]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - drivers/ata/libata-eh.c
Merge tag 'powerpc-4.17-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[mirror_ubuntu-eoan-kernel.git] / drivers / ata / libata-eh.c
1 /*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
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.
15 *
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.
20 *
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.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/driver-api/libata.rst
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
35 #include <linux/kernel.h>
36 #include <linux/blkdev.h>
37 #include <linux/export.h>
38 #include <linux/pci.h>
39 #include <scsi/scsi.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi_eh.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_cmnd.h>
44 #include <scsi/scsi_dbg.h>
45 #include "../scsi/scsi_transport_api.h"
46
47 #include <linux/libata.h>
48
49 #include <trace/events/libata.h>
50 #include "libata.h"
51
52 enum {
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
58
59 /* error flags */
60 ATA_EFLAG_IS_IO = (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
62 ATA_EFLAG_OLD_ER = (1 << 31),
63
64 /* error categories */
65 ATA_ECAT_NONE = 0,
66 ATA_ECAT_ATA_BUS = 1,
67 ATA_ECAT_TOUT_HSM = 2,
68 ATA_ECAT_UNK_DEV = 3,
69 ATA_ECAT_DUBIOUS_NONE = 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
73 ATA_ECAT_NR = 8,
74
75 ATA_EH_CMD_DFL_TIMEOUT = 5000,
76
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN = 5000,
79
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
84 */
85 ATA_EH_PRERESET_TIMEOUT = 10000,
86 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87
88 ATA_EH_UA_TRIES = 5,
89
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS = 2,
93 };
94
95 /* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
101 */
102 static const unsigned long ata_eh_reset_timeouts[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX, /* > 1 min has elapsed, give up */
108 };
109
110 static const unsigned long ata_eh_identify_timeouts[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
114 ULONG_MAX,
115 };
116
117 static const unsigned long ata_eh_flush_timeouts[] = {
118 15000, /* be generous with flush */
119 15000, /* ditto */
120 30000, /* and even more generous */
121 ULONG_MAX,
122 };
123
124 static const unsigned long ata_eh_other_timeouts[] = {
125 5000, /* same rationale as identify timeout */
126 10000, /* ditto */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
128 ULONG_MAX,
129 };
130
131 struct ata_eh_cmd_timeout_ent {
132 const u8 *commands;
133 const unsigned long *timeouts;
134 };
135
136 /* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
139 *
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
143 *
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
147 */
148 #define CMDS(cmds...) (const u8 []){ cmds, 0 }
149 static const struct ata_eh_cmd_timeout_ent
150 ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
151 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
152 .timeouts = ata_eh_identify_timeouts, },
153 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_SET_FEATURES),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
160 .timeouts = ata_eh_other_timeouts, },
161 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
162 .timeouts = ata_eh_flush_timeouts },
163 };
164 #undef CMDS
165
166 static void __ata_port_freeze(struct ata_port *ap);
167 #ifdef CONFIG_PM
168 static void ata_eh_handle_port_suspend(struct ata_port *ap);
169 static void ata_eh_handle_port_resume(struct ata_port *ap);
170 #else /* CONFIG_PM */
171 static void ata_eh_handle_port_suspend(struct ata_port *ap)
172 { }
173
174 static void ata_eh_handle_port_resume(struct ata_port *ap)
175 { }
176 #endif /* CONFIG_PM */
177
178 static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
179 va_list args)
180 {
181 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
182 ATA_EH_DESC_LEN - ehi->desc_len,
183 fmt, args);
184 }
185
186 /**
187 * __ata_ehi_push_desc - push error description without adding separator
188 * @ehi: target EHI
189 * @fmt: printf format string
190 *
191 * Format string according to @fmt and append it to @ehi->desc.
192 *
193 * LOCKING:
194 * spin_lock_irqsave(host lock)
195 */
196 void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
197 {
198 va_list args;
199
200 va_start(args, fmt);
201 __ata_ehi_pushv_desc(ehi, fmt, args);
202 va_end(args);
203 }
204
205 /**
206 * ata_ehi_push_desc - push error description with separator
207 * @ehi: target EHI
208 * @fmt: printf format string
209 *
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
212 *
213 * LOCKING:
214 * spin_lock_irqsave(host lock)
215 */
216 void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
217 {
218 va_list args;
219
220 if (ehi->desc_len)
221 __ata_ehi_push_desc(ehi, ", ");
222
223 va_start(args, fmt);
224 __ata_ehi_pushv_desc(ehi, fmt, args);
225 va_end(args);
226 }
227
228 /**
229 * ata_ehi_clear_desc - clean error description
230 * @ehi: target EHI
231 *
232 * Clear @ehi->desc.
233 *
234 * LOCKING:
235 * spin_lock_irqsave(host lock)
236 */
237 void ata_ehi_clear_desc(struct ata_eh_info *ehi)
238 {
239 ehi->desc[0] = '\0';
240 ehi->desc_len = 0;
241 }
242
243 /**
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
247 *
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
252 *
253 * LOCKING:
254 * None.
255 */
256 void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
257 {
258 va_list args;
259
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
261
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
264
265 va_start(args, fmt);
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
267 va_end(args);
268 }
269
270 #ifdef CONFIG_PCI
271
272 /**
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
278 *
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
283 *
284 * LOCKING:
285 * None.
286 */
287 void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
288 const char *name)
289 {
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
291 char *type = "";
292 unsigned long long start, len;
293
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
295 type = "m";
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
297 type = "i";
298
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
301
302 if (offset < 0)
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
304 else
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
307 }
308
309 #endif /* CONFIG_PCI */
310
311 static int ata_lookup_timeout_table(u8 cmd)
312 {
313 int i;
314
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
316 const u8 *cur;
317
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
319 if (*cur == cmd)
320 return i;
321 }
322
323 return -1;
324 }
325
326 /**
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
330 *
331 * Determine timeout for internal command @cmd for @dev.
332 *
333 * LOCKING:
334 * EH context.
335 *
336 * RETURNS:
337 * Determined timeout.
338 */
339 unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
340 {
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
343 int idx;
344
345 if (ent < 0)
346 return ATA_EH_CMD_DFL_TIMEOUT;
347
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
350 }
351
352 /**
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
356 *
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
360 *
361 * LOCKING:
362 * EH context.
363 */
364 void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
365 {
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
368 int idx;
369
370 if (ent < 0)
371 return;
372
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
376 }
377
378 static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
380 {
381 struct ata_ering_entry *ent;
382
383 WARN_ON(!err_mask);
384
385 ering->cursor++;
386 ering->cursor %= ATA_ERING_SIZE;
387
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
392 }
393
394 static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
395 {
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
397
398 if (ent->err_mask)
399 return ent;
400 return NULL;
401 }
402
403 int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
405 void *arg)
406 {
407 int idx, rc = 0;
408 struct ata_ering_entry *ent;
409
410 idx = ering->cursor;
411 do {
412 ent = &ering->ring[idx];
413 if (!ent->err_mask)
414 break;
415 rc = map_fn(ent, arg);
416 if (rc)
417 break;
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
420
421 return rc;
422 }
423
424 static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
425 {
426 ent->eflags |= ATA_EFLAG_OLD_ER;
427 return 0;
428 }
429
430 static void ata_ering_clear(struct ata_ering *ering)
431 {
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
433 }
434
435 static unsigned int ata_eh_dev_action(struct ata_device *dev)
436 {
437 struct ata_eh_context *ehc = &dev->link->eh_context;
438
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
440 }
441
442 static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
444 {
445 struct ata_device *tdev;
446
447 if (!dev) {
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
451 } else {
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
454
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
461 }
462
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
465 }
466 }
467
468 /**
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
471 *
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
475 *
476 * LOCKING:
477 * EH context.
478 */
479 void ata_eh_acquire(struct ata_port *ap)
480 {
481 mutex_lock(&ap->host->eh_mutex);
482 WARN_ON_ONCE(ap->host->eh_owner);
483 ap->host->eh_owner = current;
484 }
485
486 /**
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
489 *
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
492 *
493 * LOCKING:
494 * EH context.
495 */
496 void ata_eh_release(struct ata_port *ap)
497 {
498 WARN_ON_ONCE(ap->host->eh_owner != current);
499 ap->host->eh_owner = NULL;
500 mutex_unlock(&ap->host->eh_mutex);
501 }
502
503 /**
504 * ata_scsi_timed_out - SCSI layer time out callback
505 * @cmd: timed out SCSI command
506 *
507 * Handles SCSI layer timeout. We race with normal completion of
508 * the qc for @cmd. If the qc is already gone, we lose and let
509 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
510 * timed out and EH should be invoked. Prevent ata_qc_complete()
511 * from finishing it by setting EH_SCHEDULED and return
512 * EH_NOT_HANDLED.
513 *
514 * TODO: kill this function once old EH is gone.
515 *
516 * LOCKING:
517 * Called from timer context
518 *
519 * RETURNS:
520 * EH_HANDLED or EH_NOT_HANDLED
521 */
522 enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
523 {
524 struct Scsi_Host *host = cmd->device->host;
525 struct ata_port *ap = ata_shost_to_port(host);
526 unsigned long flags;
527 struct ata_queued_cmd *qc;
528 enum blk_eh_timer_return ret;
529
530 DPRINTK("ENTER\n");
531
532 if (ap->ops->error_handler) {
533 ret = BLK_EH_NOT_HANDLED;
534 goto out;
535 }
536
537 ret = BLK_EH_HANDLED;
538 spin_lock_irqsave(ap->lock, flags);
539 qc = ata_qc_from_tag(ap, ap->link.active_tag);
540 if (qc) {
541 WARN_ON(qc->scsicmd != cmd);
542 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
543 qc->err_mask |= AC_ERR_TIMEOUT;
544 ret = BLK_EH_NOT_HANDLED;
545 }
546 spin_unlock_irqrestore(ap->lock, flags);
547
548 out:
549 DPRINTK("EXIT, ret=%d\n", ret);
550 return ret;
551 }
552 EXPORT_SYMBOL(ata_scsi_timed_out);
553
554 static void ata_eh_unload(struct ata_port *ap)
555 {
556 struct ata_link *link;
557 struct ata_device *dev;
558 unsigned long flags;
559
560 /* Restore SControl IPM and SPD for the next driver and
561 * disable attached devices.
562 */
563 ata_for_each_link(link, ap, PMP_FIRST) {
564 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
565 ata_for_each_dev(dev, link, ALL)
566 ata_dev_disable(dev);
567 }
568
569 /* freeze and set UNLOADED */
570 spin_lock_irqsave(ap->lock, flags);
571
572 ata_port_freeze(ap); /* won't be thawed */
573 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
574 ap->pflags |= ATA_PFLAG_UNLOADED;
575
576 spin_unlock_irqrestore(ap->lock, flags);
577 }
578
579 /**
580 * ata_scsi_error - SCSI layer error handler callback
581 * @host: SCSI host on which error occurred
582 *
583 * Handles SCSI-layer-thrown error events.
584 *
585 * LOCKING:
586 * Inherited from SCSI layer (none, can sleep)
587 *
588 * RETURNS:
589 * Zero.
590 */
591 void ata_scsi_error(struct Scsi_Host *host)
592 {
593 struct ata_port *ap = ata_shost_to_port(host);
594 unsigned long flags;
595 LIST_HEAD(eh_work_q);
596
597 DPRINTK("ENTER\n");
598
599 spin_lock_irqsave(host->host_lock, flags);
600 list_splice_init(&host->eh_cmd_q, &eh_work_q);
601 spin_unlock_irqrestore(host->host_lock, flags);
602
603 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
604
605 /* If we timed raced normal completion and there is nothing to
606 recover nr_timedout == 0 why exactly are we doing error recovery ? */
607 ata_scsi_port_error_handler(host, ap);
608
609 /* finish or retry handled scmd's and clean up */
610 WARN_ON(!list_empty(&eh_work_q));
611
612 DPRINTK("EXIT\n");
613 }
614
615 /**
616 * ata_scsi_cmd_error_handler - error callback for a list of commands
617 * @host: scsi host containing the port
618 * @ap: ATA port within the host
619 * @eh_work_q: list of commands to process
620 *
621 * process the given list of commands and return those finished to the
622 * ap->eh_done_q. This function is the first part of the libata error
623 * handler which processes a given list of failed commands.
624 */
625 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
626 struct list_head *eh_work_q)
627 {
628 int i;
629 unsigned long flags;
630
631 /* make sure sff pio task is not running */
632 ata_sff_flush_pio_task(ap);
633
634 /* synchronize with host lock and sort out timeouts */
635
636 /* For new EH, all qcs are finished in one of three ways -
637 * normal completion, error completion, and SCSI timeout.
638 * Both completions can race against SCSI timeout. When normal
639 * completion wins, the qc never reaches EH. When error
640 * completion wins, the qc has ATA_QCFLAG_FAILED set.
641 *
642 * When SCSI timeout wins, things are a bit more complex.
643 * Normal or error completion can occur after the timeout but
644 * before this point. In such cases, both types of
645 * completions are honored. A scmd is determined to have
646 * timed out iff its associated qc is active and not failed.
647 */
648 spin_lock_irqsave(ap->lock, flags);
649 if (ap->ops->error_handler) {
650 struct scsi_cmnd *scmd, *tmp;
651 int nr_timedout = 0;
652
653 /* This must occur under the ap->lock as we don't want
654 a polled recovery to race the real interrupt handler
655
656 The lost_interrupt handler checks for any completed but
657 non-notified command and completes much like an IRQ handler.
658
659 We then fall into the error recovery code which will treat
660 this as if normal completion won the race */
661
662 if (ap->ops->lost_interrupt)
663 ap->ops->lost_interrupt(ap);
664
665 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
666 struct ata_queued_cmd *qc;
667
668 for (i = 0; i < ATA_MAX_QUEUE; i++) {
669 qc = __ata_qc_from_tag(ap, i);
670 if (qc->flags & ATA_QCFLAG_ACTIVE &&
671 qc->scsicmd == scmd)
672 break;
673 }
674
675 if (i < ATA_MAX_QUEUE) {
676 /* the scmd has an associated qc */
677 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
678 /* which hasn't failed yet, timeout */
679 qc->err_mask |= AC_ERR_TIMEOUT;
680 qc->flags |= ATA_QCFLAG_FAILED;
681 nr_timedout++;
682 }
683 } else {
684 /* Normal completion occurred after
685 * SCSI timeout but before this point.
686 * Successfully complete it.
687 */
688 scmd->retries = scmd->allowed;
689 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
690 }
691 }
692
693 /* If we have timed out qcs. They belong to EH from
694 * this point but the state of the controller is
695 * unknown. Freeze the port to make sure the IRQ
696 * handler doesn't diddle with those qcs. This must
697 * be done atomically w.r.t. setting QCFLAG_FAILED.
698 */
699 if (nr_timedout)
700 __ata_port_freeze(ap);
701
702
703 /* initialize eh_tries */
704 ap->eh_tries = ATA_EH_MAX_TRIES;
705 }
706 spin_unlock_irqrestore(ap->lock, flags);
707
708 }
709 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
710
711 /**
712 * ata_scsi_port_error_handler - recover the port after the commands
713 * @host: SCSI host containing the port
714 * @ap: the ATA port
715 *
716 * Handle the recovery of the port @ap after all the commands
717 * have been recovered.
718 */
719 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
720 {
721 unsigned long flags;
722
723 /* invoke error handler */
724 if (ap->ops->error_handler) {
725 struct ata_link *link;
726
727 /* acquire EH ownership */
728 ata_eh_acquire(ap);
729 repeat:
730 /* kill fast drain timer */
731 del_timer_sync(&ap->fastdrain_timer);
732
733 /* process port resume request */
734 ata_eh_handle_port_resume(ap);
735
736 /* fetch & clear EH info */
737 spin_lock_irqsave(ap->lock, flags);
738
739 ata_for_each_link(link, ap, HOST_FIRST) {
740 struct ata_eh_context *ehc = &link->eh_context;
741 struct ata_device *dev;
742
743 memset(&link->eh_context, 0, sizeof(link->eh_context));
744 link->eh_context.i = link->eh_info;
745 memset(&link->eh_info, 0, sizeof(link->eh_info));
746
747 ata_for_each_dev(dev, link, ENABLED) {
748 int devno = dev->devno;
749
750 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
751 if (ata_ncq_enabled(dev))
752 ehc->saved_ncq_enabled |= 1 << devno;
753 }
754 }
755
756 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
757 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
758 ap->excl_link = NULL; /* don't maintain exclusion over EH */
759
760 spin_unlock_irqrestore(ap->lock, flags);
761
762 /* invoke EH, skip if unloading or suspended */
763 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
764 ap->ops->error_handler(ap);
765 else {
766 /* if unloading, commence suicide */
767 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
768 !(ap->pflags & ATA_PFLAG_UNLOADED))
769 ata_eh_unload(ap);
770 ata_eh_finish(ap);
771 }
772
773 /* process port suspend request */
774 ata_eh_handle_port_suspend(ap);
775
776 /* Exception might have happened after ->error_handler
777 * recovered the port but before this point. Repeat
778 * EH in such case.
779 */
780 spin_lock_irqsave(ap->lock, flags);
781
782 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
783 if (--ap->eh_tries) {
784 spin_unlock_irqrestore(ap->lock, flags);
785 goto repeat;
786 }
787 ata_port_err(ap,
788 "EH pending after %d tries, giving up\n",
789 ATA_EH_MAX_TRIES);
790 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
791 }
792
793 /* this run is complete, make sure EH info is clear */
794 ata_for_each_link(link, ap, HOST_FIRST)
795 memset(&link->eh_info, 0, sizeof(link->eh_info));
796
797 /* end eh (clear host_eh_scheduled) while holding
798 * ap->lock such that if exception occurs after this
799 * point but before EH completion, SCSI midlayer will
800 * re-initiate EH.
801 */
802 ap->ops->end_eh(ap);
803
804 spin_unlock_irqrestore(ap->lock, flags);
805 ata_eh_release(ap);
806 } else {
807 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
808 ap->ops->eng_timeout(ap);
809 }
810
811 scsi_eh_flush_done_q(&ap->eh_done_q);
812
813 /* clean up */
814 spin_lock_irqsave(ap->lock, flags);
815
816 if (ap->pflags & ATA_PFLAG_LOADING)
817 ap->pflags &= ~ATA_PFLAG_LOADING;
818 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
819 !(ap->flags & ATA_FLAG_SAS_HOST))
820 schedule_delayed_work(&ap->hotplug_task, 0);
821
822 if (ap->pflags & ATA_PFLAG_RECOVERED)
823 ata_port_info(ap, "EH complete\n");
824
825 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
826
827 /* tell wait_eh that we're done */
828 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
829 wake_up_all(&ap->eh_wait_q);
830
831 spin_unlock_irqrestore(ap->lock, flags);
832 }
833 EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
834
835 /**
836 * ata_port_wait_eh - Wait for the currently pending EH to complete
837 * @ap: Port to wait EH for
838 *
839 * Wait until the currently pending EH is complete.
840 *
841 * LOCKING:
842 * Kernel thread context (may sleep).
843 */
844 void ata_port_wait_eh(struct ata_port *ap)
845 {
846 unsigned long flags;
847 DEFINE_WAIT(wait);
848
849 retry:
850 spin_lock_irqsave(ap->lock, flags);
851
852 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
853 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
854 spin_unlock_irqrestore(ap->lock, flags);
855 schedule();
856 spin_lock_irqsave(ap->lock, flags);
857 }
858 finish_wait(&ap->eh_wait_q, &wait);
859
860 spin_unlock_irqrestore(ap->lock, flags);
861
862 /* make sure SCSI EH is complete */
863 if (scsi_host_in_recovery(ap->scsi_host)) {
864 ata_msleep(ap, 10);
865 goto retry;
866 }
867 }
868 EXPORT_SYMBOL_GPL(ata_port_wait_eh);
869
870 static int ata_eh_nr_in_flight(struct ata_port *ap)
871 {
872 unsigned int tag;
873 int nr = 0;
874
875 /* count only non-internal commands */
876 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
877 if (ata_qc_from_tag(ap, tag))
878 nr++;
879
880 return nr;
881 }
882
883 void ata_eh_fastdrain_timerfn(struct timer_list *t)
884 {
885 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
886 unsigned long flags;
887 int cnt;
888
889 spin_lock_irqsave(ap->lock, flags);
890
891 cnt = ata_eh_nr_in_flight(ap);
892
893 /* are we done? */
894 if (!cnt)
895 goto out_unlock;
896
897 if (cnt == ap->fastdrain_cnt) {
898 unsigned int tag;
899
900 /* No progress during the last interval, tag all
901 * in-flight qcs as timed out and freeze the port.
902 */
903 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
904 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
905 if (qc)
906 qc->err_mask |= AC_ERR_TIMEOUT;
907 }
908
909 ata_port_freeze(ap);
910 } else {
911 /* some qcs have finished, give it another chance */
912 ap->fastdrain_cnt = cnt;
913 ap->fastdrain_timer.expires =
914 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
915 add_timer(&ap->fastdrain_timer);
916 }
917
918 out_unlock:
919 spin_unlock_irqrestore(ap->lock, flags);
920 }
921
922 /**
923 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
924 * @ap: target ATA port
925 * @fastdrain: activate fast drain
926 *
927 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
928 * is non-zero and EH wasn't pending before. Fast drain ensures
929 * that EH kicks in in timely manner.
930 *
931 * LOCKING:
932 * spin_lock_irqsave(host lock)
933 */
934 static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
935 {
936 int cnt;
937
938 /* already scheduled? */
939 if (ap->pflags & ATA_PFLAG_EH_PENDING)
940 return;
941
942 ap->pflags |= ATA_PFLAG_EH_PENDING;
943
944 if (!fastdrain)
945 return;
946
947 /* do we have in-flight qcs? */
948 cnt = ata_eh_nr_in_flight(ap);
949 if (!cnt)
950 return;
951
952 /* activate fast drain */
953 ap->fastdrain_cnt = cnt;
954 ap->fastdrain_timer.expires =
955 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
956 add_timer(&ap->fastdrain_timer);
957 }
958
959 /**
960 * ata_qc_schedule_eh - schedule qc for error handling
961 * @qc: command to schedule error handling for
962 *
963 * Schedule error handling for @qc. EH will kick in as soon as
964 * other commands are drained.
965 *
966 * LOCKING:
967 * spin_lock_irqsave(host lock)
968 */
969 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
970 {
971 struct ata_port *ap = qc->ap;
972 struct request_queue *q = qc->scsicmd->device->request_queue;
973 unsigned long flags;
974
975 WARN_ON(!ap->ops->error_handler);
976
977 qc->flags |= ATA_QCFLAG_FAILED;
978 ata_eh_set_pending(ap, 1);
979
980 /* The following will fail if timeout has already expired.
981 * ata_scsi_error() takes care of such scmds on EH entry.
982 * Note that ATA_QCFLAG_FAILED is unconditionally set after
983 * this function completes.
984 */
985 spin_lock_irqsave(q->queue_lock, flags);
986 blk_abort_request(qc->scsicmd->request);
987 spin_unlock_irqrestore(q->queue_lock, flags);
988 }
989
990 /**
991 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
992 * @ap: ATA port to schedule EH for
993 *
994 * LOCKING: inherited from ata_port_schedule_eh
995 * spin_lock_irqsave(host lock)
996 */
997 void ata_std_sched_eh(struct ata_port *ap)
998 {
999 WARN_ON(!ap->ops->error_handler);
1000
1001 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1002 return;
1003
1004 ata_eh_set_pending(ap, 1);
1005 scsi_schedule_eh(ap->scsi_host);
1006
1007 DPRINTK("port EH scheduled\n");
1008 }
1009 EXPORT_SYMBOL_GPL(ata_std_sched_eh);
1010
1011 /**
1012 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
1013 * @ap: ATA port to end EH for
1014 *
1015 * In the libata object model there is a 1:1 mapping of ata_port to
1016 * shost, so host fields can be directly manipulated under ap->lock, in
1017 * the libsas case we need to hold a lock at the ha->level to coordinate
1018 * these events.
1019 *
1020 * LOCKING:
1021 * spin_lock_irqsave(host lock)
1022 */
1023 void ata_std_end_eh(struct ata_port *ap)
1024 {
1025 struct Scsi_Host *host = ap->scsi_host;
1026
1027 host->host_eh_scheduled = 0;
1028 }
1029 EXPORT_SYMBOL(ata_std_end_eh);
1030
1031
1032 /**
1033 * ata_port_schedule_eh - schedule error handling without a qc
1034 * @ap: ATA port to schedule EH for
1035 *
1036 * Schedule error handling for @ap. EH will kick in as soon as
1037 * all commands are drained.
1038 *
1039 * LOCKING:
1040 * spin_lock_irqsave(host lock)
1041 */
1042 void ata_port_schedule_eh(struct ata_port *ap)
1043 {
1044 /* see: ata_std_sched_eh, unless you know better */
1045 ap->ops->sched_eh(ap);
1046 }
1047
1048 static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1049 {
1050 int tag, nr_aborted = 0;
1051
1052 WARN_ON(!ap->ops->error_handler);
1053
1054 /* we're gonna abort all commands, no need for fast drain */
1055 ata_eh_set_pending(ap, 0);
1056
1057 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1058 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1059
1060 if (qc && (!link || qc->dev->link == link)) {
1061 qc->flags |= ATA_QCFLAG_FAILED;
1062 ata_qc_complete(qc);
1063 nr_aborted++;
1064 }
1065 }
1066
1067 if (!nr_aborted)
1068 ata_port_schedule_eh(ap);
1069
1070 return nr_aborted;
1071 }
1072
1073 /**
1074 * ata_link_abort - abort all qc's on the link
1075 * @link: ATA link to abort qc's for
1076 *
1077 * Abort all active qc's active on @link and schedule EH.
1078 *
1079 * LOCKING:
1080 * spin_lock_irqsave(host lock)
1081 *
1082 * RETURNS:
1083 * Number of aborted qc's.
1084 */
1085 int ata_link_abort(struct ata_link *link)
1086 {
1087 return ata_do_link_abort(link->ap, link);
1088 }
1089
1090 /**
1091 * ata_port_abort - abort all qc's on the port
1092 * @ap: ATA port to abort qc's for
1093 *
1094 * Abort all active qc's of @ap and schedule EH.
1095 *
1096 * LOCKING:
1097 * spin_lock_irqsave(host_set lock)
1098 *
1099 * RETURNS:
1100 * Number of aborted qc's.
1101 */
1102 int ata_port_abort(struct ata_port *ap)
1103 {
1104 return ata_do_link_abort(ap, NULL);
1105 }
1106
1107 /**
1108 * __ata_port_freeze - freeze port
1109 * @ap: ATA port to freeze
1110 *
1111 * This function is called when HSM violation or some other
1112 * condition disrupts normal operation of the port. Frozen port
1113 * is not allowed to perform any operation until the port is
1114 * thawed, which usually follows a successful reset.
1115 *
1116 * ap->ops->freeze() callback can be used for freezing the port
1117 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1118 * port cannot be frozen hardware-wise, the interrupt handler
1119 * must ack and clear interrupts unconditionally while the port
1120 * is frozen.
1121 *
1122 * LOCKING:
1123 * spin_lock_irqsave(host lock)
1124 */
1125 static void __ata_port_freeze(struct ata_port *ap)
1126 {
1127 WARN_ON(!ap->ops->error_handler);
1128
1129 if (ap->ops->freeze)
1130 ap->ops->freeze(ap);
1131
1132 ap->pflags |= ATA_PFLAG_FROZEN;
1133
1134 DPRINTK("ata%u port frozen\n", ap->print_id);
1135 }
1136
1137 /**
1138 * ata_port_freeze - abort & freeze port
1139 * @ap: ATA port to freeze
1140 *
1141 * Abort and freeze @ap. The freeze operation must be called
1142 * first, because some hardware requires special operations
1143 * before the taskfile registers are accessible.
1144 *
1145 * LOCKING:
1146 * spin_lock_irqsave(host lock)
1147 *
1148 * RETURNS:
1149 * Number of aborted commands.
1150 */
1151 int ata_port_freeze(struct ata_port *ap)
1152 {
1153 int nr_aborted;
1154
1155 WARN_ON(!ap->ops->error_handler);
1156
1157 __ata_port_freeze(ap);
1158 nr_aborted = ata_port_abort(ap);
1159
1160 return nr_aborted;
1161 }
1162
1163 /**
1164 * sata_async_notification - SATA async notification handler
1165 * @ap: ATA port where async notification is received
1166 *
1167 * Handler to be called when async notification via SDB FIS is
1168 * received. This function schedules EH if necessary.
1169 *
1170 * LOCKING:
1171 * spin_lock_irqsave(host lock)
1172 *
1173 * RETURNS:
1174 * 1 if EH is scheduled, 0 otherwise.
1175 */
1176 int sata_async_notification(struct ata_port *ap)
1177 {
1178 u32 sntf;
1179 int rc;
1180
1181 if (!(ap->flags & ATA_FLAG_AN))
1182 return 0;
1183
1184 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1185 if (rc == 0)
1186 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1187
1188 if (!sata_pmp_attached(ap) || rc) {
1189 /* PMP is not attached or SNTF is not available */
1190 if (!sata_pmp_attached(ap)) {
1191 /* PMP is not attached. Check whether ATAPI
1192 * AN is configured. If so, notify media
1193 * change.
1194 */
1195 struct ata_device *dev = ap->link.device;
1196
1197 if ((dev->class == ATA_DEV_ATAPI) &&
1198 (dev->flags & ATA_DFLAG_AN))
1199 ata_scsi_media_change_notify(dev);
1200 return 0;
1201 } else {
1202 /* PMP is attached but SNTF is not available.
1203 * ATAPI async media change notification is
1204 * not used. The PMP must be reporting PHY
1205 * status change, schedule EH.
1206 */
1207 ata_port_schedule_eh(ap);
1208 return 1;
1209 }
1210 } else {
1211 /* PMP is attached and SNTF is available */
1212 struct ata_link *link;
1213
1214 /* check and notify ATAPI AN */
1215 ata_for_each_link(link, ap, EDGE) {
1216 if (!(sntf & (1 << link->pmp)))
1217 continue;
1218
1219 if ((link->device->class == ATA_DEV_ATAPI) &&
1220 (link->device->flags & ATA_DFLAG_AN))
1221 ata_scsi_media_change_notify(link->device);
1222 }
1223
1224 /* If PMP is reporting that PHY status of some
1225 * downstream ports has changed, schedule EH.
1226 */
1227 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1228 ata_port_schedule_eh(ap);
1229 return 1;
1230 }
1231
1232 return 0;
1233 }
1234 }
1235
1236 /**
1237 * ata_eh_freeze_port - EH helper to freeze port
1238 * @ap: ATA port to freeze
1239 *
1240 * Freeze @ap.
1241 *
1242 * LOCKING:
1243 * None.
1244 */
1245 void ata_eh_freeze_port(struct ata_port *ap)
1246 {
1247 unsigned long flags;
1248
1249 if (!ap->ops->error_handler)
1250 return;
1251
1252 spin_lock_irqsave(ap->lock, flags);
1253 __ata_port_freeze(ap);
1254 spin_unlock_irqrestore(ap->lock, flags);
1255 }
1256
1257 /**
1258 * ata_port_thaw_port - EH helper to thaw port
1259 * @ap: ATA port to thaw
1260 *
1261 * Thaw frozen port @ap.
1262 *
1263 * LOCKING:
1264 * None.
1265 */
1266 void ata_eh_thaw_port(struct ata_port *ap)
1267 {
1268 unsigned long flags;
1269
1270 if (!ap->ops->error_handler)
1271 return;
1272
1273 spin_lock_irqsave(ap->lock, flags);
1274
1275 ap->pflags &= ~ATA_PFLAG_FROZEN;
1276
1277 if (ap->ops->thaw)
1278 ap->ops->thaw(ap);
1279
1280 spin_unlock_irqrestore(ap->lock, flags);
1281
1282 DPRINTK("ata%u port thawed\n", ap->print_id);
1283 }
1284
1285 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1286 {
1287 /* nada */
1288 }
1289
1290 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1291 {
1292 struct ata_port *ap = qc->ap;
1293 struct scsi_cmnd *scmd = qc->scsicmd;
1294 unsigned long flags;
1295
1296 spin_lock_irqsave(ap->lock, flags);
1297 qc->scsidone = ata_eh_scsidone;
1298 __ata_qc_complete(qc);
1299 WARN_ON(ata_tag_valid(qc->tag));
1300 spin_unlock_irqrestore(ap->lock, flags);
1301
1302 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1303 }
1304
1305 /**
1306 * ata_eh_qc_complete - Complete an active ATA command from EH
1307 * @qc: Command to complete
1308 *
1309 * Indicate to the mid and upper layers that an ATA command has
1310 * completed. To be used from EH.
1311 */
1312 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1313 {
1314 struct scsi_cmnd *scmd = qc->scsicmd;
1315 scmd->retries = scmd->allowed;
1316 __ata_eh_qc_complete(qc);
1317 }
1318
1319 /**
1320 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1321 * @qc: Command to retry
1322 *
1323 * Indicate to the mid and upper layers that an ATA command
1324 * should be retried. To be used from EH.
1325 *
1326 * SCSI midlayer limits the number of retries to scmd->allowed.
1327 * scmd->allowed is incremented for commands which get retried
1328 * due to unrelated failures (qc->err_mask is zero).
1329 */
1330 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1331 {
1332 struct scsi_cmnd *scmd = qc->scsicmd;
1333 if (!qc->err_mask)
1334 scmd->allowed++;
1335 __ata_eh_qc_complete(qc);
1336 }
1337
1338 /**
1339 * ata_dev_disable - disable ATA device
1340 * @dev: ATA device to disable
1341 *
1342 * Disable @dev.
1343 *
1344 * Locking:
1345 * EH context.
1346 */
1347 void ata_dev_disable(struct ata_device *dev)
1348 {
1349 if (!ata_dev_enabled(dev))
1350 return;
1351
1352 if (ata_msg_drv(dev->link->ap))
1353 ata_dev_warn(dev, "disabled\n");
1354 ata_acpi_on_disable(dev);
1355 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1356 dev->class++;
1357
1358 /* From now till the next successful probe, ering is used to
1359 * track probe failures. Clear accumulated device error info.
1360 */
1361 ata_ering_clear(&dev->ering);
1362 }
1363
1364 /**
1365 * ata_eh_detach_dev - detach ATA device
1366 * @dev: ATA device to detach
1367 *
1368 * Detach @dev.
1369 *
1370 * LOCKING:
1371 * None.
1372 */
1373 void ata_eh_detach_dev(struct ata_device *dev)
1374 {
1375 struct ata_link *link = dev->link;
1376 struct ata_port *ap = link->ap;
1377 struct ata_eh_context *ehc = &link->eh_context;
1378 unsigned long flags;
1379
1380 ata_dev_disable(dev);
1381
1382 spin_lock_irqsave(ap->lock, flags);
1383
1384 dev->flags &= ~ATA_DFLAG_DETACH;
1385
1386 if (ata_scsi_offline_dev(dev)) {
1387 dev->flags |= ATA_DFLAG_DETACHED;
1388 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1389 }
1390
1391 /* clear per-dev EH info */
1392 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1393 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1394 ehc->saved_xfer_mode[dev->devno] = 0;
1395 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1396
1397 spin_unlock_irqrestore(ap->lock, flags);
1398 }
1399
1400 /**
1401 * ata_eh_about_to_do - about to perform eh_action
1402 * @link: target ATA link
1403 * @dev: target ATA dev for per-dev action (can be NULL)
1404 * @action: action about to be performed
1405 *
1406 * Called just before performing EH actions to clear related bits
1407 * in @link->eh_info such that eh actions are not unnecessarily
1408 * repeated.
1409 *
1410 * LOCKING:
1411 * None.
1412 */
1413 void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1414 unsigned int action)
1415 {
1416 struct ata_port *ap = link->ap;
1417 struct ata_eh_info *ehi = &link->eh_info;
1418 struct ata_eh_context *ehc = &link->eh_context;
1419 unsigned long flags;
1420
1421 spin_lock_irqsave(ap->lock, flags);
1422
1423 ata_eh_clear_action(link, dev, ehi, action);
1424
1425 /* About to take EH action, set RECOVERED. Ignore actions on
1426 * slave links as master will do them again.
1427 */
1428 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1429 ap->pflags |= ATA_PFLAG_RECOVERED;
1430
1431 spin_unlock_irqrestore(ap->lock, flags);
1432 }
1433
1434 /**
1435 * ata_eh_done - EH action complete
1436 * @link: ATA link for which EH actions are complete
1437 * @dev: target ATA dev for per-dev action (can be NULL)
1438 * @action: action just completed
1439 *
1440 * Called right after performing EH actions to clear related bits
1441 * in @link->eh_context.
1442 *
1443 * LOCKING:
1444 * None.
1445 */
1446 void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1447 unsigned int action)
1448 {
1449 struct ata_eh_context *ehc = &link->eh_context;
1450
1451 ata_eh_clear_action(link, dev, &ehc->i, action);
1452 }
1453
1454 /**
1455 * ata_err_string - convert err_mask to descriptive string
1456 * @err_mask: error mask to convert to string
1457 *
1458 * Convert @err_mask to descriptive string. Errors are
1459 * prioritized according to severity and only the most severe
1460 * error is reported.
1461 *
1462 * LOCKING:
1463 * None.
1464 *
1465 * RETURNS:
1466 * Descriptive string for @err_mask
1467 */
1468 static const char *ata_err_string(unsigned int err_mask)
1469 {
1470 if (err_mask & AC_ERR_HOST_BUS)
1471 return "host bus error";
1472 if (err_mask & AC_ERR_ATA_BUS)
1473 return "ATA bus error";
1474 if (err_mask & AC_ERR_TIMEOUT)
1475 return "timeout";
1476 if (err_mask & AC_ERR_HSM)
1477 return "HSM violation";
1478 if (err_mask & AC_ERR_SYSTEM)
1479 return "internal error";
1480 if (err_mask & AC_ERR_MEDIA)
1481 return "media error";
1482 if (err_mask & AC_ERR_INVALID)
1483 return "invalid argument";
1484 if (err_mask & AC_ERR_DEV)
1485 return "device error";
1486 return "unknown error";
1487 }
1488
1489 /**
1490 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1491 * @dev: Device to read log page 10h from
1492 * @tag: Resulting tag of the failed command
1493 * @tf: Resulting taskfile registers of the failed command
1494 *
1495 * Read log page 10h to obtain NCQ error details and clear error
1496 * condition.
1497 *
1498 * LOCKING:
1499 * Kernel thread context (may sleep).
1500 *
1501 * RETURNS:
1502 * 0 on success, -errno otherwise.
1503 */
1504 static int ata_eh_read_log_10h(struct ata_device *dev,
1505 int *tag, struct ata_taskfile *tf)
1506 {
1507 u8 *buf = dev->link->ap->sector_buf;
1508 unsigned int err_mask;
1509 u8 csum;
1510 int i;
1511
1512 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1513 if (err_mask)
1514 return -EIO;
1515
1516 csum = 0;
1517 for (i = 0; i < ATA_SECT_SIZE; i++)
1518 csum += buf[i];
1519 if (csum)
1520 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1521 csum);
1522
1523 if (buf[0] & 0x80)
1524 return -ENOENT;
1525
1526 *tag = buf[0] & 0x1f;
1527
1528 tf->command = buf[2];
1529 tf->feature = buf[3];
1530 tf->lbal = buf[4];
1531 tf->lbam = buf[5];
1532 tf->lbah = buf[6];
1533 tf->device = buf[7];
1534 tf->hob_lbal = buf[8];
1535 tf->hob_lbam = buf[9];
1536 tf->hob_lbah = buf[10];
1537 tf->nsect = buf[12];
1538 tf->hob_nsect = buf[13];
1539 if (ata_id_has_ncq_autosense(dev->id))
1540 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1541
1542 return 0;
1543 }
1544
1545 /**
1546 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1547 * @dev: target ATAPI device
1548 * @r_sense_key: out parameter for sense_key
1549 *
1550 * Perform ATAPI TEST_UNIT_READY.
1551 *
1552 * LOCKING:
1553 * EH context (may sleep).
1554 *
1555 * RETURNS:
1556 * 0 on success, AC_ERR_* mask on failure.
1557 */
1558 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1559 {
1560 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1561 struct ata_taskfile tf;
1562 unsigned int err_mask;
1563
1564 ata_tf_init(dev, &tf);
1565
1566 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1567 tf.command = ATA_CMD_PACKET;
1568 tf.protocol = ATAPI_PROT_NODATA;
1569
1570 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1571 if (err_mask == AC_ERR_DEV)
1572 *r_sense_key = tf.feature >> 4;
1573 return err_mask;
1574 }
1575
1576 /**
1577 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1578 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1579 * @cmd: scsi command for which the sense code should be set
1580 *
1581 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1582 * SENSE. This function is an EH helper.
1583 *
1584 * LOCKING:
1585 * Kernel thread context (may sleep).
1586 */
1587 static void ata_eh_request_sense(struct ata_queued_cmd *qc,
1588 struct scsi_cmnd *cmd)
1589 {
1590 struct ata_device *dev = qc->dev;
1591 struct ata_taskfile tf;
1592 unsigned int err_mask;
1593
1594 if (qc->ap->pflags & ATA_PFLAG_FROZEN) {
1595 ata_dev_warn(dev, "sense data available but port frozen\n");
1596 return;
1597 }
1598
1599 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1600 return;
1601
1602 if (!ata_id_sense_reporting_enabled(dev->id)) {
1603 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1604 return;
1605 }
1606
1607 DPRINTK("ATA request sense\n");
1608
1609 ata_tf_init(dev, &tf);
1610 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1611 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1612 tf.command = ATA_CMD_REQ_SENSE_DATA;
1613 tf.protocol = ATA_PROT_NODATA;
1614
1615 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1616 /* Ignore err_mask; ATA_ERR might be set */
1617 if (tf.command & ATA_SENSE) {
1618 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1619 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1620 } else {
1621 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1622 tf.command, err_mask);
1623 }
1624 }
1625
1626 /**
1627 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1628 * @dev: device to perform REQUEST_SENSE to
1629 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1630 * @dfl_sense_key: default sense key to use
1631 *
1632 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1633 * SENSE. This function is EH helper.
1634 *
1635 * LOCKING:
1636 * Kernel thread context (may sleep).
1637 *
1638 * RETURNS:
1639 * 0 on success, AC_ERR_* mask on failure
1640 */
1641 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1642 u8 *sense_buf, u8 dfl_sense_key)
1643 {
1644 u8 cdb[ATAPI_CDB_LEN] =
1645 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1646 struct ata_port *ap = dev->link->ap;
1647 struct ata_taskfile tf;
1648
1649 DPRINTK("ATAPI request sense\n");
1650
1651 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1652
1653 /* initialize sense_buf with the error register,
1654 * for the case where they are -not- overwritten
1655 */
1656 sense_buf[0] = 0x70;
1657 sense_buf[2] = dfl_sense_key;
1658
1659 /* some devices time out if garbage left in tf */
1660 ata_tf_init(dev, &tf);
1661
1662 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1663 tf.command = ATA_CMD_PACKET;
1664
1665 /* is it pointless to prefer PIO for "safety reasons"? */
1666 if (ap->flags & ATA_FLAG_PIO_DMA) {
1667 tf.protocol = ATAPI_PROT_DMA;
1668 tf.feature |= ATAPI_PKT_DMA;
1669 } else {
1670 tf.protocol = ATAPI_PROT_PIO;
1671 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1672 tf.lbah = 0;
1673 }
1674
1675 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1676 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1677 }
1678
1679 /**
1680 * ata_eh_analyze_serror - analyze SError for a failed port
1681 * @link: ATA link to analyze SError for
1682 *
1683 * Analyze SError if available and further determine cause of
1684 * failure.
1685 *
1686 * LOCKING:
1687 * None.
1688 */
1689 static void ata_eh_analyze_serror(struct ata_link *link)
1690 {
1691 struct ata_eh_context *ehc = &link->eh_context;
1692 u32 serror = ehc->i.serror;
1693 unsigned int err_mask = 0, action = 0;
1694 u32 hotplug_mask;
1695
1696 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1697 err_mask |= AC_ERR_ATA_BUS;
1698 action |= ATA_EH_RESET;
1699 }
1700 if (serror & SERR_PROTOCOL) {
1701 err_mask |= AC_ERR_HSM;
1702 action |= ATA_EH_RESET;
1703 }
1704 if (serror & SERR_INTERNAL) {
1705 err_mask |= AC_ERR_SYSTEM;
1706 action |= ATA_EH_RESET;
1707 }
1708
1709 /* Determine whether a hotplug event has occurred. Both
1710 * SError.N/X are considered hotplug events for enabled or
1711 * host links. For disabled PMP links, only N bit is
1712 * considered as X bit is left at 1 for link plugging.
1713 */
1714 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1715 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1716 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1717 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1718 else
1719 hotplug_mask = SERR_PHYRDY_CHG;
1720
1721 if (serror & hotplug_mask)
1722 ata_ehi_hotplugged(&ehc->i);
1723
1724 ehc->i.err_mask |= err_mask;
1725 ehc->i.action |= action;
1726 }
1727
1728 /**
1729 * ata_eh_analyze_ncq_error - analyze NCQ error
1730 * @link: ATA link to analyze NCQ error for
1731 *
1732 * Read log page 10h, determine the offending qc and acquire
1733 * error status TF. For NCQ device errors, all LLDDs have to do
1734 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1735 * care of the rest.
1736 *
1737 * LOCKING:
1738 * Kernel thread context (may sleep).
1739 */
1740 void ata_eh_analyze_ncq_error(struct ata_link *link)
1741 {
1742 struct ata_port *ap = link->ap;
1743 struct ata_eh_context *ehc = &link->eh_context;
1744 struct ata_device *dev = link->device;
1745 struct ata_queued_cmd *qc;
1746 struct ata_taskfile tf;
1747 int tag, rc;
1748
1749 /* if frozen, we can't do much */
1750 if (ap->pflags & ATA_PFLAG_FROZEN)
1751 return;
1752
1753 /* is it NCQ device error? */
1754 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1755 return;
1756
1757 /* has LLDD analyzed already? */
1758 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1759 qc = __ata_qc_from_tag(ap, tag);
1760
1761 if (!(qc->flags & ATA_QCFLAG_FAILED))
1762 continue;
1763
1764 if (qc->err_mask)
1765 return;
1766 }
1767
1768 /* okay, this error is ours */
1769 memset(&tf, 0, sizeof(tf));
1770 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1771 if (rc) {
1772 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1773 rc);
1774 return;
1775 }
1776
1777 if (!(link->sactive & (1 << tag))) {
1778 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1779 tag);
1780 return;
1781 }
1782
1783 /* we've got the perpetrator, condemn it */
1784 qc = __ata_qc_from_tag(ap, tag);
1785 memcpy(&qc->result_tf, &tf, sizeof(tf));
1786 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1787 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1788 if ((qc->result_tf.command & ATA_SENSE) || qc->result_tf.auxiliary) {
1789 char sense_key, asc, ascq;
1790
1791 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1792 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1793 ascq = qc->result_tf.auxiliary & 0xff;
1794 ata_scsi_set_sense(dev, qc->scsicmd, sense_key, asc, ascq);
1795 ata_scsi_set_sense_information(dev, qc->scsicmd,
1796 &qc->result_tf);
1797 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1798 }
1799
1800 ehc->i.err_mask &= ~AC_ERR_DEV;
1801 }
1802
1803 /**
1804 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1805 * @qc: qc to analyze
1806 * @tf: Taskfile registers to analyze
1807 *
1808 * Analyze taskfile of @qc and further determine cause of
1809 * failure. This function also requests ATAPI sense data if
1810 * available.
1811 *
1812 * LOCKING:
1813 * Kernel thread context (may sleep).
1814 *
1815 * RETURNS:
1816 * Determined recovery action
1817 */
1818 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1819 const struct ata_taskfile *tf)
1820 {
1821 unsigned int tmp, action = 0;
1822 u8 stat = tf->command, err = tf->feature;
1823
1824 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1825 qc->err_mask |= AC_ERR_HSM;
1826 return ATA_EH_RESET;
1827 }
1828
1829 if (stat & (ATA_ERR | ATA_DF)) {
1830 qc->err_mask |= AC_ERR_DEV;
1831 /*
1832 * Sense data reporting does not work if the
1833 * device fault bit is set.
1834 */
1835 if (stat & ATA_DF)
1836 stat &= ~ATA_SENSE;
1837 } else {
1838 return 0;
1839 }
1840
1841 switch (qc->dev->class) {
1842 case ATA_DEV_ATA:
1843 case ATA_DEV_ZAC:
1844 if (stat & ATA_SENSE)
1845 ata_eh_request_sense(qc, qc->scsicmd);
1846 if (err & ATA_ICRC)
1847 qc->err_mask |= AC_ERR_ATA_BUS;
1848 if (err & (ATA_UNC | ATA_AMNF))
1849 qc->err_mask |= AC_ERR_MEDIA;
1850 if (err & ATA_IDNF)
1851 qc->err_mask |= AC_ERR_INVALID;
1852 break;
1853
1854 case ATA_DEV_ATAPI:
1855 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1856 tmp = atapi_eh_request_sense(qc->dev,
1857 qc->scsicmd->sense_buffer,
1858 qc->result_tf.feature >> 4);
1859 if (!tmp)
1860 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1861 else
1862 qc->err_mask |= tmp;
1863 }
1864 }
1865
1866 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1867 int ret = scsi_check_sense(qc->scsicmd);
1868 /*
1869 * SUCCESS here means that the sense code could
1870 * evaluated and should be passed to the upper layers
1871 * for correct evaluation.
1872 * FAILED means the sense code could not interpreted
1873 * and the device would need to be reset.
1874 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1875 * command would need to be retried.
1876 */
1877 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1878 qc->flags |= ATA_QCFLAG_RETRY;
1879 qc->err_mask |= AC_ERR_OTHER;
1880 } else if (ret != SUCCESS) {
1881 qc->err_mask |= AC_ERR_HSM;
1882 }
1883 }
1884 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1885 action |= ATA_EH_RESET;
1886
1887 return action;
1888 }
1889
1890 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1891 int *xfer_ok)
1892 {
1893 int base = 0;
1894
1895 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1896 *xfer_ok = 1;
1897
1898 if (!*xfer_ok)
1899 base = ATA_ECAT_DUBIOUS_NONE;
1900
1901 if (err_mask & AC_ERR_ATA_BUS)
1902 return base + ATA_ECAT_ATA_BUS;
1903
1904 if (err_mask & AC_ERR_TIMEOUT)
1905 return base + ATA_ECAT_TOUT_HSM;
1906
1907 if (eflags & ATA_EFLAG_IS_IO) {
1908 if (err_mask & AC_ERR_HSM)
1909 return base + ATA_ECAT_TOUT_HSM;
1910 if ((err_mask &
1911 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1912 return base + ATA_ECAT_UNK_DEV;
1913 }
1914
1915 return 0;
1916 }
1917
1918 struct speed_down_verdict_arg {
1919 u64 since;
1920 int xfer_ok;
1921 int nr_errors[ATA_ECAT_NR];
1922 };
1923
1924 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1925 {
1926 struct speed_down_verdict_arg *arg = void_arg;
1927 int cat;
1928
1929 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1930 return -1;
1931
1932 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1933 &arg->xfer_ok);
1934 arg->nr_errors[cat]++;
1935
1936 return 0;
1937 }
1938
1939 /**
1940 * ata_eh_speed_down_verdict - Determine speed down verdict
1941 * @dev: Device of interest
1942 *
1943 * This function examines error ring of @dev and determines
1944 * whether NCQ needs to be turned off, transfer speed should be
1945 * stepped down, or falling back to PIO is necessary.
1946 *
1947 * ECAT_ATA_BUS : ATA_BUS error for any command
1948 *
1949 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1950 * IO commands
1951 *
1952 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1953 *
1954 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1955 * data transfer hasn't been verified.
1956 *
1957 * Verdicts are
1958 *
1959 * NCQ_OFF : Turn off NCQ.
1960 *
1961 * SPEED_DOWN : Speed down transfer speed but don't fall back
1962 * to PIO.
1963 *
1964 * FALLBACK_TO_PIO : Fall back to PIO.
1965 *
1966 * Even if multiple verdicts are returned, only one action is
1967 * taken per error. An action triggered by non-DUBIOUS errors
1968 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1969 * This is to expedite speed down decisions right after device is
1970 * initially configured.
1971 *
1972 * The following are speed down rules. #1 and #2 deal with
1973 * DUBIOUS errors.
1974 *
1975 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1976 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1977 *
1978 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1979 * occurred during last 5 mins, NCQ_OFF.
1980 *
1981 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1982 * occurred during last 5 mins, FALLBACK_TO_PIO
1983 *
1984 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1985 * during last 10 mins, NCQ_OFF.
1986 *
1987 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1988 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1989 *
1990 * LOCKING:
1991 * Inherited from caller.
1992 *
1993 * RETURNS:
1994 * OR of ATA_EH_SPDN_* flags.
1995 */
1996 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1997 {
1998 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1999 u64 j64 = get_jiffies_64();
2000 struct speed_down_verdict_arg arg;
2001 unsigned int verdict = 0;
2002
2003 /* scan past 5 mins of error history */
2004 memset(&arg, 0, sizeof(arg));
2005 arg.since = j64 - min(j64, j5mins);
2006 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2007
2008 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
2009 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
2010 verdict |= ATA_EH_SPDN_SPEED_DOWN |
2011 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
2012
2013 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
2014 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
2015 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
2016
2017 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2018 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2019 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2020 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
2021
2022 /* scan past 10 mins of error history */
2023 memset(&arg, 0, sizeof(arg));
2024 arg.since = j64 - min(j64, j10mins);
2025 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2026
2027 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2028 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
2029 verdict |= ATA_EH_SPDN_NCQ_OFF;
2030
2031 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2032 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
2033 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2034 verdict |= ATA_EH_SPDN_SPEED_DOWN;
2035
2036 return verdict;
2037 }
2038
2039 /**
2040 * ata_eh_speed_down - record error and speed down if necessary
2041 * @dev: Failed device
2042 * @eflags: mask of ATA_EFLAG_* flags
2043 * @err_mask: err_mask of the error
2044 *
2045 * Record error and examine error history to determine whether
2046 * adjusting transmission speed is necessary. It also sets
2047 * transmission limits appropriately if such adjustment is
2048 * necessary.
2049 *
2050 * LOCKING:
2051 * Kernel thread context (may sleep).
2052 *
2053 * RETURNS:
2054 * Determined recovery action.
2055 */
2056 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2057 unsigned int eflags, unsigned int err_mask)
2058 {
2059 struct ata_link *link = ata_dev_phys_link(dev);
2060 int xfer_ok = 0;
2061 unsigned int verdict;
2062 unsigned int action = 0;
2063
2064 /* don't bother if Cat-0 error */
2065 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2066 return 0;
2067
2068 /* record error and determine whether speed down is necessary */
2069 ata_ering_record(&dev->ering, eflags, err_mask);
2070 verdict = ata_eh_speed_down_verdict(dev);
2071
2072 /* turn off NCQ? */
2073 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2074 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2075 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2076 dev->flags |= ATA_DFLAG_NCQ_OFF;
2077 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2078 goto done;
2079 }
2080
2081 /* speed down? */
2082 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2083 /* speed down SATA link speed if possible */
2084 if (sata_down_spd_limit(link, 0) == 0) {
2085 action |= ATA_EH_RESET;
2086 goto done;
2087 }
2088
2089 /* lower transfer mode */
2090 if (dev->spdn_cnt < 2) {
2091 static const int dma_dnxfer_sel[] =
2092 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2093 static const int pio_dnxfer_sel[] =
2094 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2095 int sel;
2096
2097 if (dev->xfer_shift != ATA_SHIFT_PIO)
2098 sel = dma_dnxfer_sel[dev->spdn_cnt];
2099 else
2100 sel = pio_dnxfer_sel[dev->spdn_cnt];
2101
2102 dev->spdn_cnt++;
2103
2104 if (ata_down_xfermask_limit(dev, sel) == 0) {
2105 action |= ATA_EH_RESET;
2106 goto done;
2107 }
2108 }
2109 }
2110
2111 /* Fall back to PIO? Slowing down to PIO is meaningless for
2112 * SATA ATA devices. Consider it only for PATA and SATAPI.
2113 */
2114 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2115 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2116 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2117 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2118 dev->spdn_cnt = 0;
2119 action |= ATA_EH_RESET;
2120 goto done;
2121 }
2122 }
2123
2124 return 0;
2125 done:
2126 /* device has been slowed down, blow error history */
2127 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2128 ata_ering_clear(&dev->ering);
2129 return action;
2130 }
2131
2132 /**
2133 * ata_eh_worth_retry - analyze error and decide whether to retry
2134 * @qc: qc to possibly retry
2135 *
2136 * Look at the cause of the error and decide if a retry
2137 * might be useful or not. We don't want to retry media errors
2138 * because the drive itself has probably already taken 10-30 seconds
2139 * doing its own internal retries before reporting the failure.
2140 */
2141 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2142 {
2143 if (qc->err_mask & AC_ERR_MEDIA)
2144 return 0; /* don't retry media errors */
2145 if (qc->flags & ATA_QCFLAG_IO)
2146 return 1; /* otherwise retry anything from fs stack */
2147 if (qc->err_mask & AC_ERR_INVALID)
2148 return 0; /* don't retry these */
2149 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2150 }
2151
2152 /**
2153 * ata_eh_link_autopsy - analyze error and determine recovery action
2154 * @link: host link to perform autopsy on
2155 *
2156 * Analyze why @link failed and determine which recovery actions
2157 * are needed. This function also sets more detailed AC_ERR_*
2158 * values and fills sense data for ATAPI CHECK SENSE.
2159 *
2160 * LOCKING:
2161 * Kernel thread context (may sleep).
2162 */
2163 static void ata_eh_link_autopsy(struct ata_link *link)
2164 {
2165 struct ata_port *ap = link->ap;
2166 struct ata_eh_context *ehc = &link->eh_context;
2167 struct ata_device *dev;
2168 unsigned int all_err_mask = 0, eflags = 0;
2169 int tag;
2170 u32 serror;
2171 int rc;
2172
2173 DPRINTK("ENTER\n");
2174
2175 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2176 return;
2177
2178 /* obtain and analyze SError */
2179 rc = sata_scr_read(link, SCR_ERROR, &serror);
2180 if (rc == 0) {
2181 ehc->i.serror |= serror;
2182 ata_eh_analyze_serror(link);
2183 } else if (rc != -EOPNOTSUPP) {
2184 /* SError read failed, force reset and probing */
2185 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2186 ehc->i.action |= ATA_EH_RESET;
2187 ehc->i.err_mask |= AC_ERR_OTHER;
2188 }
2189
2190 /* analyze NCQ failure */
2191 ata_eh_analyze_ncq_error(link);
2192
2193 /* any real error trumps AC_ERR_OTHER */
2194 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2195 ehc->i.err_mask &= ~AC_ERR_OTHER;
2196
2197 all_err_mask |= ehc->i.err_mask;
2198
2199 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2200 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2201
2202 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2203 ata_dev_phys_link(qc->dev) != link)
2204 continue;
2205
2206 /* inherit upper level err_mask */
2207 qc->err_mask |= ehc->i.err_mask;
2208
2209 /* analyze TF */
2210 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2211
2212 /* DEV errors are probably spurious in case of ATA_BUS error */
2213 if (qc->err_mask & AC_ERR_ATA_BUS)
2214 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2215 AC_ERR_INVALID);
2216
2217 /* any real error trumps unknown error */
2218 if (qc->err_mask & ~AC_ERR_OTHER)
2219 qc->err_mask &= ~AC_ERR_OTHER;
2220
2221 /* SENSE_VALID trumps dev/unknown error and revalidation */
2222 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2223 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2224
2225 /* determine whether the command is worth retrying */
2226 if (ata_eh_worth_retry(qc))
2227 qc->flags |= ATA_QCFLAG_RETRY;
2228
2229 /* accumulate error info */
2230 ehc->i.dev = qc->dev;
2231 all_err_mask |= qc->err_mask;
2232 if (qc->flags & ATA_QCFLAG_IO)
2233 eflags |= ATA_EFLAG_IS_IO;
2234 trace_ata_eh_link_autopsy_qc(qc);
2235 }
2236
2237 /* enforce default EH actions */
2238 if (ap->pflags & ATA_PFLAG_FROZEN ||
2239 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2240 ehc->i.action |= ATA_EH_RESET;
2241 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2242 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2243 ehc->i.action |= ATA_EH_REVALIDATE;
2244
2245 /* If we have offending qcs and the associated failed device,
2246 * perform per-dev EH action only on the offending device.
2247 */
2248 if (ehc->i.dev) {
2249 ehc->i.dev_action[ehc->i.dev->devno] |=
2250 ehc->i.action & ATA_EH_PERDEV_MASK;
2251 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2252 }
2253
2254 /* propagate timeout to host link */
2255 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2256 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2257
2258 /* record error and consider speeding down */
2259 dev = ehc->i.dev;
2260 if (!dev && ((ata_link_max_devices(link) == 1 &&
2261 ata_dev_enabled(link->device))))
2262 dev = link->device;
2263
2264 if (dev) {
2265 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2266 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2267 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2268 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2269 }
2270 DPRINTK("EXIT\n");
2271 }
2272
2273 /**
2274 * ata_eh_autopsy - analyze error and determine recovery action
2275 * @ap: host port to perform autopsy on
2276 *
2277 * Analyze all links of @ap and determine why they failed and
2278 * which recovery actions are needed.
2279 *
2280 * LOCKING:
2281 * Kernel thread context (may sleep).
2282 */
2283 void ata_eh_autopsy(struct ata_port *ap)
2284 {
2285 struct ata_link *link;
2286
2287 ata_for_each_link(link, ap, EDGE)
2288 ata_eh_link_autopsy(link);
2289
2290 /* Handle the frigging slave link. Autopsy is done similarly
2291 * but actions and flags are transferred over to the master
2292 * link and handled from there.
2293 */
2294 if (ap->slave_link) {
2295 struct ata_eh_context *mehc = &ap->link.eh_context;
2296 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2297
2298 /* transfer control flags from master to slave */
2299 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2300
2301 /* perform autopsy on the slave link */
2302 ata_eh_link_autopsy(ap->slave_link);
2303
2304 /* transfer actions from slave to master and clear slave */
2305 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2306 mehc->i.action |= sehc->i.action;
2307 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2308 mehc->i.flags |= sehc->i.flags;
2309 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2310 }
2311
2312 /* Autopsy of fanout ports can affect host link autopsy.
2313 * Perform host link autopsy last.
2314 */
2315 if (sata_pmp_attached(ap))
2316 ata_eh_link_autopsy(&ap->link);
2317 }
2318
2319 /**
2320 * ata_get_cmd_descript - get description for ATA command
2321 * @command: ATA command code to get description for
2322 *
2323 * Return a textual description of the given command, or NULL if the
2324 * command is not known.
2325 *
2326 * LOCKING:
2327 * None
2328 */
2329 const char *ata_get_cmd_descript(u8 command)
2330 {
2331 #ifdef CONFIG_ATA_VERBOSE_ERROR
2332 static const struct
2333 {
2334 u8 command;
2335 const char *text;
2336 } cmd_descr[] = {
2337 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2338 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2339 { ATA_CMD_STANDBY, "STANDBY" },
2340 { ATA_CMD_IDLE, "IDLE" },
2341 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2342 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2343 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2344 { ATA_CMD_NOP, "NOP" },
2345 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2346 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2347 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2348 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2349 { ATA_CMD_SERVICE, "SERVICE" },
2350 { ATA_CMD_READ, "READ DMA" },
2351 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2352 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2353 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2354 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2355 { ATA_CMD_WRITE, "WRITE DMA" },
2356 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2357 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2358 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2359 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2360 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2361 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2362 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2363 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2364 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2365 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2366 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2367 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2368 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2369 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2370 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2371 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2372 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2373 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2374 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2375 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2376 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2377 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2378 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2379 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2380 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2381 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2382 { ATA_CMD_SLEEP, "SLEEP" },
2383 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2384 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2385 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2386 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2387 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2388 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2389 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2390 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2391 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2392 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2393 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2394 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2395 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2396 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2397 { ATA_CMD_PMP_READ, "READ BUFFER" },
2398 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2399 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2400 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2401 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2402 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2403 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2404 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2405 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2406 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2407 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2408 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2409 { ATA_CMD_SMART, "SMART" },
2410 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2411 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2412 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2413 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2414 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2415 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2416 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2417 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2418 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2419 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2420 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2421 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2422 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2423 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2424 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2425 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2426 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2427 { ATA_CMD_RESTORE, "RECALIBRATE" },
2428 { 0, NULL } /* terminate list */
2429 };
2430
2431 unsigned int i;
2432 for (i = 0; cmd_descr[i].text; i++)
2433 if (cmd_descr[i].command == command)
2434 return cmd_descr[i].text;
2435 #endif
2436
2437 return NULL;
2438 }
2439 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2440
2441 /**
2442 * ata_eh_link_report - report error handling to user
2443 * @link: ATA link EH is going on
2444 *
2445 * Report EH to user.
2446 *
2447 * LOCKING:
2448 * None.
2449 */
2450 static void ata_eh_link_report(struct ata_link *link)
2451 {
2452 struct ata_port *ap = link->ap;
2453 struct ata_eh_context *ehc = &link->eh_context;
2454 const char *frozen, *desc;
2455 char tries_buf[6] = "";
2456 int tag, nr_failed = 0;
2457
2458 if (ehc->i.flags & ATA_EHI_QUIET)
2459 return;
2460
2461 desc = NULL;
2462 if (ehc->i.desc[0] != '\0')
2463 desc = ehc->i.desc;
2464
2465 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2466 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2467
2468 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2469 ata_dev_phys_link(qc->dev) != link ||
2470 ((qc->flags & ATA_QCFLAG_QUIET) &&
2471 qc->err_mask == AC_ERR_DEV))
2472 continue;
2473 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2474 continue;
2475
2476 nr_failed++;
2477 }
2478
2479 if (!nr_failed && !ehc->i.err_mask)
2480 return;
2481
2482 frozen = "";
2483 if (ap->pflags & ATA_PFLAG_FROZEN)
2484 frozen = " frozen";
2485
2486 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2487 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2488 ap->eh_tries);
2489
2490 if (ehc->i.dev) {
2491 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2492 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2493 ehc->i.err_mask, link->sactive, ehc->i.serror,
2494 ehc->i.action, frozen, tries_buf);
2495 if (desc)
2496 ata_dev_err(ehc->i.dev, "%s\n", desc);
2497 } else {
2498 ata_link_err(link, "exception Emask 0x%x "
2499 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2500 ehc->i.err_mask, link->sactive, ehc->i.serror,
2501 ehc->i.action, frozen, tries_buf);
2502 if (desc)
2503 ata_link_err(link, "%s\n", desc);
2504 }
2505
2506 #ifdef CONFIG_ATA_VERBOSE_ERROR
2507 if (ehc->i.serror)
2508 ata_link_err(link,
2509 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2510 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2511 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2512 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2513 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2514 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2515 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2516 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2517 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2518 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2519 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2520 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2521 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2522 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2523 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2524 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2525 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2526 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2527 #endif
2528
2529 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2530 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2531 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2532 char data_buf[20] = "";
2533 char cdb_buf[70] = "";
2534
2535 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2536 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2537 continue;
2538
2539 if (qc->dma_dir != DMA_NONE) {
2540 static const char *dma_str[] = {
2541 [DMA_BIDIRECTIONAL] = "bidi",
2542 [DMA_TO_DEVICE] = "out",
2543 [DMA_FROM_DEVICE] = "in",
2544 };
2545 const char *prot_str = NULL;
2546
2547 switch (qc->tf.protocol) {
2548 case ATA_PROT_UNKNOWN:
2549 prot_str = "unknown";
2550 break;
2551 case ATA_PROT_NODATA:
2552 prot_str = "nodata";
2553 break;
2554 case ATA_PROT_PIO:
2555 prot_str = "pio";
2556 break;
2557 case ATA_PROT_DMA:
2558 prot_str = "dma";
2559 break;
2560 case ATA_PROT_NCQ:
2561 prot_str = "ncq dma";
2562 break;
2563 case ATA_PROT_NCQ_NODATA:
2564 prot_str = "ncq nodata";
2565 break;
2566 case ATAPI_PROT_NODATA:
2567 prot_str = "nodata";
2568 break;
2569 case ATAPI_PROT_PIO:
2570 prot_str = "pio";
2571 break;
2572 case ATAPI_PROT_DMA:
2573 prot_str = "dma";
2574 break;
2575 }
2576 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2577 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2578 }
2579
2580 if (ata_is_atapi(qc->tf.protocol)) {
2581 const u8 *cdb = qc->cdb;
2582 size_t cdb_len = qc->dev->cdb_len;
2583
2584 if (qc->scsicmd) {
2585 cdb = qc->scsicmd->cmnd;
2586 cdb_len = qc->scsicmd->cmd_len;
2587 }
2588 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2589 cdb, cdb_len);
2590 } else {
2591 const char *descr = ata_get_cmd_descript(cmd->command);
2592 if (descr)
2593 ata_dev_err(qc->dev, "failed command: %s\n",
2594 descr);
2595 }
2596
2597 ata_dev_err(qc->dev,
2598 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2599 "tag %d%s\n %s"
2600 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2601 "Emask 0x%x (%s)%s\n",
2602 cmd->command, cmd->feature, cmd->nsect,
2603 cmd->lbal, cmd->lbam, cmd->lbah,
2604 cmd->hob_feature, cmd->hob_nsect,
2605 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2606 cmd->device, qc->tag, data_buf, cdb_buf,
2607 res->command, res->feature, res->nsect,
2608 res->lbal, res->lbam, res->lbah,
2609 res->hob_feature, res->hob_nsect,
2610 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2611 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2612 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2613
2614 #ifdef CONFIG_ATA_VERBOSE_ERROR
2615 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2616 ATA_SENSE | ATA_ERR)) {
2617 if (res->command & ATA_BUSY)
2618 ata_dev_err(qc->dev, "status: { Busy }\n");
2619 else
2620 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2621 res->command & ATA_DRDY ? "DRDY " : "",
2622 res->command & ATA_DF ? "DF " : "",
2623 res->command & ATA_DRQ ? "DRQ " : "",
2624 res->command & ATA_SENSE ? "SENSE " : "",
2625 res->command & ATA_ERR ? "ERR " : "");
2626 }
2627
2628 if (cmd->command != ATA_CMD_PACKET &&
2629 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2630 ATA_IDNF | ATA_ABORTED)))
2631 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2632 res->feature & ATA_ICRC ? "ICRC " : "",
2633 res->feature & ATA_UNC ? "UNC " : "",
2634 res->feature & ATA_AMNF ? "AMNF " : "",
2635 res->feature & ATA_IDNF ? "IDNF " : "",
2636 res->feature & ATA_ABORTED ? "ABRT " : "");
2637 #endif
2638 }
2639 }
2640
2641 /**
2642 * ata_eh_report - report error handling to user
2643 * @ap: ATA port to report EH about
2644 *
2645 * Report EH to user.
2646 *
2647 * LOCKING:
2648 * None.
2649 */
2650 void ata_eh_report(struct ata_port *ap)
2651 {
2652 struct ata_link *link;
2653
2654 ata_for_each_link(link, ap, HOST_FIRST)
2655 ata_eh_link_report(link);
2656 }
2657
2658 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2659 unsigned int *classes, unsigned long deadline,
2660 bool clear_classes)
2661 {
2662 struct ata_device *dev;
2663
2664 if (clear_classes)
2665 ata_for_each_dev(dev, link, ALL)
2666 classes[dev->devno] = ATA_DEV_UNKNOWN;
2667
2668 return reset(link, classes, deadline);
2669 }
2670
2671 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2672 {
2673 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2674 return 0;
2675 if (rc == -EAGAIN)
2676 return 1;
2677 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2678 return 1;
2679 return 0;
2680 }
2681
2682 int ata_eh_reset(struct ata_link *link, int classify,
2683 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2684 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2685 {
2686 struct ata_port *ap = link->ap;
2687 struct ata_link *slave = ap->slave_link;
2688 struct ata_eh_context *ehc = &link->eh_context;
2689 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2690 unsigned int *classes = ehc->classes;
2691 unsigned int lflags = link->flags;
2692 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2693 int max_tries = 0, try = 0;
2694 struct ata_link *failed_link;
2695 struct ata_device *dev;
2696 unsigned long deadline, now;
2697 ata_reset_fn_t reset;
2698 unsigned long flags;
2699 u32 sstatus;
2700 int nr_unknown, rc;
2701
2702 /*
2703 * Prepare to reset
2704 */
2705 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2706 max_tries++;
2707 if (link->flags & ATA_LFLAG_RST_ONCE)
2708 max_tries = 1;
2709 if (link->flags & ATA_LFLAG_NO_HRST)
2710 hardreset = NULL;
2711 if (link->flags & ATA_LFLAG_NO_SRST)
2712 softreset = NULL;
2713
2714 /* make sure each reset attempt is at least COOL_DOWN apart */
2715 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2716 now = jiffies;
2717 WARN_ON(time_after(ehc->last_reset, now));
2718 deadline = ata_deadline(ehc->last_reset,
2719 ATA_EH_RESET_COOL_DOWN);
2720 if (time_before(now, deadline))
2721 schedule_timeout_uninterruptible(deadline - now);
2722 }
2723
2724 spin_lock_irqsave(ap->lock, flags);
2725 ap->pflags |= ATA_PFLAG_RESETTING;
2726 spin_unlock_irqrestore(ap->lock, flags);
2727
2728 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2729
2730 ata_for_each_dev(dev, link, ALL) {
2731 /* If we issue an SRST then an ATA drive (not ATAPI)
2732 * may change configuration and be in PIO0 timing. If
2733 * we do a hard reset (or are coming from power on)
2734 * this is true for ATA or ATAPI. Until we've set a
2735 * suitable controller mode we should not touch the
2736 * bus as we may be talking too fast.
2737 */
2738 dev->pio_mode = XFER_PIO_0;
2739 dev->dma_mode = 0xff;
2740
2741 /* If the controller has a pio mode setup function
2742 * then use it to set the chipset to rights. Don't
2743 * touch the DMA setup as that will be dealt with when
2744 * configuring devices.
2745 */
2746 if (ap->ops->set_piomode)
2747 ap->ops->set_piomode(ap, dev);
2748 }
2749
2750 /* prefer hardreset */
2751 reset = NULL;
2752 ehc->i.action &= ~ATA_EH_RESET;
2753 if (hardreset) {
2754 reset = hardreset;
2755 ehc->i.action |= ATA_EH_HARDRESET;
2756 } else if (softreset) {
2757 reset = softreset;
2758 ehc->i.action |= ATA_EH_SOFTRESET;
2759 }
2760
2761 if (prereset) {
2762 unsigned long deadline = ata_deadline(jiffies,
2763 ATA_EH_PRERESET_TIMEOUT);
2764
2765 if (slave) {
2766 sehc->i.action &= ~ATA_EH_RESET;
2767 sehc->i.action |= ehc->i.action;
2768 }
2769
2770 rc = prereset(link, deadline);
2771
2772 /* If present, do prereset on slave link too. Reset
2773 * is skipped iff both master and slave links report
2774 * -ENOENT or clear ATA_EH_RESET.
2775 */
2776 if (slave && (rc == 0 || rc == -ENOENT)) {
2777 int tmp;
2778
2779 tmp = prereset(slave, deadline);
2780 if (tmp != -ENOENT)
2781 rc = tmp;
2782
2783 ehc->i.action |= sehc->i.action;
2784 }
2785
2786 if (rc) {
2787 if (rc == -ENOENT) {
2788 ata_link_dbg(link, "port disabled--ignoring\n");
2789 ehc->i.action &= ~ATA_EH_RESET;
2790
2791 ata_for_each_dev(dev, link, ALL)
2792 classes[dev->devno] = ATA_DEV_NONE;
2793
2794 rc = 0;
2795 } else
2796 ata_link_err(link,
2797 "prereset failed (errno=%d)\n",
2798 rc);
2799 goto out;
2800 }
2801
2802 /* prereset() might have cleared ATA_EH_RESET. If so,
2803 * bang classes, thaw and return.
2804 */
2805 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2806 ata_for_each_dev(dev, link, ALL)
2807 classes[dev->devno] = ATA_DEV_NONE;
2808 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2809 ata_is_host_link(link))
2810 ata_eh_thaw_port(ap);
2811 rc = 0;
2812 goto out;
2813 }
2814 }
2815
2816 retry:
2817 /*
2818 * Perform reset
2819 */
2820 if (ata_is_host_link(link))
2821 ata_eh_freeze_port(ap);
2822
2823 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2824
2825 if (reset) {
2826 if (verbose)
2827 ata_link_info(link, "%s resetting link\n",
2828 reset == softreset ? "soft" : "hard");
2829
2830 /* mark that this EH session started with reset */
2831 ehc->last_reset = jiffies;
2832 if (reset == hardreset)
2833 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2834 else
2835 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2836
2837 rc = ata_do_reset(link, reset, classes, deadline, true);
2838 if (rc && rc != -EAGAIN) {
2839 failed_link = link;
2840 goto fail;
2841 }
2842
2843 /* hardreset slave link if existent */
2844 if (slave && reset == hardreset) {
2845 int tmp;
2846
2847 if (verbose)
2848 ata_link_info(slave, "hard resetting link\n");
2849
2850 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2851 tmp = ata_do_reset(slave, reset, classes, deadline,
2852 false);
2853 switch (tmp) {
2854 case -EAGAIN:
2855 rc = -EAGAIN;
2856 case 0:
2857 break;
2858 default:
2859 failed_link = slave;
2860 rc = tmp;
2861 goto fail;
2862 }
2863 }
2864
2865 /* perform follow-up SRST if necessary */
2866 if (reset == hardreset &&
2867 ata_eh_followup_srst_needed(link, rc)) {
2868 reset = softreset;
2869
2870 if (!reset) {
2871 ata_link_err(link,
2872 "follow-up softreset required but no softreset available\n");
2873 failed_link = link;
2874 rc = -EINVAL;
2875 goto fail;
2876 }
2877
2878 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2879 rc = ata_do_reset(link, reset, classes, deadline, true);
2880 if (rc) {
2881 failed_link = link;
2882 goto fail;
2883 }
2884 }
2885 } else {
2886 if (verbose)
2887 ata_link_info(link,
2888 "no reset method available, skipping reset\n");
2889 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2890 lflags |= ATA_LFLAG_ASSUME_ATA;
2891 }
2892
2893 /*
2894 * Post-reset processing
2895 */
2896 ata_for_each_dev(dev, link, ALL) {
2897 /* After the reset, the device state is PIO 0 and the
2898 * controller state is undefined. Reset also wakes up
2899 * drives from sleeping mode.
2900 */
2901 dev->pio_mode = XFER_PIO_0;
2902 dev->flags &= ~ATA_DFLAG_SLEEPING;
2903
2904 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2905 continue;
2906
2907 /* apply class override */
2908 if (lflags & ATA_LFLAG_ASSUME_ATA)
2909 classes[dev->devno] = ATA_DEV_ATA;
2910 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2911 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2912 }
2913
2914 /* record current link speed */
2915 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2916 link->sata_spd = (sstatus >> 4) & 0xf;
2917 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2918 slave->sata_spd = (sstatus >> 4) & 0xf;
2919
2920 /* thaw the port */
2921 if (ata_is_host_link(link))
2922 ata_eh_thaw_port(ap);
2923
2924 /* postreset() should clear hardware SError. Although SError
2925 * is cleared during link resume, clearing SError here is
2926 * necessary as some PHYs raise hotplug events after SRST.
2927 * This introduces race condition where hotplug occurs between
2928 * reset and here. This race is mediated by cross checking
2929 * link onlineness and classification result later.
2930 */
2931 if (postreset) {
2932 postreset(link, classes);
2933 if (slave)
2934 postreset(slave, classes);
2935 }
2936
2937 /*
2938 * Some controllers can't be frozen very well and may set spurious
2939 * error conditions during reset. Clear accumulated error
2940 * information and re-thaw the port if frozen. As reset is the
2941 * final recovery action and we cross check link onlineness against
2942 * device classification later, no hotplug event is lost by this.
2943 */
2944 spin_lock_irqsave(link->ap->lock, flags);
2945 memset(&link->eh_info, 0, sizeof(link->eh_info));
2946 if (slave)
2947 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2948 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2949 spin_unlock_irqrestore(link->ap->lock, flags);
2950
2951 if (ap->pflags & ATA_PFLAG_FROZEN)
2952 ata_eh_thaw_port(ap);
2953
2954 /*
2955 * Make sure onlineness and classification result correspond.
2956 * Hotplug could have happened during reset and some
2957 * controllers fail to wait while a drive is spinning up after
2958 * being hotplugged causing misdetection. By cross checking
2959 * link on/offlineness and classification result, those
2960 * conditions can be reliably detected and retried.
2961 */
2962 nr_unknown = 0;
2963 ata_for_each_dev(dev, link, ALL) {
2964 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2965 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2966 ata_dev_dbg(dev, "link online but device misclassified\n");
2967 classes[dev->devno] = ATA_DEV_NONE;
2968 nr_unknown++;
2969 }
2970 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2971 if (ata_class_enabled(classes[dev->devno]))
2972 ata_dev_dbg(dev,
2973 "link offline, clearing class %d to NONE\n",
2974 classes[dev->devno]);
2975 classes[dev->devno] = ATA_DEV_NONE;
2976 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2977 ata_dev_dbg(dev,
2978 "link status unknown, clearing UNKNOWN to NONE\n");
2979 classes[dev->devno] = ATA_DEV_NONE;
2980 }
2981 }
2982
2983 if (classify && nr_unknown) {
2984 if (try < max_tries) {
2985 ata_link_warn(link,
2986 "link online but %d devices misclassified, retrying\n",
2987 nr_unknown);
2988 failed_link = link;
2989 rc = -EAGAIN;
2990 goto fail;
2991 }
2992 ata_link_warn(link,
2993 "link online but %d devices misclassified, "
2994 "device detection might fail\n", nr_unknown);
2995 }
2996
2997 /* reset successful, schedule revalidation */
2998 ata_eh_done(link, NULL, ATA_EH_RESET);
2999 if (slave)
3000 ata_eh_done(slave, NULL, ATA_EH_RESET);
3001 ehc->last_reset = jiffies; /* update to completion time */
3002 ehc->i.action |= ATA_EH_REVALIDATE;
3003 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
3004
3005 rc = 0;
3006 out:
3007 /* clear hotplug flag */
3008 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3009 if (slave)
3010 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
3011
3012 spin_lock_irqsave(ap->lock, flags);
3013 ap->pflags &= ~ATA_PFLAG_RESETTING;
3014 spin_unlock_irqrestore(ap->lock, flags);
3015
3016 return rc;
3017
3018 fail:
3019 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
3020 if (!ata_is_host_link(link) &&
3021 sata_scr_read(link, SCR_STATUS, &sstatus))
3022 rc = -ERESTART;
3023
3024 if (try >= max_tries) {
3025 /*
3026 * Thaw host port even if reset failed, so that the port
3027 * can be retried on the next phy event. This risks
3028 * repeated EH runs but seems to be a better tradeoff than
3029 * shutting down a port after a botched hotplug attempt.
3030 */
3031 if (ata_is_host_link(link))
3032 ata_eh_thaw_port(ap);
3033 goto out;
3034 }
3035
3036 now = jiffies;
3037 if (time_before(now, deadline)) {
3038 unsigned long delta = deadline - now;
3039
3040 ata_link_warn(failed_link,
3041 "reset failed (errno=%d), retrying in %u secs\n",
3042 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3043
3044 ata_eh_release(ap);
3045 while (delta)
3046 delta = schedule_timeout_uninterruptible(delta);
3047 ata_eh_acquire(ap);
3048 }
3049
3050 /*
3051 * While disks spinup behind PMP, some controllers fail sending SRST.
3052 * They need to be reset - as well as the PMP - before retrying.
3053 */
3054 if (rc == -ERESTART) {
3055 if (ata_is_host_link(link))
3056 ata_eh_thaw_port(ap);
3057 goto out;
3058 }
3059
3060 if (try == max_tries - 1) {
3061 sata_down_spd_limit(link, 0);
3062 if (slave)
3063 sata_down_spd_limit(slave, 0);
3064 } else if (rc == -EPIPE)
3065 sata_down_spd_limit(failed_link, 0);
3066
3067 if (hardreset)
3068 reset = hardreset;
3069 goto retry;
3070 }
3071
3072 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3073 {
3074 struct ata_link *link;
3075 struct ata_device *dev;
3076 unsigned long flags;
3077
3078 /*
3079 * This function can be thought of as an extended version of
3080 * ata_eh_about_to_do() specially crafted to accommodate the
3081 * requirements of ATA_EH_PARK handling. Since the EH thread
3082 * does not leave the do {} while () loop in ata_eh_recover as
3083 * long as the timeout for a park request to *one* device on
3084 * the port has not expired, and since we still want to pick
3085 * up park requests to other devices on the same port or
3086 * timeout updates for the same device, we have to pull
3087 * ATA_EH_PARK actions from eh_info into eh_context.i
3088 * ourselves at the beginning of each pass over the loop.
3089 *
3090 * Additionally, all write accesses to &ap->park_req_pending
3091 * through reinit_completion() (see below) or complete_all()
3092 * (see ata_scsi_park_store()) are protected by the host lock.
3093 * As a result we have that park_req_pending.done is zero on
3094 * exit from this function, i.e. when ATA_EH_PARK actions for
3095 * *all* devices on port ap have been pulled into the
3096 * respective eh_context structs. If, and only if,
3097 * park_req_pending.done is non-zero by the time we reach
3098 * wait_for_completion_timeout(), another ATA_EH_PARK action
3099 * has been scheduled for at least one of the devices on port
3100 * ap and we have to cycle over the do {} while () loop in
3101 * ata_eh_recover() again.
3102 */
3103
3104 spin_lock_irqsave(ap->lock, flags);
3105 reinit_completion(&ap->park_req_pending);
3106 ata_for_each_link(link, ap, EDGE) {
3107 ata_for_each_dev(dev, link, ALL) {
3108 struct ata_eh_info *ehi = &link->eh_info;
3109
3110 link->eh_context.i.dev_action[dev->devno] |=
3111 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3112 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3113 }
3114 }
3115 spin_unlock_irqrestore(ap->lock, flags);
3116 }
3117
3118 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3119 {
3120 struct ata_eh_context *ehc = &dev->link->eh_context;
3121 struct ata_taskfile tf;
3122 unsigned int err_mask;
3123
3124 ata_tf_init(dev, &tf);
3125 if (park) {
3126 ehc->unloaded_mask |= 1 << dev->devno;
3127 tf.command = ATA_CMD_IDLEIMMEDIATE;
3128 tf.feature = 0x44;
3129 tf.lbal = 0x4c;
3130 tf.lbam = 0x4e;
3131 tf.lbah = 0x55;
3132 } else {
3133 ehc->unloaded_mask &= ~(1 << dev->devno);
3134 tf.command = ATA_CMD_CHK_POWER;
3135 }
3136
3137 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3138 tf.protocol = ATA_PROT_NODATA;
3139 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3140 if (park && (err_mask || tf.lbal != 0xc4)) {
3141 ata_dev_err(dev, "head unload failed!\n");
3142 ehc->unloaded_mask &= ~(1 << dev->devno);
3143 }
3144 }
3145
3146 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3147 struct ata_device **r_failed_dev)
3148 {
3149 struct ata_port *ap = link->ap;
3150 struct ata_eh_context *ehc = &link->eh_context;
3151 struct ata_device *dev;
3152 unsigned int new_mask = 0;
3153 unsigned long flags;
3154 int rc = 0;
3155
3156 DPRINTK("ENTER\n");
3157
3158 /* For PATA drive side cable detection to work, IDENTIFY must
3159 * be done backwards such that PDIAG- is released by the slave
3160 * device before the master device is identified.
3161 */
3162 ata_for_each_dev(dev, link, ALL_REVERSE) {
3163 unsigned int action = ata_eh_dev_action(dev);
3164 unsigned int readid_flags = 0;
3165
3166 if (ehc->i.flags & ATA_EHI_DID_RESET)
3167 readid_flags |= ATA_READID_POSTRESET;
3168
3169 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3170 WARN_ON(dev->class == ATA_DEV_PMP);
3171
3172 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3173 rc = -EIO;
3174 goto err;
3175 }
3176
3177 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3178 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3179 readid_flags);
3180 if (rc)
3181 goto err;
3182
3183 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3184
3185 /* Configuration may have changed, reconfigure
3186 * transfer mode.
3187 */
3188 ehc->i.flags |= ATA_EHI_SETMODE;
3189
3190 /* schedule the scsi_rescan_device() here */
3191 schedule_work(&(ap->scsi_rescan_task));
3192 } else if (dev->class == ATA_DEV_UNKNOWN &&
3193 ehc->tries[dev->devno] &&
3194 ata_class_enabled(ehc->classes[dev->devno])) {
3195 /* Temporarily set dev->class, it will be
3196 * permanently set once all configurations are
3197 * complete. This is necessary because new
3198 * device configuration is done in two
3199 * separate loops.
3200 */
3201 dev->class = ehc->classes[dev->devno];
3202
3203 if (dev->class == ATA_DEV_PMP)
3204 rc = sata_pmp_attach(dev);
3205 else
3206 rc = ata_dev_read_id(dev, &dev->class,
3207 readid_flags, dev->id);
3208
3209 /* read_id might have changed class, store and reset */
3210 ehc->classes[dev->devno] = dev->class;
3211 dev->class = ATA_DEV_UNKNOWN;
3212
3213 switch (rc) {
3214 case 0:
3215 /* clear error info accumulated during probe */
3216 ata_ering_clear(&dev->ering);
3217 new_mask |= 1 << dev->devno;
3218 break;
3219 case -ENOENT:
3220 /* IDENTIFY was issued to non-existent
3221 * device. No need to reset. Just
3222 * thaw and ignore the device.
3223 */
3224 ata_eh_thaw_port(ap);
3225 break;
3226 default:
3227 goto err;
3228 }
3229 }
3230 }
3231
3232 /* PDIAG- should have been released, ask cable type if post-reset */
3233 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3234 if (ap->ops->cable_detect)
3235 ap->cbl = ap->ops->cable_detect(ap);
3236 ata_force_cbl(ap);
3237 }
3238
3239 /* Configure new devices forward such that user doesn't see
3240 * device detection messages backwards.
3241 */
3242 ata_for_each_dev(dev, link, ALL) {
3243 if (!(new_mask & (1 << dev->devno)))
3244 continue;
3245
3246 dev->class = ehc->classes[dev->devno];
3247
3248 if (dev->class == ATA_DEV_PMP)
3249 continue;
3250
3251 ehc->i.flags |= ATA_EHI_PRINTINFO;
3252 rc = ata_dev_configure(dev);
3253 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3254 if (rc) {
3255 dev->class = ATA_DEV_UNKNOWN;
3256 goto err;
3257 }
3258
3259 spin_lock_irqsave(ap->lock, flags);
3260 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3261 spin_unlock_irqrestore(ap->lock, flags);
3262
3263 /* new device discovered, configure xfermode */
3264 ehc->i.flags |= ATA_EHI_SETMODE;
3265 }
3266
3267 return 0;
3268
3269 err:
3270 *r_failed_dev = dev;
3271 DPRINTK("EXIT rc=%d\n", rc);
3272 return rc;
3273 }
3274
3275 /**
3276 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3277 * @link: link on which timings will be programmed
3278 * @r_failed_dev: out parameter for failed device
3279 *
3280 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3281 * ata_set_mode() fails, pointer to the failing device is
3282 * returned in @r_failed_dev.
3283 *
3284 * LOCKING:
3285 * PCI/etc. bus probe sem.
3286 *
3287 * RETURNS:
3288 * 0 on success, negative errno otherwise
3289 */
3290 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3291 {
3292 struct ata_port *ap = link->ap;
3293 struct ata_device *dev;
3294 int rc;
3295
3296 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3297 ata_for_each_dev(dev, link, ENABLED) {
3298 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3299 struct ata_ering_entry *ent;
3300
3301 ent = ata_ering_top(&dev->ering);
3302 if (ent)
3303 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3304 }
3305 }
3306
3307 /* has private set_mode? */
3308 if (ap->ops->set_mode)
3309 rc = ap->ops->set_mode(link, r_failed_dev);
3310 else
3311 rc = ata_do_set_mode(link, r_failed_dev);
3312
3313 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3314 ata_for_each_dev(dev, link, ENABLED) {
3315 struct ata_eh_context *ehc = &link->eh_context;
3316 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3317 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3318
3319 if (dev->xfer_mode != saved_xfer_mode ||
3320 ata_ncq_enabled(dev) != saved_ncq)
3321 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3322 }
3323
3324 return rc;
3325 }
3326
3327 /**
3328 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3329 * @dev: ATAPI device to clear UA for
3330 *
3331 * Resets and other operations can make an ATAPI device raise
3332 * UNIT ATTENTION which causes the next operation to fail. This
3333 * function clears UA.
3334 *
3335 * LOCKING:
3336 * EH context (may sleep).
3337 *
3338 * RETURNS:
3339 * 0 on success, -errno on failure.
3340 */
3341 static int atapi_eh_clear_ua(struct ata_device *dev)
3342 {
3343 int i;
3344
3345 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3346 u8 *sense_buffer = dev->link->ap->sector_buf;
3347 u8 sense_key = 0;
3348 unsigned int err_mask;
3349
3350 err_mask = atapi_eh_tur(dev, &sense_key);
3351 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3352 ata_dev_warn(dev,
3353 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3354 err_mask);
3355 return -EIO;
3356 }
3357
3358 if (!err_mask || sense_key != UNIT_ATTENTION)
3359 return 0;
3360
3361 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3362 if (err_mask) {
3363 ata_dev_warn(dev, "failed to clear "
3364 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3365 return -EIO;
3366 }
3367 }
3368
3369 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3370 ATA_EH_UA_TRIES);
3371
3372 return 0;
3373 }
3374
3375 /**
3376 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3377 * @dev: ATA device which may need FLUSH retry
3378 *
3379 * If @dev failed FLUSH, it needs to be reported upper layer
3380 * immediately as it means that @dev failed to remap and already
3381 * lost at least a sector and further FLUSH retrials won't make
3382 * any difference to the lost sector. However, if FLUSH failed
3383 * for other reasons, for example transmission error, FLUSH needs
3384 * to be retried.
3385 *
3386 * This function determines whether FLUSH failure retry is
3387 * necessary and performs it if so.
3388 *
3389 * RETURNS:
3390 * 0 if EH can continue, -errno if EH needs to be repeated.
3391 */
3392 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3393 {
3394 struct ata_link *link = dev->link;
3395 struct ata_port *ap = link->ap;
3396 struct ata_queued_cmd *qc;
3397 struct ata_taskfile tf;
3398 unsigned int err_mask;
3399 int rc = 0;
3400
3401 /* did flush fail for this device? */
3402 if (!ata_tag_valid(link->active_tag))
3403 return 0;
3404
3405 qc = __ata_qc_from_tag(ap, link->active_tag);
3406 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3407 qc->tf.command != ATA_CMD_FLUSH))
3408 return 0;
3409
3410 /* if the device failed it, it should be reported to upper layers */
3411 if (qc->err_mask & AC_ERR_DEV)
3412 return 0;
3413
3414 /* flush failed for some other reason, give it another shot */
3415 ata_tf_init(dev, &tf);
3416
3417 tf.command = qc->tf.command;
3418 tf.flags |= ATA_TFLAG_DEVICE;
3419 tf.protocol = ATA_PROT_NODATA;
3420
3421 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3422 tf.command, qc->err_mask);
3423
3424 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3425 if (!err_mask) {
3426 /*
3427 * FLUSH is complete but there's no way to
3428 * successfully complete a failed command from EH.
3429 * Making sure retry is allowed at least once and
3430 * retrying it should do the trick - whatever was in
3431 * the cache is already on the platter and this won't
3432 * cause infinite loop.
3433 */
3434 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3435 } else {
3436 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3437 err_mask);
3438 rc = -EIO;
3439
3440 /* if device failed it, report it to upper layers */
3441 if (err_mask & AC_ERR_DEV) {
3442 qc->err_mask |= AC_ERR_DEV;
3443 qc->result_tf = tf;
3444 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3445 rc = 0;
3446 }
3447 }
3448 return rc;
3449 }
3450
3451 /**
3452 * ata_eh_set_lpm - configure SATA interface power management
3453 * @link: link to configure power management
3454 * @policy: the link power management policy
3455 * @r_failed_dev: out parameter for failed device
3456 *
3457 * Enable SATA Interface power management. This will enable
3458 * Device Interface Power Management (DIPM) for min_power and
3459 * medium_power_with_dipm policies, and then call driver specific
3460 * callbacks for enabling Host Initiated Power management.
3461 *
3462 * LOCKING:
3463 * EH context.
3464 *
3465 * RETURNS:
3466 * 0 on success, -errno on failure.
3467 */
3468 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3469 struct ata_device **r_failed_dev)
3470 {
3471 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3472 struct ata_eh_context *ehc = &link->eh_context;
3473 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3474 enum ata_lpm_policy old_policy = link->lpm_policy;
3475 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3476 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3477 unsigned int err_mask;
3478 int rc;
3479
3480 /* if the link or host doesn't do LPM, noop */
3481 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3482 return 0;
3483
3484 /*
3485 * DIPM is enabled only for MIN_POWER as some devices
3486 * misbehave when the host NACKs transition to SLUMBER. Order
3487 * device and link configurations such that the host always
3488 * allows DIPM requests.
3489 */
3490 ata_for_each_dev(dev, link, ENABLED) {
3491 bool hipm = ata_id_has_hipm(dev->id);
3492 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3493
3494 /* find the first enabled and LPM enabled devices */
3495 if (!link_dev)
3496 link_dev = dev;
3497
3498 if (!lpm_dev && (hipm || dipm))
3499 lpm_dev = dev;
3500
3501 hints &= ~ATA_LPM_EMPTY;
3502 if (!hipm)
3503 hints &= ~ATA_LPM_HIPM;
3504
3505 /* disable DIPM before changing link config */
3506 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3507 err_mask = ata_dev_set_feature(dev,
3508 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3509 if (err_mask && err_mask != AC_ERR_DEV) {
3510 ata_dev_warn(dev,
3511 "failed to disable DIPM, Emask 0x%x\n",
3512 err_mask);
3513 rc = -EIO;
3514 goto fail;
3515 }
3516 }
3517 }
3518
3519 if (ap) {
3520 rc = ap->ops->set_lpm(link, policy, hints);
3521 if (!rc && ap->slave_link)
3522 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3523 } else
3524 rc = sata_pmp_set_lpm(link, policy, hints);
3525
3526 /*
3527 * Attribute link config failure to the first (LPM) enabled
3528 * device on the link.
3529 */
3530 if (rc) {
3531 if (rc == -EOPNOTSUPP) {
3532 link->flags |= ATA_LFLAG_NO_LPM;
3533 return 0;
3534 }
3535 dev = lpm_dev ? lpm_dev : link_dev;
3536 goto fail;
3537 }
3538
3539 /*
3540 * Low level driver acked the transition. Issue DIPM command
3541 * with the new policy set.
3542 */
3543 link->lpm_policy = policy;
3544 if (ap && ap->slave_link)
3545 ap->slave_link->lpm_policy = policy;
3546
3547 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3548 ata_for_each_dev(dev, link, ENABLED) {
3549 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3550 ata_id_has_dipm(dev->id)) {
3551 err_mask = ata_dev_set_feature(dev,
3552 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3553 if (err_mask && err_mask != AC_ERR_DEV) {
3554 ata_dev_warn(dev,
3555 "failed to enable DIPM, Emask 0x%x\n",
3556 err_mask);
3557 rc = -EIO;
3558 goto fail;
3559 }
3560 }
3561 }
3562
3563 link->last_lpm_change = jiffies;
3564 link->flags |= ATA_LFLAG_CHANGED;
3565
3566 return 0;
3567
3568 fail:
3569 /* restore the old policy */
3570 link->lpm_policy = old_policy;
3571 if (ap && ap->slave_link)
3572 ap->slave_link->lpm_policy = old_policy;
3573
3574 /* if no device or only one more chance is left, disable LPM */
3575 if (!dev || ehc->tries[dev->devno] <= 2) {
3576 ata_link_warn(link, "disabling LPM on the link\n");
3577 link->flags |= ATA_LFLAG_NO_LPM;
3578 }
3579 if (r_failed_dev)
3580 *r_failed_dev = dev;
3581 return rc;
3582 }
3583
3584 int ata_link_nr_enabled(struct ata_link *link)
3585 {
3586 struct ata_device *dev;
3587 int cnt = 0;
3588
3589 ata_for_each_dev(dev, link, ENABLED)
3590 cnt++;
3591 return cnt;
3592 }
3593
3594 static int ata_link_nr_vacant(struct ata_link *link)
3595 {
3596 struct ata_device *dev;
3597 int cnt = 0;
3598
3599 ata_for_each_dev(dev, link, ALL)
3600 if (dev->class == ATA_DEV_UNKNOWN)
3601 cnt++;
3602 return cnt;
3603 }
3604
3605 static int ata_eh_skip_recovery(struct ata_link *link)
3606 {
3607 struct ata_port *ap = link->ap;
3608 struct ata_eh_context *ehc = &link->eh_context;
3609 struct ata_device *dev;
3610
3611 /* skip disabled links */
3612 if (link->flags & ATA_LFLAG_DISABLED)
3613 return 1;
3614
3615 /* skip if explicitly requested */
3616 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3617 return 1;
3618
3619 /* thaw frozen port and recover failed devices */
3620 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3621 return 0;
3622
3623 /* reset at least once if reset is requested */
3624 if ((ehc->i.action & ATA_EH_RESET) &&
3625 !(ehc->i.flags & ATA_EHI_DID_RESET))
3626 return 0;
3627
3628 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3629 ata_for_each_dev(dev, link, ALL) {
3630 if (dev->class == ATA_DEV_UNKNOWN &&
3631 ehc->classes[dev->devno] != ATA_DEV_NONE)
3632 return 0;
3633 }
3634
3635 return 1;
3636 }
3637
3638 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3639 {
3640 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3641 u64 now = get_jiffies_64();
3642 int *trials = void_arg;
3643
3644 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3645 (ent->timestamp < now - min(now, interval)))
3646 return -1;
3647
3648 (*trials)++;
3649 return 0;
3650 }
3651
3652 static int ata_eh_schedule_probe(struct ata_device *dev)
3653 {
3654 struct ata_eh_context *ehc = &dev->link->eh_context;
3655 struct ata_link *link = ata_dev_phys_link(dev);
3656 int trials = 0;
3657
3658 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3659 (ehc->did_probe_mask & (1 << dev->devno)))
3660 return 0;
3661
3662 ata_eh_detach_dev(dev);
3663 ata_dev_init(dev);
3664 ehc->did_probe_mask |= (1 << dev->devno);
3665 ehc->i.action |= ATA_EH_RESET;
3666 ehc->saved_xfer_mode[dev->devno] = 0;
3667 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3668
3669 /* the link maybe in a deep sleep, wake it up */
3670 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3671 if (ata_is_host_link(link))
3672 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3673 ATA_LPM_EMPTY);
3674 else
3675 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3676 ATA_LPM_EMPTY);
3677 }
3678
3679 /* Record and count probe trials on the ering. The specific
3680 * error mask used is irrelevant. Because a successful device
3681 * detection clears the ering, this count accumulates only if
3682 * there are consecutive failed probes.
3683 *
3684 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3685 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3686 * forced to 1.5Gbps.
3687 *
3688 * This is to work around cases where failed link speed
3689 * negotiation results in device misdetection leading to
3690 * infinite DEVXCHG or PHRDY CHG events.
3691 */
3692 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3693 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3694
3695 if (trials > ATA_EH_PROBE_TRIALS)
3696 sata_down_spd_limit(link, 1);
3697
3698 return 1;
3699 }
3700
3701 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3702 {
3703 struct ata_eh_context *ehc = &dev->link->eh_context;
3704
3705 /* -EAGAIN from EH routine indicates retry without prejudice.
3706 * The requester is responsible for ensuring forward progress.
3707 */
3708 if (err != -EAGAIN)
3709 ehc->tries[dev->devno]--;
3710
3711 switch (err) {
3712 case -ENODEV:
3713 /* device missing or wrong IDENTIFY data, schedule probing */
3714 ehc->i.probe_mask |= (1 << dev->devno);
3715 /* fall through */
3716 case -EINVAL:
3717 /* give it just one more chance */
3718 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3719 /* fall through */
3720 case -EIO:
3721 if (ehc->tries[dev->devno] == 1) {
3722 /* This is the last chance, better to slow
3723 * down than lose it.
3724 */
3725 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3726 if (dev->pio_mode > XFER_PIO_0)
3727 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3728 }
3729 }
3730
3731 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3732 /* disable device if it has used up all its chances */
3733 ata_dev_disable(dev);
3734
3735 /* detach if offline */
3736 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3737 ata_eh_detach_dev(dev);
3738
3739 /* schedule probe if necessary */
3740 if (ata_eh_schedule_probe(dev)) {
3741 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3742 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3743 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3744 }
3745
3746 return 1;
3747 } else {
3748 ehc->i.action |= ATA_EH_RESET;
3749 return 0;
3750 }
3751 }
3752
3753 /**
3754 * ata_eh_recover - recover host port after error
3755 * @ap: host port to recover
3756 * @prereset: prereset method (can be NULL)
3757 * @softreset: softreset method (can be NULL)
3758 * @hardreset: hardreset method (can be NULL)
3759 * @postreset: postreset method (can be NULL)
3760 * @r_failed_link: out parameter for failed link
3761 *
3762 * This is the alpha and omega, eum and yang, heart and soul of
3763 * libata exception handling. On entry, actions required to
3764 * recover each link and hotplug requests are recorded in the
3765 * link's eh_context. This function executes all the operations
3766 * with appropriate retrials and fallbacks to resurrect failed
3767 * devices, detach goners and greet newcomers.
3768 *
3769 * LOCKING:
3770 * Kernel thread context (may sleep).
3771 *
3772 * RETURNS:
3773 * 0 on success, -errno on failure.
3774 */
3775 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3776 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3777 ata_postreset_fn_t postreset,
3778 struct ata_link **r_failed_link)
3779 {
3780 struct ata_link *link;
3781 struct ata_device *dev;
3782 int rc, nr_fails;
3783 unsigned long flags, deadline;
3784
3785 DPRINTK("ENTER\n");
3786
3787 /* prep for recovery */
3788 ata_for_each_link(link, ap, EDGE) {
3789 struct ata_eh_context *ehc = &link->eh_context;
3790
3791 /* re-enable link? */
3792 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3793 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3794 spin_lock_irqsave(ap->lock, flags);
3795 link->flags &= ~ATA_LFLAG_DISABLED;
3796 spin_unlock_irqrestore(ap->lock, flags);
3797 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3798 }
3799
3800 ata_for_each_dev(dev, link, ALL) {
3801 if (link->flags & ATA_LFLAG_NO_RETRY)
3802 ehc->tries[dev->devno] = 1;
3803 else
3804 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3805
3806 /* collect port action mask recorded in dev actions */
3807 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3808 ~ATA_EH_PERDEV_MASK;
3809 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3810
3811 /* process hotplug request */
3812 if (dev->flags & ATA_DFLAG_DETACH)
3813 ata_eh_detach_dev(dev);
3814
3815 /* schedule probe if necessary */
3816 if (!ata_dev_enabled(dev))
3817 ata_eh_schedule_probe(dev);
3818 }
3819 }
3820
3821 retry:
3822 rc = 0;
3823
3824 /* if UNLOADING, finish immediately */
3825 if (ap->pflags & ATA_PFLAG_UNLOADING)
3826 goto out;
3827
3828 /* prep for EH */
3829 ata_for_each_link(link, ap, EDGE) {
3830 struct ata_eh_context *ehc = &link->eh_context;
3831
3832 /* skip EH if possible. */
3833 if (ata_eh_skip_recovery(link))
3834 ehc->i.action = 0;
3835
3836 ata_for_each_dev(dev, link, ALL)
3837 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3838 }
3839
3840 /* reset */
3841 ata_for_each_link(link, ap, EDGE) {
3842 struct ata_eh_context *ehc = &link->eh_context;
3843
3844 if (!(ehc->i.action & ATA_EH_RESET))
3845 continue;
3846
3847 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3848 prereset, softreset, hardreset, postreset);
3849 if (rc) {
3850 ata_link_err(link, "reset failed, giving up\n");
3851 goto out;
3852 }
3853 }
3854
3855 do {
3856 unsigned long now;
3857
3858 /*
3859 * clears ATA_EH_PARK in eh_info and resets
3860 * ap->park_req_pending
3861 */
3862 ata_eh_pull_park_action(ap);
3863
3864 deadline = jiffies;
3865 ata_for_each_link(link, ap, EDGE) {
3866 ata_for_each_dev(dev, link, ALL) {
3867 struct ata_eh_context *ehc = &link->eh_context;
3868 unsigned long tmp;
3869
3870 if (dev->class != ATA_DEV_ATA &&
3871 dev->class != ATA_DEV_ZAC)
3872 continue;
3873 if (!(ehc->i.dev_action[dev->devno] &
3874 ATA_EH_PARK))
3875 continue;
3876 tmp = dev->unpark_deadline;
3877 if (time_before(deadline, tmp))
3878 deadline = tmp;
3879 else if (time_before_eq(tmp, jiffies))
3880 continue;
3881 if (ehc->unloaded_mask & (1 << dev->devno))
3882 continue;
3883
3884 ata_eh_park_issue_cmd(dev, 1);
3885 }
3886 }
3887
3888 now = jiffies;
3889 if (time_before_eq(deadline, now))
3890 break;
3891
3892 ata_eh_release(ap);
3893 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3894 deadline - now);
3895 ata_eh_acquire(ap);
3896 } while (deadline);
3897 ata_for_each_link(link, ap, EDGE) {
3898 ata_for_each_dev(dev, link, ALL) {
3899 if (!(link->eh_context.unloaded_mask &
3900 (1 << dev->devno)))
3901 continue;
3902
3903 ata_eh_park_issue_cmd(dev, 0);
3904 ata_eh_done(link, dev, ATA_EH_PARK);
3905 }
3906 }
3907
3908 /* the rest */
3909 nr_fails = 0;
3910 ata_for_each_link(link, ap, PMP_FIRST) {
3911 struct ata_eh_context *ehc = &link->eh_context;
3912
3913 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3914 goto config_lpm;
3915
3916 /* revalidate existing devices and attach new ones */
3917 rc = ata_eh_revalidate_and_attach(link, &dev);
3918 if (rc)
3919 goto rest_fail;
3920
3921 /* if PMP got attached, return, pmp EH will take care of it */
3922 if (link->device->class == ATA_DEV_PMP) {
3923 ehc->i.action = 0;
3924 return 0;
3925 }
3926
3927 /* configure transfer mode if necessary */
3928 if (ehc->i.flags & ATA_EHI_SETMODE) {
3929 rc = ata_set_mode(link, &dev);
3930 if (rc)
3931 goto rest_fail;
3932 ehc->i.flags &= ~ATA_EHI_SETMODE;
3933 }
3934
3935 /* If reset has been issued, clear UA to avoid
3936 * disrupting the current users of the device.
3937 */
3938 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3939 ata_for_each_dev(dev, link, ALL) {
3940 if (dev->class != ATA_DEV_ATAPI)
3941 continue;
3942 rc = atapi_eh_clear_ua(dev);
3943 if (rc)
3944 goto rest_fail;
3945 if (zpodd_dev_enabled(dev))
3946 zpodd_post_poweron(dev);
3947 }
3948 }
3949
3950 /* retry flush if necessary */
3951 ata_for_each_dev(dev, link, ALL) {
3952 if (dev->class != ATA_DEV_ATA &&
3953 dev->class != ATA_DEV_ZAC)
3954 continue;
3955 rc = ata_eh_maybe_retry_flush(dev);
3956 if (rc)
3957 goto rest_fail;
3958 }
3959
3960 config_lpm:
3961 /* configure link power saving */
3962 if (link->lpm_policy != ap->target_lpm_policy) {
3963 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3964 if (rc)
3965 goto rest_fail;
3966 }
3967
3968 /* this link is okay now */
3969 ehc->i.flags = 0;
3970 continue;
3971
3972 rest_fail:
3973 nr_fails++;
3974 if (dev)
3975 ata_eh_handle_dev_fail(dev, rc);
3976
3977 if (ap->pflags & ATA_PFLAG_FROZEN) {
3978 /* PMP reset requires working host port.
3979 * Can't retry if it's frozen.
3980 */
3981 if (sata_pmp_attached(ap))
3982 goto out;
3983 break;
3984 }
3985 }
3986
3987 if (nr_fails)
3988 goto retry;
3989
3990 out:
3991 if (rc && r_failed_link)
3992 *r_failed_link = link;
3993
3994 DPRINTK("EXIT, rc=%d\n", rc);
3995 return rc;
3996 }
3997
3998 /**
3999 * ata_eh_finish - finish up EH
4000 * @ap: host port to finish EH for
4001 *
4002 * Recovery is complete. Clean up EH states and retry or finish
4003 * failed qcs.
4004 *
4005 * LOCKING:
4006 * None.
4007 */
4008 void ata_eh_finish(struct ata_port *ap)
4009 {
4010 int tag;
4011
4012 /* retry or finish qcs */
4013 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
4014 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
4015
4016 if (!(qc->flags & ATA_QCFLAG_FAILED))
4017 continue;
4018
4019 if (qc->err_mask) {
4020 /* FIXME: Once EH migration is complete,
4021 * generate sense data in this function,
4022 * considering both err_mask and tf.
4023 */
4024 if (qc->flags & ATA_QCFLAG_RETRY)
4025 ata_eh_qc_retry(qc);
4026 else
4027 ata_eh_qc_complete(qc);
4028 } else {
4029 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
4030 ata_eh_qc_complete(qc);
4031 } else {
4032 /* feed zero TF to sense generation */
4033 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4034 ata_eh_qc_retry(qc);
4035 }
4036 }
4037 }
4038
4039 /* make sure nr_active_links is zero after EH */
4040 WARN_ON(ap->nr_active_links);
4041 ap->nr_active_links = 0;
4042 }
4043
4044 /**
4045 * ata_do_eh - do standard error handling
4046 * @ap: host port to handle error for
4047 *
4048 * @prereset: prereset method (can be NULL)
4049 * @softreset: softreset method (can be NULL)
4050 * @hardreset: hardreset method (can be NULL)
4051 * @postreset: postreset method (can be NULL)
4052 *
4053 * Perform standard error handling sequence.
4054 *
4055 * LOCKING:
4056 * Kernel thread context (may sleep).
4057 */
4058 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4059 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4060 ata_postreset_fn_t postreset)
4061 {
4062 struct ata_device *dev;
4063 int rc;
4064
4065 ata_eh_autopsy(ap);
4066 ata_eh_report(ap);
4067
4068 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4069 NULL);
4070 if (rc) {
4071 ata_for_each_dev(dev, &ap->link, ALL)
4072 ata_dev_disable(dev);
4073 }
4074
4075 ata_eh_finish(ap);
4076 }
4077
4078 /**
4079 * ata_std_error_handler - standard error handler
4080 * @ap: host port to handle error for
4081 *
4082 * Standard error handler
4083 *
4084 * LOCKING:
4085 * Kernel thread context (may sleep).
4086 */
4087 void ata_std_error_handler(struct ata_port *ap)
4088 {
4089 struct ata_port_operations *ops = ap->ops;
4090 ata_reset_fn_t hardreset = ops->hardreset;
4091
4092 /* ignore built-in hardreset if SCR access is not available */
4093 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4094 hardreset = NULL;
4095
4096 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4097 }
4098
4099 #ifdef CONFIG_PM
4100 /**
4101 * ata_eh_handle_port_suspend - perform port suspend operation
4102 * @ap: port to suspend
4103 *
4104 * Suspend @ap.
4105 *
4106 * LOCKING:
4107 * Kernel thread context (may sleep).
4108 */
4109 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4110 {
4111 unsigned long flags;
4112 int rc = 0;
4113 struct ata_device *dev;
4114
4115 /* are we suspending? */
4116 spin_lock_irqsave(ap->lock, flags);
4117 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4118 ap->pm_mesg.event & PM_EVENT_RESUME) {
4119 spin_unlock_irqrestore(ap->lock, flags);
4120 return;
4121 }
4122 spin_unlock_irqrestore(ap->lock, flags);
4123
4124 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4125
4126 /*
4127 * If we have a ZPODD attached, check its zero
4128 * power ready status before the port is frozen.
4129 * Only needed for runtime suspend.
4130 */
4131 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4132 ata_for_each_dev(dev, &ap->link, ENABLED) {
4133 if (zpodd_dev_enabled(dev))
4134 zpodd_on_suspend(dev);
4135 }
4136 }
4137
4138 /* tell ACPI we're suspending */
4139 rc = ata_acpi_on_suspend(ap);
4140 if (rc)
4141 goto out;
4142
4143 /* suspend */
4144 ata_eh_freeze_port(ap);
4145
4146 if (ap->ops->port_suspend)
4147 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4148
4149 ata_acpi_set_state(ap, ap->pm_mesg);
4150 out:
4151 /* update the flags */
4152 spin_lock_irqsave(ap->lock, flags);
4153
4154 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4155 if (rc == 0)
4156 ap->pflags |= ATA_PFLAG_SUSPENDED;
4157 else if (ap->pflags & ATA_PFLAG_FROZEN)
4158 ata_port_schedule_eh(ap);
4159
4160 spin_unlock_irqrestore(ap->lock, flags);
4161
4162 return;
4163 }
4164
4165 /**
4166 * ata_eh_handle_port_resume - perform port resume operation
4167 * @ap: port to resume
4168 *
4169 * Resume @ap.
4170 *
4171 * LOCKING:
4172 * Kernel thread context (may sleep).
4173 */
4174 static void ata_eh_handle_port_resume(struct ata_port *ap)
4175 {
4176 struct ata_link *link;
4177 struct ata_device *dev;
4178 unsigned long flags;
4179
4180 /* are we resuming? */
4181 spin_lock_irqsave(ap->lock, flags);
4182 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4183 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4184 spin_unlock_irqrestore(ap->lock, flags);
4185 return;
4186 }
4187 spin_unlock_irqrestore(ap->lock, flags);
4188
4189 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4190
4191 /*
4192 * Error timestamps are in jiffies which doesn't run while
4193 * suspended and PHY events during resume isn't too uncommon.
4194 * When the two are combined, it can lead to unnecessary speed
4195 * downs if the machine is suspended and resumed repeatedly.
4196 * Clear error history.
4197 */
4198 ata_for_each_link(link, ap, HOST_FIRST)
4199 ata_for_each_dev(dev, link, ALL)
4200 ata_ering_clear(&dev->ering);
4201
4202 ata_acpi_set_state(ap, ap->pm_mesg);
4203
4204 if (ap->ops->port_resume)
4205 ap->ops->port_resume(ap);
4206
4207 /* tell ACPI that we're resuming */
4208 ata_acpi_on_resume(ap);
4209
4210 /* update the flags */
4211 spin_lock_irqsave(ap->lock, flags);
4212 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4213 spin_unlock_irqrestore(ap->lock, flags);
4214 }
4215 #endif /* CONFIG_PM */