ext3: Add journal error check in resize.c
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / ata / libata-eh.c
blob17a637877d0311abb7c95c7cd6e2359be4a2c70b
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),
60 ATA_EFLAG_OLD_ER = (1 << 31),
62 /* error categories */
63 ATA_ECAT_NONE = 0,
64 ATA_ECAT_ATA_BUS = 1,
65 ATA_ECAT_TOUT_HSM = 2,
66 ATA_ECAT_UNK_DEV = 3,
67 ATA_ECAT_DUBIOUS_NONE = 4,
68 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
69 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
70 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_ECAT_NR = 8,
73 ATA_EH_CMD_DFL_TIMEOUT = 5000,
75 /* always put at least this amount of time between resets */
76 ATA_EH_RESET_COOL_DOWN = 5000,
78 /* Waiting in ->prereset can never be reliable. It's
79 * sometimes nice to wait there but it can't be depended upon;
80 * otherwise, we wouldn't be resetting. Just give it enough
81 * time for most drives to spin up.
83 ATA_EH_PRERESET_TIMEOUT = 10000,
84 ATA_EH_FASTDRAIN_INTERVAL = 3000,
86 ATA_EH_UA_TRIES = 5,
88 /* probe speed down parameters, see ata_eh_schedule_probe() */
89 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
90 ATA_EH_PROBE_TRIALS = 2,
93 /* The following table determines how we sequence resets. Each entry
94 * represents timeout for that try. The first try can be soft or
95 * hardreset. All others are hardreset if available. In most cases
96 * the first reset w/ 10sec timeout should succeed. Following entries
97 * are mostly for error handling, hotplug and retarded devices.
99 static const unsigned long ata_eh_reset_timeouts[] = {
100 10000, /* most drives spin up by 10sec */
101 10000, /* > 99% working drives spin up before 20sec */
102 35000, /* give > 30 secs of idleness for retarded devices */
103 5000, /* and sweet one last chance */
104 ULONG_MAX, /* > 1 min has elapsed, give up */
107 static const unsigned long ata_eh_identify_timeouts[] = {
108 5000, /* covers > 99% of successes and not too boring on failures */
109 10000, /* combined time till here is enough even for media access */
110 30000, /* for true idiots */
111 ULONG_MAX,
114 static const unsigned long ata_eh_flush_timeouts[] = {
115 15000, /* be generous with flush */
116 15000, /* ditto */
117 30000, /* and even more generous */
118 ULONG_MAX,
121 static const unsigned long ata_eh_other_timeouts[] = {
122 5000, /* same rationale as identify timeout */
123 10000, /* ditto */
124 /* but no merciful 30sec for other commands, it just isn't worth it */
125 ULONG_MAX,
128 struct ata_eh_cmd_timeout_ent {
129 const u8 *commands;
130 const unsigned long *timeouts;
133 /* The following table determines timeouts to use for EH internal
134 * commands. Each table entry is a command class and matches the
135 * commands the entry applies to and the timeout table to use.
137 * On the retry after a command timed out, the next timeout value from
138 * the table is used. If the table doesn't contain further entries,
139 * the last value is used.
141 * ehc->cmd_timeout_idx keeps track of which timeout to use per
142 * command class, so if SET_FEATURES times out on the first try, the
143 * next try will use the second timeout value only for that class.
145 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
146 static const struct ata_eh_cmd_timeout_ent
147 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
148 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
149 .timeouts = ata_eh_identify_timeouts, },
150 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
151 .timeouts = ata_eh_other_timeouts, },
152 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_SET_FEATURES),
155 .timeouts = ata_eh_other_timeouts, },
156 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
157 .timeouts = ata_eh_other_timeouts, },
158 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
159 .timeouts = ata_eh_flush_timeouts },
161 #undef CMDS
163 static void __ata_port_freeze(struct ata_port *ap);
164 #ifdef CONFIG_PM
165 static void ata_eh_handle_port_suspend(struct ata_port *ap);
166 static void ata_eh_handle_port_resume(struct ata_port *ap);
167 #else /* CONFIG_PM */
168 static void ata_eh_handle_port_suspend(struct ata_port *ap)
171 static void ata_eh_handle_port_resume(struct ata_port *ap)
173 #endif /* CONFIG_PM */
175 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
176 va_list args)
178 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
179 ATA_EH_DESC_LEN - ehi->desc_len,
180 fmt, args);
184 * __ata_ehi_push_desc - push error description without adding separator
185 * @ehi: target EHI
186 * @fmt: printf format string
188 * Format string according to @fmt and append it to @ehi->desc.
190 * LOCKING:
191 * spin_lock_irqsave(host lock)
193 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
195 va_list args;
197 va_start(args, fmt);
198 __ata_ehi_pushv_desc(ehi, fmt, args);
199 va_end(args);
203 * ata_ehi_push_desc - push error description with separator
204 * @ehi: target EHI
205 * @fmt: printf format string
207 * Format string according to @fmt and append it to @ehi->desc.
208 * If @ehi->desc is not empty, ", " is added in-between.
210 * LOCKING:
211 * spin_lock_irqsave(host lock)
213 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
215 va_list args;
217 if (ehi->desc_len)
218 __ata_ehi_push_desc(ehi, ", ");
220 va_start(args, fmt);
221 __ata_ehi_pushv_desc(ehi, fmt, args);
222 va_end(args);
226 * ata_ehi_clear_desc - clean error description
227 * @ehi: target EHI
229 * Clear @ehi->desc.
231 * LOCKING:
232 * spin_lock_irqsave(host lock)
234 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
236 ehi->desc[0] = '\0';
237 ehi->desc_len = 0;
241 * ata_port_desc - append port description
242 * @ap: target ATA port
243 * @fmt: printf format string
245 * Format string according to @fmt and append it to port
246 * description. If port description is not empty, " " is added
247 * in-between. This function is to be used while initializing
248 * ata_host. The description is printed on host registration.
250 * LOCKING:
251 * None.
253 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
255 va_list args;
257 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
259 if (ap->link.eh_info.desc_len)
260 __ata_ehi_push_desc(&ap->link.eh_info, " ");
262 va_start(args, fmt);
263 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
264 va_end(args);
267 #ifdef CONFIG_PCI
270 * ata_port_pbar_desc - append PCI BAR description
271 * @ap: target ATA port
272 * @bar: target PCI BAR
273 * @offset: offset into PCI BAR
274 * @name: name of the area
276 * If @offset is negative, this function formats a string which
277 * contains the name, address, size and type of the BAR and
278 * appends it to the port description. If @offset is zero or
279 * positive, only name and offsetted address is appended.
281 * LOCKING:
282 * None.
284 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
285 const char *name)
287 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
288 char *type = "";
289 unsigned long long start, len;
291 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
292 type = "m";
293 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
294 type = "i";
296 start = (unsigned long long)pci_resource_start(pdev, bar);
297 len = (unsigned long long)pci_resource_len(pdev, bar);
299 if (offset < 0)
300 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
301 else
302 ata_port_desc(ap, "%s 0x%llx", name,
303 start + (unsigned long long)offset);
306 #endif /* CONFIG_PCI */
308 static int ata_lookup_timeout_table(u8 cmd)
310 int i;
312 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
313 const u8 *cur;
315 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
316 if (*cur == cmd)
317 return i;
320 return -1;
324 * ata_internal_cmd_timeout - determine timeout for an internal command
325 * @dev: target device
326 * @cmd: internal command to be issued
328 * Determine timeout for internal command @cmd for @dev.
330 * LOCKING:
331 * EH context.
333 * RETURNS:
334 * Determined timeout.
336 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
338 struct ata_eh_context *ehc = &dev->link->eh_context;
339 int ent = ata_lookup_timeout_table(cmd);
340 int idx;
342 if (ent < 0)
343 return ATA_EH_CMD_DFL_TIMEOUT;
345 idx = ehc->cmd_timeout_idx[dev->devno][ent];
346 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
350 * ata_internal_cmd_timed_out - notification for internal command timeout
351 * @dev: target device
352 * @cmd: internal command which timed out
354 * Notify EH that internal command @cmd for @dev timed out. This
355 * function should be called only for commands whose timeouts are
356 * determined using ata_internal_cmd_timeout().
358 * LOCKING:
359 * EH context.
361 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
363 struct ata_eh_context *ehc = &dev->link->eh_context;
364 int ent = ata_lookup_timeout_table(cmd);
365 int idx;
367 if (ent < 0)
368 return;
370 idx = ehc->cmd_timeout_idx[dev->devno][ent];
371 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
372 ehc->cmd_timeout_idx[dev->devno][ent]++;
375 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
376 unsigned int err_mask)
378 struct ata_ering_entry *ent;
380 WARN_ON(!err_mask);
382 ering->cursor++;
383 ering->cursor %= ATA_ERING_SIZE;
385 ent = &ering->ring[ering->cursor];
386 ent->eflags = eflags;
387 ent->err_mask = err_mask;
388 ent->timestamp = get_jiffies_64();
391 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
393 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
395 if (ent->err_mask)
396 return ent;
397 return NULL;
400 int ata_ering_map(struct ata_ering *ering,
401 int (*map_fn)(struct ata_ering_entry *, void *),
402 void *arg)
404 int idx, rc = 0;
405 struct ata_ering_entry *ent;
407 idx = ering->cursor;
408 do {
409 ent = &ering->ring[idx];
410 if (!ent->err_mask)
411 break;
412 rc = map_fn(ent, arg);
413 if (rc)
414 break;
415 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
416 } while (idx != ering->cursor);
418 return rc;
421 int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
423 ent->eflags |= ATA_EFLAG_OLD_ER;
424 return 0;
427 static void ata_ering_clear(struct ata_ering *ering)
429 ata_ering_map(ering, ata_ering_clear_cb, NULL);
432 static unsigned int ata_eh_dev_action(struct ata_device *dev)
434 struct ata_eh_context *ehc = &dev->link->eh_context;
436 return ehc->i.action | ehc->i.dev_action[dev->devno];
439 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
440 struct ata_eh_info *ehi, unsigned int action)
442 struct ata_device *tdev;
444 if (!dev) {
445 ehi->action &= ~action;
446 ata_for_each_dev(tdev, link, ALL)
447 ehi->dev_action[tdev->devno] &= ~action;
448 } else {
449 /* doesn't make sense for port-wide EH actions */
450 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
452 /* break ehi->action into ehi->dev_action */
453 if (ehi->action & action) {
454 ata_for_each_dev(tdev, link, ALL)
455 ehi->dev_action[tdev->devno] |=
456 ehi->action & action;
457 ehi->action &= ~action;
460 /* turn off the specified per-dev action */
461 ehi->dev_action[dev->devno] &= ~action;
466 * ata_eh_acquire - acquire EH ownership
467 * @ap: ATA port to acquire EH ownership for
469 * Acquire EH ownership for @ap. This is the basic exclusion
470 * mechanism for ports sharing a host. Only one port hanging off
471 * the same host can claim the ownership of EH.
473 * LOCKING:
474 * EH context.
476 void ata_eh_acquire(struct ata_port *ap)
478 mutex_lock(&ap->host->eh_mutex);
479 WARN_ON_ONCE(ap->host->eh_owner);
480 ap->host->eh_owner = current;
484 * ata_eh_release - release EH ownership
485 * @ap: ATA port to release EH ownership for
487 * Release EH ownership for @ap if the caller. The caller must
488 * have acquired EH ownership using ata_eh_acquire() previously.
490 * LOCKING:
491 * EH context.
493 void ata_eh_release(struct ata_port *ap)
495 WARN_ON_ONCE(ap->host->eh_owner != current);
496 ap->host->eh_owner = NULL;
497 mutex_unlock(&ap->host->eh_mutex);
501 * ata_scsi_timed_out - SCSI layer time out callback
502 * @cmd: timed out SCSI command
504 * Handles SCSI layer timeout. We race with normal completion of
505 * the qc for @cmd. If the qc is already gone, we lose and let
506 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
507 * timed out and EH should be invoked. Prevent ata_qc_complete()
508 * from finishing it by setting EH_SCHEDULED and return
509 * EH_NOT_HANDLED.
511 * TODO: kill this function once old EH is gone.
513 * LOCKING:
514 * Called from timer context
516 * RETURNS:
517 * EH_HANDLED or EH_NOT_HANDLED
519 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
521 struct Scsi_Host *host = cmd->device->host;
522 struct ata_port *ap = ata_shost_to_port(host);
523 unsigned long flags;
524 struct ata_queued_cmd *qc;
525 enum blk_eh_timer_return ret;
527 DPRINTK("ENTER\n");
529 if (ap->ops->error_handler) {
530 ret = BLK_EH_NOT_HANDLED;
531 goto out;
534 ret = BLK_EH_HANDLED;
535 spin_lock_irqsave(ap->lock, flags);
536 qc = ata_qc_from_tag(ap, ap->link.active_tag);
537 if (qc) {
538 WARN_ON(qc->scsicmd != cmd);
539 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
540 qc->err_mask |= AC_ERR_TIMEOUT;
541 ret = BLK_EH_NOT_HANDLED;
543 spin_unlock_irqrestore(ap->lock, flags);
545 out:
546 DPRINTK("EXIT, ret=%d\n", ret);
547 return ret;
550 static void ata_eh_unload(struct ata_port *ap)
552 struct ata_link *link;
553 struct ata_device *dev;
554 unsigned long flags;
556 /* Restore SControl IPM and SPD for the next driver and
557 * disable attached devices.
559 ata_for_each_link(link, ap, PMP_FIRST) {
560 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
561 ata_for_each_dev(dev, link, ALL)
562 ata_dev_disable(dev);
565 /* freeze and set UNLOADED */
566 spin_lock_irqsave(ap->lock, flags);
568 ata_port_freeze(ap); /* won't be thawed */
569 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
570 ap->pflags |= ATA_PFLAG_UNLOADED;
572 spin_unlock_irqrestore(ap->lock, flags);
576 * ata_scsi_error - SCSI layer error handler callback
577 * @host: SCSI host on which error occurred
579 * Handles SCSI-layer-thrown error events.
581 * LOCKING:
582 * Inherited from SCSI layer (none, can sleep)
584 * RETURNS:
585 * Zero.
587 void ata_scsi_error(struct Scsi_Host *host)
589 struct ata_port *ap = ata_shost_to_port(host);
590 int i;
591 unsigned long flags;
593 DPRINTK("ENTER\n");
595 /* make sure sff pio task is not running */
596 ata_sff_flush_pio_task(ap);
598 /* synchronize with host lock and sort out timeouts */
600 /* For new EH, all qcs are finished in one of three ways -
601 * normal completion, error completion, and SCSI timeout.
602 * Both completions can race against SCSI timeout. When normal
603 * completion wins, the qc never reaches EH. When error
604 * completion wins, the qc has ATA_QCFLAG_FAILED set.
606 * When SCSI timeout wins, things are a bit more complex.
607 * Normal or error completion can occur after the timeout but
608 * before this point. In such cases, both types of
609 * completions are honored. A scmd is determined to have
610 * timed out iff its associated qc is active and not failed.
612 if (ap->ops->error_handler) {
613 struct scsi_cmnd *scmd, *tmp;
614 int nr_timedout = 0;
616 spin_lock_irqsave(ap->lock, flags);
618 /* This must occur under the ap->lock as we don't want
619 a polled recovery to race the real interrupt handler
621 The lost_interrupt handler checks for any completed but
622 non-notified command and completes much like an IRQ handler.
624 We then fall into the error recovery code which will treat
625 this as if normal completion won the race */
627 if (ap->ops->lost_interrupt)
628 ap->ops->lost_interrupt(ap);
630 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
631 struct ata_queued_cmd *qc;
633 for (i = 0; i < ATA_MAX_QUEUE; i++) {
634 qc = __ata_qc_from_tag(ap, i);
635 if (qc->flags & ATA_QCFLAG_ACTIVE &&
636 qc->scsicmd == scmd)
637 break;
640 if (i < ATA_MAX_QUEUE) {
641 /* the scmd has an associated qc */
642 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
643 /* which hasn't failed yet, timeout */
644 qc->err_mask |= AC_ERR_TIMEOUT;
645 qc->flags |= ATA_QCFLAG_FAILED;
646 nr_timedout++;
648 } else {
649 /* Normal completion occurred after
650 * SCSI timeout but before this point.
651 * Successfully complete it.
653 scmd->retries = scmd->allowed;
654 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
658 /* If we have timed out qcs. They belong to EH from
659 * this point but the state of the controller is
660 * unknown. Freeze the port to make sure the IRQ
661 * handler doesn't diddle with those qcs. This must
662 * be done atomically w.r.t. setting QCFLAG_FAILED.
664 if (nr_timedout)
665 __ata_port_freeze(ap);
667 spin_unlock_irqrestore(ap->lock, flags);
669 /* initialize eh_tries */
670 ap->eh_tries = ATA_EH_MAX_TRIES;
671 } else
672 spin_unlock_wait(ap->lock);
674 /* If we timed raced normal completion and there is nothing to
675 recover nr_timedout == 0 why exactly are we doing error recovery ? */
677 /* invoke error handler */
678 if (ap->ops->error_handler) {
679 struct ata_link *link;
681 /* acquire EH ownership */
682 ata_eh_acquire(ap);
683 repeat:
684 /* kill fast drain timer */
685 del_timer_sync(&ap->fastdrain_timer);
687 /* process port resume request */
688 ata_eh_handle_port_resume(ap);
690 /* fetch & clear EH info */
691 spin_lock_irqsave(ap->lock, flags);
693 ata_for_each_link(link, ap, HOST_FIRST) {
694 struct ata_eh_context *ehc = &link->eh_context;
695 struct ata_device *dev;
697 memset(&link->eh_context, 0, sizeof(link->eh_context));
698 link->eh_context.i = link->eh_info;
699 memset(&link->eh_info, 0, sizeof(link->eh_info));
701 ata_for_each_dev(dev, link, ENABLED) {
702 int devno = dev->devno;
704 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
705 if (ata_ncq_enabled(dev))
706 ehc->saved_ncq_enabled |= 1 << devno;
710 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
711 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
712 ap->excl_link = NULL; /* don't maintain exclusion over EH */
714 spin_unlock_irqrestore(ap->lock, flags);
716 /* invoke EH, skip if unloading or suspended */
717 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
718 ap->ops->error_handler(ap);
719 else {
720 /* if unloading, commence suicide */
721 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
722 !(ap->pflags & ATA_PFLAG_UNLOADED))
723 ata_eh_unload(ap);
724 ata_eh_finish(ap);
727 /* process port suspend request */
728 ata_eh_handle_port_suspend(ap);
730 /* Exception might have happend after ->error_handler
731 * recovered the port but before this point. Repeat
732 * EH in such case.
734 spin_lock_irqsave(ap->lock, flags);
736 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
737 if (--ap->eh_tries) {
738 spin_unlock_irqrestore(ap->lock, flags);
739 goto repeat;
741 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
742 "tries, giving up\n", ATA_EH_MAX_TRIES);
743 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
746 /* this run is complete, make sure EH info is clear */
747 ata_for_each_link(link, ap, HOST_FIRST)
748 memset(&link->eh_info, 0, sizeof(link->eh_info));
750 /* Clear host_eh_scheduled while holding ap->lock such
751 * that if exception occurs after this point but
752 * before EH completion, SCSI midlayer will
753 * re-initiate EH.
755 host->host_eh_scheduled = 0;
757 spin_unlock_irqrestore(ap->lock, flags);
758 ata_eh_release(ap);
759 } else {
760 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
761 ap->ops->eng_timeout(ap);
764 /* finish or retry handled scmd's and clean up */
765 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
767 scsi_eh_flush_done_q(&ap->eh_done_q);
769 /* clean up */
770 spin_lock_irqsave(ap->lock, flags);
772 if (ap->pflags & ATA_PFLAG_LOADING)
773 ap->pflags &= ~ATA_PFLAG_LOADING;
774 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
775 schedule_delayed_work(&ap->hotplug_task, 0);
777 if (ap->pflags & ATA_PFLAG_RECOVERED)
778 ata_port_printk(ap, KERN_INFO, "EH complete\n");
780 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
782 /* tell wait_eh that we're done */
783 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
784 wake_up_all(&ap->eh_wait_q);
786 spin_unlock_irqrestore(ap->lock, flags);
788 DPRINTK("EXIT\n");
792 * ata_port_wait_eh - Wait for the currently pending EH to complete
793 * @ap: Port to wait EH for
795 * Wait until the currently pending EH is complete.
797 * LOCKING:
798 * Kernel thread context (may sleep).
800 void ata_port_wait_eh(struct ata_port *ap)
802 unsigned long flags;
803 DEFINE_WAIT(wait);
805 retry:
806 spin_lock_irqsave(ap->lock, flags);
808 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
809 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
810 spin_unlock_irqrestore(ap->lock, flags);
811 schedule();
812 spin_lock_irqsave(ap->lock, flags);
814 finish_wait(&ap->eh_wait_q, &wait);
816 spin_unlock_irqrestore(ap->lock, flags);
818 /* make sure SCSI EH is complete */
819 if (scsi_host_in_recovery(ap->scsi_host)) {
820 ata_msleep(ap, 10);
821 goto retry;
825 static int ata_eh_nr_in_flight(struct ata_port *ap)
827 unsigned int tag;
828 int nr = 0;
830 /* count only non-internal commands */
831 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
832 if (ata_qc_from_tag(ap, tag))
833 nr++;
835 return nr;
838 void ata_eh_fastdrain_timerfn(unsigned long arg)
840 struct ata_port *ap = (void *)arg;
841 unsigned long flags;
842 int cnt;
844 spin_lock_irqsave(ap->lock, flags);
846 cnt = ata_eh_nr_in_flight(ap);
848 /* are we done? */
849 if (!cnt)
850 goto out_unlock;
852 if (cnt == ap->fastdrain_cnt) {
853 unsigned int tag;
855 /* No progress during the last interval, tag all
856 * in-flight qcs as timed out and freeze the port.
858 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
859 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
860 if (qc)
861 qc->err_mask |= AC_ERR_TIMEOUT;
864 ata_port_freeze(ap);
865 } else {
866 /* some qcs have finished, give it another chance */
867 ap->fastdrain_cnt = cnt;
868 ap->fastdrain_timer.expires =
869 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
870 add_timer(&ap->fastdrain_timer);
873 out_unlock:
874 spin_unlock_irqrestore(ap->lock, flags);
878 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
879 * @ap: target ATA port
880 * @fastdrain: activate fast drain
882 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
883 * is non-zero and EH wasn't pending before. Fast drain ensures
884 * that EH kicks in in timely manner.
886 * LOCKING:
887 * spin_lock_irqsave(host lock)
889 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
891 int cnt;
893 /* already scheduled? */
894 if (ap->pflags & ATA_PFLAG_EH_PENDING)
895 return;
897 ap->pflags |= ATA_PFLAG_EH_PENDING;
899 if (!fastdrain)
900 return;
902 /* do we have in-flight qcs? */
903 cnt = ata_eh_nr_in_flight(ap);
904 if (!cnt)
905 return;
907 /* activate fast drain */
908 ap->fastdrain_cnt = cnt;
909 ap->fastdrain_timer.expires =
910 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
911 add_timer(&ap->fastdrain_timer);
915 * ata_qc_schedule_eh - schedule qc for error handling
916 * @qc: command to schedule error handling for
918 * Schedule error handling for @qc. EH will kick in as soon as
919 * other commands are drained.
921 * LOCKING:
922 * spin_lock_irqsave(host lock)
924 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
926 struct ata_port *ap = qc->ap;
927 struct request_queue *q = qc->scsicmd->device->request_queue;
928 unsigned long flags;
930 WARN_ON(!ap->ops->error_handler);
932 qc->flags |= ATA_QCFLAG_FAILED;
933 ata_eh_set_pending(ap, 1);
935 /* The following will fail if timeout has already expired.
936 * ata_scsi_error() takes care of such scmds on EH entry.
937 * Note that ATA_QCFLAG_FAILED is unconditionally set after
938 * this function completes.
940 spin_lock_irqsave(q->queue_lock, flags);
941 blk_abort_request(qc->scsicmd->request);
942 spin_unlock_irqrestore(q->queue_lock, flags);
946 * ata_port_schedule_eh - schedule error handling without a qc
947 * @ap: ATA port to schedule EH for
949 * Schedule error handling for @ap. EH will kick in as soon as
950 * all commands are drained.
952 * LOCKING:
953 * spin_lock_irqsave(host lock)
955 void ata_port_schedule_eh(struct ata_port *ap)
957 WARN_ON(!ap->ops->error_handler);
959 if (ap->pflags & ATA_PFLAG_INITIALIZING)
960 return;
962 ata_eh_set_pending(ap, 1);
963 scsi_schedule_eh(ap->scsi_host);
965 DPRINTK("port EH scheduled\n");
968 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
970 int tag, nr_aborted = 0;
972 WARN_ON(!ap->ops->error_handler);
974 /* we're gonna abort all commands, no need for fast drain */
975 ata_eh_set_pending(ap, 0);
977 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
978 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
980 if (qc && (!link || qc->dev->link == link)) {
981 qc->flags |= ATA_QCFLAG_FAILED;
982 ata_qc_complete(qc);
983 nr_aborted++;
987 if (!nr_aborted)
988 ata_port_schedule_eh(ap);
990 return nr_aborted;
994 * ata_link_abort - abort all qc's on the link
995 * @link: ATA link to abort qc's for
997 * Abort all active qc's active on @link and schedule EH.
999 * LOCKING:
1000 * spin_lock_irqsave(host lock)
1002 * RETURNS:
1003 * Number of aborted qc's.
1005 int ata_link_abort(struct ata_link *link)
1007 return ata_do_link_abort(link->ap, link);
1011 * ata_port_abort - abort all qc's on the port
1012 * @ap: ATA port to abort qc's for
1014 * Abort all active qc's of @ap and schedule EH.
1016 * LOCKING:
1017 * spin_lock_irqsave(host_set lock)
1019 * RETURNS:
1020 * Number of aborted qc's.
1022 int ata_port_abort(struct ata_port *ap)
1024 return ata_do_link_abort(ap, NULL);
1028 * __ata_port_freeze - freeze port
1029 * @ap: ATA port to freeze
1031 * This function is called when HSM violation or some other
1032 * condition disrupts normal operation of the port. Frozen port
1033 * is not allowed to perform any operation until the port is
1034 * thawed, which usually follows a successful reset.
1036 * ap->ops->freeze() callback can be used for freezing the port
1037 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1038 * port cannot be frozen hardware-wise, the interrupt handler
1039 * must ack and clear interrupts unconditionally while the port
1040 * is frozen.
1042 * LOCKING:
1043 * spin_lock_irqsave(host lock)
1045 static void __ata_port_freeze(struct ata_port *ap)
1047 WARN_ON(!ap->ops->error_handler);
1049 if (ap->ops->freeze)
1050 ap->ops->freeze(ap);
1052 ap->pflags |= ATA_PFLAG_FROZEN;
1054 DPRINTK("ata%u port frozen\n", ap->print_id);
1058 * ata_port_freeze - abort & freeze port
1059 * @ap: ATA port to freeze
1061 * Abort and freeze @ap. The freeze operation must be called
1062 * first, because some hardware requires special operations
1063 * before the taskfile registers are accessible.
1065 * LOCKING:
1066 * spin_lock_irqsave(host lock)
1068 * RETURNS:
1069 * Number of aborted commands.
1071 int ata_port_freeze(struct ata_port *ap)
1073 int nr_aborted;
1075 WARN_ON(!ap->ops->error_handler);
1077 __ata_port_freeze(ap);
1078 nr_aborted = ata_port_abort(ap);
1080 return nr_aborted;
1084 * sata_async_notification - SATA async notification handler
1085 * @ap: ATA port where async notification is received
1087 * Handler to be called when async notification via SDB FIS is
1088 * received. This function schedules EH if necessary.
1090 * LOCKING:
1091 * spin_lock_irqsave(host lock)
1093 * RETURNS:
1094 * 1 if EH is scheduled, 0 otherwise.
1096 int sata_async_notification(struct ata_port *ap)
1098 u32 sntf;
1099 int rc;
1101 if (!(ap->flags & ATA_FLAG_AN))
1102 return 0;
1104 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1105 if (rc == 0)
1106 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1108 if (!sata_pmp_attached(ap) || rc) {
1109 /* PMP is not attached or SNTF is not available */
1110 if (!sata_pmp_attached(ap)) {
1111 /* PMP is not attached. Check whether ATAPI
1112 * AN is configured. If so, notify media
1113 * change.
1115 struct ata_device *dev = ap->link.device;
1117 if ((dev->class == ATA_DEV_ATAPI) &&
1118 (dev->flags & ATA_DFLAG_AN))
1119 ata_scsi_media_change_notify(dev);
1120 return 0;
1121 } else {
1122 /* PMP is attached but SNTF is not available.
1123 * ATAPI async media change notification is
1124 * not used. The PMP must be reporting PHY
1125 * status change, schedule EH.
1127 ata_port_schedule_eh(ap);
1128 return 1;
1130 } else {
1131 /* PMP is attached and SNTF is available */
1132 struct ata_link *link;
1134 /* check and notify ATAPI AN */
1135 ata_for_each_link(link, ap, EDGE) {
1136 if (!(sntf & (1 << link->pmp)))
1137 continue;
1139 if ((link->device->class == ATA_DEV_ATAPI) &&
1140 (link->device->flags & ATA_DFLAG_AN))
1141 ata_scsi_media_change_notify(link->device);
1144 /* If PMP is reporting that PHY status of some
1145 * downstream ports has changed, schedule EH.
1147 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1148 ata_port_schedule_eh(ap);
1149 return 1;
1152 return 0;
1157 * ata_eh_freeze_port - EH helper to freeze port
1158 * @ap: ATA port to freeze
1160 * Freeze @ap.
1162 * LOCKING:
1163 * None.
1165 void ata_eh_freeze_port(struct ata_port *ap)
1167 unsigned long flags;
1169 if (!ap->ops->error_handler)
1170 return;
1172 spin_lock_irqsave(ap->lock, flags);
1173 __ata_port_freeze(ap);
1174 spin_unlock_irqrestore(ap->lock, flags);
1178 * ata_port_thaw_port - EH helper to thaw port
1179 * @ap: ATA port to thaw
1181 * Thaw frozen port @ap.
1183 * LOCKING:
1184 * None.
1186 void ata_eh_thaw_port(struct ata_port *ap)
1188 unsigned long flags;
1190 if (!ap->ops->error_handler)
1191 return;
1193 spin_lock_irqsave(ap->lock, flags);
1195 ap->pflags &= ~ATA_PFLAG_FROZEN;
1197 if (ap->ops->thaw)
1198 ap->ops->thaw(ap);
1200 spin_unlock_irqrestore(ap->lock, flags);
1202 DPRINTK("ata%u port thawed\n", ap->print_id);
1205 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1207 /* nada */
1210 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1212 struct ata_port *ap = qc->ap;
1213 struct scsi_cmnd *scmd = qc->scsicmd;
1214 unsigned long flags;
1216 spin_lock_irqsave(ap->lock, flags);
1217 qc->scsidone = ata_eh_scsidone;
1218 __ata_qc_complete(qc);
1219 WARN_ON(ata_tag_valid(qc->tag));
1220 spin_unlock_irqrestore(ap->lock, flags);
1222 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1226 * ata_eh_qc_complete - Complete an active ATA command from EH
1227 * @qc: Command to complete
1229 * Indicate to the mid and upper layers that an ATA command has
1230 * completed. To be used from EH.
1232 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1234 struct scsi_cmnd *scmd = qc->scsicmd;
1235 scmd->retries = scmd->allowed;
1236 __ata_eh_qc_complete(qc);
1240 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1241 * @qc: Command to retry
1243 * Indicate to the mid and upper layers that an ATA command
1244 * should be retried. To be used from EH.
1246 * SCSI midlayer limits the number of retries to scmd->allowed.
1247 * scmd->retries is decremented for commands which get retried
1248 * due to unrelated failures (qc->err_mask is zero).
1250 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1252 struct scsi_cmnd *scmd = qc->scsicmd;
1253 if (!qc->err_mask && scmd->retries)
1254 scmd->retries--;
1255 __ata_eh_qc_complete(qc);
1259 * ata_dev_disable - disable ATA device
1260 * @dev: ATA device to disable
1262 * Disable @dev.
1264 * Locking:
1265 * EH context.
1267 void ata_dev_disable(struct ata_device *dev)
1269 if (!ata_dev_enabled(dev))
1270 return;
1272 if (ata_msg_drv(dev->link->ap))
1273 ata_dev_printk(dev, KERN_WARNING, "disabled\n");
1274 ata_acpi_on_disable(dev);
1275 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1276 dev->class++;
1278 /* From now till the next successful probe, ering is used to
1279 * track probe failures. Clear accumulated device error info.
1281 ata_ering_clear(&dev->ering);
1285 * ata_eh_detach_dev - detach ATA device
1286 * @dev: ATA device to detach
1288 * Detach @dev.
1290 * LOCKING:
1291 * None.
1293 void ata_eh_detach_dev(struct ata_device *dev)
1295 struct ata_link *link = dev->link;
1296 struct ata_port *ap = link->ap;
1297 struct ata_eh_context *ehc = &link->eh_context;
1298 unsigned long flags;
1300 ata_dev_disable(dev);
1302 spin_lock_irqsave(ap->lock, flags);
1304 dev->flags &= ~ATA_DFLAG_DETACH;
1306 if (ata_scsi_offline_dev(dev)) {
1307 dev->flags |= ATA_DFLAG_DETACHED;
1308 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1311 /* clear per-dev EH info */
1312 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1313 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1314 ehc->saved_xfer_mode[dev->devno] = 0;
1315 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1317 spin_unlock_irqrestore(ap->lock, flags);
1321 * ata_eh_about_to_do - about to perform eh_action
1322 * @link: target ATA link
1323 * @dev: target ATA dev for per-dev action (can be NULL)
1324 * @action: action about to be performed
1326 * Called just before performing EH actions to clear related bits
1327 * in @link->eh_info such that eh actions are not unnecessarily
1328 * repeated.
1330 * LOCKING:
1331 * None.
1333 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1334 unsigned int action)
1336 struct ata_port *ap = link->ap;
1337 struct ata_eh_info *ehi = &link->eh_info;
1338 struct ata_eh_context *ehc = &link->eh_context;
1339 unsigned long flags;
1341 spin_lock_irqsave(ap->lock, flags);
1343 ata_eh_clear_action(link, dev, ehi, action);
1345 /* About to take EH action, set RECOVERED. Ignore actions on
1346 * slave links as master will do them again.
1348 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1349 ap->pflags |= ATA_PFLAG_RECOVERED;
1351 spin_unlock_irqrestore(ap->lock, flags);
1355 * ata_eh_done - EH action complete
1356 * @ap: target ATA port
1357 * @dev: target ATA dev for per-dev action (can be NULL)
1358 * @action: action just completed
1360 * Called right after performing EH actions to clear related bits
1361 * in @link->eh_context.
1363 * LOCKING:
1364 * None.
1366 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1367 unsigned int action)
1369 struct ata_eh_context *ehc = &link->eh_context;
1371 ata_eh_clear_action(link, dev, &ehc->i, action);
1375 * ata_err_string - convert err_mask to descriptive string
1376 * @err_mask: error mask to convert to string
1378 * Convert @err_mask to descriptive string. Errors are
1379 * prioritized according to severity and only the most severe
1380 * error is reported.
1382 * LOCKING:
1383 * None.
1385 * RETURNS:
1386 * Descriptive string for @err_mask
1388 static const char *ata_err_string(unsigned int err_mask)
1390 if (err_mask & AC_ERR_HOST_BUS)
1391 return "host bus error";
1392 if (err_mask & AC_ERR_ATA_BUS)
1393 return "ATA bus error";
1394 if (err_mask & AC_ERR_TIMEOUT)
1395 return "timeout";
1396 if (err_mask & AC_ERR_HSM)
1397 return "HSM violation";
1398 if (err_mask & AC_ERR_SYSTEM)
1399 return "internal error";
1400 if (err_mask & AC_ERR_MEDIA)
1401 return "media error";
1402 if (err_mask & AC_ERR_INVALID)
1403 return "invalid argument";
1404 if (err_mask & AC_ERR_DEV)
1405 return "device error";
1406 return "unknown error";
1410 * ata_read_log_page - read a specific log page
1411 * @dev: target device
1412 * @page: page to read
1413 * @buf: buffer to store read page
1414 * @sectors: number of sectors to read
1416 * Read log page using READ_LOG_EXT command.
1418 * LOCKING:
1419 * Kernel thread context (may sleep).
1421 * RETURNS:
1422 * 0 on success, AC_ERR_* mask otherwise.
1424 static unsigned int ata_read_log_page(struct ata_device *dev,
1425 u8 page, void *buf, unsigned int sectors)
1427 struct ata_taskfile tf;
1428 unsigned int err_mask;
1430 DPRINTK("read log page - page %d\n", page);
1432 ata_tf_init(dev, &tf);
1433 tf.command = ATA_CMD_READ_LOG_EXT;
1434 tf.lbal = page;
1435 tf.nsect = sectors;
1436 tf.hob_nsect = sectors >> 8;
1437 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1438 tf.protocol = ATA_PROT_PIO;
1440 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1441 buf, sectors * ATA_SECT_SIZE, 0);
1443 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1444 return err_mask;
1448 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1449 * @dev: Device to read log page 10h from
1450 * @tag: Resulting tag of the failed command
1451 * @tf: Resulting taskfile registers of the failed command
1453 * Read log page 10h to obtain NCQ error details and clear error
1454 * condition.
1456 * LOCKING:
1457 * Kernel thread context (may sleep).
1459 * RETURNS:
1460 * 0 on success, -errno otherwise.
1462 static int ata_eh_read_log_10h(struct ata_device *dev,
1463 int *tag, struct ata_taskfile *tf)
1465 u8 *buf = dev->link->ap->sector_buf;
1466 unsigned int err_mask;
1467 u8 csum;
1468 int i;
1470 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1471 if (err_mask)
1472 return -EIO;
1474 csum = 0;
1475 for (i = 0; i < ATA_SECT_SIZE; i++)
1476 csum += buf[i];
1477 if (csum)
1478 ata_dev_printk(dev, KERN_WARNING,
1479 "invalid checksum 0x%x on log page 10h\n", csum);
1481 if (buf[0] & 0x80)
1482 return -ENOENT;
1484 *tag = buf[0] & 0x1f;
1486 tf->command = buf[2];
1487 tf->feature = buf[3];
1488 tf->lbal = buf[4];
1489 tf->lbam = buf[5];
1490 tf->lbah = buf[6];
1491 tf->device = buf[7];
1492 tf->hob_lbal = buf[8];
1493 tf->hob_lbam = buf[9];
1494 tf->hob_lbah = buf[10];
1495 tf->nsect = buf[12];
1496 tf->hob_nsect = buf[13];
1498 return 0;
1502 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1503 * @dev: target ATAPI device
1504 * @r_sense_key: out parameter for sense_key
1506 * Perform ATAPI TEST_UNIT_READY.
1508 * LOCKING:
1509 * EH context (may sleep).
1511 * RETURNS:
1512 * 0 on success, AC_ERR_* mask on failure.
1514 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1516 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1517 struct ata_taskfile tf;
1518 unsigned int err_mask;
1520 ata_tf_init(dev, &tf);
1522 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1523 tf.command = ATA_CMD_PACKET;
1524 tf.protocol = ATAPI_PROT_NODATA;
1526 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1527 if (err_mask == AC_ERR_DEV)
1528 *r_sense_key = tf.feature >> 4;
1529 return err_mask;
1533 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1534 * @dev: device to perform REQUEST_SENSE to
1535 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1536 * @dfl_sense_key: default sense key to use
1538 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1539 * SENSE. This function is EH helper.
1541 * LOCKING:
1542 * Kernel thread context (may sleep).
1544 * RETURNS:
1545 * 0 on success, AC_ERR_* mask on failure
1547 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1548 u8 *sense_buf, u8 dfl_sense_key)
1550 u8 cdb[ATAPI_CDB_LEN] =
1551 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1552 struct ata_port *ap = dev->link->ap;
1553 struct ata_taskfile tf;
1555 DPRINTK("ATAPI request sense\n");
1557 /* FIXME: is this needed? */
1558 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1560 /* initialize sense_buf with the error register,
1561 * for the case where they are -not- overwritten
1563 sense_buf[0] = 0x70;
1564 sense_buf[2] = dfl_sense_key;
1566 /* some devices time out if garbage left in tf */
1567 ata_tf_init(dev, &tf);
1569 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1570 tf.command = ATA_CMD_PACKET;
1572 /* is it pointless to prefer PIO for "safety reasons"? */
1573 if (ap->flags & ATA_FLAG_PIO_DMA) {
1574 tf.protocol = ATAPI_PROT_DMA;
1575 tf.feature |= ATAPI_PKT_DMA;
1576 } else {
1577 tf.protocol = ATAPI_PROT_PIO;
1578 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1579 tf.lbah = 0;
1582 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1583 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1587 * ata_eh_analyze_serror - analyze SError for a failed port
1588 * @link: ATA link to analyze SError for
1590 * Analyze SError if available and further determine cause of
1591 * failure.
1593 * LOCKING:
1594 * None.
1596 static void ata_eh_analyze_serror(struct ata_link *link)
1598 struct ata_eh_context *ehc = &link->eh_context;
1599 u32 serror = ehc->i.serror;
1600 unsigned int err_mask = 0, action = 0;
1601 u32 hotplug_mask;
1603 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1604 err_mask |= AC_ERR_ATA_BUS;
1605 action |= ATA_EH_RESET;
1607 if (serror & SERR_PROTOCOL) {
1608 err_mask |= AC_ERR_HSM;
1609 action |= ATA_EH_RESET;
1611 if (serror & SERR_INTERNAL) {
1612 err_mask |= AC_ERR_SYSTEM;
1613 action |= ATA_EH_RESET;
1616 /* Determine whether a hotplug event has occurred. Both
1617 * SError.N/X are considered hotplug events for enabled or
1618 * host links. For disabled PMP links, only N bit is
1619 * considered as X bit is left at 1 for link plugging.
1621 if (link->lpm_policy != ATA_LPM_MAX_POWER)
1622 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1623 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1624 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1625 else
1626 hotplug_mask = SERR_PHYRDY_CHG;
1628 if (serror & hotplug_mask)
1629 ata_ehi_hotplugged(&ehc->i);
1631 ehc->i.err_mask |= err_mask;
1632 ehc->i.action |= action;
1636 * ata_eh_analyze_ncq_error - analyze NCQ error
1637 * @link: ATA link to analyze NCQ error for
1639 * Read log page 10h, determine the offending qc and acquire
1640 * error status TF. For NCQ device errors, all LLDDs have to do
1641 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1642 * care of the rest.
1644 * LOCKING:
1645 * Kernel thread context (may sleep).
1647 void ata_eh_analyze_ncq_error(struct ata_link *link)
1649 struct ata_port *ap = link->ap;
1650 struct ata_eh_context *ehc = &link->eh_context;
1651 struct ata_device *dev = link->device;
1652 struct ata_queued_cmd *qc;
1653 struct ata_taskfile tf;
1654 int tag, rc;
1656 /* if frozen, we can't do much */
1657 if (ap->pflags & ATA_PFLAG_FROZEN)
1658 return;
1660 /* is it NCQ device error? */
1661 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1662 return;
1664 /* has LLDD analyzed already? */
1665 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1666 qc = __ata_qc_from_tag(ap, tag);
1668 if (!(qc->flags & ATA_QCFLAG_FAILED))
1669 continue;
1671 if (qc->err_mask)
1672 return;
1675 /* okay, this error is ours */
1676 memset(&tf, 0, sizeof(tf));
1677 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1678 if (rc) {
1679 ata_link_printk(link, KERN_ERR, "failed to read log page 10h "
1680 "(errno=%d)\n", rc);
1681 return;
1684 if (!(link->sactive & (1 << tag))) {
1685 ata_link_printk(link, KERN_ERR, "log page 10h reported "
1686 "inactive tag %d\n", tag);
1687 return;
1690 /* we've got the perpetrator, condemn it */
1691 qc = __ata_qc_from_tag(ap, tag);
1692 memcpy(&qc->result_tf, &tf, sizeof(tf));
1693 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1694 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1695 ehc->i.err_mask &= ~AC_ERR_DEV;
1699 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1700 * @qc: qc to analyze
1701 * @tf: Taskfile registers to analyze
1703 * Analyze taskfile of @qc and further determine cause of
1704 * failure. This function also requests ATAPI sense data if
1705 * avaliable.
1707 * LOCKING:
1708 * Kernel thread context (may sleep).
1710 * RETURNS:
1711 * Determined recovery action
1713 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1714 const struct ata_taskfile *tf)
1716 unsigned int tmp, action = 0;
1717 u8 stat = tf->command, err = tf->feature;
1719 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1720 qc->err_mask |= AC_ERR_HSM;
1721 return ATA_EH_RESET;
1724 if (stat & (ATA_ERR | ATA_DF))
1725 qc->err_mask |= AC_ERR_DEV;
1726 else
1727 return 0;
1729 switch (qc->dev->class) {
1730 case ATA_DEV_ATA:
1731 if (err & ATA_ICRC)
1732 qc->err_mask |= AC_ERR_ATA_BUS;
1733 if (err & ATA_UNC)
1734 qc->err_mask |= AC_ERR_MEDIA;
1735 if (err & ATA_IDNF)
1736 qc->err_mask |= AC_ERR_INVALID;
1737 break;
1739 case ATA_DEV_ATAPI:
1740 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1741 tmp = atapi_eh_request_sense(qc->dev,
1742 qc->scsicmd->sense_buffer,
1743 qc->result_tf.feature >> 4);
1744 if (!tmp) {
1745 /* ATA_QCFLAG_SENSE_VALID is used to
1746 * tell atapi_qc_complete() that sense
1747 * data is already valid.
1749 * TODO: interpret sense data and set
1750 * appropriate err_mask.
1752 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1753 } else
1754 qc->err_mask |= tmp;
1758 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1759 action |= ATA_EH_RESET;
1761 return action;
1764 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1765 int *xfer_ok)
1767 int base = 0;
1769 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1770 *xfer_ok = 1;
1772 if (!*xfer_ok)
1773 base = ATA_ECAT_DUBIOUS_NONE;
1775 if (err_mask & AC_ERR_ATA_BUS)
1776 return base + ATA_ECAT_ATA_BUS;
1778 if (err_mask & AC_ERR_TIMEOUT)
1779 return base + ATA_ECAT_TOUT_HSM;
1781 if (eflags & ATA_EFLAG_IS_IO) {
1782 if (err_mask & AC_ERR_HSM)
1783 return base + ATA_ECAT_TOUT_HSM;
1784 if ((err_mask &
1785 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1786 return base + ATA_ECAT_UNK_DEV;
1789 return 0;
1792 struct speed_down_verdict_arg {
1793 u64 since;
1794 int xfer_ok;
1795 int nr_errors[ATA_ECAT_NR];
1798 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1800 struct speed_down_verdict_arg *arg = void_arg;
1801 int cat;
1803 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1804 return -1;
1806 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1807 &arg->xfer_ok);
1808 arg->nr_errors[cat]++;
1810 return 0;
1814 * ata_eh_speed_down_verdict - Determine speed down verdict
1815 * @dev: Device of interest
1817 * This function examines error ring of @dev and determines
1818 * whether NCQ needs to be turned off, transfer speed should be
1819 * stepped down, or falling back to PIO is necessary.
1821 * ECAT_ATA_BUS : ATA_BUS error for any command
1823 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1824 * IO commands
1826 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1828 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1829 * data transfer hasn't been verified.
1831 * Verdicts are
1833 * NCQ_OFF : Turn off NCQ.
1835 * SPEED_DOWN : Speed down transfer speed but don't fall back
1836 * to PIO.
1838 * FALLBACK_TO_PIO : Fall back to PIO.
1840 * Even if multiple verdicts are returned, only one action is
1841 * taken per error. An action triggered by non-DUBIOUS errors
1842 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1843 * This is to expedite speed down decisions right after device is
1844 * initially configured.
1846 * The followings are speed down rules. #1 and #2 deal with
1847 * DUBIOUS errors.
1849 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1850 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1852 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1853 * occurred during last 5 mins, NCQ_OFF.
1855 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1856 * ocurred during last 5 mins, FALLBACK_TO_PIO
1858 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1859 * during last 10 mins, NCQ_OFF.
1861 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1862 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1864 * LOCKING:
1865 * Inherited from caller.
1867 * RETURNS:
1868 * OR of ATA_EH_SPDN_* flags.
1870 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1872 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1873 u64 j64 = get_jiffies_64();
1874 struct speed_down_verdict_arg arg;
1875 unsigned int verdict = 0;
1877 /* scan past 5 mins of error history */
1878 memset(&arg, 0, sizeof(arg));
1879 arg.since = j64 - min(j64, j5mins);
1880 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1882 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1883 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1884 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1885 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1887 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1888 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1889 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1891 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1892 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1893 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1894 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1896 /* scan past 10 mins of error history */
1897 memset(&arg, 0, sizeof(arg));
1898 arg.since = j64 - min(j64, j10mins);
1899 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1901 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1902 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1903 verdict |= ATA_EH_SPDN_NCQ_OFF;
1905 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1906 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1907 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1908 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1910 return verdict;
1914 * ata_eh_speed_down - record error and speed down if necessary
1915 * @dev: Failed device
1916 * @eflags: mask of ATA_EFLAG_* flags
1917 * @err_mask: err_mask of the error
1919 * Record error and examine error history to determine whether
1920 * adjusting transmission speed is necessary. It also sets
1921 * transmission limits appropriately if such adjustment is
1922 * necessary.
1924 * LOCKING:
1925 * Kernel thread context (may sleep).
1927 * RETURNS:
1928 * Determined recovery action.
1930 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1931 unsigned int eflags, unsigned int err_mask)
1933 struct ata_link *link = ata_dev_phys_link(dev);
1934 int xfer_ok = 0;
1935 unsigned int verdict;
1936 unsigned int action = 0;
1938 /* don't bother if Cat-0 error */
1939 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1940 return 0;
1942 /* record error and determine whether speed down is necessary */
1943 ata_ering_record(&dev->ering, eflags, err_mask);
1944 verdict = ata_eh_speed_down_verdict(dev);
1946 /* turn off NCQ? */
1947 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1948 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1949 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1950 dev->flags |= ATA_DFLAG_NCQ_OFF;
1951 ata_dev_printk(dev, KERN_WARNING,
1952 "NCQ disabled due to excessive errors\n");
1953 goto done;
1956 /* speed down? */
1957 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1958 /* speed down SATA link speed if possible */
1959 if (sata_down_spd_limit(link, 0) == 0) {
1960 action |= ATA_EH_RESET;
1961 goto done;
1964 /* lower transfer mode */
1965 if (dev->spdn_cnt < 2) {
1966 static const int dma_dnxfer_sel[] =
1967 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1968 static const int pio_dnxfer_sel[] =
1969 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1970 int sel;
1972 if (dev->xfer_shift != ATA_SHIFT_PIO)
1973 sel = dma_dnxfer_sel[dev->spdn_cnt];
1974 else
1975 sel = pio_dnxfer_sel[dev->spdn_cnt];
1977 dev->spdn_cnt++;
1979 if (ata_down_xfermask_limit(dev, sel) == 0) {
1980 action |= ATA_EH_RESET;
1981 goto done;
1986 /* Fall back to PIO? Slowing down to PIO is meaningless for
1987 * SATA ATA devices. Consider it only for PATA and SATAPI.
1989 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1990 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1991 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1992 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1993 dev->spdn_cnt = 0;
1994 action |= ATA_EH_RESET;
1995 goto done;
1999 return 0;
2000 done:
2001 /* device has been slowed down, blow error history */
2002 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2003 ata_ering_clear(&dev->ering);
2004 return action;
2008 * ata_eh_link_autopsy - analyze error and determine recovery action
2009 * @link: host link to perform autopsy on
2011 * Analyze why @link failed and determine which recovery actions
2012 * are needed. This function also sets more detailed AC_ERR_*
2013 * values and fills sense data for ATAPI CHECK SENSE.
2015 * LOCKING:
2016 * Kernel thread context (may sleep).
2018 static void ata_eh_link_autopsy(struct ata_link *link)
2020 struct ata_port *ap = link->ap;
2021 struct ata_eh_context *ehc = &link->eh_context;
2022 struct ata_device *dev;
2023 unsigned int all_err_mask = 0, eflags = 0;
2024 int tag;
2025 u32 serror;
2026 int rc;
2028 DPRINTK("ENTER\n");
2030 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2031 return;
2033 /* obtain and analyze SError */
2034 rc = sata_scr_read(link, SCR_ERROR, &serror);
2035 if (rc == 0) {
2036 ehc->i.serror |= serror;
2037 ata_eh_analyze_serror(link);
2038 } else if (rc != -EOPNOTSUPP) {
2039 /* SError read failed, force reset and probing */
2040 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2041 ehc->i.action |= ATA_EH_RESET;
2042 ehc->i.err_mask |= AC_ERR_OTHER;
2045 /* analyze NCQ failure */
2046 ata_eh_analyze_ncq_error(link);
2048 /* any real error trumps AC_ERR_OTHER */
2049 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2050 ehc->i.err_mask &= ~AC_ERR_OTHER;
2052 all_err_mask |= ehc->i.err_mask;
2054 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2055 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2057 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2058 ata_dev_phys_link(qc->dev) != link)
2059 continue;
2061 /* inherit upper level err_mask */
2062 qc->err_mask |= ehc->i.err_mask;
2064 /* analyze TF */
2065 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2067 /* DEV errors are probably spurious in case of ATA_BUS error */
2068 if (qc->err_mask & AC_ERR_ATA_BUS)
2069 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2070 AC_ERR_INVALID);
2072 /* any real error trumps unknown error */
2073 if (qc->err_mask & ~AC_ERR_OTHER)
2074 qc->err_mask &= ~AC_ERR_OTHER;
2076 /* SENSE_VALID trumps dev/unknown error and revalidation */
2077 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2078 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2080 /* determine whether the command is worth retrying */
2081 if (qc->flags & ATA_QCFLAG_IO ||
2082 (!(qc->err_mask & AC_ERR_INVALID) &&
2083 qc->err_mask != AC_ERR_DEV))
2084 qc->flags |= ATA_QCFLAG_RETRY;
2086 /* accumulate error info */
2087 ehc->i.dev = qc->dev;
2088 all_err_mask |= qc->err_mask;
2089 if (qc->flags & ATA_QCFLAG_IO)
2090 eflags |= ATA_EFLAG_IS_IO;
2093 /* enforce default EH actions */
2094 if (ap->pflags & ATA_PFLAG_FROZEN ||
2095 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2096 ehc->i.action |= ATA_EH_RESET;
2097 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2098 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2099 ehc->i.action |= ATA_EH_REVALIDATE;
2101 /* If we have offending qcs and the associated failed device,
2102 * perform per-dev EH action only on the offending device.
2104 if (ehc->i.dev) {
2105 ehc->i.dev_action[ehc->i.dev->devno] |=
2106 ehc->i.action & ATA_EH_PERDEV_MASK;
2107 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2110 /* propagate timeout to host link */
2111 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2112 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2114 /* record error and consider speeding down */
2115 dev = ehc->i.dev;
2116 if (!dev && ((ata_link_max_devices(link) == 1 &&
2117 ata_dev_enabled(link->device))))
2118 dev = link->device;
2120 if (dev) {
2121 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2122 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2123 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2126 DPRINTK("EXIT\n");
2130 * ata_eh_autopsy - analyze error and determine recovery action
2131 * @ap: host port to perform autopsy on
2133 * Analyze all links of @ap and determine why they failed and
2134 * which recovery actions are needed.
2136 * LOCKING:
2137 * Kernel thread context (may sleep).
2139 void ata_eh_autopsy(struct ata_port *ap)
2141 struct ata_link *link;
2143 ata_for_each_link(link, ap, EDGE)
2144 ata_eh_link_autopsy(link);
2146 /* Handle the frigging slave link. Autopsy is done similarly
2147 * but actions and flags are transferred over to the master
2148 * link and handled from there.
2150 if (ap->slave_link) {
2151 struct ata_eh_context *mehc = &ap->link.eh_context;
2152 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2154 /* transfer control flags from master to slave */
2155 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2157 /* perform autopsy on the slave link */
2158 ata_eh_link_autopsy(ap->slave_link);
2160 /* transfer actions from slave to master and clear slave */
2161 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2162 mehc->i.action |= sehc->i.action;
2163 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2164 mehc->i.flags |= sehc->i.flags;
2165 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2168 /* Autopsy of fanout ports can affect host link autopsy.
2169 * Perform host link autopsy last.
2171 if (sata_pmp_attached(ap))
2172 ata_eh_link_autopsy(&ap->link);
2176 * ata_get_cmd_descript - get description for ATA command
2177 * @command: ATA command code to get description for
2179 * Return a textual description of the given command, or NULL if the
2180 * command is not known.
2182 * LOCKING:
2183 * None
2185 const char *ata_get_cmd_descript(u8 command)
2187 #ifdef CONFIG_ATA_VERBOSE_ERROR
2188 static const struct
2190 u8 command;
2191 const char *text;
2192 } cmd_descr[] = {
2193 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2194 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2195 { ATA_CMD_STANDBY, "STANDBY" },
2196 { ATA_CMD_IDLE, "IDLE" },
2197 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2198 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2199 { ATA_CMD_NOP, "NOP" },
2200 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2201 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2202 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2203 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2204 { ATA_CMD_SERVICE, "SERVICE" },
2205 { ATA_CMD_READ, "READ DMA" },
2206 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2207 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2208 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2209 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2210 { ATA_CMD_WRITE, "WRITE DMA" },
2211 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2212 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2213 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2214 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2215 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2216 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2217 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2218 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2219 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2220 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2221 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2222 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2223 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2224 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2225 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2226 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2227 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2228 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2229 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2230 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2231 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2232 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2233 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2234 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2235 { ATA_CMD_SLEEP, "SLEEP" },
2236 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2237 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2238 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2239 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2240 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2241 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2242 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2243 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2244 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2245 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2246 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2247 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2248 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2249 { ATA_CMD_PMP_READ, "READ BUFFER" },
2250 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2251 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2252 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2253 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2254 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2255 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2256 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2257 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2258 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2259 { ATA_CMD_SMART, "SMART" },
2260 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2261 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2262 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2263 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2264 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2265 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2266 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2267 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2268 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2269 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2270 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2271 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2272 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2273 { ATA_CMD_RESTORE, "RECALIBRATE" },
2274 { 0, NULL } /* terminate list */
2277 unsigned int i;
2278 for (i = 0; cmd_descr[i].text; i++)
2279 if (cmd_descr[i].command == command)
2280 return cmd_descr[i].text;
2281 #endif
2283 return NULL;
2287 * ata_eh_link_report - report error handling to user
2288 * @link: ATA link EH is going on
2290 * Report EH to user.
2292 * LOCKING:
2293 * None.
2295 static void ata_eh_link_report(struct ata_link *link)
2297 struct ata_port *ap = link->ap;
2298 struct ata_eh_context *ehc = &link->eh_context;
2299 const char *frozen, *desc;
2300 char tries_buf[6];
2301 int tag, nr_failed = 0;
2303 if (ehc->i.flags & ATA_EHI_QUIET)
2304 return;
2306 desc = NULL;
2307 if (ehc->i.desc[0] != '\0')
2308 desc = ehc->i.desc;
2310 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2311 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2313 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2314 ata_dev_phys_link(qc->dev) != link ||
2315 ((qc->flags & ATA_QCFLAG_QUIET) &&
2316 qc->err_mask == AC_ERR_DEV))
2317 continue;
2318 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2319 continue;
2321 nr_failed++;
2324 if (!nr_failed && !ehc->i.err_mask)
2325 return;
2327 frozen = "";
2328 if (ap->pflags & ATA_PFLAG_FROZEN)
2329 frozen = " frozen";
2331 memset(tries_buf, 0, sizeof(tries_buf));
2332 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2333 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2334 ap->eh_tries);
2336 if (ehc->i.dev) {
2337 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
2338 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2339 ehc->i.err_mask, link->sactive, ehc->i.serror,
2340 ehc->i.action, frozen, tries_buf);
2341 if (desc)
2342 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
2343 } else {
2344 ata_link_printk(link, KERN_ERR, "exception Emask 0x%x "
2345 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2346 ehc->i.err_mask, link->sactive, ehc->i.serror,
2347 ehc->i.action, frozen, tries_buf);
2348 if (desc)
2349 ata_link_printk(link, KERN_ERR, "%s\n", desc);
2352 #ifdef CONFIG_ATA_VERBOSE_ERROR
2353 if (ehc->i.serror)
2354 ata_link_printk(link, KERN_ERR,
2355 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2356 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2357 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2358 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2359 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2360 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2361 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2362 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2363 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2364 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2365 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2366 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2367 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2368 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2369 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2370 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2371 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2372 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2373 #endif
2375 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2376 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2377 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2378 const u8 *cdb = qc->cdb;
2379 char data_buf[20] = "";
2380 char cdb_buf[70] = "";
2382 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2383 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2384 continue;
2386 if (qc->dma_dir != DMA_NONE) {
2387 static const char *dma_str[] = {
2388 [DMA_BIDIRECTIONAL] = "bidi",
2389 [DMA_TO_DEVICE] = "out",
2390 [DMA_FROM_DEVICE] = "in",
2392 static const char *prot_str[] = {
2393 [ATA_PROT_PIO] = "pio",
2394 [ATA_PROT_DMA] = "dma",
2395 [ATA_PROT_NCQ] = "ncq",
2396 [ATAPI_PROT_PIO] = "pio",
2397 [ATAPI_PROT_DMA] = "dma",
2400 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2401 prot_str[qc->tf.protocol], qc->nbytes,
2402 dma_str[qc->dma_dir]);
2405 if (ata_is_atapi(qc->tf.protocol)) {
2406 if (qc->scsicmd)
2407 scsi_print_command(qc->scsicmd);
2408 else
2409 snprintf(cdb_buf, sizeof(cdb_buf),
2410 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2411 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2412 cdb[0], cdb[1], cdb[2], cdb[3],
2413 cdb[4], cdb[5], cdb[6], cdb[7],
2414 cdb[8], cdb[9], cdb[10], cdb[11],
2415 cdb[12], cdb[13], cdb[14], cdb[15]);
2416 } else {
2417 const char *descr = ata_get_cmd_descript(cmd->command);
2418 if (descr)
2419 ata_dev_printk(qc->dev, KERN_ERR,
2420 "failed command: %s\n", descr);
2423 ata_dev_printk(qc->dev, KERN_ERR,
2424 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2425 "tag %d%s\n %s"
2426 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2427 "Emask 0x%x (%s)%s\n",
2428 cmd->command, cmd->feature, cmd->nsect,
2429 cmd->lbal, cmd->lbam, cmd->lbah,
2430 cmd->hob_feature, cmd->hob_nsect,
2431 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2432 cmd->device, qc->tag, data_buf, cdb_buf,
2433 res->command, res->feature, res->nsect,
2434 res->lbal, res->lbam, res->lbah,
2435 res->hob_feature, res->hob_nsect,
2436 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2437 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2438 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2440 #ifdef CONFIG_ATA_VERBOSE_ERROR
2441 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2442 ATA_ERR)) {
2443 if (res->command & ATA_BUSY)
2444 ata_dev_printk(qc->dev, KERN_ERR,
2445 "status: { Busy }\n");
2446 else
2447 ata_dev_printk(qc->dev, KERN_ERR,
2448 "status: { %s%s%s%s}\n",
2449 res->command & ATA_DRDY ? "DRDY " : "",
2450 res->command & ATA_DF ? "DF " : "",
2451 res->command & ATA_DRQ ? "DRQ " : "",
2452 res->command & ATA_ERR ? "ERR " : "");
2455 if (cmd->command != ATA_CMD_PACKET &&
2456 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2457 ATA_ABORTED)))
2458 ata_dev_printk(qc->dev, KERN_ERR,
2459 "error: { %s%s%s%s}\n",
2460 res->feature & ATA_ICRC ? "ICRC " : "",
2461 res->feature & ATA_UNC ? "UNC " : "",
2462 res->feature & ATA_IDNF ? "IDNF " : "",
2463 res->feature & ATA_ABORTED ? "ABRT " : "");
2464 #endif
2469 * ata_eh_report - report error handling to user
2470 * @ap: ATA port to report EH about
2472 * Report EH to user.
2474 * LOCKING:
2475 * None.
2477 void ata_eh_report(struct ata_port *ap)
2479 struct ata_link *link;
2481 ata_for_each_link(link, ap, HOST_FIRST)
2482 ata_eh_link_report(link);
2485 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2486 unsigned int *classes, unsigned long deadline,
2487 bool clear_classes)
2489 struct ata_device *dev;
2491 if (clear_classes)
2492 ata_for_each_dev(dev, link, ALL)
2493 classes[dev->devno] = ATA_DEV_UNKNOWN;
2495 return reset(link, classes, deadline);
2498 static int ata_eh_followup_srst_needed(struct ata_link *link,
2499 int rc, const unsigned int *classes)
2501 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2502 return 0;
2503 if (rc == -EAGAIN)
2504 return 1;
2505 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2506 return 1;
2507 return 0;
2510 int ata_eh_reset(struct ata_link *link, int classify,
2511 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2512 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2514 struct ata_port *ap = link->ap;
2515 struct ata_link *slave = ap->slave_link;
2516 struct ata_eh_context *ehc = &link->eh_context;
2517 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2518 unsigned int *classes = ehc->classes;
2519 unsigned int lflags = link->flags;
2520 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2521 int max_tries = 0, try = 0;
2522 struct ata_link *failed_link;
2523 struct ata_device *dev;
2524 unsigned long deadline, now;
2525 ata_reset_fn_t reset;
2526 unsigned long flags;
2527 u32 sstatus;
2528 int nr_unknown, rc;
2531 * Prepare to reset
2533 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2534 max_tries++;
2535 if (link->flags & ATA_LFLAG_NO_HRST)
2536 hardreset = NULL;
2537 if (link->flags & ATA_LFLAG_NO_SRST)
2538 softreset = NULL;
2540 /* make sure each reset attemp is at least COOL_DOWN apart */
2541 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2542 now = jiffies;
2543 WARN_ON(time_after(ehc->last_reset, now));
2544 deadline = ata_deadline(ehc->last_reset,
2545 ATA_EH_RESET_COOL_DOWN);
2546 if (time_before(now, deadline))
2547 schedule_timeout_uninterruptible(deadline - now);
2550 spin_lock_irqsave(ap->lock, flags);
2551 ap->pflags |= ATA_PFLAG_RESETTING;
2552 spin_unlock_irqrestore(ap->lock, flags);
2554 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2556 ata_for_each_dev(dev, link, ALL) {
2557 /* If we issue an SRST then an ATA drive (not ATAPI)
2558 * may change configuration and be in PIO0 timing. If
2559 * we do a hard reset (or are coming from power on)
2560 * this is true for ATA or ATAPI. Until we've set a
2561 * suitable controller mode we should not touch the
2562 * bus as we may be talking too fast.
2564 dev->pio_mode = XFER_PIO_0;
2566 /* If the controller has a pio mode setup function
2567 * then use it to set the chipset to rights. Don't
2568 * touch the DMA setup as that will be dealt with when
2569 * configuring devices.
2571 if (ap->ops->set_piomode)
2572 ap->ops->set_piomode(ap, dev);
2575 /* prefer hardreset */
2576 reset = NULL;
2577 ehc->i.action &= ~ATA_EH_RESET;
2578 if (hardreset) {
2579 reset = hardreset;
2580 ehc->i.action |= ATA_EH_HARDRESET;
2581 } else if (softreset) {
2582 reset = softreset;
2583 ehc->i.action |= ATA_EH_SOFTRESET;
2586 if (prereset) {
2587 unsigned long deadline = ata_deadline(jiffies,
2588 ATA_EH_PRERESET_TIMEOUT);
2590 if (slave) {
2591 sehc->i.action &= ~ATA_EH_RESET;
2592 sehc->i.action |= ehc->i.action;
2595 rc = prereset(link, deadline);
2597 /* If present, do prereset on slave link too. Reset
2598 * is skipped iff both master and slave links report
2599 * -ENOENT or clear ATA_EH_RESET.
2601 if (slave && (rc == 0 || rc == -ENOENT)) {
2602 int tmp;
2604 tmp = prereset(slave, deadline);
2605 if (tmp != -ENOENT)
2606 rc = tmp;
2608 ehc->i.action |= sehc->i.action;
2611 if (rc) {
2612 if (rc == -ENOENT) {
2613 ata_link_printk(link, KERN_DEBUG,
2614 "port disabled. ignoring.\n");
2615 ehc->i.action &= ~ATA_EH_RESET;
2617 ata_for_each_dev(dev, link, ALL)
2618 classes[dev->devno] = ATA_DEV_NONE;
2620 rc = 0;
2621 } else
2622 ata_link_printk(link, KERN_ERR,
2623 "prereset failed (errno=%d)\n", rc);
2624 goto out;
2627 /* prereset() might have cleared ATA_EH_RESET. If so,
2628 * bang classes, thaw and return.
2630 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2631 ata_for_each_dev(dev, link, ALL)
2632 classes[dev->devno] = ATA_DEV_NONE;
2633 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2634 ata_is_host_link(link))
2635 ata_eh_thaw_port(ap);
2636 rc = 0;
2637 goto out;
2641 retry:
2643 * Perform reset
2645 if (ata_is_host_link(link))
2646 ata_eh_freeze_port(ap);
2648 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2650 if (reset) {
2651 if (verbose)
2652 ata_link_printk(link, KERN_INFO, "%s resetting link\n",
2653 reset == softreset ? "soft" : "hard");
2655 /* mark that this EH session started with reset */
2656 ehc->last_reset = jiffies;
2657 if (reset == hardreset)
2658 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2659 else
2660 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2662 rc = ata_do_reset(link, reset, classes, deadline, true);
2663 if (rc && rc != -EAGAIN) {
2664 failed_link = link;
2665 goto fail;
2668 /* hardreset slave link if existent */
2669 if (slave && reset == hardreset) {
2670 int tmp;
2672 if (verbose)
2673 ata_link_printk(slave, KERN_INFO,
2674 "hard resetting link\n");
2676 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2677 tmp = ata_do_reset(slave, reset, classes, deadline,
2678 false);
2679 switch (tmp) {
2680 case -EAGAIN:
2681 rc = -EAGAIN;
2682 case 0:
2683 break;
2684 default:
2685 failed_link = slave;
2686 rc = tmp;
2687 goto fail;
2691 /* perform follow-up SRST if necessary */
2692 if (reset == hardreset &&
2693 ata_eh_followup_srst_needed(link, rc, classes)) {
2694 reset = softreset;
2696 if (!reset) {
2697 ata_link_printk(link, KERN_ERR,
2698 "follow-up softreset required "
2699 "but no softreset avaliable\n");
2700 failed_link = link;
2701 rc = -EINVAL;
2702 goto fail;
2705 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2706 rc = ata_do_reset(link, reset, classes, deadline, true);
2707 if (rc) {
2708 failed_link = link;
2709 goto fail;
2712 } else {
2713 if (verbose)
2714 ata_link_printk(link, KERN_INFO, "no reset method "
2715 "available, skipping reset\n");
2716 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2717 lflags |= ATA_LFLAG_ASSUME_ATA;
2721 * Post-reset processing
2723 ata_for_each_dev(dev, link, ALL) {
2724 /* After the reset, the device state is PIO 0 and the
2725 * controller state is undefined. Reset also wakes up
2726 * drives from sleeping mode.
2728 dev->pio_mode = XFER_PIO_0;
2729 dev->flags &= ~ATA_DFLAG_SLEEPING;
2731 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2732 continue;
2734 /* apply class override */
2735 if (lflags & ATA_LFLAG_ASSUME_ATA)
2736 classes[dev->devno] = ATA_DEV_ATA;
2737 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2738 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2741 /* record current link speed */
2742 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2743 link->sata_spd = (sstatus >> 4) & 0xf;
2744 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2745 slave->sata_spd = (sstatus >> 4) & 0xf;
2747 /* thaw the port */
2748 if (ata_is_host_link(link))
2749 ata_eh_thaw_port(ap);
2751 /* postreset() should clear hardware SError. Although SError
2752 * is cleared during link resume, clearing SError here is
2753 * necessary as some PHYs raise hotplug events after SRST.
2754 * This introduces race condition where hotplug occurs between
2755 * reset and here. This race is mediated by cross checking
2756 * link onlineness and classification result later.
2758 if (postreset) {
2759 postreset(link, classes);
2760 if (slave)
2761 postreset(slave, classes);
2765 * Some controllers can't be frozen very well and may set
2766 * spuruious error conditions during reset. Clear accumulated
2767 * error information. As reset is the final recovery action,
2768 * nothing is lost by doing this.
2770 spin_lock_irqsave(link->ap->lock, flags);
2771 memset(&link->eh_info, 0, sizeof(link->eh_info));
2772 if (slave)
2773 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2774 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2775 spin_unlock_irqrestore(link->ap->lock, flags);
2778 * Make sure onlineness and classification result correspond.
2779 * Hotplug could have happened during reset and some
2780 * controllers fail to wait while a drive is spinning up after
2781 * being hotplugged causing misdetection. By cross checking
2782 * link on/offlineness and classification result, those
2783 * conditions can be reliably detected and retried.
2785 nr_unknown = 0;
2786 ata_for_each_dev(dev, link, ALL) {
2787 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2788 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2789 ata_dev_printk(dev, KERN_DEBUG, "link online "
2790 "but device misclassifed\n");
2791 classes[dev->devno] = ATA_DEV_NONE;
2792 nr_unknown++;
2794 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2795 if (ata_class_enabled(classes[dev->devno]))
2796 ata_dev_printk(dev, KERN_DEBUG, "link offline, "
2797 "clearing class %d to NONE\n",
2798 classes[dev->devno]);
2799 classes[dev->devno] = ATA_DEV_NONE;
2800 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2801 ata_dev_printk(dev, KERN_DEBUG, "link status unknown, "
2802 "clearing UNKNOWN to NONE\n");
2803 classes[dev->devno] = ATA_DEV_NONE;
2807 if (classify && nr_unknown) {
2808 if (try < max_tries) {
2809 ata_link_printk(link, KERN_WARNING, "link online but "
2810 "%d devices misclassified, retrying\n",
2811 nr_unknown);
2812 failed_link = link;
2813 rc = -EAGAIN;
2814 goto fail;
2816 ata_link_printk(link, KERN_WARNING,
2817 "link online but %d devices misclassified, "
2818 "device detection might fail\n", nr_unknown);
2821 /* reset successful, schedule revalidation */
2822 ata_eh_done(link, NULL, ATA_EH_RESET);
2823 if (slave)
2824 ata_eh_done(slave, NULL, ATA_EH_RESET);
2825 ehc->last_reset = jiffies; /* update to completion time */
2826 ehc->i.action |= ATA_EH_REVALIDATE;
2827 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2829 rc = 0;
2830 out:
2831 /* clear hotplug flag */
2832 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2833 if (slave)
2834 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2836 spin_lock_irqsave(ap->lock, flags);
2837 ap->pflags &= ~ATA_PFLAG_RESETTING;
2838 spin_unlock_irqrestore(ap->lock, flags);
2840 return rc;
2842 fail:
2843 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2844 if (!ata_is_host_link(link) &&
2845 sata_scr_read(link, SCR_STATUS, &sstatus))
2846 rc = -ERESTART;
2848 if (rc == -ERESTART || try >= max_tries)
2849 goto out;
2851 now = jiffies;
2852 if (time_before(now, deadline)) {
2853 unsigned long delta = deadline - now;
2855 ata_link_printk(failed_link, KERN_WARNING,
2856 "reset failed (errno=%d), retrying in %u secs\n",
2857 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2859 ata_eh_release(ap);
2860 while (delta)
2861 delta = schedule_timeout_uninterruptible(delta);
2862 ata_eh_acquire(ap);
2865 if (try == max_tries - 1) {
2866 sata_down_spd_limit(link, 0);
2867 if (slave)
2868 sata_down_spd_limit(slave, 0);
2869 } else if (rc == -EPIPE)
2870 sata_down_spd_limit(failed_link, 0);
2872 if (hardreset)
2873 reset = hardreset;
2874 goto retry;
2877 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2879 struct ata_link *link;
2880 struct ata_device *dev;
2881 unsigned long flags;
2884 * This function can be thought of as an extended version of
2885 * ata_eh_about_to_do() specially crafted to accommodate the
2886 * requirements of ATA_EH_PARK handling. Since the EH thread
2887 * does not leave the do {} while () loop in ata_eh_recover as
2888 * long as the timeout for a park request to *one* device on
2889 * the port has not expired, and since we still want to pick
2890 * up park requests to other devices on the same port or
2891 * timeout updates for the same device, we have to pull
2892 * ATA_EH_PARK actions from eh_info into eh_context.i
2893 * ourselves at the beginning of each pass over the loop.
2895 * Additionally, all write accesses to &ap->park_req_pending
2896 * through INIT_COMPLETION() (see below) or complete_all()
2897 * (see ata_scsi_park_store()) are protected by the host lock.
2898 * As a result we have that park_req_pending.done is zero on
2899 * exit from this function, i.e. when ATA_EH_PARK actions for
2900 * *all* devices on port ap have been pulled into the
2901 * respective eh_context structs. If, and only if,
2902 * park_req_pending.done is non-zero by the time we reach
2903 * wait_for_completion_timeout(), another ATA_EH_PARK action
2904 * has been scheduled for at least one of the devices on port
2905 * ap and we have to cycle over the do {} while () loop in
2906 * ata_eh_recover() again.
2909 spin_lock_irqsave(ap->lock, flags);
2910 INIT_COMPLETION(ap->park_req_pending);
2911 ata_for_each_link(link, ap, EDGE) {
2912 ata_for_each_dev(dev, link, ALL) {
2913 struct ata_eh_info *ehi = &link->eh_info;
2915 link->eh_context.i.dev_action[dev->devno] |=
2916 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2917 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2920 spin_unlock_irqrestore(ap->lock, flags);
2923 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2925 struct ata_eh_context *ehc = &dev->link->eh_context;
2926 struct ata_taskfile tf;
2927 unsigned int err_mask;
2929 ata_tf_init(dev, &tf);
2930 if (park) {
2931 ehc->unloaded_mask |= 1 << dev->devno;
2932 tf.command = ATA_CMD_IDLEIMMEDIATE;
2933 tf.feature = 0x44;
2934 tf.lbal = 0x4c;
2935 tf.lbam = 0x4e;
2936 tf.lbah = 0x55;
2937 } else {
2938 ehc->unloaded_mask &= ~(1 << dev->devno);
2939 tf.command = ATA_CMD_CHK_POWER;
2942 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2943 tf.protocol |= ATA_PROT_NODATA;
2944 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2945 if (park && (err_mask || tf.lbal != 0xc4)) {
2946 ata_dev_printk(dev, KERN_ERR, "head unload failed!\n");
2947 ehc->unloaded_mask &= ~(1 << dev->devno);
2951 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2952 struct ata_device **r_failed_dev)
2954 struct ata_port *ap = link->ap;
2955 struct ata_eh_context *ehc = &link->eh_context;
2956 struct ata_device *dev;
2957 unsigned int new_mask = 0;
2958 unsigned long flags;
2959 int rc = 0;
2961 DPRINTK("ENTER\n");
2963 /* For PATA drive side cable detection to work, IDENTIFY must
2964 * be done backwards such that PDIAG- is released by the slave
2965 * device before the master device is identified.
2967 ata_for_each_dev(dev, link, ALL_REVERSE) {
2968 unsigned int action = ata_eh_dev_action(dev);
2969 unsigned int readid_flags = 0;
2971 if (ehc->i.flags & ATA_EHI_DID_RESET)
2972 readid_flags |= ATA_READID_POSTRESET;
2974 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2975 WARN_ON(dev->class == ATA_DEV_PMP);
2977 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2978 rc = -EIO;
2979 goto err;
2982 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2983 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2984 readid_flags);
2985 if (rc)
2986 goto err;
2988 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2990 /* Configuration may have changed, reconfigure
2991 * transfer mode.
2993 ehc->i.flags |= ATA_EHI_SETMODE;
2995 /* schedule the scsi_rescan_device() here */
2996 schedule_work(&(ap->scsi_rescan_task));
2997 } else if (dev->class == ATA_DEV_UNKNOWN &&
2998 ehc->tries[dev->devno] &&
2999 ata_class_enabled(ehc->classes[dev->devno])) {
3000 /* Temporarily set dev->class, it will be
3001 * permanently set once all configurations are
3002 * complete. This is necessary because new
3003 * device configuration is done in two
3004 * separate loops.
3006 dev->class = ehc->classes[dev->devno];
3008 if (dev->class == ATA_DEV_PMP)
3009 rc = sata_pmp_attach(dev);
3010 else
3011 rc = ata_dev_read_id(dev, &dev->class,
3012 readid_flags, dev->id);
3014 /* read_id might have changed class, store and reset */
3015 ehc->classes[dev->devno] = dev->class;
3016 dev->class = ATA_DEV_UNKNOWN;
3018 switch (rc) {
3019 case 0:
3020 /* clear error info accumulated during probe */
3021 ata_ering_clear(&dev->ering);
3022 new_mask |= 1 << dev->devno;
3023 break;
3024 case -ENOENT:
3025 /* IDENTIFY was issued to non-existent
3026 * device. No need to reset. Just
3027 * thaw and ignore the device.
3029 ata_eh_thaw_port(ap);
3030 break;
3031 default:
3032 goto err;
3037 /* PDIAG- should have been released, ask cable type if post-reset */
3038 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3039 if (ap->ops->cable_detect)
3040 ap->cbl = ap->ops->cable_detect(ap);
3041 ata_force_cbl(ap);
3044 /* Configure new devices forward such that user doesn't see
3045 * device detection messages backwards.
3047 ata_for_each_dev(dev, link, ALL) {
3048 if (!(new_mask & (1 << dev->devno)))
3049 continue;
3051 dev->class = ehc->classes[dev->devno];
3053 if (dev->class == ATA_DEV_PMP)
3054 continue;
3056 ehc->i.flags |= ATA_EHI_PRINTINFO;
3057 rc = ata_dev_configure(dev);
3058 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3059 if (rc) {
3060 dev->class = ATA_DEV_UNKNOWN;
3061 goto err;
3064 spin_lock_irqsave(ap->lock, flags);
3065 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3066 spin_unlock_irqrestore(ap->lock, flags);
3068 /* new device discovered, configure xfermode */
3069 ehc->i.flags |= ATA_EHI_SETMODE;
3072 return 0;
3074 err:
3075 *r_failed_dev = dev;
3076 DPRINTK("EXIT rc=%d\n", rc);
3077 return rc;
3081 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3082 * @link: link on which timings will be programmed
3083 * @r_failed_dev: out parameter for failed device
3085 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3086 * ata_set_mode() fails, pointer to the failing device is
3087 * returned in @r_failed_dev.
3089 * LOCKING:
3090 * PCI/etc. bus probe sem.
3092 * RETURNS:
3093 * 0 on success, negative errno otherwise
3095 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3097 struct ata_port *ap = link->ap;
3098 struct ata_device *dev;
3099 int rc;
3101 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3102 ata_for_each_dev(dev, link, ENABLED) {
3103 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3104 struct ata_ering_entry *ent;
3106 ent = ata_ering_top(&dev->ering);
3107 if (ent)
3108 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3112 /* has private set_mode? */
3113 if (ap->ops->set_mode)
3114 rc = ap->ops->set_mode(link, r_failed_dev);
3115 else
3116 rc = ata_do_set_mode(link, r_failed_dev);
3118 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3119 ata_for_each_dev(dev, link, ENABLED) {
3120 struct ata_eh_context *ehc = &link->eh_context;
3121 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3122 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3124 if (dev->xfer_mode != saved_xfer_mode ||
3125 ata_ncq_enabled(dev) != saved_ncq)
3126 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3129 return rc;
3133 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3134 * @dev: ATAPI device to clear UA for
3136 * Resets and other operations can make an ATAPI device raise
3137 * UNIT ATTENTION which causes the next operation to fail. This
3138 * function clears UA.
3140 * LOCKING:
3141 * EH context (may sleep).
3143 * RETURNS:
3144 * 0 on success, -errno on failure.
3146 static int atapi_eh_clear_ua(struct ata_device *dev)
3148 int i;
3150 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3151 u8 *sense_buffer = dev->link->ap->sector_buf;
3152 u8 sense_key = 0;
3153 unsigned int err_mask;
3155 err_mask = atapi_eh_tur(dev, &sense_key);
3156 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3157 ata_dev_printk(dev, KERN_WARNING, "TEST_UNIT_READY "
3158 "failed (err_mask=0x%x)\n", err_mask);
3159 return -EIO;
3162 if (!err_mask || sense_key != UNIT_ATTENTION)
3163 return 0;
3165 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3166 if (err_mask) {
3167 ata_dev_printk(dev, KERN_WARNING, "failed to clear "
3168 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3169 return -EIO;
3173 ata_dev_printk(dev, KERN_WARNING,
3174 "UNIT ATTENTION persists after %d tries\n", ATA_EH_UA_TRIES);
3176 return 0;
3180 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3181 * @dev: ATA device which may need FLUSH retry
3183 * If @dev failed FLUSH, it needs to be reported upper layer
3184 * immediately as it means that @dev failed to remap and already
3185 * lost at least a sector and further FLUSH retrials won't make
3186 * any difference to the lost sector. However, if FLUSH failed
3187 * for other reasons, for example transmission error, FLUSH needs
3188 * to be retried.
3190 * This function determines whether FLUSH failure retry is
3191 * necessary and performs it if so.
3193 * RETURNS:
3194 * 0 if EH can continue, -errno if EH needs to be repeated.
3196 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3198 struct ata_link *link = dev->link;
3199 struct ata_port *ap = link->ap;
3200 struct ata_queued_cmd *qc;
3201 struct ata_taskfile tf;
3202 unsigned int err_mask;
3203 int rc = 0;
3205 /* did flush fail for this device? */
3206 if (!ata_tag_valid(link->active_tag))
3207 return 0;
3209 qc = __ata_qc_from_tag(ap, link->active_tag);
3210 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3211 qc->tf.command != ATA_CMD_FLUSH))
3212 return 0;
3214 /* if the device failed it, it should be reported to upper layers */
3215 if (qc->err_mask & AC_ERR_DEV)
3216 return 0;
3218 /* flush failed for some other reason, give it another shot */
3219 ata_tf_init(dev, &tf);
3221 tf.command = qc->tf.command;
3222 tf.flags |= ATA_TFLAG_DEVICE;
3223 tf.protocol = ATA_PROT_NODATA;
3225 ata_dev_printk(dev, KERN_WARNING, "retrying FLUSH 0x%x Emask 0x%x\n",
3226 tf.command, qc->err_mask);
3228 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3229 if (!err_mask) {
3231 * FLUSH is complete but there's no way to
3232 * successfully complete a failed command from EH.
3233 * Making sure retry is allowed at least once and
3234 * retrying it should do the trick - whatever was in
3235 * the cache is already on the platter and this won't
3236 * cause infinite loop.
3238 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3239 } else {
3240 ata_dev_printk(dev, KERN_WARNING, "FLUSH failed Emask 0x%x\n",
3241 err_mask);
3242 rc = -EIO;
3244 /* if device failed it, report it to upper layers */
3245 if (err_mask & AC_ERR_DEV) {
3246 qc->err_mask |= AC_ERR_DEV;
3247 qc->result_tf = tf;
3248 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3249 rc = 0;
3252 return rc;
3256 * ata_eh_set_lpm - configure SATA interface power management
3257 * @link: link to configure power management
3258 * @policy: the link power management policy
3259 * @r_failed_dev: out parameter for failed device
3261 * Enable SATA Interface power management. This will enable
3262 * Device Interface Power Management (DIPM) for min_power
3263 * policy, and then call driver specific callbacks for
3264 * enabling Host Initiated Power management.
3266 * LOCKING:
3267 * EH context.
3269 * RETURNS:
3270 * 0 on success, -errno on failure.
3272 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3273 struct ata_device **r_failed_dev)
3275 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3276 struct ata_eh_context *ehc = &link->eh_context;
3277 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3278 enum ata_lpm_policy old_policy = link->lpm_policy;
3279 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3280 unsigned int err_mask;
3281 int rc;
3283 /* if the link or host doesn't do LPM, noop */
3284 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3285 return 0;
3288 * DIPM is enabled only for MIN_POWER as some devices
3289 * misbehave when the host NACKs transition to SLUMBER. Order
3290 * device and link configurations such that the host always
3291 * allows DIPM requests.
3293 ata_for_each_dev(dev, link, ENABLED) {
3294 bool hipm = ata_id_has_hipm(dev->id);
3295 bool dipm = ata_id_has_dipm(dev->id);
3297 /* find the first enabled and LPM enabled devices */
3298 if (!link_dev)
3299 link_dev = dev;
3301 if (!lpm_dev && (hipm || dipm))
3302 lpm_dev = dev;
3304 hints &= ~ATA_LPM_EMPTY;
3305 if (!hipm)
3306 hints &= ~ATA_LPM_HIPM;
3308 /* disable DIPM before changing link config */
3309 if (policy != ATA_LPM_MIN_POWER && dipm) {
3310 err_mask = ata_dev_set_feature(dev,
3311 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3312 if (err_mask && err_mask != AC_ERR_DEV) {
3313 ata_dev_printk(dev, KERN_WARNING,
3314 "failed to disable DIPM, Emask 0x%x\n",
3315 err_mask);
3316 rc = -EIO;
3317 goto fail;
3322 if (ap) {
3323 rc = ap->ops->set_lpm(link, policy, hints);
3324 if (!rc && ap->slave_link)
3325 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3326 } else
3327 rc = sata_pmp_set_lpm(link, policy, hints);
3330 * Attribute link config failure to the first (LPM) enabled
3331 * device on the link.
3333 if (rc) {
3334 if (rc == -EOPNOTSUPP) {
3335 link->flags |= ATA_LFLAG_NO_LPM;
3336 return 0;
3338 dev = lpm_dev ? lpm_dev : link_dev;
3339 goto fail;
3343 * Low level driver acked the transition. Issue DIPM command
3344 * with the new policy set.
3346 link->lpm_policy = policy;
3347 if (ap && ap->slave_link)
3348 ap->slave_link->lpm_policy = policy;
3350 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3351 ata_for_each_dev(dev, link, ENABLED) {
3352 if (policy == ATA_LPM_MIN_POWER && ata_id_has_dipm(dev->id)) {
3353 err_mask = ata_dev_set_feature(dev,
3354 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3355 if (err_mask && err_mask != AC_ERR_DEV) {
3356 ata_dev_printk(dev, KERN_WARNING,
3357 "failed to enable DIPM, Emask 0x%x\n",
3358 err_mask);
3359 rc = -EIO;
3360 goto fail;
3365 return 0;
3367 fail:
3368 /* restore the old policy */
3369 link->lpm_policy = old_policy;
3370 if (ap && ap->slave_link)
3371 ap->slave_link->lpm_policy = old_policy;
3373 /* if no device or only one more chance is left, disable LPM */
3374 if (!dev || ehc->tries[dev->devno] <= 2) {
3375 ata_link_printk(link, KERN_WARNING,
3376 "disabling LPM on the link\n");
3377 link->flags |= ATA_LFLAG_NO_LPM;
3379 if (r_failed_dev)
3380 *r_failed_dev = dev;
3381 return rc;
3384 static int ata_link_nr_enabled(struct ata_link *link)
3386 struct ata_device *dev;
3387 int cnt = 0;
3389 ata_for_each_dev(dev, link, ENABLED)
3390 cnt++;
3391 return cnt;
3394 static int ata_link_nr_vacant(struct ata_link *link)
3396 struct ata_device *dev;
3397 int cnt = 0;
3399 ata_for_each_dev(dev, link, ALL)
3400 if (dev->class == ATA_DEV_UNKNOWN)
3401 cnt++;
3402 return cnt;
3405 static int ata_eh_skip_recovery(struct ata_link *link)
3407 struct ata_port *ap = link->ap;
3408 struct ata_eh_context *ehc = &link->eh_context;
3409 struct ata_device *dev;
3411 /* skip disabled links */
3412 if (link->flags & ATA_LFLAG_DISABLED)
3413 return 1;
3415 /* skip if explicitly requested */
3416 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3417 return 1;
3419 /* thaw frozen port and recover failed devices */
3420 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3421 return 0;
3423 /* reset at least once if reset is requested */
3424 if ((ehc->i.action & ATA_EH_RESET) &&
3425 !(ehc->i.flags & ATA_EHI_DID_RESET))
3426 return 0;
3428 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3429 ata_for_each_dev(dev, link, ALL) {
3430 if (dev->class == ATA_DEV_UNKNOWN &&
3431 ehc->classes[dev->devno] != ATA_DEV_NONE)
3432 return 0;
3435 return 1;
3438 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3440 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3441 u64 now = get_jiffies_64();
3442 int *trials = void_arg;
3444 if (ent->timestamp < now - min(now, interval))
3445 return -1;
3447 (*trials)++;
3448 return 0;
3451 static int ata_eh_schedule_probe(struct ata_device *dev)
3453 struct ata_eh_context *ehc = &dev->link->eh_context;
3454 struct ata_link *link = ata_dev_phys_link(dev);
3455 int trials = 0;
3457 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3458 (ehc->did_probe_mask & (1 << dev->devno)))
3459 return 0;
3461 ata_eh_detach_dev(dev);
3462 ata_dev_init(dev);
3463 ehc->did_probe_mask |= (1 << dev->devno);
3464 ehc->i.action |= ATA_EH_RESET;
3465 ehc->saved_xfer_mode[dev->devno] = 0;
3466 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3468 /* the link maybe in a deep sleep, wake it up */
3469 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3470 if (ata_is_host_link(link))
3471 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3472 ATA_LPM_EMPTY);
3473 else
3474 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3475 ATA_LPM_EMPTY);
3478 /* Record and count probe trials on the ering. The specific
3479 * error mask used is irrelevant. Because a successful device
3480 * detection clears the ering, this count accumulates only if
3481 * there are consecutive failed probes.
3483 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3484 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3485 * forced to 1.5Gbps.
3487 * This is to work around cases where failed link speed
3488 * negotiation results in device misdetection leading to
3489 * infinite DEVXCHG or PHRDY CHG events.
3491 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3492 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3494 if (trials > ATA_EH_PROBE_TRIALS)
3495 sata_down_spd_limit(link, 1);
3497 return 1;
3500 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3502 struct ata_eh_context *ehc = &dev->link->eh_context;
3504 /* -EAGAIN from EH routine indicates retry without prejudice.
3505 * The requester is responsible for ensuring forward progress.
3507 if (err != -EAGAIN)
3508 ehc->tries[dev->devno]--;
3510 switch (err) {
3511 case -ENODEV:
3512 /* device missing or wrong IDENTIFY data, schedule probing */
3513 ehc->i.probe_mask |= (1 << dev->devno);
3514 case -EINVAL:
3515 /* give it just one more chance */
3516 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3517 case -EIO:
3518 if (ehc->tries[dev->devno] == 1) {
3519 /* This is the last chance, better to slow
3520 * down than lose it.
3522 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3523 if (dev->pio_mode > XFER_PIO_0)
3524 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3528 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3529 /* disable device if it has used up all its chances */
3530 ata_dev_disable(dev);
3532 /* detach if offline */
3533 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3534 ata_eh_detach_dev(dev);
3536 /* schedule probe if necessary */
3537 if (ata_eh_schedule_probe(dev)) {
3538 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3539 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3540 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3543 return 1;
3544 } else {
3545 ehc->i.action |= ATA_EH_RESET;
3546 return 0;
3551 * ata_eh_recover - recover host port after error
3552 * @ap: host port to recover
3553 * @prereset: prereset method (can be NULL)
3554 * @softreset: softreset method (can be NULL)
3555 * @hardreset: hardreset method (can be NULL)
3556 * @postreset: postreset method (can be NULL)
3557 * @r_failed_link: out parameter for failed link
3559 * This is the alpha and omega, eum and yang, heart and soul of
3560 * libata exception handling. On entry, actions required to
3561 * recover each link and hotplug requests are recorded in the
3562 * link's eh_context. This function executes all the operations
3563 * with appropriate retrials and fallbacks to resurrect failed
3564 * devices, detach goners and greet newcomers.
3566 * LOCKING:
3567 * Kernel thread context (may sleep).
3569 * RETURNS:
3570 * 0 on success, -errno on failure.
3572 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3573 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3574 ata_postreset_fn_t postreset,
3575 struct ata_link **r_failed_link)
3577 struct ata_link *link;
3578 struct ata_device *dev;
3579 int rc, nr_fails;
3580 unsigned long flags, deadline;
3582 DPRINTK("ENTER\n");
3584 /* prep for recovery */
3585 ata_for_each_link(link, ap, EDGE) {
3586 struct ata_eh_context *ehc = &link->eh_context;
3588 /* re-enable link? */
3589 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3590 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3591 spin_lock_irqsave(ap->lock, flags);
3592 link->flags &= ~ATA_LFLAG_DISABLED;
3593 spin_unlock_irqrestore(ap->lock, flags);
3594 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3597 ata_for_each_dev(dev, link, ALL) {
3598 if (link->flags & ATA_LFLAG_NO_RETRY)
3599 ehc->tries[dev->devno] = 1;
3600 else
3601 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3603 /* collect port action mask recorded in dev actions */
3604 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3605 ~ATA_EH_PERDEV_MASK;
3606 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3608 /* process hotplug request */
3609 if (dev->flags & ATA_DFLAG_DETACH)
3610 ata_eh_detach_dev(dev);
3612 /* schedule probe if necessary */
3613 if (!ata_dev_enabled(dev))
3614 ata_eh_schedule_probe(dev);
3618 retry:
3619 rc = 0;
3621 /* if UNLOADING, finish immediately */
3622 if (ap->pflags & ATA_PFLAG_UNLOADING)
3623 goto out;
3625 /* prep for EH */
3626 ata_for_each_link(link, ap, EDGE) {
3627 struct ata_eh_context *ehc = &link->eh_context;
3629 /* skip EH if possible. */
3630 if (ata_eh_skip_recovery(link))
3631 ehc->i.action = 0;
3633 ata_for_each_dev(dev, link, ALL)
3634 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3637 /* reset */
3638 ata_for_each_link(link, ap, EDGE) {
3639 struct ata_eh_context *ehc = &link->eh_context;
3641 if (!(ehc->i.action & ATA_EH_RESET))
3642 continue;
3644 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3645 prereset, softreset, hardreset, postreset);
3646 if (rc) {
3647 ata_link_printk(link, KERN_ERR,
3648 "reset failed, giving up\n");
3649 goto out;
3653 do {
3654 unsigned long now;
3657 * clears ATA_EH_PARK in eh_info and resets
3658 * ap->park_req_pending
3660 ata_eh_pull_park_action(ap);
3662 deadline = jiffies;
3663 ata_for_each_link(link, ap, EDGE) {
3664 ata_for_each_dev(dev, link, ALL) {
3665 struct ata_eh_context *ehc = &link->eh_context;
3666 unsigned long tmp;
3668 if (dev->class != ATA_DEV_ATA)
3669 continue;
3670 if (!(ehc->i.dev_action[dev->devno] &
3671 ATA_EH_PARK))
3672 continue;
3673 tmp = dev->unpark_deadline;
3674 if (time_before(deadline, tmp))
3675 deadline = tmp;
3676 else if (time_before_eq(tmp, jiffies))
3677 continue;
3678 if (ehc->unloaded_mask & (1 << dev->devno))
3679 continue;
3681 ata_eh_park_issue_cmd(dev, 1);
3685 now = jiffies;
3686 if (time_before_eq(deadline, now))
3687 break;
3689 ata_eh_release(ap);
3690 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3691 deadline - now);
3692 ata_eh_acquire(ap);
3693 } while (deadline);
3694 ata_for_each_link(link, ap, EDGE) {
3695 ata_for_each_dev(dev, link, ALL) {
3696 if (!(link->eh_context.unloaded_mask &
3697 (1 << dev->devno)))
3698 continue;
3700 ata_eh_park_issue_cmd(dev, 0);
3701 ata_eh_done(link, dev, ATA_EH_PARK);
3705 /* the rest */
3706 nr_fails = 0;
3707 ata_for_each_link(link, ap, PMP_FIRST) {
3708 struct ata_eh_context *ehc = &link->eh_context;
3710 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3711 goto config_lpm;
3713 /* revalidate existing devices and attach new ones */
3714 rc = ata_eh_revalidate_and_attach(link, &dev);
3715 if (rc)
3716 goto rest_fail;
3718 /* if PMP got attached, return, pmp EH will take care of it */
3719 if (link->device->class == ATA_DEV_PMP) {
3720 ehc->i.action = 0;
3721 return 0;
3724 /* configure transfer mode if necessary */
3725 if (ehc->i.flags & ATA_EHI_SETMODE) {
3726 rc = ata_set_mode(link, &dev);
3727 if (rc)
3728 goto rest_fail;
3729 ehc->i.flags &= ~ATA_EHI_SETMODE;
3732 /* If reset has been issued, clear UA to avoid
3733 * disrupting the current users of the device.
3735 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3736 ata_for_each_dev(dev, link, ALL) {
3737 if (dev->class != ATA_DEV_ATAPI)
3738 continue;
3739 rc = atapi_eh_clear_ua(dev);
3740 if (rc)
3741 goto rest_fail;
3745 /* retry flush if necessary */
3746 ata_for_each_dev(dev, link, ALL) {
3747 if (dev->class != ATA_DEV_ATA)
3748 continue;
3749 rc = ata_eh_maybe_retry_flush(dev);
3750 if (rc)
3751 goto rest_fail;
3754 config_lpm:
3755 /* configure link power saving */
3756 if (link->lpm_policy != ap->target_lpm_policy) {
3757 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3758 if (rc)
3759 goto rest_fail;
3762 /* this link is okay now */
3763 ehc->i.flags = 0;
3764 continue;
3766 rest_fail:
3767 nr_fails++;
3768 if (dev)
3769 ata_eh_handle_dev_fail(dev, rc);
3771 if (ap->pflags & ATA_PFLAG_FROZEN) {
3772 /* PMP reset requires working host port.
3773 * Can't retry if it's frozen.
3775 if (sata_pmp_attached(ap))
3776 goto out;
3777 break;
3781 if (nr_fails)
3782 goto retry;
3784 out:
3785 if (rc && r_failed_link)
3786 *r_failed_link = link;
3788 DPRINTK("EXIT, rc=%d\n", rc);
3789 return rc;
3793 * ata_eh_finish - finish up EH
3794 * @ap: host port to finish EH for
3796 * Recovery is complete. Clean up EH states and retry or finish
3797 * failed qcs.
3799 * LOCKING:
3800 * None.
3802 void ata_eh_finish(struct ata_port *ap)
3804 int tag;
3806 /* retry or finish qcs */
3807 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3808 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3810 if (!(qc->flags & ATA_QCFLAG_FAILED))
3811 continue;
3813 if (qc->err_mask) {
3814 /* FIXME: Once EH migration is complete,
3815 * generate sense data in this function,
3816 * considering both err_mask and tf.
3818 if (qc->flags & ATA_QCFLAG_RETRY)
3819 ata_eh_qc_retry(qc);
3820 else
3821 ata_eh_qc_complete(qc);
3822 } else {
3823 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3824 ata_eh_qc_complete(qc);
3825 } else {
3826 /* feed zero TF to sense generation */
3827 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3828 ata_eh_qc_retry(qc);
3833 /* make sure nr_active_links is zero after EH */
3834 WARN_ON(ap->nr_active_links);
3835 ap->nr_active_links = 0;
3839 * ata_do_eh - do standard error handling
3840 * @ap: host port to handle error for
3842 * @prereset: prereset method (can be NULL)
3843 * @softreset: softreset method (can be NULL)
3844 * @hardreset: hardreset method (can be NULL)
3845 * @postreset: postreset method (can be NULL)
3847 * Perform standard error handling sequence.
3849 * LOCKING:
3850 * Kernel thread context (may sleep).
3852 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3853 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3854 ata_postreset_fn_t postreset)
3856 struct ata_device *dev;
3857 int rc;
3859 ata_eh_autopsy(ap);
3860 ata_eh_report(ap);
3862 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3863 NULL);
3864 if (rc) {
3865 ata_for_each_dev(dev, &ap->link, ALL)
3866 ata_dev_disable(dev);
3869 ata_eh_finish(ap);
3873 * ata_std_error_handler - standard error handler
3874 * @ap: host port to handle error for
3876 * Standard error handler
3878 * LOCKING:
3879 * Kernel thread context (may sleep).
3881 void ata_std_error_handler(struct ata_port *ap)
3883 struct ata_port_operations *ops = ap->ops;
3884 ata_reset_fn_t hardreset = ops->hardreset;
3886 /* ignore built-in hardreset if SCR access is not available */
3887 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3888 hardreset = NULL;
3890 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3893 #ifdef CONFIG_PM
3895 * ata_eh_handle_port_suspend - perform port suspend operation
3896 * @ap: port to suspend
3898 * Suspend @ap.
3900 * LOCKING:
3901 * Kernel thread context (may sleep).
3903 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3905 unsigned long flags;
3906 int rc = 0;
3908 /* are we suspending? */
3909 spin_lock_irqsave(ap->lock, flags);
3910 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3911 ap->pm_mesg.event == PM_EVENT_ON) {
3912 spin_unlock_irqrestore(ap->lock, flags);
3913 return;
3915 spin_unlock_irqrestore(ap->lock, flags);
3917 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3919 /* tell ACPI we're suspending */
3920 rc = ata_acpi_on_suspend(ap);
3921 if (rc)
3922 goto out;
3924 /* suspend */
3925 ata_eh_freeze_port(ap);
3927 if (ap->ops->port_suspend)
3928 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3930 ata_acpi_set_state(ap, PMSG_SUSPEND);
3931 out:
3932 /* report result */
3933 spin_lock_irqsave(ap->lock, flags);
3935 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3936 if (rc == 0)
3937 ap->pflags |= ATA_PFLAG_SUSPENDED;
3938 else if (ap->pflags & ATA_PFLAG_FROZEN)
3939 ata_port_schedule_eh(ap);
3941 if (ap->pm_result) {
3942 *ap->pm_result = rc;
3943 ap->pm_result = NULL;
3946 spin_unlock_irqrestore(ap->lock, flags);
3948 return;
3952 * ata_eh_handle_port_resume - perform port resume operation
3953 * @ap: port to resume
3955 * Resume @ap.
3957 * LOCKING:
3958 * Kernel thread context (may sleep).
3960 static void ata_eh_handle_port_resume(struct ata_port *ap)
3962 struct ata_link *link;
3963 struct ata_device *dev;
3964 unsigned long flags;
3965 int rc = 0;
3967 /* are we resuming? */
3968 spin_lock_irqsave(ap->lock, flags);
3969 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3970 ap->pm_mesg.event != PM_EVENT_ON) {
3971 spin_unlock_irqrestore(ap->lock, flags);
3972 return;
3974 spin_unlock_irqrestore(ap->lock, flags);
3976 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3979 * Error timestamps are in jiffies which doesn't run while
3980 * suspended and PHY events during resume isn't too uncommon.
3981 * When the two are combined, it can lead to unnecessary speed
3982 * downs if the machine is suspended and resumed repeatedly.
3983 * Clear error history.
3985 ata_for_each_link(link, ap, HOST_FIRST)
3986 ata_for_each_dev(dev, link, ALL)
3987 ata_ering_clear(&dev->ering);
3989 ata_acpi_set_state(ap, PMSG_ON);
3991 if (ap->ops->port_resume)
3992 rc = ap->ops->port_resume(ap);
3994 /* tell ACPI that we're resuming */
3995 ata_acpi_on_resume(ap);
3997 /* report result */
3998 spin_lock_irqsave(ap->lock, flags);
3999 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4000 if (ap->pm_result) {
4001 *ap->pm_result = rc;
4002 ap->pm_result = NULL;
4004 spin_unlock_irqrestore(ap->lock, flags);
4006 #endif /* CONFIG_PM */