Loading...
1/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
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/pci.h>
38#include <scsi/scsi.h>
39#include <scsi/scsi_host.h>
40#include <scsi/scsi_eh.h>
41#include <scsi/scsi_device.h>
42#include <scsi/scsi_cmnd.h>
43#include <scsi/scsi_dbg.h>
44#include "../scsi/scsi_transport_api.h"
45
46#include <linux/libata.h>
47
48#include "libata.h"
49
50enum {
51 /* speed down verdicts */
52 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
53 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
54 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
55 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
56
57 /* error flags */
58 ATA_EFLAG_IS_IO = (1 << 0),
59 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
60 ATA_EFLAG_OLD_ER = (1 << 31),
61
62 /* error categories */
63 ATA_ECAT_NONE = 0,
64 ATA_ECAT_ATA_BUS = 1,
65 ATA_ECAT_TOUT_HSM = 2,
66 ATA_ECAT_UNK_DEV = 3,
67 ATA_ECAT_DUBIOUS_NONE = 4,
68 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
69 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
70 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
71 ATA_ECAT_NR = 8,
72
73 ATA_EH_CMD_DFL_TIMEOUT = 5000,
74
75 /* always put at least this amount of time between resets */
76 ATA_EH_RESET_COOL_DOWN = 5000,
77
78 /* Waiting in ->prereset can never be reliable. It's
79 * sometimes nice to wait there but it can't be depended upon;
80 * otherwise, we wouldn't be resetting. Just give it enough
81 * time for most drives to spin up.
82 */
83 ATA_EH_PRERESET_TIMEOUT = 10000,
84 ATA_EH_FASTDRAIN_INTERVAL = 3000,
85
86 ATA_EH_UA_TRIES = 5,
87
88 /* probe speed down parameters, see ata_eh_schedule_probe() */
89 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
90 ATA_EH_PROBE_TRIALS = 2,
91};
92
93/* The following table determines how we sequence resets. Each entry
94 * represents timeout for that try. The first try can be soft or
95 * hardreset. All others are hardreset if available. In most cases
96 * the first reset w/ 10sec timeout should succeed. Following entries
97 * are mostly for error handling, hotplug and retarded devices.
98 */
99static const unsigned long ata_eh_reset_timeouts[] = {
100 10000, /* most drives spin up by 10sec */
101 10000, /* > 99% working drives spin up before 20sec */
102 35000, /* give > 30 secs of idleness for retarded devices */
103 5000, /* and sweet one last chance */
104 ULONG_MAX, /* > 1 min has elapsed, give up */
105};
106
107static const unsigned long ata_eh_identify_timeouts[] = {
108 5000, /* covers > 99% of successes and not too boring on failures */
109 10000, /* combined time till here is enough even for media access */
110 30000, /* for true idiots */
111 ULONG_MAX,
112};
113
114static const unsigned long ata_eh_flush_timeouts[] = {
115 15000, /* be generous with flush */
116 15000, /* ditto */
117 30000, /* and even more generous */
118 ULONG_MAX,
119};
120
121static const unsigned long ata_eh_other_timeouts[] = {
122 5000, /* same rationale as identify timeout */
123 10000, /* ditto */
124 /* but no merciful 30sec for other commands, it just isn't worth it */
125 ULONG_MAX,
126};
127
128struct ata_eh_cmd_timeout_ent {
129 const u8 *commands;
130 const unsigned long *timeouts;
131};
132
133/* The following table determines timeouts to use for EH internal
134 * commands. Each table entry is a command class and matches the
135 * commands the entry applies to and the timeout table to use.
136 *
137 * On the retry after a command timed out, the next timeout value from
138 * the table is used. If the table doesn't contain further entries,
139 * the last value is used.
140 *
141 * ehc->cmd_timeout_idx keeps track of which timeout to use per
142 * command class, so if SET_FEATURES times out on the first try, the
143 * next try will use the second timeout value only for that class.
144 */
145#define CMDS(cmds...) (const u8 []){ cmds, 0 }
146static const struct ata_eh_cmd_timeout_ent
147ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
148 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
149 .timeouts = ata_eh_identify_timeouts, },
150 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
151 .timeouts = ata_eh_other_timeouts, },
152 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
153 .timeouts = ata_eh_other_timeouts, },
154 { .commands = CMDS(ATA_CMD_SET_FEATURES),
155 .timeouts = ata_eh_other_timeouts, },
156 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
157 .timeouts = ata_eh_other_timeouts, },
158 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
159 .timeouts = ata_eh_flush_timeouts },
160};
161#undef CMDS
162
163static void __ata_port_freeze(struct ata_port *ap);
164#ifdef CONFIG_PM
165static void ata_eh_handle_port_suspend(struct ata_port *ap);
166static void ata_eh_handle_port_resume(struct ata_port *ap);
167#else /* CONFIG_PM */
168static void ata_eh_handle_port_suspend(struct ata_port *ap)
169{ }
170
171static void ata_eh_handle_port_resume(struct ata_port *ap)
172{ }
173#endif /* CONFIG_PM */
174
175static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
176 va_list args)
177{
178 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
179 ATA_EH_DESC_LEN - ehi->desc_len,
180 fmt, args);
181}
182
183/**
184 * __ata_ehi_push_desc - push error description without adding separator
185 * @ehi: target EHI
186 * @fmt: printf format string
187 *
188 * Format string according to @fmt and append it to @ehi->desc.
189 *
190 * LOCKING:
191 * spin_lock_irqsave(host lock)
192 */
193void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
194{
195 va_list args;
196
197 va_start(args, fmt);
198 __ata_ehi_pushv_desc(ehi, fmt, args);
199 va_end(args);
200}
201
202/**
203 * ata_ehi_push_desc - push error description with separator
204 * @ehi: target EHI
205 * @fmt: printf format string
206 *
207 * Format string according to @fmt and append it to @ehi->desc.
208 * If @ehi->desc is not empty, ", " is added in-between.
209 *
210 * LOCKING:
211 * spin_lock_irqsave(host lock)
212 */
213void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
214{
215 va_list args;
216
217 if (ehi->desc_len)
218 __ata_ehi_push_desc(ehi, ", ");
219
220 va_start(args, fmt);
221 __ata_ehi_pushv_desc(ehi, fmt, args);
222 va_end(args);
223}
224
225/**
226 * ata_ehi_clear_desc - clean error description
227 * @ehi: target EHI
228 *
229 * Clear @ehi->desc.
230 *
231 * LOCKING:
232 * spin_lock_irqsave(host lock)
233 */
234void ata_ehi_clear_desc(struct ata_eh_info *ehi)
235{
236 ehi->desc[0] = '\0';
237 ehi->desc_len = 0;
238}
239
240/**
241 * ata_port_desc - append port description
242 * @ap: target ATA port
243 * @fmt: printf format string
244 *
245 * Format string according to @fmt and append it to port
246 * description. If port description is not empty, " " is added
247 * in-between. This function is to be used while initializing
248 * ata_host. The description is printed on host registration.
249 *
250 * LOCKING:
251 * None.
252 */
253void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
254{
255 va_list args;
256
257 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
258
259 if (ap->link.eh_info.desc_len)
260 __ata_ehi_push_desc(&ap->link.eh_info, " ");
261
262 va_start(args, fmt);
263 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
264 va_end(args);
265}
266
267#ifdef CONFIG_PCI
268
269/**
270 * ata_port_pbar_desc - append PCI BAR description
271 * @ap: target ATA port
272 * @bar: target PCI BAR
273 * @offset: offset into PCI BAR
274 * @name: name of the area
275 *
276 * If @offset is negative, this function formats a string which
277 * contains the name, address, size and type of the BAR and
278 * appends it to the port description. If @offset is zero or
279 * positive, only name and offsetted address is appended.
280 *
281 * LOCKING:
282 * None.
283 */
284void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
285 const char *name)
286{
287 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
288 char *type = "";
289 unsigned long long start, len;
290
291 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
292 type = "m";
293 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
294 type = "i";
295
296 start = (unsigned long long)pci_resource_start(pdev, bar);
297 len = (unsigned long long)pci_resource_len(pdev, bar);
298
299 if (offset < 0)
300 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
301 else
302 ata_port_desc(ap, "%s 0x%llx", name,
303 start + (unsigned long long)offset);
304}
305
306#endif /* CONFIG_PCI */
307
308static int ata_lookup_timeout_table(u8 cmd)
309{
310 int i;
311
312 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
313 const u8 *cur;
314
315 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
316 if (*cur == cmd)
317 return i;
318 }
319
320 return -1;
321}
322
323/**
324 * ata_internal_cmd_timeout - determine timeout for an internal command
325 * @dev: target device
326 * @cmd: internal command to be issued
327 *
328 * Determine timeout for internal command @cmd for @dev.
329 *
330 * LOCKING:
331 * EH context.
332 *
333 * RETURNS:
334 * Determined timeout.
335 */
336unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
337{
338 struct ata_eh_context *ehc = &dev->link->eh_context;
339 int ent = ata_lookup_timeout_table(cmd);
340 int idx;
341
342 if (ent < 0)
343 return ATA_EH_CMD_DFL_TIMEOUT;
344
345 idx = ehc->cmd_timeout_idx[dev->devno][ent];
346 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
347}
348
349/**
350 * ata_internal_cmd_timed_out - notification for internal command timeout
351 * @dev: target device
352 * @cmd: internal command which timed out
353 *
354 * Notify EH that internal command @cmd for @dev timed out. This
355 * function should be called only for commands whose timeouts are
356 * determined using ata_internal_cmd_timeout().
357 *
358 * LOCKING:
359 * EH context.
360 */
361void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
362{
363 struct ata_eh_context *ehc = &dev->link->eh_context;
364 int ent = ata_lookup_timeout_table(cmd);
365 int idx;
366
367 if (ent < 0)
368 return;
369
370 idx = ehc->cmd_timeout_idx[dev->devno][ent];
371 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
372 ehc->cmd_timeout_idx[dev->devno][ent]++;
373}
374
375static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
376 unsigned int err_mask)
377{
378 struct ata_ering_entry *ent;
379
380 WARN_ON(!err_mask);
381
382 ering->cursor++;
383 ering->cursor %= ATA_ERING_SIZE;
384
385 ent = &ering->ring[ering->cursor];
386 ent->eflags = eflags;
387 ent->err_mask = err_mask;
388 ent->timestamp = get_jiffies_64();
389}
390
391static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
392{
393 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
394
395 if (ent->err_mask)
396 return ent;
397 return NULL;
398}
399
400int ata_ering_map(struct ata_ering *ering,
401 int (*map_fn)(struct ata_ering_entry *, void *),
402 void *arg)
403{
404 int idx, rc = 0;
405 struct ata_ering_entry *ent;
406
407 idx = ering->cursor;
408 do {
409 ent = &ering->ring[idx];
410 if (!ent->err_mask)
411 break;
412 rc = map_fn(ent, arg);
413 if (rc)
414 break;
415 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
416 } while (idx != ering->cursor);
417
418 return rc;
419}
420
421int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
422{
423 ent->eflags |= ATA_EFLAG_OLD_ER;
424 return 0;
425}
426
427static void ata_ering_clear(struct ata_ering *ering)
428{
429 ata_ering_map(ering, ata_ering_clear_cb, NULL);
430}
431
432static unsigned int ata_eh_dev_action(struct ata_device *dev)
433{
434 struct ata_eh_context *ehc = &dev->link->eh_context;
435
436 return ehc->i.action | ehc->i.dev_action[dev->devno];
437}
438
439static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
440 struct ata_eh_info *ehi, unsigned int action)
441{
442 struct ata_device *tdev;
443
444 if (!dev) {
445 ehi->action &= ~action;
446 ata_for_each_dev(tdev, link, ALL)
447 ehi->dev_action[tdev->devno] &= ~action;
448 } else {
449 /* doesn't make sense for port-wide EH actions */
450 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
451
452 /* break ehi->action into ehi->dev_action */
453 if (ehi->action & action) {
454 ata_for_each_dev(tdev, link, ALL)
455 ehi->dev_action[tdev->devno] |=
456 ehi->action & action;
457 ehi->action &= ~action;
458 }
459
460 /* turn off the specified per-dev action */
461 ehi->dev_action[dev->devno] &= ~action;
462 }
463}
464
465/**
466 * ata_eh_acquire - acquire EH ownership
467 * @ap: ATA port to acquire EH ownership for
468 *
469 * Acquire EH ownership for @ap. This is the basic exclusion
470 * mechanism for ports sharing a host. Only one port hanging off
471 * the same host can claim the ownership of EH.
472 *
473 * LOCKING:
474 * EH context.
475 */
476void ata_eh_acquire(struct ata_port *ap)
477{
478 mutex_lock(&ap->host->eh_mutex);
479 WARN_ON_ONCE(ap->host->eh_owner);
480 ap->host->eh_owner = current;
481}
482
483/**
484 * ata_eh_release - release EH ownership
485 * @ap: ATA port to release EH ownership for
486 *
487 * Release EH ownership for @ap if the caller. The caller must
488 * have acquired EH ownership using ata_eh_acquire() previously.
489 *
490 * LOCKING:
491 * EH context.
492 */
493void ata_eh_release(struct ata_port *ap)
494{
495 WARN_ON_ONCE(ap->host->eh_owner != current);
496 ap->host->eh_owner = NULL;
497 mutex_unlock(&ap->host->eh_mutex);
498}
499
500/**
501 * ata_scsi_timed_out - SCSI layer time out callback
502 * @cmd: timed out SCSI command
503 *
504 * Handles SCSI layer timeout. We race with normal completion of
505 * the qc for @cmd. If the qc is already gone, we lose and let
506 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
507 * timed out and EH should be invoked. Prevent ata_qc_complete()
508 * from finishing it by setting EH_SCHEDULED and return
509 * EH_NOT_HANDLED.
510 *
511 * TODO: kill this function once old EH is gone.
512 *
513 * LOCKING:
514 * Called from timer context
515 *
516 * RETURNS:
517 * EH_HANDLED or EH_NOT_HANDLED
518 */
519enum blk_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
520{
521 struct Scsi_Host *host = cmd->device->host;
522 struct ata_port *ap = ata_shost_to_port(host);
523 unsigned long flags;
524 struct ata_queued_cmd *qc;
525 enum blk_eh_timer_return ret;
526
527 DPRINTK("ENTER\n");
528
529 if (ap->ops->error_handler) {
530 ret = BLK_EH_NOT_HANDLED;
531 goto out;
532 }
533
534 ret = BLK_EH_HANDLED;
535 spin_lock_irqsave(ap->lock, flags);
536 qc = ata_qc_from_tag(ap, ap->link.active_tag);
537 if (qc) {
538 WARN_ON(qc->scsicmd != cmd);
539 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
540 qc->err_mask |= AC_ERR_TIMEOUT;
541 ret = BLK_EH_NOT_HANDLED;
542 }
543 spin_unlock_irqrestore(ap->lock, flags);
544
545 out:
546 DPRINTK("EXIT, ret=%d\n", ret);
547 return ret;
548}
549
550static void ata_eh_unload(struct ata_port *ap)
551{
552 struct ata_link *link;
553 struct ata_device *dev;
554 unsigned long flags;
555
556 /* Restore SControl IPM and SPD for the next driver and
557 * disable attached devices.
558 */
559 ata_for_each_link(link, ap, PMP_FIRST) {
560 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
561 ata_for_each_dev(dev, link, ALL)
562 ata_dev_disable(dev);
563 }
564
565 /* freeze and set UNLOADED */
566 spin_lock_irqsave(ap->lock, flags);
567
568 ata_port_freeze(ap); /* won't be thawed */
569 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
570 ap->pflags |= ATA_PFLAG_UNLOADED;
571
572 spin_unlock_irqrestore(ap->lock, flags);
573}
574
575/**
576 * ata_scsi_error - SCSI layer error handler callback
577 * @host: SCSI host on which error occurred
578 *
579 * Handles SCSI-layer-thrown error events.
580 *
581 * LOCKING:
582 * Inherited from SCSI layer (none, can sleep)
583 *
584 * RETURNS:
585 * Zero.
586 */
587void ata_scsi_error(struct Scsi_Host *host)
588{
589 struct ata_port *ap = ata_shost_to_port(host);
590 unsigned long flags;
591 LIST_HEAD(eh_work_q);
592
593 DPRINTK("ENTER\n");
594
595 spin_lock_irqsave(host->host_lock, flags);
596 list_splice_init(&host->eh_cmd_q, &eh_work_q);
597 spin_unlock_irqrestore(host->host_lock, flags);
598
599 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
600
601 /* If we timed raced normal completion and there is nothing to
602 recover nr_timedout == 0 why exactly are we doing error recovery ? */
603 ata_scsi_port_error_handler(host, ap);
604
605 /* finish or retry handled scmd's and clean up */
606 WARN_ON(host->host_failed || !list_empty(&eh_work_q));
607
608 DPRINTK("EXIT\n");
609}
610
611/**
612 * ata_scsi_cmd_error_handler - error callback for a list of commands
613 * @host: scsi host containing the port
614 * @ap: ATA port within the host
615 * @eh_work_q: list of commands to process
616 *
617 * process the given list of commands and return those finished to the
618 * ap->eh_done_q. This function is the first part of the libata error
619 * handler which processes a given list of failed commands.
620 */
621void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
622 struct list_head *eh_work_q)
623{
624 int i;
625 unsigned long flags;
626
627 /* make sure sff pio task is not running */
628 ata_sff_flush_pio_task(ap);
629
630 /* synchronize with host lock and sort out timeouts */
631
632 /* For new EH, all qcs are finished in one of three ways -
633 * normal completion, error completion, and SCSI timeout.
634 * Both completions can race against SCSI timeout. When normal
635 * completion wins, the qc never reaches EH. When error
636 * completion wins, the qc has ATA_QCFLAG_FAILED set.
637 *
638 * When SCSI timeout wins, things are a bit more complex.
639 * Normal or error completion can occur after the timeout but
640 * before this point. In such cases, both types of
641 * completions are honored. A scmd is determined to have
642 * timed out iff its associated qc is active and not failed.
643 */
644 if (ap->ops->error_handler) {
645 struct scsi_cmnd *scmd, *tmp;
646 int nr_timedout = 0;
647
648 spin_lock_irqsave(ap->lock, flags);
649
650 /* This must occur under the ap->lock as we don't want
651 a polled recovery to race the real interrupt handler
652
653 The lost_interrupt handler checks for any completed but
654 non-notified command and completes much like an IRQ handler.
655
656 We then fall into the error recovery code which will treat
657 this as if normal completion won the race */
658
659 if (ap->ops->lost_interrupt)
660 ap->ops->lost_interrupt(ap);
661
662 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
663 struct ata_queued_cmd *qc;
664
665 for (i = 0; i < ATA_MAX_QUEUE; i++) {
666 qc = __ata_qc_from_tag(ap, i);
667 if (qc->flags & ATA_QCFLAG_ACTIVE &&
668 qc->scsicmd == scmd)
669 break;
670 }
671
672 if (i < ATA_MAX_QUEUE) {
673 /* the scmd has an associated qc */
674 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
675 /* which hasn't failed yet, timeout */
676 qc->err_mask |= AC_ERR_TIMEOUT;
677 qc->flags |= ATA_QCFLAG_FAILED;
678 nr_timedout++;
679 }
680 } else {
681 /* Normal completion occurred after
682 * SCSI timeout but before this point.
683 * Successfully complete it.
684 */
685 scmd->retries = scmd->allowed;
686 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
687 }
688 }
689
690 /* If we have timed out qcs. They belong to EH from
691 * this point but the state of the controller is
692 * unknown. Freeze the port to make sure the IRQ
693 * handler doesn't diddle with those qcs. This must
694 * be done atomically w.r.t. setting QCFLAG_FAILED.
695 */
696 if (nr_timedout)
697 __ata_port_freeze(ap);
698
699 spin_unlock_irqrestore(ap->lock, flags);
700
701 /* initialize eh_tries */
702 ap->eh_tries = ATA_EH_MAX_TRIES;
703 } else
704 spin_unlock_wait(ap->lock);
705
706}
707EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
708
709/**
710 * ata_scsi_port_error_handler - recover the port after the commands
711 * @host: SCSI host containing the port
712 * @ap: the ATA port
713 *
714 * Handle the recovery of the port @ap after all the commands
715 * have been recovered.
716 */
717void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
718{
719 unsigned long flags;
720
721 /* invoke error handler */
722 if (ap->ops->error_handler) {
723 struct ata_link *link;
724
725 /* acquire EH ownership */
726 ata_eh_acquire(ap);
727 repeat:
728 /* kill fast drain timer */
729 del_timer_sync(&ap->fastdrain_timer);
730
731 /* process port resume request */
732 ata_eh_handle_port_resume(ap);
733
734 /* fetch & clear EH info */
735 spin_lock_irqsave(ap->lock, flags);
736
737 ata_for_each_link(link, ap, HOST_FIRST) {
738 struct ata_eh_context *ehc = &link->eh_context;
739 struct ata_device *dev;
740
741 memset(&link->eh_context, 0, sizeof(link->eh_context));
742 link->eh_context.i = link->eh_info;
743 memset(&link->eh_info, 0, sizeof(link->eh_info));
744
745 ata_for_each_dev(dev, link, ENABLED) {
746 int devno = dev->devno;
747
748 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
749 if (ata_ncq_enabled(dev))
750 ehc->saved_ncq_enabled |= 1 << devno;
751 }
752 }
753
754 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
755 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
756 ap->excl_link = NULL; /* don't maintain exclusion over EH */
757
758 spin_unlock_irqrestore(ap->lock, flags);
759
760 /* invoke EH, skip if unloading or suspended */
761 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
762 ap->ops->error_handler(ap);
763 else {
764 /* if unloading, commence suicide */
765 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
766 !(ap->pflags & ATA_PFLAG_UNLOADED))
767 ata_eh_unload(ap);
768 ata_eh_finish(ap);
769 }
770
771 /* process port suspend request */
772 ata_eh_handle_port_suspend(ap);
773
774 /* Exception might have happened after ->error_handler
775 * recovered the port but before this point. Repeat
776 * EH in such case.
777 */
778 spin_lock_irqsave(ap->lock, flags);
779
780 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
781 if (--ap->eh_tries) {
782 spin_unlock_irqrestore(ap->lock, flags);
783 goto repeat;
784 }
785 ata_port_err(ap,
786 "EH pending after %d tries, giving up\n",
787 ATA_EH_MAX_TRIES);
788 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
789 }
790
791 /* this run is complete, make sure EH info is clear */
792 ata_for_each_link(link, ap, HOST_FIRST)
793 memset(&link->eh_info, 0, sizeof(link->eh_info));
794
795 /* Clear host_eh_scheduled while holding ap->lock such
796 * that if exception occurs after this point but
797 * before EH completion, SCSI midlayer will
798 * re-initiate EH.
799 */
800 host->host_eh_scheduled = 0;
801
802 spin_unlock_irqrestore(ap->lock, flags);
803 ata_eh_release(ap);
804 } else {
805 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
806 ap->ops->eng_timeout(ap);
807 }
808
809 scsi_eh_flush_done_q(&ap->eh_done_q);
810
811 /* clean up */
812 spin_lock_irqsave(ap->lock, flags);
813
814 if (ap->pflags & ATA_PFLAG_LOADING)
815 ap->pflags &= ~ATA_PFLAG_LOADING;
816 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
817 schedule_delayed_work(&ap->hotplug_task, 0);
818
819 if (ap->pflags & ATA_PFLAG_RECOVERED)
820 ata_port_info(ap, "EH complete\n");
821
822 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
823
824 /* tell wait_eh that we're done */
825 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
826 wake_up_all(&ap->eh_wait_q);
827
828 spin_unlock_irqrestore(ap->lock, flags);
829}
830EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
831
832/**
833 * ata_port_wait_eh - Wait for the currently pending EH to complete
834 * @ap: Port to wait EH for
835 *
836 * Wait until the currently pending EH is complete.
837 *
838 * LOCKING:
839 * Kernel thread context (may sleep).
840 */
841void ata_port_wait_eh(struct ata_port *ap)
842{
843 unsigned long flags;
844 DEFINE_WAIT(wait);
845
846 retry:
847 spin_lock_irqsave(ap->lock, flags);
848
849 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
850 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
851 spin_unlock_irqrestore(ap->lock, flags);
852 schedule();
853 spin_lock_irqsave(ap->lock, flags);
854 }
855 finish_wait(&ap->eh_wait_q, &wait);
856
857 spin_unlock_irqrestore(ap->lock, flags);
858
859 /* make sure SCSI EH is complete */
860 if (scsi_host_in_recovery(ap->scsi_host)) {
861 ata_msleep(ap, 10);
862 goto retry;
863 }
864}
865
866static int ata_eh_nr_in_flight(struct ata_port *ap)
867{
868 unsigned int tag;
869 int nr = 0;
870
871 /* count only non-internal commands */
872 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++)
873 if (ata_qc_from_tag(ap, tag))
874 nr++;
875
876 return nr;
877}
878
879void ata_eh_fastdrain_timerfn(unsigned long arg)
880{
881 struct ata_port *ap = (void *)arg;
882 unsigned long flags;
883 int cnt;
884
885 spin_lock_irqsave(ap->lock, flags);
886
887 cnt = ata_eh_nr_in_flight(ap);
888
889 /* are we done? */
890 if (!cnt)
891 goto out_unlock;
892
893 if (cnt == ap->fastdrain_cnt) {
894 unsigned int tag;
895
896 /* No progress during the last interval, tag all
897 * in-flight qcs as timed out and freeze the port.
898 */
899 for (tag = 0; tag < ATA_MAX_QUEUE - 1; tag++) {
900 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
901 if (qc)
902 qc->err_mask |= AC_ERR_TIMEOUT;
903 }
904
905 ata_port_freeze(ap);
906 } else {
907 /* some qcs have finished, give it another chance */
908 ap->fastdrain_cnt = cnt;
909 ap->fastdrain_timer.expires =
910 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
911 add_timer(&ap->fastdrain_timer);
912 }
913
914 out_unlock:
915 spin_unlock_irqrestore(ap->lock, flags);
916}
917
918/**
919 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
920 * @ap: target ATA port
921 * @fastdrain: activate fast drain
922 *
923 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
924 * is non-zero and EH wasn't pending before. Fast drain ensures
925 * that EH kicks in in timely manner.
926 *
927 * LOCKING:
928 * spin_lock_irqsave(host lock)
929 */
930static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
931{
932 int cnt;
933
934 /* already scheduled? */
935 if (ap->pflags & ATA_PFLAG_EH_PENDING)
936 return;
937
938 ap->pflags |= ATA_PFLAG_EH_PENDING;
939
940 if (!fastdrain)
941 return;
942
943 /* do we have in-flight qcs? */
944 cnt = ata_eh_nr_in_flight(ap);
945 if (!cnt)
946 return;
947
948 /* activate fast drain */
949 ap->fastdrain_cnt = cnt;
950 ap->fastdrain_timer.expires =
951 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
952 add_timer(&ap->fastdrain_timer);
953}
954
955/**
956 * ata_qc_schedule_eh - schedule qc for error handling
957 * @qc: command to schedule error handling for
958 *
959 * Schedule error handling for @qc. EH will kick in as soon as
960 * other commands are drained.
961 *
962 * LOCKING:
963 * spin_lock_irqsave(host lock)
964 */
965void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
966{
967 struct ata_port *ap = qc->ap;
968 struct request_queue *q = qc->scsicmd->device->request_queue;
969 unsigned long flags;
970
971 WARN_ON(!ap->ops->error_handler);
972
973 qc->flags |= ATA_QCFLAG_FAILED;
974 ata_eh_set_pending(ap, 1);
975
976 /* The following will fail if timeout has already expired.
977 * ata_scsi_error() takes care of such scmds on EH entry.
978 * Note that ATA_QCFLAG_FAILED is unconditionally set after
979 * this function completes.
980 */
981 spin_lock_irqsave(q->queue_lock, flags);
982 blk_abort_request(qc->scsicmd->request);
983 spin_unlock_irqrestore(q->queue_lock, flags);
984}
985
986/**
987 * ata_port_schedule_eh - schedule error handling without a qc
988 * @ap: ATA port to schedule EH for
989 *
990 * Schedule error handling for @ap. EH will kick in as soon as
991 * all commands are drained.
992 *
993 * LOCKING:
994 * spin_lock_irqsave(host lock)
995 */
996void ata_port_schedule_eh(struct ata_port *ap)
997{
998 WARN_ON(!ap->ops->error_handler);
999
1000 if (ap->pflags & ATA_PFLAG_INITIALIZING)
1001 return;
1002
1003 ata_eh_set_pending(ap, 1);
1004 scsi_schedule_eh(ap->scsi_host);
1005
1006 DPRINTK("port EH scheduled\n");
1007}
1008
1009static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
1010{
1011 int tag, nr_aborted = 0;
1012
1013 WARN_ON(!ap->ops->error_handler);
1014
1015 /* we're gonna abort all commands, no need for fast drain */
1016 ata_eh_set_pending(ap, 0);
1017
1018 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1019 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
1020
1021 if (qc && (!link || qc->dev->link == link)) {
1022 qc->flags |= ATA_QCFLAG_FAILED;
1023 ata_qc_complete(qc);
1024 nr_aborted++;
1025 }
1026 }
1027
1028 if (!nr_aborted)
1029 ata_port_schedule_eh(ap);
1030
1031 return nr_aborted;
1032}
1033
1034/**
1035 * ata_link_abort - abort all qc's on the link
1036 * @link: ATA link to abort qc's for
1037 *
1038 * Abort all active qc's active on @link and schedule EH.
1039 *
1040 * LOCKING:
1041 * spin_lock_irqsave(host lock)
1042 *
1043 * RETURNS:
1044 * Number of aborted qc's.
1045 */
1046int ata_link_abort(struct ata_link *link)
1047{
1048 return ata_do_link_abort(link->ap, link);
1049}
1050
1051/**
1052 * ata_port_abort - abort all qc's on the port
1053 * @ap: ATA port to abort qc's for
1054 *
1055 * Abort all active qc's of @ap and schedule EH.
1056 *
1057 * LOCKING:
1058 * spin_lock_irqsave(host_set lock)
1059 *
1060 * RETURNS:
1061 * Number of aborted qc's.
1062 */
1063int ata_port_abort(struct ata_port *ap)
1064{
1065 return ata_do_link_abort(ap, NULL);
1066}
1067
1068/**
1069 * __ata_port_freeze - freeze port
1070 * @ap: ATA port to freeze
1071 *
1072 * This function is called when HSM violation or some other
1073 * condition disrupts normal operation of the port. Frozen port
1074 * is not allowed to perform any operation until the port is
1075 * thawed, which usually follows a successful reset.
1076 *
1077 * ap->ops->freeze() callback can be used for freezing the port
1078 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1079 * port cannot be frozen hardware-wise, the interrupt handler
1080 * must ack and clear interrupts unconditionally while the port
1081 * is frozen.
1082 *
1083 * LOCKING:
1084 * spin_lock_irqsave(host lock)
1085 */
1086static void __ata_port_freeze(struct ata_port *ap)
1087{
1088 WARN_ON(!ap->ops->error_handler);
1089
1090 if (ap->ops->freeze)
1091 ap->ops->freeze(ap);
1092
1093 ap->pflags |= ATA_PFLAG_FROZEN;
1094
1095 DPRINTK("ata%u port frozen\n", ap->print_id);
1096}
1097
1098/**
1099 * ata_port_freeze - abort & freeze port
1100 * @ap: ATA port to freeze
1101 *
1102 * Abort and freeze @ap. The freeze operation must be called
1103 * first, because some hardware requires special operations
1104 * before the taskfile registers are accessible.
1105 *
1106 * LOCKING:
1107 * spin_lock_irqsave(host lock)
1108 *
1109 * RETURNS:
1110 * Number of aborted commands.
1111 */
1112int ata_port_freeze(struct ata_port *ap)
1113{
1114 int nr_aborted;
1115
1116 WARN_ON(!ap->ops->error_handler);
1117
1118 __ata_port_freeze(ap);
1119 nr_aborted = ata_port_abort(ap);
1120
1121 return nr_aborted;
1122}
1123
1124/**
1125 * sata_async_notification - SATA async notification handler
1126 * @ap: ATA port where async notification is received
1127 *
1128 * Handler to be called when async notification via SDB FIS is
1129 * received. This function schedules EH if necessary.
1130 *
1131 * LOCKING:
1132 * spin_lock_irqsave(host lock)
1133 *
1134 * RETURNS:
1135 * 1 if EH is scheduled, 0 otherwise.
1136 */
1137int sata_async_notification(struct ata_port *ap)
1138{
1139 u32 sntf;
1140 int rc;
1141
1142 if (!(ap->flags & ATA_FLAG_AN))
1143 return 0;
1144
1145 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1146 if (rc == 0)
1147 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1148
1149 if (!sata_pmp_attached(ap) || rc) {
1150 /* PMP is not attached or SNTF is not available */
1151 if (!sata_pmp_attached(ap)) {
1152 /* PMP is not attached. Check whether ATAPI
1153 * AN is configured. If so, notify media
1154 * change.
1155 */
1156 struct ata_device *dev = ap->link.device;
1157
1158 if ((dev->class == ATA_DEV_ATAPI) &&
1159 (dev->flags & ATA_DFLAG_AN))
1160 ata_scsi_media_change_notify(dev);
1161 return 0;
1162 } else {
1163 /* PMP is attached but SNTF is not available.
1164 * ATAPI async media change notification is
1165 * not used. The PMP must be reporting PHY
1166 * status change, schedule EH.
1167 */
1168 ata_port_schedule_eh(ap);
1169 return 1;
1170 }
1171 } else {
1172 /* PMP is attached and SNTF is available */
1173 struct ata_link *link;
1174
1175 /* check and notify ATAPI AN */
1176 ata_for_each_link(link, ap, EDGE) {
1177 if (!(sntf & (1 << link->pmp)))
1178 continue;
1179
1180 if ((link->device->class == ATA_DEV_ATAPI) &&
1181 (link->device->flags & ATA_DFLAG_AN))
1182 ata_scsi_media_change_notify(link->device);
1183 }
1184
1185 /* If PMP is reporting that PHY status of some
1186 * downstream ports has changed, schedule EH.
1187 */
1188 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1189 ata_port_schedule_eh(ap);
1190 return 1;
1191 }
1192
1193 return 0;
1194 }
1195}
1196
1197/**
1198 * ata_eh_freeze_port - EH helper to freeze port
1199 * @ap: ATA port to freeze
1200 *
1201 * Freeze @ap.
1202 *
1203 * LOCKING:
1204 * None.
1205 */
1206void ata_eh_freeze_port(struct ata_port *ap)
1207{
1208 unsigned long flags;
1209
1210 if (!ap->ops->error_handler)
1211 return;
1212
1213 spin_lock_irqsave(ap->lock, flags);
1214 __ata_port_freeze(ap);
1215 spin_unlock_irqrestore(ap->lock, flags);
1216}
1217
1218/**
1219 * ata_port_thaw_port - EH helper to thaw port
1220 * @ap: ATA port to thaw
1221 *
1222 * Thaw frozen port @ap.
1223 *
1224 * LOCKING:
1225 * None.
1226 */
1227void ata_eh_thaw_port(struct ata_port *ap)
1228{
1229 unsigned long flags;
1230
1231 if (!ap->ops->error_handler)
1232 return;
1233
1234 spin_lock_irqsave(ap->lock, flags);
1235
1236 ap->pflags &= ~ATA_PFLAG_FROZEN;
1237
1238 if (ap->ops->thaw)
1239 ap->ops->thaw(ap);
1240
1241 spin_unlock_irqrestore(ap->lock, flags);
1242
1243 DPRINTK("ata%u port thawed\n", ap->print_id);
1244}
1245
1246static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1247{
1248 /* nada */
1249}
1250
1251static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1252{
1253 struct ata_port *ap = qc->ap;
1254 struct scsi_cmnd *scmd = qc->scsicmd;
1255 unsigned long flags;
1256
1257 spin_lock_irqsave(ap->lock, flags);
1258 qc->scsidone = ata_eh_scsidone;
1259 __ata_qc_complete(qc);
1260 WARN_ON(ata_tag_valid(qc->tag));
1261 spin_unlock_irqrestore(ap->lock, flags);
1262
1263 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1264}
1265
1266/**
1267 * ata_eh_qc_complete - Complete an active ATA command from EH
1268 * @qc: Command to complete
1269 *
1270 * Indicate to the mid and upper layers that an ATA command has
1271 * completed. To be used from EH.
1272 */
1273void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1274{
1275 struct scsi_cmnd *scmd = qc->scsicmd;
1276 scmd->retries = scmd->allowed;
1277 __ata_eh_qc_complete(qc);
1278}
1279
1280/**
1281 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1282 * @qc: Command to retry
1283 *
1284 * Indicate to the mid and upper layers that an ATA command
1285 * should be retried. To be used from EH.
1286 *
1287 * SCSI midlayer limits the number of retries to scmd->allowed.
1288 * scmd->retries is decremented for commands which get retried
1289 * due to unrelated failures (qc->err_mask is zero).
1290 */
1291void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1292{
1293 struct scsi_cmnd *scmd = qc->scsicmd;
1294 if (!qc->err_mask && scmd->retries)
1295 scmd->retries--;
1296 __ata_eh_qc_complete(qc);
1297}
1298
1299/**
1300 * ata_dev_disable - disable ATA device
1301 * @dev: ATA device to disable
1302 *
1303 * Disable @dev.
1304 *
1305 * Locking:
1306 * EH context.
1307 */
1308void ata_dev_disable(struct ata_device *dev)
1309{
1310 if (!ata_dev_enabled(dev))
1311 return;
1312
1313 if (ata_msg_drv(dev->link->ap))
1314 ata_dev_warn(dev, "disabled\n");
1315 ata_acpi_on_disable(dev);
1316 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1317 dev->class++;
1318
1319 /* From now till the next successful probe, ering is used to
1320 * track probe failures. Clear accumulated device error info.
1321 */
1322 ata_ering_clear(&dev->ering);
1323}
1324
1325/**
1326 * ata_eh_detach_dev - detach ATA device
1327 * @dev: ATA device to detach
1328 *
1329 * Detach @dev.
1330 *
1331 * LOCKING:
1332 * None.
1333 */
1334void ata_eh_detach_dev(struct ata_device *dev)
1335{
1336 struct ata_link *link = dev->link;
1337 struct ata_port *ap = link->ap;
1338 struct ata_eh_context *ehc = &link->eh_context;
1339 unsigned long flags;
1340
1341 ata_dev_disable(dev);
1342
1343 spin_lock_irqsave(ap->lock, flags);
1344
1345 dev->flags &= ~ATA_DFLAG_DETACH;
1346
1347 if (ata_scsi_offline_dev(dev)) {
1348 dev->flags |= ATA_DFLAG_DETACHED;
1349 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1350 }
1351
1352 /* clear per-dev EH info */
1353 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1354 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1355 ehc->saved_xfer_mode[dev->devno] = 0;
1356 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1357
1358 spin_unlock_irqrestore(ap->lock, flags);
1359}
1360
1361/**
1362 * ata_eh_about_to_do - about to perform eh_action
1363 * @link: target ATA link
1364 * @dev: target ATA dev for per-dev action (can be NULL)
1365 * @action: action about to be performed
1366 *
1367 * Called just before performing EH actions to clear related bits
1368 * in @link->eh_info such that eh actions are not unnecessarily
1369 * repeated.
1370 *
1371 * LOCKING:
1372 * None.
1373 */
1374void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1375 unsigned int action)
1376{
1377 struct ata_port *ap = link->ap;
1378 struct ata_eh_info *ehi = &link->eh_info;
1379 struct ata_eh_context *ehc = &link->eh_context;
1380 unsigned long flags;
1381
1382 spin_lock_irqsave(ap->lock, flags);
1383
1384 ata_eh_clear_action(link, dev, ehi, action);
1385
1386 /* About to take EH action, set RECOVERED. Ignore actions on
1387 * slave links as master will do them again.
1388 */
1389 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1390 ap->pflags |= ATA_PFLAG_RECOVERED;
1391
1392 spin_unlock_irqrestore(ap->lock, flags);
1393}
1394
1395/**
1396 * ata_eh_done - EH action complete
1397* @ap: target ATA port
1398 * @dev: target ATA dev for per-dev action (can be NULL)
1399 * @action: action just completed
1400 *
1401 * Called right after performing EH actions to clear related bits
1402 * in @link->eh_context.
1403 *
1404 * LOCKING:
1405 * None.
1406 */
1407void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1408 unsigned int action)
1409{
1410 struct ata_eh_context *ehc = &link->eh_context;
1411
1412 ata_eh_clear_action(link, dev, &ehc->i, action);
1413}
1414
1415/**
1416 * ata_err_string - convert err_mask to descriptive string
1417 * @err_mask: error mask to convert to string
1418 *
1419 * Convert @err_mask to descriptive string. Errors are
1420 * prioritized according to severity and only the most severe
1421 * error is reported.
1422 *
1423 * LOCKING:
1424 * None.
1425 *
1426 * RETURNS:
1427 * Descriptive string for @err_mask
1428 */
1429static const char *ata_err_string(unsigned int err_mask)
1430{
1431 if (err_mask & AC_ERR_HOST_BUS)
1432 return "host bus error";
1433 if (err_mask & AC_ERR_ATA_BUS)
1434 return "ATA bus error";
1435 if (err_mask & AC_ERR_TIMEOUT)
1436 return "timeout";
1437 if (err_mask & AC_ERR_HSM)
1438 return "HSM violation";
1439 if (err_mask & AC_ERR_SYSTEM)
1440 return "internal error";
1441 if (err_mask & AC_ERR_MEDIA)
1442 return "media error";
1443 if (err_mask & AC_ERR_INVALID)
1444 return "invalid argument";
1445 if (err_mask & AC_ERR_DEV)
1446 return "device error";
1447 return "unknown error";
1448}
1449
1450/**
1451 * ata_read_log_page - read a specific log page
1452 * @dev: target device
1453 * @page: page to read
1454 * @buf: buffer to store read page
1455 * @sectors: number of sectors to read
1456 *
1457 * Read log page using READ_LOG_EXT command.
1458 *
1459 * LOCKING:
1460 * Kernel thread context (may sleep).
1461 *
1462 * RETURNS:
1463 * 0 on success, AC_ERR_* mask otherwise.
1464 */
1465static unsigned int ata_read_log_page(struct ata_device *dev,
1466 u8 page, void *buf, unsigned int sectors)
1467{
1468 struct ata_taskfile tf;
1469 unsigned int err_mask;
1470
1471 DPRINTK("read log page - page %d\n", page);
1472
1473 ata_tf_init(dev, &tf);
1474 tf.command = ATA_CMD_READ_LOG_EXT;
1475 tf.lbal = page;
1476 tf.nsect = sectors;
1477 tf.hob_nsect = sectors >> 8;
1478 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
1479 tf.protocol = ATA_PROT_PIO;
1480
1481 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
1482 buf, sectors * ATA_SECT_SIZE, 0);
1483
1484 DPRINTK("EXIT, err_mask=%x\n", err_mask);
1485 return err_mask;
1486}
1487
1488/**
1489 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1490 * @dev: Device to read log page 10h from
1491 * @tag: Resulting tag of the failed command
1492 * @tf: Resulting taskfile registers of the failed command
1493 *
1494 * Read log page 10h to obtain NCQ error details and clear error
1495 * condition.
1496 *
1497 * LOCKING:
1498 * Kernel thread context (may sleep).
1499 *
1500 * RETURNS:
1501 * 0 on success, -errno otherwise.
1502 */
1503static int ata_eh_read_log_10h(struct ata_device *dev,
1504 int *tag, struct ata_taskfile *tf)
1505{
1506 u8 *buf = dev->link->ap->sector_buf;
1507 unsigned int err_mask;
1508 u8 csum;
1509 int i;
1510
1511 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1512 if (err_mask)
1513 return -EIO;
1514
1515 csum = 0;
1516 for (i = 0; i < ATA_SECT_SIZE; i++)
1517 csum += buf[i];
1518 if (csum)
1519 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1520 csum);
1521
1522 if (buf[0] & 0x80)
1523 return -ENOENT;
1524
1525 *tag = buf[0] & 0x1f;
1526
1527 tf->command = buf[2];
1528 tf->feature = buf[3];
1529 tf->lbal = buf[4];
1530 tf->lbam = buf[5];
1531 tf->lbah = buf[6];
1532 tf->device = buf[7];
1533 tf->hob_lbal = buf[8];
1534 tf->hob_lbam = buf[9];
1535 tf->hob_lbah = buf[10];
1536 tf->nsect = buf[12];
1537 tf->hob_nsect = buf[13];
1538
1539 return 0;
1540}
1541
1542/**
1543 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1544 * @dev: target ATAPI device
1545 * @r_sense_key: out parameter for sense_key
1546 *
1547 * Perform ATAPI TEST_UNIT_READY.
1548 *
1549 * LOCKING:
1550 * EH context (may sleep).
1551 *
1552 * RETURNS:
1553 * 0 on success, AC_ERR_* mask on failure.
1554 */
1555static unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1556{
1557 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1558 struct ata_taskfile tf;
1559 unsigned int err_mask;
1560
1561 ata_tf_init(dev, &tf);
1562
1563 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1564 tf.command = ATA_CMD_PACKET;
1565 tf.protocol = ATAPI_PROT_NODATA;
1566
1567 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1568 if (err_mask == AC_ERR_DEV)
1569 *r_sense_key = tf.feature >> 4;
1570 return err_mask;
1571}
1572
1573/**
1574 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1575 * @dev: device to perform REQUEST_SENSE to
1576 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1577 * @dfl_sense_key: default sense key to use
1578 *
1579 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1580 * SENSE. This function is EH helper.
1581 *
1582 * LOCKING:
1583 * Kernel thread context (may sleep).
1584 *
1585 * RETURNS:
1586 * 0 on success, AC_ERR_* mask on failure
1587 */
1588static unsigned int atapi_eh_request_sense(struct ata_device *dev,
1589 u8 *sense_buf, u8 dfl_sense_key)
1590{
1591 u8 cdb[ATAPI_CDB_LEN] =
1592 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1593 struct ata_port *ap = dev->link->ap;
1594 struct ata_taskfile tf;
1595
1596 DPRINTK("ATAPI request sense\n");
1597
1598 /* FIXME: is this needed? */
1599 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1600
1601 /* initialize sense_buf with the error register,
1602 * for the case where they are -not- overwritten
1603 */
1604 sense_buf[0] = 0x70;
1605 sense_buf[2] = dfl_sense_key;
1606
1607 /* some devices time out if garbage left in tf */
1608 ata_tf_init(dev, &tf);
1609
1610 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1611 tf.command = ATA_CMD_PACKET;
1612
1613 /* is it pointless to prefer PIO for "safety reasons"? */
1614 if (ap->flags & ATA_FLAG_PIO_DMA) {
1615 tf.protocol = ATAPI_PROT_DMA;
1616 tf.feature |= ATAPI_PKT_DMA;
1617 } else {
1618 tf.protocol = ATAPI_PROT_PIO;
1619 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1620 tf.lbah = 0;
1621 }
1622
1623 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1624 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1625}
1626
1627/**
1628 * ata_eh_analyze_serror - analyze SError for a failed port
1629 * @link: ATA link to analyze SError for
1630 *
1631 * Analyze SError if available and further determine cause of
1632 * failure.
1633 *
1634 * LOCKING:
1635 * None.
1636 */
1637static void ata_eh_analyze_serror(struct ata_link *link)
1638{
1639 struct ata_eh_context *ehc = &link->eh_context;
1640 u32 serror = ehc->i.serror;
1641 unsigned int err_mask = 0, action = 0;
1642 u32 hotplug_mask;
1643
1644 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1645 err_mask |= AC_ERR_ATA_BUS;
1646 action |= ATA_EH_RESET;
1647 }
1648 if (serror & SERR_PROTOCOL) {
1649 err_mask |= AC_ERR_HSM;
1650 action |= ATA_EH_RESET;
1651 }
1652 if (serror & SERR_INTERNAL) {
1653 err_mask |= AC_ERR_SYSTEM;
1654 action |= ATA_EH_RESET;
1655 }
1656
1657 /* Determine whether a hotplug event has occurred. Both
1658 * SError.N/X are considered hotplug events for enabled or
1659 * host links. For disabled PMP links, only N bit is
1660 * considered as X bit is left at 1 for link plugging.
1661 */
1662 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1663 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1664 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1665 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1666 else
1667 hotplug_mask = SERR_PHYRDY_CHG;
1668
1669 if (serror & hotplug_mask)
1670 ata_ehi_hotplugged(&ehc->i);
1671
1672 ehc->i.err_mask |= err_mask;
1673 ehc->i.action |= action;
1674}
1675
1676/**
1677 * ata_eh_analyze_ncq_error - analyze NCQ error
1678 * @link: ATA link to analyze NCQ error for
1679 *
1680 * Read log page 10h, determine the offending qc and acquire
1681 * error status TF. For NCQ device errors, all LLDDs have to do
1682 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1683 * care of the rest.
1684 *
1685 * LOCKING:
1686 * Kernel thread context (may sleep).
1687 */
1688void ata_eh_analyze_ncq_error(struct ata_link *link)
1689{
1690 struct ata_port *ap = link->ap;
1691 struct ata_eh_context *ehc = &link->eh_context;
1692 struct ata_device *dev = link->device;
1693 struct ata_queued_cmd *qc;
1694 struct ata_taskfile tf;
1695 int tag, rc;
1696
1697 /* if frozen, we can't do much */
1698 if (ap->pflags & ATA_PFLAG_FROZEN)
1699 return;
1700
1701 /* is it NCQ device error? */
1702 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1703 return;
1704
1705 /* has LLDD analyzed already? */
1706 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1707 qc = __ata_qc_from_tag(ap, tag);
1708
1709 if (!(qc->flags & ATA_QCFLAG_FAILED))
1710 continue;
1711
1712 if (qc->err_mask)
1713 return;
1714 }
1715
1716 /* okay, this error is ours */
1717 memset(&tf, 0, sizeof(tf));
1718 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1719 if (rc) {
1720 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1721 rc);
1722 return;
1723 }
1724
1725 if (!(link->sactive & (1 << tag))) {
1726 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1727 tag);
1728 return;
1729 }
1730
1731 /* we've got the perpetrator, condemn it */
1732 qc = __ata_qc_from_tag(ap, tag);
1733 memcpy(&qc->result_tf, &tf, sizeof(tf));
1734 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1735 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1736 ehc->i.err_mask &= ~AC_ERR_DEV;
1737}
1738
1739/**
1740 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1741 * @qc: qc to analyze
1742 * @tf: Taskfile registers to analyze
1743 *
1744 * Analyze taskfile of @qc and further determine cause of
1745 * failure. This function also requests ATAPI sense data if
1746 * available.
1747 *
1748 * LOCKING:
1749 * Kernel thread context (may sleep).
1750 *
1751 * RETURNS:
1752 * Determined recovery action
1753 */
1754static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1755 const struct ata_taskfile *tf)
1756{
1757 unsigned int tmp, action = 0;
1758 u8 stat = tf->command, err = tf->feature;
1759
1760 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1761 qc->err_mask |= AC_ERR_HSM;
1762 return ATA_EH_RESET;
1763 }
1764
1765 if (stat & (ATA_ERR | ATA_DF))
1766 qc->err_mask |= AC_ERR_DEV;
1767 else
1768 return 0;
1769
1770 switch (qc->dev->class) {
1771 case ATA_DEV_ATA:
1772 if (err & ATA_ICRC)
1773 qc->err_mask |= AC_ERR_ATA_BUS;
1774 if (err & ATA_UNC)
1775 qc->err_mask |= AC_ERR_MEDIA;
1776 if (err & ATA_IDNF)
1777 qc->err_mask |= AC_ERR_INVALID;
1778 break;
1779
1780 case ATA_DEV_ATAPI:
1781 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1782 tmp = atapi_eh_request_sense(qc->dev,
1783 qc->scsicmd->sense_buffer,
1784 qc->result_tf.feature >> 4);
1785 if (!tmp) {
1786 /* ATA_QCFLAG_SENSE_VALID is used to
1787 * tell atapi_qc_complete() that sense
1788 * data is already valid.
1789 *
1790 * TODO: interpret sense data and set
1791 * appropriate err_mask.
1792 */
1793 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1794 } else
1795 qc->err_mask |= tmp;
1796 }
1797 }
1798
1799 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1800 action |= ATA_EH_RESET;
1801
1802 return action;
1803}
1804
1805static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1806 int *xfer_ok)
1807{
1808 int base = 0;
1809
1810 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1811 *xfer_ok = 1;
1812
1813 if (!*xfer_ok)
1814 base = ATA_ECAT_DUBIOUS_NONE;
1815
1816 if (err_mask & AC_ERR_ATA_BUS)
1817 return base + ATA_ECAT_ATA_BUS;
1818
1819 if (err_mask & AC_ERR_TIMEOUT)
1820 return base + ATA_ECAT_TOUT_HSM;
1821
1822 if (eflags & ATA_EFLAG_IS_IO) {
1823 if (err_mask & AC_ERR_HSM)
1824 return base + ATA_ECAT_TOUT_HSM;
1825 if ((err_mask &
1826 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1827 return base + ATA_ECAT_UNK_DEV;
1828 }
1829
1830 return 0;
1831}
1832
1833struct speed_down_verdict_arg {
1834 u64 since;
1835 int xfer_ok;
1836 int nr_errors[ATA_ECAT_NR];
1837};
1838
1839static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1840{
1841 struct speed_down_verdict_arg *arg = void_arg;
1842 int cat;
1843
1844 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1845 return -1;
1846
1847 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1848 &arg->xfer_ok);
1849 arg->nr_errors[cat]++;
1850
1851 return 0;
1852}
1853
1854/**
1855 * ata_eh_speed_down_verdict - Determine speed down verdict
1856 * @dev: Device of interest
1857 *
1858 * This function examines error ring of @dev and determines
1859 * whether NCQ needs to be turned off, transfer speed should be
1860 * stepped down, or falling back to PIO is necessary.
1861 *
1862 * ECAT_ATA_BUS : ATA_BUS error for any command
1863 *
1864 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1865 * IO commands
1866 *
1867 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1868 *
1869 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1870 * data transfer hasn't been verified.
1871 *
1872 * Verdicts are
1873 *
1874 * NCQ_OFF : Turn off NCQ.
1875 *
1876 * SPEED_DOWN : Speed down transfer speed but don't fall back
1877 * to PIO.
1878 *
1879 * FALLBACK_TO_PIO : Fall back to PIO.
1880 *
1881 * Even if multiple verdicts are returned, only one action is
1882 * taken per error. An action triggered by non-DUBIOUS errors
1883 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1884 * This is to expedite speed down decisions right after device is
1885 * initially configured.
1886 *
1887 * The followings are speed down rules. #1 and #2 deal with
1888 * DUBIOUS errors.
1889 *
1890 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1891 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1892 *
1893 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1894 * occurred during last 5 mins, NCQ_OFF.
1895 *
1896 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1897 * occurred during last 5 mins, FALLBACK_TO_PIO
1898 *
1899 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1900 * during last 10 mins, NCQ_OFF.
1901 *
1902 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1903 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1904 *
1905 * LOCKING:
1906 * Inherited from caller.
1907 *
1908 * RETURNS:
1909 * OR of ATA_EH_SPDN_* flags.
1910 */
1911static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1912{
1913 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1914 u64 j64 = get_jiffies_64();
1915 struct speed_down_verdict_arg arg;
1916 unsigned int verdict = 0;
1917
1918 /* scan past 5 mins of error history */
1919 memset(&arg, 0, sizeof(arg));
1920 arg.since = j64 - min(j64, j5mins);
1921 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1922
1923 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1924 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1925 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1926 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1927
1928 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1929 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1930 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1931
1932 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1933 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1934 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1935 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1936
1937 /* scan past 10 mins of error history */
1938 memset(&arg, 0, sizeof(arg));
1939 arg.since = j64 - min(j64, j10mins);
1940 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1941
1942 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1943 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1944 verdict |= ATA_EH_SPDN_NCQ_OFF;
1945
1946 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1947 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1948 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1949 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1950
1951 return verdict;
1952}
1953
1954/**
1955 * ata_eh_speed_down - record error and speed down if necessary
1956 * @dev: Failed device
1957 * @eflags: mask of ATA_EFLAG_* flags
1958 * @err_mask: err_mask of the error
1959 *
1960 * Record error and examine error history to determine whether
1961 * adjusting transmission speed is necessary. It also sets
1962 * transmission limits appropriately if such adjustment is
1963 * necessary.
1964 *
1965 * LOCKING:
1966 * Kernel thread context (may sleep).
1967 *
1968 * RETURNS:
1969 * Determined recovery action.
1970 */
1971static unsigned int ata_eh_speed_down(struct ata_device *dev,
1972 unsigned int eflags, unsigned int err_mask)
1973{
1974 struct ata_link *link = ata_dev_phys_link(dev);
1975 int xfer_ok = 0;
1976 unsigned int verdict;
1977 unsigned int action = 0;
1978
1979 /* don't bother if Cat-0 error */
1980 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1981 return 0;
1982
1983 /* record error and determine whether speed down is necessary */
1984 ata_ering_record(&dev->ering, eflags, err_mask);
1985 verdict = ata_eh_speed_down_verdict(dev);
1986
1987 /* turn off NCQ? */
1988 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1989 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1990 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1991 dev->flags |= ATA_DFLAG_NCQ_OFF;
1992 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1993 goto done;
1994 }
1995
1996 /* speed down? */
1997 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1998 /* speed down SATA link speed if possible */
1999 if (sata_down_spd_limit(link, 0) == 0) {
2000 action |= ATA_EH_RESET;
2001 goto done;
2002 }
2003
2004 /* lower transfer mode */
2005 if (dev->spdn_cnt < 2) {
2006 static const int dma_dnxfer_sel[] =
2007 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2008 static const int pio_dnxfer_sel[] =
2009 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2010 int sel;
2011
2012 if (dev->xfer_shift != ATA_SHIFT_PIO)
2013 sel = dma_dnxfer_sel[dev->spdn_cnt];
2014 else
2015 sel = pio_dnxfer_sel[dev->spdn_cnt];
2016
2017 dev->spdn_cnt++;
2018
2019 if (ata_down_xfermask_limit(dev, sel) == 0) {
2020 action |= ATA_EH_RESET;
2021 goto done;
2022 }
2023 }
2024 }
2025
2026 /* Fall back to PIO? Slowing down to PIO is meaningless for
2027 * SATA ATA devices. Consider it only for PATA and SATAPI.
2028 */
2029 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2030 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2031 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2032 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2033 dev->spdn_cnt = 0;
2034 action |= ATA_EH_RESET;
2035 goto done;
2036 }
2037 }
2038
2039 return 0;
2040 done:
2041 /* device has been slowed down, blow error history */
2042 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2043 ata_ering_clear(&dev->ering);
2044 return action;
2045}
2046
2047/**
2048 * ata_eh_link_autopsy - analyze error and determine recovery action
2049 * @link: host link to perform autopsy on
2050 *
2051 * Analyze why @link failed and determine which recovery actions
2052 * are needed. This function also sets more detailed AC_ERR_*
2053 * values and fills sense data for ATAPI CHECK SENSE.
2054 *
2055 * LOCKING:
2056 * Kernel thread context (may sleep).
2057 */
2058static void ata_eh_link_autopsy(struct ata_link *link)
2059{
2060 struct ata_port *ap = link->ap;
2061 struct ata_eh_context *ehc = &link->eh_context;
2062 struct ata_device *dev;
2063 unsigned int all_err_mask = 0, eflags = 0;
2064 int tag;
2065 u32 serror;
2066 int rc;
2067
2068 DPRINTK("ENTER\n");
2069
2070 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2071 return;
2072
2073 /* obtain and analyze SError */
2074 rc = sata_scr_read(link, SCR_ERROR, &serror);
2075 if (rc == 0) {
2076 ehc->i.serror |= serror;
2077 ata_eh_analyze_serror(link);
2078 } else if (rc != -EOPNOTSUPP) {
2079 /* SError read failed, force reset and probing */
2080 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2081 ehc->i.action |= ATA_EH_RESET;
2082 ehc->i.err_mask |= AC_ERR_OTHER;
2083 }
2084
2085 /* analyze NCQ failure */
2086 ata_eh_analyze_ncq_error(link);
2087
2088 /* any real error trumps AC_ERR_OTHER */
2089 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2090 ehc->i.err_mask &= ~AC_ERR_OTHER;
2091
2092 all_err_mask |= ehc->i.err_mask;
2093
2094 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2095 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2096
2097 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2098 ata_dev_phys_link(qc->dev) != link)
2099 continue;
2100
2101 /* inherit upper level err_mask */
2102 qc->err_mask |= ehc->i.err_mask;
2103
2104 /* analyze TF */
2105 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2106
2107 /* DEV errors are probably spurious in case of ATA_BUS error */
2108 if (qc->err_mask & AC_ERR_ATA_BUS)
2109 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2110 AC_ERR_INVALID);
2111
2112 /* any real error trumps unknown error */
2113 if (qc->err_mask & ~AC_ERR_OTHER)
2114 qc->err_mask &= ~AC_ERR_OTHER;
2115
2116 /* SENSE_VALID trumps dev/unknown error and revalidation */
2117 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2118 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2119
2120 /* determine whether the command is worth retrying */
2121 if (qc->flags & ATA_QCFLAG_IO ||
2122 (!(qc->err_mask & AC_ERR_INVALID) &&
2123 qc->err_mask != AC_ERR_DEV))
2124 qc->flags |= ATA_QCFLAG_RETRY;
2125
2126 /* accumulate error info */
2127 ehc->i.dev = qc->dev;
2128 all_err_mask |= qc->err_mask;
2129 if (qc->flags & ATA_QCFLAG_IO)
2130 eflags |= ATA_EFLAG_IS_IO;
2131 }
2132
2133 /* enforce default EH actions */
2134 if (ap->pflags & ATA_PFLAG_FROZEN ||
2135 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2136 ehc->i.action |= ATA_EH_RESET;
2137 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2138 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2139 ehc->i.action |= ATA_EH_REVALIDATE;
2140
2141 /* If we have offending qcs and the associated failed device,
2142 * perform per-dev EH action only on the offending device.
2143 */
2144 if (ehc->i.dev) {
2145 ehc->i.dev_action[ehc->i.dev->devno] |=
2146 ehc->i.action & ATA_EH_PERDEV_MASK;
2147 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2148 }
2149
2150 /* propagate timeout to host link */
2151 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2152 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2153
2154 /* record error and consider speeding down */
2155 dev = ehc->i.dev;
2156 if (!dev && ((ata_link_max_devices(link) == 1 &&
2157 ata_dev_enabled(link->device))))
2158 dev = link->device;
2159
2160 if (dev) {
2161 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2162 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2163 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2164 }
2165
2166 DPRINTK("EXIT\n");
2167}
2168
2169/**
2170 * ata_eh_autopsy - analyze error and determine recovery action
2171 * @ap: host port to perform autopsy on
2172 *
2173 * Analyze all links of @ap and determine why they failed and
2174 * which recovery actions are needed.
2175 *
2176 * LOCKING:
2177 * Kernel thread context (may sleep).
2178 */
2179void ata_eh_autopsy(struct ata_port *ap)
2180{
2181 struct ata_link *link;
2182
2183 ata_for_each_link(link, ap, EDGE)
2184 ata_eh_link_autopsy(link);
2185
2186 /* Handle the frigging slave link. Autopsy is done similarly
2187 * but actions and flags are transferred over to the master
2188 * link and handled from there.
2189 */
2190 if (ap->slave_link) {
2191 struct ata_eh_context *mehc = &ap->link.eh_context;
2192 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2193
2194 /* transfer control flags from master to slave */
2195 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2196
2197 /* perform autopsy on the slave link */
2198 ata_eh_link_autopsy(ap->slave_link);
2199
2200 /* transfer actions from slave to master and clear slave */
2201 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2202 mehc->i.action |= sehc->i.action;
2203 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2204 mehc->i.flags |= sehc->i.flags;
2205 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2206 }
2207
2208 /* Autopsy of fanout ports can affect host link autopsy.
2209 * Perform host link autopsy last.
2210 */
2211 if (sata_pmp_attached(ap))
2212 ata_eh_link_autopsy(&ap->link);
2213}
2214
2215/**
2216 * ata_get_cmd_descript - get description for ATA command
2217 * @command: ATA command code to get description for
2218 *
2219 * Return a textual description of the given command, or NULL if the
2220 * command is not known.
2221 *
2222 * LOCKING:
2223 * None
2224 */
2225const char *ata_get_cmd_descript(u8 command)
2226{
2227#ifdef CONFIG_ATA_VERBOSE_ERROR
2228 static const struct
2229 {
2230 u8 command;
2231 const char *text;
2232 } cmd_descr[] = {
2233 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2234 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2235 { ATA_CMD_STANDBY, "STANDBY" },
2236 { ATA_CMD_IDLE, "IDLE" },
2237 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2238 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2239 { ATA_CMD_NOP, "NOP" },
2240 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2241 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2242 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2243 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2244 { ATA_CMD_SERVICE, "SERVICE" },
2245 { ATA_CMD_READ, "READ DMA" },
2246 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2247 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2248 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2249 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2250 { ATA_CMD_WRITE, "WRITE DMA" },
2251 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2252 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2253 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2254 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2255 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2256 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2257 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2258 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2259 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2260 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2261 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2262 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2263 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2264 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2265 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2266 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2267 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2268 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2269 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2270 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2271 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2272 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2273 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2274 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2275 { ATA_CMD_SLEEP, "SLEEP" },
2276 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2277 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2278 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2279 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2280 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2281 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2282 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2283 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2284 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2285 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2286 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2287 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2288 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2289 { ATA_CMD_PMP_READ, "READ BUFFER" },
2290 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2291 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2292 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2293 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2294 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2295 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2296 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2297 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2298 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2299 { ATA_CMD_SMART, "SMART" },
2300 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2301 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2302 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2303 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2304 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2305 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2306 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2307 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2308 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2309 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2310 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2311 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2312 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2313 { ATA_CMD_RESTORE, "RECALIBRATE" },
2314 { 0, NULL } /* terminate list */
2315 };
2316
2317 unsigned int i;
2318 for (i = 0; cmd_descr[i].text; i++)
2319 if (cmd_descr[i].command == command)
2320 return cmd_descr[i].text;
2321#endif
2322
2323 return NULL;
2324}
2325
2326/**
2327 * ata_eh_link_report - report error handling to user
2328 * @link: ATA link EH is going on
2329 *
2330 * Report EH to user.
2331 *
2332 * LOCKING:
2333 * None.
2334 */
2335static void ata_eh_link_report(struct ata_link *link)
2336{
2337 struct ata_port *ap = link->ap;
2338 struct ata_eh_context *ehc = &link->eh_context;
2339 const char *frozen, *desc;
2340 char tries_buf[6];
2341 int tag, nr_failed = 0;
2342
2343 if (ehc->i.flags & ATA_EHI_QUIET)
2344 return;
2345
2346 desc = NULL;
2347 if (ehc->i.desc[0] != '\0')
2348 desc = ehc->i.desc;
2349
2350 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2351 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2352
2353 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2354 ata_dev_phys_link(qc->dev) != link ||
2355 ((qc->flags & ATA_QCFLAG_QUIET) &&
2356 qc->err_mask == AC_ERR_DEV))
2357 continue;
2358 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2359 continue;
2360
2361 nr_failed++;
2362 }
2363
2364 if (!nr_failed && !ehc->i.err_mask)
2365 return;
2366
2367 frozen = "";
2368 if (ap->pflags & ATA_PFLAG_FROZEN)
2369 frozen = " frozen";
2370
2371 memset(tries_buf, 0, sizeof(tries_buf));
2372 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2373 snprintf(tries_buf, sizeof(tries_buf) - 1, " t%d",
2374 ap->eh_tries);
2375
2376 if (ehc->i.dev) {
2377 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2378 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2379 ehc->i.err_mask, link->sactive, ehc->i.serror,
2380 ehc->i.action, frozen, tries_buf);
2381 if (desc)
2382 ata_dev_err(ehc->i.dev, "%s\n", desc);
2383 } else {
2384 ata_link_err(link, "exception Emask 0x%x "
2385 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2386 ehc->i.err_mask, link->sactive, ehc->i.serror,
2387 ehc->i.action, frozen, tries_buf);
2388 if (desc)
2389 ata_link_err(link, "%s\n", desc);
2390 }
2391
2392#ifdef CONFIG_ATA_VERBOSE_ERROR
2393 if (ehc->i.serror)
2394 ata_link_err(link,
2395 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2396 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2397 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2398 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2399 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2400 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2401 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2402 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2403 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2404 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2405 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2406 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2407 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2408 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2409 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2410 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2411 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2412 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2413#endif
2414
2415 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2416 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2417 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2418 const u8 *cdb = qc->cdb;
2419 char data_buf[20] = "";
2420 char cdb_buf[70] = "";
2421
2422 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2423 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2424 continue;
2425
2426 if (qc->dma_dir != DMA_NONE) {
2427 static const char *dma_str[] = {
2428 [DMA_BIDIRECTIONAL] = "bidi",
2429 [DMA_TO_DEVICE] = "out",
2430 [DMA_FROM_DEVICE] = "in",
2431 };
2432 static const char *prot_str[] = {
2433 [ATA_PROT_PIO] = "pio",
2434 [ATA_PROT_DMA] = "dma",
2435 [ATA_PROT_NCQ] = "ncq",
2436 [ATAPI_PROT_PIO] = "pio",
2437 [ATAPI_PROT_DMA] = "dma",
2438 };
2439
2440 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2441 prot_str[qc->tf.protocol], qc->nbytes,
2442 dma_str[qc->dma_dir]);
2443 }
2444
2445 if (ata_is_atapi(qc->tf.protocol)) {
2446 if (qc->scsicmd)
2447 scsi_print_command(qc->scsicmd);
2448 else
2449 snprintf(cdb_buf, sizeof(cdb_buf),
2450 "cdb %02x %02x %02x %02x %02x %02x %02x %02x "
2451 "%02x %02x %02x %02x %02x %02x %02x %02x\n ",
2452 cdb[0], cdb[1], cdb[2], cdb[3],
2453 cdb[4], cdb[5], cdb[6], cdb[7],
2454 cdb[8], cdb[9], cdb[10], cdb[11],
2455 cdb[12], cdb[13], cdb[14], cdb[15]);
2456 } else {
2457 const char *descr = ata_get_cmd_descript(cmd->command);
2458 if (descr)
2459 ata_dev_err(qc->dev, "failed command: %s\n",
2460 descr);
2461 }
2462
2463 ata_dev_err(qc->dev,
2464 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2465 "tag %d%s\n %s"
2466 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2467 "Emask 0x%x (%s)%s\n",
2468 cmd->command, cmd->feature, cmd->nsect,
2469 cmd->lbal, cmd->lbam, cmd->lbah,
2470 cmd->hob_feature, cmd->hob_nsect,
2471 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2472 cmd->device, qc->tag, data_buf, cdb_buf,
2473 res->command, res->feature, res->nsect,
2474 res->lbal, res->lbam, res->lbah,
2475 res->hob_feature, res->hob_nsect,
2476 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2477 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2478 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2479
2480#ifdef CONFIG_ATA_VERBOSE_ERROR
2481 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2482 ATA_ERR)) {
2483 if (res->command & ATA_BUSY)
2484 ata_dev_err(qc->dev, "status: { Busy }\n");
2485 else
2486 ata_dev_err(qc->dev, "status: { %s%s%s%s}\n",
2487 res->command & ATA_DRDY ? "DRDY " : "",
2488 res->command & ATA_DF ? "DF " : "",
2489 res->command & ATA_DRQ ? "DRQ " : "",
2490 res->command & ATA_ERR ? "ERR " : "");
2491 }
2492
2493 if (cmd->command != ATA_CMD_PACKET &&
2494 (res->feature & (ATA_ICRC | ATA_UNC | ATA_IDNF |
2495 ATA_ABORTED)))
2496 ata_dev_err(qc->dev, "error: { %s%s%s%s}\n",
2497 res->feature & ATA_ICRC ? "ICRC " : "",
2498 res->feature & ATA_UNC ? "UNC " : "",
2499 res->feature & ATA_IDNF ? "IDNF " : "",
2500 res->feature & ATA_ABORTED ? "ABRT " : "");
2501#endif
2502 }
2503}
2504
2505/**
2506 * ata_eh_report - report error handling to user
2507 * @ap: ATA port to report EH about
2508 *
2509 * Report EH to user.
2510 *
2511 * LOCKING:
2512 * None.
2513 */
2514void ata_eh_report(struct ata_port *ap)
2515{
2516 struct ata_link *link;
2517
2518 ata_for_each_link(link, ap, HOST_FIRST)
2519 ata_eh_link_report(link);
2520}
2521
2522static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2523 unsigned int *classes, unsigned long deadline,
2524 bool clear_classes)
2525{
2526 struct ata_device *dev;
2527
2528 if (clear_classes)
2529 ata_for_each_dev(dev, link, ALL)
2530 classes[dev->devno] = ATA_DEV_UNKNOWN;
2531
2532 return reset(link, classes, deadline);
2533}
2534
2535static int ata_eh_followup_srst_needed(struct ata_link *link,
2536 int rc, const unsigned int *classes)
2537{
2538 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2539 return 0;
2540 if (rc == -EAGAIN)
2541 return 1;
2542 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2543 return 1;
2544 return 0;
2545}
2546
2547int ata_eh_reset(struct ata_link *link, int classify,
2548 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2549 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2550{
2551 struct ata_port *ap = link->ap;
2552 struct ata_link *slave = ap->slave_link;
2553 struct ata_eh_context *ehc = &link->eh_context;
2554 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2555 unsigned int *classes = ehc->classes;
2556 unsigned int lflags = link->flags;
2557 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2558 int max_tries = 0, try = 0;
2559 struct ata_link *failed_link;
2560 struct ata_device *dev;
2561 unsigned long deadline, now;
2562 ata_reset_fn_t reset;
2563 unsigned long flags;
2564 u32 sstatus;
2565 int nr_unknown, rc;
2566
2567 /*
2568 * Prepare to reset
2569 */
2570 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2571 max_tries++;
2572 if (link->flags & ATA_LFLAG_NO_HRST)
2573 hardreset = NULL;
2574 if (link->flags & ATA_LFLAG_NO_SRST)
2575 softreset = NULL;
2576
2577 /* make sure each reset attempt is at least COOL_DOWN apart */
2578 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2579 now = jiffies;
2580 WARN_ON(time_after(ehc->last_reset, now));
2581 deadline = ata_deadline(ehc->last_reset,
2582 ATA_EH_RESET_COOL_DOWN);
2583 if (time_before(now, deadline))
2584 schedule_timeout_uninterruptible(deadline - now);
2585 }
2586
2587 spin_lock_irqsave(ap->lock, flags);
2588 ap->pflags |= ATA_PFLAG_RESETTING;
2589 spin_unlock_irqrestore(ap->lock, flags);
2590
2591 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2592
2593 ata_for_each_dev(dev, link, ALL) {
2594 /* If we issue an SRST then an ATA drive (not ATAPI)
2595 * may change configuration and be in PIO0 timing. If
2596 * we do a hard reset (or are coming from power on)
2597 * this is true for ATA or ATAPI. Until we've set a
2598 * suitable controller mode we should not touch the
2599 * bus as we may be talking too fast.
2600 */
2601 dev->pio_mode = XFER_PIO_0;
2602
2603 /* If the controller has a pio mode setup function
2604 * then use it to set the chipset to rights. Don't
2605 * touch the DMA setup as that will be dealt with when
2606 * configuring devices.
2607 */
2608 if (ap->ops->set_piomode)
2609 ap->ops->set_piomode(ap, dev);
2610 }
2611
2612 /* prefer hardreset */
2613 reset = NULL;
2614 ehc->i.action &= ~ATA_EH_RESET;
2615 if (hardreset) {
2616 reset = hardreset;
2617 ehc->i.action |= ATA_EH_HARDRESET;
2618 } else if (softreset) {
2619 reset = softreset;
2620 ehc->i.action |= ATA_EH_SOFTRESET;
2621 }
2622
2623 if (prereset) {
2624 unsigned long deadline = ata_deadline(jiffies,
2625 ATA_EH_PRERESET_TIMEOUT);
2626
2627 if (slave) {
2628 sehc->i.action &= ~ATA_EH_RESET;
2629 sehc->i.action |= ehc->i.action;
2630 }
2631
2632 rc = prereset(link, deadline);
2633
2634 /* If present, do prereset on slave link too. Reset
2635 * is skipped iff both master and slave links report
2636 * -ENOENT or clear ATA_EH_RESET.
2637 */
2638 if (slave && (rc == 0 || rc == -ENOENT)) {
2639 int tmp;
2640
2641 tmp = prereset(slave, deadline);
2642 if (tmp != -ENOENT)
2643 rc = tmp;
2644
2645 ehc->i.action |= sehc->i.action;
2646 }
2647
2648 if (rc) {
2649 if (rc == -ENOENT) {
2650 ata_link_dbg(link, "port disabled--ignoring\n");
2651 ehc->i.action &= ~ATA_EH_RESET;
2652
2653 ata_for_each_dev(dev, link, ALL)
2654 classes[dev->devno] = ATA_DEV_NONE;
2655
2656 rc = 0;
2657 } else
2658 ata_link_err(link,
2659 "prereset failed (errno=%d)\n",
2660 rc);
2661 goto out;
2662 }
2663
2664 /* prereset() might have cleared ATA_EH_RESET. If so,
2665 * bang classes, thaw and return.
2666 */
2667 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2668 ata_for_each_dev(dev, link, ALL)
2669 classes[dev->devno] = ATA_DEV_NONE;
2670 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2671 ata_is_host_link(link))
2672 ata_eh_thaw_port(ap);
2673 rc = 0;
2674 goto out;
2675 }
2676 }
2677
2678 retry:
2679 /*
2680 * Perform reset
2681 */
2682 if (ata_is_host_link(link))
2683 ata_eh_freeze_port(ap);
2684
2685 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2686
2687 if (reset) {
2688 if (verbose)
2689 ata_link_info(link, "%s resetting link\n",
2690 reset == softreset ? "soft" : "hard");
2691
2692 /* mark that this EH session started with reset */
2693 ehc->last_reset = jiffies;
2694 if (reset == hardreset)
2695 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2696 else
2697 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2698
2699 rc = ata_do_reset(link, reset, classes, deadline, true);
2700 if (rc && rc != -EAGAIN) {
2701 failed_link = link;
2702 goto fail;
2703 }
2704
2705 /* hardreset slave link if existent */
2706 if (slave && reset == hardreset) {
2707 int tmp;
2708
2709 if (verbose)
2710 ata_link_info(slave, "hard resetting link\n");
2711
2712 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2713 tmp = ata_do_reset(slave, reset, classes, deadline,
2714 false);
2715 switch (tmp) {
2716 case -EAGAIN:
2717 rc = -EAGAIN;
2718 case 0:
2719 break;
2720 default:
2721 failed_link = slave;
2722 rc = tmp;
2723 goto fail;
2724 }
2725 }
2726
2727 /* perform follow-up SRST if necessary */
2728 if (reset == hardreset &&
2729 ata_eh_followup_srst_needed(link, rc, classes)) {
2730 reset = softreset;
2731
2732 if (!reset) {
2733 ata_link_err(link,
2734 "follow-up softreset required but no softreset available\n");
2735 failed_link = link;
2736 rc = -EINVAL;
2737 goto fail;
2738 }
2739
2740 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2741 rc = ata_do_reset(link, reset, classes, deadline, true);
2742 if (rc) {
2743 failed_link = link;
2744 goto fail;
2745 }
2746 }
2747 } else {
2748 if (verbose)
2749 ata_link_info(link,
2750 "no reset method available, skipping reset\n");
2751 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2752 lflags |= ATA_LFLAG_ASSUME_ATA;
2753 }
2754
2755 /*
2756 * Post-reset processing
2757 */
2758 ata_for_each_dev(dev, link, ALL) {
2759 /* After the reset, the device state is PIO 0 and the
2760 * controller state is undefined. Reset also wakes up
2761 * drives from sleeping mode.
2762 */
2763 dev->pio_mode = XFER_PIO_0;
2764 dev->flags &= ~ATA_DFLAG_SLEEPING;
2765
2766 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2767 continue;
2768
2769 /* apply class override */
2770 if (lflags & ATA_LFLAG_ASSUME_ATA)
2771 classes[dev->devno] = ATA_DEV_ATA;
2772 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2773 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2774 }
2775
2776 /* record current link speed */
2777 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2778 link->sata_spd = (sstatus >> 4) & 0xf;
2779 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2780 slave->sata_spd = (sstatus >> 4) & 0xf;
2781
2782 /* thaw the port */
2783 if (ata_is_host_link(link))
2784 ata_eh_thaw_port(ap);
2785
2786 /* postreset() should clear hardware SError. Although SError
2787 * is cleared during link resume, clearing SError here is
2788 * necessary as some PHYs raise hotplug events after SRST.
2789 * This introduces race condition where hotplug occurs between
2790 * reset and here. This race is mediated by cross checking
2791 * link onlineness and classification result later.
2792 */
2793 if (postreset) {
2794 postreset(link, classes);
2795 if (slave)
2796 postreset(slave, classes);
2797 }
2798
2799 /*
2800 * Some controllers can't be frozen very well and may set spurious
2801 * error conditions during reset. Clear accumulated error
2802 * information and re-thaw the port if frozen. As reset is the
2803 * final recovery action and we cross check link onlineness against
2804 * device classification later, no hotplug event is lost by this.
2805 */
2806 spin_lock_irqsave(link->ap->lock, flags);
2807 memset(&link->eh_info, 0, sizeof(link->eh_info));
2808 if (slave)
2809 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2810 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2811 spin_unlock_irqrestore(link->ap->lock, flags);
2812
2813 if (ap->pflags & ATA_PFLAG_FROZEN)
2814 ata_eh_thaw_port(ap);
2815
2816 /*
2817 * Make sure onlineness and classification result correspond.
2818 * Hotplug could have happened during reset and some
2819 * controllers fail to wait while a drive is spinning up after
2820 * being hotplugged causing misdetection. By cross checking
2821 * link on/offlineness and classification result, those
2822 * conditions can be reliably detected and retried.
2823 */
2824 nr_unknown = 0;
2825 ata_for_each_dev(dev, link, ALL) {
2826 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2827 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2828 ata_dev_dbg(dev, "link online but device misclassified\n");
2829 classes[dev->devno] = ATA_DEV_NONE;
2830 nr_unknown++;
2831 }
2832 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2833 if (ata_class_enabled(classes[dev->devno]))
2834 ata_dev_dbg(dev,
2835 "link offline, clearing class %d to NONE\n",
2836 classes[dev->devno]);
2837 classes[dev->devno] = ATA_DEV_NONE;
2838 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2839 ata_dev_dbg(dev,
2840 "link status unknown, clearing UNKNOWN to NONE\n");
2841 classes[dev->devno] = ATA_DEV_NONE;
2842 }
2843 }
2844
2845 if (classify && nr_unknown) {
2846 if (try < max_tries) {
2847 ata_link_warn(link,
2848 "link online but %d devices misclassified, retrying\n",
2849 nr_unknown);
2850 failed_link = link;
2851 rc = -EAGAIN;
2852 goto fail;
2853 }
2854 ata_link_warn(link,
2855 "link online but %d devices misclassified, "
2856 "device detection might fail\n", nr_unknown);
2857 }
2858
2859 /* reset successful, schedule revalidation */
2860 ata_eh_done(link, NULL, ATA_EH_RESET);
2861 if (slave)
2862 ata_eh_done(slave, NULL, ATA_EH_RESET);
2863 ehc->last_reset = jiffies; /* update to completion time */
2864 ehc->i.action |= ATA_EH_REVALIDATE;
2865 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2866
2867 rc = 0;
2868 out:
2869 /* clear hotplug flag */
2870 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2871 if (slave)
2872 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2873
2874 spin_lock_irqsave(ap->lock, flags);
2875 ap->pflags &= ~ATA_PFLAG_RESETTING;
2876 spin_unlock_irqrestore(ap->lock, flags);
2877
2878 return rc;
2879
2880 fail:
2881 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2882 if (!ata_is_host_link(link) &&
2883 sata_scr_read(link, SCR_STATUS, &sstatus))
2884 rc = -ERESTART;
2885
2886 if (rc == -ERESTART || try >= max_tries) {
2887 /*
2888 * Thaw host port even if reset failed, so that the port
2889 * can be retried on the next phy event. This risks
2890 * repeated EH runs but seems to be a better tradeoff than
2891 * shutting down a port after a botched hotplug attempt.
2892 */
2893 if (ata_is_host_link(link))
2894 ata_eh_thaw_port(ap);
2895 goto out;
2896 }
2897
2898 now = jiffies;
2899 if (time_before(now, deadline)) {
2900 unsigned long delta = deadline - now;
2901
2902 ata_link_warn(failed_link,
2903 "reset failed (errno=%d), retrying in %u secs\n",
2904 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2905
2906 ata_eh_release(ap);
2907 while (delta)
2908 delta = schedule_timeout_uninterruptible(delta);
2909 ata_eh_acquire(ap);
2910 }
2911
2912 if (try == max_tries - 1) {
2913 sata_down_spd_limit(link, 0);
2914 if (slave)
2915 sata_down_spd_limit(slave, 0);
2916 } else if (rc == -EPIPE)
2917 sata_down_spd_limit(failed_link, 0);
2918
2919 if (hardreset)
2920 reset = hardreset;
2921 goto retry;
2922}
2923
2924static inline void ata_eh_pull_park_action(struct ata_port *ap)
2925{
2926 struct ata_link *link;
2927 struct ata_device *dev;
2928 unsigned long flags;
2929
2930 /*
2931 * This function can be thought of as an extended version of
2932 * ata_eh_about_to_do() specially crafted to accommodate the
2933 * requirements of ATA_EH_PARK handling. Since the EH thread
2934 * does not leave the do {} while () loop in ata_eh_recover as
2935 * long as the timeout for a park request to *one* device on
2936 * the port has not expired, and since we still want to pick
2937 * up park requests to other devices on the same port or
2938 * timeout updates for the same device, we have to pull
2939 * ATA_EH_PARK actions from eh_info into eh_context.i
2940 * ourselves at the beginning of each pass over the loop.
2941 *
2942 * Additionally, all write accesses to &ap->park_req_pending
2943 * through INIT_COMPLETION() (see below) or complete_all()
2944 * (see ata_scsi_park_store()) are protected by the host lock.
2945 * As a result we have that park_req_pending.done is zero on
2946 * exit from this function, i.e. when ATA_EH_PARK actions for
2947 * *all* devices on port ap have been pulled into the
2948 * respective eh_context structs. If, and only if,
2949 * park_req_pending.done is non-zero by the time we reach
2950 * wait_for_completion_timeout(), another ATA_EH_PARK action
2951 * has been scheduled for at least one of the devices on port
2952 * ap and we have to cycle over the do {} while () loop in
2953 * ata_eh_recover() again.
2954 */
2955
2956 spin_lock_irqsave(ap->lock, flags);
2957 INIT_COMPLETION(ap->park_req_pending);
2958 ata_for_each_link(link, ap, EDGE) {
2959 ata_for_each_dev(dev, link, ALL) {
2960 struct ata_eh_info *ehi = &link->eh_info;
2961
2962 link->eh_context.i.dev_action[dev->devno] |=
2963 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2964 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2965 }
2966 }
2967 spin_unlock_irqrestore(ap->lock, flags);
2968}
2969
2970static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2971{
2972 struct ata_eh_context *ehc = &dev->link->eh_context;
2973 struct ata_taskfile tf;
2974 unsigned int err_mask;
2975
2976 ata_tf_init(dev, &tf);
2977 if (park) {
2978 ehc->unloaded_mask |= 1 << dev->devno;
2979 tf.command = ATA_CMD_IDLEIMMEDIATE;
2980 tf.feature = 0x44;
2981 tf.lbal = 0x4c;
2982 tf.lbam = 0x4e;
2983 tf.lbah = 0x55;
2984 } else {
2985 ehc->unloaded_mask &= ~(1 << dev->devno);
2986 tf.command = ATA_CMD_CHK_POWER;
2987 }
2988
2989 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2990 tf.protocol |= ATA_PROT_NODATA;
2991 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2992 if (park && (err_mask || tf.lbal != 0xc4)) {
2993 ata_dev_err(dev, "head unload failed!\n");
2994 ehc->unloaded_mask &= ~(1 << dev->devno);
2995 }
2996}
2997
2998static int ata_eh_revalidate_and_attach(struct ata_link *link,
2999 struct ata_device **r_failed_dev)
3000{
3001 struct ata_port *ap = link->ap;
3002 struct ata_eh_context *ehc = &link->eh_context;
3003 struct ata_device *dev;
3004 unsigned int new_mask = 0;
3005 unsigned long flags;
3006 int rc = 0;
3007
3008 DPRINTK("ENTER\n");
3009
3010 /* For PATA drive side cable detection to work, IDENTIFY must
3011 * be done backwards such that PDIAG- is released by the slave
3012 * device before the master device is identified.
3013 */
3014 ata_for_each_dev(dev, link, ALL_REVERSE) {
3015 unsigned int action = ata_eh_dev_action(dev);
3016 unsigned int readid_flags = 0;
3017
3018 if (ehc->i.flags & ATA_EHI_DID_RESET)
3019 readid_flags |= ATA_READID_POSTRESET;
3020
3021 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3022 WARN_ON(dev->class == ATA_DEV_PMP);
3023
3024 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3025 rc = -EIO;
3026 goto err;
3027 }
3028
3029 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3030 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3031 readid_flags);
3032 if (rc)
3033 goto err;
3034
3035 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3036
3037 /* Configuration may have changed, reconfigure
3038 * transfer mode.
3039 */
3040 ehc->i.flags |= ATA_EHI_SETMODE;
3041
3042 /* schedule the scsi_rescan_device() here */
3043 schedule_work(&(ap->scsi_rescan_task));
3044 } else if (dev->class == ATA_DEV_UNKNOWN &&
3045 ehc->tries[dev->devno] &&
3046 ata_class_enabled(ehc->classes[dev->devno])) {
3047 /* Temporarily set dev->class, it will be
3048 * permanently set once all configurations are
3049 * complete. This is necessary because new
3050 * device configuration is done in two
3051 * separate loops.
3052 */
3053 dev->class = ehc->classes[dev->devno];
3054
3055 if (dev->class == ATA_DEV_PMP)
3056 rc = sata_pmp_attach(dev);
3057 else
3058 rc = ata_dev_read_id(dev, &dev->class,
3059 readid_flags, dev->id);
3060
3061 /* read_id might have changed class, store and reset */
3062 ehc->classes[dev->devno] = dev->class;
3063 dev->class = ATA_DEV_UNKNOWN;
3064
3065 switch (rc) {
3066 case 0:
3067 /* clear error info accumulated during probe */
3068 ata_ering_clear(&dev->ering);
3069 new_mask |= 1 << dev->devno;
3070 break;
3071 case -ENOENT:
3072 /* IDENTIFY was issued to non-existent
3073 * device. No need to reset. Just
3074 * thaw and ignore the device.
3075 */
3076 ata_eh_thaw_port(ap);
3077 break;
3078 default:
3079 goto err;
3080 }
3081 }
3082 }
3083
3084 /* PDIAG- should have been released, ask cable type if post-reset */
3085 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3086 if (ap->ops->cable_detect)
3087 ap->cbl = ap->ops->cable_detect(ap);
3088 ata_force_cbl(ap);
3089 }
3090
3091 /* Configure new devices forward such that user doesn't see
3092 * device detection messages backwards.
3093 */
3094 ata_for_each_dev(dev, link, ALL) {
3095 if (!(new_mask & (1 << dev->devno)))
3096 continue;
3097
3098 dev->class = ehc->classes[dev->devno];
3099
3100 if (dev->class == ATA_DEV_PMP)
3101 continue;
3102
3103 ehc->i.flags |= ATA_EHI_PRINTINFO;
3104 rc = ata_dev_configure(dev);
3105 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3106 if (rc) {
3107 dev->class = ATA_DEV_UNKNOWN;
3108 goto err;
3109 }
3110
3111 spin_lock_irqsave(ap->lock, flags);
3112 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3113 spin_unlock_irqrestore(ap->lock, flags);
3114
3115 /* new device discovered, configure xfermode */
3116 ehc->i.flags |= ATA_EHI_SETMODE;
3117 }
3118
3119 return 0;
3120
3121 err:
3122 *r_failed_dev = dev;
3123 DPRINTK("EXIT rc=%d\n", rc);
3124 return rc;
3125}
3126
3127/**
3128 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3129 * @link: link on which timings will be programmed
3130 * @r_failed_dev: out parameter for failed device
3131 *
3132 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3133 * ata_set_mode() fails, pointer to the failing device is
3134 * returned in @r_failed_dev.
3135 *
3136 * LOCKING:
3137 * PCI/etc. bus probe sem.
3138 *
3139 * RETURNS:
3140 * 0 on success, negative errno otherwise
3141 */
3142int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3143{
3144 struct ata_port *ap = link->ap;
3145 struct ata_device *dev;
3146 int rc;
3147
3148 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3149 ata_for_each_dev(dev, link, ENABLED) {
3150 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3151 struct ata_ering_entry *ent;
3152
3153 ent = ata_ering_top(&dev->ering);
3154 if (ent)
3155 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3156 }
3157 }
3158
3159 /* has private set_mode? */
3160 if (ap->ops->set_mode)
3161 rc = ap->ops->set_mode(link, r_failed_dev);
3162 else
3163 rc = ata_do_set_mode(link, r_failed_dev);
3164
3165 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3166 ata_for_each_dev(dev, link, ENABLED) {
3167 struct ata_eh_context *ehc = &link->eh_context;
3168 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3169 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3170
3171 if (dev->xfer_mode != saved_xfer_mode ||
3172 ata_ncq_enabled(dev) != saved_ncq)
3173 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3174 }
3175
3176 return rc;
3177}
3178
3179/**
3180 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3181 * @dev: ATAPI device to clear UA for
3182 *
3183 * Resets and other operations can make an ATAPI device raise
3184 * UNIT ATTENTION which causes the next operation to fail. This
3185 * function clears UA.
3186 *
3187 * LOCKING:
3188 * EH context (may sleep).
3189 *
3190 * RETURNS:
3191 * 0 on success, -errno on failure.
3192 */
3193static int atapi_eh_clear_ua(struct ata_device *dev)
3194{
3195 int i;
3196
3197 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3198 u8 *sense_buffer = dev->link->ap->sector_buf;
3199 u8 sense_key = 0;
3200 unsigned int err_mask;
3201
3202 err_mask = atapi_eh_tur(dev, &sense_key);
3203 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3204 ata_dev_warn(dev,
3205 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3206 err_mask);
3207 return -EIO;
3208 }
3209
3210 if (!err_mask || sense_key != UNIT_ATTENTION)
3211 return 0;
3212
3213 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3214 if (err_mask) {
3215 ata_dev_warn(dev, "failed to clear "
3216 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3217 return -EIO;
3218 }
3219 }
3220
3221 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3222 ATA_EH_UA_TRIES);
3223
3224 return 0;
3225}
3226
3227/**
3228 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3229 * @dev: ATA device which may need FLUSH retry
3230 *
3231 * If @dev failed FLUSH, it needs to be reported upper layer
3232 * immediately as it means that @dev failed to remap and already
3233 * lost at least a sector and further FLUSH retrials won't make
3234 * any difference to the lost sector. However, if FLUSH failed
3235 * for other reasons, for example transmission error, FLUSH needs
3236 * to be retried.
3237 *
3238 * This function determines whether FLUSH failure retry is
3239 * necessary and performs it if so.
3240 *
3241 * RETURNS:
3242 * 0 if EH can continue, -errno if EH needs to be repeated.
3243 */
3244static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3245{
3246 struct ata_link *link = dev->link;
3247 struct ata_port *ap = link->ap;
3248 struct ata_queued_cmd *qc;
3249 struct ata_taskfile tf;
3250 unsigned int err_mask;
3251 int rc = 0;
3252
3253 /* did flush fail for this device? */
3254 if (!ata_tag_valid(link->active_tag))
3255 return 0;
3256
3257 qc = __ata_qc_from_tag(ap, link->active_tag);
3258 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3259 qc->tf.command != ATA_CMD_FLUSH))
3260 return 0;
3261
3262 /* if the device failed it, it should be reported to upper layers */
3263 if (qc->err_mask & AC_ERR_DEV)
3264 return 0;
3265
3266 /* flush failed for some other reason, give it another shot */
3267 ata_tf_init(dev, &tf);
3268
3269 tf.command = qc->tf.command;
3270 tf.flags |= ATA_TFLAG_DEVICE;
3271 tf.protocol = ATA_PROT_NODATA;
3272
3273 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3274 tf.command, qc->err_mask);
3275
3276 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3277 if (!err_mask) {
3278 /*
3279 * FLUSH is complete but there's no way to
3280 * successfully complete a failed command from EH.
3281 * Making sure retry is allowed at least once and
3282 * retrying it should do the trick - whatever was in
3283 * the cache is already on the platter and this won't
3284 * cause infinite loop.
3285 */
3286 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3287 } else {
3288 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3289 err_mask);
3290 rc = -EIO;
3291
3292 /* if device failed it, report it to upper layers */
3293 if (err_mask & AC_ERR_DEV) {
3294 qc->err_mask |= AC_ERR_DEV;
3295 qc->result_tf = tf;
3296 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3297 rc = 0;
3298 }
3299 }
3300 return rc;
3301}
3302
3303/**
3304 * ata_eh_set_lpm - configure SATA interface power management
3305 * @link: link to configure power management
3306 * @policy: the link power management policy
3307 * @r_failed_dev: out parameter for failed device
3308 *
3309 * Enable SATA Interface power management. This will enable
3310 * Device Interface Power Management (DIPM) for min_power
3311 * policy, and then call driver specific callbacks for
3312 * enabling Host Initiated Power management.
3313 *
3314 * LOCKING:
3315 * EH context.
3316 *
3317 * RETURNS:
3318 * 0 on success, -errno on failure.
3319 */
3320static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3321 struct ata_device **r_failed_dev)
3322{
3323 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3324 struct ata_eh_context *ehc = &link->eh_context;
3325 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3326 enum ata_lpm_policy old_policy = link->lpm_policy;
3327 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3328 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3329 unsigned int err_mask;
3330 int rc;
3331
3332 /* if the link or host doesn't do LPM, noop */
3333 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3334 return 0;
3335
3336 /*
3337 * DIPM is enabled only for MIN_POWER as some devices
3338 * misbehave when the host NACKs transition to SLUMBER. Order
3339 * device and link configurations such that the host always
3340 * allows DIPM requests.
3341 */
3342 ata_for_each_dev(dev, link, ENABLED) {
3343 bool hipm = ata_id_has_hipm(dev->id);
3344 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3345
3346 /* find the first enabled and LPM enabled devices */
3347 if (!link_dev)
3348 link_dev = dev;
3349
3350 if (!lpm_dev && (hipm || dipm))
3351 lpm_dev = dev;
3352
3353 hints &= ~ATA_LPM_EMPTY;
3354 if (!hipm)
3355 hints &= ~ATA_LPM_HIPM;
3356
3357 /* disable DIPM before changing link config */
3358 if (policy != ATA_LPM_MIN_POWER && dipm) {
3359 err_mask = ata_dev_set_feature(dev,
3360 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3361 if (err_mask && err_mask != AC_ERR_DEV) {
3362 ata_dev_warn(dev,
3363 "failed to disable DIPM, Emask 0x%x\n",
3364 err_mask);
3365 rc = -EIO;
3366 goto fail;
3367 }
3368 }
3369 }
3370
3371 if (ap) {
3372 rc = ap->ops->set_lpm(link, policy, hints);
3373 if (!rc && ap->slave_link)
3374 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3375 } else
3376 rc = sata_pmp_set_lpm(link, policy, hints);
3377
3378 /*
3379 * Attribute link config failure to the first (LPM) enabled
3380 * device on the link.
3381 */
3382 if (rc) {
3383 if (rc == -EOPNOTSUPP) {
3384 link->flags |= ATA_LFLAG_NO_LPM;
3385 return 0;
3386 }
3387 dev = lpm_dev ? lpm_dev : link_dev;
3388 goto fail;
3389 }
3390
3391 /*
3392 * Low level driver acked the transition. Issue DIPM command
3393 * with the new policy set.
3394 */
3395 link->lpm_policy = policy;
3396 if (ap && ap->slave_link)
3397 ap->slave_link->lpm_policy = policy;
3398
3399 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3400 ata_for_each_dev(dev, link, ENABLED) {
3401 if (policy == ATA_LPM_MIN_POWER && !no_dipm &&
3402 ata_id_has_dipm(dev->id)) {
3403 err_mask = ata_dev_set_feature(dev,
3404 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3405 if (err_mask && err_mask != AC_ERR_DEV) {
3406 ata_dev_warn(dev,
3407 "failed to enable DIPM, Emask 0x%x\n",
3408 err_mask);
3409 rc = -EIO;
3410 goto fail;
3411 }
3412 }
3413 }
3414
3415 return 0;
3416
3417fail:
3418 /* restore the old policy */
3419 link->lpm_policy = old_policy;
3420 if (ap && ap->slave_link)
3421 ap->slave_link->lpm_policy = old_policy;
3422
3423 /* if no device or only one more chance is left, disable LPM */
3424 if (!dev || ehc->tries[dev->devno] <= 2) {
3425 ata_link_warn(link, "disabling LPM on the link\n");
3426 link->flags |= ATA_LFLAG_NO_LPM;
3427 }
3428 if (r_failed_dev)
3429 *r_failed_dev = dev;
3430 return rc;
3431}
3432
3433int ata_link_nr_enabled(struct ata_link *link)
3434{
3435 struct ata_device *dev;
3436 int cnt = 0;
3437
3438 ata_for_each_dev(dev, link, ENABLED)
3439 cnt++;
3440 return cnt;
3441}
3442
3443static int ata_link_nr_vacant(struct ata_link *link)
3444{
3445 struct ata_device *dev;
3446 int cnt = 0;
3447
3448 ata_for_each_dev(dev, link, ALL)
3449 if (dev->class == ATA_DEV_UNKNOWN)
3450 cnt++;
3451 return cnt;
3452}
3453
3454static int ata_eh_skip_recovery(struct ata_link *link)
3455{
3456 struct ata_port *ap = link->ap;
3457 struct ata_eh_context *ehc = &link->eh_context;
3458 struct ata_device *dev;
3459
3460 /* skip disabled links */
3461 if (link->flags & ATA_LFLAG_DISABLED)
3462 return 1;
3463
3464 /* skip if explicitly requested */
3465 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3466 return 1;
3467
3468 /* thaw frozen port and recover failed devices */
3469 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3470 return 0;
3471
3472 /* reset at least once if reset is requested */
3473 if ((ehc->i.action & ATA_EH_RESET) &&
3474 !(ehc->i.flags & ATA_EHI_DID_RESET))
3475 return 0;
3476
3477 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3478 ata_for_each_dev(dev, link, ALL) {
3479 if (dev->class == ATA_DEV_UNKNOWN &&
3480 ehc->classes[dev->devno] != ATA_DEV_NONE)
3481 return 0;
3482 }
3483
3484 return 1;
3485}
3486
3487static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3488{
3489 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3490 u64 now = get_jiffies_64();
3491 int *trials = void_arg;
3492
3493 if (ent->timestamp < now - min(now, interval))
3494 return -1;
3495
3496 (*trials)++;
3497 return 0;
3498}
3499
3500static int ata_eh_schedule_probe(struct ata_device *dev)
3501{
3502 struct ata_eh_context *ehc = &dev->link->eh_context;
3503 struct ata_link *link = ata_dev_phys_link(dev);
3504 int trials = 0;
3505
3506 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3507 (ehc->did_probe_mask & (1 << dev->devno)))
3508 return 0;
3509
3510 ata_eh_detach_dev(dev);
3511 ata_dev_init(dev);
3512 ehc->did_probe_mask |= (1 << dev->devno);
3513 ehc->i.action |= ATA_EH_RESET;
3514 ehc->saved_xfer_mode[dev->devno] = 0;
3515 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3516
3517 /* the link maybe in a deep sleep, wake it up */
3518 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3519 if (ata_is_host_link(link))
3520 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3521 ATA_LPM_EMPTY);
3522 else
3523 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3524 ATA_LPM_EMPTY);
3525 }
3526
3527 /* Record and count probe trials on the ering. The specific
3528 * error mask used is irrelevant. Because a successful device
3529 * detection clears the ering, this count accumulates only if
3530 * there are consecutive failed probes.
3531 *
3532 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3533 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3534 * forced to 1.5Gbps.
3535 *
3536 * This is to work around cases where failed link speed
3537 * negotiation results in device misdetection leading to
3538 * infinite DEVXCHG or PHRDY CHG events.
3539 */
3540 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3541 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3542
3543 if (trials > ATA_EH_PROBE_TRIALS)
3544 sata_down_spd_limit(link, 1);
3545
3546 return 1;
3547}
3548
3549static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3550{
3551 struct ata_eh_context *ehc = &dev->link->eh_context;
3552
3553 /* -EAGAIN from EH routine indicates retry without prejudice.
3554 * The requester is responsible for ensuring forward progress.
3555 */
3556 if (err != -EAGAIN)
3557 ehc->tries[dev->devno]--;
3558
3559 switch (err) {
3560 case -ENODEV:
3561 /* device missing or wrong IDENTIFY data, schedule probing */
3562 ehc->i.probe_mask |= (1 << dev->devno);
3563 case -EINVAL:
3564 /* give it just one more chance */
3565 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3566 case -EIO:
3567 if (ehc->tries[dev->devno] == 1) {
3568 /* This is the last chance, better to slow
3569 * down than lose it.
3570 */
3571 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3572 if (dev->pio_mode > XFER_PIO_0)
3573 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3574 }
3575 }
3576
3577 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3578 /* disable device if it has used up all its chances */
3579 ata_dev_disable(dev);
3580
3581 /* detach if offline */
3582 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3583 ata_eh_detach_dev(dev);
3584
3585 /* schedule probe if necessary */
3586 if (ata_eh_schedule_probe(dev)) {
3587 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3588 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3589 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3590 }
3591
3592 return 1;
3593 } else {
3594 ehc->i.action |= ATA_EH_RESET;
3595 return 0;
3596 }
3597}
3598
3599/**
3600 * ata_eh_recover - recover host port after error
3601 * @ap: host port to recover
3602 * @prereset: prereset method (can be NULL)
3603 * @softreset: softreset method (can be NULL)
3604 * @hardreset: hardreset method (can be NULL)
3605 * @postreset: postreset method (can be NULL)
3606 * @r_failed_link: out parameter for failed link
3607 *
3608 * This is the alpha and omega, eum and yang, heart and soul of
3609 * libata exception handling. On entry, actions required to
3610 * recover each link and hotplug requests are recorded in the
3611 * link's eh_context. This function executes all the operations
3612 * with appropriate retrials and fallbacks to resurrect failed
3613 * devices, detach goners and greet newcomers.
3614 *
3615 * LOCKING:
3616 * Kernel thread context (may sleep).
3617 *
3618 * RETURNS:
3619 * 0 on success, -errno on failure.
3620 */
3621int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3622 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3623 ata_postreset_fn_t postreset,
3624 struct ata_link **r_failed_link)
3625{
3626 struct ata_link *link;
3627 struct ata_device *dev;
3628 int rc, nr_fails;
3629 unsigned long flags, deadline;
3630
3631 DPRINTK("ENTER\n");
3632
3633 /* prep for recovery */
3634 ata_for_each_link(link, ap, EDGE) {
3635 struct ata_eh_context *ehc = &link->eh_context;
3636
3637 /* re-enable link? */
3638 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3639 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3640 spin_lock_irqsave(ap->lock, flags);
3641 link->flags &= ~ATA_LFLAG_DISABLED;
3642 spin_unlock_irqrestore(ap->lock, flags);
3643 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3644 }
3645
3646 ata_for_each_dev(dev, link, ALL) {
3647 if (link->flags & ATA_LFLAG_NO_RETRY)
3648 ehc->tries[dev->devno] = 1;
3649 else
3650 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3651
3652 /* collect port action mask recorded in dev actions */
3653 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3654 ~ATA_EH_PERDEV_MASK;
3655 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3656
3657 /* process hotplug request */
3658 if (dev->flags & ATA_DFLAG_DETACH)
3659 ata_eh_detach_dev(dev);
3660
3661 /* schedule probe if necessary */
3662 if (!ata_dev_enabled(dev))
3663 ata_eh_schedule_probe(dev);
3664 }
3665 }
3666
3667 retry:
3668 rc = 0;
3669
3670 /* if UNLOADING, finish immediately */
3671 if (ap->pflags & ATA_PFLAG_UNLOADING)
3672 goto out;
3673
3674 /* prep for EH */
3675 ata_for_each_link(link, ap, EDGE) {
3676 struct ata_eh_context *ehc = &link->eh_context;
3677
3678 /* skip EH if possible. */
3679 if (ata_eh_skip_recovery(link))
3680 ehc->i.action = 0;
3681
3682 ata_for_each_dev(dev, link, ALL)
3683 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3684 }
3685
3686 /* reset */
3687 ata_for_each_link(link, ap, EDGE) {
3688 struct ata_eh_context *ehc = &link->eh_context;
3689
3690 if (!(ehc->i.action & ATA_EH_RESET))
3691 continue;
3692
3693 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3694 prereset, softreset, hardreset, postreset);
3695 if (rc) {
3696 ata_link_err(link, "reset failed, giving up\n");
3697 goto out;
3698 }
3699 }
3700
3701 do {
3702 unsigned long now;
3703
3704 /*
3705 * clears ATA_EH_PARK in eh_info and resets
3706 * ap->park_req_pending
3707 */
3708 ata_eh_pull_park_action(ap);
3709
3710 deadline = jiffies;
3711 ata_for_each_link(link, ap, EDGE) {
3712 ata_for_each_dev(dev, link, ALL) {
3713 struct ata_eh_context *ehc = &link->eh_context;
3714 unsigned long tmp;
3715
3716 if (dev->class != ATA_DEV_ATA)
3717 continue;
3718 if (!(ehc->i.dev_action[dev->devno] &
3719 ATA_EH_PARK))
3720 continue;
3721 tmp = dev->unpark_deadline;
3722 if (time_before(deadline, tmp))
3723 deadline = tmp;
3724 else if (time_before_eq(tmp, jiffies))
3725 continue;
3726 if (ehc->unloaded_mask & (1 << dev->devno))
3727 continue;
3728
3729 ata_eh_park_issue_cmd(dev, 1);
3730 }
3731 }
3732
3733 now = jiffies;
3734 if (time_before_eq(deadline, now))
3735 break;
3736
3737 ata_eh_release(ap);
3738 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3739 deadline - now);
3740 ata_eh_acquire(ap);
3741 } while (deadline);
3742 ata_for_each_link(link, ap, EDGE) {
3743 ata_for_each_dev(dev, link, ALL) {
3744 if (!(link->eh_context.unloaded_mask &
3745 (1 << dev->devno)))
3746 continue;
3747
3748 ata_eh_park_issue_cmd(dev, 0);
3749 ata_eh_done(link, dev, ATA_EH_PARK);
3750 }
3751 }
3752
3753 /* the rest */
3754 nr_fails = 0;
3755 ata_for_each_link(link, ap, PMP_FIRST) {
3756 struct ata_eh_context *ehc = &link->eh_context;
3757
3758 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3759 goto config_lpm;
3760
3761 /* revalidate existing devices and attach new ones */
3762 rc = ata_eh_revalidate_and_attach(link, &dev);
3763 if (rc)
3764 goto rest_fail;
3765
3766 /* if PMP got attached, return, pmp EH will take care of it */
3767 if (link->device->class == ATA_DEV_PMP) {
3768 ehc->i.action = 0;
3769 return 0;
3770 }
3771
3772 /* configure transfer mode if necessary */
3773 if (ehc->i.flags & ATA_EHI_SETMODE) {
3774 rc = ata_set_mode(link, &dev);
3775 if (rc)
3776 goto rest_fail;
3777 ehc->i.flags &= ~ATA_EHI_SETMODE;
3778 }
3779
3780 /* If reset has been issued, clear UA to avoid
3781 * disrupting the current users of the device.
3782 */
3783 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3784 ata_for_each_dev(dev, link, ALL) {
3785 if (dev->class != ATA_DEV_ATAPI)
3786 continue;
3787 rc = atapi_eh_clear_ua(dev);
3788 if (rc)
3789 goto rest_fail;
3790 }
3791 }
3792
3793 /* retry flush if necessary */
3794 ata_for_each_dev(dev, link, ALL) {
3795 if (dev->class != ATA_DEV_ATA)
3796 continue;
3797 rc = ata_eh_maybe_retry_flush(dev);
3798 if (rc)
3799 goto rest_fail;
3800 }
3801
3802 config_lpm:
3803 /* configure link power saving */
3804 if (link->lpm_policy != ap->target_lpm_policy) {
3805 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3806 if (rc)
3807 goto rest_fail;
3808 }
3809
3810 /* this link is okay now */
3811 ehc->i.flags = 0;
3812 continue;
3813
3814 rest_fail:
3815 nr_fails++;
3816 if (dev)
3817 ata_eh_handle_dev_fail(dev, rc);
3818
3819 if (ap->pflags & ATA_PFLAG_FROZEN) {
3820 /* PMP reset requires working host port.
3821 * Can't retry if it's frozen.
3822 */
3823 if (sata_pmp_attached(ap))
3824 goto out;
3825 break;
3826 }
3827 }
3828
3829 if (nr_fails)
3830 goto retry;
3831
3832 out:
3833 if (rc && r_failed_link)
3834 *r_failed_link = link;
3835
3836 DPRINTK("EXIT, rc=%d\n", rc);
3837 return rc;
3838}
3839
3840/**
3841 * ata_eh_finish - finish up EH
3842 * @ap: host port to finish EH for
3843 *
3844 * Recovery is complete. Clean up EH states and retry or finish
3845 * failed qcs.
3846 *
3847 * LOCKING:
3848 * None.
3849 */
3850void ata_eh_finish(struct ata_port *ap)
3851{
3852 int tag;
3853
3854 /* retry or finish qcs */
3855 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
3856 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
3857
3858 if (!(qc->flags & ATA_QCFLAG_FAILED))
3859 continue;
3860
3861 if (qc->err_mask) {
3862 /* FIXME: Once EH migration is complete,
3863 * generate sense data in this function,
3864 * considering both err_mask and tf.
3865 */
3866 if (qc->flags & ATA_QCFLAG_RETRY)
3867 ata_eh_qc_retry(qc);
3868 else
3869 ata_eh_qc_complete(qc);
3870 } else {
3871 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3872 ata_eh_qc_complete(qc);
3873 } else {
3874 /* feed zero TF to sense generation */
3875 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3876 ata_eh_qc_retry(qc);
3877 }
3878 }
3879 }
3880
3881 /* make sure nr_active_links is zero after EH */
3882 WARN_ON(ap->nr_active_links);
3883 ap->nr_active_links = 0;
3884}
3885
3886/**
3887 * ata_do_eh - do standard error handling
3888 * @ap: host port to handle error for
3889 *
3890 * @prereset: prereset method (can be NULL)
3891 * @softreset: softreset method (can be NULL)
3892 * @hardreset: hardreset method (can be NULL)
3893 * @postreset: postreset method (can be NULL)
3894 *
3895 * Perform standard error handling sequence.
3896 *
3897 * LOCKING:
3898 * Kernel thread context (may sleep).
3899 */
3900void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3901 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3902 ata_postreset_fn_t postreset)
3903{
3904 struct ata_device *dev;
3905 int rc;
3906
3907 ata_eh_autopsy(ap);
3908 ata_eh_report(ap);
3909
3910 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3911 NULL);
3912 if (rc) {
3913 ata_for_each_dev(dev, &ap->link, ALL)
3914 ata_dev_disable(dev);
3915 }
3916
3917 ata_eh_finish(ap);
3918}
3919
3920/**
3921 * ata_std_error_handler - standard error handler
3922 * @ap: host port to handle error for
3923 *
3924 * Standard error handler
3925 *
3926 * LOCKING:
3927 * Kernel thread context (may sleep).
3928 */
3929void ata_std_error_handler(struct ata_port *ap)
3930{
3931 struct ata_port_operations *ops = ap->ops;
3932 ata_reset_fn_t hardreset = ops->hardreset;
3933
3934 /* ignore built-in hardreset if SCR access is not available */
3935 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3936 hardreset = NULL;
3937
3938 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3939}
3940
3941#ifdef CONFIG_PM
3942/**
3943 * ata_eh_handle_port_suspend - perform port suspend operation
3944 * @ap: port to suspend
3945 *
3946 * Suspend @ap.
3947 *
3948 * LOCKING:
3949 * Kernel thread context (may sleep).
3950 */
3951static void ata_eh_handle_port_suspend(struct ata_port *ap)
3952{
3953 unsigned long flags;
3954 int rc = 0;
3955
3956 /* are we suspending? */
3957 spin_lock_irqsave(ap->lock, flags);
3958 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3959 ap->pm_mesg.event == PM_EVENT_ON) {
3960 spin_unlock_irqrestore(ap->lock, flags);
3961 return;
3962 }
3963 spin_unlock_irqrestore(ap->lock, flags);
3964
3965 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3966
3967 /* tell ACPI we're suspending */
3968 rc = ata_acpi_on_suspend(ap);
3969 if (rc)
3970 goto out;
3971
3972 /* suspend */
3973 ata_eh_freeze_port(ap);
3974
3975 if (ap->ops->port_suspend)
3976 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3977
3978 ata_acpi_set_state(ap, PMSG_SUSPEND);
3979 out:
3980 /* report result */
3981 spin_lock_irqsave(ap->lock, flags);
3982
3983 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3984 if (rc == 0)
3985 ap->pflags |= ATA_PFLAG_SUSPENDED;
3986 else if (ap->pflags & ATA_PFLAG_FROZEN)
3987 ata_port_schedule_eh(ap);
3988
3989 if (ap->pm_result) {
3990 *ap->pm_result = rc;
3991 ap->pm_result = NULL;
3992 }
3993
3994 spin_unlock_irqrestore(ap->lock, flags);
3995
3996 return;
3997}
3998
3999/**
4000 * ata_eh_handle_port_resume - perform port resume operation
4001 * @ap: port to resume
4002 *
4003 * Resume @ap.
4004 *
4005 * LOCKING:
4006 * Kernel thread context (may sleep).
4007 */
4008static void ata_eh_handle_port_resume(struct ata_port *ap)
4009{
4010 struct ata_link *link;
4011 struct ata_device *dev;
4012 unsigned long flags;
4013 int rc = 0;
4014
4015 /* are we resuming? */
4016 spin_lock_irqsave(ap->lock, flags);
4017 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4018 ap->pm_mesg.event != PM_EVENT_ON) {
4019 spin_unlock_irqrestore(ap->lock, flags);
4020 return;
4021 }
4022 spin_unlock_irqrestore(ap->lock, flags);
4023
4024 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4025
4026 /*
4027 * Error timestamps are in jiffies which doesn't run while
4028 * suspended and PHY events during resume isn't too uncommon.
4029 * When the two are combined, it can lead to unnecessary speed
4030 * downs if the machine is suspended and resumed repeatedly.
4031 * Clear error history.
4032 */
4033 ata_for_each_link(link, ap, HOST_FIRST)
4034 ata_for_each_dev(dev, link, ALL)
4035 ata_ering_clear(&dev->ering);
4036
4037 ata_acpi_set_state(ap, PMSG_ON);
4038
4039 if (ap->ops->port_resume)
4040 rc = ap->ops->port_resume(ap);
4041
4042 /* tell ACPI that we're resuming */
4043 ata_acpi_on_resume(ap);
4044
4045 /* report result */
4046 spin_lock_irqsave(ap->lock, flags);
4047 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4048 if (ap->pm_result) {
4049 *ap->pm_result = rc;
4050 ap->pm_result = NULL;
4051 }
4052 spin_unlock_irqrestore(ap->lock, flags);
4053}
4054#endif /* CONFIG_PM */
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * libata-eh.c - libata error handling
4 *
5 * Copyright 2006 Tejun Heo <htejun@gmail.com>
6 *
7 * libata documentation is available via 'make {ps|pdf}docs',
8 * as Documentation/driver-api/libata.rst
9 *
10 * Hardware documentation available from http://www.t13.org/ and
11 * http://www.sata-io.org/
12 */
13
14#include <linux/kernel.h>
15#include <linux/blkdev.h>
16#include <linux/export.h>
17#include <linux/pci.h>
18#include <scsi/scsi.h>
19#include <scsi/scsi_host.h>
20#include <scsi/scsi_eh.h>
21#include <scsi/scsi_device.h>
22#include <scsi/scsi_cmnd.h>
23#include <scsi/scsi_dbg.h>
24#include "../scsi/scsi_transport_api.h"
25
26#include <linux/libata.h>
27
28#include <trace/events/libata.h>
29#include "libata.h"
30
31enum {
32 /* speed down verdicts */
33 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
34 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
35 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
36 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
37
38 /* error flags */
39 ATA_EFLAG_IS_IO = (1 << 0),
40 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
41 ATA_EFLAG_OLD_ER = (1 << 31),
42
43 /* error categories */
44 ATA_ECAT_NONE = 0,
45 ATA_ECAT_ATA_BUS = 1,
46 ATA_ECAT_TOUT_HSM = 2,
47 ATA_ECAT_UNK_DEV = 3,
48 ATA_ECAT_DUBIOUS_NONE = 4,
49 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
50 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
51 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
52 ATA_ECAT_NR = 8,
53
54 ATA_EH_CMD_DFL_TIMEOUT = 5000,
55
56 /* always put at least this amount of time between resets */
57 ATA_EH_RESET_COOL_DOWN = 5000,
58
59 /* Waiting in ->prereset can never be reliable. It's
60 * sometimes nice to wait there but it can't be depended upon;
61 * otherwise, we wouldn't be resetting. Just give it enough
62 * time for most drives to spin up.
63 */
64 ATA_EH_PRERESET_TIMEOUT = 10000,
65 ATA_EH_FASTDRAIN_INTERVAL = 3000,
66
67 ATA_EH_UA_TRIES = 5,
68
69 /* probe speed down parameters, see ata_eh_schedule_probe() */
70 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
71 ATA_EH_PROBE_TRIALS = 2,
72};
73
74/* The following table determines how we sequence resets. Each entry
75 * represents timeout for that try. The first try can be soft or
76 * hardreset. All others are hardreset if available. In most cases
77 * the first reset w/ 10sec timeout should succeed. Following entries
78 * are mostly for error handling, hotplug and those outlier devices that
79 * take an exceptionally long time to recover from reset.
80 */
81static const unsigned long ata_eh_reset_timeouts[] = {
82 10000, /* most drives spin up by 10sec */
83 10000, /* > 99% working drives spin up before 20sec */
84 35000, /* give > 30 secs of idleness for outlier devices */
85 5000, /* and sweet one last chance */
86 ULONG_MAX, /* > 1 min has elapsed, give up */
87};
88
89static const unsigned int ata_eh_identify_timeouts[] = {
90 5000, /* covers > 99% of successes and not too boring on failures */
91 10000, /* combined time till here is enough even for media access */
92 30000, /* for true idiots */
93 UINT_MAX,
94};
95
96static const unsigned int ata_eh_revalidate_timeouts[] = {
97 15000, /* Some drives are slow to read log pages when waking-up */
98 15000, /* combined time till here is enough even for media access */
99 UINT_MAX,
100};
101
102static const unsigned int ata_eh_flush_timeouts[] = {
103 15000, /* be generous with flush */
104 15000, /* ditto */
105 30000, /* and even more generous */
106 UINT_MAX,
107};
108
109static const unsigned int ata_eh_other_timeouts[] = {
110 5000, /* same rationale as identify timeout */
111 10000, /* ditto */
112 /* but no merciful 30sec for other commands, it just isn't worth it */
113 UINT_MAX,
114};
115
116struct ata_eh_cmd_timeout_ent {
117 const u8 *commands;
118 const unsigned int *timeouts;
119};
120
121/* The following table determines timeouts to use for EH internal
122 * commands. Each table entry is a command class and matches the
123 * commands the entry applies to and the timeout table to use.
124 *
125 * On the retry after a command timed out, the next timeout value from
126 * the table is used. If the table doesn't contain further entries,
127 * the last value is used.
128 *
129 * ehc->cmd_timeout_idx keeps track of which timeout to use per
130 * command class, so if SET_FEATURES times out on the first try, the
131 * next try will use the second timeout value only for that class.
132 */
133#define CMDS(cmds...) (const u8 []){ cmds, 0 }
134static const struct ata_eh_cmd_timeout_ent
135ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
136 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
137 .timeouts = ata_eh_identify_timeouts, },
138 { .commands = CMDS(ATA_CMD_READ_LOG_EXT, ATA_CMD_READ_LOG_DMA_EXT),
139 .timeouts = ata_eh_revalidate_timeouts, },
140 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
141 .timeouts = ata_eh_other_timeouts, },
142 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
143 .timeouts = ata_eh_other_timeouts, },
144 { .commands = CMDS(ATA_CMD_SET_FEATURES),
145 .timeouts = ata_eh_other_timeouts, },
146 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
147 .timeouts = ata_eh_other_timeouts, },
148 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
149 .timeouts = ata_eh_flush_timeouts },
150};
151#undef CMDS
152
153static void __ata_port_freeze(struct ata_port *ap);
154static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
155 struct ata_device **r_failed_dev);
156#ifdef CONFIG_PM
157static void ata_eh_handle_port_suspend(struct ata_port *ap);
158static void ata_eh_handle_port_resume(struct ata_port *ap);
159#else /* CONFIG_PM */
160static void ata_eh_handle_port_suspend(struct ata_port *ap)
161{ }
162
163static void ata_eh_handle_port_resume(struct ata_port *ap)
164{ }
165#endif /* CONFIG_PM */
166
167static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
168 const char *fmt, va_list args)
169{
170 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
171 ATA_EH_DESC_LEN - ehi->desc_len,
172 fmt, args);
173}
174
175/**
176 * __ata_ehi_push_desc - push error description without adding separator
177 * @ehi: target EHI
178 * @fmt: printf format string
179 *
180 * Format string according to @fmt and append it to @ehi->desc.
181 *
182 * LOCKING:
183 * spin_lock_irqsave(host lock)
184 */
185void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
186{
187 va_list args;
188
189 va_start(args, fmt);
190 __ata_ehi_pushv_desc(ehi, fmt, args);
191 va_end(args);
192}
193EXPORT_SYMBOL_GPL(__ata_ehi_push_desc);
194
195/**
196 * ata_ehi_push_desc - push error description with separator
197 * @ehi: target EHI
198 * @fmt: printf format string
199 *
200 * Format string according to @fmt and append it to @ehi->desc.
201 * If @ehi->desc is not empty, ", " is added in-between.
202 *
203 * LOCKING:
204 * spin_lock_irqsave(host lock)
205 */
206void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
207{
208 va_list args;
209
210 if (ehi->desc_len)
211 __ata_ehi_push_desc(ehi, ", ");
212
213 va_start(args, fmt);
214 __ata_ehi_pushv_desc(ehi, fmt, args);
215 va_end(args);
216}
217EXPORT_SYMBOL_GPL(ata_ehi_push_desc);
218
219/**
220 * ata_ehi_clear_desc - clean error description
221 * @ehi: target EHI
222 *
223 * Clear @ehi->desc.
224 *
225 * LOCKING:
226 * spin_lock_irqsave(host lock)
227 */
228void ata_ehi_clear_desc(struct ata_eh_info *ehi)
229{
230 ehi->desc[0] = '\0';
231 ehi->desc_len = 0;
232}
233EXPORT_SYMBOL_GPL(ata_ehi_clear_desc);
234
235/**
236 * ata_port_desc - append port description
237 * @ap: target ATA port
238 * @fmt: printf format string
239 *
240 * Format string according to @fmt and append it to port
241 * description. If port description is not empty, " " is added
242 * in-between. This function is to be used while initializing
243 * ata_host. The description is printed on host registration.
244 *
245 * LOCKING:
246 * None.
247 */
248void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
249{
250 va_list args;
251
252 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
253
254 if (ap->link.eh_info.desc_len)
255 __ata_ehi_push_desc(&ap->link.eh_info, " ");
256
257 va_start(args, fmt);
258 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
259 va_end(args);
260}
261EXPORT_SYMBOL_GPL(ata_port_desc);
262
263#ifdef CONFIG_PCI
264/**
265 * ata_port_pbar_desc - append PCI BAR description
266 * @ap: target ATA port
267 * @bar: target PCI BAR
268 * @offset: offset into PCI BAR
269 * @name: name of the area
270 *
271 * If @offset is negative, this function formats a string which
272 * contains the name, address, size and type of the BAR and
273 * appends it to the port description. If @offset is zero or
274 * positive, only name and offsetted address is appended.
275 *
276 * LOCKING:
277 * None.
278 */
279void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
280 const char *name)
281{
282 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
283 char *type = "";
284 unsigned long long start, len;
285
286 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
287 type = "m";
288 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
289 type = "i";
290
291 start = (unsigned long long)pci_resource_start(pdev, bar);
292 len = (unsigned long long)pci_resource_len(pdev, bar);
293
294 if (offset < 0)
295 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
296 else
297 ata_port_desc(ap, "%s 0x%llx", name,
298 start + (unsigned long long)offset);
299}
300EXPORT_SYMBOL_GPL(ata_port_pbar_desc);
301#endif /* CONFIG_PCI */
302
303static int ata_lookup_timeout_table(u8 cmd)
304{
305 int i;
306
307 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
308 const u8 *cur;
309
310 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
311 if (*cur == cmd)
312 return i;
313 }
314
315 return -1;
316}
317
318/**
319 * ata_internal_cmd_timeout - determine timeout for an internal command
320 * @dev: target device
321 * @cmd: internal command to be issued
322 *
323 * Determine timeout for internal command @cmd for @dev.
324 *
325 * LOCKING:
326 * EH context.
327 *
328 * RETURNS:
329 * Determined timeout.
330 */
331unsigned int ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
332{
333 struct ata_eh_context *ehc = &dev->link->eh_context;
334 int ent = ata_lookup_timeout_table(cmd);
335 int idx;
336
337 if (ent < 0)
338 return ATA_EH_CMD_DFL_TIMEOUT;
339
340 idx = ehc->cmd_timeout_idx[dev->devno][ent];
341 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
342}
343
344/**
345 * ata_internal_cmd_timed_out - notification for internal command timeout
346 * @dev: target device
347 * @cmd: internal command which timed out
348 *
349 * Notify EH that internal command @cmd for @dev timed out. This
350 * function should be called only for commands whose timeouts are
351 * determined using ata_internal_cmd_timeout().
352 *
353 * LOCKING:
354 * EH context.
355 */
356void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
357{
358 struct ata_eh_context *ehc = &dev->link->eh_context;
359 int ent = ata_lookup_timeout_table(cmd);
360 int idx;
361
362 if (ent < 0)
363 return;
364
365 idx = ehc->cmd_timeout_idx[dev->devno][ent];
366 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != UINT_MAX)
367 ehc->cmd_timeout_idx[dev->devno][ent]++;
368}
369
370static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
371 unsigned int err_mask)
372{
373 struct ata_ering_entry *ent;
374
375 WARN_ON(!err_mask);
376
377 ering->cursor++;
378 ering->cursor %= ATA_ERING_SIZE;
379
380 ent = &ering->ring[ering->cursor];
381 ent->eflags = eflags;
382 ent->err_mask = err_mask;
383 ent->timestamp = get_jiffies_64();
384}
385
386static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
387{
388 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
389
390 if (ent->err_mask)
391 return ent;
392 return NULL;
393}
394
395int ata_ering_map(struct ata_ering *ering,
396 int (*map_fn)(struct ata_ering_entry *, void *),
397 void *arg)
398{
399 int idx, rc = 0;
400 struct ata_ering_entry *ent;
401
402 idx = ering->cursor;
403 do {
404 ent = &ering->ring[idx];
405 if (!ent->err_mask)
406 break;
407 rc = map_fn(ent, arg);
408 if (rc)
409 break;
410 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
411 } while (idx != ering->cursor);
412
413 return rc;
414}
415
416static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
417{
418 ent->eflags |= ATA_EFLAG_OLD_ER;
419 return 0;
420}
421
422static void ata_ering_clear(struct ata_ering *ering)
423{
424 ata_ering_map(ering, ata_ering_clear_cb, NULL);
425}
426
427static unsigned int ata_eh_dev_action(struct ata_device *dev)
428{
429 struct ata_eh_context *ehc = &dev->link->eh_context;
430
431 return ehc->i.action | ehc->i.dev_action[dev->devno];
432}
433
434static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
435 struct ata_eh_info *ehi, unsigned int action)
436{
437 struct ata_device *tdev;
438
439 if (!dev) {
440 ehi->action &= ~action;
441 ata_for_each_dev(tdev, link, ALL)
442 ehi->dev_action[tdev->devno] &= ~action;
443 } else {
444 /* doesn't make sense for port-wide EH actions */
445 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
446
447 /* break ehi->action into ehi->dev_action */
448 if (ehi->action & action) {
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] |=
451 ehi->action & action;
452 ehi->action &= ~action;
453 }
454
455 /* turn off the specified per-dev action */
456 ehi->dev_action[dev->devno] &= ~action;
457 }
458}
459
460/**
461 * ata_eh_acquire - acquire EH ownership
462 * @ap: ATA port to acquire EH ownership for
463 *
464 * Acquire EH ownership for @ap. This is the basic exclusion
465 * mechanism for ports sharing a host. Only one port hanging off
466 * the same host can claim the ownership of EH.
467 *
468 * LOCKING:
469 * EH context.
470 */
471void ata_eh_acquire(struct ata_port *ap)
472{
473 mutex_lock(&ap->host->eh_mutex);
474 WARN_ON_ONCE(ap->host->eh_owner);
475 ap->host->eh_owner = current;
476}
477
478/**
479 * ata_eh_release - release EH ownership
480 * @ap: ATA port to release EH ownership for
481 *
482 * Release EH ownership for @ap if the caller. The caller must
483 * have acquired EH ownership using ata_eh_acquire() previously.
484 *
485 * LOCKING:
486 * EH context.
487 */
488void ata_eh_release(struct ata_port *ap)
489{
490 WARN_ON_ONCE(ap->host->eh_owner != current);
491 ap->host->eh_owner = NULL;
492 mutex_unlock(&ap->host->eh_mutex);
493}
494
495static void ata_eh_unload(struct ata_port *ap)
496{
497 struct ata_link *link;
498 struct ata_device *dev;
499 unsigned long flags;
500
501 /* Restore SControl IPM and SPD for the next driver and
502 * disable attached devices.
503 */
504 ata_for_each_link(link, ap, PMP_FIRST) {
505 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
506 ata_for_each_dev(dev, link, ALL)
507 ata_dev_disable(dev);
508 }
509
510 /* freeze and set UNLOADED */
511 spin_lock_irqsave(ap->lock, flags);
512
513 ata_port_freeze(ap); /* won't be thawed */
514 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
515 ap->pflags |= ATA_PFLAG_UNLOADED;
516
517 spin_unlock_irqrestore(ap->lock, flags);
518}
519
520/**
521 * ata_scsi_error - SCSI layer error handler callback
522 * @host: SCSI host on which error occurred
523 *
524 * Handles SCSI-layer-thrown error events.
525 *
526 * LOCKING:
527 * Inherited from SCSI layer (none, can sleep)
528 *
529 * RETURNS:
530 * Zero.
531 */
532void ata_scsi_error(struct Scsi_Host *host)
533{
534 struct ata_port *ap = ata_shost_to_port(host);
535 unsigned long flags;
536 LIST_HEAD(eh_work_q);
537
538 spin_lock_irqsave(host->host_lock, flags);
539 list_splice_init(&host->eh_cmd_q, &eh_work_q);
540 spin_unlock_irqrestore(host->host_lock, flags);
541
542 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
543
544 /* If we timed raced normal completion and there is nothing to
545 recover nr_timedout == 0 why exactly are we doing error recovery ? */
546 ata_scsi_port_error_handler(host, ap);
547
548 /* finish or retry handled scmd's and clean up */
549 WARN_ON(!list_empty(&eh_work_q));
550
551}
552
553/**
554 * ata_scsi_cmd_error_handler - error callback for a list of commands
555 * @host: scsi host containing the port
556 * @ap: ATA port within the host
557 * @eh_work_q: list of commands to process
558 *
559 * process the given list of commands and return those finished to the
560 * ap->eh_done_q. This function is the first part of the libata error
561 * handler which processes a given list of failed commands.
562 */
563void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
564 struct list_head *eh_work_q)
565{
566 int i;
567 unsigned long flags;
568
569 /* make sure sff pio task is not running */
570 ata_sff_flush_pio_task(ap);
571
572 /* synchronize with host lock and sort out timeouts */
573
574 /* For new EH, all qcs are finished in one of three ways -
575 * normal completion, error completion, and SCSI timeout.
576 * Both completions can race against SCSI timeout. When normal
577 * completion wins, the qc never reaches EH. When error
578 * completion wins, the qc has ATA_QCFLAG_FAILED set.
579 *
580 * When SCSI timeout wins, things are a bit more complex.
581 * Normal or error completion can occur after the timeout but
582 * before this point. In such cases, both types of
583 * completions are honored. A scmd is determined to have
584 * timed out iff its associated qc is active and not failed.
585 */
586 spin_lock_irqsave(ap->lock, flags);
587 if (ap->ops->error_handler) {
588 struct scsi_cmnd *scmd, *tmp;
589 int nr_timedout = 0;
590
591 /* This must occur under the ap->lock as we don't want
592 a polled recovery to race the real interrupt handler
593
594 The lost_interrupt handler checks for any completed but
595 non-notified command and completes much like an IRQ handler.
596
597 We then fall into the error recovery code which will treat
598 this as if normal completion won the race */
599
600 if (ap->ops->lost_interrupt)
601 ap->ops->lost_interrupt(ap);
602
603 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
604 struct ata_queued_cmd *qc;
605
606 ata_qc_for_each_raw(ap, qc, i) {
607 if (qc->flags & ATA_QCFLAG_ACTIVE &&
608 qc->scsicmd == scmd)
609 break;
610 }
611
612 if (i < ATA_MAX_QUEUE) {
613 /* the scmd has an associated qc */
614 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
615 /* which hasn't failed yet, timeout */
616 qc->err_mask |= AC_ERR_TIMEOUT;
617 qc->flags |= ATA_QCFLAG_FAILED;
618 nr_timedout++;
619 }
620 } else {
621 /* Normal completion occurred after
622 * SCSI timeout but before this point.
623 * Successfully complete it.
624 */
625 scmd->retries = scmd->allowed;
626 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
627 }
628 }
629
630 /* If we have timed out qcs. They belong to EH from
631 * this point but the state of the controller is
632 * unknown. Freeze the port to make sure the IRQ
633 * handler doesn't diddle with those qcs. This must
634 * be done atomically w.r.t. setting QCFLAG_FAILED.
635 */
636 if (nr_timedout)
637 __ata_port_freeze(ap);
638
639
640 /* initialize eh_tries */
641 ap->eh_tries = ATA_EH_MAX_TRIES;
642 }
643 spin_unlock_irqrestore(ap->lock, flags);
644
645}
646EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
647
648/**
649 * ata_scsi_port_error_handler - recover the port after the commands
650 * @host: SCSI host containing the port
651 * @ap: the ATA port
652 *
653 * Handle the recovery of the port @ap after all the commands
654 * have been recovered.
655 */
656void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
657{
658 unsigned long flags;
659
660 /* invoke error handler */
661 if (ap->ops->error_handler) {
662 struct ata_link *link;
663
664 /* acquire EH ownership */
665 ata_eh_acquire(ap);
666 repeat:
667 /* kill fast drain timer */
668 del_timer_sync(&ap->fastdrain_timer);
669
670 /* process port resume request */
671 ata_eh_handle_port_resume(ap);
672
673 /* fetch & clear EH info */
674 spin_lock_irqsave(ap->lock, flags);
675
676 ata_for_each_link(link, ap, HOST_FIRST) {
677 struct ata_eh_context *ehc = &link->eh_context;
678 struct ata_device *dev;
679
680 memset(&link->eh_context, 0, sizeof(link->eh_context));
681 link->eh_context.i = link->eh_info;
682 memset(&link->eh_info, 0, sizeof(link->eh_info));
683
684 ata_for_each_dev(dev, link, ENABLED) {
685 int devno = dev->devno;
686
687 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
688 if (ata_ncq_enabled(dev))
689 ehc->saved_ncq_enabled |= 1 << devno;
690 }
691 }
692
693 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
694 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
695 ap->excl_link = NULL; /* don't maintain exclusion over EH */
696
697 spin_unlock_irqrestore(ap->lock, flags);
698
699 /* invoke EH, skip if unloading or suspended */
700 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
701 ap->ops->error_handler(ap);
702 else {
703 /* if unloading, commence suicide */
704 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
705 !(ap->pflags & ATA_PFLAG_UNLOADED))
706 ata_eh_unload(ap);
707 ata_eh_finish(ap);
708 }
709
710 /* process port suspend request */
711 ata_eh_handle_port_suspend(ap);
712
713 /* Exception might have happened after ->error_handler
714 * recovered the port but before this point. Repeat
715 * EH in such case.
716 */
717 spin_lock_irqsave(ap->lock, flags);
718
719 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
720 if (--ap->eh_tries) {
721 spin_unlock_irqrestore(ap->lock, flags);
722 goto repeat;
723 }
724 ata_port_err(ap,
725 "EH pending after %d tries, giving up\n",
726 ATA_EH_MAX_TRIES);
727 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
728 }
729
730 /* this run is complete, make sure EH info is clear */
731 ata_for_each_link(link, ap, HOST_FIRST)
732 memset(&link->eh_info, 0, sizeof(link->eh_info));
733
734 /* end eh (clear host_eh_scheduled) while holding
735 * ap->lock such that if exception occurs after this
736 * point but before EH completion, SCSI midlayer will
737 * re-initiate EH.
738 */
739 ap->ops->end_eh(ap);
740
741 spin_unlock_irqrestore(ap->lock, flags);
742 ata_eh_release(ap);
743 } else {
744 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
745 ap->ops->eng_timeout(ap);
746 }
747
748 scsi_eh_flush_done_q(&ap->eh_done_q);
749
750 /* clean up */
751 spin_lock_irqsave(ap->lock, flags);
752
753 if (ap->pflags & ATA_PFLAG_LOADING)
754 ap->pflags &= ~ATA_PFLAG_LOADING;
755 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
756 !(ap->flags & ATA_FLAG_SAS_HOST))
757 schedule_delayed_work(&ap->hotplug_task, 0);
758
759 if (ap->pflags & ATA_PFLAG_RECOVERED)
760 ata_port_info(ap, "EH complete\n");
761
762 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
763
764 /* tell wait_eh that we're done */
765 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
766 wake_up_all(&ap->eh_wait_q);
767
768 spin_unlock_irqrestore(ap->lock, flags);
769}
770EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
771
772/**
773 * ata_port_wait_eh - Wait for the currently pending EH to complete
774 * @ap: Port to wait EH for
775 *
776 * Wait until the currently pending EH is complete.
777 *
778 * LOCKING:
779 * Kernel thread context (may sleep).
780 */
781void ata_port_wait_eh(struct ata_port *ap)
782{
783 unsigned long flags;
784 DEFINE_WAIT(wait);
785
786 retry:
787 spin_lock_irqsave(ap->lock, flags);
788
789 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
790 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
791 spin_unlock_irqrestore(ap->lock, flags);
792 schedule();
793 spin_lock_irqsave(ap->lock, flags);
794 }
795 finish_wait(&ap->eh_wait_q, &wait);
796
797 spin_unlock_irqrestore(ap->lock, flags);
798
799 /* make sure SCSI EH is complete */
800 if (scsi_host_in_recovery(ap->scsi_host)) {
801 ata_msleep(ap, 10);
802 goto retry;
803 }
804}
805EXPORT_SYMBOL_GPL(ata_port_wait_eh);
806
807static unsigned int ata_eh_nr_in_flight(struct ata_port *ap)
808{
809 struct ata_queued_cmd *qc;
810 unsigned int tag;
811 unsigned int nr = 0;
812
813 /* count only non-internal commands */
814 ata_qc_for_each(ap, qc, tag) {
815 if (qc)
816 nr++;
817 }
818
819 return nr;
820}
821
822void ata_eh_fastdrain_timerfn(struct timer_list *t)
823{
824 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
825 unsigned long flags;
826 unsigned int cnt;
827
828 spin_lock_irqsave(ap->lock, flags);
829
830 cnt = ata_eh_nr_in_flight(ap);
831
832 /* are we done? */
833 if (!cnt)
834 goto out_unlock;
835
836 if (cnt == ap->fastdrain_cnt) {
837 struct ata_queued_cmd *qc;
838 unsigned int tag;
839
840 /* No progress during the last interval, tag all
841 * in-flight qcs as timed out and freeze the port.
842 */
843 ata_qc_for_each(ap, qc, tag) {
844 if (qc)
845 qc->err_mask |= AC_ERR_TIMEOUT;
846 }
847
848 ata_port_freeze(ap);
849 } else {
850 /* some qcs have finished, give it another chance */
851 ap->fastdrain_cnt = cnt;
852 ap->fastdrain_timer.expires =
853 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
854 add_timer(&ap->fastdrain_timer);
855 }
856
857 out_unlock:
858 spin_unlock_irqrestore(ap->lock, flags);
859}
860
861/**
862 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
863 * @ap: target ATA port
864 * @fastdrain: activate fast drain
865 *
866 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
867 * is non-zero and EH wasn't pending before. Fast drain ensures
868 * that EH kicks in in timely manner.
869 *
870 * LOCKING:
871 * spin_lock_irqsave(host lock)
872 */
873static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
874{
875 unsigned int cnt;
876
877 /* already scheduled? */
878 if (ap->pflags & ATA_PFLAG_EH_PENDING)
879 return;
880
881 ap->pflags |= ATA_PFLAG_EH_PENDING;
882
883 if (!fastdrain)
884 return;
885
886 /* do we have in-flight qcs? */
887 cnt = ata_eh_nr_in_flight(ap);
888 if (!cnt)
889 return;
890
891 /* activate fast drain */
892 ap->fastdrain_cnt = cnt;
893 ap->fastdrain_timer.expires =
894 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
895 add_timer(&ap->fastdrain_timer);
896}
897
898/**
899 * ata_qc_schedule_eh - schedule qc for error handling
900 * @qc: command to schedule error handling for
901 *
902 * Schedule error handling for @qc. EH will kick in as soon as
903 * other commands are drained.
904 *
905 * LOCKING:
906 * spin_lock_irqsave(host lock)
907 */
908void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
909{
910 struct ata_port *ap = qc->ap;
911
912 WARN_ON(!ap->ops->error_handler);
913
914 qc->flags |= ATA_QCFLAG_FAILED;
915 ata_eh_set_pending(ap, 1);
916
917 /* The following will fail if timeout has already expired.
918 * ata_scsi_error() takes care of such scmds on EH entry.
919 * Note that ATA_QCFLAG_FAILED is unconditionally set after
920 * this function completes.
921 */
922 blk_abort_request(scsi_cmd_to_rq(qc->scsicmd));
923}
924
925/**
926 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
927 * @ap: ATA port to schedule EH for
928 *
929 * LOCKING: inherited from ata_port_schedule_eh
930 * spin_lock_irqsave(host lock)
931 */
932void ata_std_sched_eh(struct ata_port *ap)
933{
934 WARN_ON(!ap->ops->error_handler);
935
936 if (ap->pflags & ATA_PFLAG_INITIALIZING)
937 return;
938
939 ata_eh_set_pending(ap, 1);
940 scsi_schedule_eh(ap->scsi_host);
941
942 trace_ata_std_sched_eh(ap);
943}
944EXPORT_SYMBOL_GPL(ata_std_sched_eh);
945
946/**
947 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
948 * @ap: ATA port to end EH for
949 *
950 * In the libata object model there is a 1:1 mapping of ata_port to
951 * shost, so host fields can be directly manipulated under ap->lock, in
952 * the libsas case we need to hold a lock at the ha->level to coordinate
953 * these events.
954 *
955 * LOCKING:
956 * spin_lock_irqsave(host lock)
957 */
958void ata_std_end_eh(struct ata_port *ap)
959{
960 struct Scsi_Host *host = ap->scsi_host;
961
962 host->host_eh_scheduled = 0;
963}
964EXPORT_SYMBOL(ata_std_end_eh);
965
966
967/**
968 * ata_port_schedule_eh - schedule error handling without a qc
969 * @ap: ATA port to schedule EH for
970 *
971 * Schedule error handling for @ap. EH will kick in as soon as
972 * all commands are drained.
973 *
974 * LOCKING:
975 * spin_lock_irqsave(host lock)
976 */
977void ata_port_schedule_eh(struct ata_port *ap)
978{
979 /* see: ata_std_sched_eh, unless you know better */
980 ap->ops->sched_eh(ap);
981}
982EXPORT_SYMBOL_GPL(ata_port_schedule_eh);
983
984static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
985{
986 struct ata_queued_cmd *qc;
987 int tag, nr_aborted = 0;
988
989 WARN_ON(!ap->ops->error_handler);
990
991 /* we're gonna abort all commands, no need for fast drain */
992 ata_eh_set_pending(ap, 0);
993
994 /* include internal tag in iteration */
995 ata_qc_for_each_with_internal(ap, qc, tag) {
996 if (qc && (!link || qc->dev->link == link)) {
997 qc->flags |= ATA_QCFLAG_FAILED;
998 ata_qc_complete(qc);
999 nr_aborted++;
1000 }
1001 }
1002
1003 if (!nr_aborted)
1004 ata_port_schedule_eh(ap);
1005
1006 return nr_aborted;
1007}
1008
1009/**
1010 * ata_link_abort - abort all qc's on the link
1011 * @link: ATA link to abort qc's for
1012 *
1013 * Abort all active qc's active on @link and schedule EH.
1014 *
1015 * LOCKING:
1016 * spin_lock_irqsave(host lock)
1017 *
1018 * RETURNS:
1019 * Number of aborted qc's.
1020 */
1021int ata_link_abort(struct ata_link *link)
1022{
1023 return ata_do_link_abort(link->ap, link);
1024}
1025EXPORT_SYMBOL_GPL(ata_link_abort);
1026
1027/**
1028 * ata_port_abort - abort all qc's on the port
1029 * @ap: ATA port to abort qc's for
1030 *
1031 * Abort all active qc's of @ap and schedule EH.
1032 *
1033 * LOCKING:
1034 * spin_lock_irqsave(host_set lock)
1035 *
1036 * RETURNS:
1037 * Number of aborted qc's.
1038 */
1039int ata_port_abort(struct ata_port *ap)
1040{
1041 return ata_do_link_abort(ap, NULL);
1042}
1043EXPORT_SYMBOL_GPL(ata_port_abort);
1044
1045/**
1046 * __ata_port_freeze - freeze port
1047 * @ap: ATA port to freeze
1048 *
1049 * This function is called when HSM violation or some other
1050 * condition disrupts normal operation of the port. Frozen port
1051 * is not allowed to perform any operation until the port is
1052 * thawed, which usually follows a successful reset.
1053 *
1054 * ap->ops->freeze() callback can be used for freezing the port
1055 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1056 * port cannot be frozen hardware-wise, the interrupt handler
1057 * must ack and clear interrupts unconditionally while the port
1058 * is frozen.
1059 *
1060 * LOCKING:
1061 * spin_lock_irqsave(host lock)
1062 */
1063static void __ata_port_freeze(struct ata_port *ap)
1064{
1065 WARN_ON(!ap->ops->error_handler);
1066
1067 if (ap->ops->freeze)
1068 ap->ops->freeze(ap);
1069
1070 ap->pflags |= ATA_PFLAG_FROZEN;
1071
1072 trace_ata_port_freeze(ap);
1073}
1074
1075/**
1076 * ata_port_freeze - abort & freeze port
1077 * @ap: ATA port to freeze
1078 *
1079 * Abort and freeze @ap. The freeze operation must be called
1080 * first, because some hardware requires special operations
1081 * before the taskfile registers are accessible.
1082 *
1083 * LOCKING:
1084 * spin_lock_irqsave(host lock)
1085 *
1086 * RETURNS:
1087 * Number of aborted commands.
1088 */
1089int ata_port_freeze(struct ata_port *ap)
1090{
1091 WARN_ON(!ap->ops->error_handler);
1092
1093 __ata_port_freeze(ap);
1094
1095 return ata_port_abort(ap);
1096}
1097EXPORT_SYMBOL_GPL(ata_port_freeze);
1098
1099/**
1100 * ata_eh_freeze_port - EH helper to freeze port
1101 * @ap: ATA port to freeze
1102 *
1103 * Freeze @ap.
1104 *
1105 * LOCKING:
1106 * None.
1107 */
1108void ata_eh_freeze_port(struct ata_port *ap)
1109{
1110 unsigned long flags;
1111
1112 if (!ap->ops->error_handler)
1113 return;
1114
1115 spin_lock_irqsave(ap->lock, flags);
1116 __ata_port_freeze(ap);
1117 spin_unlock_irqrestore(ap->lock, flags);
1118}
1119EXPORT_SYMBOL_GPL(ata_eh_freeze_port);
1120
1121/**
1122 * ata_eh_thaw_port - EH helper to thaw port
1123 * @ap: ATA port to thaw
1124 *
1125 * Thaw frozen port @ap.
1126 *
1127 * LOCKING:
1128 * None.
1129 */
1130void ata_eh_thaw_port(struct ata_port *ap)
1131{
1132 unsigned long flags;
1133
1134 if (!ap->ops->error_handler)
1135 return;
1136
1137 spin_lock_irqsave(ap->lock, flags);
1138
1139 ap->pflags &= ~ATA_PFLAG_FROZEN;
1140
1141 if (ap->ops->thaw)
1142 ap->ops->thaw(ap);
1143
1144 spin_unlock_irqrestore(ap->lock, flags);
1145
1146 trace_ata_port_thaw(ap);
1147}
1148
1149static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1150{
1151 /* nada */
1152}
1153
1154static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1155{
1156 struct ata_port *ap = qc->ap;
1157 struct scsi_cmnd *scmd = qc->scsicmd;
1158 unsigned long flags;
1159
1160 spin_lock_irqsave(ap->lock, flags);
1161 qc->scsidone = ata_eh_scsidone;
1162 __ata_qc_complete(qc);
1163 WARN_ON(ata_tag_valid(qc->tag));
1164 spin_unlock_irqrestore(ap->lock, flags);
1165
1166 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1167}
1168
1169/**
1170 * ata_eh_qc_complete - Complete an active ATA command from EH
1171 * @qc: Command to complete
1172 *
1173 * Indicate to the mid and upper layers that an ATA command has
1174 * completed. To be used from EH.
1175 */
1176void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1177{
1178 struct scsi_cmnd *scmd = qc->scsicmd;
1179 scmd->retries = scmd->allowed;
1180 __ata_eh_qc_complete(qc);
1181}
1182
1183/**
1184 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1185 * @qc: Command to retry
1186 *
1187 * Indicate to the mid and upper layers that an ATA command
1188 * should be retried. To be used from EH.
1189 *
1190 * SCSI midlayer limits the number of retries to scmd->allowed.
1191 * scmd->allowed is incremented for commands which get retried
1192 * due to unrelated failures (qc->err_mask is zero).
1193 */
1194void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1195{
1196 struct scsi_cmnd *scmd = qc->scsicmd;
1197 if (!qc->err_mask)
1198 scmd->allowed++;
1199 __ata_eh_qc_complete(qc);
1200}
1201
1202/**
1203 * ata_dev_disable - disable ATA device
1204 * @dev: ATA device to disable
1205 *
1206 * Disable @dev.
1207 *
1208 * Locking:
1209 * EH context.
1210 */
1211void ata_dev_disable(struct ata_device *dev)
1212{
1213 if (!ata_dev_enabled(dev))
1214 return;
1215
1216 ata_dev_warn(dev, "disable device\n");
1217 ata_acpi_on_disable(dev);
1218 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1219 dev->class++;
1220
1221 /* From now till the next successful probe, ering is used to
1222 * track probe failures. Clear accumulated device error info.
1223 */
1224 ata_ering_clear(&dev->ering);
1225}
1226EXPORT_SYMBOL_GPL(ata_dev_disable);
1227
1228/**
1229 * ata_eh_detach_dev - detach ATA device
1230 * @dev: ATA device to detach
1231 *
1232 * Detach @dev.
1233 *
1234 * LOCKING:
1235 * None.
1236 */
1237void ata_eh_detach_dev(struct ata_device *dev)
1238{
1239 struct ata_link *link = dev->link;
1240 struct ata_port *ap = link->ap;
1241 struct ata_eh_context *ehc = &link->eh_context;
1242 unsigned long flags;
1243
1244 ata_dev_disable(dev);
1245
1246 spin_lock_irqsave(ap->lock, flags);
1247
1248 dev->flags &= ~ATA_DFLAG_DETACH;
1249
1250 if (ata_scsi_offline_dev(dev)) {
1251 dev->flags |= ATA_DFLAG_DETACHED;
1252 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1253 }
1254
1255 /* clear per-dev EH info */
1256 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1257 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1258 ehc->saved_xfer_mode[dev->devno] = 0;
1259 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1260
1261 spin_unlock_irqrestore(ap->lock, flags);
1262}
1263
1264/**
1265 * ata_eh_about_to_do - about to perform eh_action
1266 * @link: target ATA link
1267 * @dev: target ATA dev for per-dev action (can be NULL)
1268 * @action: action about to be performed
1269 *
1270 * Called just before performing EH actions to clear related bits
1271 * in @link->eh_info such that eh actions are not unnecessarily
1272 * repeated.
1273 *
1274 * LOCKING:
1275 * None.
1276 */
1277void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1278 unsigned int action)
1279{
1280 struct ata_port *ap = link->ap;
1281 struct ata_eh_info *ehi = &link->eh_info;
1282 struct ata_eh_context *ehc = &link->eh_context;
1283 unsigned long flags;
1284
1285 trace_ata_eh_about_to_do(link, dev ? dev->devno : 0, action);
1286
1287 spin_lock_irqsave(ap->lock, flags);
1288
1289 ata_eh_clear_action(link, dev, ehi, action);
1290
1291 /* About to take EH action, set RECOVERED. Ignore actions on
1292 * slave links as master will do them again.
1293 */
1294 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1295 ap->pflags |= ATA_PFLAG_RECOVERED;
1296
1297 spin_unlock_irqrestore(ap->lock, flags);
1298}
1299
1300/**
1301 * ata_eh_done - EH action complete
1302 * @link: ATA link for which EH actions are complete
1303 * @dev: target ATA dev for per-dev action (can be NULL)
1304 * @action: action just completed
1305 *
1306 * Called right after performing EH actions to clear related bits
1307 * in @link->eh_context.
1308 *
1309 * LOCKING:
1310 * None.
1311 */
1312void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1313 unsigned int action)
1314{
1315 struct ata_eh_context *ehc = &link->eh_context;
1316
1317 trace_ata_eh_done(link, dev ? dev->devno : 0, action);
1318
1319 ata_eh_clear_action(link, dev, &ehc->i, action);
1320}
1321
1322/**
1323 * ata_err_string - convert err_mask to descriptive string
1324 * @err_mask: error mask to convert to string
1325 *
1326 * Convert @err_mask to descriptive string. Errors are
1327 * prioritized according to severity and only the most severe
1328 * error is reported.
1329 *
1330 * LOCKING:
1331 * None.
1332 *
1333 * RETURNS:
1334 * Descriptive string for @err_mask
1335 */
1336static const char *ata_err_string(unsigned int err_mask)
1337{
1338 if (err_mask & AC_ERR_HOST_BUS)
1339 return "host bus error";
1340 if (err_mask & AC_ERR_ATA_BUS)
1341 return "ATA bus error";
1342 if (err_mask & AC_ERR_TIMEOUT)
1343 return "timeout";
1344 if (err_mask & AC_ERR_HSM)
1345 return "HSM violation";
1346 if (err_mask & AC_ERR_SYSTEM)
1347 return "internal error";
1348 if (err_mask & AC_ERR_MEDIA)
1349 return "media error";
1350 if (err_mask & AC_ERR_INVALID)
1351 return "invalid argument";
1352 if (err_mask & AC_ERR_DEV)
1353 return "device error";
1354 if (err_mask & AC_ERR_NCQ)
1355 return "NCQ error";
1356 if (err_mask & AC_ERR_NODEV_HINT)
1357 return "Polling detection error";
1358 return "unknown error";
1359}
1360
1361/**
1362 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1363 * @dev: target ATAPI device
1364 * @r_sense_key: out parameter for sense_key
1365 *
1366 * Perform ATAPI TEST_UNIT_READY.
1367 *
1368 * LOCKING:
1369 * EH context (may sleep).
1370 *
1371 * RETURNS:
1372 * 0 on success, AC_ERR_* mask on failure.
1373 */
1374unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1375{
1376 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1377 struct ata_taskfile tf;
1378 unsigned int err_mask;
1379
1380 ata_tf_init(dev, &tf);
1381
1382 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1383 tf.command = ATA_CMD_PACKET;
1384 tf.protocol = ATAPI_PROT_NODATA;
1385
1386 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1387 if (err_mask == AC_ERR_DEV)
1388 *r_sense_key = tf.error >> 4;
1389 return err_mask;
1390}
1391
1392/**
1393 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1394 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1395 *
1396 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1397 * SENSE. This function is an EH helper.
1398 *
1399 * LOCKING:
1400 * Kernel thread context (may sleep).
1401 */
1402static void ata_eh_request_sense(struct ata_queued_cmd *qc)
1403{
1404 struct scsi_cmnd *cmd = qc->scsicmd;
1405 struct ata_device *dev = qc->dev;
1406 struct ata_taskfile tf;
1407 unsigned int err_mask;
1408
1409 if (ata_port_is_frozen(qc->ap)) {
1410 ata_dev_warn(dev, "sense data available but port frozen\n");
1411 return;
1412 }
1413
1414 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1415 return;
1416
1417 if (!ata_id_sense_reporting_enabled(dev->id)) {
1418 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1419 return;
1420 }
1421
1422 ata_tf_init(dev, &tf);
1423 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1424 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1425 tf.command = ATA_CMD_REQ_SENSE_DATA;
1426 tf.protocol = ATA_PROT_NODATA;
1427
1428 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1429 /* Ignore err_mask; ATA_ERR might be set */
1430 if (tf.status & ATA_SENSE) {
1431 if (ata_scsi_sense_is_valid(tf.lbah, tf.lbam, tf.lbal)) {
1432 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1433 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1434 }
1435 } else {
1436 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1437 tf.status, err_mask);
1438 }
1439}
1440
1441/**
1442 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1443 * @dev: device to perform REQUEST_SENSE to
1444 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1445 * @dfl_sense_key: default sense key to use
1446 *
1447 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1448 * SENSE. This function is EH helper.
1449 *
1450 * LOCKING:
1451 * Kernel thread context (may sleep).
1452 *
1453 * RETURNS:
1454 * 0 on success, AC_ERR_* mask on failure
1455 */
1456unsigned int atapi_eh_request_sense(struct ata_device *dev,
1457 u8 *sense_buf, u8 dfl_sense_key)
1458{
1459 u8 cdb[ATAPI_CDB_LEN] =
1460 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1461 struct ata_port *ap = dev->link->ap;
1462 struct ata_taskfile tf;
1463
1464 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1465
1466 /* initialize sense_buf with the error register,
1467 * for the case where they are -not- overwritten
1468 */
1469 sense_buf[0] = 0x70;
1470 sense_buf[2] = dfl_sense_key;
1471
1472 /* some devices time out if garbage left in tf */
1473 ata_tf_init(dev, &tf);
1474
1475 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1476 tf.command = ATA_CMD_PACKET;
1477
1478 /* is it pointless to prefer PIO for "safety reasons"? */
1479 if (ap->flags & ATA_FLAG_PIO_DMA) {
1480 tf.protocol = ATAPI_PROT_DMA;
1481 tf.feature |= ATAPI_PKT_DMA;
1482 } else {
1483 tf.protocol = ATAPI_PROT_PIO;
1484 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1485 tf.lbah = 0;
1486 }
1487
1488 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1489 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1490}
1491
1492/**
1493 * ata_eh_analyze_serror - analyze SError for a failed port
1494 * @link: ATA link to analyze SError for
1495 *
1496 * Analyze SError if available and further determine cause of
1497 * failure.
1498 *
1499 * LOCKING:
1500 * None.
1501 */
1502static void ata_eh_analyze_serror(struct ata_link *link)
1503{
1504 struct ata_eh_context *ehc = &link->eh_context;
1505 u32 serror = ehc->i.serror;
1506 unsigned int err_mask = 0, action = 0;
1507 u32 hotplug_mask;
1508
1509 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1510 err_mask |= AC_ERR_ATA_BUS;
1511 action |= ATA_EH_RESET;
1512 }
1513 if (serror & SERR_PROTOCOL) {
1514 err_mask |= AC_ERR_HSM;
1515 action |= ATA_EH_RESET;
1516 }
1517 if (serror & SERR_INTERNAL) {
1518 err_mask |= AC_ERR_SYSTEM;
1519 action |= ATA_EH_RESET;
1520 }
1521
1522 /* Determine whether a hotplug event has occurred. Both
1523 * SError.N/X are considered hotplug events for enabled or
1524 * host links. For disabled PMP links, only N bit is
1525 * considered as X bit is left at 1 for link plugging.
1526 */
1527 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1528 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1529 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1530 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1531 else
1532 hotplug_mask = SERR_PHYRDY_CHG;
1533
1534 if (serror & hotplug_mask)
1535 ata_ehi_hotplugged(&ehc->i);
1536
1537 ehc->i.err_mask |= err_mask;
1538 ehc->i.action |= action;
1539}
1540
1541/**
1542 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1543 * @qc: qc to analyze
1544 *
1545 * Analyze taskfile of @qc and further determine cause of
1546 * failure. This function also requests ATAPI sense data if
1547 * available.
1548 *
1549 * LOCKING:
1550 * Kernel thread context (may sleep).
1551 *
1552 * RETURNS:
1553 * Determined recovery action
1554 */
1555static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc)
1556{
1557 const struct ata_taskfile *tf = &qc->result_tf;
1558 unsigned int tmp, action = 0;
1559 u8 stat = tf->status, err = tf->error;
1560
1561 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1562 qc->err_mask |= AC_ERR_HSM;
1563 return ATA_EH_RESET;
1564 }
1565
1566 if (stat & (ATA_ERR | ATA_DF)) {
1567 qc->err_mask |= AC_ERR_DEV;
1568 /*
1569 * Sense data reporting does not work if the
1570 * device fault bit is set.
1571 */
1572 if (stat & ATA_DF)
1573 stat &= ~ATA_SENSE;
1574 } else {
1575 return 0;
1576 }
1577
1578 switch (qc->dev->class) {
1579 case ATA_DEV_ATA:
1580 case ATA_DEV_ZAC:
1581 /*
1582 * Fetch the sense data explicitly if:
1583 * -It was a non-NCQ command that failed, or
1584 * -It was a NCQ command that failed, but the sense data
1585 * was not included in the NCQ command error log
1586 * (i.e. NCQ autosense is not supported by the device).
1587 */
1588 if (!(qc->flags & ATA_QCFLAG_SENSE_VALID) && (stat & ATA_SENSE))
1589 ata_eh_request_sense(qc);
1590 if (err & ATA_ICRC)
1591 qc->err_mask |= AC_ERR_ATA_BUS;
1592 if (err & (ATA_UNC | ATA_AMNF))
1593 qc->err_mask |= AC_ERR_MEDIA;
1594 if (err & ATA_IDNF)
1595 qc->err_mask |= AC_ERR_INVALID;
1596 break;
1597
1598 case ATA_DEV_ATAPI:
1599 if (!ata_port_is_frozen(qc->ap)) {
1600 tmp = atapi_eh_request_sense(qc->dev,
1601 qc->scsicmd->sense_buffer,
1602 qc->result_tf.error >> 4);
1603 if (!tmp)
1604 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1605 else
1606 qc->err_mask |= tmp;
1607 }
1608 }
1609
1610 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1611 enum scsi_disposition ret = scsi_check_sense(qc->scsicmd);
1612 /*
1613 * SUCCESS here means that the sense code could be
1614 * evaluated and should be passed to the upper layers
1615 * for correct evaluation.
1616 * FAILED means the sense code could not be interpreted
1617 * and the device would need to be reset.
1618 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1619 * command would need to be retried.
1620 */
1621 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1622 qc->flags |= ATA_QCFLAG_RETRY;
1623 qc->err_mask |= AC_ERR_OTHER;
1624 } else if (ret != SUCCESS) {
1625 qc->err_mask |= AC_ERR_HSM;
1626 }
1627 }
1628 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1629 action |= ATA_EH_RESET;
1630
1631 return action;
1632}
1633
1634static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1635 int *xfer_ok)
1636{
1637 int base = 0;
1638
1639 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1640 *xfer_ok = 1;
1641
1642 if (!*xfer_ok)
1643 base = ATA_ECAT_DUBIOUS_NONE;
1644
1645 if (err_mask & AC_ERR_ATA_BUS)
1646 return base + ATA_ECAT_ATA_BUS;
1647
1648 if (err_mask & AC_ERR_TIMEOUT)
1649 return base + ATA_ECAT_TOUT_HSM;
1650
1651 if (eflags & ATA_EFLAG_IS_IO) {
1652 if (err_mask & AC_ERR_HSM)
1653 return base + ATA_ECAT_TOUT_HSM;
1654 if ((err_mask &
1655 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1656 return base + ATA_ECAT_UNK_DEV;
1657 }
1658
1659 return 0;
1660}
1661
1662struct speed_down_verdict_arg {
1663 u64 since;
1664 int xfer_ok;
1665 int nr_errors[ATA_ECAT_NR];
1666};
1667
1668static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1669{
1670 struct speed_down_verdict_arg *arg = void_arg;
1671 int cat;
1672
1673 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1674 return -1;
1675
1676 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1677 &arg->xfer_ok);
1678 arg->nr_errors[cat]++;
1679
1680 return 0;
1681}
1682
1683/**
1684 * ata_eh_speed_down_verdict - Determine speed down verdict
1685 * @dev: Device of interest
1686 *
1687 * This function examines error ring of @dev and determines
1688 * whether NCQ needs to be turned off, transfer speed should be
1689 * stepped down, or falling back to PIO is necessary.
1690 *
1691 * ECAT_ATA_BUS : ATA_BUS error for any command
1692 *
1693 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1694 * IO commands
1695 *
1696 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1697 *
1698 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1699 * data transfer hasn't been verified.
1700 *
1701 * Verdicts are
1702 *
1703 * NCQ_OFF : Turn off NCQ.
1704 *
1705 * SPEED_DOWN : Speed down transfer speed but don't fall back
1706 * to PIO.
1707 *
1708 * FALLBACK_TO_PIO : Fall back to PIO.
1709 *
1710 * Even if multiple verdicts are returned, only one action is
1711 * taken per error. An action triggered by non-DUBIOUS errors
1712 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1713 * This is to expedite speed down decisions right after device is
1714 * initially configured.
1715 *
1716 * The following are speed down rules. #1 and #2 deal with
1717 * DUBIOUS errors.
1718 *
1719 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1720 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1721 *
1722 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1723 * occurred during last 5 mins, NCQ_OFF.
1724 *
1725 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1726 * occurred during last 5 mins, FALLBACK_TO_PIO
1727 *
1728 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1729 * during last 10 mins, NCQ_OFF.
1730 *
1731 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1732 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1733 *
1734 * LOCKING:
1735 * Inherited from caller.
1736 *
1737 * RETURNS:
1738 * OR of ATA_EH_SPDN_* flags.
1739 */
1740static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1741{
1742 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1743 u64 j64 = get_jiffies_64();
1744 struct speed_down_verdict_arg arg;
1745 unsigned int verdict = 0;
1746
1747 /* scan past 5 mins of error history */
1748 memset(&arg, 0, sizeof(arg));
1749 arg.since = j64 - min(j64, j5mins);
1750 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1751
1752 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1753 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1754 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1755 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1756
1757 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1758 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1759 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1760
1761 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1762 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1763 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1764 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1765
1766 /* scan past 10 mins of error history */
1767 memset(&arg, 0, sizeof(arg));
1768 arg.since = j64 - min(j64, j10mins);
1769 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1770
1771 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1772 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1773 verdict |= ATA_EH_SPDN_NCQ_OFF;
1774
1775 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1776 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1777 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1778 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1779
1780 return verdict;
1781}
1782
1783/**
1784 * ata_eh_speed_down - record error and speed down if necessary
1785 * @dev: Failed device
1786 * @eflags: mask of ATA_EFLAG_* flags
1787 * @err_mask: err_mask of the error
1788 *
1789 * Record error and examine error history to determine whether
1790 * adjusting transmission speed is necessary. It also sets
1791 * transmission limits appropriately if such adjustment is
1792 * necessary.
1793 *
1794 * LOCKING:
1795 * Kernel thread context (may sleep).
1796 *
1797 * RETURNS:
1798 * Determined recovery action.
1799 */
1800static unsigned int ata_eh_speed_down(struct ata_device *dev,
1801 unsigned int eflags, unsigned int err_mask)
1802{
1803 struct ata_link *link = ata_dev_phys_link(dev);
1804 int xfer_ok = 0;
1805 unsigned int verdict;
1806 unsigned int action = 0;
1807
1808 /* don't bother if Cat-0 error */
1809 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
1810 return 0;
1811
1812 /* record error and determine whether speed down is necessary */
1813 ata_ering_record(&dev->ering, eflags, err_mask);
1814 verdict = ata_eh_speed_down_verdict(dev);
1815
1816 /* turn off NCQ? */
1817 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1818 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1819 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1820 dev->flags |= ATA_DFLAG_NCQ_OFF;
1821 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
1822 goto done;
1823 }
1824
1825 /* speed down? */
1826 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1827 /* speed down SATA link speed if possible */
1828 if (sata_down_spd_limit(link, 0) == 0) {
1829 action |= ATA_EH_RESET;
1830 goto done;
1831 }
1832
1833 /* lower transfer mode */
1834 if (dev->spdn_cnt < 2) {
1835 static const int dma_dnxfer_sel[] =
1836 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1837 static const int pio_dnxfer_sel[] =
1838 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1839 int sel;
1840
1841 if (dev->xfer_shift != ATA_SHIFT_PIO)
1842 sel = dma_dnxfer_sel[dev->spdn_cnt];
1843 else
1844 sel = pio_dnxfer_sel[dev->spdn_cnt];
1845
1846 dev->spdn_cnt++;
1847
1848 if (ata_down_xfermask_limit(dev, sel) == 0) {
1849 action |= ATA_EH_RESET;
1850 goto done;
1851 }
1852 }
1853 }
1854
1855 /* Fall back to PIO? Slowing down to PIO is meaningless for
1856 * SATA ATA devices. Consider it only for PATA and SATAPI.
1857 */
1858 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1859 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
1860 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1861 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1862 dev->spdn_cnt = 0;
1863 action |= ATA_EH_RESET;
1864 goto done;
1865 }
1866 }
1867
1868 return 0;
1869 done:
1870 /* device has been slowed down, blow error history */
1871 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
1872 ata_ering_clear(&dev->ering);
1873 return action;
1874}
1875
1876/**
1877 * ata_eh_worth_retry - analyze error and decide whether to retry
1878 * @qc: qc to possibly retry
1879 *
1880 * Look at the cause of the error and decide if a retry
1881 * might be useful or not. We don't want to retry media errors
1882 * because the drive itself has probably already taken 10-30 seconds
1883 * doing its own internal retries before reporting the failure.
1884 */
1885static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
1886{
1887 if (qc->err_mask & AC_ERR_MEDIA)
1888 return 0; /* don't retry media errors */
1889 if (qc->flags & ATA_QCFLAG_IO)
1890 return 1; /* otherwise retry anything from fs stack */
1891 if (qc->err_mask & AC_ERR_INVALID)
1892 return 0; /* don't retry these */
1893 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
1894}
1895
1896/**
1897 * ata_eh_quiet - check if we need to be quiet about a command error
1898 * @qc: qc to check
1899 *
1900 * Look at the qc flags anbd its scsi command request flags to determine
1901 * if we need to be quiet about the command failure.
1902 */
1903static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
1904{
1905 if (qc->scsicmd && scsi_cmd_to_rq(qc->scsicmd)->rq_flags & RQF_QUIET)
1906 qc->flags |= ATA_QCFLAG_QUIET;
1907 return qc->flags & ATA_QCFLAG_QUIET;
1908}
1909
1910/**
1911 * ata_eh_link_autopsy - analyze error and determine recovery action
1912 * @link: host link to perform autopsy on
1913 *
1914 * Analyze why @link failed and determine which recovery actions
1915 * are needed. This function also sets more detailed AC_ERR_*
1916 * values and fills sense data for ATAPI CHECK SENSE.
1917 *
1918 * LOCKING:
1919 * Kernel thread context (may sleep).
1920 */
1921static void ata_eh_link_autopsy(struct ata_link *link)
1922{
1923 struct ata_port *ap = link->ap;
1924 struct ata_eh_context *ehc = &link->eh_context;
1925 struct ata_queued_cmd *qc;
1926 struct ata_device *dev;
1927 unsigned int all_err_mask = 0, eflags = 0;
1928 int tag, nr_failed = 0, nr_quiet = 0;
1929 u32 serror;
1930 int rc;
1931
1932 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1933 return;
1934
1935 /* obtain and analyze SError */
1936 rc = sata_scr_read(link, SCR_ERROR, &serror);
1937 if (rc == 0) {
1938 ehc->i.serror |= serror;
1939 ata_eh_analyze_serror(link);
1940 } else if (rc != -EOPNOTSUPP) {
1941 /* SError read failed, force reset and probing */
1942 ehc->i.probe_mask |= ATA_ALL_DEVICES;
1943 ehc->i.action |= ATA_EH_RESET;
1944 ehc->i.err_mask |= AC_ERR_OTHER;
1945 }
1946
1947 /* analyze NCQ failure */
1948 ata_eh_analyze_ncq_error(link);
1949
1950 /* any real error trumps AC_ERR_OTHER */
1951 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1952 ehc->i.err_mask &= ~AC_ERR_OTHER;
1953
1954 all_err_mask |= ehc->i.err_mask;
1955
1956 ata_qc_for_each_raw(ap, qc, tag) {
1957 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
1958 qc->flags & ATA_QCFLAG_RETRY ||
1959 ata_dev_phys_link(qc->dev) != link)
1960 continue;
1961
1962 /* inherit upper level err_mask */
1963 qc->err_mask |= ehc->i.err_mask;
1964
1965 /* analyze TF */
1966 ehc->i.action |= ata_eh_analyze_tf(qc);
1967
1968 /* DEV errors are probably spurious in case of ATA_BUS error */
1969 if (qc->err_mask & AC_ERR_ATA_BUS)
1970 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1971 AC_ERR_INVALID);
1972
1973 /* any real error trumps unknown error */
1974 if (qc->err_mask & ~AC_ERR_OTHER)
1975 qc->err_mask &= ~AC_ERR_OTHER;
1976
1977 /*
1978 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
1979 * layers will determine whether the command is worth retrying
1980 * based on the sense data and device class/type. Otherwise,
1981 * determine directly if the command is worth retrying using its
1982 * error mask and flags.
1983 */
1984 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
1985 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1986 else if (ata_eh_worth_retry(qc))
1987 qc->flags |= ATA_QCFLAG_RETRY;
1988
1989 /* accumulate error info */
1990 ehc->i.dev = qc->dev;
1991 all_err_mask |= qc->err_mask;
1992 if (qc->flags & ATA_QCFLAG_IO)
1993 eflags |= ATA_EFLAG_IS_IO;
1994 trace_ata_eh_link_autopsy_qc(qc);
1995
1996 /* Count quiet errors */
1997 if (ata_eh_quiet(qc))
1998 nr_quiet++;
1999 nr_failed++;
2000 }
2001
2002 /* If all failed commands requested silence, then be quiet */
2003 if (nr_quiet == nr_failed)
2004 ehc->i.flags |= ATA_EHI_QUIET;
2005
2006 /* enforce default EH actions */
2007 if (ata_port_is_frozen(ap) ||
2008 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2009 ehc->i.action |= ATA_EH_RESET;
2010 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2011 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2012 ehc->i.action |= ATA_EH_REVALIDATE;
2013
2014 /* If we have offending qcs and the associated failed device,
2015 * perform per-dev EH action only on the offending device.
2016 */
2017 if (ehc->i.dev) {
2018 ehc->i.dev_action[ehc->i.dev->devno] |=
2019 ehc->i.action & ATA_EH_PERDEV_MASK;
2020 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2021 }
2022
2023 /* propagate timeout to host link */
2024 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2025 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2026
2027 /* record error and consider speeding down */
2028 dev = ehc->i.dev;
2029 if (!dev && ((ata_link_max_devices(link) == 1 &&
2030 ata_dev_enabled(link->device))))
2031 dev = link->device;
2032
2033 if (dev) {
2034 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2035 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2036 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2037 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2038 }
2039}
2040
2041/**
2042 * ata_eh_autopsy - analyze error and determine recovery action
2043 * @ap: host port to perform autopsy on
2044 *
2045 * Analyze all links of @ap and determine why they failed and
2046 * which recovery actions are needed.
2047 *
2048 * LOCKING:
2049 * Kernel thread context (may sleep).
2050 */
2051void ata_eh_autopsy(struct ata_port *ap)
2052{
2053 struct ata_link *link;
2054
2055 ata_for_each_link(link, ap, EDGE)
2056 ata_eh_link_autopsy(link);
2057
2058 /* Handle the frigging slave link. Autopsy is done similarly
2059 * but actions and flags are transferred over to the master
2060 * link and handled from there.
2061 */
2062 if (ap->slave_link) {
2063 struct ata_eh_context *mehc = &ap->link.eh_context;
2064 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2065
2066 /* transfer control flags from master to slave */
2067 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2068
2069 /* perform autopsy on the slave link */
2070 ata_eh_link_autopsy(ap->slave_link);
2071
2072 /* transfer actions from slave to master and clear slave */
2073 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2074 mehc->i.action |= sehc->i.action;
2075 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2076 mehc->i.flags |= sehc->i.flags;
2077 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2078 }
2079
2080 /* Autopsy of fanout ports can affect host link autopsy.
2081 * Perform host link autopsy last.
2082 */
2083 if (sata_pmp_attached(ap))
2084 ata_eh_link_autopsy(&ap->link);
2085}
2086
2087/**
2088 * ata_get_cmd_name - get name for ATA command
2089 * @command: ATA command code to get name for
2090 *
2091 * Return a textual name of the given command or "unknown"
2092 *
2093 * LOCKING:
2094 * None
2095 */
2096const char *ata_get_cmd_name(u8 command)
2097{
2098#ifdef CONFIG_ATA_VERBOSE_ERROR
2099 static const struct
2100 {
2101 u8 command;
2102 const char *text;
2103 } cmd_descr[] = {
2104 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2105 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2106 { ATA_CMD_STANDBY, "STANDBY" },
2107 { ATA_CMD_IDLE, "IDLE" },
2108 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2109 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2110 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2111 { ATA_CMD_NOP, "NOP" },
2112 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2113 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2114 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2115 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2116 { ATA_CMD_SERVICE, "SERVICE" },
2117 { ATA_CMD_READ, "READ DMA" },
2118 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2119 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2120 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2121 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2122 { ATA_CMD_WRITE, "WRITE DMA" },
2123 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2124 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2125 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2126 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2127 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2128 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2129 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2130 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2131 { ATA_CMD_NCQ_NON_DATA, "NCQ NON-DATA" },
2132 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2133 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2134 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2135 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2136 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2137 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2138 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2139 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2140 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2141 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2142 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2143 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2144 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2145 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2146 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2147 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2148 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2149 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2150 { ATA_CMD_SLEEP, "SLEEP" },
2151 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2152 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2153 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2154 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2155 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2156 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2157 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2158 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2159 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2160 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2161 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2162 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2163 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2164 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2165 { ATA_CMD_PMP_READ, "READ BUFFER" },
2166 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2167 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2168 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2169 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2170 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2171 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2172 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2173 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2174 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2175 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2176 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2177 { ATA_CMD_SMART, "SMART" },
2178 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2179 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2180 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2181 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2182 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2183 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2184 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2185 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2186 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2187 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2188 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2189 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2190 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2191 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2192 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2193 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2194 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2195 { ATA_CMD_RESTORE, "RECALIBRATE" },
2196 { 0, NULL } /* terminate list */
2197 };
2198
2199 unsigned int i;
2200 for (i = 0; cmd_descr[i].text; i++)
2201 if (cmd_descr[i].command == command)
2202 return cmd_descr[i].text;
2203#endif
2204
2205 return "unknown";
2206}
2207EXPORT_SYMBOL_GPL(ata_get_cmd_name);
2208
2209/**
2210 * ata_eh_link_report - report error handling to user
2211 * @link: ATA link EH is going on
2212 *
2213 * Report EH to user.
2214 *
2215 * LOCKING:
2216 * None.
2217 */
2218static void ata_eh_link_report(struct ata_link *link)
2219{
2220 struct ata_port *ap = link->ap;
2221 struct ata_eh_context *ehc = &link->eh_context;
2222 struct ata_queued_cmd *qc;
2223 const char *frozen, *desc;
2224 char tries_buf[6] = "";
2225 int tag, nr_failed = 0;
2226
2227 if (ehc->i.flags & ATA_EHI_QUIET)
2228 return;
2229
2230 desc = NULL;
2231 if (ehc->i.desc[0] != '\0')
2232 desc = ehc->i.desc;
2233
2234 ata_qc_for_each_raw(ap, qc, tag) {
2235 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2236 ata_dev_phys_link(qc->dev) != link ||
2237 ((qc->flags & ATA_QCFLAG_QUIET) &&
2238 qc->err_mask == AC_ERR_DEV))
2239 continue;
2240 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2241 continue;
2242
2243 nr_failed++;
2244 }
2245
2246 if (!nr_failed && !ehc->i.err_mask)
2247 return;
2248
2249 frozen = "";
2250 if (ata_port_is_frozen(ap))
2251 frozen = " frozen";
2252
2253 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2254 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2255 ap->eh_tries);
2256
2257 if (ehc->i.dev) {
2258 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2259 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2260 ehc->i.err_mask, link->sactive, ehc->i.serror,
2261 ehc->i.action, frozen, tries_buf);
2262 if (desc)
2263 ata_dev_err(ehc->i.dev, "%s\n", desc);
2264 } else {
2265 ata_link_err(link, "exception Emask 0x%x "
2266 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2267 ehc->i.err_mask, link->sactive, ehc->i.serror,
2268 ehc->i.action, frozen, tries_buf);
2269 if (desc)
2270 ata_link_err(link, "%s\n", desc);
2271 }
2272
2273#ifdef CONFIG_ATA_VERBOSE_ERROR
2274 if (ehc->i.serror)
2275 ata_link_err(link,
2276 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2277 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2278 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2279 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2280 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2281 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2282 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2283 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2284 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2285 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2286 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2287 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2288 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2289 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2290 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2291 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2292 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2293 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2294#endif
2295
2296 ata_qc_for_each_raw(ap, qc, tag) {
2297 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2298 char data_buf[20] = "";
2299 char cdb_buf[70] = "";
2300
2301 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2302 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2303 continue;
2304
2305 if (qc->dma_dir != DMA_NONE) {
2306 static const char *dma_str[] = {
2307 [DMA_BIDIRECTIONAL] = "bidi",
2308 [DMA_TO_DEVICE] = "out",
2309 [DMA_FROM_DEVICE] = "in",
2310 };
2311 const char *prot_str = NULL;
2312
2313 switch (qc->tf.protocol) {
2314 case ATA_PROT_UNKNOWN:
2315 prot_str = "unknown";
2316 break;
2317 case ATA_PROT_NODATA:
2318 prot_str = "nodata";
2319 break;
2320 case ATA_PROT_PIO:
2321 prot_str = "pio";
2322 break;
2323 case ATA_PROT_DMA:
2324 prot_str = "dma";
2325 break;
2326 case ATA_PROT_NCQ:
2327 prot_str = "ncq dma";
2328 break;
2329 case ATA_PROT_NCQ_NODATA:
2330 prot_str = "ncq nodata";
2331 break;
2332 case ATAPI_PROT_NODATA:
2333 prot_str = "nodata";
2334 break;
2335 case ATAPI_PROT_PIO:
2336 prot_str = "pio";
2337 break;
2338 case ATAPI_PROT_DMA:
2339 prot_str = "dma";
2340 break;
2341 }
2342 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2343 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2344 }
2345
2346 if (ata_is_atapi(qc->tf.protocol)) {
2347 const u8 *cdb = qc->cdb;
2348 size_t cdb_len = qc->dev->cdb_len;
2349
2350 if (qc->scsicmd) {
2351 cdb = qc->scsicmd->cmnd;
2352 cdb_len = qc->scsicmd->cmd_len;
2353 }
2354 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2355 cdb, cdb_len);
2356 } else
2357 ata_dev_err(qc->dev, "failed command: %s\n",
2358 ata_get_cmd_name(cmd->command));
2359
2360 ata_dev_err(qc->dev,
2361 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2362 "tag %d%s\n %s"
2363 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2364 "Emask 0x%x (%s)%s\n",
2365 cmd->command, cmd->feature, cmd->nsect,
2366 cmd->lbal, cmd->lbam, cmd->lbah,
2367 cmd->hob_feature, cmd->hob_nsect,
2368 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2369 cmd->device, qc->tag, data_buf, cdb_buf,
2370 res->status, res->error, res->nsect,
2371 res->lbal, res->lbam, res->lbah,
2372 res->hob_feature, res->hob_nsect,
2373 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2374 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2375 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2376
2377#ifdef CONFIG_ATA_VERBOSE_ERROR
2378 if (res->status & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2379 ATA_SENSE | ATA_ERR)) {
2380 if (res->status & ATA_BUSY)
2381 ata_dev_err(qc->dev, "status: { Busy }\n");
2382 else
2383 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2384 res->status & ATA_DRDY ? "DRDY " : "",
2385 res->status & ATA_DF ? "DF " : "",
2386 res->status & ATA_DRQ ? "DRQ " : "",
2387 res->status & ATA_SENSE ? "SENSE " : "",
2388 res->status & ATA_ERR ? "ERR " : "");
2389 }
2390
2391 if (cmd->command != ATA_CMD_PACKET &&
2392 (res->error & (ATA_ICRC | ATA_UNC | ATA_AMNF | ATA_IDNF |
2393 ATA_ABORTED)))
2394 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2395 res->error & ATA_ICRC ? "ICRC " : "",
2396 res->error & ATA_UNC ? "UNC " : "",
2397 res->error & ATA_AMNF ? "AMNF " : "",
2398 res->error & ATA_IDNF ? "IDNF " : "",
2399 res->error & ATA_ABORTED ? "ABRT " : "");
2400#endif
2401 }
2402}
2403
2404/**
2405 * ata_eh_report - report error handling to user
2406 * @ap: ATA port to report EH about
2407 *
2408 * Report EH to user.
2409 *
2410 * LOCKING:
2411 * None.
2412 */
2413void ata_eh_report(struct ata_port *ap)
2414{
2415 struct ata_link *link;
2416
2417 ata_for_each_link(link, ap, HOST_FIRST)
2418 ata_eh_link_report(link);
2419}
2420
2421static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2422 unsigned int *classes, unsigned long deadline,
2423 bool clear_classes)
2424{
2425 struct ata_device *dev;
2426
2427 if (clear_classes)
2428 ata_for_each_dev(dev, link, ALL)
2429 classes[dev->devno] = ATA_DEV_UNKNOWN;
2430
2431 return reset(link, classes, deadline);
2432}
2433
2434static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2435{
2436 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2437 return 0;
2438 if (rc == -EAGAIN)
2439 return 1;
2440 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2441 return 1;
2442 return 0;
2443}
2444
2445int ata_eh_reset(struct ata_link *link, int classify,
2446 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2447 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2448{
2449 struct ata_port *ap = link->ap;
2450 struct ata_link *slave = ap->slave_link;
2451 struct ata_eh_context *ehc = &link->eh_context;
2452 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2453 unsigned int *classes = ehc->classes;
2454 unsigned int lflags = link->flags;
2455 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2456 int max_tries = 0, try = 0;
2457 struct ata_link *failed_link;
2458 struct ata_device *dev;
2459 unsigned long deadline, now;
2460 ata_reset_fn_t reset;
2461 unsigned long flags;
2462 u32 sstatus;
2463 int nr_unknown, rc;
2464
2465 /*
2466 * Prepare to reset
2467 */
2468 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2469 max_tries++;
2470 if (link->flags & ATA_LFLAG_RST_ONCE)
2471 max_tries = 1;
2472 if (link->flags & ATA_LFLAG_NO_HRST)
2473 hardreset = NULL;
2474 if (link->flags & ATA_LFLAG_NO_SRST)
2475 softreset = NULL;
2476
2477 /* make sure each reset attempt is at least COOL_DOWN apart */
2478 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2479 now = jiffies;
2480 WARN_ON(time_after(ehc->last_reset, now));
2481 deadline = ata_deadline(ehc->last_reset,
2482 ATA_EH_RESET_COOL_DOWN);
2483 if (time_before(now, deadline))
2484 schedule_timeout_uninterruptible(deadline - now);
2485 }
2486
2487 spin_lock_irqsave(ap->lock, flags);
2488 ap->pflags |= ATA_PFLAG_RESETTING;
2489 spin_unlock_irqrestore(ap->lock, flags);
2490
2491 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2492
2493 ata_for_each_dev(dev, link, ALL) {
2494 /* If we issue an SRST then an ATA drive (not ATAPI)
2495 * may change configuration and be in PIO0 timing. If
2496 * we do a hard reset (or are coming from power on)
2497 * this is true for ATA or ATAPI. Until we've set a
2498 * suitable controller mode we should not touch the
2499 * bus as we may be talking too fast.
2500 */
2501 dev->pio_mode = XFER_PIO_0;
2502 dev->dma_mode = 0xff;
2503
2504 /* If the controller has a pio mode setup function
2505 * then use it to set the chipset to rights. Don't
2506 * touch the DMA setup as that will be dealt with when
2507 * configuring devices.
2508 */
2509 if (ap->ops->set_piomode)
2510 ap->ops->set_piomode(ap, dev);
2511 }
2512
2513 /* prefer hardreset */
2514 reset = NULL;
2515 ehc->i.action &= ~ATA_EH_RESET;
2516 if (hardreset) {
2517 reset = hardreset;
2518 ehc->i.action |= ATA_EH_HARDRESET;
2519 } else if (softreset) {
2520 reset = softreset;
2521 ehc->i.action |= ATA_EH_SOFTRESET;
2522 }
2523
2524 if (prereset) {
2525 unsigned long deadline = ata_deadline(jiffies,
2526 ATA_EH_PRERESET_TIMEOUT);
2527
2528 if (slave) {
2529 sehc->i.action &= ~ATA_EH_RESET;
2530 sehc->i.action |= ehc->i.action;
2531 }
2532
2533 rc = prereset(link, deadline);
2534
2535 /* If present, do prereset on slave link too. Reset
2536 * is skipped iff both master and slave links report
2537 * -ENOENT or clear ATA_EH_RESET.
2538 */
2539 if (slave && (rc == 0 || rc == -ENOENT)) {
2540 int tmp;
2541
2542 tmp = prereset(slave, deadline);
2543 if (tmp != -ENOENT)
2544 rc = tmp;
2545
2546 ehc->i.action |= sehc->i.action;
2547 }
2548
2549 if (rc) {
2550 if (rc == -ENOENT) {
2551 ata_link_dbg(link, "port disabled--ignoring\n");
2552 ehc->i.action &= ~ATA_EH_RESET;
2553
2554 ata_for_each_dev(dev, link, ALL)
2555 classes[dev->devno] = ATA_DEV_NONE;
2556
2557 rc = 0;
2558 } else
2559 ata_link_err(link,
2560 "prereset failed (errno=%d)\n",
2561 rc);
2562 goto out;
2563 }
2564
2565 /* prereset() might have cleared ATA_EH_RESET. If so,
2566 * bang classes, thaw and return.
2567 */
2568 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2569 ata_for_each_dev(dev, link, ALL)
2570 classes[dev->devno] = ATA_DEV_NONE;
2571 if (ata_port_is_frozen(ap) && ata_is_host_link(link))
2572 ata_eh_thaw_port(ap);
2573 rc = 0;
2574 goto out;
2575 }
2576 }
2577
2578 retry:
2579 /*
2580 * Perform reset
2581 */
2582 if (ata_is_host_link(link))
2583 ata_eh_freeze_port(ap);
2584
2585 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2586
2587 if (reset) {
2588 if (verbose)
2589 ata_link_info(link, "%s resetting link\n",
2590 reset == softreset ? "soft" : "hard");
2591
2592 /* mark that this EH session started with reset */
2593 ehc->last_reset = jiffies;
2594 if (reset == hardreset) {
2595 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2596 trace_ata_link_hardreset_begin(link, classes, deadline);
2597 } else {
2598 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2599 trace_ata_link_softreset_begin(link, classes, deadline);
2600 }
2601
2602 rc = ata_do_reset(link, reset, classes, deadline, true);
2603 if (reset == hardreset)
2604 trace_ata_link_hardreset_end(link, classes, rc);
2605 else
2606 trace_ata_link_softreset_end(link, classes, rc);
2607 if (rc && rc != -EAGAIN) {
2608 failed_link = link;
2609 goto fail;
2610 }
2611
2612 /* hardreset slave link if existent */
2613 if (slave && reset == hardreset) {
2614 int tmp;
2615
2616 if (verbose)
2617 ata_link_info(slave, "hard resetting link\n");
2618
2619 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2620 trace_ata_slave_hardreset_begin(slave, classes,
2621 deadline);
2622 tmp = ata_do_reset(slave, reset, classes, deadline,
2623 false);
2624 trace_ata_slave_hardreset_end(slave, classes, tmp);
2625 switch (tmp) {
2626 case -EAGAIN:
2627 rc = -EAGAIN;
2628 break;
2629 case 0:
2630 break;
2631 default:
2632 failed_link = slave;
2633 rc = tmp;
2634 goto fail;
2635 }
2636 }
2637
2638 /* perform follow-up SRST if necessary */
2639 if (reset == hardreset &&
2640 ata_eh_followup_srst_needed(link, rc)) {
2641 reset = softreset;
2642
2643 if (!reset) {
2644 ata_link_err(link,
2645 "follow-up softreset required but no softreset available\n");
2646 failed_link = link;
2647 rc = -EINVAL;
2648 goto fail;
2649 }
2650
2651 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2652 trace_ata_link_softreset_begin(link, classes, deadline);
2653 rc = ata_do_reset(link, reset, classes, deadline, true);
2654 trace_ata_link_softreset_end(link, classes, rc);
2655 if (rc) {
2656 failed_link = link;
2657 goto fail;
2658 }
2659 }
2660 } else {
2661 if (verbose)
2662 ata_link_info(link,
2663 "no reset method available, skipping reset\n");
2664 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2665 lflags |= ATA_LFLAG_ASSUME_ATA;
2666 }
2667
2668 /*
2669 * Post-reset processing
2670 */
2671 ata_for_each_dev(dev, link, ALL) {
2672 /* After the reset, the device state is PIO 0 and the
2673 * controller state is undefined. Reset also wakes up
2674 * drives from sleeping mode.
2675 */
2676 dev->pio_mode = XFER_PIO_0;
2677 dev->flags &= ~ATA_DFLAG_SLEEPING;
2678
2679 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2680 continue;
2681
2682 /* apply class override */
2683 if (lflags & ATA_LFLAG_ASSUME_ATA)
2684 classes[dev->devno] = ATA_DEV_ATA;
2685 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2686 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2687 }
2688
2689 /* record current link speed */
2690 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2691 link->sata_spd = (sstatus >> 4) & 0xf;
2692 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2693 slave->sata_spd = (sstatus >> 4) & 0xf;
2694
2695 /* thaw the port */
2696 if (ata_is_host_link(link))
2697 ata_eh_thaw_port(ap);
2698
2699 /* postreset() should clear hardware SError. Although SError
2700 * is cleared during link resume, clearing SError here is
2701 * necessary as some PHYs raise hotplug events after SRST.
2702 * This introduces race condition where hotplug occurs between
2703 * reset and here. This race is mediated by cross checking
2704 * link onlineness and classification result later.
2705 */
2706 if (postreset) {
2707 postreset(link, classes);
2708 trace_ata_link_postreset(link, classes, rc);
2709 if (slave) {
2710 postreset(slave, classes);
2711 trace_ata_slave_postreset(slave, classes, rc);
2712 }
2713 }
2714
2715 /*
2716 * Some controllers can't be frozen very well and may set spurious
2717 * error conditions during reset. Clear accumulated error
2718 * information and re-thaw the port if frozen. As reset is the
2719 * final recovery action and we cross check link onlineness against
2720 * device classification later, no hotplug event is lost by this.
2721 */
2722 spin_lock_irqsave(link->ap->lock, flags);
2723 memset(&link->eh_info, 0, sizeof(link->eh_info));
2724 if (slave)
2725 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2726 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2727 spin_unlock_irqrestore(link->ap->lock, flags);
2728
2729 if (ata_port_is_frozen(ap))
2730 ata_eh_thaw_port(ap);
2731
2732 /*
2733 * Make sure onlineness and classification result correspond.
2734 * Hotplug could have happened during reset and some
2735 * controllers fail to wait while a drive is spinning up after
2736 * being hotplugged causing misdetection. By cross checking
2737 * link on/offlineness and classification result, those
2738 * conditions can be reliably detected and retried.
2739 */
2740 nr_unknown = 0;
2741 ata_for_each_dev(dev, link, ALL) {
2742 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2743 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2744 ata_dev_dbg(dev, "link online but device misclassified\n");
2745 classes[dev->devno] = ATA_DEV_NONE;
2746 nr_unknown++;
2747 }
2748 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2749 if (ata_class_enabled(classes[dev->devno]))
2750 ata_dev_dbg(dev,
2751 "link offline, clearing class %d to NONE\n",
2752 classes[dev->devno]);
2753 classes[dev->devno] = ATA_DEV_NONE;
2754 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2755 ata_dev_dbg(dev,
2756 "link status unknown, clearing UNKNOWN to NONE\n");
2757 classes[dev->devno] = ATA_DEV_NONE;
2758 }
2759 }
2760
2761 if (classify && nr_unknown) {
2762 if (try < max_tries) {
2763 ata_link_warn(link,
2764 "link online but %d devices misclassified, retrying\n",
2765 nr_unknown);
2766 failed_link = link;
2767 rc = -EAGAIN;
2768 goto fail;
2769 }
2770 ata_link_warn(link,
2771 "link online but %d devices misclassified, "
2772 "device detection might fail\n", nr_unknown);
2773 }
2774
2775 /* reset successful, schedule revalidation */
2776 ata_eh_done(link, NULL, ATA_EH_RESET);
2777 if (slave)
2778 ata_eh_done(slave, NULL, ATA_EH_RESET);
2779 ehc->last_reset = jiffies; /* update to completion time */
2780 ehc->i.action |= ATA_EH_REVALIDATE;
2781 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2782
2783 rc = 0;
2784 out:
2785 /* clear hotplug flag */
2786 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2787 if (slave)
2788 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2789
2790 spin_lock_irqsave(ap->lock, flags);
2791 ap->pflags &= ~ATA_PFLAG_RESETTING;
2792 spin_unlock_irqrestore(ap->lock, flags);
2793
2794 return rc;
2795
2796 fail:
2797 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2798 if (!ata_is_host_link(link) &&
2799 sata_scr_read(link, SCR_STATUS, &sstatus))
2800 rc = -ERESTART;
2801
2802 if (try >= max_tries) {
2803 /*
2804 * Thaw host port even if reset failed, so that the port
2805 * can be retried on the next phy event. This risks
2806 * repeated EH runs but seems to be a better tradeoff than
2807 * shutting down a port after a botched hotplug attempt.
2808 */
2809 if (ata_is_host_link(link))
2810 ata_eh_thaw_port(ap);
2811 goto out;
2812 }
2813
2814 now = jiffies;
2815 if (time_before(now, deadline)) {
2816 unsigned long delta = deadline - now;
2817
2818 ata_link_warn(failed_link,
2819 "reset failed (errno=%d), retrying in %u secs\n",
2820 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
2821
2822 ata_eh_release(ap);
2823 while (delta)
2824 delta = schedule_timeout_uninterruptible(delta);
2825 ata_eh_acquire(ap);
2826 }
2827
2828 /*
2829 * While disks spinup behind PMP, some controllers fail sending SRST.
2830 * They need to be reset - as well as the PMP - before retrying.
2831 */
2832 if (rc == -ERESTART) {
2833 if (ata_is_host_link(link))
2834 ata_eh_thaw_port(ap);
2835 goto out;
2836 }
2837
2838 if (try == max_tries - 1) {
2839 sata_down_spd_limit(link, 0);
2840 if (slave)
2841 sata_down_spd_limit(slave, 0);
2842 } else if (rc == -EPIPE)
2843 sata_down_spd_limit(failed_link, 0);
2844
2845 if (hardreset)
2846 reset = hardreset;
2847 goto retry;
2848}
2849
2850static inline void ata_eh_pull_park_action(struct ata_port *ap)
2851{
2852 struct ata_link *link;
2853 struct ata_device *dev;
2854 unsigned long flags;
2855
2856 /*
2857 * This function can be thought of as an extended version of
2858 * ata_eh_about_to_do() specially crafted to accommodate the
2859 * requirements of ATA_EH_PARK handling. Since the EH thread
2860 * does not leave the do {} while () loop in ata_eh_recover as
2861 * long as the timeout for a park request to *one* device on
2862 * the port has not expired, and since we still want to pick
2863 * up park requests to other devices on the same port or
2864 * timeout updates for the same device, we have to pull
2865 * ATA_EH_PARK actions from eh_info into eh_context.i
2866 * ourselves at the beginning of each pass over the loop.
2867 *
2868 * Additionally, all write accesses to &ap->park_req_pending
2869 * through reinit_completion() (see below) or complete_all()
2870 * (see ata_scsi_park_store()) are protected by the host lock.
2871 * As a result we have that park_req_pending.done is zero on
2872 * exit from this function, i.e. when ATA_EH_PARK actions for
2873 * *all* devices on port ap have been pulled into the
2874 * respective eh_context structs. If, and only if,
2875 * park_req_pending.done is non-zero by the time we reach
2876 * wait_for_completion_timeout(), another ATA_EH_PARK action
2877 * has been scheduled for at least one of the devices on port
2878 * ap and we have to cycle over the do {} while () loop in
2879 * ata_eh_recover() again.
2880 */
2881
2882 spin_lock_irqsave(ap->lock, flags);
2883 reinit_completion(&ap->park_req_pending);
2884 ata_for_each_link(link, ap, EDGE) {
2885 ata_for_each_dev(dev, link, ALL) {
2886 struct ata_eh_info *ehi = &link->eh_info;
2887
2888 link->eh_context.i.dev_action[dev->devno] |=
2889 ehi->dev_action[dev->devno] & ATA_EH_PARK;
2890 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
2891 }
2892 }
2893 spin_unlock_irqrestore(ap->lock, flags);
2894}
2895
2896static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
2897{
2898 struct ata_eh_context *ehc = &dev->link->eh_context;
2899 struct ata_taskfile tf;
2900 unsigned int err_mask;
2901
2902 ata_tf_init(dev, &tf);
2903 if (park) {
2904 ehc->unloaded_mask |= 1 << dev->devno;
2905 tf.command = ATA_CMD_IDLEIMMEDIATE;
2906 tf.feature = 0x44;
2907 tf.lbal = 0x4c;
2908 tf.lbam = 0x4e;
2909 tf.lbah = 0x55;
2910 } else {
2911 ehc->unloaded_mask &= ~(1 << dev->devno);
2912 tf.command = ATA_CMD_CHK_POWER;
2913 }
2914
2915 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
2916 tf.protocol = ATA_PROT_NODATA;
2917 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
2918 if (park && (err_mask || tf.lbal != 0xc4)) {
2919 ata_dev_err(dev, "head unload failed!\n");
2920 ehc->unloaded_mask &= ~(1 << dev->devno);
2921 }
2922}
2923
2924static int ata_eh_revalidate_and_attach(struct ata_link *link,
2925 struct ata_device **r_failed_dev)
2926{
2927 struct ata_port *ap = link->ap;
2928 struct ata_eh_context *ehc = &link->eh_context;
2929 struct ata_device *dev;
2930 unsigned int new_mask = 0;
2931 unsigned long flags;
2932 int rc = 0;
2933
2934 /* For PATA drive side cable detection to work, IDENTIFY must
2935 * be done backwards such that PDIAG- is released by the slave
2936 * device before the master device is identified.
2937 */
2938 ata_for_each_dev(dev, link, ALL_REVERSE) {
2939 unsigned int action = ata_eh_dev_action(dev);
2940 unsigned int readid_flags = 0;
2941
2942 if (ehc->i.flags & ATA_EHI_DID_RESET)
2943 readid_flags |= ATA_READID_POSTRESET;
2944
2945 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
2946 WARN_ON(dev->class == ATA_DEV_PMP);
2947
2948 /*
2949 * The link may be in a deep sleep, wake it up.
2950 *
2951 * If the link is in deep sleep, ata_phys_link_offline()
2952 * will return true, causing the revalidation to fail,
2953 * which leads to a (potentially) needless hard reset.
2954 *
2955 * ata_eh_recover() will later restore the link policy
2956 * to ap->target_lpm_policy after revalidation is done.
2957 */
2958 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
2959 rc = ata_eh_set_lpm(link, ATA_LPM_MAX_POWER,
2960 r_failed_dev);
2961 if (rc)
2962 goto err;
2963 }
2964
2965 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2966 rc = -EIO;
2967 goto err;
2968 }
2969
2970 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
2971 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
2972 readid_flags);
2973 if (rc)
2974 goto err;
2975
2976 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
2977
2978 /* Configuration may have changed, reconfigure
2979 * transfer mode.
2980 */
2981 ehc->i.flags |= ATA_EHI_SETMODE;
2982
2983 /* schedule the scsi_rescan_device() here */
2984 schedule_work(&(ap->scsi_rescan_task));
2985 } else if (dev->class == ATA_DEV_UNKNOWN &&
2986 ehc->tries[dev->devno] &&
2987 ata_class_enabled(ehc->classes[dev->devno])) {
2988 /* Temporarily set dev->class, it will be
2989 * permanently set once all configurations are
2990 * complete. This is necessary because new
2991 * device configuration is done in two
2992 * separate loops.
2993 */
2994 dev->class = ehc->classes[dev->devno];
2995
2996 if (dev->class == ATA_DEV_PMP)
2997 rc = sata_pmp_attach(dev);
2998 else
2999 rc = ata_dev_read_id(dev, &dev->class,
3000 readid_flags, dev->id);
3001
3002 /* read_id might have changed class, store and reset */
3003 ehc->classes[dev->devno] = dev->class;
3004 dev->class = ATA_DEV_UNKNOWN;
3005
3006 switch (rc) {
3007 case 0:
3008 /* clear error info accumulated during probe */
3009 ata_ering_clear(&dev->ering);
3010 new_mask |= 1 << dev->devno;
3011 break;
3012 case -ENOENT:
3013 /* IDENTIFY was issued to non-existent
3014 * device. No need to reset. Just
3015 * thaw and ignore the device.
3016 */
3017 ata_eh_thaw_port(ap);
3018 break;
3019 default:
3020 goto err;
3021 }
3022 }
3023 }
3024
3025 /* PDIAG- should have been released, ask cable type if post-reset */
3026 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3027 if (ap->ops->cable_detect)
3028 ap->cbl = ap->ops->cable_detect(ap);
3029 ata_force_cbl(ap);
3030 }
3031
3032 /* Configure new devices forward such that user doesn't see
3033 * device detection messages backwards.
3034 */
3035 ata_for_each_dev(dev, link, ALL) {
3036 if (!(new_mask & (1 << dev->devno)))
3037 continue;
3038
3039 dev->class = ehc->classes[dev->devno];
3040
3041 if (dev->class == ATA_DEV_PMP)
3042 continue;
3043
3044 ehc->i.flags |= ATA_EHI_PRINTINFO;
3045 rc = ata_dev_configure(dev);
3046 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3047 if (rc) {
3048 dev->class = ATA_DEV_UNKNOWN;
3049 goto err;
3050 }
3051
3052 spin_lock_irqsave(ap->lock, flags);
3053 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3054 spin_unlock_irqrestore(ap->lock, flags);
3055
3056 /* new device discovered, configure xfermode */
3057 ehc->i.flags |= ATA_EHI_SETMODE;
3058 }
3059
3060 return 0;
3061
3062 err:
3063 *r_failed_dev = dev;
3064 return rc;
3065}
3066
3067/**
3068 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3069 * @link: link on which timings will be programmed
3070 * @r_failed_dev: out parameter for failed device
3071 *
3072 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3073 * ata_set_mode() fails, pointer to the failing device is
3074 * returned in @r_failed_dev.
3075 *
3076 * LOCKING:
3077 * PCI/etc. bus probe sem.
3078 *
3079 * RETURNS:
3080 * 0 on success, negative errno otherwise
3081 */
3082int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3083{
3084 struct ata_port *ap = link->ap;
3085 struct ata_device *dev;
3086 int rc;
3087
3088 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3089 ata_for_each_dev(dev, link, ENABLED) {
3090 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3091 struct ata_ering_entry *ent;
3092
3093 ent = ata_ering_top(&dev->ering);
3094 if (ent)
3095 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3096 }
3097 }
3098
3099 /* has private set_mode? */
3100 if (ap->ops->set_mode)
3101 rc = ap->ops->set_mode(link, r_failed_dev);
3102 else
3103 rc = ata_do_set_mode(link, r_failed_dev);
3104
3105 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3106 ata_for_each_dev(dev, link, ENABLED) {
3107 struct ata_eh_context *ehc = &link->eh_context;
3108 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3109 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3110
3111 if (dev->xfer_mode != saved_xfer_mode ||
3112 ata_ncq_enabled(dev) != saved_ncq)
3113 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3114 }
3115
3116 return rc;
3117}
3118
3119/**
3120 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3121 * @dev: ATAPI device to clear UA for
3122 *
3123 * Resets and other operations can make an ATAPI device raise
3124 * UNIT ATTENTION which causes the next operation to fail. This
3125 * function clears UA.
3126 *
3127 * LOCKING:
3128 * EH context (may sleep).
3129 *
3130 * RETURNS:
3131 * 0 on success, -errno on failure.
3132 */
3133static int atapi_eh_clear_ua(struct ata_device *dev)
3134{
3135 int i;
3136
3137 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3138 u8 *sense_buffer = dev->link->ap->sector_buf;
3139 u8 sense_key = 0;
3140 unsigned int err_mask;
3141
3142 err_mask = atapi_eh_tur(dev, &sense_key);
3143 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3144 ata_dev_warn(dev,
3145 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3146 err_mask);
3147 return -EIO;
3148 }
3149
3150 if (!err_mask || sense_key != UNIT_ATTENTION)
3151 return 0;
3152
3153 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3154 if (err_mask) {
3155 ata_dev_warn(dev, "failed to clear "
3156 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3157 return -EIO;
3158 }
3159 }
3160
3161 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3162 ATA_EH_UA_TRIES);
3163
3164 return 0;
3165}
3166
3167/**
3168 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3169 * @dev: ATA device which may need FLUSH retry
3170 *
3171 * If @dev failed FLUSH, it needs to be reported upper layer
3172 * immediately as it means that @dev failed to remap and already
3173 * lost at least a sector and further FLUSH retrials won't make
3174 * any difference to the lost sector. However, if FLUSH failed
3175 * for other reasons, for example transmission error, FLUSH needs
3176 * to be retried.
3177 *
3178 * This function determines whether FLUSH failure retry is
3179 * necessary and performs it if so.
3180 *
3181 * RETURNS:
3182 * 0 if EH can continue, -errno if EH needs to be repeated.
3183 */
3184static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3185{
3186 struct ata_link *link = dev->link;
3187 struct ata_port *ap = link->ap;
3188 struct ata_queued_cmd *qc;
3189 struct ata_taskfile tf;
3190 unsigned int err_mask;
3191 int rc = 0;
3192
3193 /* did flush fail for this device? */
3194 if (!ata_tag_valid(link->active_tag))
3195 return 0;
3196
3197 qc = __ata_qc_from_tag(ap, link->active_tag);
3198 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3199 qc->tf.command != ATA_CMD_FLUSH))
3200 return 0;
3201
3202 /* if the device failed it, it should be reported to upper layers */
3203 if (qc->err_mask & AC_ERR_DEV)
3204 return 0;
3205
3206 /* flush failed for some other reason, give it another shot */
3207 ata_tf_init(dev, &tf);
3208
3209 tf.command = qc->tf.command;
3210 tf.flags |= ATA_TFLAG_DEVICE;
3211 tf.protocol = ATA_PROT_NODATA;
3212
3213 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3214 tf.command, qc->err_mask);
3215
3216 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3217 if (!err_mask) {
3218 /*
3219 * FLUSH is complete but there's no way to
3220 * successfully complete a failed command from EH.
3221 * Making sure retry is allowed at least once and
3222 * retrying it should do the trick - whatever was in
3223 * the cache is already on the platter and this won't
3224 * cause infinite loop.
3225 */
3226 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3227 } else {
3228 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3229 err_mask);
3230 rc = -EIO;
3231
3232 /* if device failed it, report it to upper layers */
3233 if (err_mask & AC_ERR_DEV) {
3234 qc->err_mask |= AC_ERR_DEV;
3235 qc->result_tf = tf;
3236 if (!ata_port_is_frozen(ap))
3237 rc = 0;
3238 }
3239 }
3240 return rc;
3241}
3242
3243/**
3244 * ata_eh_set_lpm - configure SATA interface power management
3245 * @link: link to configure power management
3246 * @policy: the link power management policy
3247 * @r_failed_dev: out parameter for failed device
3248 *
3249 * Enable SATA Interface power management. This will enable
3250 * Device Interface Power Management (DIPM) for min_power and
3251 * medium_power_with_dipm policies, and then call driver specific
3252 * callbacks for enabling Host Initiated Power management.
3253 *
3254 * LOCKING:
3255 * EH context.
3256 *
3257 * RETURNS:
3258 * 0 on success, -errno on failure.
3259 */
3260static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3261 struct ata_device **r_failed_dev)
3262{
3263 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3264 struct ata_eh_context *ehc = &link->eh_context;
3265 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3266 enum ata_lpm_policy old_policy = link->lpm_policy;
3267 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3268 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3269 unsigned int err_mask;
3270 int rc;
3271
3272 /* if the link or host doesn't do LPM, noop */
3273 if (!IS_ENABLED(CONFIG_SATA_HOST) ||
3274 (link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3275 return 0;
3276
3277 /*
3278 * DIPM is enabled only for MIN_POWER as some devices
3279 * misbehave when the host NACKs transition to SLUMBER. Order
3280 * device and link configurations such that the host always
3281 * allows DIPM requests.
3282 */
3283 ata_for_each_dev(dev, link, ENABLED) {
3284 bool hipm = ata_id_has_hipm(dev->id);
3285 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3286
3287 /* find the first enabled and LPM enabled devices */
3288 if (!link_dev)
3289 link_dev = dev;
3290
3291 if (!lpm_dev && (hipm || dipm))
3292 lpm_dev = dev;
3293
3294 hints &= ~ATA_LPM_EMPTY;
3295 if (!hipm)
3296 hints &= ~ATA_LPM_HIPM;
3297
3298 /* disable DIPM before changing link config */
3299 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3300 err_mask = ata_dev_set_feature(dev,
3301 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3302 if (err_mask && err_mask != AC_ERR_DEV) {
3303 ata_dev_warn(dev,
3304 "failed to disable DIPM, Emask 0x%x\n",
3305 err_mask);
3306 rc = -EIO;
3307 goto fail;
3308 }
3309 }
3310 }
3311
3312 if (ap) {
3313 rc = ap->ops->set_lpm(link, policy, hints);
3314 if (!rc && ap->slave_link)
3315 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3316 } else
3317 rc = sata_pmp_set_lpm(link, policy, hints);
3318
3319 /*
3320 * Attribute link config failure to the first (LPM) enabled
3321 * device on the link.
3322 */
3323 if (rc) {
3324 if (rc == -EOPNOTSUPP) {
3325 link->flags |= ATA_LFLAG_NO_LPM;
3326 return 0;
3327 }
3328 dev = lpm_dev ? lpm_dev : link_dev;
3329 goto fail;
3330 }
3331
3332 /*
3333 * Low level driver acked the transition. Issue DIPM command
3334 * with the new policy set.
3335 */
3336 link->lpm_policy = policy;
3337 if (ap && ap->slave_link)
3338 ap->slave_link->lpm_policy = policy;
3339
3340 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3341 ata_for_each_dev(dev, link, ENABLED) {
3342 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3343 ata_id_has_dipm(dev->id)) {
3344 err_mask = ata_dev_set_feature(dev,
3345 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3346 if (err_mask && err_mask != AC_ERR_DEV) {
3347 ata_dev_warn(dev,
3348 "failed to enable DIPM, Emask 0x%x\n",
3349 err_mask);
3350 rc = -EIO;
3351 goto fail;
3352 }
3353 }
3354 }
3355
3356 link->last_lpm_change = jiffies;
3357 link->flags |= ATA_LFLAG_CHANGED;
3358
3359 return 0;
3360
3361fail:
3362 /* restore the old policy */
3363 link->lpm_policy = old_policy;
3364 if (ap && ap->slave_link)
3365 ap->slave_link->lpm_policy = old_policy;
3366
3367 /* if no device or only one more chance is left, disable LPM */
3368 if (!dev || ehc->tries[dev->devno] <= 2) {
3369 ata_link_warn(link, "disabling LPM on the link\n");
3370 link->flags |= ATA_LFLAG_NO_LPM;
3371 }
3372 if (r_failed_dev)
3373 *r_failed_dev = dev;
3374 return rc;
3375}
3376
3377int ata_link_nr_enabled(struct ata_link *link)
3378{
3379 struct ata_device *dev;
3380 int cnt = 0;
3381
3382 ata_for_each_dev(dev, link, ENABLED)
3383 cnt++;
3384 return cnt;
3385}
3386
3387static int ata_link_nr_vacant(struct ata_link *link)
3388{
3389 struct ata_device *dev;
3390 int cnt = 0;
3391
3392 ata_for_each_dev(dev, link, ALL)
3393 if (dev->class == ATA_DEV_UNKNOWN)
3394 cnt++;
3395 return cnt;
3396}
3397
3398static int ata_eh_skip_recovery(struct ata_link *link)
3399{
3400 struct ata_port *ap = link->ap;
3401 struct ata_eh_context *ehc = &link->eh_context;
3402 struct ata_device *dev;
3403
3404 /* skip disabled links */
3405 if (link->flags & ATA_LFLAG_DISABLED)
3406 return 1;
3407
3408 /* skip if explicitly requested */
3409 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3410 return 1;
3411
3412 /* thaw frozen port and recover failed devices */
3413 if (ata_port_is_frozen(ap) || ata_link_nr_enabled(link))
3414 return 0;
3415
3416 /* reset at least once if reset is requested */
3417 if ((ehc->i.action & ATA_EH_RESET) &&
3418 !(ehc->i.flags & ATA_EHI_DID_RESET))
3419 return 0;
3420
3421 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3422 ata_for_each_dev(dev, link, ALL) {
3423 if (dev->class == ATA_DEV_UNKNOWN &&
3424 ehc->classes[dev->devno] != ATA_DEV_NONE)
3425 return 0;
3426 }
3427
3428 return 1;
3429}
3430
3431static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3432{
3433 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3434 u64 now = get_jiffies_64();
3435 int *trials = void_arg;
3436
3437 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3438 (ent->timestamp < now - min(now, interval)))
3439 return -1;
3440
3441 (*trials)++;
3442 return 0;
3443}
3444
3445static int ata_eh_schedule_probe(struct ata_device *dev)
3446{
3447 struct ata_eh_context *ehc = &dev->link->eh_context;
3448 struct ata_link *link = ata_dev_phys_link(dev);
3449 int trials = 0;
3450
3451 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3452 (ehc->did_probe_mask & (1 << dev->devno)))
3453 return 0;
3454
3455 ata_eh_detach_dev(dev);
3456 ata_dev_init(dev);
3457 ehc->did_probe_mask |= (1 << dev->devno);
3458 ehc->i.action |= ATA_EH_RESET;
3459 ehc->saved_xfer_mode[dev->devno] = 0;
3460 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3461
3462 /* the link maybe in a deep sleep, wake it up */
3463 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3464 if (ata_is_host_link(link))
3465 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3466 ATA_LPM_EMPTY);
3467 else
3468 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3469 ATA_LPM_EMPTY);
3470 }
3471
3472 /* Record and count probe trials on the ering. The specific
3473 * error mask used is irrelevant. Because a successful device
3474 * detection clears the ering, this count accumulates only if
3475 * there are consecutive failed probes.
3476 *
3477 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3478 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3479 * forced to 1.5Gbps.
3480 *
3481 * This is to work around cases where failed link speed
3482 * negotiation results in device misdetection leading to
3483 * infinite DEVXCHG or PHRDY CHG events.
3484 */
3485 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3486 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3487
3488 if (trials > ATA_EH_PROBE_TRIALS)
3489 sata_down_spd_limit(link, 1);
3490
3491 return 1;
3492}
3493
3494static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3495{
3496 struct ata_eh_context *ehc = &dev->link->eh_context;
3497
3498 /* -EAGAIN from EH routine indicates retry without prejudice.
3499 * The requester is responsible for ensuring forward progress.
3500 */
3501 if (err != -EAGAIN)
3502 ehc->tries[dev->devno]--;
3503
3504 switch (err) {
3505 case -ENODEV:
3506 /* device missing or wrong IDENTIFY data, schedule probing */
3507 ehc->i.probe_mask |= (1 << dev->devno);
3508 fallthrough;
3509 case -EINVAL:
3510 /* give it just one more chance */
3511 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3512 fallthrough;
3513 case -EIO:
3514 if (ehc->tries[dev->devno] == 1) {
3515 /* This is the last chance, better to slow
3516 * down than lose it.
3517 */
3518 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3519 if (dev->pio_mode > XFER_PIO_0)
3520 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3521 }
3522 }
3523
3524 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3525 /* disable device if it has used up all its chances */
3526 ata_dev_disable(dev);
3527
3528 /* detach if offline */
3529 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3530 ata_eh_detach_dev(dev);
3531
3532 /* schedule probe if necessary */
3533 if (ata_eh_schedule_probe(dev)) {
3534 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3535 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3536 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3537 }
3538
3539 return 1;
3540 } else {
3541 ehc->i.action |= ATA_EH_RESET;
3542 return 0;
3543 }
3544}
3545
3546/**
3547 * ata_eh_recover - recover host port after error
3548 * @ap: host port to recover
3549 * @prereset: prereset method (can be NULL)
3550 * @softreset: softreset method (can be NULL)
3551 * @hardreset: hardreset method (can be NULL)
3552 * @postreset: postreset method (can be NULL)
3553 * @r_failed_link: out parameter for failed link
3554 *
3555 * This is the alpha and omega, eum and yang, heart and soul of
3556 * libata exception handling. On entry, actions required to
3557 * recover each link and hotplug requests are recorded in the
3558 * link's eh_context. This function executes all the operations
3559 * with appropriate retrials and fallbacks to resurrect failed
3560 * devices, detach goners and greet newcomers.
3561 *
3562 * LOCKING:
3563 * Kernel thread context (may sleep).
3564 *
3565 * RETURNS:
3566 * 0 on success, -errno on failure.
3567 */
3568int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3569 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3570 ata_postreset_fn_t postreset,
3571 struct ata_link **r_failed_link)
3572{
3573 struct ata_link *link;
3574 struct ata_device *dev;
3575 int rc, nr_fails;
3576 unsigned long flags, deadline;
3577
3578 /* prep for recovery */
3579 ata_for_each_link(link, ap, EDGE) {
3580 struct ata_eh_context *ehc = &link->eh_context;
3581
3582 /* re-enable link? */
3583 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3584 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3585 spin_lock_irqsave(ap->lock, flags);
3586 link->flags &= ~ATA_LFLAG_DISABLED;
3587 spin_unlock_irqrestore(ap->lock, flags);
3588 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3589 }
3590
3591 ata_for_each_dev(dev, link, ALL) {
3592 if (link->flags & ATA_LFLAG_NO_RETRY)
3593 ehc->tries[dev->devno] = 1;
3594 else
3595 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3596
3597 /* collect port action mask recorded in dev actions */
3598 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3599 ~ATA_EH_PERDEV_MASK;
3600 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3601
3602 /* process hotplug request */
3603 if (dev->flags & ATA_DFLAG_DETACH)
3604 ata_eh_detach_dev(dev);
3605
3606 /* schedule probe if necessary */
3607 if (!ata_dev_enabled(dev))
3608 ata_eh_schedule_probe(dev);
3609 }
3610 }
3611
3612 retry:
3613 rc = 0;
3614
3615 /* if UNLOADING, finish immediately */
3616 if (ap->pflags & ATA_PFLAG_UNLOADING)
3617 goto out;
3618
3619 /* prep for EH */
3620 ata_for_each_link(link, ap, EDGE) {
3621 struct ata_eh_context *ehc = &link->eh_context;
3622
3623 /* skip EH if possible. */
3624 if (ata_eh_skip_recovery(link))
3625 ehc->i.action = 0;
3626
3627 ata_for_each_dev(dev, link, ALL)
3628 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3629 }
3630
3631 /* reset */
3632 ata_for_each_link(link, ap, EDGE) {
3633 struct ata_eh_context *ehc = &link->eh_context;
3634
3635 if (!(ehc->i.action & ATA_EH_RESET))
3636 continue;
3637
3638 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3639 prereset, softreset, hardreset, postreset);
3640 if (rc) {
3641 ata_link_err(link, "reset failed, giving up\n");
3642 goto out;
3643 }
3644 }
3645
3646 do {
3647 unsigned long now;
3648
3649 /*
3650 * clears ATA_EH_PARK in eh_info and resets
3651 * ap->park_req_pending
3652 */
3653 ata_eh_pull_park_action(ap);
3654
3655 deadline = jiffies;
3656 ata_for_each_link(link, ap, EDGE) {
3657 ata_for_each_dev(dev, link, ALL) {
3658 struct ata_eh_context *ehc = &link->eh_context;
3659 unsigned long tmp;
3660
3661 if (dev->class != ATA_DEV_ATA &&
3662 dev->class != ATA_DEV_ZAC)
3663 continue;
3664 if (!(ehc->i.dev_action[dev->devno] &
3665 ATA_EH_PARK))
3666 continue;
3667 tmp = dev->unpark_deadline;
3668 if (time_before(deadline, tmp))
3669 deadline = tmp;
3670 else if (time_before_eq(tmp, jiffies))
3671 continue;
3672 if (ehc->unloaded_mask & (1 << dev->devno))
3673 continue;
3674
3675 ata_eh_park_issue_cmd(dev, 1);
3676 }
3677 }
3678
3679 now = jiffies;
3680 if (time_before_eq(deadline, now))
3681 break;
3682
3683 ata_eh_release(ap);
3684 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3685 deadline - now);
3686 ata_eh_acquire(ap);
3687 } while (deadline);
3688 ata_for_each_link(link, ap, EDGE) {
3689 ata_for_each_dev(dev, link, ALL) {
3690 if (!(link->eh_context.unloaded_mask &
3691 (1 << dev->devno)))
3692 continue;
3693
3694 ata_eh_park_issue_cmd(dev, 0);
3695 ata_eh_done(link, dev, ATA_EH_PARK);
3696 }
3697 }
3698
3699 /* the rest */
3700 nr_fails = 0;
3701 ata_for_each_link(link, ap, PMP_FIRST) {
3702 struct ata_eh_context *ehc = &link->eh_context;
3703
3704 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3705 goto config_lpm;
3706
3707 /* revalidate existing devices and attach new ones */
3708 rc = ata_eh_revalidate_and_attach(link, &dev);
3709 if (rc)
3710 goto rest_fail;
3711
3712 /* if PMP got attached, return, pmp EH will take care of it */
3713 if (link->device->class == ATA_DEV_PMP) {
3714 ehc->i.action = 0;
3715 return 0;
3716 }
3717
3718 /* configure transfer mode if necessary */
3719 if (ehc->i.flags & ATA_EHI_SETMODE) {
3720 rc = ata_set_mode(link, &dev);
3721 if (rc)
3722 goto rest_fail;
3723 ehc->i.flags &= ~ATA_EHI_SETMODE;
3724 }
3725
3726 /* If reset has been issued, clear UA to avoid
3727 * disrupting the current users of the device.
3728 */
3729 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3730 ata_for_each_dev(dev, link, ALL) {
3731 if (dev->class != ATA_DEV_ATAPI)
3732 continue;
3733 rc = atapi_eh_clear_ua(dev);
3734 if (rc)
3735 goto rest_fail;
3736 if (zpodd_dev_enabled(dev))
3737 zpodd_post_poweron(dev);
3738 }
3739 }
3740
3741 /* retry flush if necessary */
3742 ata_for_each_dev(dev, link, ALL) {
3743 if (dev->class != ATA_DEV_ATA &&
3744 dev->class != ATA_DEV_ZAC)
3745 continue;
3746 rc = ata_eh_maybe_retry_flush(dev);
3747 if (rc)
3748 goto rest_fail;
3749 }
3750
3751 config_lpm:
3752 /* configure link power saving */
3753 if (link->lpm_policy != ap->target_lpm_policy) {
3754 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3755 if (rc)
3756 goto rest_fail;
3757 }
3758
3759 /* this link is okay now */
3760 ehc->i.flags = 0;
3761 continue;
3762
3763 rest_fail:
3764 nr_fails++;
3765 if (dev)
3766 ata_eh_handle_dev_fail(dev, rc);
3767
3768 if (ata_port_is_frozen(ap)) {
3769 /* PMP reset requires working host port.
3770 * Can't retry if it's frozen.
3771 */
3772 if (sata_pmp_attached(ap))
3773 goto out;
3774 break;
3775 }
3776 }
3777
3778 if (nr_fails)
3779 goto retry;
3780
3781 out:
3782 if (rc && r_failed_link)
3783 *r_failed_link = link;
3784
3785 return rc;
3786}
3787
3788/**
3789 * ata_eh_finish - finish up EH
3790 * @ap: host port to finish EH for
3791 *
3792 * Recovery is complete. Clean up EH states and retry or finish
3793 * failed qcs.
3794 *
3795 * LOCKING:
3796 * None.
3797 */
3798void ata_eh_finish(struct ata_port *ap)
3799{
3800 struct ata_queued_cmd *qc;
3801 int tag;
3802
3803 /* retry or finish qcs */
3804 ata_qc_for_each_raw(ap, qc, tag) {
3805 if (!(qc->flags & ATA_QCFLAG_FAILED))
3806 continue;
3807
3808 if (qc->err_mask) {
3809 /* FIXME: Once EH migration is complete,
3810 * generate sense data in this function,
3811 * considering both err_mask and tf.
3812 */
3813 if (qc->flags & ATA_QCFLAG_RETRY)
3814 ata_eh_qc_retry(qc);
3815 else
3816 ata_eh_qc_complete(qc);
3817 } else {
3818 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
3819 ata_eh_qc_complete(qc);
3820 } else {
3821 /* feed zero TF to sense generation */
3822 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
3823 ata_eh_qc_retry(qc);
3824 }
3825 }
3826 }
3827
3828 /* make sure nr_active_links is zero after EH */
3829 WARN_ON(ap->nr_active_links);
3830 ap->nr_active_links = 0;
3831}
3832
3833/**
3834 * ata_do_eh - do standard error handling
3835 * @ap: host port to handle error for
3836 *
3837 * @prereset: prereset method (can be NULL)
3838 * @softreset: softreset method (can be NULL)
3839 * @hardreset: hardreset method (can be NULL)
3840 * @postreset: postreset method (can be NULL)
3841 *
3842 * Perform standard error handling sequence.
3843 *
3844 * LOCKING:
3845 * Kernel thread context (may sleep).
3846 */
3847void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
3848 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3849 ata_postreset_fn_t postreset)
3850{
3851 struct ata_device *dev;
3852 int rc;
3853
3854 ata_eh_autopsy(ap);
3855 ata_eh_report(ap);
3856
3857 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
3858 NULL);
3859 if (rc) {
3860 ata_for_each_dev(dev, &ap->link, ALL)
3861 ata_dev_disable(dev);
3862 }
3863
3864 ata_eh_finish(ap);
3865}
3866
3867/**
3868 * ata_std_error_handler - standard error handler
3869 * @ap: host port to handle error for
3870 *
3871 * Standard error handler
3872 *
3873 * LOCKING:
3874 * Kernel thread context (may sleep).
3875 */
3876void ata_std_error_handler(struct ata_port *ap)
3877{
3878 struct ata_port_operations *ops = ap->ops;
3879 ata_reset_fn_t hardreset = ops->hardreset;
3880
3881 /* ignore built-in hardreset if SCR access is not available */
3882 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
3883 hardreset = NULL;
3884
3885 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
3886}
3887EXPORT_SYMBOL_GPL(ata_std_error_handler);
3888
3889#ifdef CONFIG_PM
3890/**
3891 * ata_eh_handle_port_suspend - perform port suspend operation
3892 * @ap: port to suspend
3893 *
3894 * Suspend @ap.
3895 *
3896 * LOCKING:
3897 * Kernel thread context (may sleep).
3898 */
3899static void ata_eh_handle_port_suspend(struct ata_port *ap)
3900{
3901 unsigned long flags;
3902 int rc = 0;
3903 struct ata_device *dev;
3904
3905 /* are we suspending? */
3906 spin_lock_irqsave(ap->lock, flags);
3907 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3908 ap->pm_mesg.event & PM_EVENT_RESUME) {
3909 spin_unlock_irqrestore(ap->lock, flags);
3910 return;
3911 }
3912 spin_unlock_irqrestore(ap->lock, flags);
3913
3914 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
3915
3916 /*
3917 * If we have a ZPODD attached, check its zero
3918 * power ready status before the port is frozen.
3919 * Only needed for runtime suspend.
3920 */
3921 if (PMSG_IS_AUTO(ap->pm_mesg)) {
3922 ata_for_each_dev(dev, &ap->link, ENABLED) {
3923 if (zpodd_dev_enabled(dev))
3924 zpodd_on_suspend(dev);
3925 }
3926 }
3927
3928 /* suspend */
3929 ata_eh_freeze_port(ap);
3930
3931 if (ap->ops->port_suspend)
3932 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
3933
3934 ata_acpi_set_state(ap, ap->pm_mesg);
3935
3936 /* update the flags */
3937 spin_lock_irqsave(ap->lock, flags);
3938
3939 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
3940 if (rc == 0)
3941 ap->pflags |= ATA_PFLAG_SUSPENDED;
3942 else if (ata_port_is_frozen(ap))
3943 ata_port_schedule_eh(ap);
3944
3945 spin_unlock_irqrestore(ap->lock, flags);
3946
3947 return;
3948}
3949
3950/**
3951 * ata_eh_handle_port_resume - perform port resume operation
3952 * @ap: port to resume
3953 *
3954 * Resume @ap.
3955 *
3956 * LOCKING:
3957 * Kernel thread context (may sleep).
3958 */
3959static void ata_eh_handle_port_resume(struct ata_port *ap)
3960{
3961 struct ata_link *link;
3962 struct ata_device *dev;
3963 unsigned long flags;
3964
3965 /* are we resuming? */
3966 spin_lock_irqsave(ap->lock, flags);
3967 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
3968 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
3969 spin_unlock_irqrestore(ap->lock, flags);
3970 return;
3971 }
3972 spin_unlock_irqrestore(ap->lock, flags);
3973
3974 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
3975
3976 /*
3977 * Error timestamps are in jiffies which doesn't run while
3978 * suspended and PHY events during resume isn't too uncommon.
3979 * When the two are combined, it can lead to unnecessary speed
3980 * downs if the machine is suspended and resumed repeatedly.
3981 * Clear error history.
3982 */
3983 ata_for_each_link(link, ap, HOST_FIRST)
3984 ata_for_each_dev(dev, link, ALL)
3985 ata_ering_clear(&dev->ering);
3986
3987 ata_acpi_set_state(ap, ap->pm_mesg);
3988
3989 if (ap->ops->port_resume)
3990 ap->ops->port_resume(ap);
3991
3992 /* tell ACPI that we're resuming */
3993 ata_acpi_on_resume(ap);
3994
3995 /* update the flags */
3996 spin_lock_irqsave(ap->lock, flags);
3997 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
3998 spin_unlock_irqrestore(ap->lock, flags);
3999}
4000#endif /* CONFIG_PM */