Merge branch 'upstream' of git://git.linux-mips.org/pub/scm/upstream-linus
[linux-btrfs-devel.git] / drivers / ata / libata-eh.c
blobed16fbedaabd78020044996a70f4f4ad1d8879c8
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 unsigned long flags;
591 LIST_HEAD(eh_work_q);
593 DPRINTK("ENTER\n");
595 spin_lock_irqsave(host->host_lock, flags);
596 list_splice_init(&host->eh_cmd_q, &eh_work_q);
597 spin_unlock_irqrestore(host->host_lock, flags);
599 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
601 /* If we timed raced normal completion and there is nothing to
602 recover nr_timedout == 0 why exactly are we doing error recovery ? */
603 ata_scsi_port_error_handler(host, ap);
605 /* finish or retry handled scmd's and clean up */
606 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
608 DPRINTK("EXIT\n");
612 * ata_scsi_cmd_error_handler - error callback for a list of commands
613 * @host: scsi host containing the port
614 * @ap: ATA port within the host
615 * @eh_work_q: list of commands to process
617 * process the given list of commands and return those finished to the
618 * ap->eh_done_q. This function is the first part of the libata error
619 * handler which processes a given list of failed commands.
621 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
622 struct list_head *eh_work_q)
624 int i;
625 unsigned long flags;
627 /* make sure sff pio task is not running */
628 ata_sff_flush_pio_task(ap);
630 /* synchronize with host lock and sort out timeouts */
632 /* For new EH, all qcs are finished in one of three ways -
633 * normal completion, error completion, and SCSI timeout.
634 * Both completions can race against SCSI timeout. When normal
635 * completion wins, the qc never reaches EH. When error
636 * completion wins, the qc has ATA_QCFLAG_FAILED set.
638 * When SCSI timeout wins, things are a bit more complex.
639 * Normal or error completion can occur after the timeout but
640 * before this point. In such cases, both types of
641 * completions are honored. A scmd is determined to have
642 * timed out iff its associated qc is active and not failed.
644 if (ap->ops->error_handler) {
645 struct scsi_cmnd *scmd, *tmp;
646 int nr_timedout = 0;
648 spin_lock_irqsave(ap->lock, flags);
650 /* This must occur under the ap->lock as we don't want
651 a polled recovery to race the real interrupt handler
653 The lost_interrupt handler checks for any completed but
654 non-notified command and completes much like an IRQ handler.
656 We then fall into the error recovery code which will treat
657 this as if normal completion won the race */
659 if (ap->ops->lost_interrupt)
660 ap->ops->lost_interrupt(ap);
662 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
663 struct ata_queued_cmd *qc;
665 for (i = 0; i < ATA_MAX_QUEUE; i++) {
666 qc = __ata_qc_from_tag(ap, i);
667 if (qc->flags & ATA_QCFLAG_ACTIVE &&
668 qc->scsicmd == scmd)
669 break;
672 if (i < ATA_MAX_QUEUE) {
673 /* the scmd has an associated qc */
674 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
675 /* which hasn't failed yet, timeout */
676 qc->err_mask |= AC_ERR_TIMEOUT;
677 qc->flags |= ATA_QCFLAG_FAILED;
678 nr_timedout++;
680 } else {
681 /* Normal completion occurred after
682 * SCSI timeout but before this point.
683 * Successfully complete it.
685 scmd->retries = scmd->allowed;
686 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
690 /* If we have timed out qcs. They belong to EH from
691 * this point but the state of the controller is
692 * unknown. Freeze the port to make sure the IRQ
693 * handler doesn't diddle with those qcs. This must
694 * be done atomically w.r.t. setting QCFLAG_FAILED.
696 if (nr_timedout)
697 __ata_port_freeze(ap);
699 spin_unlock_irqrestore(ap->lock, flags);
701 /* initialize eh_tries */
702 ap->eh_tries = ATA_EH_MAX_TRIES;
703 } else
704 spin_unlock_wait(ap->lock);
707 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
710 * ata_scsi_port_error_handler - recover the port after the commands
711 * @host: SCSI host containing the port
712 * @ap: the ATA port
714 * Handle the recovery of the port @ap after all the commands
715 * have been recovered.
717 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
719 unsigned long flags;
721 /* invoke error handler */
722 if (ap->ops->error_handler) {
723 struct ata_link *link;
725 /* acquire EH ownership */
726 ata_eh_acquire(ap);
727 repeat:
728 /* kill fast drain timer */
729 del_timer_sync(&ap->fastdrain_timer);
731 /* process port resume request */
732 ata_eh_handle_port_resume(ap);
734 /* fetch & clear EH info */
735 spin_lock_irqsave(ap->lock, flags);
737 ata_for_each_link(link, ap, HOST_FIRST) {
738 struct ata_eh_context *ehc = &link->eh_context;
739 struct ata_device *dev;
741 memset(&link->eh_context, 0, sizeof(link->eh_context));
742 link->eh_context.i = link->eh_info;
743 memset(&link->eh_info, 0, sizeof(link->eh_info));
745 ata_for_each_dev(dev, link, ENABLED) {
746 int devno = dev->devno;
748 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
749 if (ata_ncq_enabled(dev))
750 ehc->saved_ncq_enabled |= 1 << devno;
754 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
755 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
756 ap->excl_link = NULL; /* don't maintain exclusion over EH */
758 spin_unlock_irqrestore(ap->lock, flags);
760 /* invoke EH, skip if unloading or suspended */
761 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
762 ap->ops->error_handler(ap);
763 else {
764 /* if unloading, commence suicide */
765 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
766 !(ap->pflags & ATA_PFLAG_UNLOADED))
767 ata_eh_unload(ap);
768 ata_eh_finish(ap);
771 /* process port suspend request */
772 ata_eh_handle_port_suspend(ap);
774 /* Exception might have happened after ->error_handler
775 * recovered the port but before this point. Repeat
776 * EH in such case.
778 spin_lock_irqsave(ap->lock, flags);
780 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
781 if (--ap->eh_tries) {
782 spin_unlock_irqrestore(ap->lock, flags);
783 goto repeat;
785 ata_port_err(ap,
786 "EH pending after %d tries, giving up\n",
787 ATA_EH_MAX_TRIES);
788 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
791 /* this run is complete, make sure EH info is clear */
792 ata_for_each_link(link, ap, HOST_FIRST)
793 memset(&link->eh_info, 0, sizeof(link->eh_info));
795 /* Clear host_eh_scheduled while holding ap->lock such
796 * that if exception occurs after this point but
797 * before EH completion, SCSI midlayer will
798 * re-initiate EH.
800 host->host_eh_scheduled = 0;
802 spin_unlock_irqrestore(ap->lock, flags);
803 ata_eh_release(ap);
804 } else {
805 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
806 ap->ops->eng_timeout(ap);
809 scsi_eh_flush_done_q(&ap->eh_done_q);
811 /* clean up */
812 spin_lock_irqsave(ap->lock, flags);
814 if (ap->pflags & ATA_PFLAG_LOADING)
815 ap->pflags &= ~ATA_PFLAG_LOADING;
816 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
817 schedule_delayed_work(&ap->hotplug_task, 0);
819 if (ap->pflags & ATA_PFLAG_RECOVERED)
820 ata_port_info(ap, "EH complete\n");
822 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
824 /* tell wait_eh that we're done */
825 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
826 wake_up_all(&ap->eh_wait_q);
828 spin_unlock_irqrestore(ap->lock, flags);
830 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
833 * ata_port_wait_eh - Wait for the currently pending EH to complete
834 * @ap: Port to wait EH for
836 * Wait until the currently pending EH is complete.
838 * LOCKING:
839 * Kernel thread context (may sleep).
841 void ata_port_wait_eh(struct ata_port *ap)
843 unsigned long flags;
844 DEFINE_WAIT(wait);
846 retry:
847 spin_lock_irqsave(ap->lock, flags);
849 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
850 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
851 spin_unlock_irqrestore(ap->lock, flags);
852 schedule();
853 spin_lock_irqsave(ap->lock, flags);
855 finish_wait(&ap->eh_wait_q, &wait);
857 spin_unlock_irqrestore(ap->lock, flags);
859 /* make sure SCSI EH is complete */
860 if (scsi_host_in_recovery(ap->scsi_host)) {
861 ata_msleep(ap, 10);
862 goto retry;
866 static int ata_eh_nr_in_flight(struct ata_port *ap)
868 unsigned int tag;
869 int nr = 0;
871 /* count only non-internal commands */
872 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
873 if (ata_qc_from_tag(ap, tag))
874 nr++;
876 return nr;
879 void ata_eh_fastdrain_timerfn(unsigned long arg)
881 struct ata_port *ap = (void *)arg;
882 unsigned long flags;
883 int cnt;
885 spin_lock_irqsave(ap->lock, flags);
887 cnt = ata_eh_nr_in_flight(ap);
889 /* are we done? */
890 if (!cnt)
891 goto out_unlock;
893 if (cnt == ap->fastdrain_cnt) {
894 unsigned int tag;
896 /* No progress during the last interval, tag all
897 * in-flight qcs as timed out and freeze the port.
899 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
900 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
901 if (qc)
902 qc->err_mask |= AC_ERR_TIMEOUT;
905 ata_port_freeze(ap);
906 } else {
907 /* some qcs have finished, give it another chance */
908 ap->fastdrain_cnt = cnt;
909 ap->fastdrain_timer.expires =
910 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
911 add_timer(&ap->fastdrain_timer);
914 out_unlock:
915 spin_unlock_irqrestore(ap->lock, flags);
919 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
920 * @ap: target ATA port
921 * @fastdrain: activate fast drain
923 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
924 * is non-zero and EH wasn't pending before. Fast drain ensures
925 * that EH kicks in in timely manner.
927 * LOCKING:
928 * spin_lock_irqsave(host lock)
930 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
932 int cnt;
934 /* already scheduled? */
935 if (ap->pflags & ATA_PFLAG_EH_PENDING)
936 return;
938 ap->pflags |= ATA_PFLAG_EH_PENDING;
940 if (!fastdrain)
941 return;
943 /* do we have in-flight qcs? */
944 cnt = ata_eh_nr_in_flight(ap);
945 if (!cnt)
946 return;
948 /* activate fast drain */
949 ap->fastdrain_cnt = cnt;
950 ap->fastdrain_timer.expires =
951 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
952 add_timer(&ap->fastdrain_timer);
956 * ata_qc_schedule_eh - schedule qc for error handling
957 * @qc: command to schedule error handling for
959 * Schedule error handling for @qc. EH will kick in as soon as
960 * other commands are drained.
962 * LOCKING:
963 * spin_lock_irqsave(host lock)
965 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
967 struct ata_port *ap = qc->ap;
968 struct request_queue *q = qc->scsicmd->device->request_queue;
969 unsigned long flags;
971 WARN_ON(!ap->ops->error_handler);
973 qc->flags |= ATA_QCFLAG_FAILED;
974 ata_eh_set_pending(ap, 1);
976 /* The following will fail if timeout has already expired.
977 * ata_scsi_error() takes care of such scmds on EH entry.
978 * Note that ATA_QCFLAG_FAILED is unconditionally set after
979 * this function completes.
981 spin_lock_irqsave(q->queue_lock, flags);
982 blk_abort_request(qc->scsicmd->request);
983 spin_unlock_irqrestore(q->queue_lock, flags);
987 * ata_port_schedule_eh - schedule error handling without a qc
988 * @ap: ATA port to schedule EH for
990 * Schedule error handling for @ap. EH will kick in as soon as
991 * all commands are drained.
993 * LOCKING:
994 * spin_lock_irqsave(host lock)
996 void ata_port_schedule_eh(struct ata_port *ap)
998 WARN_ON(!ap->ops->error_handler);
1000 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1001 return;
1003 ata_eh_set_pending(ap, 1);
1004 scsi_schedule_eh(ap->scsi_host);
1006 DPRINTK("port EH scheduled\n");
1009 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1011 int tag, nr_aborted = 0;
1013 WARN_ON(!ap->ops->error_handler);
1015 /* we're gonna abort all commands, no need for fast drain */
1016 ata_eh_set_pending(ap, 0);
1018 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1019 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1021 if (qc && (!link || qc->dev->link == link)) {
1022 qc->flags |= ATA_QCFLAG_FAILED;
1023 ata_qc_complete(qc);
1024 nr_aborted++;
1028 if (!nr_aborted)
1029 ata_port_schedule_eh(ap);
1031 return nr_aborted;
1035 * ata_link_abort - abort all qc's on the link
1036 * @link: ATA link to abort qc's for
1038 * Abort all active qc's active on @link and schedule EH.
1040 * LOCKING:
1041 * spin_lock_irqsave(host lock)
1043 * RETURNS:
1044 * Number of aborted qc's.
1046 int ata_link_abort(struct ata_link *link)
1048 return ata_do_link_abort(link->ap, link);
1052 * ata_port_abort - abort all qc's on the port
1053 * @ap: ATA port to abort qc's for
1055 * Abort all active qc's of @ap and schedule EH.
1057 * LOCKING:
1058 * spin_lock_irqsave(host_set lock)
1060 * RETURNS:
1061 * Number of aborted qc's.
1063 int ata_port_abort(struct ata_port *ap)
1065 return ata_do_link_abort(ap, NULL);
1069 * __ata_port_freeze - freeze port
1070 * @ap: ATA port to freeze
1072 * This function is called when HSM violation or some other
1073 * condition disrupts normal operation of the port. Frozen port
1074 * is not allowed to perform any operation until the port is
1075 * thawed, which usually follows a successful reset.
1077 * ap->ops->freeze() callback can be used for freezing the port
1078 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1079 * port cannot be frozen hardware-wise, the interrupt handler
1080 * must ack and clear interrupts unconditionally while the port
1081 * is frozen.
1083 * LOCKING:
1084 * spin_lock_irqsave(host lock)
1086 static void __ata_port_freeze(struct ata_port *ap)
1088 WARN_ON(!ap->ops->error_handler);
1090 if (ap->ops->freeze)
1091 ap->ops->freeze(ap);
1093 ap->pflags |= ATA_PFLAG_FROZEN;
1095 DPRINTK("ata%u port frozen\n", ap->print_id);
1099 * ata_port_freeze - abort & freeze port
1100 * @ap: ATA port to freeze
1102 * Abort and freeze @ap. The freeze operation must be called
1103 * first, because some hardware requires special operations
1104 * before the taskfile registers are accessible.
1106 * LOCKING:
1107 * spin_lock_irqsave(host lock)
1109 * RETURNS:
1110 * Number of aborted commands.
1112 int ata_port_freeze(struct ata_port *ap)
1114 int nr_aborted;
1116 WARN_ON(!ap->ops->error_handler);
1118 __ata_port_freeze(ap);
1119 nr_aborted = ata_port_abort(ap);
1121 return nr_aborted;
1125 * sata_async_notification - SATA async notification handler
1126 * @ap: ATA port where async notification is received
1128 * Handler to be called when async notification via SDB FIS is
1129 * received. This function schedules EH if necessary.
1131 * LOCKING:
1132 * spin_lock_irqsave(host lock)
1134 * RETURNS:
1135 * 1 if EH is scheduled, 0 otherwise.
1137 int sata_async_notification(struct ata_port *ap)
1139 u32 sntf;
1140 int rc;
1142 if (!(ap->flags & ATA_FLAG_AN))
1143 return 0;
1145 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1146 if (rc == 0)
1147 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1149 if (!sata_pmp_attached(ap) || rc) {
1150 /* PMP is not attached or SNTF is not available */
1151 if (!sata_pmp_attached(ap)) {
1152 /* PMP is not attached. Check whether ATAPI
1153 * AN is configured. If so, notify media
1154 * change.
1156 struct ata_device *dev = ap->link.device;
1158 if ((dev->class == ATA_DEV_ATAPI) &&
1159 (dev->flags & ATA_DFLAG_AN))
1160 ata_scsi_media_change_notify(dev);
1161 return 0;
1162 } else {
1163 /* PMP is attached but SNTF is not available.
1164 * ATAPI async media change notification is
1165 * not used. The PMP must be reporting PHY
1166 * status change, schedule EH.
1168 ata_port_schedule_eh(ap);
1169 return 1;
1171 } else {
1172 /* PMP is attached and SNTF is available */
1173 struct ata_link *link;
1175 /* check and notify ATAPI AN */
1176 ata_for_each_link(link, ap, EDGE) {
1177 if (!(sntf & (1 << link->pmp)))
1178 continue;
1180 if ((link->device->class == ATA_DEV_ATAPI) &&
1181 (link->device->flags & ATA_DFLAG_AN))
1182 ata_scsi_media_change_notify(link->device);
1185 /* If PMP is reporting that PHY status of some
1186 * downstream ports has changed, schedule EH.
1188 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1189 ata_port_schedule_eh(ap);
1190 return 1;
1193 return 0;
1198 * ata_eh_freeze_port - EH helper to freeze port
1199 * @ap: ATA port to freeze
1201 * Freeze @ap.
1203 * LOCKING:
1204 * None.
1206 void ata_eh_freeze_port(struct ata_port *ap)
1208 unsigned long flags;
1210 if (!ap->ops->error_handler)
1211 return;
1213 spin_lock_irqsave(ap->lock, flags);
1214 __ata_port_freeze(ap);
1215 spin_unlock_irqrestore(ap->lock, flags);
1219 * ata_port_thaw_port - EH helper to thaw port
1220 * @ap: ATA port to thaw
1222 * Thaw frozen port @ap.
1224 * LOCKING:
1225 * None.
1227 void ata_eh_thaw_port(struct ata_port *ap)
1229 unsigned long flags;
1231 if (!ap->ops->error_handler)
1232 return;
1234 spin_lock_irqsave(ap->lock, flags);
1236 ap->pflags &= ~ATA_PFLAG_FROZEN;
1238 if (ap->ops->thaw)
1239 ap->ops->thaw(ap);
1241 spin_unlock_irqrestore(ap->lock, flags);
1243 DPRINTK("ata%u port thawed\n", ap->print_id);
1246 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1248 /* nada */
1251 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1253 struct ata_port *ap = qc->ap;
1254 struct scsi_cmnd *scmd = qc->scsicmd;
1255 unsigned long flags;
1257 spin_lock_irqsave(ap->lock, flags);
1258 qc->scsidone = ata_eh_scsidone;
1259 __ata_qc_complete(qc);
1260 WARN_ON(ata_tag_valid(qc->tag));
1261 spin_unlock_irqrestore(ap->lock, flags);
1263 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1267 * ata_eh_qc_complete - Complete an active ATA command from EH
1268 * @qc: Command to complete
1270 * Indicate to the mid and upper layers that an ATA command has
1271 * completed. To be used from EH.
1273 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1275 struct scsi_cmnd *scmd = qc->scsicmd;
1276 scmd->retries = scmd->allowed;
1277 __ata_eh_qc_complete(qc);
1281 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1282 * @qc: Command to retry
1284 * Indicate to the mid and upper layers that an ATA command
1285 * should be retried. To be used from EH.
1287 * SCSI midlayer limits the number of retries to scmd->allowed.
1288 * scmd->retries is decremented for commands which get retried
1289 * due to unrelated failures (qc->err_mask is zero).
1291 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1293 struct scsi_cmnd *scmd = qc->scsicmd;
1294 if (!qc->err_mask && scmd->retries)
1295 scmd->retries--;
1296 __ata_eh_qc_complete(qc);
1300 * ata_dev_disable - disable ATA device
1301 * @dev: ATA device to disable
1303 * Disable @dev.
1305 * Locking:
1306 * EH context.
1308 void ata_dev_disable(struct ata_device *dev)
1310 if (!ata_dev_enabled(dev))
1311 return;
1313 if (ata_msg_drv(dev->link->ap))
1314 ata_dev_warn(dev, "disabled\n");
1315 ata_acpi_on_disable(dev);
1316 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1317 dev->class++;
1319 /* From now till the next successful probe, ering is used to
1320 * track probe failures. Clear accumulated device error info.
1322 ata_ering_clear(&dev->ering);
1326 * ata_eh_detach_dev - detach ATA device
1327 * @dev: ATA device to detach
1329 * Detach @dev.
1331 * LOCKING:
1332 * None.
1334 void ata_eh_detach_dev(struct ata_device *dev)
1336 struct ata_link *link = dev->link;
1337 struct ata_port *ap = link->ap;
1338 struct ata_eh_context *ehc = &link->eh_context;
1339 unsigned long flags;
1341 ata_dev_disable(dev);
1343 spin_lock_irqsave(ap->lock, flags);
1345 dev->flags &= ~ATA_DFLAG_DETACH;
1347 if (ata_scsi_offline_dev(dev)) {
1348 dev->flags |= ATA_DFLAG_DETACHED;
1349 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1352 /* clear per-dev EH info */
1353 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1354 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1355 ehc->saved_xfer_mode[dev->devno] = 0;
1356 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1358 spin_unlock_irqrestore(ap->lock, flags);
1362 * ata_eh_about_to_do - about to perform eh_action
1363 * @link: target ATA link
1364 * @dev: target ATA dev for per-dev action (can be NULL)
1365 * @action: action about to be performed
1367 * Called just before performing EH actions to clear related bits
1368 * in @link->eh_info such that eh actions are not unnecessarily
1369 * repeated.
1371 * LOCKING:
1372 * None.
1374 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1375 unsigned int action)
1377 struct ata_port *ap = link->ap;
1378 struct ata_eh_info *ehi = &link->eh_info;
1379 struct ata_eh_context *ehc = &link->eh_context;
1380 unsigned long flags;
1382 spin_lock_irqsave(ap->lock, flags);
1384 ata_eh_clear_action(link, dev, ehi, action);
1386 /* About to take EH action, set RECOVERED. Ignore actions on
1387 * slave links as master will do them again.
1389 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1390 ap->pflags |= ATA_PFLAG_RECOVERED;
1392 spin_unlock_irqrestore(ap->lock, flags);
1396 * ata_eh_done - EH action complete
1397 * @ap: target ATA port
1398 * @dev: target ATA dev for per-dev action (can be NULL)
1399 * @action: action just completed
1401 * Called right after performing EH actions to clear related bits
1402 * in @link->eh_context.
1404 * LOCKING:
1405 * None.
1407 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1408 unsigned int action)
1410 struct ata_eh_context *ehc = &link->eh_context;
1412 ata_eh_clear_action(link, dev, &ehc->i, action);
1416 * ata_err_string - convert err_mask to descriptive string
1417 * @err_mask: error mask to convert to string
1419 * Convert @err_mask to descriptive string. Errors are
1420 * prioritized according to severity and only the most severe
1421 * error is reported.
1423 * LOCKING:
1424 * None.
1426 * RETURNS:
1427 * Descriptive string for @err_mask
1429 static const char *ata_err_string(unsigned int err_mask)
1431 if (err_mask & AC_ERR_HOST_BUS)
1432 return "host bus error";
1433 if (err_mask & AC_ERR_ATA_BUS)
1434 return "ATA bus error";
1435 if (err_mask & AC_ERR_TIMEOUT)
1436 return "timeout";
1437 if (err_mask & AC_ERR_HSM)
1438 return "HSM violation";
1439 if (err_mask & AC_ERR_SYSTEM)
1440 return "internal error";
1441 if (err_mask & AC_ERR_MEDIA)
1442 return "media error";
1443 if (err_mask & AC_ERR_INVALID)
1444 return "invalid argument";
1445 if (err_mask & AC_ERR_DEV)
1446 return "device error";
1447 return "unknown error";
1451 * ata_read_log_page - read a specific log page
1452 * @dev: target device
1453 * @page: page to read
1454 * @buf: buffer to store read page
1455 * @sectors: number of sectors to read
1457 * Read log page using READ_LOG_EXT command.
1459 * LOCKING:
1460 * Kernel thread context (may sleep).
1462 * RETURNS:
1463 * 0 on success, AC_ERR_* mask otherwise.
1465 static unsigned int ata_read_log_page(struct ata_device *dev,
1466 u8 page, void *buf, unsigned int sectors)
1468 struct ata_taskfile tf;
1469 unsigned int err_mask;
1471 DPRINTK("read log page - page %d\n", page);
1473 ata_tf_init(dev, &tf);
1474 tf.command = ATA_CMD_READ_LOG_EXT;
1475 tf.lbal = page;
1476 tf.nsect = sectors;
1477 tf.hob_nsect = sectors >> 8;
1478 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1479 tf.protocol = ATA_PROT_PIO;
1481 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1482 buf, sectors * ATA_SECT_SIZE, 0);
1484 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1485 return err_mask;
1489 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1490 * @dev: Device to read log page 10h from
1491 * @tag: Resulting tag of the failed command
1492 * @tf: Resulting taskfile registers of the failed command
1494 * Read log page 10h to obtain NCQ error details and clear error
1495 * condition.
1497 * LOCKING:
1498 * Kernel thread context (may sleep).
1500 * RETURNS:
1501 * 0 on success, -errno otherwise.
1503 static int ata_eh_read_log_10h(struct ata_device *dev,
1504 int *tag, struct ata_taskfile *tf)
1506 u8 *buf = dev->link->ap->sector_buf;
1507 unsigned int err_mask;
1508 u8 csum;
1509 int i;
1511 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1512 if (err_mask)
1513 return -EIO;
1515 csum = 0;
1516 for (i = 0; i < ATA_SECT_SIZE; i++)
1517 csum += buf[i];
1518 if (csum)
1519 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1520 csum);
1522 if (buf[0] & 0x80)
1523 return -ENOENT;
1525 *tag = buf[0] & 0x1f;
1527 tf->command = buf[2];
1528 tf->feature = buf[3];
1529 tf->lbal = buf[4];
1530 tf->lbam = buf[5];
1531 tf->lbah = buf[6];
1532 tf->device = buf[7];
1533 tf->hob_lbal = buf[8];
1534 tf->hob_lbam = buf[9];
1535 tf->hob_lbah = buf[10];
1536 tf->nsect = buf[12];
1537 tf->hob_nsect = buf[13];
1539 return 0;
1543 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1544 * @dev: target ATAPI device
1545 * @r_sense_key: out parameter for sense_key
1547 * Perform ATAPI TEST_UNIT_READY.
1549 * LOCKING:
1550 * EH context (may sleep).
1552 * RETURNS:
1553 * 0 on success, AC_ERR_* mask on failure.
1555 static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1557 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1558 struct ata_taskfile tf;
1559 unsigned int err_mask;
1561 ata_tf_init(dev, &tf);
1563 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1564 tf.command = ATA_CMD_PACKET;
1565 tf.protocol = ATAPI_PROT_NODATA;
1567 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1568 if (err_mask == AC_ERR_DEV)
1569 *r_sense_key = tf.feature >> 4;
1570 return err_mask;
1574 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1575 * @dev: device to perform REQUEST_SENSE to
1576 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1577 * @dfl_sense_key: default sense key to use
1579 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1580 * SENSE. This function is EH helper.
1582 * LOCKING:
1583 * Kernel thread context (may sleep).
1585 * RETURNS:
1586 * 0 on success, AC_ERR_* mask on failure
1588 static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1589 u8 *sense_buf, u8 dfl_sense_key)
1591 u8 cdb[ATAPI_CDB_LEN] =
1592 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1593 struct ata_port *ap = dev->link->ap;
1594 struct ata_taskfile tf;
1596 DPRINTK("ATAPI request sense\n");
1598 /* FIXME: is this needed? */
1599 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1601 /* initialize sense_buf with the error register,
1602 * for the case where they are -not- overwritten
1604 sense_buf[0] = 0x70;
1605 sense_buf[2] = dfl_sense_key;
1607 /* some devices time out if garbage left in tf */
1608 ata_tf_init(dev, &tf);
1610 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1611 tf.command = ATA_CMD_PACKET;
1613 /* is it pointless to prefer PIO for "safety reasons"? */
1614 if (ap->flags & ATA_FLAG_PIO_DMA) {
1615 tf.protocol = ATAPI_PROT_DMA;
1616 tf.feature |= ATAPI_PKT_DMA;
1617 } else {
1618 tf.protocol = ATAPI_PROT_PIO;
1619 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1620 tf.lbah = 0;
1623 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1624 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1628 * ata_eh_analyze_serror - analyze SError for a failed port
1629 * @link: ATA link to analyze SError for
1631 * Analyze SError if available and further determine cause of
1632 * failure.
1634 * LOCKING:
1635 * None.
1637 static void ata_eh_analyze_serror(struct ata_link *link)
1639 struct ata_eh_context *ehc = &link->eh_context;
1640 u32 serror = ehc->i.serror;
1641 unsigned int err_mask = 0, action = 0;
1642 u32 hotplug_mask;
1644 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1645 err_mask |= AC_ERR_ATA_BUS;
1646 action |= ATA_EH_RESET;
1648 if (serror & SERR_PROTOCOL) {
1649 err_mask |= AC_ERR_HSM;
1650 action |= ATA_EH_RESET;
1652 if (serror & SERR_INTERNAL) {
1653 err_mask |= AC_ERR_SYSTEM;
1654 action |= ATA_EH_RESET;
1657 /* Determine whether a hotplug event has occurred. Both
1658 * SError.N/X are considered hotplug events for enabled or
1659 * host links. For disabled PMP links, only N bit is
1660 * considered as X bit is left at 1 for link plugging.
1662 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1663 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1664 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1665 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1666 else
1667 hotplug_mask = SERR_PHYRDY_CHG;
1669 if (serror & hotplug_mask)
1670 ata_ehi_hotplugged(&ehc->i);
1672 ehc->i.err_mask |= err_mask;
1673 ehc->i.action |= action;
1677 * ata_eh_analyze_ncq_error - analyze NCQ error
1678 * @link: ATA link to analyze NCQ error for
1680 * Read log page 10h, determine the offending qc and acquire
1681 * error status TF. For NCQ device errors, all LLDDs have to do
1682 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1683 * care of the rest.
1685 * LOCKING:
1686 * Kernel thread context (may sleep).
1688 void ata_eh_analyze_ncq_error(struct ata_link *link)
1690 struct ata_port *ap = link->ap;
1691 struct ata_eh_context *ehc = &link->eh_context;
1692 struct ata_device *dev = link->device;
1693 struct ata_queued_cmd *qc;
1694 struct ata_taskfile tf;
1695 int tag, rc;
1697 /* if frozen, we can't do much */
1698 if (ap->pflags & ATA_PFLAG_FROZEN)
1699 return;
1701 /* is it NCQ device error? */
1702 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1703 return;
1705 /* has LLDD analyzed already? */
1706 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1707 qc = __ata_qc_from_tag(ap, tag);
1709 if (!(qc->flags & ATA_QCFLAG_FAILED))
1710 continue;
1712 if (qc->err_mask)
1713 return;
1716 /* okay, this error is ours */
1717 memset(&tf, 0, sizeof(tf));
1718 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1719 if (rc) {
1720 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1721 rc);
1722 return;
1725 if (!(link->sactive & (1 << tag))) {
1726 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1727 tag);
1728 return;
1731 /* we've got the perpetrator, condemn it */
1732 qc = __ata_qc_from_tag(ap, tag);
1733 memcpy(&qc->result_tf, &tf, sizeof(tf));
1734 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1735 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1736 ehc->i.err_mask &= ~AC_ERR_DEV;
1740 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1741 * @qc: qc to analyze
1742 * @tf: Taskfile registers to analyze
1744 * Analyze taskfile of @qc and further determine cause of
1745 * failure. This function also requests ATAPI sense data if
1746 * available.
1748 * LOCKING:
1749 * Kernel thread context (may sleep).
1751 * RETURNS:
1752 * Determined recovery action
1754 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1755 const struct ata_taskfile *tf)
1757 unsigned int tmp, action = 0;
1758 u8 stat = tf->command, err = tf->feature;
1760 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1761 qc->err_mask |= AC_ERR_HSM;
1762 return ATA_EH_RESET;
1765 if (stat & (ATA_ERR | ATA_DF))
1766 qc->err_mask |= AC_ERR_DEV;
1767 else
1768 return 0;
1770 switch (qc->dev->class) {
1771 case ATA_DEV_ATA:
1772 if (err & ATA_ICRC)
1773 qc->err_mask |= AC_ERR_ATA_BUS;
1774 if (err & ATA_UNC)
1775 qc->err_mask |= AC_ERR_MEDIA;
1776 if (err & ATA_IDNF)
1777 qc->err_mask |= AC_ERR_INVALID;
1778 break;
1780 case ATA_DEV_ATAPI:
1781 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1782 tmp = atapi_eh_request_sense(qc->dev,
1783 qc->scsicmd->sense_buffer,
1784 qc->result_tf.feature >> 4);
1785 if (!tmp) {
1786 /* ATA_QCFLAG_SENSE_VALID is used to
1787 * tell atapi_qc_complete() that sense
1788 * data is already valid.
1790 * TODO: interpret sense data and set
1791 * appropriate err_mask.
1793 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1794 } else
1795 qc->err_mask |= tmp;
1799 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1800 action |= ATA_EH_RESET;
1802 return action;
1805 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1806 int *xfer_ok)
1808 int base = 0;
1810 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1811 *xfer_ok = 1;
1813 if (!*xfer_ok)
1814 base = ATA_ECAT_DUBIOUS_NONE;
1816 if (err_mask & AC_ERR_ATA_BUS)
1817 return base + ATA_ECAT_ATA_BUS;
1819 if (err_mask & AC_ERR_TIMEOUT)
1820 return base + ATA_ECAT_TOUT_HSM;
1822 if (eflags & ATA_EFLAG_IS_IO) {
1823 if (err_mask & AC_ERR_HSM)
1824 return base + ATA_ECAT_TOUT_HSM;
1825 if ((err_mask &
1826 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1827 return base + ATA_ECAT_UNK_DEV;
1830 return 0;
1833 struct speed_down_verdict_arg {
1834 u64 since;
1835 int xfer_ok;
1836 int nr_errors[ATA_ECAT_NR];
1839 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1841 struct speed_down_verdict_arg *arg = void_arg;
1842 int cat;
1844 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1845 return -1;
1847 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1848 &arg->xfer_ok);
1849 arg->nr_errors[cat]++;
1851 return 0;
1855 * ata_eh_speed_down_verdict - Determine speed down verdict
1856 * @dev: Device of interest
1858 * This function examines error ring of @dev and determines
1859 * whether NCQ needs to be turned off, transfer speed should be
1860 * stepped down, or falling back to PIO is necessary.
1862 * ECAT_ATA_BUS : ATA_BUS error for any command
1864 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1865 * IO commands
1867 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1869 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1870 * data transfer hasn't been verified.
1872 * Verdicts are
1874 * NCQ_OFF : Turn off NCQ.
1876 * SPEED_DOWN : Speed down transfer speed but don't fall back
1877 * to PIO.
1879 * FALLBACK_TO_PIO : Fall back to PIO.
1881 * Even if multiple verdicts are returned, only one action is
1882 * taken per error. An action triggered by non-DUBIOUS errors
1883 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1884 * This is to expedite speed down decisions right after device is
1885 * initially configured.
1887 * The followings are speed down rules. #1 and #2 deal with
1888 * DUBIOUS errors.
1890 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1891 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1893 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1894 * occurred during last 5 mins, NCQ_OFF.
1896 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1897 * occurred during last 5 mins, FALLBACK_TO_PIO
1899 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1900 * during last 10 mins, NCQ_OFF.
1902 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1903 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1905 * LOCKING:
1906 * Inherited from caller.
1908 * RETURNS:
1909 * OR of ATA_EH_SPDN_* flags.
1911 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1913 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1914 u64 j64 = get_jiffies_64();
1915 struct speed_down_verdict_arg arg;
1916 unsigned int verdict = 0;
1918 /* scan past 5 mins of error history */
1919 memset(&arg, 0, sizeof(arg));
1920 arg.since = j64 - min(j64, j5mins);
1921 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1923 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1924 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1925 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1926 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1928 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1929 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1930 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1932 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1933 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1934 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1935 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1937 /* scan past 10 mins of error history */
1938 memset(&arg, 0, sizeof(arg));
1939 arg.since = j64 - min(j64, j10mins);
1940 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1942 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1943 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1944 verdict |= ATA_EH_SPDN_NCQ_OFF;
1946 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1947 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1948 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1949 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1951 return verdict;
1955 * ata_eh_speed_down - record error and speed down if necessary
1956 * @dev: Failed device
1957 * @eflags: mask of ATA_EFLAG_* flags
1958 * @err_mask: err_mask of the error
1960 * Record error and examine error history to determine whether
1961 * adjusting transmission speed is necessary. It also sets
1962 * transmission limits appropriately if such adjustment is
1963 * necessary.
1965 * LOCKING:
1966 * Kernel thread context (may sleep).
1968 * RETURNS:
1969 * Determined recovery action.
1971 static unsigned int ata_eh_speed_down(struct ata_device *dev,
1972 unsigned int eflags, unsigned int err_mask)
1974 struct ata_link *link = ata_dev_phys_link(dev);
1975 int xfer_ok = 0;
1976 unsigned int verdict;
1977 unsigned int action = 0;
1979 /* don't bother if Cat-0 error */
1980 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1981 return 0;
1983 /* record error and determine whether speed down is necessary */
1984 ata_ering_record(&dev->ering, eflags, err_mask);
1985 verdict = ata_eh_speed_down_verdict(dev);
1987 /* turn off NCQ? */
1988 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1989 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1990 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1991 dev->flags |= ATA_DFLAG_NCQ_OFF;
1992 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1993 goto done;
1996 /* speed down? */
1997 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1998 /* speed down SATA link speed if possible */
1999 if (sata_down_spd_limit(link, 0) == 0) {
2000 action |= ATA_EH_RESET;
2001 goto done;
2004 /* lower transfer mode */
2005 if (dev->spdn_cnt < 2) {
2006 static const int dma_dnxfer_sel[] =
2007 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2008 static const int pio_dnxfer_sel[] =
2009 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2010 int sel;
2012 if (dev->xfer_shift != ATA_SHIFT_PIO)
2013 sel = dma_dnxfer_sel[dev->spdn_cnt];
2014 else
2015 sel = pio_dnxfer_sel[dev->spdn_cnt];
2017 dev->spdn_cnt++;
2019 if (ata_down_xfermask_limit(dev, sel) == 0) {
2020 action |= ATA_EH_RESET;
2021 goto done;
2026 /* Fall back to PIO? Slowing down to PIO is meaningless for
2027 * SATA ATA devices. Consider it only for PATA and SATAPI.
2029 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2030 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2031 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2032 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2033 dev->spdn_cnt = 0;
2034 action |= ATA_EH_RESET;
2035 goto done;
2039 return 0;
2040 done:
2041 /* device has been slowed down, blow error history */
2042 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2043 ata_ering_clear(&dev->ering);
2044 return action;
2048 * ata_eh_link_autopsy - analyze error and determine recovery action
2049 * @link: host link to perform autopsy on
2051 * Analyze why @link failed and determine which recovery actions
2052 * are needed. This function also sets more detailed AC_ERR_*
2053 * values and fills sense data for ATAPI CHECK SENSE.
2055 * LOCKING:
2056 * Kernel thread context (may sleep).
2058 static void ata_eh_link_autopsy(struct ata_link *link)
2060 struct ata_port *ap = link->ap;
2061 struct ata_eh_context *ehc = &link->eh_context;
2062 struct ata_device *dev;
2063 unsigned int all_err_mask = 0, eflags = 0;
2064 int tag;
2065 u32 serror;
2066 int rc;
2068 DPRINTK("ENTER\n");
2070 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2071 return;
2073 /* obtain and analyze SError */
2074 rc = sata_scr_read(link, SCR_ERROR, &serror);
2075 if (rc == 0) {
2076 ehc->i.serror |= serror;
2077 ata_eh_analyze_serror(link);
2078 } else if (rc != -EOPNOTSUPP) {
2079 /* SError read failed, force reset and probing */
2080 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2081 ehc->i.action |= ATA_EH_RESET;
2082 ehc->i.err_mask |= AC_ERR_OTHER;
2085 /* analyze NCQ failure */
2086 ata_eh_analyze_ncq_error(link);
2088 /* any real error trumps AC_ERR_OTHER */
2089 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2090 ehc->i.err_mask &= ~AC_ERR_OTHER;
2092 all_err_mask |= ehc->i.err_mask;
2094 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2095 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2097 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2098 ata_dev_phys_link(qc->dev) != link)
2099 continue;
2101 /* inherit upper level err_mask */
2102 qc->err_mask |= ehc->i.err_mask;
2104 /* analyze TF */
2105 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2107 /* DEV errors are probably spurious in case of ATA_BUS error */
2108 if (qc->err_mask & AC_ERR_ATA_BUS)
2109 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2110 AC_ERR_INVALID);
2112 /* any real error trumps unknown error */
2113 if (qc->err_mask & ~AC_ERR_OTHER)
2114 qc->err_mask &= ~AC_ERR_OTHER;
2116 /* SENSE_VALID trumps dev/unknown error and revalidation */
2117 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2118 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2120 /* determine whether the command is worth retrying */
2121 if (qc->flags & ATA_QCFLAG_IO ||
2122 (!(qc->err_mask & AC_ERR_INVALID) &&
2123 qc->err_mask != AC_ERR_DEV))
2124 qc->flags |= ATA_QCFLAG_RETRY;
2126 /* accumulate error info */
2127 ehc->i.dev = qc->dev;
2128 all_err_mask |= qc->err_mask;
2129 if (qc->flags & ATA_QCFLAG_IO)
2130 eflags |= ATA_EFLAG_IS_IO;
2133 /* enforce default EH actions */
2134 if (ap->pflags & ATA_PFLAG_FROZEN ||
2135 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2136 ehc->i.action |= ATA_EH_RESET;
2137 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2138 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2139 ehc->i.action |= ATA_EH_REVALIDATE;
2141 /* If we have offending qcs and the associated failed device,
2142 * perform per-dev EH action only on the offending device.
2144 if (ehc->i.dev) {
2145 ehc->i.dev_action[ehc->i.dev->devno] |=
2146 ehc->i.action & ATA_EH_PERDEV_MASK;
2147 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2150 /* propagate timeout to host link */
2151 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2152 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2154 /* record error and consider speeding down */
2155 dev = ehc->i.dev;
2156 if (!dev && ((ata_link_max_devices(link) == 1 &&
2157 ata_dev_enabled(link->device))))
2158 dev = link->device;
2160 if (dev) {
2161 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2162 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2163 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2166 DPRINTK("EXIT\n");
2170 * ata_eh_autopsy - analyze error and determine recovery action
2171 * @ap: host port to perform autopsy on
2173 * Analyze all links of @ap and determine why they failed and
2174 * which recovery actions are needed.
2176 * LOCKING:
2177 * Kernel thread context (may sleep).
2179 void ata_eh_autopsy(struct ata_port *ap)
2181 struct ata_link *link;
2183 ata_for_each_link(link, ap, EDGE)
2184 ata_eh_link_autopsy(link);
2186 /* Handle the frigging slave link. Autopsy is done similarly
2187 * but actions and flags are transferred over to the master
2188 * link and handled from there.
2190 if (ap->slave_link) {
2191 struct ata_eh_context *mehc = &ap->link.eh_context;
2192 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2194 /* transfer control flags from master to slave */
2195 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2197 /* perform autopsy on the slave link */
2198 ata_eh_link_autopsy(ap->slave_link);
2200 /* transfer actions from slave to master and clear slave */
2201 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2202 mehc->i.action |= sehc->i.action;
2203 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2204 mehc->i.flags |= sehc->i.flags;
2205 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2208 /* Autopsy of fanout ports can affect host link autopsy.
2209 * Perform host link autopsy last.
2211 if (sata_pmp_attached(ap))
2212 ata_eh_link_autopsy(&ap->link);
2216 * ata_get_cmd_descript - get description for ATA command
2217 * @command: ATA command code to get description for
2219 * Return a textual description of the given command, or NULL if the
2220 * command is not known.
2222 * LOCKING:
2223 * None
2225 const char *ata_get_cmd_descript(u8 command)
2227 #ifdef CONFIG_ATA_VERBOSE_ERROR
2228 static const struct
2230 u8 command;
2231 const char *text;
2232 } cmd_descr[] = {
2233 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2234 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2235 { ATA_CMD_STANDBY, "STANDBY" },
2236 { ATA_CMD_IDLE, "IDLE" },
2237 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2238 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2239 { ATA_CMD_NOP, "NOP" },
2240 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2241 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2242 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2243 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2244 { ATA_CMD_SERVICE, "SERVICE" },
2245 { ATA_CMD_READ, "READ DMA" },
2246 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2247 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2248 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2249 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2250 { ATA_CMD_WRITE, "WRITE DMA" },
2251 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2252 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2253 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2254 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2255 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2256 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2257 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2258 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2259 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2260 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2261 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2262 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2263 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2264 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2265 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2266 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2267 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2268 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2269 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2270 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2271 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2272 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2273 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2274 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2275 { ATA_CMD_SLEEP, "SLEEP" },
2276 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2277 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2278 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2279 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2280 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2281 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2282 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2283 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2284 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2285 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2286 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2287 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2288 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2289 { ATA_CMD_PMP_READ, "READ BUFFER" },
2290 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2291 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2292 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2293 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2294 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2295 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2296 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2297 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2298 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2299 { ATA_CMD_SMART, "SMART" },
2300 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2301 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2302 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2303 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2304 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2305 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2306 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2307 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2308 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2309 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2310 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2311 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2312 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2313 { ATA_CMD_RESTORE, "RECALIBRATE" },
2314 { 0, NULL } /* terminate list */
2317 unsigned int i;
2318 for (i = 0; cmd_descr[i].text; i++)
2319 if (cmd_descr[i].command == command)
2320 return cmd_descr[i].text;
2321 #endif
2323 return NULL;
2327 * ata_eh_link_report - report error handling to user
2328 * @link: ATA link EH is going on
2330 * Report EH to user.
2332 * LOCKING:
2333 * None.
2335 static void ata_eh_link_report(struct ata_link *link)
2337 struct ata_port *ap = link->ap;
2338 struct ata_eh_context *ehc = &link->eh_context;
2339 const char *frozen, *desc;
2340 char tries_buf[6];
2341 int tag, nr_failed = 0;
2343 if (ehc->i.flags & ATA_EHI_QUIET)
2344 return;
2346 desc = NULL;
2347 if (ehc->i.desc[0] != '\0')
2348 desc = ehc->i.desc;
2350 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2351 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2353 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2354 ata_dev_phys_link(qc->dev) != link ||
2355 ((qc->flags & ATA_QCFLAG_QUIET) &&
2356 qc->err_mask == AC_ERR_DEV))
2357 continue;
2358 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2359 continue;
2361 nr_failed++;
2364 if (!nr_failed && !ehc->i.err_mask)
2365 return;
2367 frozen = "";
2368 if (ap->pflags & ATA_PFLAG_FROZEN)
2369 frozen = " frozen";
2371 memset(tries_buf, 0, sizeof(tries_buf));
2372 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2373 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2374 ap->eh_tries);
2376 if (ehc->i.dev) {
2377 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2378 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2379 ehc->i.err_mask, link->sactive, ehc->i.serror,
2380 ehc->i.action, frozen, tries_buf);
2381 if (desc)
2382 ata_dev_err(ehc->i.dev, "%s\n", desc);
2383 } else {
2384 ata_link_err(link, "exception Emask 0x%x "
2385 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2386 ehc->i.err_mask, link->sactive, ehc->i.serror,
2387 ehc->i.action, frozen, tries_buf);
2388 if (desc)
2389 ata_link_err(link, "%s\n", desc);
2392 #ifdef CONFIG_ATA_VERBOSE_ERROR
2393 if (ehc->i.serror)
2394 ata_link_err(link,
2395 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2396 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2397 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2398 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2399 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2400 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2401 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2402 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2403 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2404 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2405 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2406 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2407 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2408 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2409 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2410 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2411 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2412 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2413 #endif
2415 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2416 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2417 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2418 const u8 *cdb = qc->cdb;
2419 char data_buf[20] = "";
2420 char cdb_buf[70] = "";
2422 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2423 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2424 continue;
2426 if (qc->dma_dir != DMA_NONE) {
2427 static const char *dma_str[] = {
2428 [DMA_BIDIRECTIONAL] = "bidi",
2429 [DMA_TO_DEVICE] = "out",
2430 [DMA_FROM_DEVICE] = "in",
2432 static const char *prot_str[] = {
2433 [ATA_PROT_PIO] = "pio",
2434 [ATA_PROT_DMA] = "dma",
2435 [ATA_PROT_NCQ] = "ncq",
2436 [ATAPI_PROT_PIO] = "pio",
2437 [ATAPI_PROT_DMA] = "dma",
2440 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2441 prot_str[qc->tf.protocol], qc->nbytes,
2442 dma_str[qc->dma_dir]);
2445 if (ata_is_atapi(qc->tf.protocol)) {
2446 if (qc->scsicmd)
2447 scsi_print_command(qc->scsicmd);
2448 else
2449 snprintf(cdb_buf, sizeof(cdb_buf),
2450 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2451 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2452 cdb[0], cdb[1], cdb[2], cdb[3],
2453 cdb[4], cdb[5], cdb[6], cdb[7],
2454 cdb[8], cdb[9], cdb[10], cdb[11],
2455 cdb[12], cdb[13], cdb[14], cdb[15]);
2456 } else {
2457 const char *descr = ata_get_cmd_descript(cmd->command);
2458 if (descr)
2459 ata_dev_err(qc->dev, "failed command: %s\n",
2460 descr);
2463 ata_dev_err(qc->dev,
2464 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2465 "tag %d%s\n %s"
2466 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2467 "Emask 0x%x (%s)%s\n",
2468 cmd->command, cmd->feature, cmd->nsect,
2469 cmd->lbal, cmd->lbam, cmd->lbah,
2470 cmd->hob_feature, cmd->hob_nsect,
2471 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2472 cmd->device, qc->tag, data_buf, cdb_buf,
2473 res->command, res->feature, res->nsect,
2474 res->lbal, res->lbam, res->lbah,
2475 res->hob_feature, res->hob_nsect,
2476 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2477 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2478 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2480 #ifdef CONFIG_ATA_VERBOSE_ERROR
2481 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2482 ATA_ERR)) {
2483 if (res->command & ATA_BUSY)
2484 ata_dev_err(qc->dev, "status: { Busy }\n");
2485 else
2486 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
2487 res->command & ATA_DRDY ? "DRDY " : "",
2488 res->command & ATA_DF ? "DF " : "",
2489 res->command & ATA_DRQ ? "DRQ " : "",
2490 res->command & ATA_ERR ? "ERR " : "");
2493 if (cmd->command != ATA_CMD_PACKET &&
2494 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2495 ATA_ABORTED)))
2496 ata_dev_err(qc->dev, "error: { %s%s%s%s}\n",
2497 res->feature & ATA_ICRC ? "ICRC " : "",
2498 res->feature & ATA_UNC ? "UNC " : "",
2499 res->feature & ATA_IDNF ? "IDNF " : "",
2500 res->feature & ATA_ABORTED ? "ABRT " : "");
2501 #endif
2506 * ata_eh_report - report error handling to user
2507 * @ap: ATA port to report EH about
2509 * Report EH to user.
2511 * LOCKING:
2512 * None.
2514 void ata_eh_report(struct ata_port *ap)
2516 struct ata_link *link;
2518 ata_for_each_link(link, ap, HOST_FIRST)
2519 ata_eh_link_report(link);
2522 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2523 unsigned int *classes, unsigned long deadline,
2524 bool clear_classes)
2526 struct ata_device *dev;
2528 if (clear_classes)
2529 ata_for_each_dev(dev, link, ALL)
2530 classes[dev->devno] = ATA_DEV_UNKNOWN;
2532 return reset(link, classes, deadline);
2535 static int ata_eh_followup_srst_needed(struct ata_link *link,
2536 int rc, const unsigned int *classes)
2538 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2539 return 0;
2540 if (rc == -EAGAIN)
2541 return 1;
2542 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2543 return 1;
2544 return 0;
2547 int ata_eh_reset(struct ata_link *link, int classify,
2548 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2549 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2551 struct ata_port *ap = link->ap;
2552 struct ata_link *slave = ap->slave_link;
2553 struct ata_eh_context *ehc = &link->eh_context;
2554 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2555 unsigned int *classes = ehc->classes;
2556 unsigned int lflags = link->flags;
2557 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2558 int max_tries = 0, try = 0;
2559 struct ata_link *failed_link;
2560 struct ata_device *dev;
2561 unsigned long deadline, now;
2562 ata_reset_fn_t reset;
2563 unsigned long flags;
2564 u32 sstatus;
2565 int nr_unknown, rc;
2568 * Prepare to reset
2570 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2571 max_tries++;
2572 if (link->flags & ATA_LFLAG_NO_HRST)
2573 hardreset = NULL;
2574 if (link->flags & ATA_LFLAG_NO_SRST)
2575 softreset = NULL;
2577 /* make sure each reset attempt is at least COOL_DOWN apart */
2578 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2579 now = jiffies;
2580 WARN_ON(time_after(ehc->last_reset, now));
2581 deadline = ata_deadline(ehc->last_reset,
2582 ATA_EH_RESET_COOL_DOWN);
2583 if (time_before(now, deadline))
2584 schedule_timeout_uninterruptible(deadline - now);
2587 spin_lock_irqsave(ap->lock, flags);
2588 ap->pflags |= ATA_PFLAG_RESETTING;
2589 spin_unlock_irqrestore(ap->lock, flags);
2591 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2593 ata_for_each_dev(dev, link, ALL) {
2594 /* If we issue an SRST then an ATA drive (not ATAPI)
2595 * may change configuration and be in PIO0 timing. If
2596 * we do a hard reset (or are coming from power on)
2597 * this is true for ATA or ATAPI. Until we've set a
2598 * suitable controller mode we should not touch the
2599 * bus as we may be talking too fast.
2601 dev->pio_mode = XFER_PIO_0;
2603 /* If the controller has a pio mode setup function
2604 * then use it to set the chipset to rights. Don't
2605 * touch the DMA setup as that will be dealt with when
2606 * configuring devices.
2608 if (ap->ops->set_piomode)
2609 ap->ops->set_piomode(ap, dev);
2612 /* prefer hardreset */
2613 reset = NULL;
2614 ehc->i.action &= ~ATA_EH_RESET;
2615 if (hardreset) {
2616 reset = hardreset;
2617 ehc->i.action |= ATA_EH_HARDRESET;
2618 } else if (softreset) {
2619 reset = softreset;
2620 ehc->i.action |= ATA_EH_SOFTRESET;
2623 if (prereset) {
2624 unsigned long deadline = ata_deadline(jiffies,
2625 ATA_EH_PRERESET_TIMEOUT);
2627 if (slave) {
2628 sehc->i.action &= ~ATA_EH_RESET;
2629 sehc->i.action |= ehc->i.action;
2632 rc = prereset(link, deadline);
2634 /* If present, do prereset on slave link too. Reset
2635 * is skipped iff both master and slave links report
2636 * -ENOENT or clear ATA_EH_RESET.
2638 if (slave && (rc == 0 || rc == -ENOENT)) {
2639 int tmp;
2641 tmp = prereset(slave, deadline);
2642 if (tmp != -ENOENT)
2643 rc = tmp;
2645 ehc->i.action |= sehc->i.action;
2648 if (rc) {
2649 if (rc == -ENOENT) {
2650 ata_link_dbg(link, "port disabled--ignoring\n");
2651 ehc->i.action &= ~ATA_EH_RESET;
2653 ata_for_each_dev(dev, link, ALL)
2654 classes[dev->devno] = ATA_DEV_NONE;
2656 rc = 0;
2657 } else
2658 ata_link_err(link,
2659 "prereset failed (errno=%d)\n",
2660 rc);
2661 goto out;
2664 /* prereset() might have cleared ATA_EH_RESET. If so,
2665 * bang classes, thaw and return.
2667 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2668 ata_for_each_dev(dev, link, ALL)
2669 classes[dev->devno] = ATA_DEV_NONE;
2670 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2671 ata_is_host_link(link))
2672 ata_eh_thaw_port(ap);
2673 rc = 0;
2674 goto out;
2678 retry:
2680 * Perform reset
2682 if (ata_is_host_link(link))
2683 ata_eh_freeze_port(ap);
2685 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2687 if (reset) {
2688 if (verbose)
2689 ata_link_info(link, "%s resetting link\n",
2690 reset == softreset ? "soft" : "hard");
2692 /* mark that this EH session started with reset */
2693 ehc->last_reset = jiffies;
2694 if (reset == hardreset)
2695 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2696 else
2697 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2699 rc = ata_do_reset(link, reset, classes, deadline, true);
2700 if (rc && rc != -EAGAIN) {
2701 failed_link = link;
2702 goto fail;
2705 /* hardreset slave link if existent */
2706 if (slave && reset == hardreset) {
2707 int tmp;
2709 if (verbose)
2710 ata_link_info(slave, "hard resetting link\n");
2712 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2713 tmp = ata_do_reset(slave, reset, classes, deadline,
2714 false);
2715 switch (tmp) {
2716 case -EAGAIN:
2717 rc = -EAGAIN;
2718 case 0:
2719 break;
2720 default:
2721 failed_link = slave;
2722 rc = tmp;
2723 goto fail;
2727 /* perform follow-up SRST if necessary */
2728 if (reset == hardreset &&
2729 ata_eh_followup_srst_needed(link, rc, classes)) {
2730 reset = softreset;
2732 if (!reset) {
2733 ata_link_err(link,
2734 "follow-up softreset required but no softreset available\n");
2735 failed_link = link;
2736 rc = -EINVAL;
2737 goto fail;
2740 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2741 rc = ata_do_reset(link, reset, classes, deadline, true);
2742 if (rc) {
2743 failed_link = link;
2744 goto fail;
2747 } else {
2748 if (verbose)
2749 ata_link_info(link,
2750 "no reset method available, skipping reset\n");
2751 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2752 lflags |= ATA_LFLAG_ASSUME_ATA;
2756 * Post-reset processing
2758 ata_for_each_dev(dev, link, ALL) {
2759 /* After the reset, the device state is PIO 0 and the
2760 * controller state is undefined. Reset also wakes up
2761 * drives from sleeping mode.
2763 dev->pio_mode = XFER_PIO_0;
2764 dev->flags &= ~ATA_DFLAG_SLEEPING;
2766 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2767 continue;
2769 /* apply class override */
2770 if (lflags & ATA_LFLAG_ASSUME_ATA)
2771 classes[dev->devno] = ATA_DEV_ATA;
2772 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2773 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2776 /* record current link speed */
2777 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2778 link->sata_spd = (sstatus >> 4) & 0xf;
2779 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2780 slave->sata_spd = (sstatus >> 4) & 0xf;
2782 /* thaw the port */
2783 if (ata_is_host_link(link))
2784 ata_eh_thaw_port(ap);
2786 /* postreset() should clear hardware SError. Although SError
2787 * is cleared during link resume, clearing SError here is
2788 * necessary as some PHYs raise hotplug events after SRST.
2789 * This introduces race condition where hotplug occurs between
2790 * reset and here. This race is mediated by cross checking
2791 * link onlineness and classification result later.
2793 if (postreset) {
2794 postreset(link, classes);
2795 if (slave)
2796 postreset(slave, classes);
2800 * Some controllers can't be frozen very well and may set spurious
2801 * error conditions during reset. Clear accumulated error
2802 * information and re-thaw the port if frozen. As reset is the
2803 * final recovery action and we cross check link onlineness against
2804 * device classification later, no hotplug event is lost by this.
2806 spin_lock_irqsave(link->ap->lock, flags);
2807 memset(&link->eh_info, 0, sizeof(link->eh_info));
2808 if (slave)
2809 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2810 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2811 spin_unlock_irqrestore(link->ap->lock, flags);
2813 if (ap->pflags & ATA_PFLAG_FROZEN)
2814 ata_eh_thaw_port(ap);
2817 * Make sure onlineness and classification result correspond.
2818 * Hotplug could have happened during reset and some
2819 * controllers fail to wait while a drive is spinning up after
2820 * being hotplugged causing misdetection. By cross checking
2821 * link on/offlineness and classification result, those
2822 * conditions can be reliably detected and retried.
2824 nr_unknown = 0;
2825 ata_for_each_dev(dev, link, ALL) {
2826 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2827 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2828 ata_dev_dbg(dev, "link online but device misclassified\n");
2829 classes[dev->devno] = ATA_DEV_NONE;
2830 nr_unknown++;
2832 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2833 if (ata_class_enabled(classes[dev->devno]))
2834 ata_dev_dbg(dev,
2835 "link offline, clearing class %d to NONE\n",
2836 classes[dev->devno]);
2837 classes[dev->devno] = ATA_DEV_NONE;
2838 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2839 ata_dev_dbg(dev,
2840 "link status unknown, clearing UNKNOWN to NONE\n");
2841 classes[dev->devno] = ATA_DEV_NONE;
2845 if (classify && nr_unknown) {
2846 if (try < max_tries) {
2847 ata_link_warn(link,
2848 "link online but %d devices misclassified, retrying\n",
2849 nr_unknown);
2850 failed_link = link;
2851 rc = -EAGAIN;
2852 goto fail;
2854 ata_link_warn(link,
2855 "link online but %d devices misclassified, "
2856 "device detection might fail\n", nr_unknown);
2859 /* reset successful, schedule revalidation */
2860 ata_eh_done(link, NULL, ATA_EH_RESET);
2861 if (slave)
2862 ata_eh_done(slave, NULL, ATA_EH_RESET);
2863 ehc->last_reset = jiffies; /* update to completion time */
2864 ehc->i.action |= ATA_EH_REVALIDATE;
2865 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2867 rc = 0;
2868 out:
2869 /* clear hotplug flag */
2870 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2871 if (slave)
2872 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2874 spin_lock_irqsave(ap->lock, flags);
2875 ap->pflags &= ~ATA_PFLAG_RESETTING;
2876 spin_unlock_irqrestore(ap->lock, flags);
2878 return rc;
2880 fail:
2881 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2882 if (!ata_is_host_link(link) &&
2883 sata_scr_read(link, SCR_STATUS, &sstatus))
2884 rc = -ERESTART;
2886 if (rc == -ERESTART || try >= max_tries) {
2888 * Thaw host port even if reset failed, so that the port
2889 * can be retried on the next phy event. This risks
2890 * repeated EH runs but seems to be a better tradeoff than
2891 * shutting down a port after a botched hotplug attempt.
2893 if (ata_is_host_link(link))
2894 ata_eh_thaw_port(ap);
2895 goto out;
2898 now = jiffies;
2899 if (time_before(now, deadline)) {
2900 unsigned long delta = deadline - now;
2902 ata_link_warn(failed_link,
2903 "reset failed (errno=%d), retrying in %u secs\n",
2904 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2906 ata_eh_release(ap);
2907 while (delta)
2908 delta = schedule_timeout_uninterruptible(delta);
2909 ata_eh_acquire(ap);
2912 if (try == max_tries - 1) {
2913 sata_down_spd_limit(link, 0);
2914 if (slave)
2915 sata_down_spd_limit(slave, 0);
2916 } else if (rc == -EPIPE)
2917 sata_down_spd_limit(failed_link, 0);
2919 if (hardreset)
2920 reset = hardreset;
2921 goto retry;
2924 static inline void ata_eh_pull_park_action(struct ata_port *ap)
2926 struct ata_link *link;
2927 struct ata_device *dev;
2928 unsigned long flags;
2931 * This function can be thought of as an extended version of
2932 * ata_eh_about_to_do() specially crafted to accommodate the
2933 * requirements of ATA_EH_PARK handling. Since the EH thread
2934 * does not leave the do {} while () loop in ata_eh_recover as
2935 * long as the timeout for a park request to *one* device on
2936 * the port has not expired, and since we still want to pick
2937 * up park requests to other devices on the same port or
2938 * timeout updates for the same device, we have to pull
2939 * ATA_EH_PARK actions from eh_info into eh_context.i
2940 * ourselves at the beginning of each pass over the loop.
2942 * Additionally, all write accesses to &ap->park_req_pending
2943 * through INIT_COMPLETION() (see below) or complete_all()
2944 * (see ata_scsi_park_store()) are protected by the host lock.
2945 * As a result we have that park_req_pending.done is zero on
2946 * exit from this function, i.e. when ATA_EH_PARK actions for
2947 * *all* devices on port ap have been pulled into the
2948 * respective eh_context structs. If, and only if,
2949 * park_req_pending.done is non-zero by the time we reach
2950 * wait_for_completion_timeout(), another ATA_EH_PARK action
2951 * has been scheduled for at least one of the devices on port
2952 * ap and we have to cycle over the do {} while () loop in
2953 * ata_eh_recover() again.
2956 spin_lock_irqsave(ap->lock, flags);
2957 INIT_COMPLETION(ap->park_req_pending);
2958 ata_for_each_link(link, ap, EDGE) {
2959 ata_for_each_dev(dev, link, ALL) {
2960 struct ata_eh_info *ehi = &link->eh_info;
2962 link->eh_context.i.dev_action[dev->devno] |=
2963 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2964 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2967 spin_unlock_irqrestore(ap->lock, flags);
2970 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2972 struct ata_eh_context *ehc = &dev->link->eh_context;
2973 struct ata_taskfile tf;
2974 unsigned int err_mask;
2976 ata_tf_init(dev, &tf);
2977 if (park) {
2978 ehc->unloaded_mask |= 1 << dev->devno;
2979 tf.command = ATA_CMD_IDLEIMMEDIATE;
2980 tf.feature = 0x44;
2981 tf.lbal = 0x4c;
2982 tf.lbam = 0x4e;
2983 tf.lbah = 0x55;
2984 } else {
2985 ehc->unloaded_mask &= ~(1 << dev->devno);
2986 tf.command = ATA_CMD_CHK_POWER;
2989 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2990 tf.protocol |= ATA_PROT_NODATA;
2991 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2992 if (park && (err_mask || tf.lbal != 0xc4)) {
2993 ata_dev_err(dev, "head unload failed!\n");
2994 ehc->unloaded_mask &= ~(1 << dev->devno);
2998 static int ata_eh_revalidate_and_attach(struct ata_link *link,
2999 struct ata_device **r_failed_dev)
3001 struct ata_port *ap = link->ap;
3002 struct ata_eh_context *ehc = &link->eh_context;
3003 struct ata_device *dev;
3004 unsigned int new_mask = 0;
3005 unsigned long flags;
3006 int rc = 0;
3008 DPRINTK("ENTER\n");
3010 /* For PATA drive side cable detection to work, IDENTIFY must
3011 * be done backwards such that PDIAG- is released by the slave
3012 * device before the master device is identified.
3014 ata_for_each_dev(dev, link, ALL_REVERSE) {
3015 unsigned int action = ata_eh_dev_action(dev);
3016 unsigned int readid_flags = 0;
3018 if (ehc->i.flags & ATA_EHI_DID_RESET)
3019 readid_flags |= ATA_READID_POSTRESET;
3021 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3022 WARN_ON(dev->class == ATA_DEV_PMP);
3024 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3025 rc = -EIO;
3026 goto err;
3029 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3030 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3031 readid_flags);
3032 if (rc)
3033 goto err;
3035 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3037 /* Configuration may have changed, reconfigure
3038 * transfer mode.
3040 ehc->i.flags |= ATA_EHI_SETMODE;
3042 /* schedule the scsi_rescan_device() here */
3043 schedule_work(&(ap->scsi_rescan_task));
3044 } else if (dev->class == ATA_DEV_UNKNOWN &&
3045 ehc->tries[dev->devno] &&
3046 ata_class_enabled(ehc->classes[dev->devno])) {
3047 /* Temporarily set dev->class, it will be
3048 * permanently set once all configurations are
3049 * complete. This is necessary because new
3050 * device configuration is done in two
3051 * separate loops.
3053 dev->class = ehc->classes[dev->devno];
3055 if (dev->class == ATA_DEV_PMP)
3056 rc = sata_pmp_attach(dev);
3057 else
3058 rc = ata_dev_read_id(dev, &dev->class,
3059 readid_flags, dev->id);
3061 /* read_id might have changed class, store and reset */
3062 ehc->classes[dev->devno] = dev->class;
3063 dev->class = ATA_DEV_UNKNOWN;
3065 switch (rc) {
3066 case 0:
3067 /* clear error info accumulated during probe */
3068 ata_ering_clear(&dev->ering);
3069 new_mask |= 1 << dev->devno;
3070 break;
3071 case -ENOENT:
3072 /* IDENTIFY was issued to non-existent
3073 * device. No need to reset. Just
3074 * thaw and ignore the device.
3076 ata_eh_thaw_port(ap);
3077 break;
3078 default:
3079 goto err;
3084 /* PDIAG- should have been released, ask cable type if post-reset */
3085 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3086 if (ap->ops->cable_detect)
3087 ap->cbl = ap->ops->cable_detect(ap);
3088 ata_force_cbl(ap);
3091 /* Configure new devices forward such that user doesn't see
3092 * device detection messages backwards.
3094 ata_for_each_dev(dev, link, ALL) {
3095 if (!(new_mask & (1 << dev->devno)))
3096 continue;
3098 dev->class = ehc->classes[dev->devno];
3100 if (dev->class == ATA_DEV_PMP)
3101 continue;
3103 ehc->i.flags |= ATA_EHI_PRINTINFO;
3104 rc = ata_dev_configure(dev);
3105 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3106 if (rc) {
3107 dev->class = ATA_DEV_UNKNOWN;
3108 goto err;
3111 spin_lock_irqsave(ap->lock, flags);
3112 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3113 spin_unlock_irqrestore(ap->lock, flags);
3115 /* new device discovered, configure xfermode */
3116 ehc->i.flags |= ATA_EHI_SETMODE;
3119 return 0;
3121 err:
3122 *r_failed_dev = dev;
3123 DPRINTK("EXIT rc=%d\n", rc);
3124 return rc;
3128 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3129 * @link: link on which timings will be programmed
3130 * @r_failed_dev: out parameter for failed device
3132 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3133 * ata_set_mode() fails, pointer to the failing device is
3134 * returned in @r_failed_dev.
3136 * LOCKING:
3137 * PCI/etc. bus probe sem.
3139 * RETURNS:
3140 * 0 on success, negative errno otherwise
3142 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3144 struct ata_port *ap = link->ap;
3145 struct ata_device *dev;
3146 int rc;
3148 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3149 ata_for_each_dev(dev, link, ENABLED) {
3150 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3151 struct ata_ering_entry *ent;
3153 ent = ata_ering_top(&dev->ering);
3154 if (ent)
3155 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3159 /* has private set_mode? */
3160 if (ap->ops->set_mode)
3161 rc = ap->ops->set_mode(link, r_failed_dev);
3162 else
3163 rc = ata_do_set_mode(link, r_failed_dev);
3165 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3166 ata_for_each_dev(dev, link, ENABLED) {
3167 struct ata_eh_context *ehc = &link->eh_context;
3168 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3169 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3171 if (dev->xfer_mode != saved_xfer_mode ||
3172 ata_ncq_enabled(dev) != saved_ncq)
3173 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3176 return rc;
3180 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3181 * @dev: ATAPI device to clear UA for
3183 * Resets and other operations can make an ATAPI device raise
3184 * UNIT ATTENTION which causes the next operation to fail. This
3185 * function clears UA.
3187 * LOCKING:
3188 * EH context (may sleep).
3190 * RETURNS:
3191 * 0 on success, -errno on failure.
3193 static int atapi_eh_clear_ua(struct ata_device *dev)
3195 int i;
3197 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3198 u8 *sense_buffer = dev->link->ap->sector_buf;
3199 u8 sense_key = 0;
3200 unsigned int err_mask;
3202 err_mask = atapi_eh_tur(dev, &sense_key);
3203 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3204 ata_dev_warn(dev,
3205 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3206 err_mask);
3207 return -EIO;
3210 if (!err_mask || sense_key != UNIT_ATTENTION)
3211 return 0;
3213 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3214 if (err_mask) {
3215 ata_dev_warn(dev, "failed to clear "
3216 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3217 return -EIO;
3221 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3222 ATA_EH_UA_TRIES);
3224 return 0;
3228 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3229 * @dev: ATA device which may need FLUSH retry
3231 * If @dev failed FLUSH, it needs to be reported upper layer
3232 * immediately as it means that @dev failed to remap and already
3233 * lost at least a sector and further FLUSH retrials won't make
3234 * any difference to the lost sector. However, if FLUSH failed
3235 * for other reasons, for example transmission error, FLUSH needs
3236 * to be retried.
3238 * This function determines whether FLUSH failure retry is
3239 * necessary and performs it if so.
3241 * RETURNS:
3242 * 0 if EH can continue, -errno if EH needs to be repeated.
3244 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3246 struct ata_link *link = dev->link;
3247 struct ata_port *ap = link->ap;
3248 struct ata_queued_cmd *qc;
3249 struct ata_taskfile tf;
3250 unsigned int err_mask;
3251 int rc = 0;
3253 /* did flush fail for this device? */
3254 if (!ata_tag_valid(link->active_tag))
3255 return 0;
3257 qc = __ata_qc_from_tag(ap, link->active_tag);
3258 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3259 qc->tf.command != ATA_CMD_FLUSH))
3260 return 0;
3262 /* if the device failed it, it should be reported to upper layers */
3263 if (qc->err_mask & AC_ERR_DEV)
3264 return 0;
3266 /* flush failed for some other reason, give it another shot */
3267 ata_tf_init(dev, &tf);
3269 tf.command = qc->tf.command;
3270 tf.flags |= ATA_TFLAG_DEVICE;
3271 tf.protocol = ATA_PROT_NODATA;
3273 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3274 tf.command, qc->err_mask);
3276 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3277 if (!err_mask) {
3279 * FLUSH is complete but there's no way to
3280 * successfully complete a failed command from EH.
3281 * Making sure retry is allowed at least once and
3282 * retrying it should do the trick - whatever was in
3283 * the cache is already on the platter and this won't
3284 * cause infinite loop.
3286 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3287 } else {
3288 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3289 err_mask);
3290 rc = -EIO;
3292 /* if device failed it, report it to upper layers */
3293 if (err_mask & AC_ERR_DEV) {
3294 qc->err_mask |= AC_ERR_DEV;
3295 qc->result_tf = tf;
3296 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3297 rc = 0;
3300 return rc;
3304 * ata_eh_set_lpm - configure SATA interface power management
3305 * @link: link to configure power management
3306 * @policy: the link power management policy
3307 * @r_failed_dev: out parameter for failed device
3309 * Enable SATA Interface power management. This will enable
3310 * Device Interface Power Management (DIPM) for min_power
3311 * policy, and then call driver specific callbacks for
3312 * enabling Host Initiated Power management.
3314 * LOCKING:
3315 * EH context.
3317 * RETURNS:
3318 * 0 on success, -errno on failure.
3320 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3321 struct ata_device **r_failed_dev)
3323 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3324 struct ata_eh_context *ehc = &link->eh_context;
3325 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3326 enum ata_lpm_policy old_policy = link->lpm_policy;
3327 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3328 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3329 unsigned int err_mask;
3330 int rc;
3332 /* if the link or host doesn't do LPM, noop */
3333 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3334 return 0;
3337 * DIPM is enabled only for MIN_POWER as some devices
3338 * misbehave when the host NACKs transition to SLUMBER. Order
3339 * device and link configurations such that the host always
3340 * allows DIPM requests.
3342 ata_for_each_dev(dev, link, ENABLED) {
3343 bool hipm = ata_id_has_hipm(dev->id);
3344 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3346 /* find the first enabled and LPM enabled devices */
3347 if (!link_dev)
3348 link_dev = dev;
3350 if (!lpm_dev && (hipm || dipm))
3351 lpm_dev = dev;
3353 hints &= ~ATA_LPM_EMPTY;
3354 if (!hipm)
3355 hints &= ~ATA_LPM_HIPM;
3357 /* disable DIPM before changing link config */
3358 if (policy != ATA_LPM_MIN_POWER && dipm) {
3359 err_mask = ata_dev_set_feature(dev,
3360 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3361 if (err_mask && err_mask != AC_ERR_DEV) {
3362 ata_dev_warn(dev,
3363 "failed to disable DIPM, Emask 0x%x\n",
3364 err_mask);
3365 rc = -EIO;
3366 goto fail;
3371 if (ap) {
3372 rc = ap->ops->set_lpm(link, policy, hints);
3373 if (!rc && ap->slave_link)
3374 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3375 } else
3376 rc = sata_pmp_set_lpm(link, policy, hints);
3379 * Attribute link config failure to the first (LPM) enabled
3380 * device on the link.
3382 if (rc) {
3383 if (rc == -EOPNOTSUPP) {
3384 link->flags |= ATA_LFLAG_NO_LPM;
3385 return 0;
3387 dev = lpm_dev ? lpm_dev : link_dev;
3388 goto fail;
3392 * Low level driver acked the transition. Issue DIPM command
3393 * with the new policy set.
3395 link->lpm_policy = policy;
3396 if (ap && ap->slave_link)
3397 ap->slave_link->lpm_policy = policy;
3399 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3400 ata_for_each_dev(dev, link, ENABLED) {
3401 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3402 ata_id_has_dipm(dev->id)) {
3403 err_mask = ata_dev_set_feature(dev,
3404 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3405 if (err_mask && err_mask != AC_ERR_DEV) {
3406 ata_dev_warn(dev,
3407 "failed to enable DIPM, Emask 0x%x\n",
3408 err_mask);
3409 rc = -EIO;
3410 goto fail;
3415 return 0;
3417 fail:
3418 /* restore the old policy */
3419 link->lpm_policy = old_policy;
3420 if (ap && ap->slave_link)
3421 ap->slave_link->lpm_policy = old_policy;
3423 /* if no device or only one more chance is left, disable LPM */
3424 if (!dev || ehc->tries[dev->devno] <= 2) {
3425 ata_link_warn(link, "disabling LPM on the link\n");
3426 link->flags |= ATA_LFLAG_NO_LPM;
3428 if (r_failed_dev)
3429 *r_failed_dev = dev;
3430 return rc;
3433 int ata_link_nr_enabled(struct ata_link *link)
3435 struct ata_device *dev;
3436 int cnt = 0;
3438 ata_for_each_dev(dev, link, ENABLED)
3439 cnt++;
3440 return cnt;
3443 static int ata_link_nr_vacant(struct ata_link *link)
3445 struct ata_device *dev;
3446 int cnt = 0;
3448 ata_for_each_dev(dev, link, ALL)
3449 if (dev->class == ATA_DEV_UNKNOWN)
3450 cnt++;
3451 return cnt;
3454 static int ata_eh_skip_recovery(struct ata_link *link)
3456 struct ata_port *ap = link->ap;
3457 struct ata_eh_context *ehc = &link->eh_context;
3458 struct ata_device *dev;
3460 /* skip disabled links */
3461 if (link->flags & ATA_LFLAG_DISABLED)
3462 return 1;
3464 /* skip if explicitly requested */
3465 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3466 return 1;
3468 /* thaw frozen port and recover failed devices */
3469 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3470 return 0;
3472 /* reset at least once if reset is requested */
3473 if ((ehc->i.action & ATA_EH_RESET) &&
3474 !(ehc->i.flags & ATA_EHI_DID_RESET))
3475 return 0;
3477 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3478 ata_for_each_dev(dev, link, ALL) {
3479 if (dev->class == ATA_DEV_UNKNOWN &&
3480 ehc->classes[dev->devno] != ATA_DEV_NONE)
3481 return 0;
3484 return 1;
3487 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3489 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3490 u64 now = get_jiffies_64();
3491 int *trials = void_arg;
3493 if (ent->timestamp < now - min(now, interval))
3494 return -1;
3496 (*trials)++;
3497 return 0;
3500 static int ata_eh_schedule_probe(struct ata_device *dev)
3502 struct ata_eh_context *ehc = &dev->link->eh_context;
3503 struct ata_link *link = ata_dev_phys_link(dev);
3504 int trials = 0;
3506 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3507 (ehc->did_probe_mask & (1 << dev->devno)))
3508 return 0;
3510 ata_eh_detach_dev(dev);
3511 ata_dev_init(dev);
3512 ehc->did_probe_mask |= (1 << dev->devno);
3513 ehc->i.action |= ATA_EH_RESET;
3514 ehc->saved_xfer_mode[dev->devno] = 0;
3515 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3517 /* the link maybe in a deep sleep, wake it up */
3518 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3519 if (ata_is_host_link(link))
3520 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3521 ATA_LPM_EMPTY);
3522 else
3523 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3524 ATA_LPM_EMPTY);
3527 /* Record and count probe trials on the ering. The specific
3528 * error mask used is irrelevant. Because a successful device
3529 * detection clears the ering, this count accumulates only if
3530 * there are consecutive failed probes.
3532 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3533 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3534 * forced to 1.5Gbps.
3536 * This is to work around cases where failed link speed
3537 * negotiation results in device misdetection leading to
3538 * infinite DEVXCHG or PHRDY CHG events.
3540 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3541 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3543 if (trials > ATA_EH_PROBE_TRIALS)
3544 sata_down_spd_limit(link, 1);
3546 return 1;
3549 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3551 struct ata_eh_context *ehc = &dev->link->eh_context;
3553 /* -EAGAIN from EH routine indicates retry without prejudice.
3554 * The requester is responsible for ensuring forward progress.
3556 if (err != -EAGAIN)
3557 ehc->tries[dev->devno]--;
3559 switch (err) {
3560 case -ENODEV:
3561 /* device missing or wrong IDENTIFY data, schedule probing */
3562 ehc->i.probe_mask |= (1 << dev->devno);
3563 case -EINVAL:
3564 /* give it just one more chance */
3565 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3566 case -EIO:
3567 if (ehc->tries[dev->devno] == 1) {
3568 /* This is the last chance, better to slow
3569 * down than lose it.
3571 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3572 if (dev->pio_mode > XFER_PIO_0)
3573 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3577 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3578 /* disable device if it has used up all its chances */
3579 ata_dev_disable(dev);
3581 /* detach if offline */
3582 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3583 ata_eh_detach_dev(dev);
3585 /* schedule probe if necessary */
3586 if (ata_eh_schedule_probe(dev)) {
3587 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3588 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3589 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3592 return 1;
3593 } else {
3594 ehc->i.action |= ATA_EH_RESET;
3595 return 0;
3600 * ata_eh_recover - recover host port after error
3601 * @ap: host port to recover
3602 * @prereset: prereset method (can be NULL)
3603 * @softreset: softreset method (can be NULL)
3604 * @hardreset: hardreset method (can be NULL)
3605 * @postreset: postreset method (can be NULL)
3606 * @r_failed_link: out parameter for failed link
3608 * This is the alpha and omega, eum and yang, heart and soul of
3609 * libata exception handling. On entry, actions required to
3610 * recover each link and hotplug requests are recorded in the
3611 * link's eh_context. This function executes all the operations
3612 * with appropriate retrials and fallbacks to resurrect failed
3613 * devices, detach goners and greet newcomers.
3615 * LOCKING:
3616 * Kernel thread context (may sleep).
3618 * RETURNS:
3619 * 0 on success, -errno on failure.
3621 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3622 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3623 ata_postreset_fn_t postreset,
3624 struct ata_link **r_failed_link)
3626 struct ata_link *link;
3627 struct ata_device *dev;
3628 int rc, nr_fails;
3629 unsigned long flags, deadline;
3631 DPRINTK("ENTER\n");
3633 /* prep for recovery */
3634 ata_for_each_link(link, ap, EDGE) {
3635 struct ata_eh_context *ehc = &link->eh_context;
3637 /* re-enable link? */
3638 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3639 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3640 spin_lock_irqsave(ap->lock, flags);
3641 link->flags &= ~ATA_LFLAG_DISABLED;
3642 spin_unlock_irqrestore(ap->lock, flags);
3643 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3646 ata_for_each_dev(dev, link, ALL) {
3647 if (link->flags & ATA_LFLAG_NO_RETRY)
3648 ehc->tries[dev->devno] = 1;
3649 else
3650 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3652 /* collect port action mask recorded in dev actions */
3653 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3654 ~ATA_EH_PERDEV_MASK;
3655 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3657 /* process hotplug request */
3658 if (dev->flags & ATA_DFLAG_DETACH)
3659 ata_eh_detach_dev(dev);
3661 /* schedule probe if necessary */
3662 if (!ata_dev_enabled(dev))
3663 ata_eh_schedule_probe(dev);
3667 retry:
3668 rc = 0;
3670 /* if UNLOADING, finish immediately */
3671 if (ap->pflags & ATA_PFLAG_UNLOADING)
3672 goto out;
3674 /* prep for EH */
3675 ata_for_each_link(link, ap, EDGE) {
3676 struct ata_eh_context *ehc = &link->eh_context;
3678 /* skip EH if possible. */
3679 if (ata_eh_skip_recovery(link))
3680 ehc->i.action = 0;
3682 ata_for_each_dev(dev, link, ALL)
3683 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3686 /* reset */
3687 ata_for_each_link(link, ap, EDGE) {
3688 struct ata_eh_context *ehc = &link->eh_context;
3690 if (!(ehc->i.action & ATA_EH_RESET))
3691 continue;
3693 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3694 prereset, softreset, hardreset, postreset);
3695 if (rc) {
3696 ata_link_err(link, "reset failed, giving up\n");
3697 goto out;
3701 do {
3702 unsigned long now;
3705 * clears ATA_EH_PARK in eh_info and resets
3706 * ap->park_req_pending
3708 ata_eh_pull_park_action(ap);
3710 deadline = jiffies;
3711 ata_for_each_link(link, ap, EDGE) {
3712 ata_for_each_dev(dev, link, ALL) {
3713 struct ata_eh_context *ehc = &link->eh_context;
3714 unsigned long tmp;
3716 if (dev->class != ATA_DEV_ATA)
3717 continue;
3718 if (!(ehc->i.dev_action[dev->devno] &
3719 ATA_EH_PARK))
3720 continue;
3721 tmp = dev->unpark_deadline;
3722 if (time_before(deadline, tmp))
3723 deadline = tmp;
3724 else if (time_before_eq(tmp, jiffies))
3725 continue;
3726 if (ehc->unloaded_mask & (1 << dev->devno))
3727 continue;
3729 ata_eh_park_issue_cmd(dev, 1);
3733 now = jiffies;
3734 if (time_before_eq(deadline, now))
3735 break;
3737 ata_eh_release(ap);
3738 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3739 deadline - now);
3740 ata_eh_acquire(ap);
3741 } while (deadline);
3742 ata_for_each_link(link, ap, EDGE) {
3743 ata_for_each_dev(dev, link, ALL) {
3744 if (!(link->eh_context.unloaded_mask &
3745 (1 << dev->devno)))
3746 continue;
3748 ata_eh_park_issue_cmd(dev, 0);
3749 ata_eh_done(link, dev, ATA_EH_PARK);
3753 /* the rest */
3754 nr_fails = 0;
3755 ata_for_each_link(link, ap, PMP_FIRST) {
3756 struct ata_eh_context *ehc = &link->eh_context;
3758 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3759 goto config_lpm;
3761 /* revalidate existing devices and attach new ones */
3762 rc = ata_eh_revalidate_and_attach(link, &dev);
3763 if (rc)
3764 goto rest_fail;
3766 /* if PMP got attached, return, pmp EH will take care of it */
3767 if (link->device->class == ATA_DEV_PMP) {
3768 ehc->i.action = 0;
3769 return 0;
3772 /* configure transfer mode if necessary */
3773 if (ehc->i.flags & ATA_EHI_SETMODE) {
3774 rc = ata_set_mode(link, &dev);
3775 if (rc)
3776 goto rest_fail;
3777 ehc->i.flags &= ~ATA_EHI_SETMODE;
3780 /* If reset has been issued, clear UA to avoid
3781 * disrupting the current users of the device.
3783 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3784 ata_for_each_dev(dev, link, ALL) {
3785 if (dev->class != ATA_DEV_ATAPI)
3786 continue;
3787 rc = atapi_eh_clear_ua(dev);
3788 if (rc)
3789 goto rest_fail;
3793 /* retry flush if necessary */
3794 ata_for_each_dev(dev, link, ALL) {
3795 if (dev->class != ATA_DEV_ATA)
3796 continue;
3797 rc = ata_eh_maybe_retry_flush(dev);
3798 if (rc)
3799 goto rest_fail;
3802 config_lpm:
3803 /* configure link power saving */
3804 if (link->lpm_policy != ap->target_lpm_policy) {
3805 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3806 if (rc)
3807 goto rest_fail;
3810 /* this link is okay now */
3811 ehc->i.flags = 0;
3812 continue;
3814 rest_fail:
3815 nr_fails++;
3816 if (dev)
3817 ata_eh_handle_dev_fail(dev, rc);
3819 if (ap->pflags & ATA_PFLAG_FROZEN) {
3820 /* PMP reset requires working host port.
3821 * Can't retry if it's frozen.
3823 if (sata_pmp_attached(ap))
3824 goto out;
3825 break;
3829 if (nr_fails)
3830 goto retry;
3832 out:
3833 if (rc && r_failed_link)
3834 *r_failed_link = link;
3836 DPRINTK("EXIT, rc=%d\n", rc);
3837 return rc;
3841 * ata_eh_finish - finish up EH
3842 * @ap: host port to finish EH for
3844 * Recovery is complete. Clean up EH states and retry or finish
3845 * failed qcs.
3847 * LOCKING:
3848 * None.
3850 void ata_eh_finish(struct ata_port *ap)
3852 int tag;
3854 /* retry or finish qcs */
3855 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3856 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3858 if (!(qc->flags & ATA_QCFLAG_FAILED))
3859 continue;
3861 if (qc->err_mask) {
3862 /* FIXME: Once EH migration is complete,
3863 * generate sense data in this function,
3864 * considering both err_mask and tf.
3866 if (qc->flags & ATA_QCFLAG_RETRY)
3867 ata_eh_qc_retry(qc);
3868 else
3869 ata_eh_qc_complete(qc);
3870 } else {
3871 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3872 ata_eh_qc_complete(qc);
3873 } else {
3874 /* feed zero TF to sense generation */
3875 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3876 ata_eh_qc_retry(qc);
3881 /* make sure nr_active_links is zero after EH */
3882 WARN_ON(ap->nr_active_links);
3883 ap->nr_active_links = 0;
3887 * ata_do_eh - do standard error handling
3888 * @ap: host port to handle error for
3890 * @prereset: prereset method (can be NULL)
3891 * @softreset: softreset method (can be NULL)
3892 * @hardreset: hardreset method (can be NULL)
3893 * @postreset: postreset method (can be NULL)
3895 * Perform standard error handling sequence.
3897 * LOCKING:
3898 * Kernel thread context (may sleep).
3900 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3901 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3902 ata_postreset_fn_t postreset)
3904 struct ata_device *dev;
3905 int rc;
3907 ata_eh_autopsy(ap);
3908 ata_eh_report(ap);
3910 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3911 NULL);
3912 if (rc) {
3913 ata_for_each_dev(dev, &ap->link, ALL)
3914 ata_dev_disable(dev);
3917 ata_eh_finish(ap);
3921 * ata_std_error_handler - standard error handler
3922 * @ap: host port to handle error for
3924 * Standard error handler
3926 * LOCKING:
3927 * Kernel thread context (may sleep).
3929 void ata_std_error_handler(struct ata_port *ap)
3931 struct ata_port_operations *ops = ap->ops;
3932 ata_reset_fn_t hardreset = ops->hardreset;
3934 /* ignore built-in hardreset if SCR access is not available */
3935 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3936 hardreset = NULL;
3938 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3941 #ifdef CONFIG_PM
3943 * ata_eh_handle_port_suspend - perform port suspend operation
3944 * @ap: port to suspend
3946 * Suspend @ap.
3948 * LOCKING:
3949 * Kernel thread context (may sleep).
3951 static void ata_eh_handle_port_suspend(struct ata_port *ap)
3953 unsigned long flags;
3954 int rc = 0;
3956 /* are we suspending? */
3957 spin_lock_irqsave(ap->lock, flags);
3958 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3959 ap->pm_mesg.event == PM_EVENT_ON) {
3960 spin_unlock_irqrestore(ap->lock, flags);
3961 return;
3963 spin_unlock_irqrestore(ap->lock, flags);
3965 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3967 /* tell ACPI we're suspending */
3968 rc = ata_acpi_on_suspend(ap);
3969 if (rc)
3970 goto out;
3972 /* suspend */
3973 ata_eh_freeze_port(ap);
3975 if (ap->ops->port_suspend)
3976 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3978 ata_acpi_set_state(ap, PMSG_SUSPEND);
3979 out:
3980 /* report result */
3981 spin_lock_irqsave(ap->lock, flags);
3983 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3984 if (rc == 0)
3985 ap->pflags |= ATA_PFLAG_SUSPENDED;
3986 else if (ap->pflags & ATA_PFLAG_FROZEN)
3987 ata_port_schedule_eh(ap);
3989 if (ap->pm_result) {
3990 *ap->pm_result = rc;
3991 ap->pm_result = NULL;
3994 spin_unlock_irqrestore(ap->lock, flags);
3996 return;
4000 * ata_eh_handle_port_resume - perform port resume operation
4001 * @ap: port to resume
4003 * Resume @ap.
4005 * LOCKING:
4006 * Kernel thread context (may sleep).
4008 static void ata_eh_handle_port_resume(struct ata_port *ap)
4010 struct ata_link *link;
4011 struct ata_device *dev;
4012 unsigned long flags;
4013 int rc = 0;
4015 /* are we resuming? */
4016 spin_lock_irqsave(ap->lock, flags);
4017 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4018 ap->pm_mesg.event != PM_EVENT_ON) {
4019 spin_unlock_irqrestore(ap->lock, flags);
4020 return;
4022 spin_unlock_irqrestore(ap->lock, flags);
4024 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4027 * Error timestamps are in jiffies which doesn't run while
4028 * suspended and PHY events during resume isn't too uncommon.
4029 * When the two are combined, it can lead to unnecessary speed
4030 * downs if the machine is suspended and resumed repeatedly.
4031 * Clear error history.
4033 ata_for_each_link(link, ap, HOST_FIRST)
4034 ata_for_each_dev(dev, link, ALL)
4035 ata_ering_clear(&dev->ering);
4037 ata_acpi_set_state(ap, PMSG_ON);
4039 if (ap->ops->port_resume)
4040 rc = ap->ops->port_resume(ap);
4042 /* tell ACPI that we're resuming */
4043 ata_acpi_on_resume(ap);
4045 /* report result */
4046 spin_lock_irqsave(ap->lock, flags);
4047 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4048 if (ap->pm_result) {
4049 *ap->pm_result = rc;
4050 ap->pm_result = NULL;
4052 spin_unlock_irqrestore(ap->lock, flags);
4054 #endif /* CONFIG_PM */