blob: ccc80ff57eb2018adf9657aaa284a5aa3b1e52a4 [file] [log] [blame]
xjb04a4022021-11-25 15:01:52 +08001/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Tejun Heo <tj@kernel.org>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/driver-api/libata.rst
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
35#include <linux/kernel.h>
36#include <linux/blkdev.h>
37#include <linux/export.h>
38#include <linux/pci.h>
39#include <scsi/scsi.h>
40#include <scsi/scsi_host.h>
41#include <scsi/scsi_eh.h>
42#include <scsi/scsi_device.h>
43#include <scsi/scsi_cmnd.h>
44#include <scsi/scsi_dbg.h>
45#include "../scsi/scsi_transport_api.h"
46
47#include <linux/libata.h>
48
49#include <trace/events/libata.h>
50#include "libata.h"
51
52enum {
53 /* speed down verdicts */
54 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
55 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
56 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
57 ATA_EH_SPDN_KEEP_ERRORS = (1 << 3),
58
59 /* error flags */
60 ATA_EFLAG_IS_IO = (1 << 0),
61 ATA_EFLAG_DUBIOUS_XFER = (1 << 1),
62 ATA_EFLAG_OLD_ER = (1 << 31),
63
64 /* error categories */
65 ATA_ECAT_NONE = 0,
66 ATA_ECAT_ATA_BUS = 1,
67 ATA_ECAT_TOUT_HSM = 2,
68 ATA_ECAT_UNK_DEV = 3,
69 ATA_ECAT_DUBIOUS_NONE = 4,
70 ATA_ECAT_DUBIOUS_ATA_BUS = 5,
71 ATA_ECAT_DUBIOUS_TOUT_HSM = 6,
72 ATA_ECAT_DUBIOUS_UNK_DEV = 7,
73 ATA_ECAT_NR = 8,
74
75 ATA_EH_CMD_DFL_TIMEOUT = 5000,
76
77 /* always put at least this amount of time between resets */
78 ATA_EH_RESET_COOL_DOWN = 5000,
79
80 /* Waiting in ->prereset can never be reliable. It's
81 * sometimes nice to wait there but it can't be depended upon;
82 * otherwise, we wouldn't be resetting. Just give it enough
83 * time for most drives to spin up.
84 */
85 ATA_EH_PRERESET_TIMEOUT = 10000,
86 ATA_EH_FASTDRAIN_INTERVAL = 3000,
87
88 ATA_EH_UA_TRIES = 5,
89
90 /* probe speed down parameters, see ata_eh_schedule_probe() */
91 ATA_EH_PROBE_TRIAL_INTERVAL = 60000, /* 1 min */
92 ATA_EH_PROBE_TRIALS = 2,
93};
94
95/* The following table determines how we sequence resets. Each entry
96 * represents timeout for that try. The first try can be soft or
97 * hardreset. All others are hardreset if available. In most cases
98 * the first reset w/ 10sec timeout should succeed. Following entries
99 * are mostly for error handling, hotplug and those outlier devices that
100 * take an exceptionally long time to recover from reset.
101 */
102static const unsigned long ata_eh_reset_timeouts[] = {
103 10000, /* most drives spin up by 10sec */
104 10000, /* > 99% working drives spin up before 20sec */
105 35000, /* give > 30 secs of idleness for outlier devices */
106 5000, /* and sweet one last chance */
107 ULONG_MAX, /* > 1 min has elapsed, give up */
108};
109
110static const unsigned long ata_eh_identify_timeouts[] = {
111 5000, /* covers > 99% of successes and not too boring on failures */
112 10000, /* combined time till here is enough even for media access */
113 30000, /* for true idiots */
114 ULONG_MAX,
115};
116
117static const unsigned long ata_eh_flush_timeouts[] = {
118 15000, /* be generous with flush */
119 15000, /* ditto */
120 30000, /* and even more generous */
121 ULONG_MAX,
122};
123
124static const unsigned long ata_eh_other_timeouts[] = {
125 5000, /* same rationale as identify timeout */
126 10000, /* ditto */
127 /* but no merciful 30sec for other commands, it just isn't worth it */
128 ULONG_MAX,
129};
130
131struct ata_eh_cmd_timeout_ent {
132 const u8 *commands;
133 const unsigned long *timeouts;
134};
135
136/* The following table determines timeouts to use for EH internal
137 * commands. Each table entry is a command class and matches the
138 * commands the entry applies to and the timeout table to use.
139 *
140 * On the retry after a command timed out, the next timeout value from
141 * the table is used. If the table doesn't contain further entries,
142 * the last value is used.
143 *
144 * ehc->cmd_timeout_idx keeps track of which timeout to use per
145 * command class, so if SET_FEATURES times out on the first try, the
146 * next try will use the second timeout value only for that class.
147 */
148#define CMDS(cmds...) (const u8 []){ cmds, 0 }
149static const struct ata_eh_cmd_timeout_ent
150ata_eh_cmd_timeout_table[ATA_EH_CMD_TIMEOUT_TABLE_SIZE] = {
151 { .commands = CMDS(ATA_CMD_ID_ATA, ATA_CMD_ID_ATAPI),
152 .timeouts = ata_eh_identify_timeouts, },
153 { .commands = CMDS(ATA_CMD_READ_NATIVE_MAX, ATA_CMD_READ_NATIVE_MAX_EXT),
154 .timeouts = ata_eh_other_timeouts, },
155 { .commands = CMDS(ATA_CMD_SET_MAX, ATA_CMD_SET_MAX_EXT),
156 .timeouts = ata_eh_other_timeouts, },
157 { .commands = CMDS(ATA_CMD_SET_FEATURES),
158 .timeouts = ata_eh_other_timeouts, },
159 { .commands = CMDS(ATA_CMD_INIT_DEV_PARAMS),
160 .timeouts = ata_eh_other_timeouts, },
161 { .commands = CMDS(ATA_CMD_FLUSH, ATA_CMD_FLUSH_EXT),
162 .timeouts = ata_eh_flush_timeouts },
163};
164#undef CMDS
165
166static void __ata_port_freeze(struct ata_port *ap);
167#ifdef CONFIG_PM
168static void ata_eh_handle_port_suspend(struct ata_port *ap);
169static void ata_eh_handle_port_resume(struct ata_port *ap);
170#else /* CONFIG_PM */
171static void ata_eh_handle_port_suspend(struct ata_port *ap)
172{ }
173
174static void ata_eh_handle_port_resume(struct ata_port *ap)
175{ }
176#endif /* CONFIG_PM */
177
178static __printf(2, 0) void __ata_ehi_pushv_desc(struct ata_eh_info *ehi,
179 const char *fmt, va_list args)
180{
181 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
182 ATA_EH_DESC_LEN - ehi->desc_len,
183 fmt, args);
184}
185
186/**
187 * __ata_ehi_push_desc - push error description without adding separator
188 * @ehi: target EHI
189 * @fmt: printf format string
190 *
191 * Format string according to @fmt and append it to @ehi->desc.
192 *
193 * LOCKING:
194 * spin_lock_irqsave(host lock)
195 */
196void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
197{
198 va_list args;
199
200 va_start(args, fmt);
201 __ata_ehi_pushv_desc(ehi, fmt, args);
202 va_end(args);
203}
204
205/**
206 * ata_ehi_push_desc - push error description with separator
207 * @ehi: target EHI
208 * @fmt: printf format string
209 *
210 * Format string according to @fmt and append it to @ehi->desc.
211 * If @ehi->desc is not empty, ", " is added in-between.
212 *
213 * LOCKING:
214 * spin_lock_irqsave(host lock)
215 */
216void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
217{
218 va_list args;
219
220 if (ehi->desc_len)
221 __ata_ehi_push_desc(ehi, ", ");
222
223 va_start(args, fmt);
224 __ata_ehi_pushv_desc(ehi, fmt, args);
225 va_end(args);
226}
227
228/**
229 * ata_ehi_clear_desc - clean error description
230 * @ehi: target EHI
231 *
232 * Clear @ehi->desc.
233 *
234 * LOCKING:
235 * spin_lock_irqsave(host lock)
236 */
237void ata_ehi_clear_desc(struct ata_eh_info *ehi)
238{
239 ehi->desc[0] = '\0';
240 ehi->desc_len = 0;
241}
242
243/**
244 * ata_port_desc - append port description
245 * @ap: target ATA port
246 * @fmt: printf format string
247 *
248 * Format string according to @fmt and append it to port
249 * description. If port description is not empty, " " is added
250 * in-between. This function is to be used while initializing
251 * ata_host. The description is printed on host registration.
252 *
253 * LOCKING:
254 * None.
255 */
256void ata_port_desc(struct ata_port *ap, const char *fmt, ...)
257{
258 va_list args;
259
260 WARN_ON(!(ap->pflags & ATA_PFLAG_INITIALIZING));
261
262 if (ap->link.eh_info.desc_len)
263 __ata_ehi_push_desc(&ap->link.eh_info, " ");
264
265 va_start(args, fmt);
266 __ata_ehi_pushv_desc(&ap->link.eh_info, fmt, args);
267 va_end(args);
268}
269
270#ifdef CONFIG_PCI
271
272/**
273 * ata_port_pbar_desc - append PCI BAR description
274 * @ap: target ATA port
275 * @bar: target PCI BAR
276 * @offset: offset into PCI BAR
277 * @name: name of the area
278 *
279 * If @offset is negative, this function formats a string which
280 * contains the name, address, size and type of the BAR and
281 * appends it to the port description. If @offset is zero or
282 * positive, only name and offsetted address is appended.
283 *
284 * LOCKING:
285 * None.
286 */
287void ata_port_pbar_desc(struct ata_port *ap, int bar, ssize_t offset,
288 const char *name)
289{
290 struct pci_dev *pdev = to_pci_dev(ap->host->dev);
291 char *type = "";
292 unsigned long long start, len;
293
294 if (pci_resource_flags(pdev, bar) & IORESOURCE_MEM)
295 type = "m";
296 else if (pci_resource_flags(pdev, bar) & IORESOURCE_IO)
297 type = "i";
298
299 start = (unsigned long long)pci_resource_start(pdev, bar);
300 len = (unsigned long long)pci_resource_len(pdev, bar);
301
302 if (offset < 0)
303 ata_port_desc(ap, "%s %s%llu@0x%llx", name, type, len, start);
304 else
305 ata_port_desc(ap, "%s 0x%llx", name,
306 start + (unsigned long long)offset);
307}
308
309#endif /* CONFIG_PCI */
310
311static int ata_lookup_timeout_table(u8 cmd)
312{
313 int i;
314
315 for (i = 0; i < ATA_EH_CMD_TIMEOUT_TABLE_SIZE; i++) {
316 const u8 *cur;
317
318 for (cur = ata_eh_cmd_timeout_table[i].commands; *cur; cur++)
319 if (*cur == cmd)
320 return i;
321 }
322
323 return -1;
324}
325
326/**
327 * ata_internal_cmd_timeout - determine timeout for an internal command
328 * @dev: target device
329 * @cmd: internal command to be issued
330 *
331 * Determine timeout for internal command @cmd for @dev.
332 *
333 * LOCKING:
334 * EH context.
335 *
336 * RETURNS:
337 * Determined timeout.
338 */
339unsigned long ata_internal_cmd_timeout(struct ata_device *dev, u8 cmd)
340{
341 struct ata_eh_context *ehc = &dev->link->eh_context;
342 int ent = ata_lookup_timeout_table(cmd);
343 int idx;
344
345 if (ent < 0)
346 return ATA_EH_CMD_DFL_TIMEOUT;
347
348 idx = ehc->cmd_timeout_idx[dev->devno][ent];
349 return ata_eh_cmd_timeout_table[ent].timeouts[idx];
350}
351
352/**
353 * ata_internal_cmd_timed_out - notification for internal command timeout
354 * @dev: target device
355 * @cmd: internal command which timed out
356 *
357 * Notify EH that internal command @cmd for @dev timed out. This
358 * function should be called only for commands whose timeouts are
359 * determined using ata_internal_cmd_timeout().
360 *
361 * LOCKING:
362 * EH context.
363 */
364void ata_internal_cmd_timed_out(struct ata_device *dev, u8 cmd)
365{
366 struct ata_eh_context *ehc = &dev->link->eh_context;
367 int ent = ata_lookup_timeout_table(cmd);
368 int idx;
369
370 if (ent < 0)
371 return;
372
373 idx = ehc->cmd_timeout_idx[dev->devno][ent];
374 if (ata_eh_cmd_timeout_table[ent].timeouts[idx + 1] != ULONG_MAX)
375 ehc->cmd_timeout_idx[dev->devno][ent]++;
376}
377
378static void ata_ering_record(struct ata_ering *ering, unsigned int eflags,
379 unsigned int err_mask)
380{
381 struct ata_ering_entry *ent;
382
383 WARN_ON(!err_mask);
384
385 ering->cursor++;
386 ering->cursor %= ATA_ERING_SIZE;
387
388 ent = &ering->ring[ering->cursor];
389 ent->eflags = eflags;
390 ent->err_mask = err_mask;
391 ent->timestamp = get_jiffies_64();
392}
393
394static struct ata_ering_entry *ata_ering_top(struct ata_ering *ering)
395{
396 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
397
398 if (ent->err_mask)
399 return ent;
400 return NULL;
401}
402
403int ata_ering_map(struct ata_ering *ering,
404 int (*map_fn)(struct ata_ering_entry *, void *),
405 void *arg)
406{
407 int idx, rc = 0;
408 struct ata_ering_entry *ent;
409
410 idx = ering->cursor;
411 do {
412 ent = &ering->ring[idx];
413 if (!ent->err_mask)
414 break;
415 rc = map_fn(ent, arg);
416 if (rc)
417 break;
418 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
419 } while (idx != ering->cursor);
420
421 return rc;
422}
423
424static int ata_ering_clear_cb(struct ata_ering_entry *ent, void *void_arg)
425{
426 ent->eflags |= ATA_EFLAG_OLD_ER;
427 return 0;
428}
429
430static void ata_ering_clear(struct ata_ering *ering)
431{
432 ata_ering_map(ering, ata_ering_clear_cb, NULL);
433}
434
435static unsigned int ata_eh_dev_action(struct ata_device *dev)
436{
437 struct ata_eh_context *ehc = &dev->link->eh_context;
438
439 return ehc->i.action | ehc->i.dev_action[dev->devno];
440}
441
442static void ata_eh_clear_action(struct ata_link *link, struct ata_device *dev,
443 struct ata_eh_info *ehi, unsigned int action)
444{
445 struct ata_device *tdev;
446
447 if (!dev) {
448 ehi->action &= ~action;
449 ata_for_each_dev(tdev, link, ALL)
450 ehi->dev_action[tdev->devno] &= ~action;
451 } else {
452 /* doesn't make sense for port-wide EH actions */
453 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
454
455 /* break ehi->action into ehi->dev_action */
456 if (ehi->action & action) {
457 ata_for_each_dev(tdev, link, ALL)
458 ehi->dev_action[tdev->devno] |=
459 ehi->action & action;
460 ehi->action &= ~action;
461 }
462
463 /* turn off the specified per-dev action */
464 ehi->dev_action[dev->devno] &= ~action;
465 }
466}
467
468/**
469 * ata_eh_acquire - acquire EH ownership
470 * @ap: ATA port to acquire EH ownership for
471 *
472 * Acquire EH ownership for @ap. This is the basic exclusion
473 * mechanism for ports sharing a host. Only one port hanging off
474 * the same host can claim the ownership of EH.
475 *
476 * LOCKING:
477 * EH context.
478 */
479void ata_eh_acquire(struct ata_port *ap)
480{
481 mutex_lock(&ap->host->eh_mutex);
482 WARN_ON_ONCE(ap->host->eh_owner);
483 ap->host->eh_owner = current;
484}
485
486/**
487 * ata_eh_release - release EH ownership
488 * @ap: ATA port to release EH ownership for
489 *
490 * Release EH ownership for @ap if the caller. The caller must
491 * have acquired EH ownership using ata_eh_acquire() previously.
492 *
493 * LOCKING:
494 * EH context.
495 */
496void ata_eh_release(struct ata_port *ap)
497{
498 WARN_ON_ONCE(ap->host->eh_owner != current);
499 ap->host->eh_owner = NULL;
500 mutex_unlock(&ap->host->eh_mutex);
501}
502
503static void ata_eh_unload(struct ata_port *ap)
504{
505 struct ata_link *link;
506 struct ata_device *dev;
507 unsigned long flags;
508
509 /* Restore SControl IPM and SPD for the next driver and
510 * disable attached devices.
511 */
512 ata_for_each_link(link, ap, PMP_FIRST) {
513 sata_scr_write(link, SCR_CONTROL, link->saved_scontrol & 0xff0);
514 ata_for_each_dev(dev, link, ALL)
515 ata_dev_disable(dev);
516 }
517
518 /* freeze and set UNLOADED */
519 spin_lock_irqsave(ap->lock, flags);
520
521 ata_port_freeze(ap); /* won't be thawed */
522 ap->pflags &= ~ATA_PFLAG_EH_PENDING; /* clear pending from freeze */
523 ap->pflags |= ATA_PFLAG_UNLOADED;
524
525 spin_unlock_irqrestore(ap->lock, flags);
526}
527
528/**
529 * ata_scsi_error - SCSI layer error handler callback
530 * @host: SCSI host on which error occurred
531 *
532 * Handles SCSI-layer-thrown error events.
533 *
534 * LOCKING:
535 * Inherited from SCSI layer (none, can sleep)
536 *
537 * RETURNS:
538 * Zero.
539 */
540void ata_scsi_error(struct Scsi_Host *host)
541{
542 struct ata_port *ap = ata_shost_to_port(host);
543 unsigned long flags;
544 LIST_HEAD(eh_work_q);
545
546 DPRINTK("ENTER\n");
547
548 spin_lock_irqsave(host->host_lock, flags);
549 list_splice_init(&host->eh_cmd_q, &eh_work_q);
550 spin_unlock_irqrestore(host->host_lock, flags);
551
552 ata_scsi_cmd_error_handler(host, ap, &eh_work_q);
553
554 /* If we timed raced normal completion and there is nothing to
555 recover nr_timedout == 0 why exactly are we doing error recovery ? */
556 ata_scsi_port_error_handler(host, ap);
557
558 /* finish or retry handled scmd's and clean up */
559 WARN_ON(!list_empty(&eh_work_q));
560
561 DPRINTK("EXIT\n");
562}
563
564/**
565 * ata_scsi_cmd_error_handler - error callback for a list of commands
566 * @host: scsi host containing the port
567 * @ap: ATA port within the host
568 * @eh_work_q: list of commands to process
569 *
570 * process the given list of commands and return those finished to the
571 * ap->eh_done_q. This function is the first part of the libata error
572 * handler which processes a given list of failed commands.
573 */
574void ata_scsi_cmd_error_handler(struct Scsi_Host *host, struct ata_port *ap,
575 struct list_head *eh_work_q)
576{
577 int i;
578 unsigned long flags;
579
580 /* make sure sff pio task is not running */
581 ata_sff_flush_pio_task(ap);
582
583 /* synchronize with host lock and sort out timeouts */
584
585 /* For new EH, all qcs are finished in one of three ways -
586 * normal completion, error completion, and SCSI timeout.
587 * Both completions can race against SCSI timeout. When normal
588 * completion wins, the qc never reaches EH. When error
589 * completion wins, the qc has ATA_QCFLAG_FAILED set.
590 *
591 * When SCSI timeout wins, things are a bit more complex.
592 * Normal or error completion can occur after the timeout but
593 * before this point. In such cases, both types of
594 * completions are honored. A scmd is determined to have
595 * timed out iff its associated qc is active and not failed.
596 */
597 spin_lock_irqsave(ap->lock, flags);
598 if (ap->ops->error_handler) {
599 struct scsi_cmnd *scmd, *tmp;
600 int nr_timedout = 0;
601
602 /* This must occur under the ap->lock as we don't want
603 a polled recovery to race the real interrupt handler
604
605 The lost_interrupt handler checks for any completed but
606 non-notified command and completes much like an IRQ handler.
607
608 We then fall into the error recovery code which will treat
609 this as if normal completion won the race */
610
611 if (ap->ops->lost_interrupt)
612 ap->ops->lost_interrupt(ap);
613
614 list_for_each_entry_safe(scmd, tmp, eh_work_q, eh_entry) {
615 struct ata_queued_cmd *qc;
616
617 ata_qc_for_each_raw(ap, qc, i) {
618 if (qc->flags & ATA_QCFLAG_ACTIVE &&
619 qc->scsicmd == scmd)
620 break;
621 }
622
623 if (i < ATA_MAX_QUEUE) {
624 /* the scmd has an associated qc */
625 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
626 /* which hasn't failed yet, timeout */
627 qc->err_mask |= AC_ERR_TIMEOUT;
628 qc->flags |= ATA_QCFLAG_FAILED;
629 nr_timedout++;
630 }
631 } else {
632 /* Normal completion occurred after
633 * SCSI timeout but before this point.
634 * Successfully complete it.
635 */
636 scmd->retries = scmd->allowed;
637 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
638 }
639 }
640
641 /* If we have timed out qcs. They belong to EH from
642 * this point but the state of the controller is
643 * unknown. Freeze the port to make sure the IRQ
644 * handler doesn't diddle with those qcs. This must
645 * be done atomically w.r.t. setting QCFLAG_FAILED.
646 */
647 if (nr_timedout)
648 __ata_port_freeze(ap);
649
650
651 /* initialize eh_tries */
652 ap->eh_tries = ATA_EH_MAX_TRIES;
653 }
654 spin_unlock_irqrestore(ap->lock, flags);
655
656}
657EXPORT_SYMBOL(ata_scsi_cmd_error_handler);
658
659/**
660 * ata_scsi_port_error_handler - recover the port after the commands
661 * @host: SCSI host containing the port
662 * @ap: the ATA port
663 *
664 * Handle the recovery of the port @ap after all the commands
665 * have been recovered.
666 */
667void ata_scsi_port_error_handler(struct Scsi_Host *host, struct ata_port *ap)
668{
669 unsigned long flags;
670
671 /* invoke error handler */
672 if (ap->ops->error_handler) {
673 struct ata_link *link;
674
675 /* acquire EH ownership */
676 ata_eh_acquire(ap);
677 repeat:
678 /* kill fast drain timer */
679 del_timer_sync(&ap->fastdrain_timer);
680
681 /* process port resume request */
682 ata_eh_handle_port_resume(ap);
683
684 /* fetch & clear EH info */
685 spin_lock_irqsave(ap->lock, flags);
686
687 ata_for_each_link(link, ap, HOST_FIRST) {
688 struct ata_eh_context *ehc = &link->eh_context;
689 struct ata_device *dev;
690
691 memset(&link->eh_context, 0, sizeof(link->eh_context));
692 link->eh_context.i = link->eh_info;
693 memset(&link->eh_info, 0, sizeof(link->eh_info));
694
695 ata_for_each_dev(dev, link, ENABLED) {
696 int devno = dev->devno;
697
698 ehc->saved_xfer_mode[devno] = dev->xfer_mode;
699 if (ata_ncq_enabled(dev))
700 ehc->saved_ncq_enabled |= 1 << devno;
701 }
702 }
703
704 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
705 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
706 ap->excl_link = NULL; /* don't maintain exclusion over EH */
707
708 spin_unlock_irqrestore(ap->lock, flags);
709
710 /* invoke EH, skip if unloading or suspended */
711 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
712 ap->ops->error_handler(ap);
713 else {
714 /* if unloading, commence suicide */
715 if ((ap->pflags & ATA_PFLAG_UNLOADING) &&
716 !(ap->pflags & ATA_PFLAG_UNLOADED))
717 ata_eh_unload(ap);
718 ata_eh_finish(ap);
719 }
720
721 /* process port suspend request */
722 ata_eh_handle_port_suspend(ap);
723
724 /* Exception might have happened after ->error_handler
725 * recovered the port but before this point. Repeat
726 * EH in such case.
727 */
728 spin_lock_irqsave(ap->lock, flags);
729
730 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
731 if (--ap->eh_tries) {
732 spin_unlock_irqrestore(ap->lock, flags);
733 goto repeat;
734 }
735 ata_port_err(ap,
736 "EH pending after %d tries, giving up\n",
737 ATA_EH_MAX_TRIES);
738 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
739 }
740
741 /* this run is complete, make sure EH info is clear */
742 ata_for_each_link(link, ap, HOST_FIRST)
743 memset(&link->eh_info, 0, sizeof(link->eh_info));
744
745 /* end eh (clear host_eh_scheduled) while holding
746 * ap->lock such that if exception occurs after this
747 * point but before EH completion, SCSI midlayer will
748 * re-initiate EH.
749 */
750 ap->ops->end_eh(ap);
751
752 spin_unlock_irqrestore(ap->lock, flags);
753 ata_eh_release(ap);
754 } else {
755 WARN_ON(ata_qc_from_tag(ap, ap->link.active_tag) == NULL);
756 ap->ops->eng_timeout(ap);
757 }
758
759 scsi_eh_flush_done_q(&ap->eh_done_q);
760
761 /* clean up */
762 spin_lock_irqsave(ap->lock, flags);
763
764 if (ap->pflags & ATA_PFLAG_LOADING)
765 ap->pflags &= ~ATA_PFLAG_LOADING;
766 else if ((ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) &&
767 !(ap->flags & ATA_FLAG_SAS_HOST))
768 schedule_delayed_work(&ap->hotplug_task, 0);
769
770 if (ap->pflags & ATA_PFLAG_RECOVERED)
771 ata_port_info(ap, "EH complete\n");
772
773 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
774
775 /* tell wait_eh that we're done */
776 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
777 wake_up_all(&ap->eh_wait_q);
778
779 spin_unlock_irqrestore(ap->lock, flags);
780}
781EXPORT_SYMBOL_GPL(ata_scsi_port_error_handler);
782
783/**
784 * ata_port_wait_eh - Wait for the currently pending EH to complete
785 * @ap: Port to wait EH for
786 *
787 * Wait until the currently pending EH is complete.
788 *
789 * LOCKING:
790 * Kernel thread context (may sleep).
791 */
792void ata_port_wait_eh(struct ata_port *ap)
793{
794 unsigned long flags;
795 DEFINE_WAIT(wait);
796
797 retry:
798 spin_lock_irqsave(ap->lock, flags);
799
800 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
801 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
802 spin_unlock_irqrestore(ap->lock, flags);
803 schedule();
804 spin_lock_irqsave(ap->lock, flags);
805 }
806 finish_wait(&ap->eh_wait_q, &wait);
807
808 spin_unlock_irqrestore(ap->lock, flags);
809
810 /* make sure SCSI EH is complete */
811 if (scsi_host_in_recovery(ap->scsi_host)) {
812 ata_msleep(ap, 10);
813 goto retry;
814 }
815}
816EXPORT_SYMBOL_GPL(ata_port_wait_eh);
817
818static int ata_eh_nr_in_flight(struct ata_port *ap)
819{
820 struct ata_queued_cmd *qc;
821 unsigned int tag;
822 int nr = 0;
823
824 /* count only non-internal commands */
825 ata_qc_for_each(ap, qc, tag) {
826 if (qc)
827 nr++;
828 }
829
830 return nr;
831}
832
833void ata_eh_fastdrain_timerfn(struct timer_list *t)
834{
835 struct ata_port *ap = from_timer(ap, t, fastdrain_timer);
836 unsigned long flags;
837 int cnt;
838
839 spin_lock_irqsave(ap->lock, flags);
840
841 cnt = ata_eh_nr_in_flight(ap);
842
843 /* are we done? */
844 if (!cnt)
845 goto out_unlock;
846
847 if (cnt == ap->fastdrain_cnt) {
848 struct ata_queued_cmd *qc;
849 unsigned int tag;
850
851 /* No progress during the last interval, tag all
852 * in-flight qcs as timed out and freeze the port.
853 */
854 ata_qc_for_each(ap, qc, tag) {
855 if (qc)
856 qc->err_mask |= AC_ERR_TIMEOUT;
857 }
858
859 ata_port_freeze(ap);
860 } else {
861 /* some qcs have finished, give it another chance */
862 ap->fastdrain_cnt = cnt;
863 ap->fastdrain_timer.expires =
864 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
865 add_timer(&ap->fastdrain_timer);
866 }
867
868 out_unlock:
869 spin_unlock_irqrestore(ap->lock, flags);
870}
871
872/**
873 * ata_eh_set_pending - set ATA_PFLAG_EH_PENDING and activate fast drain
874 * @ap: target ATA port
875 * @fastdrain: activate fast drain
876 *
877 * Set ATA_PFLAG_EH_PENDING and activate fast drain if @fastdrain
878 * is non-zero and EH wasn't pending before. Fast drain ensures
879 * that EH kicks in in timely manner.
880 *
881 * LOCKING:
882 * spin_lock_irqsave(host lock)
883 */
884static void ata_eh_set_pending(struct ata_port *ap, int fastdrain)
885{
886 int cnt;
887
888 /* already scheduled? */
889 if (ap->pflags & ATA_PFLAG_EH_PENDING)
890 return;
891
892 ap->pflags |= ATA_PFLAG_EH_PENDING;
893
894 if (!fastdrain)
895 return;
896
897 /* do we have in-flight qcs? */
898 cnt = ata_eh_nr_in_flight(ap);
899 if (!cnt)
900 return;
901
902 /* activate fast drain */
903 ap->fastdrain_cnt = cnt;
904 ap->fastdrain_timer.expires =
905 ata_deadline(jiffies, ATA_EH_FASTDRAIN_INTERVAL);
906 add_timer(&ap->fastdrain_timer);
907}
908
909/**
910 * ata_qc_schedule_eh - schedule qc for error handling
911 * @qc: command to schedule error handling for
912 *
913 * Schedule error handling for @qc. EH will kick in as soon as
914 * other commands are drained.
915 *
916 * LOCKING:
917 * spin_lock_irqsave(host lock)
918 */
919void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
920{
921 struct ata_port *ap = qc->ap;
922 struct request_queue *q = qc->scsicmd->device->request_queue;
923 unsigned long flags;
924
925 WARN_ON(!ap->ops->error_handler);
926
927 qc->flags |= ATA_QCFLAG_FAILED;
928 ata_eh_set_pending(ap, 1);
929
930 /* The following will fail if timeout has already expired.
931 * ata_scsi_error() takes care of such scmds on EH entry.
932 * Note that ATA_QCFLAG_FAILED is unconditionally set after
933 * this function completes.
934 */
935 spin_lock_irqsave(q->queue_lock, flags);
936 blk_abort_request(qc->scsicmd->request);
937 spin_unlock_irqrestore(q->queue_lock, flags);
938}
939
940/**
941 * ata_std_sched_eh - non-libsas ata_ports issue eh with this common routine
942 * @ap: ATA port to schedule EH for
943 *
944 * LOCKING: inherited from ata_port_schedule_eh
945 * spin_lock_irqsave(host lock)
946 */
947void ata_std_sched_eh(struct ata_port *ap)
948{
949 WARN_ON(!ap->ops->error_handler);
950
951 if (ap->pflags & ATA_PFLAG_INITIALIZING)
952 return;
953
954 ata_eh_set_pending(ap, 1);
955 scsi_schedule_eh(ap->scsi_host);
956
957 DPRINTK("port EH scheduled\n");
958}
959EXPORT_SYMBOL_GPL(ata_std_sched_eh);
960
961/**
962 * ata_std_end_eh - non-libsas ata_ports complete eh with this common routine
963 * @ap: ATA port to end EH for
964 *
965 * In the libata object model there is a 1:1 mapping of ata_port to
966 * shost, so host fields can be directly manipulated under ap->lock, in
967 * the libsas case we need to hold a lock at the ha->level to coordinate
968 * these events.
969 *
970 * LOCKING:
971 * spin_lock_irqsave(host lock)
972 */
973void ata_std_end_eh(struct ata_port *ap)
974{
975 struct Scsi_Host *host = ap->scsi_host;
976
977 host->host_eh_scheduled = 0;
978}
979EXPORT_SYMBOL(ata_std_end_eh);
980
981
982/**
983 * ata_port_schedule_eh - schedule error handling without a qc
984 * @ap: ATA port to schedule EH for
985 *
986 * Schedule error handling for @ap. EH will kick in as soon as
987 * all commands are drained.
988 *
989 * LOCKING:
990 * spin_lock_irqsave(host lock)
991 */
992void ata_port_schedule_eh(struct ata_port *ap)
993{
994 /* see: ata_std_sched_eh, unless you know better */
995 ap->ops->sched_eh(ap);
996}
997
998static int ata_do_link_abort(struct ata_port *ap, struct ata_link *link)
999{
1000 struct ata_queued_cmd *qc;
1001 int tag, nr_aborted = 0;
1002
1003 WARN_ON(!ap->ops->error_handler);
1004
1005 /* we're gonna abort all commands, no need for fast drain */
1006 ata_eh_set_pending(ap, 0);
1007
1008 /* include internal tag in iteration */
1009 ata_qc_for_each_with_internal(ap, qc, tag) {
1010 if (qc && (!link || qc->dev->link == link)) {
1011 qc->flags |= ATA_QCFLAG_FAILED;
1012 ata_qc_complete(qc);
1013 nr_aborted++;
1014 }
1015 }
1016
1017 if (!nr_aborted)
1018 ata_port_schedule_eh(ap);
1019
1020 return nr_aborted;
1021}
1022
1023/**
1024 * ata_link_abort - abort all qc's on the link
1025 * @link: ATA link to abort qc's for
1026 *
1027 * Abort all active qc's active on @link and schedule EH.
1028 *
1029 * LOCKING:
1030 * spin_lock_irqsave(host lock)
1031 *
1032 * RETURNS:
1033 * Number of aborted qc's.
1034 */
1035int ata_link_abort(struct ata_link *link)
1036{
1037 return ata_do_link_abort(link->ap, link);
1038}
1039
1040/**
1041 * ata_port_abort - abort all qc's on the port
1042 * @ap: ATA port to abort qc's for
1043 *
1044 * Abort all active qc's of @ap and schedule EH.
1045 *
1046 * LOCKING:
1047 * spin_lock_irqsave(host_set lock)
1048 *
1049 * RETURNS:
1050 * Number of aborted qc's.
1051 */
1052int ata_port_abort(struct ata_port *ap)
1053{
1054 return ata_do_link_abort(ap, NULL);
1055}
1056
1057/**
1058 * __ata_port_freeze - freeze port
1059 * @ap: ATA port to freeze
1060 *
1061 * This function is called when HSM violation or some other
1062 * condition disrupts normal operation of the port. Frozen port
1063 * is not allowed to perform any operation until the port is
1064 * thawed, which usually follows a successful reset.
1065 *
1066 * ap->ops->freeze() callback can be used for freezing the port
1067 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
1068 * port cannot be frozen hardware-wise, the interrupt handler
1069 * must ack and clear interrupts unconditionally while the port
1070 * is frozen.
1071 *
1072 * LOCKING:
1073 * spin_lock_irqsave(host lock)
1074 */
1075static void __ata_port_freeze(struct ata_port *ap)
1076{
1077 WARN_ON(!ap->ops->error_handler);
1078
1079 if (ap->ops->freeze)
1080 ap->ops->freeze(ap);
1081
1082 ap->pflags |= ATA_PFLAG_FROZEN;
1083
1084 DPRINTK("ata%u port frozen\n", ap->print_id);
1085}
1086
1087/**
1088 * ata_port_freeze - abort & freeze port
1089 * @ap: ATA port to freeze
1090 *
1091 * Abort and freeze @ap. The freeze operation must be called
1092 * first, because some hardware requires special operations
1093 * before the taskfile registers are accessible.
1094 *
1095 * LOCKING:
1096 * spin_lock_irqsave(host lock)
1097 *
1098 * RETURNS:
1099 * Number of aborted commands.
1100 */
1101int ata_port_freeze(struct ata_port *ap)
1102{
1103 int nr_aborted;
1104
1105 WARN_ON(!ap->ops->error_handler);
1106
1107 __ata_port_freeze(ap);
1108 nr_aborted = ata_port_abort(ap);
1109
1110 return nr_aborted;
1111}
1112
1113/**
1114 * sata_async_notification - SATA async notification handler
1115 * @ap: ATA port where async notification is received
1116 *
1117 * Handler to be called when async notification via SDB FIS is
1118 * received. This function schedules EH if necessary.
1119 *
1120 * LOCKING:
1121 * spin_lock_irqsave(host lock)
1122 *
1123 * RETURNS:
1124 * 1 if EH is scheduled, 0 otherwise.
1125 */
1126int sata_async_notification(struct ata_port *ap)
1127{
1128 u32 sntf;
1129 int rc;
1130
1131 if (!(ap->flags & ATA_FLAG_AN))
1132 return 0;
1133
1134 rc = sata_scr_read(&ap->link, SCR_NOTIFICATION, &sntf);
1135 if (rc == 0)
1136 sata_scr_write(&ap->link, SCR_NOTIFICATION, sntf);
1137
1138 if (!sata_pmp_attached(ap) || rc) {
1139 /* PMP is not attached or SNTF is not available */
1140 if (!sata_pmp_attached(ap)) {
1141 /* PMP is not attached. Check whether ATAPI
1142 * AN is configured. If so, notify media
1143 * change.
1144 */
1145 struct ata_device *dev = ap->link.device;
1146
1147 if ((dev->class == ATA_DEV_ATAPI) &&
1148 (dev->flags & ATA_DFLAG_AN))
1149 ata_scsi_media_change_notify(dev);
1150 return 0;
1151 } else {
1152 /* PMP is attached but SNTF is not available.
1153 * ATAPI async media change notification is
1154 * not used. The PMP must be reporting PHY
1155 * status change, schedule EH.
1156 */
1157 ata_port_schedule_eh(ap);
1158 return 1;
1159 }
1160 } else {
1161 /* PMP is attached and SNTF is available */
1162 struct ata_link *link;
1163
1164 /* check and notify ATAPI AN */
1165 ata_for_each_link(link, ap, EDGE) {
1166 if (!(sntf & (1 << link->pmp)))
1167 continue;
1168
1169 if ((link->device->class == ATA_DEV_ATAPI) &&
1170 (link->device->flags & ATA_DFLAG_AN))
1171 ata_scsi_media_change_notify(link->device);
1172 }
1173
1174 /* If PMP is reporting that PHY status of some
1175 * downstream ports has changed, schedule EH.
1176 */
1177 if (sntf & (1 << SATA_PMP_CTRL_PORT)) {
1178 ata_port_schedule_eh(ap);
1179 return 1;
1180 }
1181
1182 return 0;
1183 }
1184}
1185
1186/**
1187 * ata_eh_freeze_port - EH helper to freeze port
1188 * @ap: ATA port to freeze
1189 *
1190 * Freeze @ap.
1191 *
1192 * LOCKING:
1193 * None.
1194 */
1195void ata_eh_freeze_port(struct ata_port *ap)
1196{
1197 unsigned long flags;
1198
1199 if (!ap->ops->error_handler)
1200 return;
1201
1202 spin_lock_irqsave(ap->lock, flags);
1203 __ata_port_freeze(ap);
1204 spin_unlock_irqrestore(ap->lock, flags);
1205}
1206
1207/**
1208 * ata_port_thaw_port - EH helper to thaw port
1209 * @ap: ATA port to thaw
1210 *
1211 * Thaw frozen port @ap.
1212 *
1213 * LOCKING:
1214 * None.
1215 */
1216void ata_eh_thaw_port(struct ata_port *ap)
1217{
1218 unsigned long flags;
1219
1220 if (!ap->ops->error_handler)
1221 return;
1222
1223 spin_lock_irqsave(ap->lock, flags);
1224
1225 ap->pflags &= ~ATA_PFLAG_FROZEN;
1226
1227 if (ap->ops->thaw)
1228 ap->ops->thaw(ap);
1229
1230 spin_unlock_irqrestore(ap->lock, flags);
1231
1232 DPRINTK("ata%u port thawed\n", ap->print_id);
1233}
1234
1235static void ata_eh_scsidone(struct scsi_cmnd *scmd)
1236{
1237 /* nada */
1238}
1239
1240static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
1241{
1242 struct ata_port *ap = qc->ap;
1243 struct scsi_cmnd *scmd = qc->scsicmd;
1244 unsigned long flags;
1245
1246 spin_lock_irqsave(ap->lock, flags);
1247 qc->scsidone = ata_eh_scsidone;
1248 __ata_qc_complete(qc);
1249 WARN_ON(ata_tag_valid(qc->tag));
1250 spin_unlock_irqrestore(ap->lock, flags);
1251
1252 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
1253}
1254
1255/**
1256 * ata_eh_qc_complete - Complete an active ATA command from EH
1257 * @qc: Command to complete
1258 *
1259 * Indicate to the mid and upper layers that an ATA command has
1260 * completed. To be used from EH.
1261 */
1262void ata_eh_qc_complete(struct ata_queued_cmd *qc)
1263{
1264 struct scsi_cmnd *scmd = qc->scsicmd;
1265 scmd->retries = scmd->allowed;
1266 __ata_eh_qc_complete(qc);
1267}
1268
1269/**
1270 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
1271 * @qc: Command to retry
1272 *
1273 * Indicate to the mid and upper layers that an ATA command
1274 * should be retried. To be used from EH.
1275 *
1276 * SCSI midlayer limits the number of retries to scmd->allowed.
1277 * scmd->allowed is incremented for commands which get retried
1278 * due to unrelated failures (qc->err_mask is zero).
1279 */
1280void ata_eh_qc_retry(struct ata_queued_cmd *qc)
1281{
1282 struct scsi_cmnd *scmd = qc->scsicmd;
1283 if (!qc->err_mask)
1284 scmd->allowed++;
1285 __ata_eh_qc_complete(qc);
1286}
1287
1288/**
1289 * ata_dev_disable - disable ATA device
1290 * @dev: ATA device to disable
1291 *
1292 * Disable @dev.
1293 *
1294 * Locking:
1295 * EH context.
1296 */
1297void ata_dev_disable(struct ata_device *dev)
1298{
1299 if (!ata_dev_enabled(dev))
1300 return;
1301
1302 if (ata_msg_drv(dev->link->ap))
1303 ata_dev_warn(dev, "disabled\n");
1304 ata_acpi_on_disable(dev);
1305 ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO0 | ATA_DNXFER_QUIET);
1306 dev->class++;
1307
1308 /* From now till the next successful probe, ering is used to
1309 * track probe failures. Clear accumulated device error info.
1310 */
1311 ata_ering_clear(&dev->ering);
1312}
1313
1314/**
1315 * ata_eh_detach_dev - detach ATA device
1316 * @dev: ATA device to detach
1317 *
1318 * Detach @dev.
1319 *
1320 * LOCKING:
1321 * None.
1322 */
1323void ata_eh_detach_dev(struct ata_device *dev)
1324{
1325 struct ata_link *link = dev->link;
1326 struct ata_port *ap = link->ap;
1327 struct ata_eh_context *ehc = &link->eh_context;
1328 unsigned long flags;
1329
1330 ata_dev_disable(dev);
1331
1332 spin_lock_irqsave(ap->lock, flags);
1333
1334 dev->flags &= ~ATA_DFLAG_DETACH;
1335
1336 if (ata_scsi_offline_dev(dev)) {
1337 dev->flags |= ATA_DFLAG_DETACHED;
1338 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1339 }
1340
1341 /* clear per-dev EH info */
1342 ata_eh_clear_action(link, dev, &link->eh_info, ATA_EH_PERDEV_MASK);
1343 ata_eh_clear_action(link, dev, &link->eh_context.i, ATA_EH_PERDEV_MASK);
1344 ehc->saved_xfer_mode[dev->devno] = 0;
1345 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
1346
1347 spin_unlock_irqrestore(ap->lock, flags);
1348}
1349
1350/**
1351 * ata_eh_about_to_do - about to perform eh_action
1352 * @link: target ATA link
1353 * @dev: target ATA dev for per-dev action (can be NULL)
1354 * @action: action about to be performed
1355 *
1356 * Called just before performing EH actions to clear related bits
1357 * in @link->eh_info such that eh actions are not unnecessarily
1358 * repeated.
1359 *
1360 * LOCKING:
1361 * None.
1362 */
1363void ata_eh_about_to_do(struct ata_link *link, struct ata_device *dev,
1364 unsigned int action)
1365{
1366 struct ata_port *ap = link->ap;
1367 struct ata_eh_info *ehi = &link->eh_info;
1368 struct ata_eh_context *ehc = &link->eh_context;
1369 unsigned long flags;
1370
1371 spin_lock_irqsave(ap->lock, flags);
1372
1373 ata_eh_clear_action(link, dev, ehi, action);
1374
1375 /* About to take EH action, set RECOVERED. Ignore actions on
1376 * slave links as master will do them again.
1377 */
1378 if (!(ehc->i.flags & ATA_EHI_QUIET) && link != ap->slave_link)
1379 ap->pflags |= ATA_PFLAG_RECOVERED;
1380
1381 spin_unlock_irqrestore(ap->lock, flags);
1382}
1383
1384/**
1385 * ata_eh_done - EH action complete
1386 * @link: ATA link for which EH actions are complete
1387 * @dev: target ATA dev for per-dev action (can be NULL)
1388 * @action: action just completed
1389 *
1390 * Called right after performing EH actions to clear related bits
1391 * in @link->eh_context.
1392 *
1393 * LOCKING:
1394 * None.
1395 */
1396void ata_eh_done(struct ata_link *link, struct ata_device *dev,
1397 unsigned int action)
1398{
1399 struct ata_eh_context *ehc = &link->eh_context;
1400
1401 ata_eh_clear_action(link, dev, &ehc->i, action);
1402}
1403
1404/**
1405 * ata_err_string - convert err_mask to descriptive string
1406 * @err_mask: error mask to convert to string
1407 *
1408 * Convert @err_mask to descriptive string. Errors are
1409 * prioritized according to severity and only the most severe
1410 * error is reported.
1411 *
1412 * LOCKING:
1413 * None.
1414 *
1415 * RETURNS:
1416 * Descriptive string for @err_mask
1417 */
1418static const char *ata_err_string(unsigned int err_mask)
1419{
1420 if (err_mask & AC_ERR_HOST_BUS)
1421 return "host bus error";
1422 if (err_mask & AC_ERR_ATA_BUS)
1423 return "ATA bus error";
1424 if (err_mask & AC_ERR_TIMEOUT)
1425 return "timeout";
1426 if (err_mask & AC_ERR_HSM)
1427 return "HSM violation";
1428 if (err_mask & AC_ERR_SYSTEM)
1429 return "internal error";
1430 if (err_mask & AC_ERR_MEDIA)
1431 return "media error";
1432 if (err_mask & AC_ERR_INVALID)
1433 return "invalid argument";
1434 if (err_mask & AC_ERR_DEV)
1435 return "device error";
1436 if (err_mask & AC_ERR_NCQ)
1437 return "NCQ error";
1438 if (err_mask & AC_ERR_NODEV_HINT)
1439 return "Polling detection error";
1440 return "unknown error";
1441}
1442
1443/**
1444 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
1445 * @dev: Device to read log page 10h from
1446 * @tag: Resulting tag of the failed command
1447 * @tf: Resulting taskfile registers of the failed command
1448 *
1449 * Read log page 10h to obtain NCQ error details and clear error
1450 * condition.
1451 *
1452 * LOCKING:
1453 * Kernel thread context (may sleep).
1454 *
1455 * RETURNS:
1456 * 0 on success, -errno otherwise.
1457 */
1458static int ata_eh_read_log_10h(struct ata_device *dev,
1459 int *tag, struct ata_taskfile *tf)
1460{
1461 u8 *buf = dev->link->ap->sector_buf;
1462 unsigned int err_mask;
1463 u8 csum;
1464 int i;
1465
1466 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, 0, buf, 1);
1467 if (err_mask)
1468 return -EIO;
1469
1470 csum = 0;
1471 for (i = 0; i < ATA_SECT_SIZE; i++)
1472 csum += buf[i];
1473 if (csum)
1474 ata_dev_warn(dev, "invalid checksum 0x%x on log page 10h\n",
1475 csum);
1476
1477 if (buf[0] & 0x80)
1478 return -ENOENT;
1479
1480 *tag = buf[0] & 0x1f;
1481
1482 tf->command = buf[2];
1483 tf->feature = buf[3];
1484 tf->lbal = buf[4];
1485 tf->lbam = buf[5];
1486 tf->lbah = buf[6];
1487 tf->device = buf[7];
1488 tf->hob_lbal = buf[8];
1489 tf->hob_lbam = buf[9];
1490 tf->hob_lbah = buf[10];
1491 tf->nsect = buf[12];
1492 tf->hob_nsect = buf[13];
1493 if (dev->class == ATA_DEV_ZAC && ata_id_has_ncq_autosense(dev->id))
1494 tf->auxiliary = buf[14] << 16 | buf[15] << 8 | buf[16];
1495
1496 return 0;
1497}
1498
1499/**
1500 * atapi_eh_tur - perform ATAPI TEST_UNIT_READY
1501 * @dev: target ATAPI device
1502 * @r_sense_key: out parameter for sense_key
1503 *
1504 * Perform ATAPI TEST_UNIT_READY.
1505 *
1506 * LOCKING:
1507 * EH context (may sleep).
1508 *
1509 * RETURNS:
1510 * 0 on success, AC_ERR_* mask on failure.
1511 */
1512unsigned int atapi_eh_tur(struct ata_device *dev, u8 *r_sense_key)
1513{
1514 u8 cdb[ATAPI_CDB_LEN] = { TEST_UNIT_READY, 0, 0, 0, 0, 0 };
1515 struct ata_taskfile tf;
1516 unsigned int err_mask;
1517
1518 ata_tf_init(dev, &tf);
1519
1520 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1521 tf.command = ATA_CMD_PACKET;
1522 tf.protocol = ATAPI_PROT_NODATA;
1523
1524 err_mask = ata_exec_internal(dev, &tf, cdb, DMA_NONE, NULL, 0, 0);
1525 if (err_mask == AC_ERR_DEV)
1526 *r_sense_key = tf.feature >> 4;
1527 return err_mask;
1528}
1529
1530/**
1531 * ata_eh_request_sense - perform REQUEST_SENSE_DATA_EXT
1532 * @qc: qc to perform REQUEST_SENSE_SENSE_DATA_EXT to
1533 * @cmd: scsi command for which the sense code should be set
1534 *
1535 * Perform REQUEST_SENSE_DATA_EXT after the device reported CHECK
1536 * SENSE. This function is an EH helper.
1537 *
1538 * LOCKING:
1539 * Kernel thread context (may sleep).
1540 */
1541static void ata_eh_request_sense(struct ata_queued_cmd *qc,
1542 struct scsi_cmnd *cmd)
1543{
1544 struct ata_device *dev = qc->dev;
1545 struct ata_taskfile tf;
1546 unsigned int err_mask;
1547
1548 if (qc->ap->pflags & ATA_PFLAG_FROZEN) {
1549 ata_dev_warn(dev, "sense data available but port frozen\n");
1550 return;
1551 }
1552
1553 if (!cmd || qc->flags & ATA_QCFLAG_SENSE_VALID)
1554 return;
1555
1556 if (!ata_id_sense_reporting_enabled(dev->id)) {
1557 ata_dev_warn(qc->dev, "sense data reporting disabled\n");
1558 return;
1559 }
1560
1561 DPRINTK("ATA request sense\n");
1562
1563 ata_tf_init(dev, &tf);
1564 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1565 tf.flags |= ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1566 tf.command = ATA_CMD_REQ_SENSE_DATA;
1567 tf.protocol = ATA_PROT_NODATA;
1568
1569 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
1570 /* Ignore err_mask; ATA_ERR might be set */
1571 if (tf.command & ATA_SENSE) {
1572 ata_scsi_set_sense(dev, cmd, tf.lbah, tf.lbam, tf.lbal);
1573 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1574 } else {
1575 ata_dev_warn(dev, "request sense failed stat %02x emask %x\n",
1576 tf.command, err_mask);
1577 }
1578}
1579
1580/**
1581 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1582 * @dev: device to perform REQUEST_SENSE to
1583 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1584 * @dfl_sense_key: default sense key to use
1585 *
1586 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1587 * SENSE. This function is EH helper.
1588 *
1589 * LOCKING:
1590 * Kernel thread context (may sleep).
1591 *
1592 * RETURNS:
1593 * 0 on success, AC_ERR_* mask on failure
1594 */
1595unsigned int atapi_eh_request_sense(struct ata_device *dev,
1596 u8 *sense_buf, u8 dfl_sense_key)
1597{
1598 u8 cdb[ATAPI_CDB_LEN] =
1599 { REQUEST_SENSE, 0, 0, 0, SCSI_SENSE_BUFFERSIZE, 0 };
1600 struct ata_port *ap = dev->link->ap;
1601 struct ata_taskfile tf;
1602
1603 DPRINTK("ATAPI request sense\n");
1604
1605 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1606
1607 /* initialize sense_buf with the error register,
1608 * for the case where they are -not- overwritten
1609 */
1610 sense_buf[0] = 0x70;
1611 sense_buf[2] = dfl_sense_key;
1612
1613 /* some devices time out if garbage left in tf */
1614 ata_tf_init(dev, &tf);
1615
1616 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1617 tf.command = ATA_CMD_PACKET;
1618
1619 /* is it pointless to prefer PIO for "safety reasons"? */
1620 if (ap->flags & ATA_FLAG_PIO_DMA) {
1621 tf.protocol = ATAPI_PROT_DMA;
1622 tf.feature |= ATAPI_PKT_DMA;
1623 } else {
1624 tf.protocol = ATAPI_PROT_PIO;
1625 tf.lbam = SCSI_SENSE_BUFFERSIZE;
1626 tf.lbah = 0;
1627 }
1628
1629 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1630 sense_buf, SCSI_SENSE_BUFFERSIZE, 0);
1631}
1632
1633/**
1634 * ata_eh_analyze_serror - analyze SError for a failed port
1635 * @link: ATA link to analyze SError for
1636 *
1637 * Analyze SError if available and further determine cause of
1638 * failure.
1639 *
1640 * LOCKING:
1641 * None.
1642 */
1643static void ata_eh_analyze_serror(struct ata_link *link)
1644{
1645 struct ata_eh_context *ehc = &link->eh_context;
1646 u32 serror = ehc->i.serror;
1647 unsigned int err_mask = 0, action = 0;
1648 u32 hotplug_mask;
1649
1650 if (serror & (SERR_PERSISTENT | SERR_DATA)) {
1651 err_mask |= AC_ERR_ATA_BUS;
1652 action |= ATA_EH_RESET;
1653 }
1654 if (serror & SERR_PROTOCOL) {
1655 err_mask |= AC_ERR_HSM;
1656 action |= ATA_EH_RESET;
1657 }
1658 if (serror & SERR_INTERNAL) {
1659 err_mask |= AC_ERR_SYSTEM;
1660 action |= ATA_EH_RESET;
1661 }
1662
1663 /* Determine whether a hotplug event has occurred. Both
1664 * SError.N/X are considered hotplug events for enabled or
1665 * host links. For disabled PMP links, only N bit is
1666 * considered as X bit is left at 1 for link plugging.
1667 */
1668 if (link->lpm_policy > ATA_LPM_MAX_POWER)
1669 hotplug_mask = 0; /* hotplug doesn't work w/ LPM */
1670 else if (!(link->flags & ATA_LFLAG_DISABLED) || ata_is_host_link(link))
1671 hotplug_mask = SERR_PHYRDY_CHG | SERR_DEV_XCHG;
1672 else
1673 hotplug_mask = SERR_PHYRDY_CHG;
1674
1675 if (serror & hotplug_mask)
1676 ata_ehi_hotplugged(&ehc->i);
1677
1678 ehc->i.err_mask |= err_mask;
1679 ehc->i.action |= action;
1680}
1681
1682/**
1683 * ata_eh_analyze_ncq_error - analyze NCQ error
1684 * @link: ATA link to analyze NCQ error for
1685 *
1686 * Read log page 10h, determine the offending qc and acquire
1687 * error status TF. For NCQ device errors, all LLDDs have to do
1688 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1689 * care of the rest.
1690 *
1691 * LOCKING:
1692 * Kernel thread context (may sleep).
1693 */
1694void ata_eh_analyze_ncq_error(struct ata_link *link)
1695{
1696 struct ata_port *ap = link->ap;
1697 struct ata_eh_context *ehc = &link->eh_context;
1698 struct ata_device *dev = link->device;
1699 struct ata_queued_cmd *qc;
1700 struct ata_taskfile tf;
1701 int tag, rc;
1702
1703 /* if frozen, we can't do much */
1704 if (ap->pflags & ATA_PFLAG_FROZEN)
1705 return;
1706
1707 /* is it NCQ device error? */
1708 if (!link->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1709 return;
1710
1711 /* has LLDD analyzed already? */
1712 ata_qc_for_each_raw(ap, qc, tag) {
1713 if (!(qc->flags & ATA_QCFLAG_FAILED))
1714 continue;
1715
1716 if (qc->err_mask)
1717 return;
1718 }
1719
1720 /* okay, this error is ours */
1721 memset(&tf, 0, sizeof(tf));
1722 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1723 if (rc) {
1724 ata_link_err(link, "failed to read log page 10h (errno=%d)\n",
1725 rc);
1726 return;
1727 }
1728
1729 if (!(link->sactive & (1 << tag))) {
1730 ata_link_err(link, "log page 10h reported inactive tag %d\n",
1731 tag);
1732 return;
1733 }
1734
1735 /* we've got the perpetrator, condemn it */
1736 qc = __ata_qc_from_tag(ap, tag);
1737 memcpy(&qc->result_tf, &tf, sizeof(tf));
1738 qc->result_tf.flags = ATA_TFLAG_ISADDR | ATA_TFLAG_LBA | ATA_TFLAG_LBA48;
1739 qc->err_mask |= AC_ERR_DEV | AC_ERR_NCQ;
1740 if (dev->class == ATA_DEV_ZAC &&
1741 ((qc->result_tf.command & ATA_SENSE) || qc->result_tf.auxiliary)) {
1742 char sense_key, asc, ascq;
1743
1744 sense_key = (qc->result_tf.auxiliary >> 16) & 0xff;
1745 asc = (qc->result_tf.auxiliary >> 8) & 0xff;
1746 ascq = qc->result_tf.auxiliary & 0xff;
1747 ata_scsi_set_sense(dev, qc->scsicmd, sense_key, asc, ascq);
1748 ata_scsi_set_sense_information(dev, qc->scsicmd,
1749 &qc->result_tf);
1750 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1751 }
1752
1753 ehc->i.err_mask &= ~AC_ERR_DEV;
1754}
1755
1756/**
1757 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1758 * @qc: qc to analyze
1759 * @tf: Taskfile registers to analyze
1760 *
1761 * Analyze taskfile of @qc and further determine cause of
1762 * failure. This function also requests ATAPI sense data if
1763 * available.
1764 *
1765 * LOCKING:
1766 * Kernel thread context (may sleep).
1767 *
1768 * RETURNS:
1769 * Determined recovery action
1770 */
1771static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1772 const struct ata_taskfile *tf)
1773{
1774 unsigned int tmp, action = 0;
1775 u8 stat = tf->command, err = tf->feature;
1776
1777 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1778 qc->err_mask |= AC_ERR_HSM;
1779 return ATA_EH_RESET;
1780 }
1781
1782 if (stat & (ATA_ERR | ATA_DF)) {
1783 qc->err_mask |= AC_ERR_DEV;
1784 /*
1785 * Sense data reporting does not work if the
1786 * device fault bit is set.
1787 */
1788 if (stat & ATA_DF)
1789 stat &= ~ATA_SENSE;
1790 } else {
1791 return 0;
1792 }
1793
1794 switch (qc->dev->class) {
1795 case ATA_DEV_ZAC:
1796 if (stat & ATA_SENSE)
1797 ata_eh_request_sense(qc, qc->scsicmd);
1798 /* fall through */
1799 case ATA_DEV_ATA:
1800 if (err & ATA_ICRC)
1801 qc->err_mask |= AC_ERR_ATA_BUS;
1802 if (err & (ATA_UNC | ATA_AMNF))
1803 qc->err_mask |= AC_ERR_MEDIA;
1804 if (err & ATA_IDNF)
1805 qc->err_mask |= AC_ERR_INVALID;
1806 break;
1807
1808 case ATA_DEV_ATAPI:
1809 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1810 tmp = atapi_eh_request_sense(qc->dev,
1811 qc->scsicmd->sense_buffer,
1812 qc->result_tf.feature >> 4);
1813 if (!tmp)
1814 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1815 else
1816 qc->err_mask |= tmp;
1817 }
1818 }
1819
1820 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1821 int ret = scsi_check_sense(qc->scsicmd);
1822 /*
1823 * SUCCESS here means that the sense code could be
1824 * evaluated and should be passed to the upper layers
1825 * for correct evaluation.
1826 * FAILED means the sense code could not be interpreted
1827 * and the device would need to be reset.
1828 * NEEDS_RETRY and ADD_TO_MLQUEUE means that the
1829 * command would need to be retried.
1830 */
1831 if (ret == NEEDS_RETRY || ret == ADD_TO_MLQUEUE) {
1832 qc->flags |= ATA_QCFLAG_RETRY;
1833 qc->err_mask |= AC_ERR_OTHER;
1834 } else if (ret != SUCCESS) {
1835 qc->err_mask |= AC_ERR_HSM;
1836 }
1837 }
1838 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1839 action |= ATA_EH_RESET;
1840
1841 return action;
1842}
1843
1844static int ata_eh_categorize_error(unsigned int eflags, unsigned int err_mask,
1845 int *xfer_ok)
1846{
1847 int base = 0;
1848
1849 if (!(eflags & ATA_EFLAG_DUBIOUS_XFER))
1850 *xfer_ok = 1;
1851
1852 if (!*xfer_ok)
1853 base = ATA_ECAT_DUBIOUS_NONE;
1854
1855 if (err_mask & AC_ERR_ATA_BUS)
1856 return base + ATA_ECAT_ATA_BUS;
1857
1858 if (err_mask & AC_ERR_TIMEOUT)
1859 return base + ATA_ECAT_TOUT_HSM;
1860
1861 if (eflags & ATA_EFLAG_IS_IO) {
1862 if (err_mask & AC_ERR_HSM)
1863 return base + ATA_ECAT_TOUT_HSM;
1864 if ((err_mask &
1865 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1866 return base + ATA_ECAT_UNK_DEV;
1867 }
1868
1869 return 0;
1870}
1871
1872struct speed_down_verdict_arg {
1873 u64 since;
1874 int xfer_ok;
1875 int nr_errors[ATA_ECAT_NR];
1876};
1877
1878static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1879{
1880 struct speed_down_verdict_arg *arg = void_arg;
1881 int cat;
1882
1883 if ((ent->eflags & ATA_EFLAG_OLD_ER) || (ent->timestamp < arg->since))
1884 return -1;
1885
1886 cat = ata_eh_categorize_error(ent->eflags, ent->err_mask,
1887 &arg->xfer_ok);
1888 arg->nr_errors[cat]++;
1889
1890 return 0;
1891}
1892
1893/**
1894 * ata_eh_speed_down_verdict - Determine speed down verdict
1895 * @dev: Device of interest
1896 *
1897 * This function examines error ring of @dev and determines
1898 * whether NCQ needs to be turned off, transfer speed should be
1899 * stepped down, or falling back to PIO is necessary.
1900 *
1901 * ECAT_ATA_BUS : ATA_BUS error for any command
1902 *
1903 * ECAT_TOUT_HSM : TIMEOUT for any command or HSM violation for
1904 * IO commands
1905 *
1906 * ECAT_UNK_DEV : Unknown DEV error for IO commands
1907 *
1908 * ECAT_DUBIOUS_* : Identical to above three but occurred while
1909 * data transfer hasn't been verified.
1910 *
1911 * Verdicts are
1912 *
1913 * NCQ_OFF : Turn off NCQ.
1914 *
1915 * SPEED_DOWN : Speed down transfer speed but don't fall back
1916 * to PIO.
1917 *
1918 * FALLBACK_TO_PIO : Fall back to PIO.
1919 *
1920 * Even if multiple verdicts are returned, only one action is
1921 * taken per error. An action triggered by non-DUBIOUS errors
1922 * clears ering, while one triggered by DUBIOUS_* errors doesn't.
1923 * This is to expedite speed down decisions right after device is
1924 * initially configured.
1925 *
1926 * The following are speed down rules. #1 and #2 deal with
1927 * DUBIOUS errors.
1928 *
1929 * 1. If more than one DUBIOUS_ATA_BUS or DUBIOUS_TOUT_HSM errors
1930 * occurred during last 5 mins, SPEED_DOWN and FALLBACK_TO_PIO.
1931 *
1932 * 2. If more than one DUBIOUS_TOUT_HSM or DUBIOUS_UNK_DEV errors
1933 * occurred during last 5 mins, NCQ_OFF.
1934 *
1935 * 3. If more than 8 ATA_BUS, TOUT_HSM or UNK_DEV errors
1936 * occurred during last 5 mins, FALLBACK_TO_PIO
1937 *
1938 * 4. If more than 3 TOUT_HSM or UNK_DEV errors occurred
1939 * during last 10 mins, NCQ_OFF.
1940 *
1941 * 5. If more than 3 ATA_BUS or TOUT_HSM errors, or more than 6
1942 * UNK_DEV errors occurred during last 10 mins, SPEED_DOWN.
1943 *
1944 * LOCKING:
1945 * Inherited from caller.
1946 *
1947 * RETURNS:
1948 * OR of ATA_EH_SPDN_* flags.
1949 */
1950static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1951{
1952 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1953 u64 j64 = get_jiffies_64();
1954 struct speed_down_verdict_arg arg;
1955 unsigned int verdict = 0;
1956
1957 /* scan past 5 mins of error history */
1958 memset(&arg, 0, sizeof(arg));
1959 arg.since = j64 - min(j64, j5mins);
1960 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1961
1962 if (arg.nr_errors[ATA_ECAT_DUBIOUS_ATA_BUS] +
1963 arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] > 1)
1964 verdict |= ATA_EH_SPDN_SPEED_DOWN |
1965 ATA_EH_SPDN_FALLBACK_TO_PIO | ATA_EH_SPDN_KEEP_ERRORS;
1966
1967 if (arg.nr_errors[ATA_ECAT_DUBIOUS_TOUT_HSM] +
1968 arg.nr_errors[ATA_ECAT_DUBIOUS_UNK_DEV] > 1)
1969 verdict |= ATA_EH_SPDN_NCQ_OFF | ATA_EH_SPDN_KEEP_ERRORS;
1970
1971 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1972 arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1973 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1974 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1975
1976 /* scan past 10 mins of error history */
1977 memset(&arg, 0, sizeof(arg));
1978 arg.since = j64 - min(j64, j10mins);
1979 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1980
1981 if (arg.nr_errors[ATA_ECAT_TOUT_HSM] +
1982 arg.nr_errors[ATA_ECAT_UNK_DEV] > 3)
1983 verdict |= ATA_EH_SPDN_NCQ_OFF;
1984
1985 if (arg.nr_errors[ATA_ECAT_ATA_BUS] +
1986 arg.nr_errors[ATA_ECAT_TOUT_HSM] > 3 ||
1987 arg.nr_errors[ATA_ECAT_UNK_DEV] > 6)
1988 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1989
1990 return verdict;
1991}
1992
1993/**
1994 * ata_eh_speed_down - record error and speed down if necessary
1995 * @dev: Failed device
1996 * @eflags: mask of ATA_EFLAG_* flags
1997 * @err_mask: err_mask of the error
1998 *
1999 * Record error and examine error history to determine whether
2000 * adjusting transmission speed is necessary. It also sets
2001 * transmission limits appropriately if such adjustment is
2002 * necessary.
2003 *
2004 * LOCKING:
2005 * Kernel thread context (may sleep).
2006 *
2007 * RETURNS:
2008 * Determined recovery action.
2009 */
2010static unsigned int ata_eh_speed_down(struct ata_device *dev,
2011 unsigned int eflags, unsigned int err_mask)
2012{
2013 struct ata_link *link = ata_dev_phys_link(dev);
2014 int xfer_ok = 0;
2015 unsigned int verdict;
2016 unsigned int action = 0;
2017
2018 /* don't bother if Cat-0 error */
2019 if (ata_eh_categorize_error(eflags, err_mask, &xfer_ok) == 0)
2020 return 0;
2021
2022 /* record error and determine whether speed down is necessary */
2023 ata_ering_record(&dev->ering, eflags, err_mask);
2024 verdict = ata_eh_speed_down_verdict(dev);
2025
2026 /* turn off NCQ? */
2027 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
2028 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
2029 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
2030 dev->flags |= ATA_DFLAG_NCQ_OFF;
2031 ata_dev_warn(dev, "NCQ disabled due to excessive errors\n");
2032 goto done;
2033 }
2034
2035 /* speed down? */
2036 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
2037 /* speed down SATA link speed if possible */
2038 if (sata_down_spd_limit(link, 0) == 0) {
2039 action |= ATA_EH_RESET;
2040 goto done;
2041 }
2042
2043 /* lower transfer mode */
2044 if (dev->spdn_cnt < 2) {
2045 static const int dma_dnxfer_sel[] =
2046 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
2047 static const int pio_dnxfer_sel[] =
2048 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
2049 int sel;
2050
2051 if (dev->xfer_shift != ATA_SHIFT_PIO)
2052 sel = dma_dnxfer_sel[dev->spdn_cnt];
2053 else
2054 sel = pio_dnxfer_sel[dev->spdn_cnt];
2055
2056 dev->spdn_cnt++;
2057
2058 if (ata_down_xfermask_limit(dev, sel) == 0) {
2059 action |= ATA_EH_RESET;
2060 goto done;
2061 }
2062 }
2063 }
2064
2065 /* Fall back to PIO? Slowing down to PIO is meaningless for
2066 * SATA ATA devices. Consider it only for PATA and SATAPI.
2067 */
2068 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
2069 (link->ap->cbl != ATA_CBL_SATA || dev->class == ATA_DEV_ATAPI) &&
2070 (dev->xfer_shift != ATA_SHIFT_PIO)) {
2071 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
2072 dev->spdn_cnt = 0;
2073 action |= ATA_EH_RESET;
2074 goto done;
2075 }
2076 }
2077
2078 return 0;
2079 done:
2080 /* device has been slowed down, blow error history */
2081 if (!(verdict & ATA_EH_SPDN_KEEP_ERRORS))
2082 ata_ering_clear(&dev->ering);
2083 return action;
2084}
2085
2086/**
2087 * ata_eh_worth_retry - analyze error and decide whether to retry
2088 * @qc: qc to possibly retry
2089 *
2090 * Look at the cause of the error and decide if a retry
2091 * might be useful or not. We don't want to retry media errors
2092 * because the drive itself has probably already taken 10-30 seconds
2093 * doing its own internal retries before reporting the failure.
2094 */
2095static inline int ata_eh_worth_retry(struct ata_queued_cmd *qc)
2096{
2097 if (qc->err_mask & AC_ERR_MEDIA)
2098 return 0; /* don't retry media errors */
2099 if (qc->flags & ATA_QCFLAG_IO)
2100 return 1; /* otherwise retry anything from fs stack */
2101 if (qc->err_mask & AC_ERR_INVALID)
2102 return 0; /* don't retry these */
2103 return qc->err_mask != AC_ERR_DEV; /* retry if not dev error */
2104}
2105
2106/**
2107 * ata_eh_quiet - check if we need to be quiet about a command error
2108 * @qc: qc to check
2109 *
2110 * Look at the qc flags anbd its scsi command request flags to determine
2111 * if we need to be quiet about the command failure.
2112 */
2113static inline bool ata_eh_quiet(struct ata_queued_cmd *qc)
2114{
2115 if (qc->scsicmd &&
2116 qc->scsicmd->request->rq_flags & RQF_QUIET)
2117 qc->flags |= ATA_QCFLAG_QUIET;
2118 return qc->flags & ATA_QCFLAG_QUIET;
2119}
2120
2121/**
2122 * ata_eh_link_autopsy - analyze error and determine recovery action
2123 * @link: host link to perform autopsy on
2124 *
2125 * Analyze why @link failed and determine which recovery actions
2126 * are needed. This function also sets more detailed AC_ERR_*
2127 * values and fills sense data for ATAPI CHECK SENSE.
2128 *
2129 * LOCKING:
2130 * Kernel thread context (may sleep).
2131 */
2132static void ata_eh_link_autopsy(struct ata_link *link)
2133{
2134 struct ata_port *ap = link->ap;
2135 struct ata_eh_context *ehc = &link->eh_context;
2136 struct ata_queued_cmd *qc;
2137 struct ata_device *dev;
2138 unsigned int all_err_mask = 0, eflags = 0;
2139 int tag, nr_failed = 0, nr_quiet = 0;
2140 u32 serror;
2141 int rc;
2142
2143 DPRINTK("ENTER\n");
2144
2145 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
2146 return;
2147
2148 /* obtain and analyze SError */
2149 rc = sata_scr_read(link, SCR_ERROR, &serror);
2150 if (rc == 0) {
2151 ehc->i.serror |= serror;
2152 ata_eh_analyze_serror(link);
2153 } else if (rc != -EOPNOTSUPP) {
2154 /* SError read failed, force reset and probing */
2155 ehc->i.probe_mask |= ATA_ALL_DEVICES;
2156 ehc->i.action |= ATA_EH_RESET;
2157 ehc->i.err_mask |= AC_ERR_OTHER;
2158 }
2159
2160 /* analyze NCQ failure */
2161 ata_eh_analyze_ncq_error(link);
2162
2163 /* any real error trumps AC_ERR_OTHER */
2164 if (ehc->i.err_mask & ~AC_ERR_OTHER)
2165 ehc->i.err_mask &= ~AC_ERR_OTHER;
2166
2167 all_err_mask |= ehc->i.err_mask;
2168
2169 ata_qc_for_each_raw(ap, qc, tag) {
2170 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2171 ata_dev_phys_link(qc->dev) != link)
2172 continue;
2173
2174 /* inherit upper level err_mask */
2175 qc->err_mask |= ehc->i.err_mask;
2176
2177 /* analyze TF */
2178 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
2179
2180 /* DEV errors are probably spurious in case of ATA_BUS error */
2181 if (qc->err_mask & AC_ERR_ATA_BUS)
2182 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
2183 AC_ERR_INVALID);
2184
2185 /* any real error trumps unknown error */
2186 if (qc->err_mask & ~AC_ERR_OTHER)
2187 qc->err_mask &= ~AC_ERR_OTHER;
2188
2189 /*
2190 * SENSE_VALID trumps dev/unknown error and revalidation. Upper
2191 * layers will determine whether the command is worth retrying
2192 * based on the sense data and device class/type. Otherwise,
2193 * determine directly if the command is worth retrying using its
2194 * error mask and flags.
2195 */
2196 if (qc->flags & ATA_QCFLAG_SENSE_VALID)
2197 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
2198 else if (ata_eh_worth_retry(qc))
2199 qc->flags |= ATA_QCFLAG_RETRY;
2200
2201 /* accumulate error info */
2202 ehc->i.dev = qc->dev;
2203 all_err_mask |= qc->err_mask;
2204 if (qc->flags & ATA_QCFLAG_IO)
2205 eflags |= ATA_EFLAG_IS_IO;
2206 trace_ata_eh_link_autopsy_qc(qc);
2207
2208 /* Count quiet errors */
2209 if (ata_eh_quiet(qc))
2210 nr_quiet++;
2211 nr_failed++;
2212 }
2213
2214 /* If all failed commands requested silence, then be quiet */
2215 if (nr_quiet == nr_failed)
2216 ehc->i.flags |= ATA_EHI_QUIET;
2217
2218 /* enforce default EH actions */
2219 if (ap->pflags & ATA_PFLAG_FROZEN ||
2220 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
2221 ehc->i.action |= ATA_EH_RESET;
2222 else if (((eflags & ATA_EFLAG_IS_IO) && all_err_mask) ||
2223 (!(eflags & ATA_EFLAG_IS_IO) && (all_err_mask & ~AC_ERR_DEV)))
2224 ehc->i.action |= ATA_EH_REVALIDATE;
2225
2226 /* If we have offending qcs and the associated failed device,
2227 * perform per-dev EH action only on the offending device.
2228 */
2229 if (ehc->i.dev) {
2230 ehc->i.dev_action[ehc->i.dev->devno] |=
2231 ehc->i.action & ATA_EH_PERDEV_MASK;
2232 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
2233 }
2234
2235 /* propagate timeout to host link */
2236 if ((all_err_mask & AC_ERR_TIMEOUT) && !ata_is_host_link(link))
2237 ap->link.eh_context.i.err_mask |= AC_ERR_TIMEOUT;
2238
2239 /* record error and consider speeding down */
2240 dev = ehc->i.dev;
2241 if (!dev && ((ata_link_max_devices(link) == 1 &&
2242 ata_dev_enabled(link->device))))
2243 dev = link->device;
2244
2245 if (dev) {
2246 if (dev->flags & ATA_DFLAG_DUBIOUS_XFER)
2247 eflags |= ATA_EFLAG_DUBIOUS_XFER;
2248 ehc->i.action |= ata_eh_speed_down(dev, eflags, all_err_mask);
2249 trace_ata_eh_link_autopsy(dev, ehc->i.action, all_err_mask);
2250 }
2251 DPRINTK("EXIT\n");
2252}
2253
2254/**
2255 * ata_eh_autopsy - analyze error and determine recovery action
2256 * @ap: host port to perform autopsy on
2257 *
2258 * Analyze all links of @ap and determine why they failed and
2259 * which recovery actions are needed.
2260 *
2261 * LOCKING:
2262 * Kernel thread context (may sleep).
2263 */
2264void ata_eh_autopsy(struct ata_port *ap)
2265{
2266 struct ata_link *link;
2267
2268 ata_for_each_link(link, ap, EDGE)
2269 ata_eh_link_autopsy(link);
2270
2271 /* Handle the frigging slave link. Autopsy is done similarly
2272 * but actions and flags are transferred over to the master
2273 * link and handled from there.
2274 */
2275 if (ap->slave_link) {
2276 struct ata_eh_context *mehc = &ap->link.eh_context;
2277 struct ata_eh_context *sehc = &ap->slave_link->eh_context;
2278
2279 /* transfer control flags from master to slave */
2280 sehc->i.flags |= mehc->i.flags & ATA_EHI_TO_SLAVE_MASK;
2281
2282 /* perform autopsy on the slave link */
2283 ata_eh_link_autopsy(ap->slave_link);
2284
2285 /* transfer actions from slave to master and clear slave */
2286 ata_eh_about_to_do(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2287 mehc->i.action |= sehc->i.action;
2288 mehc->i.dev_action[1] |= sehc->i.dev_action[1];
2289 mehc->i.flags |= sehc->i.flags;
2290 ata_eh_done(ap->slave_link, NULL, ATA_EH_ALL_ACTIONS);
2291 }
2292
2293 /* Autopsy of fanout ports can affect host link autopsy.
2294 * Perform host link autopsy last.
2295 */
2296 if (sata_pmp_attached(ap))
2297 ata_eh_link_autopsy(&ap->link);
2298}
2299
2300/**
2301 * ata_get_cmd_descript - get description for ATA command
2302 * @command: ATA command code to get description for
2303 *
2304 * Return a textual description of the given command, or NULL if the
2305 * command is not known.
2306 *
2307 * LOCKING:
2308 * None
2309 */
2310const char *ata_get_cmd_descript(u8 command)
2311{
2312#ifdef CONFIG_ATA_VERBOSE_ERROR
2313 static const struct
2314 {
2315 u8 command;
2316 const char *text;
2317 } cmd_descr[] = {
2318 { ATA_CMD_DEV_RESET, "DEVICE RESET" },
2319 { ATA_CMD_CHK_POWER, "CHECK POWER MODE" },
2320 { ATA_CMD_STANDBY, "STANDBY" },
2321 { ATA_CMD_IDLE, "IDLE" },
2322 { ATA_CMD_EDD, "EXECUTE DEVICE DIAGNOSTIC" },
2323 { ATA_CMD_DOWNLOAD_MICRO, "DOWNLOAD MICROCODE" },
2324 { ATA_CMD_DOWNLOAD_MICRO_DMA, "DOWNLOAD MICROCODE DMA" },
2325 { ATA_CMD_NOP, "NOP" },
2326 { ATA_CMD_FLUSH, "FLUSH CACHE" },
2327 { ATA_CMD_FLUSH_EXT, "FLUSH CACHE EXT" },
2328 { ATA_CMD_ID_ATA, "IDENTIFY DEVICE" },
2329 { ATA_CMD_ID_ATAPI, "IDENTIFY PACKET DEVICE" },
2330 { ATA_CMD_SERVICE, "SERVICE" },
2331 { ATA_CMD_READ, "READ DMA" },
2332 { ATA_CMD_READ_EXT, "READ DMA EXT" },
2333 { ATA_CMD_READ_QUEUED, "READ DMA QUEUED" },
2334 { ATA_CMD_READ_STREAM_EXT, "READ STREAM EXT" },
2335 { ATA_CMD_READ_STREAM_DMA_EXT, "READ STREAM DMA EXT" },
2336 { ATA_CMD_WRITE, "WRITE DMA" },
2337 { ATA_CMD_WRITE_EXT, "WRITE DMA EXT" },
2338 { ATA_CMD_WRITE_QUEUED, "WRITE DMA QUEUED EXT" },
2339 { ATA_CMD_WRITE_STREAM_EXT, "WRITE STREAM EXT" },
2340 { ATA_CMD_WRITE_STREAM_DMA_EXT, "WRITE STREAM DMA EXT" },
2341 { ATA_CMD_WRITE_FUA_EXT, "WRITE DMA FUA EXT" },
2342 { ATA_CMD_WRITE_QUEUED_FUA_EXT, "WRITE DMA QUEUED FUA EXT" },
2343 { ATA_CMD_FPDMA_READ, "READ FPDMA QUEUED" },
2344 { ATA_CMD_FPDMA_WRITE, "WRITE FPDMA QUEUED" },
2345 { ATA_CMD_FPDMA_SEND, "SEND FPDMA QUEUED" },
2346 { ATA_CMD_FPDMA_RECV, "RECEIVE FPDMA QUEUED" },
2347 { ATA_CMD_PIO_READ, "READ SECTOR(S)" },
2348 { ATA_CMD_PIO_READ_EXT, "READ SECTOR(S) EXT" },
2349 { ATA_CMD_PIO_WRITE, "WRITE SECTOR(S)" },
2350 { ATA_CMD_PIO_WRITE_EXT, "WRITE SECTOR(S) EXT" },
2351 { ATA_CMD_READ_MULTI, "READ MULTIPLE" },
2352 { ATA_CMD_READ_MULTI_EXT, "READ MULTIPLE EXT" },
2353 { ATA_CMD_WRITE_MULTI, "WRITE MULTIPLE" },
2354 { ATA_CMD_WRITE_MULTI_EXT, "WRITE MULTIPLE EXT" },
2355 { ATA_CMD_WRITE_MULTI_FUA_EXT, "WRITE MULTIPLE FUA EXT" },
2356 { ATA_CMD_SET_FEATURES, "SET FEATURES" },
2357 { ATA_CMD_SET_MULTI, "SET MULTIPLE MODE" },
2358 { ATA_CMD_VERIFY, "READ VERIFY SECTOR(S)" },
2359 { ATA_CMD_VERIFY_EXT, "READ VERIFY SECTOR(S) EXT" },
2360 { ATA_CMD_WRITE_UNCORR_EXT, "WRITE UNCORRECTABLE EXT" },
2361 { ATA_CMD_STANDBYNOW1, "STANDBY IMMEDIATE" },
2362 { ATA_CMD_IDLEIMMEDIATE, "IDLE IMMEDIATE" },
2363 { ATA_CMD_SLEEP, "SLEEP" },
2364 { ATA_CMD_INIT_DEV_PARAMS, "INITIALIZE DEVICE PARAMETERS" },
2365 { ATA_CMD_READ_NATIVE_MAX, "READ NATIVE MAX ADDRESS" },
2366 { ATA_CMD_READ_NATIVE_MAX_EXT, "READ NATIVE MAX ADDRESS EXT" },
2367 { ATA_CMD_SET_MAX, "SET MAX ADDRESS" },
2368 { ATA_CMD_SET_MAX_EXT, "SET MAX ADDRESS EXT" },
2369 { ATA_CMD_READ_LOG_EXT, "READ LOG EXT" },
2370 { ATA_CMD_WRITE_LOG_EXT, "WRITE LOG EXT" },
2371 { ATA_CMD_READ_LOG_DMA_EXT, "READ LOG DMA EXT" },
2372 { ATA_CMD_WRITE_LOG_DMA_EXT, "WRITE LOG DMA EXT" },
2373 { ATA_CMD_TRUSTED_NONDATA, "TRUSTED NON-DATA" },
2374 { ATA_CMD_TRUSTED_RCV, "TRUSTED RECEIVE" },
2375 { ATA_CMD_TRUSTED_RCV_DMA, "TRUSTED RECEIVE DMA" },
2376 { ATA_CMD_TRUSTED_SND, "TRUSTED SEND" },
2377 { ATA_CMD_TRUSTED_SND_DMA, "TRUSTED SEND DMA" },
2378 { ATA_CMD_PMP_READ, "READ BUFFER" },
2379 { ATA_CMD_PMP_READ_DMA, "READ BUFFER DMA" },
2380 { ATA_CMD_PMP_WRITE, "WRITE BUFFER" },
2381 { ATA_CMD_PMP_WRITE_DMA, "WRITE BUFFER DMA" },
2382 { ATA_CMD_CONF_OVERLAY, "DEVICE CONFIGURATION OVERLAY" },
2383 { ATA_CMD_SEC_SET_PASS, "SECURITY SET PASSWORD" },
2384 { ATA_CMD_SEC_UNLOCK, "SECURITY UNLOCK" },
2385 { ATA_CMD_SEC_ERASE_PREP, "SECURITY ERASE PREPARE" },
2386 { ATA_CMD_SEC_ERASE_UNIT, "SECURITY ERASE UNIT" },
2387 { ATA_CMD_SEC_FREEZE_LOCK, "SECURITY FREEZE LOCK" },
2388 { ATA_CMD_SEC_DISABLE_PASS, "SECURITY DISABLE PASSWORD" },
2389 { ATA_CMD_CONFIG_STREAM, "CONFIGURE STREAM" },
2390 { ATA_CMD_SMART, "SMART" },
2391 { ATA_CMD_MEDIA_LOCK, "DOOR LOCK" },
2392 { ATA_CMD_MEDIA_UNLOCK, "DOOR UNLOCK" },
2393 { ATA_CMD_DSM, "DATA SET MANAGEMENT" },
2394 { ATA_CMD_CHK_MED_CRD_TYP, "CHECK MEDIA CARD TYPE" },
2395 { ATA_CMD_CFA_REQ_EXT_ERR, "CFA REQUEST EXTENDED ERROR" },
2396 { ATA_CMD_CFA_WRITE_NE, "CFA WRITE SECTORS WITHOUT ERASE" },
2397 { ATA_CMD_CFA_TRANS_SECT, "CFA TRANSLATE SECTOR" },
2398 { ATA_CMD_CFA_ERASE, "CFA ERASE SECTORS" },
2399 { ATA_CMD_CFA_WRITE_MULT_NE, "CFA WRITE MULTIPLE WITHOUT ERASE" },
2400 { ATA_CMD_REQ_SENSE_DATA, "REQUEST SENSE DATA EXT" },
2401 { ATA_CMD_SANITIZE_DEVICE, "SANITIZE DEVICE" },
2402 { ATA_CMD_ZAC_MGMT_IN, "ZAC MANAGEMENT IN" },
2403 { ATA_CMD_ZAC_MGMT_OUT, "ZAC MANAGEMENT OUT" },
2404 { ATA_CMD_READ_LONG, "READ LONG (with retries)" },
2405 { ATA_CMD_READ_LONG_ONCE, "READ LONG (without retries)" },
2406 { ATA_CMD_WRITE_LONG, "WRITE LONG (with retries)" },
2407 { ATA_CMD_WRITE_LONG_ONCE, "WRITE LONG (without retries)" },
2408 { ATA_CMD_RESTORE, "RECALIBRATE" },
2409 { 0, NULL } /* terminate list */
2410 };
2411
2412 unsigned int i;
2413 for (i = 0; cmd_descr[i].text; i++)
2414 if (cmd_descr[i].command == command)
2415 return cmd_descr[i].text;
2416#endif
2417
2418 return NULL;
2419}
2420EXPORT_SYMBOL_GPL(ata_get_cmd_descript);
2421
2422/**
2423 * ata_eh_link_report - report error handling to user
2424 * @link: ATA link EH is going on
2425 *
2426 * Report EH to user.
2427 *
2428 * LOCKING:
2429 * None.
2430 */
2431static void ata_eh_link_report(struct ata_link *link)
2432{
2433 struct ata_port *ap = link->ap;
2434 struct ata_eh_context *ehc = &link->eh_context;
2435 struct ata_queued_cmd *qc;
2436 const char *frozen, *desc;
2437 char tries_buf[6] = "";
2438 int tag, nr_failed = 0;
2439
2440 if (ehc->i.flags & ATA_EHI_QUIET)
2441 return;
2442
2443 desc = NULL;
2444 if (ehc->i.desc[0] != '\0')
2445 desc = ehc->i.desc;
2446
2447 ata_qc_for_each_raw(ap, qc, tag) {
2448 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2449 ata_dev_phys_link(qc->dev) != link ||
2450 ((qc->flags & ATA_QCFLAG_QUIET) &&
2451 qc->err_mask == AC_ERR_DEV))
2452 continue;
2453 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
2454 continue;
2455
2456 nr_failed++;
2457 }
2458
2459 if (!nr_failed && !ehc->i.err_mask)
2460 return;
2461
2462 frozen = "";
2463 if (ap->pflags & ATA_PFLAG_FROZEN)
2464 frozen = " frozen";
2465
2466 if (ap->eh_tries < ATA_EH_MAX_TRIES)
2467 snprintf(tries_buf, sizeof(tries_buf), " t%d",
2468 ap->eh_tries);
2469
2470 if (ehc->i.dev) {
2471 ata_dev_err(ehc->i.dev, "exception Emask 0x%x "
2472 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2473 ehc->i.err_mask, link->sactive, ehc->i.serror,
2474 ehc->i.action, frozen, tries_buf);
2475 if (desc)
2476 ata_dev_err(ehc->i.dev, "%s\n", desc);
2477 } else {
2478 ata_link_err(link, "exception Emask 0x%x "
2479 "SAct 0x%x SErr 0x%x action 0x%x%s%s\n",
2480 ehc->i.err_mask, link->sactive, ehc->i.serror,
2481 ehc->i.action, frozen, tries_buf);
2482 if (desc)
2483 ata_link_err(link, "%s\n", desc);
2484 }
2485
2486#ifdef CONFIG_ATA_VERBOSE_ERROR
2487 if (ehc->i.serror)
2488 ata_link_err(link,
2489 "SError: { %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s}\n",
2490 ehc->i.serror & SERR_DATA_RECOVERED ? "RecovData " : "",
2491 ehc->i.serror & SERR_COMM_RECOVERED ? "RecovComm " : "",
2492 ehc->i.serror & SERR_DATA ? "UnrecovData " : "",
2493 ehc->i.serror & SERR_PERSISTENT ? "Persist " : "",
2494 ehc->i.serror & SERR_PROTOCOL ? "Proto " : "",
2495 ehc->i.serror & SERR_INTERNAL ? "HostInt " : "",
2496 ehc->i.serror & SERR_PHYRDY_CHG ? "PHYRdyChg " : "",
2497 ehc->i.serror & SERR_PHY_INT_ERR ? "PHYInt " : "",
2498 ehc->i.serror & SERR_COMM_WAKE ? "CommWake " : "",
2499 ehc->i.serror & SERR_10B_8B_ERR ? "10B8B " : "",
2500 ehc->i.serror & SERR_DISPARITY ? "Dispar " : "",
2501 ehc->i.serror & SERR_CRC ? "BadCRC " : "",
2502 ehc->i.serror & SERR_HANDSHAKE ? "Handshk " : "",
2503 ehc->i.serror & SERR_LINK_SEQ_ERR ? "LinkSeq " : "",
2504 ehc->i.serror & SERR_TRANS_ST_ERROR ? "TrStaTrns " : "",
2505 ehc->i.serror & SERR_UNRECOG_FIS ? "UnrecFIS " : "",
2506 ehc->i.serror & SERR_DEV_XCHG ? "DevExch " : "");
2507#endif
2508
2509 ata_qc_for_each_raw(ap, qc, tag) {
2510 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
2511 char data_buf[20] = "";
2512 char cdb_buf[70] = "";
2513
2514 if (!(qc->flags & ATA_QCFLAG_FAILED) ||
2515 ata_dev_phys_link(qc->dev) != link || !qc->err_mask)
2516 continue;
2517
2518 if (qc->dma_dir != DMA_NONE) {
2519 static const char *dma_str[] = {
2520 [DMA_BIDIRECTIONAL] = "bidi",
2521 [DMA_TO_DEVICE] = "out",
2522 [DMA_FROM_DEVICE] = "in",
2523 };
2524 const char *prot_str = NULL;
2525
2526 switch (qc->tf.protocol) {
2527 case ATA_PROT_UNKNOWN:
2528 prot_str = "unknown";
2529 break;
2530 case ATA_PROT_NODATA:
2531 prot_str = "nodata";
2532 break;
2533 case ATA_PROT_PIO:
2534 prot_str = "pio";
2535 break;
2536 case ATA_PROT_DMA:
2537 prot_str = "dma";
2538 break;
2539 case ATA_PROT_NCQ:
2540 prot_str = "ncq dma";
2541 break;
2542 case ATA_PROT_NCQ_NODATA:
2543 prot_str = "ncq nodata";
2544 break;
2545 case ATAPI_PROT_NODATA:
2546 prot_str = "nodata";
2547 break;
2548 case ATAPI_PROT_PIO:
2549 prot_str = "pio";
2550 break;
2551 case ATAPI_PROT_DMA:
2552 prot_str = "dma";
2553 break;
2554 }
2555 snprintf(data_buf, sizeof(data_buf), " %s %u %s",
2556 prot_str, qc->nbytes, dma_str[qc->dma_dir]);
2557 }
2558
2559 if (ata_is_atapi(qc->tf.protocol)) {
2560 const u8 *cdb = qc->cdb;
2561 size_t cdb_len = qc->dev->cdb_len;
2562
2563 if (qc->scsicmd) {
2564 cdb = qc->scsicmd->cmnd;
2565 cdb_len = qc->scsicmd->cmd_len;
2566 }
2567 __scsi_format_command(cdb_buf, sizeof(cdb_buf),
2568 cdb, cdb_len);
2569 } else {
2570 const char *descr = ata_get_cmd_descript(cmd->command);
2571 if (descr)
2572 ata_dev_err(qc->dev, "failed command: %s\n",
2573 descr);
2574 }
2575
2576 ata_dev_err(qc->dev,
2577 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2578 "tag %d%s\n %s"
2579 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
2580 "Emask 0x%x (%s)%s\n",
2581 cmd->command, cmd->feature, cmd->nsect,
2582 cmd->lbal, cmd->lbam, cmd->lbah,
2583 cmd->hob_feature, cmd->hob_nsect,
2584 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
2585 cmd->device, qc->tag, data_buf, cdb_buf,
2586 res->command, res->feature, res->nsect,
2587 res->lbal, res->lbam, res->lbah,
2588 res->hob_feature, res->hob_nsect,
2589 res->hob_lbal, res->hob_lbam, res->hob_lbah,
2590 res->device, qc->err_mask, ata_err_string(qc->err_mask),
2591 qc->err_mask & AC_ERR_NCQ ? " <F>" : "");
2592
2593#ifdef CONFIG_ATA_VERBOSE_ERROR
2594 if (res->command & (ATA_BUSY | ATA_DRDY | ATA_DF | ATA_DRQ |
2595 ATA_SENSE | ATA_ERR)) {
2596 if (res->command & ATA_BUSY)
2597 ata_dev_err(qc->dev, "status: { Busy }\n");
2598 else
2599 ata_dev_err(qc->dev, "status: { %s%s%s%s%s}\n",
2600 res->command & ATA_DRDY ? "DRDY " : "",
2601 res->command & ATA_DF ? "DF " : "",
2602 res->command & ATA_DRQ ? "DRQ " : "",
2603 res->command & ATA_SENSE ? "SENSE " : "",
2604 res->command & ATA_ERR ? "ERR " : "");
2605 }
2606
2607 if (cmd->command != ATA_CMD_PACKET &&
2608 (res->feature & (ATA_ICRC | ATA_UNC | ATA_AMNF |
2609 ATA_IDNF | ATA_ABORTED)))
2610 ata_dev_err(qc->dev, "error: { %s%s%s%s%s}\n",
2611 res->feature & ATA_ICRC ? "ICRC " : "",
2612 res->feature & ATA_UNC ? "UNC " : "",
2613 res->feature & ATA_AMNF ? "AMNF " : "",
2614 res->feature & ATA_IDNF ? "IDNF " : "",
2615 res->feature & ATA_ABORTED ? "ABRT " : "");
2616#endif
2617 }
2618}
2619
2620/**
2621 * ata_eh_report - report error handling to user
2622 * @ap: ATA port to report EH about
2623 *
2624 * Report EH to user.
2625 *
2626 * LOCKING:
2627 * None.
2628 */
2629void ata_eh_report(struct ata_port *ap)
2630{
2631 struct ata_link *link;
2632
2633 ata_for_each_link(link, ap, HOST_FIRST)
2634 ata_eh_link_report(link);
2635}
2636
2637static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
2638 unsigned int *classes, unsigned long deadline,
2639 bool clear_classes)
2640{
2641 struct ata_device *dev;
2642
2643 if (clear_classes)
2644 ata_for_each_dev(dev, link, ALL)
2645 classes[dev->devno] = ATA_DEV_UNKNOWN;
2646
2647 return reset(link, classes, deadline);
2648}
2649
2650static int ata_eh_followup_srst_needed(struct ata_link *link, int rc)
2651{
2652 if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
2653 return 0;
2654 if (rc == -EAGAIN)
2655 return 1;
2656 if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
2657 return 1;
2658 return 0;
2659}
2660
2661int ata_eh_reset(struct ata_link *link, int classify,
2662 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
2663 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
2664{
2665 struct ata_port *ap = link->ap;
2666 struct ata_link *slave = ap->slave_link;
2667 struct ata_eh_context *ehc = &link->eh_context;
2668 struct ata_eh_context *sehc = slave ? &slave->eh_context : NULL;
2669 unsigned int *classes = ehc->classes;
2670 unsigned int lflags = link->flags;
2671 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
2672 int max_tries = 0, try = 0;
2673 struct ata_link *failed_link;
2674 struct ata_device *dev;
2675 unsigned long deadline, now;
2676 ata_reset_fn_t reset;
2677 unsigned long flags;
2678 u32 sstatus;
2679 int nr_unknown, rc;
2680
2681 /*
2682 * Prepare to reset
2683 */
2684 while (ata_eh_reset_timeouts[max_tries] != ULONG_MAX)
2685 max_tries++;
2686 if (link->flags & ATA_LFLAG_RST_ONCE)
2687 max_tries = 1;
2688 if (link->flags & ATA_LFLAG_NO_HRST)
2689 hardreset = NULL;
2690 if (link->flags & ATA_LFLAG_NO_SRST)
2691 softreset = NULL;
2692
2693 /* make sure each reset attempt is at least COOL_DOWN apart */
2694 if (ehc->i.flags & ATA_EHI_DID_RESET) {
2695 now = jiffies;
2696 WARN_ON(time_after(ehc->last_reset, now));
2697 deadline = ata_deadline(ehc->last_reset,
2698 ATA_EH_RESET_COOL_DOWN);
2699 if (time_before(now, deadline))
2700 schedule_timeout_uninterruptible(deadline - now);
2701 }
2702
2703 spin_lock_irqsave(ap->lock, flags);
2704 ap->pflags |= ATA_PFLAG_RESETTING;
2705 spin_unlock_irqrestore(ap->lock, flags);
2706
2707 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2708
2709 ata_for_each_dev(dev, link, ALL) {
2710 /* If we issue an SRST then an ATA drive (not ATAPI)
2711 * may change configuration and be in PIO0 timing. If
2712 * we do a hard reset (or are coming from power on)
2713 * this is true for ATA or ATAPI. Until we've set a
2714 * suitable controller mode we should not touch the
2715 * bus as we may be talking too fast.
2716 */
2717 dev->pio_mode = XFER_PIO_0;
2718 dev->dma_mode = 0xff;
2719
2720 /* If the controller has a pio mode setup function
2721 * then use it to set the chipset to rights. Don't
2722 * touch the DMA setup as that will be dealt with when
2723 * configuring devices.
2724 */
2725 if (ap->ops->set_piomode)
2726 ap->ops->set_piomode(ap, dev);
2727 }
2728
2729 /* prefer hardreset */
2730 reset = NULL;
2731 ehc->i.action &= ~ATA_EH_RESET;
2732 if (hardreset) {
2733 reset = hardreset;
2734 ehc->i.action |= ATA_EH_HARDRESET;
2735 } else if (softreset) {
2736 reset = softreset;
2737 ehc->i.action |= ATA_EH_SOFTRESET;
2738 }
2739
2740 if (prereset) {
2741 unsigned long deadline = ata_deadline(jiffies,
2742 ATA_EH_PRERESET_TIMEOUT);
2743
2744 if (slave) {
2745 sehc->i.action &= ~ATA_EH_RESET;
2746 sehc->i.action |= ehc->i.action;
2747 }
2748
2749 rc = prereset(link, deadline);
2750
2751 /* If present, do prereset on slave link too. Reset
2752 * is skipped iff both master and slave links report
2753 * -ENOENT or clear ATA_EH_RESET.
2754 */
2755 if (slave && (rc == 0 || rc == -ENOENT)) {
2756 int tmp;
2757
2758 tmp = prereset(slave, deadline);
2759 if (tmp != -ENOENT)
2760 rc = tmp;
2761
2762 ehc->i.action |= sehc->i.action;
2763 }
2764
2765 if (rc) {
2766 if (rc == -ENOENT) {
2767 ata_link_dbg(link, "port disabled--ignoring\n");
2768 ehc->i.action &= ~ATA_EH_RESET;
2769
2770 ata_for_each_dev(dev, link, ALL)
2771 classes[dev->devno] = ATA_DEV_NONE;
2772
2773 rc = 0;
2774 } else
2775 ata_link_err(link,
2776 "prereset failed (errno=%d)\n",
2777 rc);
2778 goto out;
2779 }
2780
2781 /* prereset() might have cleared ATA_EH_RESET. If so,
2782 * bang classes, thaw and return.
2783 */
2784 if (reset && !(ehc->i.action & ATA_EH_RESET)) {
2785 ata_for_each_dev(dev, link, ALL)
2786 classes[dev->devno] = ATA_DEV_NONE;
2787 if ((ap->pflags & ATA_PFLAG_FROZEN) &&
2788 ata_is_host_link(link))
2789 ata_eh_thaw_port(ap);
2790 rc = 0;
2791 goto out;
2792 }
2793 }
2794
2795 retry:
2796 /*
2797 * Perform reset
2798 */
2799 if (ata_is_host_link(link))
2800 ata_eh_freeze_port(ap);
2801
2802 deadline = ata_deadline(jiffies, ata_eh_reset_timeouts[try++]);
2803
2804 if (reset) {
2805 if (verbose)
2806 ata_link_info(link, "%s resetting link\n",
2807 reset == softreset ? "soft" : "hard");
2808
2809 /* mark that this EH session started with reset */
2810 ehc->last_reset = jiffies;
2811 if (reset == hardreset)
2812 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
2813 else
2814 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
2815
2816 rc = ata_do_reset(link, reset, classes, deadline, true);
2817 if (rc && rc != -EAGAIN) {
2818 failed_link = link;
2819 goto fail;
2820 }
2821
2822 /* hardreset slave link if existent */
2823 if (slave && reset == hardreset) {
2824 int tmp;
2825
2826 if (verbose)
2827 ata_link_info(slave, "hard resetting link\n");
2828
2829 ata_eh_about_to_do(slave, NULL, ATA_EH_RESET);
2830 tmp = ata_do_reset(slave, reset, classes, deadline,
2831 false);
2832 switch (tmp) {
2833 case -EAGAIN:
2834 rc = -EAGAIN;
2835 case 0:
2836 break;
2837 default:
2838 failed_link = slave;
2839 rc = tmp;
2840 goto fail;
2841 }
2842 }
2843
2844 /* perform follow-up SRST if necessary */
2845 if (reset == hardreset &&
2846 ata_eh_followup_srst_needed(link, rc)) {
2847 reset = softreset;
2848
2849 if (!reset) {
2850 ata_link_err(link,
2851 "follow-up softreset required but no softreset available\n");
2852 failed_link = link;
2853 rc = -EINVAL;
2854 goto fail;
2855 }
2856
2857 ata_eh_about_to_do(link, NULL, ATA_EH_RESET);
2858 rc = ata_do_reset(link, reset, classes, deadline, true);
2859 if (rc) {
2860 failed_link = link;
2861 goto fail;
2862 }
2863 }
2864 } else {
2865 if (verbose)
2866 ata_link_info(link,
2867 "no reset method available, skipping reset\n");
2868 if (!(lflags & ATA_LFLAG_ASSUME_CLASS))
2869 lflags |= ATA_LFLAG_ASSUME_ATA;
2870 }
2871
2872 /*
2873 * Post-reset processing
2874 */
2875 ata_for_each_dev(dev, link, ALL) {
2876 /* After the reset, the device state is PIO 0 and the
2877 * controller state is undefined. Reset also wakes up
2878 * drives from sleeping mode.
2879 */
2880 dev->pio_mode = XFER_PIO_0;
2881 dev->flags &= ~ATA_DFLAG_SLEEPING;
2882
2883 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
2884 continue;
2885
2886 /* apply class override */
2887 if (lflags & ATA_LFLAG_ASSUME_ATA)
2888 classes[dev->devno] = ATA_DEV_ATA;
2889 else if (lflags & ATA_LFLAG_ASSUME_SEMB)
2890 classes[dev->devno] = ATA_DEV_SEMB_UNSUP;
2891 }
2892
2893 /* record current link speed */
2894 if (sata_scr_read(link, SCR_STATUS, &sstatus) == 0)
2895 link->sata_spd = (sstatus >> 4) & 0xf;
2896 if (slave && sata_scr_read(slave, SCR_STATUS, &sstatus) == 0)
2897 slave->sata_spd = (sstatus >> 4) & 0xf;
2898
2899 /* thaw the port */
2900 if (ata_is_host_link(link))
2901 ata_eh_thaw_port(ap);
2902
2903 /* postreset() should clear hardware SError. Although SError
2904 * is cleared during link resume, clearing SError here is
2905 * necessary as some PHYs raise hotplug events after SRST.
2906 * This introduces race condition where hotplug occurs between
2907 * reset and here. This race is mediated by cross checking
2908 * link onlineness and classification result later.
2909 */
2910 if (postreset) {
2911 postreset(link, classes);
2912 if (slave)
2913 postreset(slave, classes);
2914 }
2915
2916 /*
2917 * Some controllers can't be frozen very well and may set spurious
2918 * error conditions during reset. Clear accumulated error
2919 * information and re-thaw the port if frozen. As reset is the
2920 * final recovery action and we cross check link onlineness against
2921 * device classification later, no hotplug event is lost by this.
2922 */
2923 spin_lock_irqsave(link->ap->lock, flags);
2924 memset(&link->eh_info, 0, sizeof(link->eh_info));
2925 if (slave)
2926 memset(&slave->eh_info, 0, sizeof(link->eh_info));
2927 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
2928 spin_unlock_irqrestore(link->ap->lock, flags);
2929
2930 if (ap->pflags & ATA_PFLAG_FROZEN)
2931 ata_eh_thaw_port(ap);
2932
2933 /*
2934 * Make sure onlineness and classification result correspond.
2935 * Hotplug could have happened during reset and some
2936 * controllers fail to wait while a drive is spinning up after
2937 * being hotplugged causing misdetection. By cross checking
2938 * link on/offlineness and classification result, those
2939 * conditions can be reliably detected and retried.
2940 */
2941 nr_unknown = 0;
2942 ata_for_each_dev(dev, link, ALL) {
2943 if (ata_phys_link_online(ata_dev_phys_link(dev))) {
2944 if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2945 ata_dev_dbg(dev, "link online but device misclassified\n");
2946 classes[dev->devno] = ATA_DEV_NONE;
2947 nr_unknown++;
2948 }
2949 } else if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
2950 if (ata_class_enabled(classes[dev->devno]))
2951 ata_dev_dbg(dev,
2952 "link offline, clearing class %d to NONE\n",
2953 classes[dev->devno]);
2954 classes[dev->devno] = ATA_DEV_NONE;
2955 } else if (classes[dev->devno] == ATA_DEV_UNKNOWN) {
2956 ata_dev_dbg(dev,
2957 "link status unknown, clearing UNKNOWN to NONE\n");
2958 classes[dev->devno] = ATA_DEV_NONE;
2959 }
2960 }
2961
2962 if (classify && nr_unknown) {
2963 if (try < max_tries) {
2964 ata_link_warn(link,
2965 "link online but %d devices misclassified, retrying\n",
2966 nr_unknown);
2967 failed_link = link;
2968 rc = -EAGAIN;
2969 goto fail;
2970 }
2971 ata_link_warn(link,
2972 "link online but %d devices misclassified, "
2973 "device detection might fail\n", nr_unknown);
2974 }
2975
2976 /* reset successful, schedule revalidation */
2977 ata_eh_done(link, NULL, ATA_EH_RESET);
2978 if (slave)
2979 ata_eh_done(slave, NULL, ATA_EH_RESET);
2980 ehc->last_reset = jiffies; /* update to completion time */
2981 ehc->i.action |= ATA_EH_REVALIDATE;
2982 link->lpm_policy = ATA_LPM_UNKNOWN; /* reset LPM state */
2983
2984 rc = 0;
2985 out:
2986 /* clear hotplug flag */
2987 ehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2988 if (slave)
2989 sehc->i.flags &= ~ATA_EHI_HOTPLUGGED;
2990
2991 spin_lock_irqsave(ap->lock, flags);
2992 ap->pflags &= ~ATA_PFLAG_RESETTING;
2993 spin_unlock_irqrestore(ap->lock, flags);
2994
2995 return rc;
2996
2997 fail:
2998 /* if SCR isn't accessible on a fan-out port, PMP needs to be reset */
2999 if (!ata_is_host_link(link) &&
3000 sata_scr_read(link, SCR_STATUS, &sstatus))
3001 rc = -ERESTART;
3002
3003 if (try >= max_tries) {
3004 /*
3005 * Thaw host port even if reset failed, so that the port
3006 * can be retried on the next phy event. This risks
3007 * repeated EH runs but seems to be a better tradeoff than
3008 * shutting down a port after a botched hotplug attempt.
3009 */
3010 if (ata_is_host_link(link))
3011 ata_eh_thaw_port(ap);
3012 goto out;
3013 }
3014
3015 now = jiffies;
3016 if (time_before(now, deadline)) {
3017 unsigned long delta = deadline - now;
3018
3019 ata_link_warn(failed_link,
3020 "reset failed (errno=%d), retrying in %u secs\n",
3021 rc, DIV_ROUND_UP(jiffies_to_msecs(delta), 1000));
3022
3023 ata_eh_release(ap);
3024 while (delta)
3025 delta = schedule_timeout_uninterruptible(delta);
3026 ata_eh_acquire(ap);
3027 }
3028
3029 /*
3030 * While disks spinup behind PMP, some controllers fail sending SRST.
3031 * They need to be reset - as well as the PMP - before retrying.
3032 */
3033 if (rc == -ERESTART) {
3034 if (ata_is_host_link(link))
3035 ata_eh_thaw_port(ap);
3036 goto out;
3037 }
3038
3039 if (try == max_tries - 1) {
3040 sata_down_spd_limit(link, 0);
3041 if (slave)
3042 sata_down_spd_limit(slave, 0);
3043 } else if (rc == -EPIPE)
3044 sata_down_spd_limit(failed_link, 0);
3045
3046 if (hardreset)
3047 reset = hardreset;
3048 goto retry;
3049}
3050
3051static inline void ata_eh_pull_park_action(struct ata_port *ap)
3052{
3053 struct ata_link *link;
3054 struct ata_device *dev;
3055 unsigned long flags;
3056
3057 /*
3058 * This function can be thought of as an extended version of
3059 * ata_eh_about_to_do() specially crafted to accommodate the
3060 * requirements of ATA_EH_PARK handling. Since the EH thread
3061 * does not leave the do {} while () loop in ata_eh_recover as
3062 * long as the timeout for a park request to *one* device on
3063 * the port has not expired, and since we still want to pick
3064 * up park requests to other devices on the same port or
3065 * timeout updates for the same device, we have to pull
3066 * ATA_EH_PARK actions from eh_info into eh_context.i
3067 * ourselves at the beginning of each pass over the loop.
3068 *
3069 * Additionally, all write accesses to &ap->park_req_pending
3070 * through reinit_completion() (see below) or complete_all()
3071 * (see ata_scsi_park_store()) are protected by the host lock.
3072 * As a result we have that park_req_pending.done is zero on
3073 * exit from this function, i.e. when ATA_EH_PARK actions for
3074 * *all* devices on port ap have been pulled into the
3075 * respective eh_context structs. If, and only if,
3076 * park_req_pending.done is non-zero by the time we reach
3077 * wait_for_completion_timeout(), another ATA_EH_PARK action
3078 * has been scheduled for at least one of the devices on port
3079 * ap and we have to cycle over the do {} while () loop in
3080 * ata_eh_recover() again.
3081 */
3082
3083 spin_lock_irqsave(ap->lock, flags);
3084 reinit_completion(&ap->park_req_pending);
3085 ata_for_each_link(link, ap, EDGE) {
3086 ata_for_each_dev(dev, link, ALL) {
3087 struct ata_eh_info *ehi = &link->eh_info;
3088
3089 link->eh_context.i.dev_action[dev->devno] |=
3090 ehi->dev_action[dev->devno] & ATA_EH_PARK;
3091 ata_eh_clear_action(link, dev, ehi, ATA_EH_PARK);
3092 }
3093 }
3094 spin_unlock_irqrestore(ap->lock, flags);
3095}
3096
3097static void ata_eh_park_issue_cmd(struct ata_device *dev, int park)
3098{
3099 struct ata_eh_context *ehc = &dev->link->eh_context;
3100 struct ata_taskfile tf;
3101 unsigned int err_mask;
3102
3103 ata_tf_init(dev, &tf);
3104 if (park) {
3105 ehc->unloaded_mask |= 1 << dev->devno;
3106 tf.command = ATA_CMD_IDLEIMMEDIATE;
3107 tf.feature = 0x44;
3108 tf.lbal = 0x4c;
3109 tf.lbam = 0x4e;
3110 tf.lbah = 0x55;
3111 } else {
3112 ehc->unloaded_mask &= ~(1 << dev->devno);
3113 tf.command = ATA_CMD_CHK_POWER;
3114 }
3115
3116 tf.flags |= ATA_TFLAG_DEVICE | ATA_TFLAG_ISADDR;
3117 tf.protocol = ATA_PROT_NODATA;
3118 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3119 if (park && (err_mask || tf.lbal != 0xc4)) {
3120 ata_dev_err(dev, "head unload failed!\n");
3121 ehc->unloaded_mask &= ~(1 << dev->devno);
3122 }
3123}
3124
3125static int ata_eh_revalidate_and_attach(struct ata_link *link,
3126 struct ata_device **r_failed_dev)
3127{
3128 struct ata_port *ap = link->ap;
3129 struct ata_eh_context *ehc = &link->eh_context;
3130 struct ata_device *dev;
3131 unsigned int new_mask = 0;
3132 unsigned long flags;
3133 int rc = 0;
3134
3135 DPRINTK("ENTER\n");
3136
3137 /* For PATA drive side cable detection to work, IDENTIFY must
3138 * be done backwards such that PDIAG- is released by the slave
3139 * device before the master device is identified.
3140 */
3141 ata_for_each_dev(dev, link, ALL_REVERSE) {
3142 unsigned int action = ata_eh_dev_action(dev);
3143 unsigned int readid_flags = 0;
3144
3145 if (ehc->i.flags & ATA_EHI_DID_RESET)
3146 readid_flags |= ATA_READID_POSTRESET;
3147
3148 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
3149 WARN_ON(dev->class == ATA_DEV_PMP);
3150
3151 if (ata_phys_link_offline(ata_dev_phys_link(dev))) {
3152 rc = -EIO;
3153 goto err;
3154 }
3155
3156 ata_eh_about_to_do(link, dev, ATA_EH_REVALIDATE);
3157 rc = ata_dev_revalidate(dev, ehc->classes[dev->devno],
3158 readid_flags);
3159 if (rc)
3160 goto err;
3161
3162 ata_eh_done(link, dev, ATA_EH_REVALIDATE);
3163
3164 /* Configuration may have changed, reconfigure
3165 * transfer mode.
3166 */
3167 ehc->i.flags |= ATA_EHI_SETMODE;
3168
3169 /* schedule the scsi_rescan_device() here */
3170 schedule_work(&(ap->scsi_rescan_task));
3171 } else if (dev->class == ATA_DEV_UNKNOWN &&
3172 ehc->tries[dev->devno] &&
3173 ata_class_enabled(ehc->classes[dev->devno])) {
3174 /* Temporarily set dev->class, it will be
3175 * permanently set once all configurations are
3176 * complete. This is necessary because new
3177 * device configuration is done in two
3178 * separate loops.
3179 */
3180 dev->class = ehc->classes[dev->devno];
3181
3182 if (dev->class == ATA_DEV_PMP)
3183 rc = sata_pmp_attach(dev);
3184 else
3185 rc = ata_dev_read_id(dev, &dev->class,
3186 readid_flags, dev->id);
3187
3188 /* read_id might have changed class, store and reset */
3189 ehc->classes[dev->devno] = dev->class;
3190 dev->class = ATA_DEV_UNKNOWN;
3191
3192 switch (rc) {
3193 case 0:
3194 /* clear error info accumulated during probe */
3195 ata_ering_clear(&dev->ering);
3196 new_mask |= 1 << dev->devno;
3197 break;
3198 case -ENOENT:
3199 /* IDENTIFY was issued to non-existent
3200 * device. No need to reset. Just
3201 * thaw and ignore the device.
3202 */
3203 ata_eh_thaw_port(ap);
3204 break;
3205 default:
3206 goto err;
3207 }
3208 }
3209 }
3210
3211 /* PDIAG- should have been released, ask cable type if post-reset */
3212 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ata_is_host_link(link)) {
3213 if (ap->ops->cable_detect)
3214 ap->cbl = ap->ops->cable_detect(ap);
3215 ata_force_cbl(ap);
3216 }
3217
3218 /* Configure new devices forward such that user doesn't see
3219 * device detection messages backwards.
3220 */
3221 ata_for_each_dev(dev, link, ALL) {
3222 if (!(new_mask & (1 << dev->devno)))
3223 continue;
3224
3225 dev->class = ehc->classes[dev->devno];
3226
3227 if (dev->class == ATA_DEV_PMP)
3228 continue;
3229
3230 ehc->i.flags |= ATA_EHI_PRINTINFO;
3231 rc = ata_dev_configure(dev);
3232 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
3233 if (rc) {
3234 dev->class = ATA_DEV_UNKNOWN;
3235 goto err;
3236 }
3237
3238 spin_lock_irqsave(ap->lock, flags);
3239 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
3240 spin_unlock_irqrestore(ap->lock, flags);
3241
3242 /* new device discovered, configure xfermode */
3243 ehc->i.flags |= ATA_EHI_SETMODE;
3244 }
3245
3246 return 0;
3247
3248 err:
3249 *r_failed_dev = dev;
3250 DPRINTK("EXIT rc=%d\n", rc);
3251 return rc;
3252}
3253
3254/**
3255 * ata_set_mode - Program timings and issue SET FEATURES - XFER
3256 * @link: link on which timings will be programmed
3257 * @r_failed_dev: out parameter for failed device
3258 *
3259 * Set ATA device disk transfer mode (PIO3, UDMA6, etc.). If
3260 * ata_set_mode() fails, pointer to the failing device is
3261 * returned in @r_failed_dev.
3262 *
3263 * LOCKING:
3264 * PCI/etc. bus probe sem.
3265 *
3266 * RETURNS:
3267 * 0 on success, negative errno otherwise
3268 */
3269int ata_set_mode(struct ata_link *link, struct ata_device **r_failed_dev)
3270{
3271 struct ata_port *ap = link->ap;
3272 struct ata_device *dev;
3273 int rc;
3274
3275 /* if data transfer is verified, clear DUBIOUS_XFER on ering top */
3276 ata_for_each_dev(dev, link, ENABLED) {
3277 if (!(dev->flags & ATA_DFLAG_DUBIOUS_XFER)) {
3278 struct ata_ering_entry *ent;
3279
3280 ent = ata_ering_top(&dev->ering);
3281 if (ent)
3282 ent->eflags &= ~ATA_EFLAG_DUBIOUS_XFER;
3283 }
3284 }
3285
3286 /* has private set_mode? */
3287 if (ap->ops->set_mode)
3288 rc = ap->ops->set_mode(link, r_failed_dev);
3289 else
3290 rc = ata_do_set_mode(link, r_failed_dev);
3291
3292 /* if transfer mode has changed, set DUBIOUS_XFER on device */
3293 ata_for_each_dev(dev, link, ENABLED) {
3294 struct ata_eh_context *ehc = &link->eh_context;
3295 u8 saved_xfer_mode = ehc->saved_xfer_mode[dev->devno];
3296 u8 saved_ncq = !!(ehc->saved_ncq_enabled & (1 << dev->devno));
3297
3298 if (dev->xfer_mode != saved_xfer_mode ||
3299 ata_ncq_enabled(dev) != saved_ncq)
3300 dev->flags |= ATA_DFLAG_DUBIOUS_XFER;
3301 }
3302
3303 return rc;
3304}
3305
3306/**
3307 * atapi_eh_clear_ua - Clear ATAPI UNIT ATTENTION after reset
3308 * @dev: ATAPI device to clear UA for
3309 *
3310 * Resets and other operations can make an ATAPI device raise
3311 * UNIT ATTENTION which causes the next operation to fail. This
3312 * function clears UA.
3313 *
3314 * LOCKING:
3315 * EH context (may sleep).
3316 *
3317 * RETURNS:
3318 * 0 on success, -errno on failure.
3319 */
3320static int atapi_eh_clear_ua(struct ata_device *dev)
3321{
3322 int i;
3323
3324 for (i = 0; i < ATA_EH_UA_TRIES; i++) {
3325 u8 *sense_buffer = dev->link->ap->sector_buf;
3326 u8 sense_key = 0;
3327 unsigned int err_mask;
3328
3329 err_mask = atapi_eh_tur(dev, &sense_key);
3330 if (err_mask != 0 && err_mask != AC_ERR_DEV) {
3331 ata_dev_warn(dev,
3332 "TEST_UNIT_READY failed (err_mask=0x%x)\n",
3333 err_mask);
3334 return -EIO;
3335 }
3336
3337 if (!err_mask || sense_key != UNIT_ATTENTION)
3338 return 0;
3339
3340 err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
3341 if (err_mask) {
3342 ata_dev_warn(dev, "failed to clear "
3343 "UNIT ATTENTION (err_mask=0x%x)\n", err_mask);
3344 return -EIO;
3345 }
3346 }
3347
3348 ata_dev_warn(dev, "UNIT ATTENTION persists after %d tries\n",
3349 ATA_EH_UA_TRIES);
3350
3351 return 0;
3352}
3353
3354/**
3355 * ata_eh_maybe_retry_flush - Retry FLUSH if necessary
3356 * @dev: ATA device which may need FLUSH retry
3357 *
3358 * If @dev failed FLUSH, it needs to be reported upper layer
3359 * immediately as it means that @dev failed to remap and already
3360 * lost at least a sector and further FLUSH retrials won't make
3361 * any difference to the lost sector. However, if FLUSH failed
3362 * for other reasons, for example transmission error, FLUSH needs
3363 * to be retried.
3364 *
3365 * This function determines whether FLUSH failure retry is
3366 * necessary and performs it if so.
3367 *
3368 * RETURNS:
3369 * 0 if EH can continue, -errno if EH needs to be repeated.
3370 */
3371static int ata_eh_maybe_retry_flush(struct ata_device *dev)
3372{
3373 struct ata_link *link = dev->link;
3374 struct ata_port *ap = link->ap;
3375 struct ata_queued_cmd *qc;
3376 struct ata_taskfile tf;
3377 unsigned int err_mask;
3378 int rc = 0;
3379
3380 /* did flush fail for this device? */
3381 if (!ata_tag_valid(link->active_tag))
3382 return 0;
3383
3384 qc = __ata_qc_from_tag(ap, link->active_tag);
3385 if (qc->dev != dev || (qc->tf.command != ATA_CMD_FLUSH_EXT &&
3386 qc->tf.command != ATA_CMD_FLUSH))
3387 return 0;
3388
3389 /* if the device failed it, it should be reported to upper layers */
3390 if (qc->err_mask & AC_ERR_DEV)
3391 return 0;
3392
3393 /* flush failed for some other reason, give it another shot */
3394 ata_tf_init(dev, &tf);
3395
3396 tf.command = qc->tf.command;
3397 tf.flags |= ATA_TFLAG_DEVICE;
3398 tf.protocol = ATA_PROT_NODATA;
3399
3400 ata_dev_warn(dev, "retrying FLUSH 0x%x Emask 0x%x\n",
3401 tf.command, qc->err_mask);
3402
3403 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_NONE, NULL, 0, 0);
3404 if (!err_mask) {
3405 /*
3406 * FLUSH is complete but there's no way to
3407 * successfully complete a failed command from EH.
3408 * Making sure retry is allowed at least once and
3409 * retrying it should do the trick - whatever was in
3410 * the cache is already on the platter and this won't
3411 * cause infinite loop.
3412 */
3413 qc->scsicmd->allowed = max(qc->scsicmd->allowed, 1);
3414 } else {
3415 ata_dev_warn(dev, "FLUSH failed Emask 0x%x\n",
3416 err_mask);
3417 rc = -EIO;
3418
3419 /* if device failed it, report it to upper layers */
3420 if (err_mask & AC_ERR_DEV) {
3421 qc->err_mask |= AC_ERR_DEV;
3422 qc->result_tf = tf;
3423 if (!(ap->pflags & ATA_PFLAG_FROZEN))
3424 rc = 0;
3425 }
3426 }
3427 return rc;
3428}
3429
3430/**
3431 * ata_eh_set_lpm - configure SATA interface power management
3432 * @link: link to configure power management
3433 * @policy: the link power management policy
3434 * @r_failed_dev: out parameter for failed device
3435 *
3436 * Enable SATA Interface power management. This will enable
3437 * Device Interface Power Management (DIPM) for min_power and
3438 * medium_power_with_dipm policies, and then call driver specific
3439 * callbacks for enabling Host Initiated Power management.
3440 *
3441 * LOCKING:
3442 * EH context.
3443 *
3444 * RETURNS:
3445 * 0 on success, -errno on failure.
3446 */
3447static int ata_eh_set_lpm(struct ata_link *link, enum ata_lpm_policy policy,
3448 struct ata_device **r_failed_dev)
3449{
3450 struct ata_port *ap = ata_is_host_link(link) ? link->ap : NULL;
3451 struct ata_eh_context *ehc = &link->eh_context;
3452 struct ata_device *dev, *link_dev = NULL, *lpm_dev = NULL;
3453 enum ata_lpm_policy old_policy = link->lpm_policy;
3454 bool no_dipm = link->ap->flags & ATA_FLAG_NO_DIPM;
3455 unsigned int hints = ATA_LPM_EMPTY | ATA_LPM_HIPM;
3456 unsigned int err_mask;
3457 int rc;
3458
3459 /* if the link or host doesn't do LPM, noop */
3460 if ((link->flags & ATA_LFLAG_NO_LPM) || (ap && !ap->ops->set_lpm))
3461 return 0;
3462
3463 /*
3464 * DIPM is enabled only for MIN_POWER as some devices
3465 * misbehave when the host NACKs transition to SLUMBER. Order
3466 * device and link configurations such that the host always
3467 * allows DIPM requests.
3468 */
3469 ata_for_each_dev(dev, link, ENABLED) {
3470 bool hipm = ata_id_has_hipm(dev->id);
3471 bool dipm = ata_id_has_dipm(dev->id) && !no_dipm;
3472
3473 /* find the first enabled and LPM enabled devices */
3474 if (!link_dev)
3475 link_dev = dev;
3476
3477 if (!lpm_dev && (hipm || dipm))
3478 lpm_dev = dev;
3479
3480 hints &= ~ATA_LPM_EMPTY;
3481 if (!hipm)
3482 hints &= ~ATA_LPM_HIPM;
3483
3484 /* disable DIPM before changing link config */
3485 if (policy < ATA_LPM_MED_POWER_WITH_DIPM && dipm) {
3486 err_mask = ata_dev_set_feature(dev,
3487 SETFEATURES_SATA_DISABLE, SATA_DIPM);
3488 if (err_mask && err_mask != AC_ERR_DEV) {
3489 ata_dev_warn(dev,
3490 "failed to disable DIPM, Emask 0x%x\n",
3491 err_mask);
3492 rc = -EIO;
3493 goto fail;
3494 }
3495 }
3496 }
3497
3498 if (ap) {
3499 rc = ap->ops->set_lpm(link, policy, hints);
3500 if (!rc && ap->slave_link)
3501 rc = ap->ops->set_lpm(ap->slave_link, policy, hints);
3502 } else
3503 rc = sata_pmp_set_lpm(link, policy, hints);
3504
3505 /*
3506 * Attribute link config failure to the first (LPM) enabled
3507 * device on the link.
3508 */
3509 if (rc) {
3510 if (rc == -EOPNOTSUPP) {
3511 link->flags |= ATA_LFLAG_NO_LPM;
3512 return 0;
3513 }
3514 dev = lpm_dev ? lpm_dev : link_dev;
3515 goto fail;
3516 }
3517
3518 /*
3519 * Low level driver acked the transition. Issue DIPM command
3520 * with the new policy set.
3521 */
3522 link->lpm_policy = policy;
3523 if (ap && ap->slave_link)
3524 ap->slave_link->lpm_policy = policy;
3525
3526 /* host config updated, enable DIPM if transitioning to MIN_POWER */
3527 ata_for_each_dev(dev, link, ENABLED) {
3528 if (policy >= ATA_LPM_MED_POWER_WITH_DIPM && !no_dipm &&
3529 ata_id_has_dipm(dev->id)) {
3530 err_mask = ata_dev_set_feature(dev,
3531 SETFEATURES_SATA_ENABLE, SATA_DIPM);
3532 if (err_mask && err_mask != AC_ERR_DEV) {
3533 ata_dev_warn(dev,
3534 "failed to enable DIPM, Emask 0x%x\n",
3535 err_mask);
3536 rc = -EIO;
3537 goto fail;
3538 }
3539 }
3540 }
3541
3542 link->last_lpm_change = jiffies;
3543 link->flags |= ATA_LFLAG_CHANGED;
3544
3545 return 0;
3546
3547fail:
3548 /* restore the old policy */
3549 link->lpm_policy = old_policy;
3550 if (ap && ap->slave_link)
3551 ap->slave_link->lpm_policy = old_policy;
3552
3553 /* if no device or only one more chance is left, disable LPM */
3554 if (!dev || ehc->tries[dev->devno] <= 2) {
3555 ata_link_warn(link, "disabling LPM on the link\n");
3556 link->flags |= ATA_LFLAG_NO_LPM;
3557 }
3558 if (r_failed_dev)
3559 *r_failed_dev = dev;
3560 return rc;
3561}
3562
3563int ata_link_nr_enabled(struct ata_link *link)
3564{
3565 struct ata_device *dev;
3566 int cnt = 0;
3567
3568 ata_for_each_dev(dev, link, ENABLED)
3569 cnt++;
3570 return cnt;
3571}
3572
3573static int ata_link_nr_vacant(struct ata_link *link)
3574{
3575 struct ata_device *dev;
3576 int cnt = 0;
3577
3578 ata_for_each_dev(dev, link, ALL)
3579 if (dev->class == ATA_DEV_UNKNOWN)
3580 cnt++;
3581 return cnt;
3582}
3583
3584static int ata_eh_skip_recovery(struct ata_link *link)
3585{
3586 struct ata_port *ap = link->ap;
3587 struct ata_eh_context *ehc = &link->eh_context;
3588 struct ata_device *dev;
3589
3590 /* skip disabled links */
3591 if (link->flags & ATA_LFLAG_DISABLED)
3592 return 1;
3593
3594 /* skip if explicitly requested */
3595 if (ehc->i.flags & ATA_EHI_NO_RECOVERY)
3596 return 1;
3597
3598 /* thaw frozen port and recover failed devices */
3599 if ((ap->pflags & ATA_PFLAG_FROZEN) || ata_link_nr_enabled(link))
3600 return 0;
3601
3602 /* reset at least once if reset is requested */
3603 if ((ehc->i.action & ATA_EH_RESET) &&
3604 !(ehc->i.flags & ATA_EHI_DID_RESET))
3605 return 0;
3606
3607 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
3608 ata_for_each_dev(dev, link, ALL) {
3609 if (dev->class == ATA_DEV_UNKNOWN &&
3610 ehc->classes[dev->devno] != ATA_DEV_NONE)
3611 return 0;
3612 }
3613
3614 return 1;
3615}
3616
3617static int ata_count_probe_trials_cb(struct ata_ering_entry *ent, void *void_arg)
3618{
3619 u64 interval = msecs_to_jiffies(ATA_EH_PROBE_TRIAL_INTERVAL);
3620 u64 now = get_jiffies_64();
3621 int *trials = void_arg;
3622
3623 if ((ent->eflags & ATA_EFLAG_OLD_ER) ||
3624 (ent->timestamp < now - min(now, interval)))
3625 return -1;
3626
3627 (*trials)++;
3628 return 0;
3629}
3630
3631static int ata_eh_schedule_probe(struct ata_device *dev)
3632{
3633 struct ata_eh_context *ehc = &dev->link->eh_context;
3634 struct ata_link *link = ata_dev_phys_link(dev);
3635 int trials = 0;
3636
3637 if (!(ehc->i.probe_mask & (1 << dev->devno)) ||
3638 (ehc->did_probe_mask & (1 << dev->devno)))
3639 return 0;
3640
3641 ata_eh_detach_dev(dev);
3642 ata_dev_init(dev);
3643 ehc->did_probe_mask |= (1 << dev->devno);
3644 ehc->i.action |= ATA_EH_RESET;
3645 ehc->saved_xfer_mode[dev->devno] = 0;
3646 ehc->saved_ncq_enabled &= ~(1 << dev->devno);
3647
3648 /* the link maybe in a deep sleep, wake it up */
3649 if (link->lpm_policy > ATA_LPM_MAX_POWER) {
3650 if (ata_is_host_link(link))
3651 link->ap->ops->set_lpm(link, ATA_LPM_MAX_POWER,
3652 ATA_LPM_EMPTY);
3653 else
3654 sata_pmp_set_lpm(link, ATA_LPM_MAX_POWER,
3655 ATA_LPM_EMPTY);
3656 }
3657
3658 /* Record and count probe trials on the ering. The specific
3659 * error mask used is irrelevant. Because a successful device
3660 * detection clears the ering, this count accumulates only if
3661 * there are consecutive failed probes.
3662 *
3663 * If the count is equal to or higher than ATA_EH_PROBE_TRIALS
3664 * in the last ATA_EH_PROBE_TRIAL_INTERVAL, link speed is
3665 * forced to 1.5Gbps.
3666 *
3667 * This is to work around cases where failed link speed
3668 * negotiation results in device misdetection leading to
3669 * infinite DEVXCHG or PHRDY CHG events.
3670 */
3671 ata_ering_record(&dev->ering, 0, AC_ERR_OTHER);
3672 ata_ering_map(&dev->ering, ata_count_probe_trials_cb, &trials);
3673
3674 if (trials > ATA_EH_PROBE_TRIALS)
3675 sata_down_spd_limit(link, 1);
3676
3677 return 1;
3678}
3679
3680static int ata_eh_handle_dev_fail(struct ata_device *dev, int err)
3681{
3682 struct ata_eh_context *ehc = &dev->link->eh_context;
3683
3684 /* -EAGAIN from EH routine indicates retry without prejudice.
3685 * The requester is responsible for ensuring forward progress.
3686 */
3687 if (err != -EAGAIN)
3688 ehc->tries[dev->devno]--;
3689
3690 switch (err) {
3691 case -ENODEV:
3692 /* device missing or wrong IDENTIFY data, schedule probing */
3693 ehc->i.probe_mask |= (1 << dev->devno);
3694 /* fall through */
3695 case -EINVAL:
3696 /* give it just one more chance */
3697 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
3698 /* fall through */
3699 case -EIO:
3700 if (ehc->tries[dev->devno] == 1) {
3701 /* This is the last chance, better to slow
3702 * down than lose it.
3703 */
3704 sata_down_spd_limit(ata_dev_phys_link(dev), 0);
3705 if (dev->pio_mode > XFER_PIO_0)
3706 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
3707 }
3708 }
3709
3710 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
3711 /* disable device if it has used up all its chances */
3712 ata_dev_disable(dev);
3713
3714 /* detach if offline */
3715 if (ata_phys_link_offline(ata_dev_phys_link(dev)))
3716 ata_eh_detach_dev(dev);
3717
3718 /* schedule probe if necessary */
3719 if (ata_eh_schedule_probe(dev)) {
3720 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3721 memset(ehc->cmd_timeout_idx[dev->devno], 0,
3722 sizeof(ehc->cmd_timeout_idx[dev->devno]));
3723 }
3724
3725 return 1;
3726 } else {
3727 ehc->i.action |= ATA_EH_RESET;
3728 return 0;
3729 }
3730}
3731
3732/**
3733 * ata_eh_recover - recover host port after error
3734 * @ap: host port to recover
3735 * @prereset: prereset method (can be NULL)
3736 * @softreset: softreset method (can be NULL)
3737 * @hardreset: hardreset method (can be NULL)
3738 * @postreset: postreset method (can be NULL)
3739 * @r_failed_link: out parameter for failed link
3740 *
3741 * This is the alpha and omega, eum and yang, heart and soul of
3742 * libata exception handling. On entry, actions required to
3743 * recover each link and hotplug requests are recorded in the
3744 * link's eh_context. This function executes all the operations
3745 * with appropriate retrials and fallbacks to resurrect failed
3746 * devices, detach goners and greet newcomers.
3747 *
3748 * LOCKING:
3749 * Kernel thread context (may sleep).
3750 *
3751 * RETURNS:
3752 * 0 on success, -errno on failure.
3753 */
3754int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
3755 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
3756 ata_postreset_fn_t postreset,
3757 struct ata_link **r_failed_link)
3758{
3759 struct ata_link *link;
3760 struct ata_device *dev;
3761 int rc, nr_fails;
3762 unsigned long flags, deadline;
3763
3764 DPRINTK("ENTER\n");
3765
3766 /* prep for recovery */
3767 ata_for_each_link(link, ap, EDGE) {
3768 struct ata_eh_context *ehc = &link->eh_context;
3769
3770 /* re-enable link? */
3771 if (ehc->i.action & ATA_EH_ENABLE_LINK) {
3772 ata_eh_about_to_do(link, NULL, ATA_EH_ENABLE_LINK);
3773 spin_lock_irqsave(ap->lock, flags);
3774 link->flags &= ~ATA_LFLAG_DISABLED;
3775 spin_unlock_irqrestore(ap->lock, flags);
3776 ata_eh_done(link, NULL, ATA_EH_ENABLE_LINK);
3777 }
3778
3779 ata_for_each_dev(dev, link, ALL) {
3780 if (link->flags & ATA_LFLAG_NO_RETRY)
3781 ehc->tries[dev->devno] = 1;
3782 else
3783 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
3784
3785 /* collect port action mask recorded in dev actions */
3786 ehc->i.action |= ehc->i.dev_action[dev->devno] &
3787 ~ATA_EH_PERDEV_MASK;
3788 ehc->i.dev_action[dev->devno] &= ATA_EH_PERDEV_MASK;
3789
3790 /* process hotplug request */
3791 if (dev->flags & ATA_DFLAG_DETACH)
3792 ata_eh_detach_dev(dev);
3793
3794 /* schedule probe if necessary */
3795 if (!ata_dev_enabled(dev))
3796 ata_eh_schedule_probe(dev);
3797 }
3798 }
3799
3800 retry:
3801 rc = 0;
3802
3803 /* if UNLOADING, finish immediately */
3804 if (ap->pflags & ATA_PFLAG_UNLOADING)
3805 goto out;
3806
3807 /* prep for EH */
3808 ata_for_each_link(link, ap, EDGE) {
3809 struct ata_eh_context *ehc = &link->eh_context;
3810
3811 /* skip EH if possible. */
3812 if (ata_eh_skip_recovery(link))
3813 ehc->i.action = 0;
3814
3815 ata_for_each_dev(dev, link, ALL)
3816 ehc->classes[dev->devno] = ATA_DEV_UNKNOWN;
3817 }
3818
3819 /* reset */
3820 ata_for_each_link(link, ap, EDGE) {
3821 struct ata_eh_context *ehc = &link->eh_context;
3822
3823 if (!(ehc->i.action & ATA_EH_RESET))
3824 continue;
3825
3826 rc = ata_eh_reset(link, ata_link_nr_vacant(link),
3827 prereset, softreset, hardreset, postreset);
3828 if (rc) {
3829 ata_link_err(link, "reset failed, giving up\n");
3830 goto out;
3831 }
3832 }
3833
3834 do {
3835 unsigned long now;
3836
3837 /*
3838 * clears ATA_EH_PARK in eh_info and resets
3839 * ap->park_req_pending
3840 */
3841 ata_eh_pull_park_action(ap);
3842
3843 deadline = jiffies;
3844 ata_for_each_link(link, ap, EDGE) {
3845 ata_for_each_dev(dev, link, ALL) {
3846 struct ata_eh_context *ehc = &link->eh_context;
3847 unsigned long tmp;
3848
3849 if (dev->class != ATA_DEV_ATA &&
3850 dev->class != ATA_DEV_ZAC)
3851 continue;
3852 if (!(ehc->i.dev_action[dev->devno] &
3853 ATA_EH_PARK))
3854 continue;
3855 tmp = dev->unpark_deadline;
3856 if (time_before(deadline, tmp))
3857 deadline = tmp;
3858 else if (time_before_eq(tmp, jiffies))
3859 continue;
3860 if (ehc->unloaded_mask & (1 << dev->devno))
3861 continue;
3862
3863 ata_eh_park_issue_cmd(dev, 1);
3864 }
3865 }
3866
3867 now = jiffies;
3868 if (time_before_eq(deadline, now))
3869 break;
3870
3871 ata_eh_release(ap);
3872 deadline = wait_for_completion_timeout(&ap->park_req_pending,
3873 deadline - now);
3874 ata_eh_acquire(ap);
3875 } while (deadline);
3876 ata_for_each_link(link, ap, EDGE) {
3877 ata_for_each_dev(dev, link, ALL) {
3878 if (!(link->eh_context.unloaded_mask &
3879 (1 << dev->devno)))
3880 continue;
3881
3882 ata_eh_park_issue_cmd(dev, 0);
3883 ata_eh_done(link, dev, ATA_EH_PARK);
3884 }
3885 }
3886
3887 /* the rest */
3888 nr_fails = 0;
3889 ata_for_each_link(link, ap, PMP_FIRST) {
3890 struct ata_eh_context *ehc = &link->eh_context;
3891
3892 if (sata_pmp_attached(ap) && ata_is_host_link(link))
3893 goto config_lpm;
3894
3895 /* revalidate existing devices and attach new ones */
3896 rc = ata_eh_revalidate_and_attach(link, &dev);
3897 if (rc)
3898 goto rest_fail;
3899
3900 /* if PMP got attached, return, pmp EH will take care of it */
3901 if (link->device->class == ATA_DEV_PMP) {
3902 ehc->i.action = 0;
3903 return 0;
3904 }
3905
3906 /* configure transfer mode if necessary */
3907 if (ehc->i.flags & ATA_EHI_SETMODE) {
3908 rc = ata_set_mode(link, &dev);
3909 if (rc)
3910 goto rest_fail;
3911 ehc->i.flags &= ~ATA_EHI_SETMODE;
3912 }
3913
3914 /* If reset has been issued, clear UA to avoid
3915 * disrupting the current users of the device.
3916 */
3917 if (ehc->i.flags & ATA_EHI_DID_RESET) {
3918 ata_for_each_dev(dev, link, ALL) {
3919 if (dev->class != ATA_DEV_ATAPI)
3920 continue;
3921 rc = atapi_eh_clear_ua(dev);
3922 if (rc)
3923 goto rest_fail;
3924 if (zpodd_dev_enabled(dev))
3925 zpodd_post_poweron(dev);
3926 }
3927 }
3928
3929 /* retry flush if necessary */
3930 ata_for_each_dev(dev, link, ALL) {
3931 if (dev->class != ATA_DEV_ATA &&
3932 dev->class != ATA_DEV_ZAC)
3933 continue;
3934 rc = ata_eh_maybe_retry_flush(dev);
3935 if (rc)
3936 goto rest_fail;
3937 }
3938
3939 config_lpm:
3940 /* configure link power saving */
3941 if (link->lpm_policy != ap->target_lpm_policy) {
3942 rc = ata_eh_set_lpm(link, ap->target_lpm_policy, &dev);
3943 if (rc)
3944 goto rest_fail;
3945 }
3946
3947 /* this link is okay now */
3948 ehc->i.flags = 0;
3949 continue;
3950
3951 rest_fail:
3952 nr_fails++;
3953 if (dev)
3954 ata_eh_handle_dev_fail(dev, rc);
3955
3956 if (ap->pflags & ATA_PFLAG_FROZEN) {
3957 /* PMP reset requires working host port.
3958 * Can't retry if it's frozen.
3959 */
3960 if (sata_pmp_attached(ap))
3961 goto out;
3962 break;
3963 }
3964 }
3965
3966 if (nr_fails)
3967 goto retry;
3968
3969 out:
3970 if (rc && r_failed_link)
3971 *r_failed_link = link;
3972
3973 DPRINTK("EXIT, rc=%d\n", rc);
3974 return rc;
3975}
3976
3977/**
3978 * ata_eh_finish - finish up EH
3979 * @ap: host port to finish EH for
3980 *
3981 * Recovery is complete. Clean up EH states and retry or finish
3982 * failed qcs.
3983 *
3984 * LOCKING:
3985 * None.
3986 */
3987void ata_eh_finish(struct ata_port *ap)
3988{
3989 struct ata_queued_cmd *qc;
3990 int tag;
3991
3992 /* retry or finish qcs */
3993 ata_qc_for_each_raw(ap, qc, tag) {
3994 if (!(qc->flags & ATA_QCFLAG_FAILED))
3995 continue;
3996
3997 if (qc->err_mask) {
3998 /* FIXME: Once EH migration is complete,
3999 * generate sense data in this function,
4000 * considering both err_mask and tf.
4001 */
4002 if (qc->flags & ATA_QCFLAG_RETRY)
4003 ata_eh_qc_retry(qc);
4004 else
4005 ata_eh_qc_complete(qc);
4006 } else {
4007 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
4008 ata_eh_qc_complete(qc);
4009 } else {
4010 /* feed zero TF to sense generation */
4011 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
4012 ata_eh_qc_retry(qc);
4013 }
4014 }
4015 }
4016
4017 /* make sure nr_active_links is zero after EH */
4018 WARN_ON(ap->nr_active_links);
4019 ap->nr_active_links = 0;
4020}
4021
4022/**
4023 * ata_do_eh - do standard error handling
4024 * @ap: host port to handle error for
4025 *
4026 * @prereset: prereset method (can be NULL)
4027 * @softreset: softreset method (can be NULL)
4028 * @hardreset: hardreset method (can be NULL)
4029 * @postreset: postreset method (can be NULL)
4030 *
4031 * Perform standard error handling sequence.
4032 *
4033 * LOCKING:
4034 * Kernel thread context (may sleep).
4035 */
4036void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
4037 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
4038 ata_postreset_fn_t postreset)
4039{
4040 struct ata_device *dev;
4041 int rc;
4042
4043 ata_eh_autopsy(ap);
4044 ata_eh_report(ap);
4045
4046 rc = ata_eh_recover(ap, prereset, softreset, hardreset, postreset,
4047 NULL);
4048 if (rc) {
4049 ata_for_each_dev(dev, &ap->link, ALL)
4050 ata_dev_disable(dev);
4051 }
4052
4053 ata_eh_finish(ap);
4054}
4055
4056/**
4057 * ata_std_error_handler - standard error handler
4058 * @ap: host port to handle error for
4059 *
4060 * Standard error handler
4061 *
4062 * LOCKING:
4063 * Kernel thread context (may sleep).
4064 */
4065void ata_std_error_handler(struct ata_port *ap)
4066{
4067 struct ata_port_operations *ops = ap->ops;
4068 ata_reset_fn_t hardreset = ops->hardreset;
4069
4070 /* ignore built-in hardreset if SCR access is not available */
4071 if (hardreset == sata_std_hardreset && !sata_scr_valid(&ap->link))
4072 hardreset = NULL;
4073
4074 ata_do_eh(ap, ops->prereset, ops->softreset, hardreset, ops->postreset);
4075}
4076
4077#ifdef CONFIG_PM
4078/**
4079 * ata_eh_handle_port_suspend - perform port suspend operation
4080 * @ap: port to suspend
4081 *
4082 * Suspend @ap.
4083 *
4084 * LOCKING:
4085 * Kernel thread context (may sleep).
4086 */
4087static void ata_eh_handle_port_suspend(struct ata_port *ap)
4088{
4089 unsigned long flags;
4090 int rc = 0;
4091 struct ata_device *dev;
4092
4093 /* are we suspending? */
4094 spin_lock_irqsave(ap->lock, flags);
4095 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4096 ap->pm_mesg.event & PM_EVENT_RESUME) {
4097 spin_unlock_irqrestore(ap->lock, flags);
4098 return;
4099 }
4100 spin_unlock_irqrestore(ap->lock, flags);
4101
4102 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
4103
4104 /*
4105 * If we have a ZPODD attached, check its zero
4106 * power ready status before the port is frozen.
4107 * Only needed for runtime suspend.
4108 */
4109 if (PMSG_IS_AUTO(ap->pm_mesg)) {
4110 ata_for_each_dev(dev, &ap->link, ENABLED) {
4111 if (zpodd_dev_enabled(dev))
4112 zpodd_on_suspend(dev);
4113 }
4114 }
4115
4116 /* tell ACPI we're suspending */
4117 rc = ata_acpi_on_suspend(ap);
4118 if (rc)
4119 goto out;
4120
4121 /* suspend */
4122 ata_eh_freeze_port(ap);
4123
4124 if (ap->ops->port_suspend)
4125 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
4126
4127 ata_acpi_set_state(ap, ap->pm_mesg);
4128 out:
4129 /* update the flags */
4130 spin_lock_irqsave(ap->lock, flags);
4131
4132 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
4133 if (rc == 0)
4134 ap->pflags |= ATA_PFLAG_SUSPENDED;
4135 else if (ap->pflags & ATA_PFLAG_FROZEN)
4136 ata_port_schedule_eh(ap);
4137
4138 spin_unlock_irqrestore(ap->lock, flags);
4139
4140 return;
4141}
4142
4143/**
4144 * ata_eh_handle_port_resume - perform port resume operation
4145 * @ap: port to resume
4146 *
4147 * Resume @ap.
4148 *
4149 * LOCKING:
4150 * Kernel thread context (may sleep).
4151 */
4152static void ata_eh_handle_port_resume(struct ata_port *ap)
4153{
4154 struct ata_link *link;
4155 struct ata_device *dev;
4156 unsigned long flags;
4157
4158 /* are we resuming? */
4159 spin_lock_irqsave(ap->lock, flags);
4160 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
4161 !(ap->pm_mesg.event & PM_EVENT_RESUME)) {
4162 spin_unlock_irqrestore(ap->lock, flags);
4163 return;
4164 }
4165 spin_unlock_irqrestore(ap->lock, flags);
4166
4167 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
4168
4169 /*
4170 * Error timestamps are in jiffies which doesn't run while
4171 * suspended and PHY events during resume isn't too uncommon.
4172 * When the two are combined, it can lead to unnecessary speed
4173 * downs if the machine is suspended and resumed repeatedly.
4174 * Clear error history.
4175 */
4176 ata_for_each_link(link, ap, HOST_FIRST)
4177 ata_for_each_dev(dev, link, ALL)
4178 ata_ering_clear(&dev->ering);
4179
4180 ata_acpi_set_state(ap, ap->pm_mesg);
4181
4182 if (ap->ops->port_resume)
4183 ap->ops->port_resume(ap);
4184
4185 /* tell ACPI that we're resuming */
4186 ata_acpi_on_resume(ap);
4187
4188 /* update the flags */
4189 spin_lock_irqsave(ap->lock, flags);
4190 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
4191 spin_unlock_irqrestore(ap->lock, flags);
4192}
4193#endif /* CONFIG_PM */