]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/ata/libata-eh.c
Merge remote-tracking branches 'regmap/topic/mmio', 'regmap/topic/rbtree' and 'regmap...
[mirror_ubuntu-artful-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/DocBook/libata.*
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
553 static void ata_eh_unload(struct ata_port *ap)
554 {
555 struct ata_link *link;
556 struct ata_device *dev;
557 unsigned long flags;
558
559 /* Restore SControl IPM and SPD for the next driver and
560 * disable attached devices.
561 */
562 ata_for_each_link(link, ap, PMP_FIRST) {
563 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
564 ata_for_each_dev(dev, link, ALL)
565 ata_dev_disable(dev);
566 }
567
568 /* freeze and set UNLOADED */
569 spin_lock_irqsave(ap->lock, flags);
570
571 ata_port_freeze(ap); /* won't be thawed */
572 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
573 ap->pflags |= ATA_PFLAG_UNLOADED;
574
575 spin_unlock_irqrestore(ap->lock, flags);
576 }
577
578 /**
579 * ata_scsi_error - SCSI layer error handler callback
580 * @host: SCSI host on which error occurred
581 *
582 * Handles SCSI-layer-thrown error events.
583 *
584 * LOCKING:
585 * Inherited from SCSI layer (none, can sleep)
586 *
587 * RETURNS:
588 * Zero.
589 */
590 void ata_scsi_error(struct Scsi_Host *host)
591 {
592 struct ata_port *ap = ata_shost_to_port(host);
593 unsigned long flags;
594 LIST_HEAD(eh_work_q);
595
596 DPRINTK("ENTER\n");
597
598 spin_lock_irqsave(host->host_lock, flags);
599 list_splice_init(&host->eh_cmd_q, &eh_work_q);
600 spin_unlock_irqrestore(host->host_lock, flags);
601
602 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
603
604 /* If we timed raced normal completion and there is nothing to
605 recover nr_timedout == 0 why exactly are we doing error recovery ? */
606 ata_scsi_port_error_handler(host, ap);
607
608 /* finish or retry handled scmd's and clean up */
609 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
610
611 DPRINTK("EXIT\n");
612 }
613
614 /**
615 * ata_scsi_cmd_error_handler - error callback for a list of commands
616 * @host: scsi host containing the port
617 * @ap: ATA port within the host
618 * @eh_work_q: list of commands to process
619 *
620 * process the given list of commands and return those finished to the
621 * ap->eh_done_q. This function is the first part of the libata error
622 * handler which processes a given list of failed commands.
623 */
624 void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
625 struct list_head *eh_work_q)
626 {
627 int i;
628 unsigned long flags;
629
630 /* make sure sff pio task is not running */
631 ata_sff_flush_pio_task(ap);
632
633 /* synchronize with host lock and sort out timeouts */
634
635 /* For new EH, all qcs are finished in one of three ways -
636 * normal completion, error completion, and SCSI timeout.
637 * Both completions can race against SCSI timeout. When normal
638 * completion wins, the qc never reaches EH. When error
639 * completion wins, the qc has ATA_QCFLAG_FAILED set.
640 *
641 * When SCSI timeout wins, things are a bit more complex.
642 * Normal or error completion can occur after the timeout but
643 * before this point. In such cases, both types of
644 * completions are honored. A scmd is determined to have
645 * timed out iff its associated qc is active and not failed.
646 */
647 if (ap->ops->error_handler) {
648 struct scsi_cmnd *scmd, *tmp;
649 int nr_timedout = 0;
650
651 spin_lock_irqsave(ap->lock, flags);
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 spin_unlock_irqrestore(ap->lock, flags);
703
704 /* initialize eh_tries */
705 ap->eh_tries = ATA_EH_MAX_TRIES;
706 } else
707 spin_unlock_wait(ap->lock);
708
709 }
710 EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
711
712 /**
713 * ata_scsi_port_error_handler - recover the port after the commands
714 * @host: SCSI host containing the port
715 * @ap: the ATA port
716 *
717 * Handle the recovery of the port @ap after all the commands
718 * have been recovered.
719 */
720 void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
721 {
722 unsigned long flags;
723
724 /* invoke error handler */
725 if (ap->ops->error_handler) {
726 struct ata_link *link;
727
728 /* acquire EH ownership */
729 ata_eh_acquire(ap);
730 repeat:
731 /* kill fast drain timer */
732 del_timer_sync(&ap->fastdrain_timer);
733
734 /* process port resume request */
735 ata_eh_handle_port_resume(ap);
736
737 /* fetch & clear EH info */
738 spin_lock_irqsave(ap->lock, flags);
739
740 ata_for_each_link(link, ap, HOST_FIRST) {
741 struct ata_eh_context *ehc = &link->eh_context;
742 struct ata_device *dev;
743
744 memset(&link->eh_context, 0, sizeof(link->eh_context));
745 link->eh_context.i = link->eh_info;
746 memset(&link->eh_info, 0, sizeof(link->eh_info));
747
748 ata_for_each_dev(dev, link, ENABLED) {
749 int devno = dev->devno;
750
751 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
752 if (ata_ncq_enabled(dev))
753 ehc->saved_ncq_enabled |= 1 << devno;
754 }
755 }
756
757 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
758 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
759 ap->excl_link = NULL; /* don't maintain exclusion over EH */
760
761 spin_unlock_irqrestore(ap->lock, flags);
762
763 /* invoke EH, skip if unloading or suspended */
764 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
765 ap->ops->error_handler(ap);
766 else {
767 /* if unloading, commence suicide */
768 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
769 !(ap->pflags & ATA_PFLAG_UNLOADED))
770 ata_eh_unload(ap);
771 ata_eh_finish(ap);
772 }
773
774 /* process port suspend request */
775 ata_eh_handle_port_suspend(ap);
776
777 /* Exception might have happened after ->error_handler
778 * recovered the port but before this point. Repeat
779 * EH in such case.
780 */
781 spin_lock_irqsave(ap->lock, flags);
782
783 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
784 if (--ap->eh_tries) {
785 spin_unlock_irqrestore(ap->lock, flags);
786 goto repeat;
787 }
788 ata_port_err(ap,
789 "EH pending after %d tries, giving up\n",
790 ATA_EH_MAX_TRIES);
791 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
792 }
793
794 /* this run is complete, make sure EH info is clear */
795 ata_for_each_link(link, ap, HOST_FIRST)
796 memset(&link->eh_info, 0, sizeof(link->eh_info));
797
798 /* end eh (clear host_eh_scheduled) while holding
799 * ap->lock such that if exception occurs after this
800 * point but before EH completion, SCSI midlayer will
801 * re-initiate EH.
802 */
803 ap->ops->end_eh(ap);
804
805 spin_unlock_irqrestore(ap->lock, flags);
806 ata_eh_release(ap);
807 } else {
808 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
809 ap->ops->eng_timeout(ap);
810 }
811
812 scsi_eh_flush_done_q(&ap->eh_done_q);
813
814 /* clean up */
815 spin_lock_irqsave(ap->lock, flags);
816
817 if (ap->pflags & ATA_PFLAG_LOADING)
818 ap->pflags &= ~ATA_PFLAG_LOADING;
819 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
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(unsigned long arg)
884 {
885 struct ata_port *ap = (void *)arg;
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 * @ap: target ATA port
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_read_log_page - read a specific log page
1491 * @dev: target device
1492 * @log: log to read
1493 * @page: page to read
1494 * @buf: buffer to store read page
1495 * @sectors: number of sectors to read
1496 *
1497 * Read log page using READ_LOG_EXT command.
1498 *
1499 * LOCKING:
1500 * Kernel thread context (may sleep).
1501 *
1502 * RETURNS:
1503 * 0 on success, AC_ERR_* mask otherwise.
1504 */
1505 unsigned int ata_read_log_page(struct ata_device *dev, u8 log,
1506 u8 page, void *buf, unsigned int sectors)
1507 {
1508 unsigned long ap_flags = dev->link->ap->flags;
1509 struct ata_taskfile tf;
1510 unsigned int err_mask;
1511 bool dma = false;
1512
1513 DPRINTK("read log page - log 0x%x, page 0x%x\n", log, page);
1514
1515 /*
1516 * Return error without actually issuing the command on controllers
1517 * which e.g. lockup on a read log page.
1518 */
1519 if (ap_flags & ATA_FLAG_NO_LOG_PAGE)
1520 return AC_ERR_DEV;
1521
1522 retry:
1523 ata_tf_init(dev, &tf);
1524 if (dev->dma_mode && ata_id_has_read_log_dma_ext(dev->id) &&
1525 !(dev->horkage & ATA_HORKAGE_NO_NCQ_LOG)) {
1526 tf.command = ATA_CMD_READ_LOG_DMA_EXT;
1527 tf.protocol = ATA_PROT_DMA;
1528 dma = true;
1529 } else {
1530 tf.command = ATA_CMD_READ_LOG_EXT;
1531 tf.protocol = ATA_PROT_PIO;
1532 dma = false;
1533 }
1534 tf.lbal = log;
1535 tf.lbam = page;
1536 tf.nsect = sectors;
1537 tf.hob_nsect = sectors >> 8;
1538 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1539
1540 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1541 buf, sectors * ATA_SECT_SIZE, 0);
1542
1543 if (err_mask && dma) {
1544 dev->horkage |= ATA_HORKAGE_NO_NCQ_LOG;
1545 ata_dev_warn(dev, "READ LOG DMA EXT failed, trying unqueued\n");
1546 goto retry;
1547 }
1548
1549 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1550 return err_mask;
1551 }
1552
1553 /**
1554 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1555 * @dev: Device to read log page 10h from
1556 * @tag: Resulting tag of the failed command
1557 * @tf: Resulting taskfile registers of the failed command
1558 *
1559 * Read log page 10h to obtain NCQ error details and clear error
1560 * condition.
1561 *
1562 * LOCKING:
1563 * Kernel thread context (may sleep).
1564 *
1565 * RETURNS:
1566 * 0 on success, -errno otherwise.
1567 */
1568 static int ata_eh_read_log_10h(struct ata_device *dev,
1569 int *tag, struct ata_taskfile *tf)
1570 {
1571 u8 *buf = dev->link->ap->sector_buf;
1572 unsigned int err_mask;
1573 u8 csum;
1574 int i;
1575
1576 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1577 if (err_mask)
1578 return -EIO;
1579
1580 csum = 0;
1581 for (i = 0; i < ATA_SECT_SIZE; i++)
1582 csum += buf[i];
1583 if (csum)
1584 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1585 csum);
1586
1587 if (buf[0] & 0x80)
1588 return -ENOENT;
1589
1590 *tag = buf[0] & 0x1f;
1591
1592 tf->command = buf[2];
1593 tf->feature = buf[3];
1594 tf->lbal = buf[4];
1595 tf->lbam = buf[5];
1596 tf->lbah = buf[6];
1597 tf->device = buf[7];
1598 tf->hob_lbal = buf[8];
1599 tf->hob_lbam = buf[9];
1600 tf->hob_lbah = buf[10];
1601 tf->nsect = buf[12];
1602 tf->hob_nsect = buf[13];
1603
1604 return 0;
1605 }
1606
1607 /**
1608 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1609 * @dev: target ATAPI device
1610 * @r_sense_key: out parameter for sense_key
1611 *
1612 * Perform ATAPI TEST_UNIT_READY.
1613 *
1614 * LOCKING:
1615 * EH context (may sleep).
1616 *
1617 * RETURNS:
1618 * 0 on success, AC_ERR_* mask on failure.
1619 */
1620 unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1621 {
1622 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1623 struct ata_taskfile tf;
1624 unsigned int err_mask;
1625
1626 ata_tf_init(dev, &tf);
1627
1628 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1629 tf.command = ATA_CMD_PACKET;
1630 tf.protocol = ATAPI_PROT_NODATA;
1631
1632 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1633 if (err_mask == AC_ERR_DEV)
1634 *r_sense_key = tf.feature >> 4;
1635 return err_mask;
1636 }
1637
1638 /**
1639 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1640 * @dev: device to perform REQUEST_SENSE to
1641 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1642 * @dfl_sense_key: default sense key to use
1643 *
1644 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1645 * SENSE. This function is EH helper.
1646 *
1647 * LOCKING:
1648 * Kernel thread context (may sleep).
1649 *
1650 * RETURNS:
1651 * 0 on success, AC_ERR_* mask on failure
1652 */
1653 unsigned int atapi_eh_request_sense(struct ata_device *dev,
1654 u8 *sense_buf, u8 dfl_sense_key)
1655 {
1656 u8 cdb[ATAPI_CDB_LEN] =
1657 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1658 struct ata_port *ap = dev->link->ap;
1659 struct ata_taskfile tf;
1660
1661 DPRINTK("ATAPI request sense\n");
1662
1663 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1664
1665 /* initialize sense_buf with the error register,
1666 * for the case where they are -not- overwritten
1667 */
1668 sense_buf[0] = 0x70;
1669 sense_buf[2] = dfl_sense_key;
1670
1671 /* some devices time out if garbage left in tf */
1672 ata_tf_init(dev, &tf);
1673
1674 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1675 tf.command = ATA_CMD_PACKET;
1676
1677 /* is it pointless to prefer PIO for "safety reasons"? */
1678 if (ap->flags & ATA_FLAG_PIO_DMA) {
1679 tf.protocol = ATAPI_PROT_DMA;
1680 tf.feature |= ATAPI_PKT_DMA;
1681 } else {
1682 tf.protocol = ATAPI_PROT_PIO;
1683 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1684 tf.lbah = 0;
1685 }
1686
1687 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1688 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1689 }
1690
1691 /**
1692 * ata_eh_analyze_serror - analyze SError for a failed port
1693 * @link: ATA link to analyze SError for
1694 *
1695 * Analyze SError if available and further determine cause of
1696 * failure.
1697 *
1698 * LOCKING:
1699 * None.
1700 */
1701 static void ata_eh_analyze_serror(struct ata_link *link)
1702 {
1703 struct ata_eh_context *ehc = &link->eh_context;
1704 u32 serror = ehc->i.serror;
1705 unsigned int err_mask = 0, action = 0;
1706 u32 hotplug_mask;
1707
1708 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1709 err_mask |= AC_ERR_ATA_BUS;
1710 action |= ATA_EH_RESET;
1711 }
1712 if (serror & SERR_PROTOCOL) {
1713 err_mask |= AC_ERR_HSM;
1714 action |= ATA_EH_RESET;
1715 }
1716 if (serror & SERR_INTERNAL) {
1717 err_mask |= AC_ERR_SYSTEM;
1718 action |= ATA_EH_RESET;
1719 }
1720
1721 /* Determine whether a hotplug event has occurred. Both
1722 * SError.N/X are considered hotplug events for enabled or
1723 * host links. For disabled PMP links, only N bit is
1724 * considered as X bit is left at 1 for link plugging.
1725 */
1726 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1727 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1728 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1729 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1730 else
1731 hotplug_mask = SERR_PHYRDY_CHG;
1732
1733 if (serror & hotplug_mask)
1734 ata_ehi_hotplugged(&ehc->i);
1735
1736 ehc->i.err_mask |= err_mask;
1737 ehc->i.action |= action;
1738 }
1739
1740 /**
1741 * ata_eh_analyze_ncq_error - analyze NCQ error
1742 * @link: ATA link to analyze NCQ error for
1743 *
1744 * Read log page 10h, determine the offending qc and acquire
1745 * error status TF. For NCQ device errors, all LLDDs have to do
1746 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1747 * care of the rest.
1748 *
1749 * LOCKING:
1750 * Kernel thread context (may sleep).
1751 */
1752 void ata_eh_analyze_ncq_error(struct ata_link *link)
1753 {
1754 struct ata_port *ap = link->ap;
1755 struct ata_eh_context *ehc = &link->eh_context;
1756 struct ata_device *dev = link->device;
1757 struct ata_queued_cmd *qc;
1758 struct ata_taskfile tf;
1759 int tag, rc;
1760
1761 /* if frozen, we can't do much */
1762 if (ap->pflags & ATA_PFLAG_FROZEN)
1763 return;
1764
1765 /* is it NCQ device error? */
1766 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1767 return;
1768
1769 /* has LLDD analyzed already? */
1770 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1771 qc = __ata_qc_from_tag(ap, tag);
1772
1773 if (!(qc->flags & ATA_QCFLAG_FAILED))
1774 continue;
1775
1776 if (qc->err_mask)
1777 return;
1778 }
1779
1780 /* okay, this error is ours */
1781 memset(&tf, 0, sizeof(tf));
1782 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1783 if (rc) {
1784 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1785 rc);
1786 return;
1787 }
1788
1789 if (!(link->sactive & (1 << tag))) {
1790 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1791 tag);
1792 return;
1793 }
1794
1795 /* we've got the perpetrator, condemn it */
1796 qc = __ata_qc_from_tag(ap, tag);
1797 memcpy(&qc->result_tf, &tf, sizeof(tf));
1798 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1799 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
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 else
1832 return 0;
1833
1834 switch (qc->dev->class) {
1835 case ATA_DEV_ATA:
1836 case ATA_DEV_ZAC:
1837 if (err & ATA_ICRC)
1838 qc->err_mask |= AC_ERR_ATA_BUS;
1839 if (err & (ATA_UNC | ATA_AMNF))
1840 qc->err_mask |= AC_ERR_MEDIA;
1841 if (err & ATA_IDNF)
1842 qc->err_mask |= AC_ERR_INVALID;
1843 break;
1844
1845 case ATA_DEV_ATAPI:
1846 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1847 tmp = atapi_eh_request_sense(qc->dev,
1848 qc->scsicmd->sense_buffer,
1849 qc->result_tf.feature >> 4);
1850 if (!tmp) {
1851 /* ATA_QCFLAG_SENSE_VALID is used to
1852 * tell atapi_qc_complete() that sense
1853 * data is already valid.
1854 *
1855 * TODO: interpret sense data and set
1856 * appropriate err_mask.
1857 */
1858 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1859 } else
1860 qc->err_mask |= tmp;
1861 }
1862 }
1863
1864 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1865 action |= ATA_EH_RESET;
1866
1867 return action;
1868 }
1869
1870 static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1871 int *xfer_ok)
1872 {
1873 int base = 0;
1874
1875 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1876 *xfer_ok = 1;
1877
1878 if (!*xfer_ok)
1879 base = ATA_ECAT_DUBIOUS_NONE;
1880
1881 if (err_mask & AC_ERR_ATA_BUS)
1882 return base + ATA_ECAT_ATA_BUS;
1883
1884 if (err_mask & AC_ERR_TIMEOUT)
1885 return base + ATA_ECAT_TOUT_HSM;
1886
1887 if (eflags & ATA_EFLAG_IS_IO) {
1888 if (err_mask & AC_ERR_HSM)
1889 return base + ATA_ECAT_TOUT_HSM;
1890 if ((err_mask &
1891 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1892 return base + ATA_ECAT_UNK_DEV;
1893 }
1894
1895 return 0;
1896 }
1897
1898 struct speed_down_verdict_arg {
1899 u64 since;
1900 int xfer_ok;
1901 int nr_errors[ATA_ECAT_NR];
1902 };
1903
1904 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1905 {
1906 struct speed_down_verdict_arg *arg = void_arg;
1907 int cat;
1908
1909 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1910 return -1;
1911
1912 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1913 &arg->xfer_ok);
1914 arg->nr_errors[cat]++;
1915
1916 return 0;
1917 }
1918
1919 /**
1920 * ata_eh_speed_down_verdict - Determine speed down verdict
1921 * @dev: Device of interest
1922 *
1923 * This function examines error ring of @dev and determines
1924 * whether NCQ needs to be turned off, transfer speed should be
1925 * stepped down, or falling back to PIO is necessary.
1926 *
1927 * ECAT_ATA_BUS : ATA_BUS error for any command
1928 *
1929 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1930 * IO commands
1931 *
1932 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1933 *
1934 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1935 * data transfer hasn't been verified.
1936 *
1937 * Verdicts are
1938 *
1939 * NCQ_OFF : Turn off NCQ.
1940 *
1941 * SPEED_DOWN : Speed down transfer speed but don't fall back
1942 * to PIO.
1943 *
1944 * FALLBACK_TO_PIO : Fall back to PIO.
1945 *
1946 * Even if multiple verdicts are returned, only one action is
1947 * taken per error. An action triggered by non-DUBIOUS errors
1948 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1949 * This is to expedite speed down decisions right after device is
1950 * initially configured.
1951 *
1952 * The followings are speed down rules. #1 and #2 deal with
1953 * DUBIOUS errors.
1954 *
1955 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1956 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1957 *
1958 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1959 * occurred during last 5 mins, NCQ_OFF.
1960 *
1961 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1962 * occurred during last 5 mins, FALLBACK_TO_PIO
1963 *
1964 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1965 * during last 10 mins, NCQ_OFF.
1966 *
1967 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1968 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1969 *
1970 * LOCKING:
1971 * Inherited from caller.
1972 *
1973 * RETURNS:
1974 * OR of ATA_EH_SPDN_* flags.
1975 */
1976 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1977 {
1978 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1979 u64 j64 = get_jiffies_64();
1980 struct speed_down_verdict_arg arg;
1981 unsigned int verdict = 0;
1982
1983 /* scan past 5 mins of error history */
1984 memset(&arg, 0, sizeof(arg));
1985 arg.since = j64 - min(j64, j5mins);
1986 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1987
1988 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1989 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1990 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1991 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1992
1993 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1994 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1995 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1996
1997 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1998 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1999 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2000 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
2001
2002 /* scan past 10 mins of error history */
2003 memset(&arg, 0, sizeof(arg));
2004 arg.since = j64 - min(j64, j10mins);
2005 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
2006
2007 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
2008 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
2009 verdict |= ATA_EH_SPDN_NCQ_OFF;
2010
2011 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
2012 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
2013 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
2014 verdict |= ATA_EH_SPDN_SPEED_DOWN;
2015
2016 return verdict;
2017 }
2018
2019 /**
2020 * ata_eh_speed_down - record error and speed down if necessary
2021 * @dev: Failed device
2022 * @eflags: mask of ATA_EFLAG_* flags
2023 * @err_mask: err_mask of the error
2024 *
2025 * Record error and examine error history to determine whether
2026 * adjusting transmission speed is necessary. It also sets
2027 * transmission limits appropriately if such adjustment is
2028 * necessary.
2029 *
2030 * LOCKING:
2031 * Kernel thread context (may sleep).
2032 *
2033 * RETURNS:
2034 * Determined recovery action.
2035 */
2036 static unsigned int ata_eh_speed_down(struct ata_device *dev,
2037 unsigned int eflags, unsigned int err_mask)
2038 {
2039 struct ata_link *link = ata_dev_phys_link(dev);
2040 int xfer_ok = 0;
2041 unsigned int verdict;
2042 unsigned int action = 0;
2043
2044 /* don't bother if Cat-0 error */
2045 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2046 return 0;
2047
2048 /* record error and determine whether speed down is necessary */
2049 ata_ering_record(&dev->ering, eflags, err_mask);
2050 verdict = ata_eh_speed_down_verdict(dev);
2051
2052 /* turn off NCQ? */
2053 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2054 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2055 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2056 dev->flags |= ATA_DFLAG_NCQ_OFF;
2057 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2058 goto done;
2059 }
2060
2061 /* speed down? */
2062 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2063 /* speed down SATA link speed if possible */
2064 if (sata_down_spd_limit(link, 0) == 0) {
2065 action |= ATA_EH_RESET;
2066 goto done;
2067 }
2068
2069 /* lower transfer mode */
2070 if (dev->spdn_cnt < 2) {
2071 static const int dma_dnxfer_sel[] =
2072 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2073 static const int pio_dnxfer_sel[] =
2074 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2075 int sel;
2076
2077 if (dev->xfer_shift != ATA_SHIFT_PIO)
2078 sel = dma_dnxfer_sel[dev->spdn_cnt];
2079 else
2080 sel = pio_dnxfer_sel[dev->spdn_cnt];
2081
2082 dev->spdn_cnt++;
2083
2084 if (ata_down_xfermask_limit(dev, sel) == 0) {
2085 action |= ATA_EH_RESET;
2086 goto done;
2087 }
2088 }
2089 }
2090
2091 /* Fall back to PIO? Slowing down to PIO is meaningless for
2092 * SATA ATA devices. Consider it only for PATA and SATAPI.
2093 */
2094 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2095 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2096 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2097 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2098 dev->spdn_cnt = 0;
2099 action |= ATA_EH_RESET;
2100 goto done;
2101 }
2102 }
2103
2104 return 0;
2105 done:
2106 /* device has been slowed down, blow error history */
2107 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2108 ata_ering_clear(&dev->ering);
2109 return action;
2110 }
2111
2112 /**
2113 * ata_eh_worth_retry - analyze error and decide whether to retry
2114 * @qc: qc to possibly retry
2115 *
2116 * Look at the cause of the error and decide if a retry
2117 * might be useful or not. We don't want to retry media errors
2118 * because the drive itself has probably already taken 10-30 seconds
2119 * doing its own internal retries before reporting the failure.
2120 */
2121 static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2122 {
2123 if (qc->err_mask & AC_ERR_MEDIA)
2124 return 0; /* don't retry media errors */
2125 if (qc->flags & ATA_QCFLAG_IO)
2126 return 1; /* otherwise retry anything from fs stack */
2127 if (qc->err_mask & AC_ERR_INVALID)
2128 return 0; /* don't retry these */
2129 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2130 }
2131
2132 /**
2133 * ata_eh_link_autopsy - analyze error and determine recovery action
2134 * @link: host link to perform autopsy on
2135 *
2136 * Analyze why @link failed and determine which recovery actions
2137 * are needed. This function also sets more detailed AC_ERR_*
2138 * values and fills sense data for ATAPI CHECK SENSE.
2139 *
2140 * LOCKING:
2141 * Kernel thread context (may sleep).
2142 */
2143 static void ata_eh_link_autopsy(struct ata_link *link)
2144 {
2145 struct ata_port *ap = link->ap;
2146 struct ata_eh_context *ehc = &link->eh_context;
2147 struct ata_device *dev;
2148 unsigned int all_err_mask = 0, eflags = 0;
2149 int tag;
2150 u32 serror;
2151 int rc;
2152
2153 DPRINTK("ENTER\n");
2154
2155 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2156 return;
2157
2158 /* obtain and analyze SError */
2159 rc = sata_scr_read(link, SCR_ERROR, &serror);
2160 if (rc == 0) {
2161 ehc->i.serror |= serror;
2162 ata_eh_analyze_serror(link);
2163 } else if (rc != -EOPNOTSUPP) {
2164 /* SError read failed, force reset and probing */
2165 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2166 ehc->i.action |= ATA_EH_RESET;
2167 ehc->i.err_mask |= AC_ERR_OTHER;
2168 }
2169
2170 /* analyze NCQ failure */
2171 ata_eh_analyze_ncq_error(link);
2172
2173 /* any real error trumps AC_ERR_OTHER */
2174 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2175 ehc->i.err_mask &= ~AC_ERR_OTHER;
2176
2177 all_err_mask |= ehc->i.err_mask;
2178
2179 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2180 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2181
2182 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2183 ata_dev_phys_link(qc->dev) != link)
2184 continue;
2185
2186 /* inherit upper level err_mask */
2187 qc->err_mask |= ehc->i.err_mask;
2188
2189 /* analyze TF */
2190 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2191
2192 /* DEV errors are probably spurious in case of ATA_BUS error */
2193 if (qc->err_mask & AC_ERR_ATA_BUS)
2194 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2195 AC_ERR_INVALID);
2196
2197 /* any real error trumps unknown error */
2198 if (qc->err_mask & ~AC_ERR_OTHER)
2199 qc->err_mask &= ~AC_ERR_OTHER;
2200
2201 /* SENSE_VALID trumps dev/unknown error and revalidation */
2202 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2203 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2204
2205 /* determine whether the command is worth retrying */
2206 if (ata_eh_worth_retry(qc))
2207 qc->flags |= ATA_QCFLAG_RETRY;
2208
2209 /* accumulate error info */
2210 ehc->i.dev = qc->dev;
2211 all_err_mask |= qc->err_mask;
2212 if (qc->flags & ATA_QCFLAG_IO)
2213 eflags |= ATA_EFLAG_IS_IO;
2214 trace_ata_eh_link_autopsy_qc(qc);
2215 }
2216
2217 /* enforce default EH actions */
2218 if (ap->pflags & ATA_PFLAG_FROZEN ||
2219 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2220 ehc->i.action |= ATA_EH_RESET;
2221 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2222 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2223 ehc->i.action |= ATA_EH_REVALIDATE;
2224
2225 /* If we have offending qcs and the associated failed device,
2226 * perform per-dev EH action only on the offending device.
2227 */
2228 if (ehc->i.dev) {
2229 ehc->i.dev_action[ehc->i.dev->devno] |=
2230 ehc->i.action & ATA_EH_PERDEV_MASK;
2231 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2232 }
2233
2234 /* propagate timeout to host link */
2235 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2236 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2237
2238 /* record error and consider speeding down */
2239 dev = ehc->i.dev;
2240 if (!dev && ((ata_link_max_devices(link) == 1 &&
2241 ata_dev_enabled(link->device))))
2242 dev = link->device;
2243
2244 if (dev) {
2245 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2246 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2247 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2248 }
2249 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2250 DPRINTK("EXIT\n");
2251 }
2252
2253 /**
2254 * ata_eh_autopsy - analyze error and determine recovery action
2255 * @ap: host port to perform autopsy on
2256 *
2257 * Analyze all links of @ap and determine why they failed and
2258 * which recovery actions are needed.
2259 *
2260 * LOCKING:
2261 * Kernel thread context (may sleep).
2262 */
2263 void ata_eh_autopsy(struct ata_port *ap)
2264 {
2265 struct ata_link *link;
2266
2267 ata_for_each_link(link, ap, EDGE)
2268 ata_eh_link_autopsy(link);
2269
2270 /* Handle the frigging slave link. Autopsy is done similarly
2271 * but actions and flags are transferred over to the master
2272 * link and handled from there.
2273 */
2274 if (ap->slave_link) {
2275 struct ata_eh_context *mehc = &ap->link.eh_context;
2276 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2277
2278 /* transfer control flags from master to slave */
2279 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2280
2281 /* perform autopsy on the slave link */
2282 ata_eh_link_autopsy(ap->slave_link);
2283
2284 /* transfer actions from slave to master and clear slave */
2285 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2286 mehc->i.action |= sehc->i.action;
2287 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2288 mehc->i.flags |= sehc->i.flags;
2289 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2290 }
2291
2292 /* Autopsy of fanout ports can affect host link autopsy.
2293 * Perform host link autopsy last.
2294 */
2295 if (sata_pmp_attached(ap))
2296 ata_eh_link_autopsy(&ap->link);
2297 }
2298
2299 /**
2300 * ata_get_cmd_descript - get description for ATA command
2301 * @command: ATA command code to get description for
2302 *
2303 * Return a textual description of the given command, or NULL if the
2304 * command is not known.
2305 *
2306 * LOCKING:
2307 * None
2308 */
2309 const char *ata_get_cmd_descript(u8 command)
2310 {
2311 #ifdef CONFIG_ATA_VERBOSE_ERROR
2312 static const struct
2313 {
2314 u8 command;
2315 const char *text;
2316 } cmd_descr[] = {
2317 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2318 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2319 { ATA_CMD_STANDBY, "STANDBY" },
2320 { ATA_CMD_IDLE, "IDLE" },
2321 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2322 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2323 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2324 { ATA_CMD_NOP, "NOP" },
2325 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2326 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2327 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2328 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2329 { ATA_CMD_SERVICE, "SERVICE" },
2330 { ATA_CMD_READ, "READ DMA" },
2331 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2332 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2333 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2334 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2335 { ATA_CMD_WRITE, "WRITE DMA" },
2336 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2337 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2338 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2339 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2340 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2341 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2342 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2343 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2344 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2345 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2346 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2347 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2348 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2349 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2350 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2351 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2352 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2353 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2354 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2355 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2356 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2357 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2358 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2359 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2360 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2361 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2362 { ATA_CMD_SLEEP, "SLEEP" },
2363 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2364 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2365 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2366 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2367 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2368 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2369 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2370 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2371 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2372 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2373 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2374 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2375 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2376 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2377 { ATA_CMD_PMP_READ, "READ BUFFER" },
2378 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2379 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2380 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2381 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2382 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2383 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2384 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2385 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2386 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2387 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2388 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2389 { ATA_CMD_SMART, "SMART" },
2390 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2391 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2392 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2393 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2394 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2395 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2396 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2397 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2398 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2399 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2400 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2401 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2402 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2403 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2404 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2405 { ATA_CMD_RESTORE, "RECALIBRATE" },
2406 { 0, NULL } /* terminate list */
2407 };
2408
2409 unsigned int i;
2410 for (i = 0; cmd_descr[i].text; i++)
2411 if (cmd_descr[i].command == command)
2412 return cmd_descr[i].text;
2413 #endif
2414
2415 return NULL;
2416 }
2417 EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2418
2419 /**
2420 * ata_eh_link_report - report error handling to user
2421 * @link: ATA link EH is going on
2422 *
2423 * Report EH to user.
2424 *
2425 * LOCKING:
2426 * None.
2427 */
2428 static void ata_eh_link_report(struct ata_link *link)
2429 {
2430 struct ata_port *ap = link->ap;
2431 struct ata_eh_context *ehc = &link->eh_context;
2432 const char *frozen, *desc;
2433 char tries_buf[6] = "";
2434 int tag, nr_failed = 0;
2435
2436 if (ehc->i.flags & ATA_EHI_QUIET)
2437 return;
2438
2439 desc = NULL;
2440 if (ehc->i.desc[0] != '\0')
2441 desc = ehc->i.desc;
2442
2443 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2444 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2445
2446 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2447 ata_dev_phys_link(qc->dev) != link ||
2448 ((qc->flags & ATA_QCFLAG_QUIET) &&
2449 qc->err_mask == AC_ERR_DEV))
2450 continue;
2451 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2452 continue;
2453
2454 nr_failed++;
2455 }
2456
2457 if (!nr_failed && !ehc->i.err_mask)
2458 return;
2459
2460 frozen = "";
2461 if (ap->pflags & ATA_PFLAG_FROZEN)
2462 frozen = " frozen";
2463
2464 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2465 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2466 ap->eh_tries);
2467
2468 if (ehc->i.dev) {
2469 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2470 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2471 ehc->i.err_mask, link->sactive, ehc->i.serror,
2472 ehc->i.action, frozen, tries_buf);
2473 if (desc)
2474 ata_dev_err(ehc->i.dev, "%s\n", desc);
2475 } else {
2476 ata_link_err(link, "exception Emask 0x%x "
2477 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2478 ehc->i.err_mask, link->sactive, ehc->i.serror,
2479 ehc->i.action, frozen, tries_buf);
2480 if (desc)
2481 ata_link_err(link, "%s\n", desc);
2482 }
2483
2484 #ifdef CONFIG_ATA_VERBOSE_ERROR
2485 if (ehc->i.serror)
2486 ata_link_err(link,
2487 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2488 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2489 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2490 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2491 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2492 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2493 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2494 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2495 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2496 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2497 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2498 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2499 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2500 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2501 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2502 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2503 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2504 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2505 #endif
2506
2507 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2508 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2509 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2510 char data_buf[20] = "";
2511 char cdb_buf[70] = "";
2512
2513 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2514 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2515 continue;
2516
2517 if (qc->dma_dir != DMA_NONE) {
2518 static const char *dma_str[] = {
2519 [DMA_BIDIRECTIONAL] = "bidi",
2520 [DMA_TO_DEVICE] = "out",
2521 [DMA_FROM_DEVICE] = "in",
2522 };
2523 static const char *prot_str[] = {
2524 [ATA_PROT_PIO] = "pio",
2525 [ATA_PROT_DMA] = "dma",
2526 [ATA_PROT_NCQ] = "ncq",
2527 [ATAPI_PROT_PIO] = "pio",
2528 [ATAPI_PROT_DMA] = "dma",
2529 };
2530
2531 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2532 prot_str[qc->tf.protocol], qc->nbytes,
2533 dma_str[qc->dma_dir]);
2534 }
2535
2536 if (ata_is_atapi(qc->tf.protocol)) {
2537 const u8 *cdb = qc->cdb;
2538 size_t cdb_len = qc->dev->cdb_len;
2539
2540 if (qc->scsicmd) {
2541 cdb = qc->scsicmd->cmnd;
2542 cdb_len = qc->scsicmd->cmd_len;
2543 }
2544 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2545 cdb, cdb_len);
2546 } else {
2547 const char *descr = ata_get_cmd_descript(cmd->command);
2548 if (descr)
2549 ata_dev_err(qc->dev, "failed command: %s\n",
2550 descr);
2551 }
2552
2553 ata_dev_err(qc->dev,
2554 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2555 "tag %d%s\n %s"
2556 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2557 "Emask 0x%x (%s)%s\n",
2558 cmd->command, cmd->feature, cmd->nsect,
2559 cmd->lbal, cmd->lbam, cmd->lbah,
2560 cmd->hob_feature, cmd->hob_nsect,
2561 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2562 cmd->device, qc->tag, data_buf, cdb_buf,
2563 res->command, res->feature, res->nsect,
2564 res->lbal, res->lbam, res->lbah,
2565 res->hob_feature, res->hob_nsect,
2566 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2567 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2568 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2569
2570 #ifdef CONFIG_ATA_VERBOSE_ERROR
2571 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2572 ATA_ERR)) {
2573 if (res->command & ATA_BUSY)
2574 ata_dev_err(qc->dev, "status: { Busy }\n");
2575 else
2576 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
2577 res->command & ATA_DRDY ? "DRDY " : "",
2578 res->command & ATA_DF ? "DF " : "",
2579 res->command & ATA_DRQ ? "DRQ " : "",
2580 res->command & ATA_ERR ? "ERR " : "");
2581 }
2582
2583 if (cmd->command != ATA_CMD_PACKET &&
2584 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2585 ATA_IDNF | ATA_ABORTED)))
2586 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2587 res->feature & ATA_ICRC ? "ICRC " : "",
2588 res->feature & ATA_UNC ? "UNC " : "",
2589 res->feature & ATA_AMNF ? "AMNF " : "",
2590 res->feature & ATA_IDNF ? "IDNF " : "",
2591 res->feature & ATA_ABORTED ? "ABRT " : "");
2592 #endif
2593 }
2594 }
2595
2596 /**
2597 * ata_eh_report - report error handling to user
2598 * @ap: ATA port to report EH about
2599 *
2600 * Report EH to user.
2601 *
2602 * LOCKING:
2603 * None.
2604 */
2605 void ata_eh_report(struct ata_port *ap)
2606 {
2607 struct ata_link *link;
2608
2609 ata_for_each_link(link, ap, HOST_FIRST)
2610 ata_eh_link_report(link);
2611 }
2612
2613 static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2614 unsigned int *classes, unsigned long deadline,
2615 bool clear_classes)
2616 {
2617 struct ata_device *dev;
2618
2619 if (clear_classes)
2620 ata_for_each_dev(dev, link, ALL)
2621 classes[dev->devno] = ATA_DEV_UNKNOWN;
2622
2623 return reset(link, classes, deadline);
2624 }
2625
2626 static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2627 {
2628 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2629 return 0;
2630 if (rc == -EAGAIN)
2631 return 1;
2632 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2633 return 1;
2634 return 0;
2635 }
2636
2637 int ata_eh_reset(struct ata_link *link, int classify,
2638 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2639 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2640 {
2641 struct ata_port *ap = link->ap;
2642 struct ata_link *slave = ap->slave_link;
2643 struct ata_eh_context *ehc = &link->eh_context;
2644 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2645 unsigned int *classes = ehc->classes;
2646 unsigned int lflags = link->flags;
2647 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2648 int max_tries = 0, try = 0;
2649 struct ata_link *failed_link;
2650 struct ata_device *dev;
2651 unsigned long deadline, now;
2652 ata_reset_fn_t reset;
2653 unsigned long flags;
2654 u32 sstatus;
2655 int nr_unknown, rc;
2656
2657 /*
2658 * Prepare to reset
2659 */
2660 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2661 max_tries++;
2662 if (link->flags & ATA_LFLAG_RST_ONCE)
2663 max_tries = 1;
2664 if (link->flags & ATA_LFLAG_NO_HRST)
2665 hardreset = NULL;
2666 if (link->flags & ATA_LFLAG_NO_SRST)
2667 softreset = NULL;
2668
2669 /* make sure each reset attempt is at least COOL_DOWN apart */
2670 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2671 now = jiffies;
2672 WARN_ON(time_after(ehc->last_reset, now));
2673 deadline = ata_deadline(ehc->last_reset,
2674 ATA_EH_RESET_COOL_DOWN);
2675 if (time_before(now, deadline))
2676 schedule_timeout_uninterruptible(deadline - now);
2677 }
2678
2679 spin_lock_irqsave(ap->lock, flags);
2680 ap->pflags |= ATA_PFLAG_RESETTING;
2681 spin_unlock_irqrestore(ap->lock, flags);
2682
2683 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2684
2685 ata_for_each_dev(dev, link, ALL) {
2686 /* If we issue an SRST then an ATA drive (not ATAPI)
2687 * may change configuration and be in PIO0 timing. If
2688 * we do a hard reset (or are coming from power on)
2689 * this is true for ATA or ATAPI. Until we've set a
2690 * suitable controller mode we should not touch the
2691 * bus as we may be talking too fast.
2692 */
2693 dev->pio_mode = XFER_PIO_0;
2694 dev->dma_mode = 0xff;
2695
2696 /* If the controller has a pio mode setup function
2697 * then use it to set the chipset to rights. Don't
2698 * touch the DMA setup as that will be dealt with when
2699 * configuring devices.
2700 */
2701 if (ap->ops->set_piomode)
2702 ap->ops->set_piomode(ap, dev);
2703 }
2704
2705 /* prefer hardreset */
2706 reset = NULL;
2707 ehc->i.action &= ~ATA_EH_RESET;
2708 if (hardreset) {
2709 reset = hardreset;
2710 ehc->i.action |= ATA_EH_HARDRESET;
2711 } else if (softreset) {
2712 reset = softreset;
2713 ehc->i.action |= ATA_EH_SOFTRESET;
2714 }
2715
2716 if (prereset) {
2717 unsigned long deadline = ata_deadline(jiffies,
2718 ATA_EH_PRERESET_TIMEOUT);
2719
2720 if (slave) {
2721 sehc->i.action &= ~ATA_EH_RESET;
2722 sehc->i.action |= ehc->i.action;
2723 }
2724
2725 rc = prereset(link, deadline);
2726
2727 /* If present, do prereset on slave link too. Reset
2728 * is skipped iff both master and slave links report
2729 * -ENOENT or clear ATA_EH_RESET.
2730 */
2731 if (slave && (rc == 0 || rc == -ENOENT)) {
2732 int tmp;
2733
2734 tmp = prereset(slave, deadline);
2735 if (tmp != -ENOENT)
2736 rc = tmp;
2737
2738 ehc->i.action |= sehc->i.action;
2739 }
2740
2741 if (rc) {
2742 if (rc == -ENOENT) {
2743 ata_link_dbg(link, "port disabled--ignoring\n");
2744 ehc->i.action &= ~ATA_EH_RESET;
2745
2746 ata_for_each_dev(dev, link, ALL)
2747 classes[dev->devno] = ATA_DEV_NONE;
2748
2749 rc = 0;
2750 } else
2751 ata_link_err(link,
2752 "prereset failed (errno=%d)\n",
2753 rc);
2754 goto out;
2755 }
2756
2757 /* prereset() might have cleared ATA_EH_RESET. If so,
2758 * bang classes, thaw and return.
2759 */
2760 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2761 ata_for_each_dev(dev, link, ALL)
2762 classes[dev->devno] = ATA_DEV_NONE;
2763 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2764 ata_is_host_link(link))
2765 ata_eh_thaw_port(ap);
2766 rc = 0;
2767 goto out;
2768 }
2769 }
2770
2771 retry:
2772 /*
2773 * Perform reset
2774 */
2775 if (ata_is_host_link(link))
2776 ata_eh_freeze_port(ap);
2777
2778 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2779
2780 if (reset) {
2781 if (verbose)
2782 ata_link_info(link, "%s resetting link\n",
2783 reset == softreset ? "soft" : "hard");
2784
2785 /* mark that this EH session started with reset */
2786 ehc->last_reset = jiffies;
2787 if (reset == hardreset)
2788 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2789 else
2790 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2791
2792 rc = ata_do_reset(link, reset, classes, deadline, true);
2793 if (rc && rc != -EAGAIN) {
2794 failed_link = link;
2795 goto fail;
2796 }
2797
2798 /* hardreset slave link if existent */
2799 if (slave && reset == hardreset) {
2800 int tmp;
2801
2802 if (verbose)
2803 ata_link_info(slave, "hard resetting link\n");
2804
2805 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2806 tmp = ata_do_reset(slave, reset, classes, deadline,
2807 false);
2808 switch (tmp) {
2809 case -EAGAIN:
2810 rc = -EAGAIN;
2811 case 0:
2812 break;
2813 default:
2814 failed_link = slave;
2815 rc = tmp;
2816 goto fail;
2817 }
2818 }
2819
2820 /* perform follow-up SRST if necessary */
2821 if (reset == hardreset &&
2822 ata_eh_followup_srst_needed(link, rc)) {
2823 reset = softreset;
2824
2825 if (!reset) {
2826 ata_link_err(link,
2827 "follow-up softreset required but no softreset available\n");
2828 failed_link = link;
2829 rc = -EINVAL;
2830 goto fail;
2831 }
2832
2833 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2834 rc = ata_do_reset(link, reset, classes, deadline, true);
2835 if (rc) {
2836 failed_link = link;
2837 goto fail;
2838 }
2839 }
2840 } else {
2841 if (verbose)
2842 ata_link_info(link,
2843 "no reset method available, skipping reset\n");
2844 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2845 lflags |= ATA_LFLAG_ASSUME_ATA;
2846 }
2847
2848 /*
2849 * Post-reset processing
2850 */
2851 ata_for_each_dev(dev, link, ALL) {
2852 /* After the reset, the device state is PIO 0 and the
2853 * controller state is undefined. Reset also wakes up
2854 * drives from sleeping mode.
2855 */
2856 dev->pio_mode = XFER_PIO_0;
2857 dev->flags &= ~ATA_DFLAG_SLEEPING;
2858
2859 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2860 continue;
2861
2862 /* apply class override */
2863 if (lflags & ATA_LFLAG_ASSUME_ATA)
2864 classes[dev->devno] = ATA_DEV_ATA;
2865 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2866 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2867 }
2868
2869 /* record current link speed */
2870 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2871 link->sata_spd = (sstatus >> 4) & 0xf;
2872 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2873 slave->sata_spd = (sstatus >> 4) & 0xf;
2874
2875 /* thaw the port */
2876 if (ata_is_host_link(link))
2877 ata_eh_thaw_port(ap);
2878
2879 /* postreset() should clear hardware SError. Although SError
2880 * is cleared during link resume, clearing SError here is
2881 * necessary as some PHYs raise hotplug events after SRST.
2882 * This introduces race condition where hotplug occurs between
2883 * reset and here. This race is mediated by cross checking
2884 * link onlineness and classification result later.
2885 */
2886 if (postreset) {
2887 postreset(link, classes);
2888 if (slave)
2889 postreset(slave, classes);
2890 }
2891
2892 /*
2893 * Some controllers can't be frozen very well and may set spurious
2894 * error conditions during reset. Clear accumulated error
2895 * information and re-thaw the port if frozen. As reset is the
2896 * final recovery action and we cross check link onlineness against
2897 * device classification later, no hotplug event is lost by this.
2898 */
2899 spin_lock_irqsave(link->ap->lock, flags);
2900 memset(&link->eh_info, 0, sizeof(link->eh_info));
2901 if (slave)
2902 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2903 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2904 spin_unlock_irqrestore(link->ap->lock, flags);
2905
2906 if (ap->pflags & ATA_PFLAG_FROZEN)
2907 ata_eh_thaw_port(ap);
2908
2909 /*
2910 * Make sure onlineness and classification result correspond.
2911 * Hotplug could have happened during reset and some
2912 * controllers fail to wait while a drive is spinning up after
2913 * being hotplugged causing misdetection. By cross checking
2914 * link on/offlineness and classification result, those
2915 * conditions can be reliably detected and retried.
2916 */
2917 nr_unknown = 0;
2918 ata_for_each_dev(dev, link, ALL) {
2919 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2920 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2921 ata_dev_dbg(dev, "link online but device misclassified\n");
2922 classes[dev->devno] = ATA_DEV_NONE;
2923 nr_unknown++;
2924 }
2925 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2926 if (ata_class_enabled(classes[dev->devno]))
2927 ata_dev_dbg(dev,
2928 "link offline, clearing class %d to NONE\n",
2929 classes[dev->devno]);
2930 classes[dev->devno] = ATA_DEV_NONE;
2931 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2932 ata_dev_dbg(dev,
2933 "link status unknown, clearing UNKNOWN to NONE\n");
2934 classes[dev->devno] = ATA_DEV_NONE;
2935 }
2936 }
2937
2938 if (classify && nr_unknown) {
2939 if (try < max_tries) {
2940 ata_link_warn(link,
2941 "link online but %d devices misclassified, retrying\n",
2942 nr_unknown);
2943 failed_link = link;
2944 rc = -EAGAIN;
2945 goto fail;
2946 }
2947 ata_link_warn(link,
2948 "link online but %d devices misclassified, "
2949 "device detection might fail\n", nr_unknown);
2950 }
2951
2952 /* reset successful, schedule revalidation */
2953 ata_eh_done(link, NULL, ATA_EH_RESET);
2954 if (slave)
2955 ata_eh_done(slave, NULL, ATA_EH_RESET);
2956 ehc->last_reset = jiffies; /* update to completion time */
2957 ehc->i.action |= ATA_EH_REVALIDATE;
2958 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2959
2960 rc = 0;
2961 out:
2962 /* clear hotplug flag */
2963 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2964 if (slave)
2965 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2966
2967 spin_lock_irqsave(ap->lock, flags);
2968 ap->pflags &= ~ATA_PFLAG_RESETTING;
2969 spin_unlock_irqrestore(ap->lock, flags);
2970
2971 return rc;
2972
2973 fail:
2974 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2975 if (!ata_is_host_link(link) &&
2976 sata_scr_read(link, SCR_STATUS, &sstatus))
2977 rc = -ERESTART;
2978
2979 if (try >= max_tries) {
2980 /*
2981 * Thaw host port even if reset failed, so that the port
2982 * can be retried on the next phy event. This risks
2983 * repeated EH runs but seems to be a better tradeoff than
2984 * shutting down a port after a botched hotplug attempt.
2985 */
2986 if (ata_is_host_link(link))
2987 ata_eh_thaw_port(ap);
2988 goto out;
2989 }
2990
2991 now = jiffies;
2992 if (time_before(now, deadline)) {
2993 unsigned long delta = deadline - now;
2994
2995 ata_link_warn(failed_link,
2996 "reset failed (errno=%d), retrying in %u secs\n",
2997 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2998
2999 ata_eh_release(ap);
3000 while (delta)
3001 delta = schedule_timeout_uninterruptible(delta);
3002 ata_eh_acquire(ap);
3003 }
3004
3005 /*
3006 * While disks spinup behind PMP, some controllers fail sending SRST.
3007 * They need to be reset - as well as the PMP - before retrying.
3008 */
3009 if (rc == -ERESTART) {
3010 if (ata_is_host_link(link))
3011 ata_eh_thaw_port(ap);
3012 goto out;
3013 }
3014
3015 if (try == max_tries - 1) {
3016 sata_down_spd_limit(link, 0);
3017 if (slave)
3018 sata_down_spd_limit(slave, 0);
3019 } else if (rc == -EPIPE)
3020 sata_down_spd_limit(failed_link, 0);
3021
3022 if (hardreset)
3023 reset = hardreset;
3024 goto retry;
3025 }
3026
3027 static inline void ata_eh_pull_park_action(struct ata_port *ap)
3028 {
3029 struct ata_link *link;
3030 struct ata_device *dev;
3031 unsigned long flags;
3032
3033 /*
3034 * This function can be thought of as an extended version of
3035 * ata_eh_about_to_do() specially crafted to accommodate the
3036 * requirements of ATA_EH_PARK handling. Since the EH thread
3037 * does not leave the do {} while () loop in ata_eh_recover as
3038 * long as the timeout for a park request to *one* device on
3039 * the port has not expired, and since we still want to pick
3040 * up park requests to other devices on the same port or
3041 * timeout updates for the same device, we have to pull
3042 * ATA_EH_PARK actions from eh_info into eh_context.i
3043 * ourselves at the beginning of each pass over the loop.
3044 *
3045 * Additionally, all write accesses to &ap->park_req_pending
3046 * through reinit_completion() (see below) or complete_all()
3047 * (see ata_scsi_park_store()) are protected by the host lock.
3048 * As a result we have that park_req_pending.done is zero on
3049 * exit from this function, i.e. when ATA_EH_PARK actions for
3050 * *all* devices on port ap have been pulled into the
3051 * respective eh_context structs. If, and only if,
3052 * park_req_pending.done is non-zero by the time we reach
3053 * wait_for_completion_timeout(), another ATA_EH_PARK action
3054 * has been scheduled for at least one of the devices on port
3055 * ap and we have to cycle over the do {} while () loop in
3056 * ata_eh_recover() again.
3057 */
3058
3059 spin_lock_irqsave(ap->lock, flags);
3060 reinit_completion(&ap->park_req_pending);
3061 ata_for_each_link(link, ap, EDGE) {
3062 ata_for_each_dev(dev, link, ALL) {
3063 struct ata_eh_info *ehi = &link->eh_info;
3064
3065 link->eh_context.i.dev_action[dev->devno] |=
3066 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3067 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3068 }
3069 }
3070 spin_unlock_irqrestore(ap->lock, flags);
3071 }
3072
3073 static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3074 {
3075 struct ata_eh_context *ehc = &dev->link->eh_context;
3076 struct ata_taskfile tf;
3077 unsigned int err_mask;
3078
3079 ata_tf_init(dev, &tf);
3080 if (park) {
3081 ehc->unloaded_mask |= 1 << dev->devno;
3082 tf.command = ATA_CMD_IDLEIMMEDIATE;
3083 tf.feature = 0x44;
3084 tf.lbal = 0x4c;
3085 tf.lbam = 0x4e;
3086 tf.lbah = 0x55;
3087 } else {
3088 ehc->unloaded_mask &= ~(1 << dev->devno);
3089 tf.command = ATA_CMD_CHK_POWER;
3090 }
3091
3092 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3093 tf.protocol |= ATA_PROT_NODATA;
3094 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3095 if (park && (err_mask || tf.lbal != 0xc4)) {
3096 ata_dev_err(dev, "head unload failed!\n");
3097 ehc->unloaded_mask &= ~(1 << dev->devno);
3098 }
3099 }
3100
3101 static int ata_eh_revalidate_and_attach(struct ata_link *link,
3102 struct ata_device **r_failed_dev)
3103 {
3104 struct ata_port *ap = link->ap;
3105 struct ata_eh_context *ehc = &link->eh_context;
3106 struct ata_device *dev;
3107 unsigned int new_mask = 0;
3108 unsigned long flags;
3109 int rc = 0;
3110
3111 DPRINTK("ENTER\n");
3112
3113 /* For PATA drive side cable detection to work, IDENTIFY must
3114 * be done backwards such that PDIAG- is released by the slave
3115 * device before the master device is identified.
3116 */
3117 ata_for_each_dev(dev, link, ALL_REVERSE) {
3118 unsigned int action = ata_eh_dev_action(dev);
3119 unsigned int readid_flags = 0;
3120
3121 if (ehc->i.flags & ATA_EHI_DID_RESET)
3122 readid_flags |= ATA_READID_POSTRESET;
3123
3124 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3125 WARN_ON(dev->class == ATA_DEV_PMP);
3126
3127 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3128 rc = -EIO;
3129 goto err;
3130 }
3131
3132 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3133 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3134 readid_flags);
3135 if (rc)
3136 goto err;
3137
3138 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3139
3140 /* Configuration may have changed, reconfigure
3141 * transfer mode.
3142 */
3143 ehc->i.flags |= ATA_EHI_SETMODE;
3144
3145 /* schedule the scsi_rescan_device() here */
3146 schedule_work(&(ap->scsi_rescan_task));
3147 } else if (dev->class == ATA_DEV_UNKNOWN &&
3148 ehc->tries[dev->devno] &&
3149 ata_class_enabled(ehc->classes[dev->devno])) {
3150 /* Temporarily set dev->class, it will be
3151 * permanently set once all configurations are
3152 * complete. This is necessary because new
3153 * device configuration is done in two
3154 * separate loops.
3155 */
3156 dev->class = ehc->classes[dev->devno];
3157
3158 if (dev->class == ATA_DEV_PMP)
3159 rc = sata_pmp_attach(dev);
3160 else
3161 rc = ata_dev_read_id(dev, &dev->class,
3162 readid_flags, dev->id);
3163
3164 /* read_id might have changed class, store and reset */
3165 ehc->classes[dev->devno] = dev->class;
3166 dev->class = ATA_DEV_UNKNOWN;
3167
3168 switch (rc) {
3169 case 0:
3170 /* clear error info accumulated during probe */
3171 ata_ering_clear(&dev->ering);
3172 new_mask |= 1 << dev->devno;
3173 break;
3174 case -ENOENT:
3175 /* IDENTIFY was issued to non-existent
3176 * device. No need to reset. Just
3177 * thaw and ignore the device.
3178 */
3179 ata_eh_thaw_port(ap);
3180 break;
3181 default:
3182 goto err;
3183 }
3184 }
3185 }
3186
3187 /* PDIAG- should have been released, ask cable type if post-reset */
3188 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3189 if (ap->ops->cable_detect)
3190 ap->cbl = ap->ops->cable_detect(ap);
3191 ata_force_cbl(ap);
3192 }
3193
3194 /* Configure new devices forward such that user doesn't see
3195 * device detection messages backwards.
3196 */
3197 ata_for_each_dev(dev, link, ALL) {
3198 if (!(new_mask & (1 << dev->devno)))
3199 continue;
3200
3201 dev->class = ehc->classes[dev->devno];
3202
3203 if (dev->class == ATA_DEV_PMP)
3204 continue;
3205
3206 ehc->i.flags |= ATA_EHI_PRINTINFO;
3207 rc = ata_dev_configure(dev);
3208 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3209 if (rc) {
3210 dev->class = ATA_DEV_UNKNOWN;
3211 goto err;
3212 }
3213
3214 spin_lock_irqsave(ap->lock, flags);
3215 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3216 spin_unlock_irqrestore(ap->lock, flags);
3217
3218 /* new device discovered, configure xfermode */
3219 ehc->i.flags |= ATA_EHI_SETMODE;
3220 }
3221
3222 return 0;
3223
3224 err:
3225 *r_failed_dev = dev;
3226 DPRINTK("EXIT rc=%d\n", rc);
3227 return rc;
3228 }
3229
3230 /**
3231 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3232 * @link: link on which timings will be programmed
3233 * @r_failed_dev: out parameter for failed device
3234 *
3235 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3236 * ata_set_mode() fails, pointer to the failing device is
3237 * returned in @r_failed_dev.
3238 *
3239 * LOCKING:
3240 * PCI/etc. bus probe sem.
3241 *
3242 * RETURNS:
3243 * 0 on success, negative errno otherwise
3244 */
3245 int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3246 {
3247 struct ata_port *ap = link->ap;
3248 struct ata_device *dev;
3249 int rc;
3250
3251 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3252 ata_for_each_dev(dev, link, ENABLED) {
3253 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3254 struct ata_ering_entry *ent;
3255
3256 ent = ata_ering_top(&dev->ering);
3257 if (ent)
3258 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3259 }
3260 }
3261
3262 /* has private set_mode? */
3263 if (ap->ops->set_mode)
3264 rc = ap->ops->set_mode(link, r_failed_dev);
3265 else
3266 rc = ata_do_set_mode(link, r_failed_dev);
3267
3268 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3269 ata_for_each_dev(dev, link, ENABLED) {
3270 struct ata_eh_context *ehc = &link->eh_context;
3271 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3272 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3273
3274 if (dev->xfer_mode != saved_xfer_mode ||
3275 ata_ncq_enabled(dev) != saved_ncq)
3276 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3277 }
3278
3279 return rc;
3280 }
3281
3282 /**
3283 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3284 * @dev: ATAPI device to clear UA for
3285 *
3286 * Resets and other operations can make an ATAPI device raise
3287 * UNIT ATTENTION which causes the next operation to fail. This
3288 * function clears UA.
3289 *
3290 * LOCKING:
3291 * EH context (may sleep).
3292 *
3293 * RETURNS:
3294 * 0 on success, -errno on failure.
3295 */
3296 static int atapi_eh_clear_ua(struct ata_device *dev)
3297 {
3298 int i;
3299
3300 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3301 u8 *sense_buffer = dev->link->ap->sector_buf;
3302 u8 sense_key = 0;
3303 unsigned int err_mask;
3304
3305 err_mask = atapi_eh_tur(dev, &sense_key);
3306 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3307 ata_dev_warn(dev,
3308 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3309 err_mask);
3310 return -EIO;
3311 }
3312
3313 if (!err_mask || sense_key != UNIT_ATTENTION)
3314 return 0;
3315
3316 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3317 if (err_mask) {
3318 ata_dev_warn(dev, "failed to clear "
3319 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3320 return -EIO;
3321 }
3322 }
3323
3324 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3325 ATA_EH_UA_TRIES);
3326
3327 return 0;
3328 }
3329
3330 /**
3331 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3332 * @dev: ATA device which may need FLUSH retry
3333 *
3334 * If @dev failed FLUSH, it needs to be reported upper layer
3335 * immediately as it means that @dev failed to remap and already
3336 * lost at least a sector and further FLUSH retrials won't make
3337 * any difference to the lost sector. However, if FLUSH failed
3338 * for other reasons, for example transmission error, FLUSH needs
3339 * to be retried.
3340 *
3341 * This function determines whether FLUSH failure retry is
3342 * necessary and performs it if so.
3343 *
3344 * RETURNS:
3345 * 0 if EH can continue, -errno if EH needs to be repeated.
3346 */
3347 static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3348 {
3349 struct ata_link *link = dev->link;
3350 struct ata_port *ap = link->ap;
3351 struct ata_queued_cmd *qc;
3352 struct ata_taskfile tf;
3353 unsigned int err_mask;
3354 int rc = 0;
3355
3356 /* did flush fail for this device? */
3357 if (!ata_tag_valid(link->active_tag))
3358 return 0;
3359
3360 qc = __ata_qc_from_tag(ap, link->active_tag);
3361 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3362 qc->tf.command != ATA_CMD_FLUSH))
3363 return 0;
3364
3365 /* if the device failed it, it should be reported to upper layers */
3366 if (qc->err_mask & AC_ERR_DEV)
3367 return 0;
3368
3369 /* flush failed for some other reason, give it another shot */
3370 ata_tf_init(dev, &tf);
3371
3372 tf.command = qc->tf.command;
3373 tf.flags |= ATA_TFLAG_DEVICE;
3374 tf.protocol = ATA_PROT_NODATA;
3375
3376 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3377 tf.command, qc->err_mask);
3378
3379 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3380 if (!err_mask) {
3381 /*
3382 * FLUSH is complete but there's no way to
3383 * successfully complete a failed command from EH.
3384 * Making sure retry is allowed at least once and
3385 * retrying it should do the trick - whatever was in
3386 * the cache is already on the platter and this won't
3387 * cause infinite loop.
3388 */
3389 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3390 } else {
3391 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3392 err_mask);
3393 rc = -EIO;
3394
3395 /* if device failed it, report it to upper layers */
3396 if (err_mask & AC_ERR_DEV) {
3397 qc->err_mask |= AC_ERR_DEV;
3398 qc->result_tf = tf;
3399 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3400 rc = 0;
3401 }
3402 }
3403 return rc;
3404 }
3405
3406 /**
3407 * ata_eh_set_lpm - configure SATA interface power management
3408 * @link: link to configure power management
3409 * @policy: the link power management policy
3410 * @r_failed_dev: out parameter for failed device
3411 *
3412 * Enable SATA Interface power management. This will enable
3413 * Device Interface Power Management (DIPM) for min_power
3414 * policy, and then call driver specific callbacks for
3415 * enabling Host Initiated Power management.
3416 *
3417 * LOCKING:
3418 * EH context.
3419 *
3420 * RETURNS:
3421 * 0 on success, -errno on failure.
3422 */
3423 static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3424 struct ata_device **r_failed_dev)
3425 {
3426 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3427 struct ata_eh_context *ehc = &link->eh_context;
3428 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3429 enum ata_lpm_policy old_policy = link->lpm_policy;
3430 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3431 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3432 unsigned int err_mask;
3433 int rc;
3434
3435 /* if the link or host doesn't do LPM, noop */
3436 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3437 return 0;
3438
3439 /*
3440 * DIPM is enabled only for MIN_POWER as some devices
3441 * misbehave when the host NACKs transition to SLUMBER. Order
3442 * device and link configurations such that the host always
3443 * allows DIPM requests.
3444 */
3445 ata_for_each_dev(dev, link, ENABLED) {
3446 bool hipm = ata_id_has_hipm(dev->id);
3447 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3448
3449 /* find the first enabled and LPM enabled devices */
3450 if (!link_dev)
3451 link_dev = dev;
3452
3453 if (!lpm_dev && (hipm || dipm))
3454 lpm_dev = dev;
3455
3456 hints &= ~ATA_LPM_EMPTY;
3457 if (!hipm)
3458 hints &= ~ATA_LPM_HIPM;
3459
3460 /* disable DIPM before changing link config */
3461 if (policy != ATA_LPM_MIN_POWER && dipm) {
3462 err_mask = ata_dev_set_feature(dev,
3463 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3464 if (err_mask && err_mask != AC_ERR_DEV) {
3465 ata_dev_warn(dev,
3466 "failed to disable DIPM, Emask 0x%x\n",
3467 err_mask);
3468 rc = -EIO;
3469 goto fail;
3470 }
3471 }
3472 }
3473
3474 if (ap) {
3475 rc = ap->ops->set_lpm(link, policy, hints);
3476 if (!rc && ap->slave_link)
3477 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3478 } else
3479 rc = sata_pmp_set_lpm(link, policy, hints);
3480
3481 /*
3482 * Attribute link config failure to the first (LPM) enabled
3483 * device on the link.
3484 */
3485 if (rc) {
3486 if (rc == -EOPNOTSUPP) {
3487 link->flags |= ATA_LFLAG_NO_LPM;
3488 return 0;
3489 }
3490 dev = lpm_dev ? lpm_dev : link_dev;
3491 goto fail;
3492 }
3493
3494 /*
3495 * Low level driver acked the transition. Issue DIPM command
3496 * with the new policy set.
3497 */
3498 link->lpm_policy = policy;
3499 if (ap && ap->slave_link)
3500 ap->slave_link->lpm_policy = policy;
3501
3502 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3503 ata_for_each_dev(dev, link, ENABLED) {
3504 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3505 ata_id_has_dipm(dev->id)) {
3506 err_mask = ata_dev_set_feature(dev,
3507 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3508 if (err_mask && err_mask != AC_ERR_DEV) {
3509 ata_dev_warn(dev,
3510 "failed to enable DIPM, Emask 0x%x\n",
3511 err_mask);
3512 rc = -EIO;
3513 goto fail;
3514 }
3515 }
3516 }
3517
3518 link->last_lpm_change = jiffies;
3519 link->flags |= ATA_LFLAG_CHANGED;
3520
3521 return 0;
3522
3523 fail:
3524 /* restore the old policy */
3525 link->lpm_policy = old_policy;
3526 if (ap && ap->slave_link)
3527 ap->slave_link->lpm_policy = old_policy;
3528
3529 /* if no device or only one more chance is left, disable LPM */
3530 if (!dev || ehc->tries[dev->devno] <= 2) {
3531 ata_link_warn(link, "disabling LPM on the link\n");
3532 link->flags |= ATA_LFLAG_NO_LPM;
3533 }
3534 if (r_failed_dev)
3535 *r_failed_dev = dev;
3536 return rc;
3537 }
3538
3539 int ata_link_nr_enabled(struct ata_link *link)
3540 {
3541 struct ata_device *dev;
3542 int cnt = 0;
3543
3544 ata_for_each_dev(dev, link, ENABLED)
3545 cnt++;
3546 return cnt;
3547 }
3548
3549 static int ata_link_nr_vacant(struct ata_link *link)
3550 {
3551 struct ata_device *dev;
3552 int cnt = 0;
3553
3554 ata_for_each_dev(dev, link, ALL)
3555 if (dev->class == ATA_DEV_UNKNOWN)
3556 cnt++;
3557 return cnt;
3558 }
3559
3560 static int ata_eh_skip_recovery(struct ata_link *link)
3561 {
3562 struct ata_port *ap = link->ap;
3563 struct ata_eh_context *ehc = &link->eh_context;
3564 struct ata_device *dev;
3565
3566 /* skip disabled links */
3567 if (link->flags & ATA_LFLAG_DISABLED)
3568 return 1;
3569
3570 /* skip if explicitly requested */
3571 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3572 return 1;
3573
3574 /* thaw frozen port and recover failed devices */
3575 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3576 return 0;
3577
3578 /* reset at least once if reset is requested */
3579 if ((ehc->i.action & ATA_EH_RESET) &&
3580 !(ehc->i.flags & ATA_EHI_DID_RESET))
3581 return 0;
3582
3583 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3584 ata_for_each_dev(dev, link, ALL) {
3585 if (dev->class == ATA_DEV_UNKNOWN &&
3586 ehc->classes[dev->devno] != ATA_DEV_NONE)
3587 return 0;
3588 }
3589
3590 return 1;
3591 }
3592
3593 static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3594 {
3595 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3596 u64 now = get_jiffies_64();
3597 int *trials = void_arg;
3598
3599 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3600 (ent->timestamp < now - min(now, interval)))
3601 return -1;
3602
3603 (*trials)++;
3604 return 0;
3605 }
3606
3607 static int ata_eh_schedule_probe(struct ata_device *dev)
3608 {
3609 struct ata_eh_context *ehc = &dev->link->eh_context;
3610 struct ata_link *link = ata_dev_phys_link(dev);
3611 int trials = 0;
3612
3613 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3614 (ehc->did_probe_mask & (1 << dev->devno)))
3615 return 0;
3616
3617 ata_eh_detach_dev(dev);
3618 ata_dev_init(dev);
3619 ehc->did_probe_mask |= (1 << dev->devno);
3620 ehc->i.action |= ATA_EH_RESET;
3621 ehc->saved_xfer_mode[dev->devno] = 0;
3622 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3623
3624 /* the link maybe in a deep sleep, wake it up */
3625 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3626 if (ata_is_host_link(link))
3627 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3628 ATA_LPM_EMPTY);
3629 else
3630 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3631 ATA_LPM_EMPTY);
3632 }
3633
3634 /* Record and count probe trials on the ering. The specific
3635 * error mask used is irrelevant. Because a successful device
3636 * detection clears the ering, this count accumulates only if
3637 * there are consecutive failed probes.
3638 *
3639 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3640 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3641 * forced to 1.5Gbps.
3642 *
3643 * This is to work around cases where failed link speed
3644 * negotiation results in device misdetection leading to
3645 * infinite DEVXCHG or PHRDY CHG events.
3646 */
3647 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3648 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3649
3650 if (trials > ATA_EH_PROBE_TRIALS)
3651 sata_down_spd_limit(link, 1);
3652
3653 return 1;
3654 }
3655
3656 static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3657 {
3658 struct ata_eh_context *ehc = &dev->link->eh_context;
3659
3660 /* -EAGAIN from EH routine indicates retry without prejudice.
3661 * The requester is responsible for ensuring forward progress.
3662 */
3663 if (err != -EAGAIN)
3664 ehc->tries[dev->devno]--;
3665
3666 switch (err) {
3667 case -ENODEV:
3668 /* device missing or wrong IDENTIFY data, schedule probing */
3669 ehc->i.probe_mask |= (1 << dev->devno);
3670 case -EINVAL:
3671 /* give it just one more chance */
3672 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3673 case -EIO:
3674 if (ehc->tries[dev->devno] == 1) {
3675 /* This is the last chance, better to slow
3676 * down than lose it.
3677 */
3678 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3679 if (dev->pio_mode > XFER_PIO_0)
3680 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3681 }
3682 }
3683
3684 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3685 /* disable device if it has used up all its chances */
3686 ata_dev_disable(dev);
3687
3688 /* detach if offline */
3689 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3690 ata_eh_detach_dev(dev);
3691
3692 /* schedule probe if necessary */
3693 if (ata_eh_schedule_probe(dev)) {
3694 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3695 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3696 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3697 }
3698
3699 return 1;
3700 } else {
3701 ehc->i.action |= ATA_EH_RESET;
3702 return 0;
3703 }
3704 }
3705
3706 /**
3707 * ata_eh_recover - recover host port after error
3708 * @ap: host port to recover
3709 * @prereset: prereset method (can be NULL)
3710 * @softreset: softreset method (can be NULL)
3711 * @hardreset: hardreset method (can be NULL)
3712 * @postreset: postreset method (can be NULL)
3713 * @r_failed_link: out parameter for failed link
3714 *
3715 * This is the alpha and omega, eum and yang, heart and soul of
3716 * libata exception handling. On entry, actions required to
3717 * recover each link and hotplug requests are recorded in the
3718 * link's eh_context. This function executes all the operations
3719 * with appropriate retrials and fallbacks to resurrect failed
3720 * devices, detach goners and greet newcomers.
3721 *
3722 * LOCKING:
3723 * Kernel thread context (may sleep).
3724 *
3725 * RETURNS:
3726 * 0 on success, -errno on failure.
3727 */
3728 int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3729 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3730 ata_postreset_fn_t postreset,
3731 struct ata_link **r_failed_link)
3732 {
3733 struct ata_link *link;
3734 struct ata_device *dev;
3735 int rc, nr_fails;
3736 unsigned long flags, deadline;
3737
3738 DPRINTK("ENTER\n");
3739
3740 /* prep for recovery */
3741 ata_for_each_link(link, ap, EDGE) {
3742 struct ata_eh_context *ehc = &link->eh_context;
3743
3744 /* re-enable link? */
3745 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3746 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3747 spin_lock_irqsave(ap->lock, flags);
3748 link->flags &= ~ATA_LFLAG_DISABLED;
3749 spin_unlock_irqrestore(ap->lock, flags);
3750 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3751 }
3752
3753 ata_for_each_dev(dev, link, ALL) {
3754 if (link->flags & ATA_LFLAG_NO_RETRY)
3755 ehc->tries[dev->devno] = 1;
3756 else
3757 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3758
3759 /* collect port action mask recorded in dev actions */
3760 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3761 ~ATA_EH_PERDEV_MASK;
3762 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3763
3764 /* process hotplug request */
3765 if (dev->flags & ATA_DFLAG_DETACH)
3766 ata_eh_detach_dev(dev);
3767
3768 /* schedule probe if necessary */
3769 if (!ata_dev_enabled(dev))
3770 ata_eh_schedule_probe(dev);
3771 }
3772 }
3773
3774 retry:
3775 rc = 0;
3776
3777 /* if UNLOADING, finish immediately */
3778 if (ap->pflags & ATA_PFLAG_UNLOADING)
3779 goto out;
3780
3781 /* prep for EH */
3782 ata_for_each_link(link, ap, EDGE) {
3783 struct ata_eh_context *ehc = &link->eh_context;
3784
3785 /* skip EH if possible. */
3786 if (ata_eh_skip_recovery(link))
3787 ehc->i.action = 0;
3788
3789 ata_for_each_dev(dev, link, ALL)
3790 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3791 }
3792
3793 /* reset */
3794 ata_for_each_link(link, ap, EDGE) {
3795 struct ata_eh_context *ehc = &link->eh_context;
3796
3797 if (!(ehc->i.action & ATA_EH_RESET))
3798 continue;
3799
3800 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3801 prereset, softreset, hardreset, postreset);
3802 if (rc) {
3803 ata_link_err(link, "reset failed, giving up\n");
3804 goto out;
3805 }
3806 }
3807
3808 do {
3809 unsigned long now;
3810
3811 /*
3812 * clears ATA_EH_PARK in eh_info and resets
3813 * ap->park_req_pending
3814 */
3815 ata_eh_pull_park_action(ap);
3816
3817 deadline = jiffies;
3818 ata_for_each_link(link, ap, EDGE) {
3819 ata_for_each_dev(dev, link, ALL) {
3820 struct ata_eh_context *ehc = &link->eh_context;
3821 unsigned long tmp;
3822
3823 if (dev->class != ATA_DEV_ATA &&
3824 dev->class != ATA_DEV_ZAC)
3825 continue;
3826 if (!(ehc->i.dev_action[dev->devno] &
3827 ATA_EH_PARK))
3828 continue;
3829 tmp = dev->unpark_deadline;
3830 if (time_before(deadline, tmp))
3831 deadline = tmp;
3832 else if (time_before_eq(tmp, jiffies))
3833 continue;
3834 if (ehc->unloaded_mask & (1 << dev->devno))
3835 continue;
3836
3837 ata_eh_park_issue_cmd(dev, 1);
3838 }
3839 }
3840
3841 now = jiffies;
3842 if (time_before_eq(deadline, now))
3843 break;
3844
3845 ata_eh_release(ap);
3846 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3847 deadline - now);
3848 ata_eh_acquire(ap);
3849 } while (deadline);
3850 ata_for_each_link(link, ap, EDGE) {
3851 ata_for_each_dev(dev, link, ALL) {
3852 if (!(link->eh_context.unloaded_mask &
3853 (1 << dev->devno)))
3854 continue;
3855
3856 ata_eh_park_issue_cmd(dev, 0);
3857 ata_eh_done(link, dev, ATA_EH_PARK);
3858 }
3859 }
3860
3861 /* the rest */
3862 nr_fails = 0;
3863 ata_for_each_link(link, ap, PMP_FIRST) {
3864 struct ata_eh_context *ehc = &link->eh_context;
3865
3866 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3867 goto config_lpm;
3868
3869 /* revalidate existing devices and attach new ones */
3870 rc = ata_eh_revalidate_and_attach(link, &dev);
3871 if (rc)
3872 goto rest_fail;
3873
3874 /* if PMP got attached, return, pmp EH will take care of it */
3875 if (link->device->class == ATA_DEV_PMP) {
3876 ehc->i.action = 0;
3877 return 0;
3878 }
3879
3880 /* configure transfer mode if necessary */
3881 if (ehc->i.flags & ATA_EHI_SETMODE) {
3882 rc = ata_set_mode(link, &dev);
3883 if (rc)
3884 goto rest_fail;
3885 ehc->i.flags &= ~ATA_EHI_SETMODE;
3886 }
3887
3888 /* If reset has been issued, clear UA to avoid
3889 * disrupting the current users of the device.
3890 */
3891 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3892 ata_for_each_dev(dev, link, ALL) {
3893 if (dev->class != ATA_DEV_ATAPI)
3894 continue;
3895 rc = atapi_eh_clear_ua(dev);
3896 if (rc)
3897 goto rest_fail;
3898 if (zpodd_dev_enabled(dev))
3899 zpodd_post_poweron(dev);
3900 }
3901 }
3902
3903 /* retry flush if necessary */
3904 ata_for_each_dev(dev, link, ALL) {
3905 if (dev->class != ATA_DEV_ATA &&
3906 dev->class != ATA_DEV_ZAC)
3907 continue;
3908 rc = ata_eh_maybe_retry_flush(dev);
3909 if (rc)
3910 goto rest_fail;
3911 }
3912
3913 config_lpm:
3914 /* configure link power saving */
3915 if (link->lpm_policy != ap->target_lpm_policy) {
3916 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3917 if (rc)
3918 goto rest_fail;
3919 }
3920
3921 /* this link is okay now */
3922 ehc->i.flags = 0;
3923 continue;
3924
3925 rest_fail:
3926 nr_fails++;
3927 if (dev)
3928 ata_eh_handle_dev_fail(dev, rc);
3929
3930 if (ap->pflags & ATA_PFLAG_FROZEN) {
3931 /* PMP reset requires working host port.
3932 * Can't retry if it's frozen.
3933 */
3934 if (sata_pmp_attached(ap))
3935 goto out;
3936 break;
3937 }
3938 }
3939
3940 if (nr_fails)
3941 goto retry;
3942
3943 out:
3944 if (rc && r_failed_link)
3945 *r_failed_link = link;
3946
3947 DPRINTK("EXIT, rc=%d\n", rc);
3948 return rc;
3949 }
3950
3951 /**
3952 * ata_eh_finish - finish up EH
3953 * @ap: host port to finish EH for
3954 *
3955 * Recovery is complete. Clean up EH states and retry or finish
3956 * failed qcs.
3957 *
3958 * LOCKING:
3959 * None.
3960 */
3961 void ata_eh_finish(struct ata_port *ap)
3962 {
3963 int tag;
3964
3965 /* retry or finish qcs */
3966 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3967 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3968
3969 if (!(qc->flags & ATA_QCFLAG_FAILED))
3970 continue;
3971
3972 if (qc->err_mask) {
3973 /* FIXME: Once EH migration is complete,
3974 * generate sense data in this function,
3975 * considering both err_mask and tf.
3976 */
3977 if (qc->flags & ATA_QCFLAG_RETRY)
3978 ata_eh_qc_retry(qc);
3979 else
3980 ata_eh_qc_complete(qc);
3981 } else {
3982 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3983 ata_eh_qc_complete(qc);
3984 } else {
3985 /* feed zero TF to sense generation */
3986 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3987 ata_eh_qc_retry(qc);
3988 }
3989 }
3990 }
3991
3992 /* make sure nr_active_links is zero after EH */
3993 WARN_ON(ap->nr_active_links);
3994 ap->nr_active_links = 0;
3995 }
3996
3997 /**
3998 * ata_do_eh - do standard error handling
3999 * @ap: host port to handle error for
4000 *
4001 * @prereset: prereset method (can be NULL)
4002 * @softreset: softreset method (can be NULL)
4003 * @hardreset: hardreset method (can be NULL)
4004 * @postreset: postreset method (can be NULL)
4005 *
4006 * Perform standard error handling sequence.
4007 *
4008 * LOCKING:
4009 * Kernel thread context (may sleep).
4010 */
4011 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4012 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4013 ata_postreset_fn_t postreset)
4014 {
4015 struct ata_device *dev;
4016 int rc;
4017
4018 ata_eh_autopsy(ap);
4019 ata_eh_report(ap);
4020
4021 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4022 NULL);
4023 if (rc) {
4024 ata_for_each_dev(dev, &ap->link, ALL)
4025 ata_dev_disable(dev);
4026 }
4027
4028 ata_eh_finish(ap);
4029 }
4030
4031 /**
4032 * ata_std_error_handler - standard error handler
4033 * @ap: host port to handle error for
4034 *
4035 * Standard error handler
4036 *
4037 * LOCKING:
4038 * Kernel thread context (may sleep).
4039 */
4040 void ata_std_error_handler(struct ata_port *ap)
4041 {
4042 struct ata_port_operations *ops = ap->ops;
4043 ata_reset_fn_t hardreset = ops->hardreset;
4044
4045 /* ignore built-in hardreset if SCR access is not available */
4046 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4047 hardreset = NULL;
4048
4049 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4050 }
4051
4052 #ifdef CONFIG_PM
4053 /**
4054 * ata_eh_handle_port_suspend - perform port suspend operation
4055 * @ap: port to suspend
4056 *
4057 * Suspend @ap.
4058 *
4059 * LOCKING:
4060 * Kernel thread context (may sleep).
4061 */
4062 static void ata_eh_handle_port_suspend(struct ata_port *ap)
4063 {
4064 unsigned long flags;
4065 int rc = 0;
4066 struct ata_device *dev;
4067
4068 /* are we suspending? */
4069 spin_lock_irqsave(ap->lock, flags);
4070 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4071 ap->pm_mesg.event & PM_EVENT_RESUME) {
4072 spin_unlock_irqrestore(ap->lock, flags);
4073 return;
4074 }
4075 spin_unlock_irqrestore(ap->lock, flags);
4076
4077 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4078
4079 /*
4080 * If we have a ZPODD attached, check its zero
4081 * power ready status before the port is frozen.
4082 * Only needed for runtime suspend.
4083 */
4084 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4085 ata_for_each_dev(dev, &ap->link, ENABLED) {
4086 if (zpodd_dev_enabled(dev))
4087 zpodd_on_suspend(dev);
4088 }
4089 }
4090
4091 /* tell ACPI we're suspending */
4092 rc = ata_acpi_on_suspend(ap);
4093 if (rc)
4094 goto out;
4095
4096 /* suspend */
4097 ata_eh_freeze_port(ap);
4098
4099 if (ap->ops->port_suspend)
4100 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4101
4102 ata_acpi_set_state(ap, ap->pm_mesg);
4103 out:
4104 /* update the flags */
4105 spin_lock_irqsave(ap->lock, flags);
4106
4107 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4108 if (rc == 0)
4109 ap->pflags |= ATA_PFLAG_SUSPENDED;
4110 else if (ap->pflags & ATA_PFLAG_FROZEN)
4111 ata_port_schedule_eh(ap);
4112
4113 spin_unlock_irqrestore(ap->lock, flags);
4114
4115 return;
4116 }
4117
4118 /**
4119 * ata_eh_handle_port_resume - perform port resume operation
4120 * @ap: port to resume
4121 *
4122 * Resume @ap.
4123 *
4124 * LOCKING:
4125 * Kernel thread context (may sleep).
4126 */
4127 static void ata_eh_handle_port_resume(struct ata_port *ap)
4128 {
4129 struct ata_link *link;
4130 struct ata_device *dev;
4131 unsigned long flags;
4132 int rc = 0;
4133
4134 /* are we resuming? */
4135 spin_lock_irqsave(ap->lock, flags);
4136 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4137 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4138 spin_unlock_irqrestore(ap->lock, flags);
4139 return;
4140 }
4141 spin_unlock_irqrestore(ap->lock, flags);
4142
4143 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4144
4145 /*
4146 * Error timestamps are in jiffies which doesn't run while
4147 * suspended and PHY events during resume isn't too uncommon.
4148 * When the two are combined, it can lead to unnecessary speed
4149 * downs if the machine is suspended and resumed repeatedly.
4150 * Clear error history.
4151 */
4152 ata_for_each_link(link, ap, HOST_FIRST)
4153 ata_for_each_dev(dev, link, ALL)
4154 ata_ering_clear(&dev->ering);
4155
4156 ata_acpi_set_state(ap, ap->pm_mesg);
4157
4158 if (ap->ops->port_resume)
4159 rc = ap->ops->port_resume(ap);
4160
4161 /* tell ACPI that we're resuming */
4162 ata_acpi_on_resume(ap);
4163
4164 /* update the flags */
4165 spin_lock_irqsave(ap->lock, flags);
4166 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4167 spin_unlock_irqrestore(ap->lock, flags);
4168 }
4169 #endif /* CONFIG_PM */