kfifo: fix Error/broken kernel-doc notation
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / ata / libata-eh.c
blob0ea97c942ceda3485670021b18b9721802949391
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->err_mask & AC_ERR_INVALID) &&
2032 ((qc->flags & ATA_QCFLAG_IO) || qc->err_mask != AC_ERR_DEV))
2033 qc->flags |= ATA_QCFLAG_RETRY;
2035 /* accumulate error info */
2036 ehc->i.dev = qc->dev;
2037 all_err_mask |= qc->err_mask;
2038 if (qc->flags & ATA_QCFLAG_IO)
2039 eflags |= ATA_EFLAG_IS_IO;
2042 /* enforce default EH actions */
2043 if (ap->pflags & ATA_PFLAG_FROZEN ||
2044 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2045 ehc->i.action |= ATA_EH_RESET;
2046 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2047 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2048 ehc->i.action |= ATA_EH_REVALIDATE;
2050 /* If we have offending qcs and the associated failed device,
2051 * perform per-dev EH action only on the offending device.
2053 if (ehc->i.dev) {
2054 ehc->i.dev_action[ehc->i.dev->devno] |=
2055 ehc->i.action & ATA_EH_PERDEV_MASK;
2056 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2059 /* propagate timeout to host link */
2060 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2061 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2063 /* record error and consider speeding down */
2064 dev = ehc->i.dev;
2065 if (!dev && ((ata_link_max_devices(link) == 1 &&
2066 ata_dev_enabled(link->device))))
2067 dev = link->device;
2069 if (dev) {
2070 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2071 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2072 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2075 DPRINTK("EXIT\n");
2079 * ata_eh_autopsy - analyze error and determine recovery action
2080 * @ap: host port to perform autopsy on
2082 * Analyze all links of @ap and determine why they failed and
2083 * which recovery actions are needed.
2085 * LOCKING:
2086 * Kernel thread context (may sleep).
2088 void ata_eh_autopsy(struct ata_port *ap)
2090 struct ata_link *link;
2092 ata_for_each_link(link, ap, EDGE)
2093 ata_eh_link_autopsy(link);
2095 /* Handle the frigging slave link. Autopsy is done similarly
2096 * but actions and flags are transferred over to the master
2097 * link and handled from there.
2099 if (ap->slave_link) {
2100 struct ata_eh_context *mehc = &ap->link.eh_context;
2101 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2103 /* transfer control flags from master to slave */
2104 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2106 /* perform autopsy on the slave link */
2107 ata_eh_link_autopsy(ap->slave_link);
2109 /* transfer actions from slave to master and clear slave */
2110 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2111 mehc->i.action |= sehc->i.action;
2112 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2113 mehc->i.flags |= sehc->i.flags;
2114 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2117 /* Autopsy of fanout ports can affect host link autopsy.
2118 * Perform host link autopsy last.
2120 if (sata_pmp_attached(ap))
2121 ata_eh_link_autopsy(&ap->link);
2125 * ata_get_cmd_descript - get description for ATA command
2126 * @command: ATA command code to get description for
2128 * Return a textual description of the given command, or NULL if the
2129 * command is not known.
2131 * LOCKING:
2132 * None
2134 const char *ata_get_cmd_descript(u8 command)
2136 #ifdef CONFIG_ATA_VERBOSE_ERROR
2137 static const struct
2139 u8 command;
2140 const char *text;
2141 } cmd_descr[] = {
2142 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2143 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2144 { ATA_CMD_STANDBY, "STANDBY" },
2145 { ATA_CMD_IDLE, "IDLE" },
2146 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2147 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2148 { ATA_CMD_NOP, "NOP" },
2149 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2150 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2151 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2152 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2153 { ATA_CMD_SERVICE, "SERVICE" },
2154 { ATA_CMD_READ, "READ DMA" },
2155 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2156 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2157 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2158 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2159 { ATA_CMD_WRITE, "WRITE DMA" },
2160 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2161 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2162 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2163 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2164 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2165 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2166 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2167 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2168 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2169 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2170 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2171 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2172 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2173 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2174 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2175 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2176 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2177 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2178 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2179 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2180 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2181 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2182 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2183 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2184 { ATA_CMD_SLEEP, "SLEEP" },
2185 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2186 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2187 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2188 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2189 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2190 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2191 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2192 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2193 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2194 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2195 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2196 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2197 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2198 { ATA_CMD_PMP_READ, "READ BUFFER" },
2199 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2200 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2201 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2202 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2203 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2204 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2205 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2206 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2207 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2208 { ATA_CMD_SMART, "SMART" },
2209 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2210 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2211 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2212 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2213 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2214 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2215 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2216 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2217 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2218 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2219 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2220 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2221 { ATA_CMD_RESTORE, "RECALIBRATE" },
2222 { 0, NULL } /* terminate list */
2225 unsigned int i;
2226 for (i = 0; cmd_descr[i].text; i++)
2227 if (cmd_descr[i].command == command)
2228 return cmd_descr[i].text;
2229 #endif
2231 return NULL;
2235 * ata_eh_link_report - report error handling to user
2236 * @link: ATA link EH is going on
2238 * Report EH to user.
2240 * LOCKING:
2241 * None.
2243 static void ata_eh_link_report(struct ata_link *link)
2245 struct ata_port *ap = link->ap;
2246 struct ata_eh_context *ehc = &link->eh_context;
2247 const char *frozen, *desc;
2248 char tries_buf[6];
2249 int tag, nr_failed = 0;
2251 if (ehc->i.flags & ATA_EHI_QUIET)
2252 return;
2254 desc = NULL;
2255 if (ehc->i.desc[0] != '\0')
2256 desc = ehc->i.desc;
2258 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2259 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2261 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2262 ata_dev_phys_link(qc->dev) != link ||
2263 ((qc->flags & ATA_QCFLAG_QUIET) &&
2264 qc->err_mask == AC_ERR_DEV))
2265 continue;
2266 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2267 continue;
2269 nr_failed++;
2272 if (!nr_failed && !ehc->i.err_mask)
2273 return;
2275 frozen = "";
2276 if (ap->pflags & ATA_PFLAG_FROZEN)
2277 frozen = " frozen";
2279 memset(tries_buf, 0, sizeof(tries_buf));
2280 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2281 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2282 ap->eh_tries);
2284 if (ehc->i.dev) {
2285 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2286 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2287 ehc->i.err_mask, link->sactive, ehc->i.serror,
2288 ehc->i.action, frozen, tries_buf);
2289 if (desc)
2290 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2291 } else {
2292 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2293 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2294 ehc->i.err_mask, link->sactive, ehc->i.serror,
2295 ehc->i.action, frozen, tries_buf);
2296 if (desc)
2297 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2300 #ifdef CONFIG_ATA_VERBOSE_ERROR
2301 if (ehc->i.serror)
2302 ata_link_printk(link, KERN_ERR,
2303 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2304 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2305 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2306 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2307 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2308 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2309 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2310 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2311 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2312 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2313 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2314 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2315 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2316 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2317 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2318 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2319 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2320 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2321 #endif
2323 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2324 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2325 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2326 const u8 *cdb = qc->cdb;
2327 char data_buf[20] = "";
2328 char cdb_buf[70] = "";
2330 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2331 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2332 continue;
2334 if (qc->dma_dir != DMA_NONE) {
2335 static const char *dma_str[] = {
2336 [DMA_BIDIRECTIONAL] = "bidi",
2337 [DMA_TO_DEVICE] = "out",
2338 [DMA_FROM_DEVICE] = "in",
2340 static const char *prot_str[] = {
2341 [ATA_PROT_PIO] = "pio",
2342 [ATA_PROT_DMA] = "dma",
2343 [ATA_PROT_NCQ] = "ncq",
2344 [ATAPI_PROT_PIO] = "pio",
2345 [ATAPI_PROT_DMA] = "dma",
2348 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2349 prot_str[qc->tf.protocol], qc->nbytes,
2350 dma_str[qc->dma_dir]);
2353 if (ata_is_atapi(qc->tf.protocol)) {
2354 if (qc->scsicmd)
2355 scsi_print_command(qc->scsicmd);
2356 else
2357 snprintf(cdb_buf, sizeof(cdb_buf),
2358 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2359 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2360 cdb[0], cdb[1], cdb[2], cdb[3],
2361 cdb[4], cdb[5], cdb[6], cdb[7],
2362 cdb[8], cdb[9], cdb[10], cdb[11],
2363 cdb[12], cdb[13], cdb[14], cdb[15]);
2364 } else {
2365 const char *descr = ata_get_cmd_descript(cmd->command);
2366 if (descr)
2367 ata_dev_printk(qc->dev, KERN_ERR,
2368 "failed command: %s\n", descr);
2371 ata_dev_printk(qc->dev, KERN_ERR,
2372 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2373 "tag %d%s\n %s"
2374 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2375 "Emask 0x%x (%s)%s\n",
2376 cmd->command, cmd->feature, cmd->nsect,
2377 cmd->lbal, cmd->lbam, cmd->lbah,
2378 cmd->hob_feature, cmd->hob_nsect,
2379 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2380 cmd->device, qc->tag, data_buf, cdb_buf,
2381 res->command, res->feature, res->nsect,
2382 res->lbal, res->lbam, res->lbah,
2383 res->hob_feature, res->hob_nsect,
2384 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2385 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2386 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2388 #ifdef CONFIG_ATA_VERBOSE_ERROR
2389 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2390 ATA_ERR)) {
2391 if (res->command & ATA_BUSY)
2392 ata_dev_printk(qc->dev, KERN_ERR,
2393 "status: { Busy }\n");
2394 else
2395 ata_dev_printk(qc->dev, KERN_ERR,
2396 "status: { %s%s%s%s}\n",
2397 res->command & ATA_DRDY ? "DRDY " : "",
2398 res->command & ATA_DF ? "DF " : "",
2399 res->command & ATA_DRQ ? "DRQ " : "",
2400 res->command & ATA_ERR ? "ERR " : "");
2403 if (cmd->command != ATA_CMD_PACKET &&
2404 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2405 ATA_ABORTED)))
2406 ata_dev_printk(qc->dev, KERN_ERR,
2407 "error: { %s%s%s%s}\n",
2408 res->feature & ATA_ICRC ? "ICRC " : "",
2409 res->feature & ATA_UNC ? "UNC " : "",
2410 res->feature & ATA_IDNF ? "IDNF " : "",
2411 res->feature & ATA_ABORTED ? "ABRT " : "");
2412 #endif
2417 * ata_eh_report - report error handling to user
2418 * @ap: ATA port to report EH about
2420 * Report EH to user.
2422 * LOCKING:
2423 * None.
2425 void ata_eh_report(struct ata_port *ap)
2427 struct ata_link *link;
2429 ata_for_each_link(link, ap, HOST_FIRST)
2430 ata_eh_link_report(link);
2433 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2434 unsigned int *classes, unsigned long deadline,
2435 bool clear_classes)
2437 struct ata_device *dev;
2439 if (clear_classes)
2440 ata_for_each_dev(dev, link, ALL)
2441 classes[dev->devno] = ATA_DEV_UNKNOWN;
2443 return reset(link, classes, deadline);
2446 static int ata_eh_followup_srst_needed(struct ata_link *link,
2447 int rc, const unsigned int *classes)
2449 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2450 return 0;
2451 if (rc == -EAGAIN)
2452 return 1;
2453 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2454 return 1;
2455 return 0;
2458 int ata_eh_reset(struct ata_link *link, int classify,
2459 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2460 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2462 struct ata_port *ap = link->ap;
2463 struct ata_link *slave = ap->slave_link;
2464 struct ata_eh_context *ehc = &link->eh_context;
2465 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2466 unsigned int *classes = ehc->classes;
2467 unsigned int lflags = link->flags;
2468 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2469 int max_tries = 0, try = 0;
2470 struct ata_link *failed_link;
2471 struct ata_device *dev;
2472 unsigned long deadline, now;
2473 ata_reset_fn_t reset;
2474 unsigned long flags;
2475 u32 sstatus;
2476 int nr_unknown, rc;
2479 * Prepare to reset
2481 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2482 max_tries++;
2483 if (link->flags & ATA_LFLAG_NO_HRST)
2484 hardreset = NULL;
2485 if (link->flags & ATA_LFLAG_NO_SRST)
2486 softreset = NULL;
2488 /* make sure each reset attemp is at least COOL_DOWN apart */
2489 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2490 now = jiffies;
2491 WARN_ON(time_after(ehc->last_reset, now));
2492 deadline = ata_deadline(ehc->last_reset,
2493 ATA_EH_RESET_COOL_DOWN);
2494 if (time_before(now, deadline))
2495 schedule_timeout_uninterruptible(deadline - now);
2498 spin_lock_irqsave(ap->lock, flags);
2499 ap->pflags |= ATA_PFLAG_RESETTING;
2500 spin_unlock_irqrestore(ap->lock, flags);
2502 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2504 ata_for_each_dev(dev, link, ALL) {
2505 /* If we issue an SRST then an ATA drive (not ATAPI)
2506 * may change configuration and be in PIO0 timing. If
2507 * we do a hard reset (or are coming from power on)
2508 * this is true for ATA or ATAPI. Until we've set a
2509 * suitable controller mode we should not touch the
2510 * bus as we may be talking too fast.
2512 dev->pio_mode = XFER_PIO_0;
2514 /* If the controller has a pio mode setup function
2515 * then use it to set the chipset to rights. Don't
2516 * touch the DMA setup as that will be dealt with when
2517 * configuring devices.
2519 if (ap->ops->set_piomode)
2520 ap->ops->set_piomode(ap, dev);
2523 /* prefer hardreset */
2524 reset = NULL;
2525 ehc->i.action &= ~ATA_EH_RESET;
2526 if (hardreset) {
2527 reset = hardreset;
2528 ehc->i.action |= ATA_EH_HARDRESET;
2529 } else if (softreset) {
2530 reset = softreset;
2531 ehc->i.action |= ATA_EH_SOFTRESET;
2534 if (prereset) {
2535 unsigned long deadline = ata_deadline(jiffies,
2536 ATA_EH_PRERESET_TIMEOUT);
2538 if (slave) {
2539 sehc->i.action &= ~ATA_EH_RESET;
2540 sehc->i.action |= ehc->i.action;
2543 rc = prereset(link, deadline);
2545 /* If present, do prereset on slave link too. Reset
2546 * is skipped iff both master and slave links report
2547 * -ENOENT or clear ATA_EH_RESET.
2549 if (slave && (rc == 0 || rc == -ENOENT)) {
2550 int tmp;
2552 tmp = prereset(slave, deadline);
2553 if (tmp != -ENOENT)
2554 rc = tmp;
2556 ehc->i.action |= sehc->i.action;
2559 if (rc) {
2560 if (rc == -ENOENT) {
2561 ata_link_printk(link, KERN_DEBUG,
2562 "port disabled. ignoring.\n");
2563 ehc->i.action &= ~ATA_EH_RESET;
2565 ata_for_each_dev(dev, link, ALL)
2566 classes[dev->devno] = ATA_DEV_NONE;
2568 rc = 0;
2569 } else
2570 ata_link_printk(link, KERN_ERR,
2571 "prereset failed (errno=%d)\n", rc);
2572 goto out;
2575 /* prereset() might have cleared ATA_EH_RESET. If so,
2576 * bang classes, thaw and return.
2578 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2579 ata_for_each_dev(dev, link, ALL)
2580 classes[dev->devno] = ATA_DEV_NONE;
2581 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2582 ata_is_host_link(link))
2583 ata_eh_thaw_port(ap);
2584 rc = 0;
2585 goto out;
2589 retry:
2591 * Perform reset
2593 if (ata_is_host_link(link))
2594 ata_eh_freeze_port(ap);
2596 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2598 if (reset) {
2599 if (verbose)
2600 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2601 reset == softreset ? "soft" : "hard");
2603 /* mark that this EH session started with reset */
2604 ehc->last_reset = jiffies;
2605 if (reset == hardreset)
2606 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2607 else
2608 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2610 rc = ata_do_reset(link, reset, classes, deadline, true);
2611 if (rc && rc != -EAGAIN) {
2612 failed_link = link;
2613 goto fail;
2616 /* hardreset slave link if existent */
2617 if (slave && reset == hardreset) {
2618 int tmp;
2620 if (verbose)
2621 ata_link_printk(slave, KERN_INFO,
2622 "hard resetting link\n");
2624 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2625 tmp = ata_do_reset(slave, reset, classes, deadline,
2626 false);
2627 switch (tmp) {
2628 case -EAGAIN:
2629 rc = -EAGAIN;
2630 case 0:
2631 break;
2632 default:
2633 failed_link = slave;
2634 rc = tmp;
2635 goto fail;
2639 /* perform follow-up SRST if necessary */
2640 if (reset == hardreset &&
2641 ata_eh_followup_srst_needed(link, rc, classes)) {
2642 reset = softreset;
2644 if (!reset) {
2645 ata_link_printk(link, KERN_ERR,
2646 "follow-up softreset required "
2647 "but no softreset avaliable\n");
2648 failed_link = link;
2649 rc = -EINVAL;
2650 goto fail;
2653 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2654 rc = ata_do_reset(link, reset, classes, deadline, true);
2655 if (rc) {
2656 failed_link = link;
2657 goto fail;
2660 } else {
2661 if (verbose)
2662 ata_link_printk(link, KERN_INFO, "no reset method "
2663 "available, skipping reset\n");
2664 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2665 lflags |= ATA_LFLAG_ASSUME_ATA;
2669 * Post-reset processing
2671 ata_for_each_dev(dev, link, ALL) {
2672 /* After the reset, the device state is PIO 0 and the
2673 * controller state is undefined. Reset also wakes up
2674 * drives from sleeping mode.
2676 dev->pio_mode = XFER_PIO_0;
2677 dev->flags &= ~ATA_DFLAG_SLEEPING;
2679 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2680 continue;
2682 /* apply class override */
2683 if (lflags & ATA_LFLAG_ASSUME_ATA)
2684 classes[dev->devno] = ATA_DEV_ATA;
2685 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2686 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2689 /* record current link speed */
2690 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2691 link->sata_spd = (sstatus >> 4) & 0xf;
2692 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2693 slave->sata_spd = (sstatus >> 4) & 0xf;
2695 /* thaw the port */
2696 if (ata_is_host_link(link))
2697 ata_eh_thaw_port(ap);
2699 /* postreset() should clear hardware SError. Although SError
2700 * is cleared during link resume, clearing SError here is
2701 * necessary as some PHYs raise hotplug events after SRST.
2702 * This introduces race condition where hotplug occurs between
2703 * reset and here. This race is mediated by cross checking
2704 * link onlineness and classification result later.
2706 if (postreset) {
2707 postreset(link, classes);
2708 if (slave)
2709 postreset(slave, classes);
2713 * Some controllers can't be frozen very well and may set
2714 * spuruious error conditions during reset. Clear accumulated
2715 * error information. As reset is the final recovery action,
2716 * nothing is lost by doing this.
2718 spin_lock_irqsave(link->ap->lock, flags);
2719 memset(&link->eh_info, 0, sizeof(link->eh_info));
2720 if (slave)
2721 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2722 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2723 spin_unlock_irqrestore(link->ap->lock, flags);
2726 * Make sure onlineness and classification result correspond.
2727 * Hotplug could have happened during reset and some
2728 * controllers fail to wait while a drive is spinning up after
2729 * being hotplugged causing misdetection. By cross checking
2730 * link on/offlineness and classification result, those
2731 * conditions can be reliably detected and retried.
2733 nr_unknown = 0;
2734 ata_for_each_dev(dev, link, ALL) {
2735 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2736 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2737 ata_dev_printk(dev, KERN_DEBUG, "link online "
2738 "but device misclassifed\n");
2739 classes[dev->devno] = ATA_DEV_NONE;
2740 nr_unknown++;
2742 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2743 if (ata_class_enabled(classes[dev->devno]))
2744 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2745 "clearing class %d to NONE\n",
2746 classes[dev->devno]);
2747 classes[dev->devno] = ATA_DEV_NONE;
2748 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2749 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2750 "clearing UNKNOWN to NONE\n");
2751 classes[dev->devno] = ATA_DEV_NONE;
2755 if (classify && nr_unknown) {
2756 if (try < max_tries) {
2757 ata_link_printk(link, KERN_WARNING, "link online but "
2758 "%d devices misclassified, retrying\n",
2759 nr_unknown);
2760 failed_link = link;
2761 rc = -EAGAIN;
2762 goto fail;
2764 ata_link_printk(link, KERN_WARNING,
2765 "link online but %d devices misclassified, "
2766 "device detection might fail\n", nr_unknown);
2769 /* reset successful, schedule revalidation */
2770 ata_eh_done(link, NULL, ATA_EH_RESET);
2771 if (slave)
2772 ata_eh_done(slave, NULL, ATA_EH_RESET);
2773 ehc->last_reset = jiffies; /* update to completion time */
2774 ehc->i.action |= ATA_EH_REVALIDATE;
2776 rc = 0;
2777 out:
2778 /* clear hotplug flag */
2779 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2780 if (slave)
2781 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2783 spin_lock_irqsave(ap->lock, flags);
2784 ap->pflags &= ~ATA_PFLAG_RESETTING;
2785 spin_unlock_irqrestore(ap->lock, flags);
2787 return rc;
2789 fail:
2790 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2791 if (!ata_is_host_link(link) &&
2792 sata_scr_read(link, SCR_STATUS, &sstatus))
2793 rc = -ERESTART;
2795 if (rc == -ERESTART || try >= max_tries)
2796 goto out;
2798 now = jiffies;
2799 if (time_before(now, deadline)) {
2800 unsigned long delta = deadline - now;
2802 ata_link_printk(failed_link, KERN_WARNING,
2803 "reset failed (errno=%d), retrying in %u secs\n",
2804 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2806 while (delta)
2807 delta = schedule_timeout_uninterruptible(delta);
2810 if (try == max_tries - 1) {
2811 sata_down_spd_limit(link, 0);
2812 if (slave)
2813 sata_down_spd_limit(slave, 0);
2814 } else if (rc == -EPIPE)
2815 sata_down_spd_limit(failed_link, 0);
2817 if (hardreset)
2818 reset = hardreset;
2819 goto retry;
2822 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2824 struct ata_link *link;
2825 struct ata_device *dev;
2826 unsigned long flags;
2829 * This function can be thought of as an extended version of
2830 * ata_eh_about_to_do() specially crafted to accommodate the
2831 * requirements of ATA_EH_PARK handling. Since the EH thread
2832 * does not leave the do {} while () loop in ata_eh_recover as
2833 * long as the timeout for a park request to *one* device on
2834 * the port has not expired, and since we still want to pick
2835 * up park requests to other devices on the same port or
2836 * timeout updates for the same device, we have to pull
2837 * ATA_EH_PARK actions from eh_info into eh_context.i
2838 * ourselves at the beginning of each pass over the loop.
2840 * Additionally, all write accesses to &ap->park_req_pending
2841 * through INIT_COMPLETION() (see below) or complete_all()
2842 * (see ata_scsi_park_store()) are protected by the host lock.
2843 * As a result we have that park_req_pending.done is zero on
2844 * exit from this function, i.e. when ATA_EH_PARK actions for
2845 * *all* devices on port ap have been pulled into the
2846 * respective eh_context structs. If, and only if,
2847 * park_req_pending.done is non-zero by the time we reach
2848 * wait_for_completion_timeout(), another ATA_EH_PARK action
2849 * has been scheduled for at least one of the devices on port
2850 * ap and we have to cycle over the do {} while () loop in
2851 * ata_eh_recover() again.
2854 spin_lock_irqsave(ap->lock, flags);
2855 INIT_COMPLETION(ap->park_req_pending);
2856 ata_for_each_link(link, ap, EDGE) {
2857 ata_for_each_dev(dev, link, ALL) {
2858 struct ata_eh_info *ehi = &link->eh_info;
2860 link->eh_context.i.dev_action[dev->devno] |=
2861 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2862 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2865 spin_unlock_irqrestore(ap->lock, flags);
2868 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2870 struct ata_eh_context *ehc = &dev->link->eh_context;
2871 struct ata_taskfile tf;
2872 unsigned int err_mask;
2874 ata_tf_init(dev, &tf);
2875 if (park) {
2876 ehc->unloaded_mask |= 1 << dev->devno;
2877 tf.command = ATA_CMD_IDLEIMMEDIATE;
2878 tf.feature = 0x44;
2879 tf.lbal = 0x4c;
2880 tf.lbam = 0x4e;
2881 tf.lbah = 0x55;
2882 } else {
2883 ehc->unloaded_mask &= ~(1 << dev->devno);
2884 tf.command = ATA_CMD_CHK_POWER;
2887 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2888 tf.protocol |= ATA_PROT_NODATA;
2889 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2890 if (park && (err_mask || tf.lbal != 0xc4)) {
2891 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2892 ehc->unloaded_mask &= ~(1 << dev->devno);
2896 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2897 struct ata_device **r_failed_dev)
2899 struct ata_port *ap = link->ap;
2900 struct ata_eh_context *ehc = &link->eh_context;
2901 struct ata_device *dev;
2902 unsigned int new_mask = 0;
2903 unsigned long flags;
2904 int rc = 0;
2906 DPRINTK("ENTER\n");
2908 /* For PATA drive side cable detection to work, IDENTIFY must
2909 * be done backwards such that PDIAG- is released by the slave
2910 * device before the master device is identified.
2912 ata_for_each_dev(dev, link, ALL_REVERSE) {
2913 unsigned int action = ata_eh_dev_action(dev);
2914 unsigned int readid_flags = 0;
2916 if (ehc->i.flags & ATA_EHI_DID_RESET)
2917 readid_flags |= ATA_READID_POSTRESET;
2919 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2920 WARN_ON(dev->class == ATA_DEV_PMP);
2922 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2923 rc = -EIO;
2924 goto err;
2927 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2928 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2929 readid_flags);
2930 if (rc)
2931 goto err;
2933 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2935 /* Configuration may have changed, reconfigure
2936 * transfer mode.
2938 ehc->i.flags |= ATA_EHI_SETMODE;
2940 /* schedule the scsi_rescan_device() here */
2941 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
2942 } else if (dev->class == ATA_DEV_UNKNOWN &&
2943 ehc->tries[dev->devno] &&
2944 ata_class_enabled(ehc->classes[dev->devno])) {
2945 /* Temporarily set dev->class, it will be
2946 * permanently set once all configurations are
2947 * complete. This is necessary because new
2948 * device configuration is done in two
2949 * separate loops.
2951 dev->class = ehc->classes[dev->devno];
2953 if (dev->class == ATA_DEV_PMP)
2954 rc = sata_pmp_attach(dev);
2955 else
2956 rc = ata_dev_read_id(dev, &dev->class,
2957 readid_flags, dev->id);
2959 /* read_id might have changed class, store and reset */
2960 ehc->classes[dev->devno] = dev->class;
2961 dev->class = ATA_DEV_UNKNOWN;
2963 switch (rc) {
2964 case 0:
2965 /* clear error info accumulated during probe */
2966 ata_ering_clear(&dev->ering);
2967 new_mask |= 1 << dev->devno;
2968 break;
2969 case -ENOENT:
2970 /* IDENTIFY was issued to non-existent
2971 * device. No need to reset. Just
2972 * thaw and ignore the device.
2974 ata_eh_thaw_port(ap);
2975 break;
2976 default:
2977 goto err;
2982 /* PDIAG- should have been released, ask cable type if post-reset */
2983 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
2984 if (ap->ops->cable_detect)
2985 ap->cbl = ap->ops->cable_detect(ap);
2986 ata_force_cbl(ap);
2989 /* Configure new devices forward such that user doesn't see
2990 * device detection messages backwards.
2992 ata_for_each_dev(dev, link, ALL) {
2993 if (!(new_mask & (1 << dev->devno)))
2994 continue;
2996 dev->class = ehc->classes[dev->devno];
2998 if (dev->class == ATA_DEV_PMP)
2999 continue;
3001 ehc->i.flags |= ATA_EHI_PRINTINFO;
3002 rc = ata_dev_configure(dev);
3003 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3004 if (rc) {
3005 dev->class = ATA_DEV_UNKNOWN;
3006 goto err;
3009 spin_lock_irqsave(ap->lock, flags);
3010 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3011 spin_unlock_irqrestore(ap->lock, flags);
3013 /* new device discovered, configure xfermode */
3014 ehc->i.flags |= ATA_EHI_SETMODE;
3017 return 0;
3019 err:
3020 *r_failed_dev = dev;
3021 DPRINTK("EXIT rc=%d\n", rc);
3022 return rc;
3026 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3027 * @link: link on which timings will be programmed
3028 * @r_failed_dev: out parameter for failed device
3030 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3031 * ata_set_mode() fails, pointer to the failing device is
3032 * returned in @r_failed_dev.
3034 * LOCKING:
3035 * PCI/etc. bus probe sem.
3037 * RETURNS:
3038 * 0 on success, negative errno otherwise
3040 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3042 struct ata_port *ap = link->ap;
3043 struct ata_device *dev;
3044 int rc;
3046 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3047 ata_for_each_dev(dev, link, ENABLED) {
3048 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3049 struct ata_ering_entry *ent;
3051 ent = ata_ering_top(&dev->ering);
3052 if (ent)
3053 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3057 /* has private set_mode? */
3058 if (ap->ops->set_mode)
3059 rc = ap->ops->set_mode(link, r_failed_dev);
3060 else
3061 rc = ata_do_set_mode(link, r_failed_dev);
3063 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3064 ata_for_each_dev(dev, link, ENABLED) {
3065 struct ata_eh_context *ehc = &link->eh_context;
3066 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3067 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3069 if (dev->xfer_mode != saved_xfer_mode ||
3070 ata_ncq_enabled(dev) != saved_ncq)
3071 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3074 return rc;
3078 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3079 * @dev: ATAPI device to clear UA for
3081 * Resets and other operations can make an ATAPI device raise
3082 * UNIT ATTENTION which causes the next operation to fail. This
3083 * function clears UA.
3085 * LOCKING:
3086 * EH context (may sleep).
3088 * RETURNS:
3089 * 0 on success, -errno on failure.
3091 static int atapi_eh_clear_ua(struct ata_device *dev)
3093 int i;
3095 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3096 u8 *sense_buffer = dev->link->ap->sector_buf;
3097 u8 sense_key = 0;
3098 unsigned int err_mask;
3100 err_mask = atapi_eh_tur(dev, &sense_key);
3101 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3102 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3103 "failed (err_mask=0x%x)\n", err_mask);
3104 return -EIO;
3107 if (!err_mask || sense_key != UNIT_ATTENTION)
3108 return 0;
3110 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3111 if (err_mask) {
3112 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3113 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3114 return -EIO;
3118 ata_dev_printk(dev, KERN_WARNING,
3119 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3121 return 0;
3125 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3126 * @dev: ATA device which may need FLUSH retry
3128 * If @dev failed FLUSH, it needs to be reported upper layer
3129 * immediately as it means that @dev failed to remap and already
3130 * lost at least a sector and further FLUSH retrials won't make
3131 * any difference to the lost sector. However, if FLUSH failed
3132 * for other reasons, for example transmission error, FLUSH needs
3133 * to be retried.
3135 * This function determines whether FLUSH failure retry is
3136 * necessary and performs it if so.
3138 * RETURNS:
3139 * 0 if EH can continue, -errno if EH needs to be repeated.
3141 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3143 struct ata_link *link = dev->link;
3144 struct ata_port *ap = link->ap;
3145 struct ata_queued_cmd *qc;
3146 struct ata_taskfile tf;
3147 unsigned int err_mask;
3148 int rc = 0;
3150 /* did flush fail for this device? */
3151 if (!ata_tag_valid(link->active_tag))
3152 return 0;
3154 qc = __ata_qc_from_tag(ap, link->active_tag);
3155 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3156 qc->tf.command != ATA_CMD_FLUSH))
3157 return 0;
3159 /* if the device failed it, it should be reported to upper layers */
3160 if (qc->err_mask & AC_ERR_DEV)
3161 return 0;
3163 /* flush failed for some other reason, give it another shot */
3164 ata_tf_init(dev, &tf);
3166 tf.command = qc->tf.command;
3167 tf.flags |= ATA_TFLAG_DEVICE;
3168 tf.protocol = ATA_PROT_NODATA;
3170 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3171 tf.command, qc->err_mask);
3173 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3174 if (!err_mask) {
3176 * FLUSH is complete but there's no way to
3177 * successfully complete a failed command from EH.
3178 * Making sure retry is allowed at least once and
3179 * retrying it should do the trick - whatever was in
3180 * the cache is already on the platter and this won't
3181 * cause infinite loop.
3183 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3184 } else {
3185 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3186 err_mask);
3187 rc = -EIO;
3189 /* if device failed it, report it to upper layers */
3190 if (err_mask & AC_ERR_DEV) {
3191 qc->err_mask |= AC_ERR_DEV;
3192 qc->result_tf = tf;
3193 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3194 rc = 0;
3197 return rc;
3200 static int ata_link_nr_enabled(struct ata_link *link)
3202 struct ata_device *dev;
3203 int cnt = 0;
3205 ata_for_each_dev(dev, link, ENABLED)
3206 cnt++;
3207 return cnt;
3210 static int ata_link_nr_vacant(struct ata_link *link)
3212 struct ata_device *dev;
3213 int cnt = 0;
3215 ata_for_each_dev(dev, link, ALL)
3216 if (dev->class == ATA_DEV_UNKNOWN)
3217 cnt++;
3218 return cnt;
3221 static int ata_eh_skip_recovery(struct ata_link *link)
3223 struct ata_port *ap = link->ap;
3224 struct ata_eh_context *ehc = &link->eh_context;
3225 struct ata_device *dev;
3227 /* skip disabled links */
3228 if (link->flags & ATA_LFLAG_DISABLED)
3229 return 1;
3231 /* thaw frozen port and recover failed devices */
3232 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3233 return 0;
3235 /* reset at least once if reset is requested */
3236 if ((ehc->i.action & ATA_EH_RESET) &&
3237 !(ehc->i.flags & ATA_EHI_DID_RESET))
3238 return 0;
3240 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3241 ata_for_each_dev(dev, link, ALL) {
3242 if (dev->class == ATA_DEV_UNKNOWN &&
3243 ehc->classes[dev->devno] != ATA_DEV_NONE)
3244 return 0;
3247 return 1;
3250 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3252 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3253 u64 now = get_jiffies_64();
3254 int *trials = void_arg;
3256 if (ent->timestamp < now - min(now, interval))
3257 return -1;
3259 (*trials)++;
3260 return 0;
3263 static int ata_eh_schedule_probe(struct ata_device *dev)
3265 struct ata_eh_context *ehc = &dev->link->eh_context;
3266 struct ata_link *link = ata_dev_phys_link(dev);
3267 int trials = 0;
3269 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3270 (ehc->did_probe_mask & (1 << dev->devno)))
3271 return 0;
3273 ata_eh_detach_dev(dev);
3274 ata_dev_init(dev);
3275 ehc->did_probe_mask |= (1 << dev->devno);
3276 ehc->i.action |= ATA_EH_RESET;
3277 ehc->saved_xfer_mode[dev->devno] = 0;
3278 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3280 /* Record and count probe trials on the ering. The specific
3281 * error mask used is irrelevant. Because a successful device
3282 * detection clears the ering, this count accumulates only if
3283 * there are consecutive failed probes.
3285 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3286 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3287 * forced to 1.5Gbps.
3289 * This is to work around cases where failed link speed
3290 * negotiation results in device misdetection leading to
3291 * infinite DEVXCHG or PHRDY CHG events.
3293 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3294 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3296 if (trials > ATA_EH_PROBE_TRIALS)
3297 sata_down_spd_limit(link, 1);
3299 return 1;
3302 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3304 struct ata_eh_context *ehc = &dev->link->eh_context;
3306 /* -EAGAIN from EH routine indicates retry without prejudice.
3307 * The requester is responsible for ensuring forward progress.
3309 if (err != -EAGAIN)
3310 ehc->tries[dev->devno]--;
3312 switch (err) {
3313 case -ENODEV:
3314 /* device missing or wrong IDENTIFY data, schedule probing */
3315 ehc->i.probe_mask |= (1 << dev->devno);
3316 case -EINVAL:
3317 /* give it just one more chance */
3318 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3319 case -EIO:
3320 if (ehc->tries[dev->devno] == 1) {
3321 /* This is the last chance, better to slow
3322 * down than lose it.
3324 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3325 if (dev->pio_mode > XFER_PIO_0)
3326 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3330 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3331 /* disable device if it has used up all its chances */
3332 ata_dev_disable(dev);
3334 /* detach if offline */
3335 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3336 ata_eh_detach_dev(dev);
3338 /* schedule probe if necessary */
3339 if (ata_eh_schedule_probe(dev)) {
3340 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3341 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3342 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3345 return 1;
3346 } else {
3347 ehc->i.action |= ATA_EH_RESET;
3348 return 0;
3353 * ata_eh_recover - recover host port after error
3354 * @ap: host port to recover
3355 * @prereset: prereset method (can be NULL)
3356 * @softreset: softreset method (can be NULL)
3357 * @hardreset: hardreset method (can be NULL)
3358 * @postreset: postreset method (can be NULL)
3359 * @r_failed_link: out parameter for failed link
3361 * This is the alpha and omega, eum and yang, heart and soul of
3362 * libata exception handling. On entry, actions required to
3363 * recover each link and hotplug requests are recorded in the
3364 * link's eh_context. This function executes all the operations
3365 * with appropriate retrials and fallbacks to resurrect failed
3366 * devices, detach goners and greet newcomers.
3368 * LOCKING:
3369 * Kernel thread context (may sleep).
3371 * RETURNS:
3372 * 0 on success, -errno on failure.
3374 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3375 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3376 ata_postreset_fn_t postreset,
3377 struct ata_link **r_failed_link)
3379 struct ata_link *link;
3380 struct ata_device *dev;
3381 int nr_failed_devs;
3382 int rc;
3383 unsigned long flags, deadline;
3385 DPRINTK("ENTER\n");
3387 /* prep for recovery */
3388 ata_for_each_link(link, ap, EDGE) {
3389 struct ata_eh_context *ehc = &link->eh_context;
3391 /* re-enable link? */
3392 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3393 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3394 spin_lock_irqsave(ap->lock, flags);
3395 link->flags &= ~ATA_LFLAG_DISABLED;
3396 spin_unlock_irqrestore(ap->lock, flags);
3397 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3400 ata_for_each_dev(dev, link, ALL) {
3401 if (link->flags & ATA_LFLAG_NO_RETRY)
3402 ehc->tries[dev->devno] = 1;
3403 else
3404 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3406 /* collect port action mask recorded in dev actions */
3407 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3408 ~ATA_EH_PERDEV_MASK;
3409 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3411 /* process hotplug request */
3412 if (dev->flags & ATA_DFLAG_DETACH)
3413 ata_eh_detach_dev(dev);
3415 /* schedule probe if necessary */
3416 if (!ata_dev_enabled(dev))
3417 ata_eh_schedule_probe(dev);
3421 retry:
3422 rc = 0;
3423 nr_failed_devs = 0;
3425 /* if UNLOADING, finish immediately */
3426 if (ap->pflags & ATA_PFLAG_UNLOADING)
3427 goto out;
3429 /* prep for EH */
3430 ata_for_each_link(link, ap, EDGE) {
3431 struct ata_eh_context *ehc = &link->eh_context;
3433 /* skip EH if possible. */
3434 if (ata_eh_skip_recovery(link))
3435 ehc->i.action = 0;
3437 ata_for_each_dev(dev, link, ALL)
3438 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3441 /* reset */
3442 ata_for_each_link(link, ap, EDGE) {
3443 struct ata_eh_context *ehc = &link->eh_context;
3445 if (!(ehc->i.action & ATA_EH_RESET))
3446 continue;
3448 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3449 prereset, softreset, hardreset, postreset);
3450 if (rc) {
3451 ata_link_printk(link, KERN_ERR,
3452 "reset failed, giving up\n");
3453 goto out;
3457 do {
3458 unsigned long now;
3461 * clears ATA_EH_PARK in eh_info and resets
3462 * ap->park_req_pending
3464 ata_eh_pull_park_action(ap);
3466 deadline = jiffies;
3467 ata_for_each_link(link, ap, EDGE) {
3468 ata_for_each_dev(dev, link, ALL) {
3469 struct ata_eh_context *ehc = &link->eh_context;
3470 unsigned long tmp;
3472 if (dev->class != ATA_DEV_ATA)
3473 continue;
3474 if (!(ehc->i.dev_action[dev->devno] &
3475 ATA_EH_PARK))
3476 continue;
3477 tmp = dev->unpark_deadline;
3478 if (time_before(deadline, tmp))
3479 deadline = tmp;
3480 else if (time_before_eq(tmp, jiffies))
3481 continue;
3482 if (ehc->unloaded_mask & (1 << dev->devno))
3483 continue;
3485 ata_eh_park_issue_cmd(dev, 1);
3489 now = jiffies;
3490 if (time_before_eq(deadline, now))
3491 break;
3493 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3494 deadline - now);
3495 } while (deadline);
3496 ata_for_each_link(link, ap, EDGE) {
3497 ata_for_each_dev(dev, link, ALL) {
3498 if (!(link->eh_context.unloaded_mask &
3499 (1 << dev->devno)))
3500 continue;
3502 ata_eh_park_issue_cmd(dev, 0);
3503 ata_eh_done(link, dev, ATA_EH_PARK);
3507 /* the rest */
3508 ata_for_each_link(link, ap, EDGE) {
3509 struct ata_eh_context *ehc = &link->eh_context;
3511 /* revalidate existing devices and attach new ones */
3512 rc = ata_eh_revalidate_and_attach(link, &dev);
3513 if (rc)
3514 goto dev_fail;
3516 /* if PMP got attached, return, pmp EH will take care of it */
3517 if (link->device->class == ATA_DEV_PMP) {
3518 ehc->i.action = 0;
3519 return 0;
3522 /* configure transfer mode if necessary */
3523 if (ehc->i.flags & ATA_EHI_SETMODE) {
3524 rc = ata_set_mode(link, &dev);
3525 if (rc)
3526 goto dev_fail;
3527 ehc->i.flags &= ~ATA_EHI_SETMODE;
3530 /* If reset has been issued, clear UA to avoid
3531 * disrupting the current users of the device.
3533 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3534 ata_for_each_dev(dev, link, ALL) {
3535 if (dev->class != ATA_DEV_ATAPI)
3536 continue;
3537 rc = atapi_eh_clear_ua(dev);
3538 if (rc)
3539 goto dev_fail;
3543 /* retry flush if necessary */
3544 ata_for_each_dev(dev, link, ALL) {
3545 if (dev->class != ATA_DEV_ATA)
3546 continue;
3547 rc = ata_eh_maybe_retry_flush(dev);
3548 if (rc)
3549 goto dev_fail;
3552 /* configure link power saving */
3553 if (ehc->i.action & ATA_EH_LPM)
3554 ata_for_each_dev(dev, link, ALL)
3555 ata_dev_enable_pm(dev, ap->pm_policy);
3557 /* this link is okay now */
3558 ehc->i.flags = 0;
3559 continue;
3561 dev_fail:
3562 nr_failed_devs++;
3563 ata_eh_handle_dev_fail(dev, rc);
3565 if (ap->pflags & ATA_PFLAG_FROZEN) {
3566 /* PMP reset requires working host port.
3567 * Can't retry if it's frozen.
3569 if (sata_pmp_attached(ap))
3570 goto out;
3571 break;
3575 if (nr_failed_devs)
3576 goto retry;
3578 out:
3579 if (rc && r_failed_link)
3580 *r_failed_link = link;
3582 DPRINTK("EXIT, rc=%d\n", rc);
3583 return rc;
3587 * ata_eh_finish - finish up EH
3588 * @ap: host port to finish EH for
3590 * Recovery is complete. Clean up EH states and retry or finish
3591 * failed qcs.
3593 * LOCKING:
3594 * None.
3596 void ata_eh_finish(struct ata_port *ap)
3598 int tag;
3600 /* retry or finish qcs */
3601 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3602 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3604 if (!(qc->flags & ATA_QCFLAG_FAILED))
3605 continue;
3607 if (qc->err_mask) {
3608 /* FIXME: Once EH migration is complete,
3609 * generate sense data in this function,
3610 * considering both err_mask and tf.
3612 if (qc->flags & ATA_QCFLAG_RETRY)
3613 ata_eh_qc_retry(qc);
3614 else
3615 ata_eh_qc_complete(qc);
3616 } else {
3617 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3618 ata_eh_qc_complete(qc);
3619 } else {
3620 /* feed zero TF to sense generation */
3621 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3622 ata_eh_qc_retry(qc);
3627 /* make sure nr_active_links is zero after EH */
3628 WARN_ON(ap->nr_active_links);
3629 ap->nr_active_links = 0;
3633 * ata_do_eh - do standard error handling
3634 * @ap: host port to handle error for
3636 * @prereset: prereset method (can be NULL)
3637 * @softreset: softreset method (can be NULL)
3638 * @hardreset: hardreset method (can be NULL)
3639 * @postreset: postreset method (can be NULL)
3641 * Perform standard error handling sequence.
3643 * LOCKING:
3644 * Kernel thread context (may sleep).
3646 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3647 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3648 ata_postreset_fn_t postreset)
3650 struct ata_device *dev;
3651 int rc;
3653 ata_eh_autopsy(ap);
3654 ata_eh_report(ap);
3656 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3657 NULL);
3658 if (rc) {
3659 ata_for_each_dev(dev, &ap->link, ALL)
3660 ata_dev_disable(dev);
3663 ata_eh_finish(ap);
3667 * ata_std_error_handler - standard error handler
3668 * @ap: host port to handle error for
3670 * Standard error handler
3672 * LOCKING:
3673 * Kernel thread context (may sleep).
3675 void ata_std_error_handler(struct ata_port *ap)
3677 struct ata_port_operations *ops = ap->ops;
3678 ata_reset_fn_t hardreset = ops->hardreset;
3680 /* ignore built-in hardreset if SCR access is not available */
3681 if (ata_is_builtin_hardreset(hardreset) && !sata_scr_valid(&ap->link))
3682 hardreset = NULL;
3684 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3687 #ifdef CONFIG_PM
3689 * ata_eh_handle_port_suspend - perform port suspend operation
3690 * @ap: port to suspend
3692 * Suspend @ap.
3694 * LOCKING:
3695 * Kernel thread context (may sleep).
3697 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3699 unsigned long flags;
3700 int rc = 0;
3702 /* are we suspending? */
3703 spin_lock_irqsave(ap->lock, flags);
3704 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3705 ap->pm_mesg.event == PM_EVENT_ON) {
3706 spin_unlock_irqrestore(ap->lock, flags);
3707 return;
3709 spin_unlock_irqrestore(ap->lock, flags);
3711 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3713 /* tell ACPI we're suspending */
3714 rc = ata_acpi_on_suspend(ap);
3715 if (rc)
3716 goto out;
3718 /* suspend */
3719 ata_eh_freeze_port(ap);
3721 if (ap->ops->port_suspend)
3722 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3724 ata_acpi_set_state(ap, PMSG_SUSPEND);
3725 out:
3726 /* report result */
3727 spin_lock_irqsave(ap->lock, flags);
3729 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3730 if (rc == 0)
3731 ap->pflags |= ATA_PFLAG_SUSPENDED;
3732 else if (ap->pflags & ATA_PFLAG_FROZEN)
3733 ata_port_schedule_eh(ap);
3735 if (ap->pm_result) {
3736 *ap->pm_result = rc;
3737 ap->pm_result = NULL;
3740 spin_unlock_irqrestore(ap->lock, flags);
3742 return;
3746 * ata_eh_handle_port_resume - perform port resume operation
3747 * @ap: port to resume
3749 * Resume @ap.
3751 * LOCKING:
3752 * Kernel thread context (may sleep).
3754 static void ata_eh_handle_port_resume(struct ata_port *ap)
3756 struct ata_link *link;
3757 struct ata_device *dev;
3758 unsigned long flags;
3759 int rc = 0;
3761 /* are we resuming? */
3762 spin_lock_irqsave(ap->lock, flags);
3763 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3764 ap->pm_mesg.event != PM_EVENT_ON) {
3765 spin_unlock_irqrestore(ap->lock, flags);
3766 return;
3768 spin_unlock_irqrestore(ap->lock, flags);
3770 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3773 * Error timestamps are in jiffies which doesn't run while
3774 * suspended and PHY events during resume isn't too uncommon.
3775 * When the two are combined, it can lead to unnecessary speed
3776 * downs if the machine is suspended and resumed repeatedly.
3777 * Clear error history.
3779 ata_for_each_link(link, ap, HOST_FIRST)
3780 ata_for_each_dev(dev, link, ALL)
3781 ata_ering_clear(&dev->ering);
3783 ata_acpi_set_state(ap, PMSG_ON);
3785 if (ap->ops->port_resume)
3786 rc = ap->ops->port_resume(ap);
3788 /* tell ACPI that we're resuming */
3789 ata_acpi_on_resume(ap);
3791 /* report result */
3792 spin_lock_irqsave(ap->lock, flags);
3793 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3794 if (ap->pm_result) {
3795 *ap->pm_result = rc;
3796 ap->pm_result = NULL;
3798 spin_unlock_irqrestore(ap->lock, flags);
3800 #endif /* CONFIG_PM */