900 * Schedule error handling for @ap. EH will kick in as soon as
901 * all commands are drained.
903 * LOCKING:
904 * spin_lock_irqsave(host lock)
906 void ata_port_schedule_eh(struct ata_port *ap)
908 WARN_ON(!ap->ops->error_handler);
910 if (ap->pflags & ATA_PFLAG_INITIALIZING)
911 return;
913 ata_eh_set_pending(ap, 1);
914 scsi_schedule_eh(ap->scsi_host);
916 DPRINTK("port EH scheduled\n");
919 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
921 int tag, nr_aborted = 0;
923 WARN_ON(!ap->ops->error_handler);
925 /* we're gonna abort all commands, no need for fast drain */
926 ata_eh_set_pending(ap, 0);
928 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
929 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
931 if (qc && (!link || qc->dev->link == link)) {
932 qc->flags |= ATA_QCFLAG_FAILED;
933 ata_qc_complete(qc);
934 nr_aborted++;
938 if (!nr_aborted)
939 ata_port_schedule_eh(ap);
941 return nr_aborted;
945 * ata_link_abort - abort all qc's on the link
946 * @link: ATA link to abort qc's for
948 * Abort all active qc's active on @link and schedule EH.
950 * LOCKING:
951 * spin_lock_irqsave(host lock)
953 * RETURNS:
954 * Number of aborted qc's.
956 int ata_link_abort(struct ata_link *link)
958 return ata_do_link_abort(link->ap, link);
962 * ata_port_abort - abort all qc's on the port
963 * @ap: ATA port to abort qc's for
965 * Abort all active qc's of @ap and schedule EH.
967 * LOCKING:
968 * spin_lock_irqsave(host_set lock)
970 * RETURNS:
971 * Number of aborted qc's.
973 int ata_port_abort(struct ata_port *ap)
975 return ata_do_link_abort(ap, NULL);
979 * __ata_port_freeze - freeze port
980 * @ap: ATA port to freeze
982 * This function is called when HSM violation or some other
983 * condition disrupts normal operation of the port. Frozen port
984 * is not allowed to perform any operation until the port is
985 * thawed, which usually follows a successful reset.
987 * ap->ops->freeze() callback can be used for freezing the port
988 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
989 * port cannot be frozen hardware-wise, the interrupt handler
990 * must ack and clear interrupts unconditionally while the port
991 * is frozen.
993 * LOCKING:
994 * spin_lock_irqsave(host lock)
996 static void __ata_port_freeze(struct ata_port *ap)
998 WARN_ON(!ap->ops->error_handler);
1000 if (ap->ops->freeze)
1001 ap->ops->freeze(ap);
1003 ap->pflags |= ATA_PFLAG_FROZEN;
1005 DPRINTK("ata%u port frozen\n", ap->print_id);
1009 * ata_port_freeze - abort & freeze port
1010 * @ap: ATA port to freeze
1012 * Abort and freeze @ap. The freeze operation must be called
1013 * first, because some hardware requires special operations
1014 * before the taskfile registers are accessible.
1016 * LOCKING:
1017 * spin_lock_irqsave(host lock)
1019 * RETURNS:
1020 * Number of aborted commands.
1022 int ata_port_freeze(struct ata_port *ap)
1024 int nr_aborted;
1026 WARN_ON(!ap->ops->error_handler);
1028 __ata_port_freeze(ap);
1029 nr_aborted = ata_port_abort(ap);
1031 return nr_aborted;
1035 * sata_async_notification - SATA async notification handler
1036 * @ap: ATA port where async notification is received
1038 * Handler to be called when async notification via SDB FIS is
1039 * received. This function schedules EH if necessary.
1041 * LOCKING:
1042 * spin_lock_irqsave(host lock)
1044 * RETURNS:
1045 * 1 if EH is scheduled, 0 otherwise.
1047 int sata_async_notification(struct ata_port *ap)
1049 u32 sntf;
1050 int rc;
1052 if (!(ap->flags & ATA_FLAG_AN))
1053 return 0;
1055 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1056 if (rc == 0)
1057 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1059 if (!sata_pmp_attached(ap) || rc) {
1060 /* PMP is not attached or SNTF is not available */
1061 if (!sata_pmp_attached(ap)) {
1062 /* PMP is not attached. Check whether ATAPI
1063 * AN is configured. If so, notify media
1064 * change.
1066 struct ata_device *dev = ap->link.device;
1068 if ((dev->class == ATA_DEV_ATAPI) &&
1069 (dev->flags & ATA_DFLAG_AN))
1070 ata_scsi_media_change_notify(dev);
1071 return 0;
1072 } else {
1073 /* PMP is attached but SNTF is not available.
1074 * ATAPI async media change notification is
1075 * not used. The PMP must be reporting PHY
1076 * status change, schedule EH.
1078 ata_port_schedule_eh(ap);
1079 return 1;
1081 } else {
1082 /* PMP is attached and SNTF is available */
1083 struct ata_link *link;
1085 /* check and notify ATAPI AN */
1086 ata_for_each_link(link, ap, EDGE) {
1087 if (!(sntf & (1 << link->pmp)))
1088 continue;
1090 if ((link->device->class == ATA_DEV_ATAPI) &&
1091 (link->device->flags & ATA_DFLAG_AN))
1092 ata_scsi_media_change_notify(link->device);
1095 /* If PMP is reporting that PHY status of some
1096 * downstream ports has changed, schedule EH.
1098 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1099 ata_port_schedule_eh(ap);
1100 return 1;
1103 return 0;
1108 * ata_eh_freeze_port - EH helper to freeze port
1109 * @ap: ATA port to freeze
1111 * Freeze @ap.
1113 * LOCKING:
1114 * None.
1116 void ata_eh_freeze_port(struct ata_port *ap)
1118 unsigned long flags;
1120 if (!ap->ops->error_handler)
1121 return;
1123 spin_lock_irqsave(ap->lock, flags);
1124 __ata_port_freeze(ap);
1125 spin_unlock_irqrestore(ap->lock, flags);
1129 * ata_port_thaw_port - EH helper to thaw port
1130 * @ap: ATA port to thaw
1132 * Thaw frozen port @ap.
1134 * LOCKING:
1135 * None.
1137 void ata_eh_thaw_port(struct ata_port *ap)
1139 unsigned long flags;
1141 if (!ap->ops->error_handler)
1142 return;
1144 spin_lock_irqsave(ap->lock, flags);
1146 ap->pflags &= ~ATA_PFLAG_FROZEN;
1148 if (ap->ops->thaw)
1149 ap->ops->thaw(ap);
1151 spin_unlock_irqrestore(ap->lock, flags);
1153 DPRINTK("ata%u port thawed\n", ap->print_id);
1156 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1158 /* nada */
1161 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1163 struct ata_port *ap = qc->ap;
1164 struct scsi_cmnd *scmd = qc->scsicmd;
1165 unsigned long flags;
1167 spin_lock_irqsave(ap->lock, flags);
1168 qc->scsidone = ata_eh_scsidone;
1169 __ata_qc_complete(qc);
1170 WARN_ON(ata_tag_valid(qc->tag));
1171 spin_unlock_irqrestore(ap->lock, flags);
1173 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1177 * ata_eh_qc_complete - Complete an active ATA command from EH
1178 * @qc: Command to complete
1180 * Indicate to the mid and upper layers that an ATA command has
1181 * completed. To be used from EH.
1183 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1185 struct scsi_cmnd *scmd = qc->scsicmd;
1186 scmd->retries = scmd->allowed;
1187 __ata_eh_qc_complete(qc);
1191 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1192 * @qc: Command to retry
1194 * Indicate to the mid and upper layers that an ATA command
1195 * should be retried. To be used from EH.
1197 * SCSI midlayer limits the number of retries to scmd->allowed.
1198 * scmd->retries is decremented for commands which get retried
1199 * due to unrelated failures (qc->err_mask is zero).
1201 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1203 struct scsi_cmnd *scmd = qc->scsicmd;
1204 if (!qc->err_mask && scmd->retries)
1205 scmd->retries--;
1206 __ata_eh_qc_complete(qc);
1210 * ata_dev_disable - disable ATA device
1211 * @dev: ATA device to disable
1213 * Disable @dev.
1215 * Locking:
1216 * EH context.
1218 void ata_dev_disable(struct ata_device *dev)
1220 if (!ata_dev_enabled(dev))
1221 return;
1223 if (ata_msg_drv(dev->link->ap))
1224 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1225 ata_acpi_on_disable(dev);
1226 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1227 dev->class++;
1229 /* From now till the next successful probe, ering is used to
1230 * track probe failures. Clear accumulated device error info.
1232 ata_ering_clear(&dev->ering);
1236 * ata_eh_detach_dev - detach ATA device
1237 * @dev: ATA device to detach
1239 * Detach @dev.
1241 * LOCKING:
1242 * None.
1244 void ata_eh_detach_dev(struct ata_device *dev)
1246 struct ata_link *link = dev->link;
1247 struct ata_port *ap = link->ap;
1248 struct ata_eh_context *ehc = &link->eh_context;
1249 unsigned long flags;
1251 ata_dev_disable(dev);
1253 spin_lock_irqsave(ap->lock, flags);
1255 dev->flags &= ~ATA_DFLAG_DETACH;
1257 if (ata_scsi_offline_dev(dev)) {
1258 dev->flags |= ATA_DFLAG_DETACHED;
1259 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1262 /* clear per-dev EH info */
1263 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1264 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1265 ehc->saved_xfer_mode[dev->devno] = 0;
1266 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1268 spin_unlock_irqrestore(ap->lock, flags);
1272 * ata_eh_about_to_do - about to perform eh_action
1273 * @link: target ATA link
1274 * @dev: target ATA dev for per-dev action (can be NULL)
1275 * @action: action about to be performed
1277 * Called just before performing EH actions to clear related bits
1278 * in @link->eh_info such that eh actions are not unnecessarily
1279 * repeated.
1281 * LOCKING:
1282 * None.
1284 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1285 unsigned int action)
1287 struct ata_port *ap = link->ap;
1288 struct ata_eh_info *ehi = &link->eh_info;
1289 struct ata_eh_context *ehc = &link->eh_context;
1290 unsigned long flags;
1292 spin_lock_irqsave(ap->lock, flags);
1294 ata_eh_clear_action(link, dev, ehi, action);
1296 /* About to take EH action, set RECOVERED. Ignore actions on
1297 * slave links as master will do them again.
1299 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1300 ap->pflags |= ATA_PFLAG_RECOVERED;
1302 spin_unlock_irqrestore(ap->lock, flags);
1306 * ata_eh_done - EH action complete
1307 * @ap: target ATA port
1308 * @dev: target ATA dev for per-dev action (can be NULL)
1309 * @action: action just completed
1311 * Called right after performing EH actions to clear related bits
1312 * in @link->eh_context.
1314 * LOCKING:
1315 * None.
1317 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1318 unsigned int action)
1320 struct ata_eh_context *ehc = &link->eh_context;
1322 ata_eh_clear_action(link, dev, &ehc->i, action);
1326 * ata_err_string - convert err_mask to descriptive string
1327 * @err_mask: error mask to convert to string
1329 * Convert @err_mask to descriptive string. Errors are
1330 * prioritized according to severity and only the most severe
1331 * error is reported.
1333 * LOCKING:
1334 * None.
1336 * RETURNS:
1337 * Descriptive string for @err_mask
1339 static const char *ata_err_string(unsigned int err_mask)
1341 if (err_mask & AC_ERR_HOST_BUS)
1342 return "host bus error";
1343 if (err_mask & AC_ERR_ATA_BUS)
1344 return "ATA bus error";
1345 if (err_mask & AC_ERR_TIMEOUT)
1346 return "timeout";
1347 if (err_mask & AC_ERR_HSM)
1348 return "HSM violation";
1349 if (err_mask & AC_ERR_SYSTEM)
1350 return "internal error";
1351 if (err_mask & AC_ERR_MEDIA)
1352 return "media error";
1353 if (err_mask & AC_ERR_INVALID)
1354 return "invalid argument";
1355 if (err_mask & AC_ERR_DEV)
1356 return "device error";
1357 return "unknown error";
1361 * ata_read_log_page - read a specific log page
1362 * @dev: target device
1363 * @page: page to read
1364 * @buf: buffer to store read page
1365 * @sectors: number of sectors to read
1367 * Read log page using READ_LOG_EXT command.
1369 * LOCKING:
1370 * Kernel thread context (may sleep).
1372 * RETURNS:
1373 * 0 on success, AC_ERR_* mask otherwise.
1375 static unsigned int ata_read_log_page(struct ata_device *dev,
1376 u8 page, void *buf, unsigned int sectors)
1378 struct ata_taskfile tf;
1379 unsigned int err_mask;
1381 DPRINTK("read log page - page %d\n", page);
1383 ata_tf_init(dev, &tf);
1384 tf.command = ATA_CMD_READ_LOG_EXT;
1385 tf.lbal = page;
1386 tf.nsect = sectors;
1387 tf.hob_nsect = sectors >> 8;
1388 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1389 tf.protocol = ATA_PROT_PIO;
1391 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1392 buf, sectors * ATA_SECT_SIZE, 0);
1394 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1395 return err_mask;
1399 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1400 * @dev: Device to read log page 10h from
1401 * @tag: Resulting tag of the failed command
1402 * @tf: Resulting taskfile registers of the failed command
1404 * Read log page 10h to obtain NCQ error details and clear error
1405 * condition.
1407 * LOCKING:
1408 * Kernel thread context (may sleep).
1410 * RETURNS:
1411 * 0 on success, -errno otherwise.
1413 static int ata_eh_read_log_10h(struct ata_device *dev,
1414 int *tag, struct ata_taskfile *tf)
1416 u8 *buf = dev->link->ap->sector_buf;
1417 unsigned int err_mask;
1418 u8 csum;
1419 int i;
1421 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1422 if (err_mask)
1423 return -EIO;
1425 csum = 0;
1426 for (i = 0; i < ATA_SECT_SIZE; i++)
1427 csum += buf[i];
1428 if (csum)
1429 ata_dev_printk(dev, KERN_WARNING,
1430 "invalid checksum 0x%x on log page 10h\n", csum);
1432 if (buf[0] & 0x80)
1433 return -ENOENT;
1435 *tag = buf[0] & 0x1f;
1437 tf->command = buf[2];
1438 tf->feature = buf[3];
1439 tf->lbal = buf[4];
1440 tf->lbam = buf[5];
1441 tf->lbah = buf[6];
1442 tf->device = buf[7];
1443 tf->hob_lbal = buf[8];
1444 tf->hob_lbam = buf[9];
1445 tf->hob_lbah = buf[10];
1446 tf->nsect = buf[12];
1447 tf->hob_nsect = buf[13];
1449 return 0;
1453 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1454 * @dev: target ATAPI device
1455 * @r_sense_key: out parameter for sense_key
1457 * Perform ATAPI TEST_UNIT_READY.
1459 * LOCKING:
1460 * EH context (may sleep).
1462 * RETURNS:
1463 * 0 on success, AC_ERR_* mask on failure.
1465 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1467 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1468 struct ata_taskfile tf;
1469 unsigned int err_mask;
1471 ata_tf_init(dev, &tf);
1473 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1474 tf.command = ATA_CMD_PACKET;
1475 tf.protocol = ATAPI_PROT_NODATA;
1477 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1478 if (err_mask == AC_ERR_DEV)
1479 *r_sense_key = tf.feature >> 4;
1480 return err_mask;
1484 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1485 * @dev: device to perform REQUEST_SENSE to
1486 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1487 * @dfl_sense_key: default sense key to use
1489 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1490 * SENSE. This function is EH helper.
1492 * LOCKING:
1493 * Kernel thread context (may sleep).
1495 * RETURNS:
1496 * 0 on success, AC_ERR_* mask on failure
1498 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1499 u8 *sense_buf, u8 dfl_sense_key)
1501 u8 cdb[ATAPI_CDB_LEN] =
1502 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1503 struct ata_port *ap = dev->link->ap;
1504 struct ata_taskfile tf;
1506 DPRINTK("ATAPI request sense\n");
1508 /* FIXME: is this needed? */
1509 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1511 /* initialize sense_buf with the error register,
1512 * for the case where they are -not- overwritten
1514 sense_buf[0] = 0x70;
1515 sense_buf[2] = dfl_sense_key;
1517 /* some devices time out if garbage left in tf */
1518 ata_tf_init(dev, &tf);
1520 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1521 tf.command = ATA_CMD_PACKET;
1523 /* is it pointless to prefer PIO for "safety reasons"? */
1524 if (ap->flags & ATA_FLAG_PIO_DMA) {
1525 tf.protocol = ATAPI_PROT_DMA;
1526 tf.feature |= ATAPI_PKT_DMA;
1527 } else {
1528 tf.protocol = ATAPI_PROT_PIO;
1529 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1530 tf.lbah = 0;
1533 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1534 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1538 * ata_eh_analyze_serror - analyze SError for a failed port
1539 * @link: ATA link to analyze SError for
1541 * Analyze SError if available and further determine cause of
1542 * failure.
1544 * LOCKING:
1545 * None.
1547 static void ata_eh_analyze_serror(struct ata_link *link)
1549 struct ata_eh_context *ehc = &link->eh_context;
1550 u32 serror = ehc->i.serror;
1551 unsigned int err_mask = 0, action = 0;
1552 u32 hotplug_mask;
1554 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1555 err_mask |= AC_ERR_ATA_BUS;
1556 action |= ATA_EH_RESET;
1558 if (serror & SERR_PROTOCOL) {
1559 err_mask |= AC_ERR_HSM;
1560 action |= ATA_EH_RESET;
1562 if (serror & SERR_INTERNAL) {
1563 err_mask |= AC_ERR_SYSTEM;
1564 action |= ATA_EH_RESET;
1567 /* Determine whether a hotplug event has occurred. Both
1568 * SError.N/X are considered hotplug events for enabled or
1569 * host links. For disabled PMP links, only N bit is
1570 * considered as X bit is left at 1 for link plugging.
1572 hotplug_mask = 0;
1574 if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1575 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1576 else
1577 hotplug_mask = SERR_PHYRDY_CHG;
1579 if (serror & hotplug_mask)
1580 ata_ehi_hotplugged(&ehc->i);
1582 ehc->i.err_mask |= err_mask;
1583 ehc->i.action |= action;
1587 * ata_eh_analyze_ncq_error - analyze NCQ error
1588 * @link: ATA link to analyze NCQ error for
1590 * Read log page 10h, determine the offending qc and acquire
1591 * error status TF. For NCQ device errors, all LLDDs have to do
1592 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1593 * care of the rest.
1595 * LOCKING:
1596 * Kernel thread context (may sleep).
1598 void ata_eh_analyze_ncq_error(struct ata_link *link)
1600 struct ata_port *ap = link->ap;
1601 struct ata_eh_context *ehc = &link->eh_context;
1602 struct ata_device *dev = link->device;
1603 struct ata_queued_cmd *qc;
1604 struct ata_taskfile tf;
1605 int tag, rc;
1607 /* if frozen, we can't do much */
1608 if (ap->pflags & ATA_PFLAG_FROZEN)
1609 return;
1611 /* is it NCQ device error? */
1612 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1613 return;
1615 /* has LLDD analyzed already? */
1616 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1617 qc = __ata_qc_from_tag(ap, tag);
1619 if (!(qc->flags & ATA_QCFLAG_FAILED))
1620 continue;
1622 if (qc->err_mask)
1623 return;
1626 /* okay, this error is ours */
1627 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1628 if (rc) {
1629 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1630 "(errno=%d)\n", rc);
1631 return;
1634 if (!(link->sactive & (1 << tag))) {
1635 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1636 "inactive tag %d\n", tag);
1637 return;
1640 /* we've got the perpetrator, condemn it */
1641 qc = __ata_qc_from_tag(ap, tag);
1642 memcpy(&qc->result_tf, &tf, sizeof(tf));
1643 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1644 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1645 ehc->i.err_mask &= ~AC_ERR_DEV;
1649 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1650 * @qc: qc to analyze
1651 * @tf: Taskfile registers to analyze
1653 * Analyze taskfile of @qc and further determine cause of
1654 * failure. This function also requests ATAPI sense data if
1655 * avaliable.
1657 * LOCKING:
1658 * Kernel thread context (may sleep).
1660 * RETURNS:
1661 * Determined recovery action
1663 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1664 const struct ata_taskfile *tf)
1666 unsigned int tmp, action = 0;
1667 u8 stat = tf->command, err = tf->feature;
1669 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1670 qc->err_mask |= AC_ERR_HSM;
1671 return ATA_EH_RESET;
1674 if (stat & (ATA_ERR | ATA_DF))
1675 qc->err_mask |= AC_ERR_DEV;
1676 else
1677 return 0;
1679 switch (qc->dev->class) {
1680 case ATA_DEV_ATA:
1681 if (err & ATA_ICRC)
1682 qc->err_mask |= AC_ERR_ATA_BUS;
1683 if (err & ATA_UNC)
1684 qc->err_mask |= AC_ERR_MEDIA;
1685 if (err & ATA_IDNF)
1686 qc->err_mask |= AC_ERR_INVALID;
1687 break;
1689 case ATA_DEV_ATAPI:
1690 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1691 tmp = atapi_eh_request_sense(qc->dev,
1692 qc->scsicmd->sense_buffer,
1693 qc->result_tf.feature >> 4);
1694 if (!tmp) {
1695 /* ATA_QCFLAG_SENSE_VALID is used to
1696 * tell atapi_qc_complete() that sense
1697 * data is already valid.
1699 * TODO: interpret sense data and set
1700 * appropriate err_mask.
1702 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1703 } else
1704 qc->err_mask |= tmp;
1708 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1709 action |= ATA_EH_RESET;
1711 return action;
1714 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1715 int *xfer_ok)
1717 int base = 0;
1719 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1720 *xfer_ok = 1;
1722 if (!*xfer_ok)
1723 base = ATA_ECAT_DUBIOUS_NONE;
1725 if (err_mask & AC_ERR_ATA_BUS)
1726 return base + ATA_ECAT_ATA_BUS;
1728 if (err_mask & AC_ERR_TIMEOUT)
1729 return base + ATA_ECAT_TOUT_HSM;
1731 if (eflags & ATA_EFLAG_IS_IO) {
1732 if (err_mask & AC_ERR_HSM)
1733 return base + ATA_ECAT_TOUT_HSM;
1734 if ((err_mask &
1735 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1736 return base + ATA_ECAT_UNK_DEV;
1739 return 0;
1742 struct speed_down_verdict_arg {
1743 u64 since;
1744 int xfer_ok;
1745 int nr_errors[ATA_ECAT_NR];
1748 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1750 struct speed_down_verdict_arg *arg = void_arg;
1751 int cat;
1753 if (ent->timestamp < arg->since)
1754 return -1;
1756 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1757 &arg->xfer_ok);
1758 arg->nr_errors[cat]++;
1760 return 0;
1764 * ata_eh_speed_down_verdict - Determine speed down verdict
1765 * @dev: Device of interest
1767 * This function examines error ring of @dev and determines
1768 * whether NCQ needs to be turned off, transfer speed should be
1769 * stepped down, or falling back to PIO is necessary.
1771 * ECAT_ATA_BUS : ATA_BUS error for any command
1773 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1774 * IO commands
1776 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1778 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1779 * data transfer hasn't been verified.
1781 * Verdicts are
1783 * NCQ_OFF : Turn off NCQ.
1785 * SPEED_DOWN : Speed down transfer speed but don't fall back
1786 * to PIO.
1788 * FALLBACK_TO_PIO : Fall back to PIO.
1790 * Even if multiple verdicts are returned, only one action is
1791 * taken per error. An action triggered by non-DUBIOUS errors
1792 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1793 * This is to expedite speed down decisions right after device is
1794 * initially configured.
1796 * The followings are speed down rules. #1 and #2 deal with
1797 * DUBIOUS errors.
1799 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1800 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1802 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1803 * occurred during last 5 mins, NCQ_OFF.
1805 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1806 * ocurred during last 5 mins, FALLBACK_TO_PIO
1808 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1809 * during last 10 mins, NCQ_OFF.
1811 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1812 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1814 * LOCKING:
1815 * Inherited from caller.
1817 * RETURNS:
1818 * OR of ATA_EH_SPDN_* flags.
1820 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1822 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1823 u64 j64 = get_jiffies_64();
1824 struct speed_down_verdict_arg arg;
1825 unsigned int verdict = 0;
1827 /* scan past 5 mins of error history */
1828 memset(&arg, 0, sizeof(arg));
1829 arg.since = j64 - min(j64, j5mins);
1830 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1832 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1833 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1834 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1835 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1837 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1838 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1839 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1841 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1842 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1843 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1844 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1846 /* scan past 10 mins of error history */
1847 memset(&arg, 0, sizeof(arg));
1848 arg.since = j64 - min(j64, j10mins);
1849 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1851 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1852 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1853 verdict |= ATA_EH_SPDN_NCQ_OFF;
1855 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1856 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1857 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1858 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1860 return verdict;
1864 * ata_eh_speed_down - record error and speed down if necessary
1865 * @dev: Failed device
1866 * @eflags: mask of ATA_EFLAG_* flags
1867 * @err_mask: err_mask of the error
1869 * Record error and examine error history to determine whether
1870 * adjusting transmission speed is necessary. It also sets
1871 * transmission limits appropriately if such adjustment is
1872 * necessary.
1874 * LOCKING:
1875 * Kernel thread context (may sleep).
1877 * RETURNS:
1878 * Determined recovery action.
1880 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1881 unsigned int eflags, unsigned int err_mask)
1883 struct ata_link *link = ata_dev_phys_link(dev);
1884 int xfer_ok = 0;
1885 unsigned int verdict;
1886 unsigned int action = 0;
1888 /* don't bother if Cat-0 error */
1889 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1890 return 0;
1892 /* record error and determine whether speed down is necessary */
1893 ata_ering_record(&dev->ering, eflags, err_mask);
1894 verdict = ata_eh_speed_down_verdict(dev);
1896 /* turn off NCQ? */
1897 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1898 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1899 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1900 dev->flags |= ATA_DFLAG_NCQ_OFF;
1901 ata_dev_printk(dev, KERN_WARNING,
1902 "NCQ disabled due to excessive errors\n");
1903 goto done;
1906 /* speed down? */
1907 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1908 /* speed down SATA link speed if possible */
1909 if (sata_down_spd_limit(link, 0) == 0) {
1910 action |= ATA_EH_RESET;
1911 goto done;
1914 /* lower transfer mode */
1915 if (dev->spdn_cnt < 2) {
1916 static const int dma_dnxfer_sel[] =
1917 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1918 static const int pio_dnxfer_sel[] =
1919 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1920 int sel;
1922 if (dev->xfer_shift != ATA_SHIFT_PIO)
1923 sel = dma_dnxfer_sel[dev->spdn_cnt];
1924 else
1925 sel = pio_dnxfer_sel[dev->spdn_cnt];
1927 dev->spdn_cnt++;
1929 if (ata_down_xfermask_limit(dev, sel) == 0) {
1930 action |= ATA_EH_RESET;
1931 goto done;
1936 /* Fall back to PIO? Slowing down to PIO is meaningless for
1937 * SATA ATA devices. Consider it only for PATA and SATAPI.
1939 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1940 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1941 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1942 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1943 dev->spdn_cnt = 0;
1944 action |= ATA_EH_RESET;
1945 goto done;
1949 return 0;
1950 done:
1951 /* device has been slowed down, blow error history */
1952 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1953 ata_ering_clear(&dev->ering);
1954 return action;
1958 * ata_eh_link_autopsy - analyze error and determine recovery action
1959 * @link: host link to perform autopsy on
1961 * Analyze why @link failed and determine which recovery actions
1962 * are needed. This function also sets more detailed AC_ERR_*
1963 * values and fills sense data for ATAPI CHECK SENSE.
1965 * LOCKING:
1966 * Kernel thread context (may sleep).
1968 static void ata_eh_link_autopsy(struct ata_link *link)
1970 struct ata_port *ap = link->ap;
1971 struct ata_eh_context *ehc = &link->eh_context;
1972 struct ata_device *dev;
1973 unsigned int all_err_mask = 0, eflags = 0;
1974 int tag;
1975 u32 serror;
1976 int rc;
1978 DPRINTK("ENTER\n");
1980 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1981 return;
1983 /* obtain and analyze SError */
1984 rc = sata_scr_read(link, SCR_ERROR, &serror);
1985 if (rc == 0) {
1986 ehc->i.serror |= serror;
1987 ata_eh_analyze_serror(link);
1988 } else if (rc != -EOPNOTSUPP) {
1989 /* SError read failed, force reset and probing */
1990 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1991 ehc->i.action |= ATA_EH_RESET;
1992 ehc->i.err_mask |= AC_ERR_OTHER;
1995 /* analyze NCQ failure */
1996 ata_eh_analyze_ncq_error(link);
1998 /* any real error trumps AC_ERR_OTHER */
1999 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2000 ehc->i.err_mask &= ~AC_ERR_OTHER;
2002 all_err_mask |= ehc->i.err_mask;
2004 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2005 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2007 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2008 ata_dev_phys_link(qc->dev) != link)
2009 continue;
2011 /* inherit upper level err_mask */
2012 qc->err_mask |= ehc->i.err_mask;
2014 /* analyze TF */
2015 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2017 /* DEV errors are probably spurious in case of ATA_BUS error */
2018 if (qc->err_mask & AC_ERR_ATA_BUS)
2019 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2020 AC_ERR_INVALID);
2022 /* any real error trumps unknown error */
2023 if (qc->err_mask & ~AC_ERR_OTHER)
2024 qc->err_mask &= ~AC_ERR_OTHER;
2026 /* SENSE_VALID trumps dev/unknown error and revalidation */
2027 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2028 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2030 /* determine whether the command is worth retrying */
2031 if (qc->flags & ATA_QCFLAG_IO ||
2032 (!(qc->err_mask & AC_ERR_INVALID) &&
2033 qc->err_mask != AC_ERR_DEV))
2034 qc->flags |= ATA_QCFLAG_RETRY;
2036 /* accumulate error info */
2037 ehc->i.dev = qc->dev;
2038 all_err_mask |= qc->err_mask;
2039 if (qc->flags & ATA_QCFLAG_IO)
2040 eflags |= ATA_EFLAG_IS_IO;
2043 /* enforce default EH actions */
2044 if (ap->pflags & ATA_PFLAG_FROZEN ||
2045 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2046 ehc->i.action |= ATA_EH_RESET;
2047 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2048 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2049 ehc->i.action |= ATA_EH_REVALIDATE;
2051 /* If we have offending qcs and the associated failed device,
2052 * perform per-dev EH action only on the offending device.
2054 if (ehc->i.dev) {
2055 ehc->i.dev_action[ehc->i.dev->devno] |=
2056 ehc->i.action & ATA_EH_PERDEV_MASK;
2057 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2060 /* propagate timeout to host link */
2061 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2062 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2064 /* record error and consider speeding down */
2065 dev = ehc->i.dev;
2066 if (!dev && ((ata_link_max_devices(link) == 1 &&
2067 ata_dev_enabled(link->device))))
2068 dev = link->device;
2070 if (dev) {
2071 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2072 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2073 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2076 DPRINTK("EXIT\n");
2080 * ata_eh_autopsy - analyze error and determine recovery action
2081 * @ap: host port to perform autopsy on
2083 * Analyze all links of @ap and determine why they failed and
2084 * which recovery actions are needed.
2086 * LOCKING:
2087 * Kernel thread context (may sleep).
2089 void ata_eh_autopsy(struct ata_port *ap)
2091 struct ata_link *link;
2093 ata_for_each_link(link, ap, EDGE)
2094 ata_eh_link_autopsy(link);
2096 /* Handle the frigging slave link. Autopsy is done similarly
2097 * but actions and flags are transferred over to the master
2098 * link and handled from there.
2100 if (ap->slave_link) {
2101 struct ata_eh_context *mehc = &ap->link.eh_context;
2102 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2104 /* transfer control flags from master to slave */
2105 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2107 /* perform autopsy on the slave link */
2108 ata_eh_link_autopsy(ap->slave_link);
2110 /* transfer actions from slave to master and clear slave */
2111 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2112 mehc->i.action |= sehc->i.action;
2113 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2114 mehc->i.flags |= sehc->i.flags;
2115 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2118 /* Autopsy of fanout ports can affect host link autopsy.
2119 * Perform host link autopsy last.
2121 if (sata_pmp_attached(ap))
2122 ata_eh_link_autopsy(&ap->link);
2126 * ata_get_cmd_descript - get description for ATA command
2127 * @command: ATA command code to get description for
2129 * Return a textual description of the given command, or NULL if the
2130 * command is not known.
2132 * LOCKING:
2133 * None
2135 const char *ata_get_cmd_descript(u8 command)
2137 #ifdef CONFIG_ATA_VERBOSE_ERROR
2138 static const struct
2140 u8 command;
2141 const char *text;
2142 } cmd_descr[] = {
2143 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2144 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2145 { ATA_CMD_STANDBY, "STANDBY" },
2146 { ATA_CMD_IDLE, "IDLE" },
2147 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2148 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2149 { ATA_CMD_NOP, "NOP" },
2150 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2151 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2152 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2153 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2154 { ATA_CMD_SERVICE, "SERVICE" },
2155 { ATA_CMD_READ, "READ DMA" },
2156 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2157 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2158 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2159 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2160 { ATA_CMD_WRITE, "WRITE DMA" },
2161 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2162 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2163 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2164 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2165 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2166 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2167 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2168 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2169 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2170 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2171 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2172 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2173 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2174 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2175 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2176 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2177 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2178 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2179 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2180 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2181 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2182 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2183 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2184 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2185 { ATA_CMD_SLEEP, "SLEEP" },
2186 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2187 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2188 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2189 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2190 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2191 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2192 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2193 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2194 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2195 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2196 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2197 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2198 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2199 { ATA_CMD_PMP_READ, "READ BUFFER" },
2200 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2201 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2202 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2203 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2204 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2205 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2206 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2207 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2208 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2209 { ATA_CMD_SMART, "SMART" },
2210 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2211 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2212 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2213 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2214 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2215 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2216 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2217 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2218 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2219 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2220 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2221 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2222 { ATA_CMD_RESTORE, "RECALIBRATE" },
2223 { 0, NULL } /* terminate list */
2226 unsigned int i;
2227 for (i = 0; cmd_descr[i].text; i++)
2228 if (cmd_descr[i].command == command)
2229 return cmd_descr[i].text;
2230 #endif
2232 return NULL;
2236 * ata_eh_link_report - report error handling to user
2237 * @link: ATA link EH is going on
2239 * Report EH to user.
2241 * LOCKING:
2242 * None.
2244 static void ata_eh_link_report(struct ata_link *link)
2246 struct ata_port *ap = link->ap;
2247 struct ata_eh_context *ehc = &link->eh_context;
2248 const char *frozen, *desc;
2249 char tries_buf[6];
2250 int tag, nr_failed = 0;
2252 if (ehc->i.flags & ATA_EHI_QUIET)
2253 return;
2255 desc = NULL;
2256 if (ehc->i.desc[0] != '\0')
2257 desc = ehc->i.desc;
2259 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2260 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2262 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2263 ata_dev_phys_link(qc->dev) != link ||
2264 ((qc->flags & ATA_QCFLAG_QUIET) &&
2265 qc->err_mask == AC_ERR_DEV))
2266 continue;
2267 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2268 continue;
2270 nr_failed++;
2273 if (!nr_failed && !ehc->i.err_mask)
2274 return;
2276 frozen = "";
2277 if (ap->pflags & ATA_PFLAG_FROZEN)
2278 frozen = " frozen";
2280 memset(tries_buf, 0, sizeof(tries_buf));
2281 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2282 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2283 ap->eh_tries);
2285 if (ehc->i.dev) {
2286 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2287 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2288 ehc->i.err_mask, link->sactive, ehc->i.serror,
2289 ehc->i.action, frozen, tries_buf);
2290 if (desc)
2291 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2292 } else {
2293 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2294 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2295 ehc->i.err_mask, link->sactive, ehc->i.serror,
2296 ehc->i.action, frozen, tries_buf);
2297 if (desc)
2298 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2301 #ifdef CONFIG_ATA_VERBOSE_ERROR
2302 if (ehc->i.serror)
2303 ata_link_printk(link, KERN_ERR,
2304 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2305 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2306 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2307 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2308 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2309 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2310 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2311 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2312 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2313 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2314 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2315 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2316 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2317 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2318 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2319 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2320 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2321 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2322 #endif
2324 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2325 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2326 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2327 const u8 *cdb = qc->cdb;
2328 char data_buf[20] = "";
2329 char cdb_buf[70] = "";
2331 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2332 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2333 continue;
2335 if (qc->dma_dir != DMA_NONE) {
2336 static const char *dma_str[] = {
2337 [DMA_BIDIRECTIONAL] = "bidi",
2338 [DMA_TO_DEVICE] = "out",
2339 [DMA_FROM_DEVICE] = "in",
2341 static const char *prot_str[] = {
2342 [ATA_PROT_PIO] = "pio",
2343 [ATA_PROT_DMA] = "dma",
2344 [ATA_PROT_NCQ] = "ncq",
2345 [ATAPI_PROT_PIO] = "pio",
2346 [ATAPI_PROT_DMA] = "dma",
2349 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2350 prot_str[qc->tf.protocol], qc->nbytes,
2351 dma_str[qc->dma_dir]);
2354 if (ata_is_atapi(qc->tf.protocol)) {
2355 if (qc->scsicmd)
2356 scsi_print_command(qc->scsicmd);
2357 else
2358 snprintf(cdb_buf, sizeof(cdb_buf),
2359 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2360 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2361 cdb[0], cdb[1], cdb[2], cdb[3],
2362 cdb[4], cdb[5], cdb[6], cdb[7],
2363 cdb[8], cdb[9], cdb[10], cdb[11],
2364 cdb[12], cdb[13], cdb[14], cdb[15]);
2365 } else {
2366 const char *descr = ata_get_cmd_descript(cmd->command);
2367 if (descr)
2368 ata_dev_printk(qc->dev, KERN_ERR,
2369 "failed command: %s\n", descr);
2372 ata_dev_printk(qc->dev, KERN_ERR,
2373 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2374 "tag %d%s\n %s"
2375 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2376 "Emask 0x%x (%s)%s\n",
2377 cmd->command, cmd->feature, cmd->nsect,
2378 cmd->lbal, cmd->lbam, cmd->lbah,
2379 cmd->hob_feature, cmd->hob_nsect,
2380 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2381 cmd->device, qc->tag, data_buf, cdb_buf,
2382 res->command, res->feature, res->nsect,
2383 res->lbal, res->lbam, res->lbah,
2384 res->hob_feature, res->hob_nsect,
2385 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2386 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2387 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2389 #ifdef CONFIG_ATA_VERBOSE_ERROR
2390 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2391 ATA_ERR)) {
2392 if (res->command & ATA_BUSY)
2393 ata_dev_printk(qc->dev, KERN_ERR,
2394 "status: { Busy }\n");
2395 else
2396 ata_dev_printk(qc->dev, KERN_ERR,
2397 "status: { %s%s%s%s}\n",
2398 res->command & ATA_DRDY ? "DRDY " : "",
2399 res->command & ATA_DF ? "DF " : "",
2400 res->command & ATA_DRQ ? "DRQ " : "",
2401 res->command & ATA_ERR ? "ERR " : "");
2404 if (cmd->command != ATA_CMD_PACKET &&
2405 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2406 ATA_ABORTED)))
2407 ata_dev_printk(qc->dev, KERN_ERR,
2408 "error: { %s%s%s%s}\n",
2409 res->feature & ATA_ICRC ? "ICRC " : "",
2410 res->feature & ATA_UNC ? "UNC " : "",
2411 res->feature & ATA_IDNF ? "IDNF " : "",
2412 res->feature & ATA_ABORTED ? "ABRT " : "");
2413 #endif
2418 * ata_eh_report - report error handling to user
2419 * @ap: ATA port to report EH about
2421 * Report EH to user.
2423 * LOCKING:
2424 * None.
2426 void ata_eh_report(struct ata_port *ap)
2428 struct ata_link *link;
2430 ata_for_each_link(link, ap, HOST_FIRST)
2431 ata_eh_link_report(link);
2434 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2435 unsigned int *classes, unsigned long deadline,
2436 bool clear_classes)
2438 struct ata_device *dev;
2440 if (clear_classes)
2441 ata_for_each_dev(dev, link, ALL)
2442 classes[dev->devno] = ATA_DEV_UNKNOWN;
2444 return reset(link, classes, deadline);
2447 static int ata_eh_followup_srst_needed(struct ata_link *link,
2448 int rc, const unsigned int *classes)
2450 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2451 return 0;
2452 if (rc == -EAGAIN)
2453 return 1;
2454 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2455 return 1;
2456 return 0;
2459 int ata_eh_reset(struct ata_link *link, int classify,
2460 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2461 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2463 struct ata_port *ap = link->ap;
2464 struct ata_link *slave = ap->slave_link;
2465 struct ata_eh_context *ehc = &link->eh_context;
2466 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2467 unsigned int *classes = ehc->classes;
2468 unsigned int lflags = link->flags;
2469 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2470 int max_tries = 0, try = 0;
2471 struct ata_link *failed_link;
2472 struct ata_device *dev;
2473 unsigned long deadline, now;
2474 ata_reset_fn_t reset;
2475 unsigned long flags;
2476 u32 sstatus;
2477 int nr_unknown, rc;
2480 * Prepare to reset
2482 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2483 max_tries++;
2484 if (link->flags & ATA_LFLAG_NO_HRST)
2485 hardreset = NULL;
2486 if (link->flags & ATA_LFLAG_NO_SRST)
2487 softreset = NULL;
2489 /* make sure each reset attemp is at least COOL_DOWN apart */
2490 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2491 now = jiffies;
2492 WARN_ON(time_after(ehc->last_reset, now));
2493 deadline = ata_deadline(ehc->last_reset,
2494 ATA_EH_RESET_COOL_DOWN);
2495 if (time_before(now, deadline))
2496 schedule_timeout_uninterruptible(deadline - now);
2499 spin_lock_irqsave(ap->lock, flags);
2500 ap->pflags |= ATA_PFLAG_RESETTING;
2501 spin_unlock_irqrestore(ap->lock, flags);
2503 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2505 ata_for_each_dev(dev, link, ALL) {
2506 /* If we issue an SRST then an ATA drive (not ATAPI)
2507 * may change configuration and be in PIO0 timing. If
2508 * we do a hard reset (or are coming from power on)
2509 * this is true for ATA or ATAPI. Until we've set a
2510 * suitable controller mode we should not touch the
2511 * bus as we may be talking too fast.
2513 dev->pio_mode = XFER_PIO_0;
2515 /* If the controller has a pio mode setup function
2516 * then use it to set the chipset to rights. Don't
2517 * touch the DMA setup as that will be dealt with when
2518 * configuring devices.
2520 if (ap->ops->set_piomode)
2521 ap->ops->set_piomode(ap, dev);
2524 /* prefer hardreset */
2525 reset = NULL;
2526 ehc->i.action &= ~ATA_EH_RESET;
2527 if (hardreset) {
2528 reset = hardreset;
2529 ehc->i.action |= ATA_EH_HARDRESET;
2530 } else if (softreset) {
2531 reset = softreset;
2532 ehc->i.action |= ATA_EH_SOFTRESET;
2535 if (prereset) {
2536 unsigned long deadline = ata_deadline(jiffies,
2537 ATA_EH_PRERESET_TIMEOUT);
2539 if (slave) {
2540 sehc->i.action &= ~ATA_EH_RESET;
2541 sehc->i.action |= ehc->i.action;
2544 rc = prereset(link, deadline);
2546 /* If present, do prereset on slave link too. Reset
2547 * is skipped iff both master and slave links report
2548 * -ENOENT or clear ATA_EH_RESET.
2550 if (slave && (rc == 0 || rc == -ENOENT)) {
2551 int tmp;
2553 tmp = prereset(slave, deadline);
2554 if (tmp != -ENOENT)
2555 rc = tmp;
2557 ehc->i.action |= sehc->i.action;
2560 if (rc) {
2561 if (rc == -ENOENT) {
2562 ata_link_printk(link, KERN_DEBUG,
2563 "port disabled. ignoring.\n");
2564 ehc->i.action &= ~ATA_EH_RESET;
2566 ata_for_each_dev(dev, link, ALL)
2567 classes[dev->devno] = ATA_DEV_NONE;
2569 rc = 0;
2570 } else
2571 ata_link_printk(link, KERN_ERR,
2572 "prereset failed (errno=%d)\n", rc);
2573 goto out;
2576 /* prereset() might have cleared ATA_EH_RESET. If so,
2577 * bang classes, thaw and return.
2579 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2580 ata_for_each_dev(dev, link, ALL)
2581 classes[dev->devno] = ATA_DEV_NONE;
2582 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2583 ata_is_host_link(link))
2584 ata_eh_thaw_port(ap);
2585 rc = 0;
2586 goto out;
2590 retry:
2592 * Perform reset
2594 if (ata_is_host_link(link))
2595 ata_eh_freeze_port(ap);
2597 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2599 if (reset) {
2600 if (verbose)
2601 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2602 reset == softreset ? "soft" : "hard");
2604 /* mark that this EH session started with reset */
2605 ehc->last_reset = jiffies;
2606 if (reset == hardreset)
2607 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2608 else
2609 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2611 rc = ata_do_reset(link, reset, classes, deadline, true);
2612 if (rc && rc != -EAGAIN) {
2613 failed_link = link;
2614 goto fail;
2617 /* hardreset slave link if existent */
2618 if (slave && reset == hardreset) {
2619 int tmp;
2621 if (verbose)
2622 ata_link_printk(slave, KERN_INFO,
2623 "hard resetting link\n");
2625 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2626 tmp = ata_do_reset(slave, reset, classes, deadline,
2627 false);
2628 switch (tmp) {
2629 case -EAGAIN:
2630 rc = -EAGAIN;
2631 case 0:
2632 break;
2633 default:
2634 failed_link = slave;
2635 rc = tmp;
2636 goto fail;
2640 /* perform follow-up SRST if necessary */
2641 if (reset == hardreset &&
2642 ata_eh_followup_srst_needed(link, rc, classes)) {
2643 reset = softreset;
2645 if (!reset) {
2646 ata_link_printk(link, KERN_ERR,
2647 "follow-up softreset required "
2648 "but no softreset avaliable\n");
2649 failed_link = link;
2650 rc = -EINVAL;
2651 goto fail;
2654 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2655 rc = ata_do_reset(link, reset, classes, deadline, true);
2656 if (rc) {
2657 failed_link = link;
2658 goto fail;
2661 } else {
2662 if (verbose)
2663 ata_link_printk(link, KERN_INFO, "no reset method "
2664 "available, skipping reset\n");
2665 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2666 lflags |= ATA_LFLAG_ASSUME_ATA;
2670 * Post-reset processing
2672 ata_for_each_dev(dev, link, ALL) {
2673 /* After the reset, the device state is PIO 0 and the
2674 * controller state is undefined. Reset also wakes up
2675 * drives from sleeping mode.
2677 dev->pio_mode = XFER_PIO_0;
2678 dev->flags &= ~ATA_DFLAG_SLEEPING;
2680 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2681 continue;
2683 /* apply class override */
2684 if (lflags & ATA_LFLAG_ASSUME_ATA)
2685 classes[dev->devno] = ATA_DEV_ATA;
2686 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2687 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2690 /* record current link speed */
2691 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2692 link->sata_spd = (sstatus >> 4) & 0xf;
2693 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2694 slave->sata_spd = (sstatus >> 4) & 0xf;
2696 /* thaw the port */
2697 if (ata_is_host_link(link))
2698 ata_eh_thaw_port(ap);
2700 /* postreset() should clear hardware SError. Although SError
2701 * is cleared during link resume, clearing SError here is
2702 * necessary as some PHYs raise hotplug events after SRST.
2703 * This introduces race condition where hotplug occurs between
2704 * reset and here. This race is mediated by cross checking
2705 * link onlineness and classification result later.
2707 if (postreset) {
2708 postreset(link, classes);
2709 if (slave)
2710 postreset(slave, classes);
2714 * Some controllers can't be frozen very well and may set
2715 * spuruious error conditions during reset. Clear accumulated
2716 * error information. As reset is the final recovery action,
2717 * nothing is lost by doing this.
2719 spin_lock_irqsave(link->ap->lock, flags);
2720 memset(&link->eh_info, 0, sizeof(link->eh_info));
2721 if (slave)
2722 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2723 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2724 spin_unlock_irqrestore(link->ap->lock, flags);
2727 * Make sure onlineness and classification result correspond.
2728 * Hotplug could have happened during reset and some
2729 * controllers fail to wait while a drive is spinning up after
2730 * being hotplugged causing misdetection. By cross checking
2731 * link on/offlineness and classification result, those
2732 * conditions can be reliably detected and retried.
2734 nr_unknown = 0;
2735 ata_for_each_dev(dev, link, ALL) {
2736 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2737 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2738 ata_dev_printk(dev, KERN_DEBUG, "link online "
2739 "but device misclassifed\n");
2740 classes[dev->devno] = ATA_DEV_NONE;
2741 nr_unknown++;
2743 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2744 if (ata_class_enabled(classes[dev->devno]))
2745 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2746 "clearing class %d to NONE\n",
2747 classes[dev->devno]);
2748 classes[dev->devno] = ATA_DEV_NONE;
2749 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2750 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2751 "clearing UNKNOWN to NONE\n");
2752 classes[dev->devno] = ATA_DEV_NONE;
2756 if (classify && nr_unknown) {
2757 if (try < max_tries) {
2758 ata_link_printk(link, KERN_WARNING, "link online but "
2759 "%d devices misclassified, retrying\n",
2760 nr_unknown);
2761 failed_link = link;
2762 rc = -EAGAIN;
2763 goto fail;
2765 ata_link_printk(link, KERN_WARNING,
2766 "link online but %d devices misclassified, "
2767 "device detection might fail\n", nr_unknown);
2770 /* reset successful, schedule revalidation */
2771 ata_eh_done(link, NULL, ATA_EH_RESET);
2772 if (slave)
2773 ata_eh_done(slave, NULL, ATA_EH_RESET);
2774 ehc->last_reset = jiffies; /* update to completion time */
2775 ehc->i.action |= ATA_EH_REVALIDATE;
2777 rc = 0;
2778 out:
2779 /* clear hotplug flag */
2780 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2781 if (slave)
2782 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2784 spin_lock_irqsave(ap->lock, flags);
2785 ap->pflags &= ~ATA_PFLAG_RESETTING;
2786 spin_unlock_irqrestore(ap->lock, flags);
2788 return rc;
2790 fail:
2791 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2792 if (!ata_is_host_link(link) &&
2793 sata_scr_read(link, SCR_STATUS, &sstatus))
2794 rc = -ERESTART;
2796 if (rc == -ERESTART || try >= max_tries)
2797 goto out;
2799 now = jiffies;
2800 if (time_before(now, deadline)) {
2801 unsigned long delta = deadline - now;
2803 ata_link_printk(failed_link, KERN_WARNING,
2804 "reset failed (errno=%d), retrying in %u secs\n",
2805 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2807 while (delta)
2808 delta = schedule_timeout_uninterruptible(delta);
2811 if (try == max_tries - 1) {
2812 sata_down_spd_limit(link, 0);
2813 if (slave)
2814 sata_down_spd_limit(slave, 0);
2815 } else if (rc == -EPIPE)
2816 sata_down_spd_limit(failed_link, 0);
2818 if (hardreset)
2819 reset = hardreset;
2820 goto retry;
2823 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2825 struct ata_link *link;
2826 struct ata_device *dev;
2827 unsigned long flags;
2830 * This function can be thought of as an extended version of
2831 * ata_eh_about_to_do() specially crafted to accommodate the
2832 * requirements of ATA_EH_PARK handling. Since the EH thread
2833 * does not leave the do {} while () loop in ata_eh_recover as
2834 * long as the timeout for a park request to *one* device on
2835 * the port has not expired, and since we still want to pick
2836 * up park requests to other devices on the same port or
2837 * timeout updates for the same device, we have to pull
2838 * ATA_EH_PARK actions from eh_info into eh_context.i
2839 * ourselves at the beginning of each pass over the loop.
2841 * Additionally, all write accesses to &ap->park_req_pending
2842 * through INIT_COMPLETION() (see below) or complete_all()
2843 * (see ata_scsi_park_store()) are protected by the host lock.
2844 * As a result we have that park_req_pending.done is zero on
2845 * exit from this function, i.e. when ATA_EH_PARK actions for
2846 * *all* devices on port ap have been pulled into the
2847 * respective eh_context structs. If, and only if,
2848 * park_req_pending.done is non-zero by the time we reach
2849 * wait_for_completion_timeout(), another ATA_EH_PARK action
2850 * has been scheduled for at least one of the devices on port
2851 * ap and we have to cycle over the do {} while () loop in
2852 * ata_eh_recover() again.
2855 spin_lock_irqsave(ap->lock, flags);
2856 INIT_COMPLETION(ap->park_req_pending);
2857 ata_for_each_link(link, ap, EDGE) {
2858 ata_for_each_dev(dev, link, ALL) {
2859 struct ata_eh_info *ehi = &link->eh_info;
2861 link->eh_context.i.dev_action[dev->devno] |=
2862 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2863 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2866 spin_unlock_irqrestore(ap->lock, flags);
2869 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2871 struct ata_eh_context *ehc = &dev->link->eh_context;
2872 struct ata_taskfile tf;
2873 unsigned int err_mask;
2875 ata_tf_init(dev, &tf);
2876 if (park) {
2877 ehc->unloaded_mask |= 1 << dev->devno;
2878 tf.command = ATA_CMD_IDLEIMMEDIATE;
2879 tf.feature = 0x44;
2880 tf.lbal = 0x4c;
2881 tf.lbam = 0x4e;
2882 tf.lbah = 0x55;
2883 } else {
2884 ehc->unloaded_mask &= ~(1 << dev->devno);
2885 tf.command = ATA_CMD_CHK_POWER;
2888 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2889 tf.protocol |= ATA_PROT_NODATA;
2890 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2891 if (park && (err_mask || tf.lbal != 0xc4)) {
2892 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2893 ehc->unloaded_mask &= ~(1 << dev->devno);
2897 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2898 struct ata_device **r_failed_dev)
2900 struct ata_port *ap = link->ap;
2901 struct ata_eh_context *ehc = &link->eh_context;
2902 struct ata_device *dev;
2903 unsigned int new_mask = 0;
2904 unsigned long flags;
2905 int rc = 0;
2907 DPRINTK("ENTER\n");
2909 /* For PATA drive side cable detection to work, IDENTIFY must
2910 * be done backwards such that PDIAG- is released by the slave
2911 * device before the master device is identified.
2913 ata_for_each_dev(dev, link, ALL_REVERSE) {
2914 unsigned int action = ata_eh_dev_action(dev);
2915 unsigned int readid_flags = 0;
2917 if (ehc->i.flags & ATA_EHI_DID_RESET)
2918 readid_flags |= ATA_READID_POSTRESET;
2920 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2921 WARN_ON(dev->class == ATA_DEV_PMP);
2923 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2924 rc = -EIO;
2925 goto err;
2928 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2929 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2930 readid_flags);
2931 if (rc)
2932 goto err;
2934 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2936 /* Configuration may have changed, reconfigure
2937 * transfer mode.
2939 ehc->i.flags |= ATA_EHI_SETMODE;
2941 /* schedule the scsi_rescan_device() here */
2942 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2943 } else if (dev->class == ATA_DEV_UNKNOWN &&
2944 ehc->tries[dev->devno] &&
2945 ata_class_enabled(ehc->classes[dev->devno])) {
2946 /* Temporarily set dev->class, it will be
2947 * permanently set once all configurations are
2948 * complete. This is necessary because new
2949 * device configuration is done in two
2950 * separate loops.
2952 dev->class = ehc->classes[dev->devno];
2954 if (dev->class == ATA_DEV_PMP)
2955 rc = sata_pmp_attach(dev);
2956 else
2957 rc = ata_dev_read_id(dev, &dev->class,
2958 readid_flags, dev->id);
2960 /* read_id might have changed class, store and reset */
2961 ehc->classes[dev->devno] = dev->class;
2962 dev->class = ATA_DEV_UNKNOWN;
2964 switch (rc) {
2965 case 0:
2966 /* clear error info accumulated during probe */
2967 ata_ering_clear(&dev->ering);
2968 new_mask |= 1 << dev->devno;
2969 break;
2970 case -ENOENT:
2971 /* IDENTIFY was issued to non-existent
2972 * device. No need to reset. Just
2973 * thaw and ignore the device.
2975 ata_eh_thaw_port(ap);
2976 break;
2977 default:
2978 goto err;
2983 /* PDIAG- should have been released, ask cable type if post-reset */
2984 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2985 if (ap->ops->cable_detect)
2986 ap->cbl = ap->ops->cable_detect(ap);
2987 ata_force_cbl(ap);
2990 /* Configure new devices forward such that user doesn't see
2991 * device detection messages backwards.
2993 ata_for_each_dev(dev, link, ALL) {
2994 if (!(new_mask & (1 << dev->devno)))
2995 continue;
2997 dev->class = ehc->classes[dev->devno];
2999 if (dev->class == ATA_DEV_PMP)
3000 continue;
3002 ehc->i.flags |= ATA_EHI_PRINTINFO;
3003 rc = ata_dev_configure(dev);
3004 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3005 if (rc) {
3006 dev->class = ATA_DEV_UNKNOWN;
3007 goto err;
3010 spin_lock_irqsave(ap->lock, flags);
3011 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3012 spin_unlock_irqrestore(ap->lock, flags);
3014 /* new device discovered, configure xfermode */
3015 ehc->i.flags |= ATA_EHI_SETMODE;
3018 return 0;
3020 err:
3021 *r_failed_dev = dev;
3022 DPRINTK("EXIT rc=%d\n", rc);
3023 return rc;
3027 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3028 * @link: link on which timings will be programmed
3029 * @r_failed_dev: out parameter for failed device
3031 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3032 * ata_set_mode() fails, pointer to the failing device is
3033 * returned in @r_failed_dev.
3035 * LOCKING:
3036 * PCI/etc. bus probe sem.
3038 * RETURNS:
3039 * 0 on success, negative errno otherwise
3041 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3043 struct ata_port *ap = link->ap;
3044 struct ata_device *dev;
3045 int rc;
3047 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3048 ata_for_each_dev(dev, link, ENABLED) {
3049 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3050 struct ata_ering_entry *ent;
3052 ent = ata_ering_top(&dev->ering);
3053 if (ent)
3054 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3058 /* has private set_mode? */
3059 if (ap->ops->set_mode)
3060 rc = ap->ops->set_mode(link, r_failed_dev);
3061 else
3062 rc = ata_do_set_mode(link, r_failed_dev);
3064 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3065 ata_for_each_dev(dev, link, ENABLED) {
3066 struct ata_eh_context *ehc = &link->eh_context;
3067 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3068 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3070 if (dev->xfer_mode != saved_xfer_mode ||
3071 ata_ncq_enabled(dev) != saved_ncq)
3072 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3075 return rc;
3079 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3080 * @dev: ATAPI device to clear UA for
3082 * Resets and other operations can make an ATAPI device raise
3083 * UNIT ATTENTION which causes the next operation to fail. This
3084 * function clears UA.
3086 * LOCKING:
3087 * EH context (may sleep).
3089 * RETURNS:
3090 * 0 on success, -errno on failure.
3092 static int atapi_eh_clear_ua(struct ata_device *dev)
3094 int i;
3096 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3097 u8 *sense_buffer = dev->link->ap->sector_buf;
3098 u8 sense_key = 0;
3099 unsigned int err_mask;
3101 err_mask = atapi_eh_tur(dev, &sense_key);
3102 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3103 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3104 "failed (err_mask=0x%x)\n", err_mask);
3105 return -EIO;
3108 if (!err_mask || sense_key != UNIT_ATTENTION)
3109 return 0;
3111 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3112 if (err_mask) {
3113 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3114 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3115 return -EIO;
3119 ata_dev_printk(dev, KERN_WARNING,
3120 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3122 return 0;
3126 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3127 * @dev: ATA device which may need FLUSH retry
3129 * If @dev failed FLUSH, it needs to be reported upper layer
3130 * immediately as it means that @dev failed to remap and already
3131 * lost at least a sector and further FLUSH retrials won't make
3132 * any difference to the lost sector. However, if FLUSH failed
3133 * for other reasons, for example transmission error, FLUSH needs
3134 * to be retried.
3136 * This function determines whether FLUSH failure retry is
3137 * necessary and performs it if so.
3139 * RETURNS:
3140 * 0 if EH can continue, -errno if EH needs to be repeated.
3142 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3144 struct ata_link *link = dev->link;
3145 struct ata_port *ap = link->ap;
3146 struct ata_queued_cmd *qc;
3147 struct ata_taskfile tf;
3148 unsigned int err_mask;
3149 int rc = 0;
3151 /* did flush fail for this device? */
3152 if (!ata_tag_valid(link->active_tag))
3153 return 0;
3155 qc = __ata_qc_from_tag(ap, link->active_tag);
3156 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3157 qc->tf.command != ATA_CMD_FLUSH))
3158 return 0;
3160 /* if the device failed it, it should be reported to upper layers */
3161 if (qc->err_mask & AC_ERR_DEV)
3162 return 0;
3164 /* flush failed for some other reason, give it another shot */
3165 ata_tf_init(dev, &tf);
3167 tf.command = qc->tf.command;
3168 tf.flags |= ATA_TFLAG_DEVICE;
3169 tf.protocol = ATA_PROT_NODATA;
3171 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3172 tf.command, qc->err_mask);
3174 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3175 if (!err_mask) {
3177 * FLUSH is complete but there's no way to
3178 * successfully complete a failed command from EH.
3179 * Making sure retry is allowed at least once and
3180 * retrying it should do the trick - whatever was in
3181 * the cache is already on the platter and this won't
3182 * cause infinite loop.
3184 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3185 } else {
3186 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3187 err_mask);
3188 rc = -EIO;
3190 /* if device failed it, report it to upper layers */
3191 if (err_mask & AC_ERR_DEV) {
3192 qc->err_mask |= AC_ERR_DEV;
3193 qc->result_tf = tf;
3194 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3195 rc = 0;
3198 return rc;
3201 static int ata_link_nr_enabled(struct ata_link *link)
3203 struct ata_device *dev;
3204 int cnt = 0;
3206 ata_for_each_dev(dev, link, ENABLED)
3207 cnt++;
3208 return cnt;
3211 static int ata_link_nr_vacant(struct ata_link *link)
3213 struct ata_device *dev;
3214 int cnt = 0;
3216 ata_for_each_dev(dev, link, ALL)
3217 if (dev->class == ATA_DEV_UNKNOWN)
3218 cnt++;
3219 return cnt;
3222 static int ata_eh_skip_recovery(struct ata_link *link)
3224 struct ata_port *ap = link->ap;
3225 struct ata_eh_context *ehc = &link->eh_context;
3226 struct ata_device *dev;
3228 /* skip disabled links */
3229 if (link->flags & ATA_LFLAG_DISABLED)
3230 return 1;
3232 /* thaw frozen port and recover failed devices */
3233 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3234 return 0;
3236 /* reset at least once if reset is requested */
3237 if ((ehc->i.action & ATA_EH_RESET) &&
3238 !(ehc->i.flags & ATA_EHI_DID_RESET))
3239 return 0;
3241 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3242 ata_for_each_dev(dev, link, ALL) {
3243 if (dev->class == ATA_DEV_UNKNOWN &&
3244 ehc->classes[dev->devno] != ATA_DEV_NONE)
3245 return 0;
3248 return 1;
3251 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3253 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3254 u64 now = get_jiffies_64();
3255 int *trials = void_arg;
3257 if (ent->timestamp < now - min(now, interval))
3258 return -1;
3260 (*trials)++;
3261 return 0;
3264 static int ata_eh_schedule_probe(struct ata_device *dev)
3266 struct ata_eh_context *ehc = &dev->link->eh_context;
3267 struct ata_link *link = ata_dev_phys_link(dev);
3268 int trials = 0;
3270 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3271 (ehc->did_probe_mask & (1 << dev->devno)))
3272 return 0;
3274 ata_eh_detach_dev(dev);
3275 ata_dev_init(dev);
3276 ehc->did_probe_mask |= (1 << dev->devno);
3277 ehc->i.action |= ATA_EH_RESET;
3278 ehc->saved_xfer_mode[dev->devno] = 0;
3279 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3281 /* Record and count probe trials on the ering. The specific
3282 * error mask used is irrelevant. Because a successful device
3283 * detection clears the ering, this count accumulates only if
3284 * there are consecutive failed probes.
3286 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3287 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3288 * forced to 1.5Gbps.
3290 * This is to work around cases where failed link speed
3291 * negotiation results in device misdetection leading to
3292 * infinite DEVXCHG or PHRDY CHG events.
3294 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3295 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3297 if (trials > ATA_EH_PROBE_TRIALS)
3298 sata_down_spd_limit(link, 1);
3300 return 1;
3303 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3305 struct ata_eh_context *ehc = &dev->link->eh_context;
3307 /* -EAGAIN from EH routine indicates retry without prejudice.
3308 * The requester is responsible for ensuring forward progress.
3310 if (err != -EAGAIN)
3311 ehc->tries[dev->devno]--;
3313 switch (err) {
3314 case -ENODEV:
3315 /* device missing or wrong IDENTIFY data, schedule probing */
3316 ehc->i.probe_mask |= (1 << dev->devno);
3317 case -EINVAL:
3318 /* give it just one more chance */
3319 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3320 case -EIO:
3321 if (ehc->tries[dev->devno] == 1) {
3322 /* This is the last chance, better to slow
3323 * down than lose it.
3325 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3326 if (dev->pio_mode > XFER_PIO_0)
3327 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3331 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3332 /* disable device if it has used up all its chances */
3333 ata_dev_disable(dev);
3335 /* detach if offline */
3336 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3337 ata_eh_detach_dev(dev);
3339 /* schedule probe if necessary */
3340 if (ata_eh_schedule_probe(dev)) {
3341 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3342 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3343 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3346 return 1;
3347 } else {
3348 ehc->i.action |= ATA_EH_RESET;
3349 return 0;
3354 * ata_eh_recover - recover host port after error
3355 * @ap: host port to recover
3356 * @prereset: prereset method (can be NULL)
3357 * @softreset: softreset method (can be NULL)
3358 * @hardreset: hardreset method (can be NULL)
3359 * @postreset: postreset method (can be NULL)
3360 * @r_failed_link: out parameter for failed link
3362 * This is the alpha and omega, eum and yang, heart and soul of
3363 * libata exception handling. On entry, actions required to
3364 * recover each link and hotplug requests are recorded in the
3365 * link's eh_context. This function executes all the operations
3366 * with appropriate retrials and fallbacks to resurrect failed
3367 * devices, detach goners and greet newcomers.
3369 * LOCKING:
3370 * Kernel thread context (may sleep).
3372 * RETURNS:
3373 * 0 on success, -errno on failure.
3375 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3376 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3377 ata_postreset_fn_t postreset,
3378 struct ata_link **r_failed_link)
3380 struct ata_link *link;
3381 struct ata_device *dev;
3382 int nr_failed_devs;
3383 int rc;
3384 unsigned long flags, deadline;
3386 DPRINTK("ENTER\n");
3388 /* prep for recovery */
3389 ata_for_each_link(link, ap, EDGE) {
3390 struct ata_eh_context *ehc = &link->eh_context;
3392 /* re-enable link? */
3393 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3394 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3395 spin_lock_irqsave(ap->lock, flags);
3396 link->flags &= ~ATA_LFLAG_DISABLED;
3397 spin_unlock_irqrestore(ap->lock, flags);
3398 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3401 ata_for_each_dev(dev, link, ALL) {
3402 if (link->flags & ATA_LFLAG_NO_RETRY)
3403 ehc->tries[dev->devno] = 1;
3404 else
3405 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3407 /* collect port action mask recorded in dev actions */
3408 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3409 ~ATA_EH_PERDEV_MASK;
3410 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3412 /* process hotplug request */
3413 if (dev->flags & ATA_DFLAG_DETACH)
3414 ata_eh_detach_dev(dev);
3416 /* schedule probe if necessary */
3417 if (!ata_dev_enabled(dev))
3418 ata_eh_schedule_probe(dev);
3422 retry:
3423 rc = 0;
3424 nr_failed_devs = 0;
3426 /* if UNLOADING, finish immediately */
3427 if (ap->pflags & ATA_PFLAG_UNLOADING)
3428 goto out;
3430 /* prep for EH */
3431 ata_for_each_link(link, ap, EDGE) {
3432 struct ata_eh_context *ehc = &link->eh_context;
3434 /* skip EH if possible. */
3435 if (ata_eh_skip_recovery(link))
3436 ehc->i.action = 0;
3438 ata_for_each_dev(dev, link, ALL)
3439 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3442 /* reset */
3443 ata_for_each_link(link, ap, EDGE) {
3444 struct ata_eh_context *ehc = &link->eh_context;
3446 if (!(ehc->i.action & ATA_EH_RESET))
3447 continue;
3449 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3450 prereset, softreset, hardreset, postreset);
3451 if (rc) {
3452 ata_link_printk(link, KERN_ERR,
3453 "reset failed, giving up\n");
3454 goto out;
3458 do {
3459 unsigned long now;
3462 * clears ATA_EH_PARK in eh_info and resets
3463 * ap->park_req_pending
3465 ata_eh_pull_park_action(ap);
3467 deadline = jiffies;
3468 ata_for_each_link(link, ap, EDGE) {
3469 ata_for_each_dev(dev, link, ALL) {
3470 struct ata_eh_context *ehc = &link->eh_context;
3471 unsigned long tmp;
3473 if (dev->class != ATA_DEV_ATA)
3474 continue;
3475 if (!(ehc->i.dev_action[dev->devno] &
3476 ATA_EH_PARK))
3477 continue;
3478 tmp = dev->unpark_deadline;
3479 if (time_before(deadline, tmp))
3480 deadline = tmp;
3481 else if (time_before_eq(tmp, jiffies))
3482 continue;
3483 if (ehc->unloaded_mask & (1 << dev->devno))
3484 continue;
3486 ata_eh_park_issue_cmd(dev, 1);
3490 now = jiffies;
3491 if (time_before_eq(deadline, now))
3492 break;
3494 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3495 deadline - now);
3496 } while (deadline);
3497 ata_for_each_link(link, ap, EDGE) {
3498 ata_for_each_dev(dev, link, ALL) {
3499 if (!(link->eh_context.unloaded_mask &
3500 (1 << dev->devno)))
3501 continue;
3503 ata_eh_park_issue_cmd(dev, 0);
3504 ata_eh_done(link, dev, ATA_EH_PARK);
3508 /* the rest */
3509 ata_for_each_link(link, ap, EDGE) {
3510 struct ata_eh_context *ehc = &link->eh_context;
3512 /* revalidate existing devices and attach new ones */
3513 rc = ata_eh_revalidate_and_attach(link, &dev);
3514 if (rc)
3515 goto dev_fail;
3517 /* if PMP got attached, return, pmp EH will take care of it */
3518 if (link->device->class == ATA_DEV_PMP) {
3519 ehc->i.action = 0;
3520 return 0;
3523 /* configure transfer mode if necessary */
3524 if (ehc->i.flags & ATA_EHI_SETMODE) {
3525 rc = ata_set_mode(link, &dev);
3526 if (rc)
3527 goto dev_fail;
3528 ehc->i.flags &= ~ATA_EHI_SETMODE;
3531 /* If reset has been issued, clear UA to avoid
3532 * disrupting the current users of the device.
3534 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3535 ata_for_each_dev(dev, link, ALL) {
3536 if (dev->class != ATA_DEV_ATAPI)
3537 continue;
3538 rc = atapi_eh_clear_ua(dev);
3539 if (rc)
3540 goto dev_fail;
3544 /* retry flush if necessary */
3545 ata_for_each_dev(dev, link, ALL) {
3546 if (dev->class != ATA_DEV_ATA)
3547 continue;
3548 rc = ata_eh_maybe_retry_flush(dev);
3549 if (rc)
3550 goto dev_fail;
3553 /* configure link power saving */
3554 if (ehc->i.action & ATA_EH_LPM)
3555 ata_for_each_dev(dev, link, ALL)
3556 ata_dev_enable_pm(dev, ap->pm_policy);
3558 /* this link is okay now */
3559 ehc->i.flags = 0;
3560 continue;
3562 dev_fail:
3563 nr_failed_devs++;
3564 ata_eh_handle_dev_fail(dev, rc);
3566 if (ap->pflags & ATA_PFLAG_FROZEN) {
3567 /* PMP reset requires working host port.
3568 * Can't retry if it's frozen.
3570 if (sata_pmp_attached(ap))
3571 goto out;
3572 break;
3576 if (nr_failed_devs)
3577 goto retry;
3579 out:
3580 if (rc && r_failed_link)
3581 *r_failed_link = link;
3583 DPRINTK("EXIT, rc=%d\n", rc);
3584 return rc;
3588 * ata_eh_finish - finish up EH
3589 * @ap: host port to finish EH for
3591 * Recovery is complete. Clean up EH states and retry or finish
3592 * failed qcs.
3594 * LOCKING:
3595 * None.
3597 void ata_eh_finish(struct ata_port *ap)
3599 int tag;
3601 /* retry or finish qcs */
3602 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3603 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3605 if (!(qc->flags & ATA_QCFLAG_FAILED))
3606 continue;
3608 if (qc->err_mask) {
3609 /* FIXME: Once EH migration is complete,
3610 * generate sense data in this function,
3611 * considering both err_mask and tf.
3613 if (qc->flags & ATA_QCFLAG_RETRY)
3614 ata_eh_qc_retry(qc);
3615 else
3616 ata_eh_qc_complete(qc);
3617 } else {
3618 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3619 ata_eh_qc_complete(qc);
3620 } else {
3621 /* feed zero TF to sense generation */
3622 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3623 ata_eh_qc_retry(qc);
3628 /* make sure nr_active_links is zero after EH */
3629 WARN_ON(ap->nr_active_links);
3630 ap->nr_active_links = 0;
3634 * ata_do_eh - do standard error handling
3635 * @ap: host port to handle error for
3637 * @prereset: prereset method (can be NULL)
3638 * @softreset: softreset method (can be NULL)
3639 * @hardreset: hardreset method (can be NULL)
3640 * @postreset: postreset method (can be NULL)
3642 * Perform standard error handling sequence.
3644 * LOCKING:
3645 * Kernel thread context (may sleep).
3647 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3648 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3649 ata_postreset_fn_t postreset)
3651 struct ata_device *dev;
3652 int rc;
3654 ata_eh_autopsy(ap);
3655 ata_eh_report(ap);
3657 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3658 NULL);
3659 if (rc) {
3660 ata_for_each_dev(dev, &ap->link, ALL)
3661 ata_dev_disable(dev);
3664 ata_eh_finish(ap);
3668 * ata_std_error_handler - standard error handler
3669 * @ap: host port to handle error for
3671 * Standard error handler
3673 * LOCKING:
3674 * Kernel thread context (may sleep).
3676 void ata_std_error_handler(struct ata_port *ap)
3678 struct ata_port_operations *ops = ap->ops;
3679 ata_reset_fn_t hardreset = ops->hardreset;
3681 /* ignore built-in hardreset if SCR access is not available */
3682 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3683 hardreset = NULL;
3685 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3688 #ifdef CONFIG_PM
3690 * ata_eh_handle_port_suspend - perform port suspend operation
3691 * @ap: port to suspend
3693 * Suspend @ap.
3695 * LOCKING:
3696 * Kernel thread context (may sleep).
3698 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3700 unsigned long flags;
3701 int rc = 0;
3703 /* are we suspending? */
3704 spin_lock_irqsave(ap->lock, flags);
3705 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3706 ap->pm_mesg.event == PM_EVENT_ON) {
3707 spin_unlock_irqrestore(ap->lock, flags);
3708 return;
3710 spin_unlock_irqrestore(ap->lock, flags);
3712 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3714 /* tell ACPI we're suspending */
3715 rc = ata_acpi_on_suspend(ap);
3716 if (rc)
3717 goto out;
3719 /* suspend */
3720 ata_eh_freeze_port(ap);
3722 if (ap->ops->port_suspend)
3723 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3725 ata_acpi_set_state(ap, PMSG_SUSPEND);
3726 out:
3727 /* report result */
3728 spin_lock_irqsave(ap->lock, flags);
3730 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3731 if (rc == 0)
3732 ap->pflags |= ATA_PFLAG_SUSPENDED;
3733 else if (ap->pflags & ATA_PFLAG_FROZEN)
3734 ata_port_schedule_eh(ap);
3736 if (ap->pm_result) {
3737 *ap->pm_result = rc;
3738 ap->pm_result = NULL;
3741 spin_unlock_irqrestore(ap->lock, flags);
3743 return;
3747 * ata_eh_handle_port_resume - perform port resume operation
3748 * @ap: port to resume
3750 * Resume @ap.
3752 * LOCKING:
3753 * Kernel thread context (may sleep).
3755 static void ata_eh_handle_port_resume(struct ata_port *ap)
3757 struct ata_link *link;
3758 struct ata_device *dev;
3759 unsigned long flags;
3760 int rc = 0;
3762 /* are we resuming? */
3763 spin_lock_irqsave(ap->lock, flags);
3764 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3765 ap->pm_mesg.event != PM_EVENT_ON) {
3766 spin_unlock_irqrestore(ap->lock, flags);
3767 return;
3769 spin_unlock_irqrestore(ap->lock, flags);
3771 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3774 * Error timestamps are in jiffies which doesn't run while
3775 * suspended and PHY events during resume isn't too uncommon.
3776 * When the two are combined, it can lead to unnecessary speed
3777 * downs if the machine is suspended and resumed repeatedly.
3778 * Clear error history.
3780 ata_for_each_link(link, ap, HOST_FIRST)
3781 ata_for_each_dev(dev, link, ALL)
3782 ata_ering_clear(&dev->ering);
3784 ata_acpi_set_state(ap, PMSG_ON);
3786 if (ap->ops->port_resume)
3787 rc = ap->ops->port_resume(ap);
3789 /* tell ACPI that we're resuming */
3790 ata_acpi_on_resume(ap);
3792 /* report result */
3793 spin_lock_irqsave(ap->lock, flags);
3794 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3795 if (ap->pm_result) {
3796 *ap->pm_result = rc;
3797 ap->pm_result = NULL;
3799 spin_unlock_irqrestore(ap->lock, flags);
3801 #endif /* CONFIG_PM */