libata: improve EH report formatting
[linux-block.git] / drivers / ata / libata-eh.c
CommitLineData
ece1d636
TH
1/*
2 * libata-eh.c - libata error handling
3 *
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
6 * on emails.
7 *
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
9 *
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
24 * USA.
25 *
26 *
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
29 *
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
32 *
33 */
34
ece1d636
TH
35#include <linux/kernel.h>
36#include <scsi/scsi.h>
37#include <scsi/scsi_host.h>
38#include <scsi/scsi_eh.h>
39#include <scsi/scsi_device.h>
40#include <scsi/scsi_cmnd.h>
c6fd2807 41#include "../scsi/scsi_transport_api.h"
ece1d636
TH
42
43#include <linux/libata.h>
44
45#include "libata.h"
46
7d47e8d4
TH
47enum {
48 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
49 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
50 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
51};
52
31daabda
TH
53/* Waiting in ->prereset can never be reliable. It's sometimes nice
54 * to wait there but it can't be depended upon; otherwise, we wouldn't
55 * be resetting. Just give it enough time for most drives to spin up.
56 */
57enum {
58 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
59};
60
61/* The following table determines how we sequence resets. Each entry
62 * represents timeout for that try. The first try can be soft or
63 * hardreset. All others are hardreset if available. In most cases
64 * the first reset w/ 10sec timeout should succeed. Following entries
65 * are mostly for error handling, hotplug and retarded devices.
66 */
67static const unsigned long ata_eh_reset_timeouts[] = {
68 10 * HZ, /* most drives spin up by 10sec */
69 10 * HZ, /* > 99% working drives spin up before 20sec */
70 35 * HZ, /* give > 30 secs of idleness for retarded devices */
71 5 * HZ, /* and sweet one last chance */
72 /* > 1 min has elapsed, give up */
73};
74
ad9e2762 75static void __ata_port_freeze(struct ata_port *ap);
720ba126 76static void ata_eh_finish(struct ata_port *ap);
6ffa01d8 77#ifdef CONFIG_PM
500530f6
TH
78static void ata_eh_handle_port_suspend(struct ata_port *ap);
79static void ata_eh_handle_port_resume(struct ata_port *ap);
6ffa01d8
TH
80#else /* CONFIG_PM */
81static void ata_eh_handle_port_suspend(struct ata_port *ap)
82{ }
83
84static void ata_eh_handle_port_resume(struct ata_port *ap)
85{ }
6ffa01d8 86#endif /* CONFIG_PM */
ad9e2762 87
b64bbc39
TH
88static void __ata_ehi_pushv_desc(struct ata_eh_info *ehi, const char *fmt,
89 va_list args)
90{
91 ehi->desc_len += vscnprintf(ehi->desc + ehi->desc_len,
92 ATA_EH_DESC_LEN - ehi->desc_len,
93 fmt, args);
94}
95
96/**
97 * __ata_ehi_push_desc - push error description without adding separator
98 * @ehi: target EHI
99 * @fmt: printf format string
100 *
101 * Format string according to @fmt and append it to @ehi->desc.
102 *
103 * LOCKING:
104 * spin_lock_irqsave(host lock)
105 */
106void __ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
107{
108 va_list args;
109
110 va_start(args, fmt);
111 __ata_ehi_pushv_desc(ehi, fmt, args);
112 va_end(args);
113}
114
115/**
116 * ata_ehi_push_desc - push error description with separator
117 * @ehi: target EHI
118 * @fmt: printf format string
119 *
120 * Format string according to @fmt and append it to @ehi->desc.
121 * If @ehi->desc is not empty, ", " is added in-between.
122 *
123 * LOCKING:
124 * spin_lock_irqsave(host lock)
125 */
126void ata_ehi_push_desc(struct ata_eh_info *ehi, const char *fmt, ...)
127{
128 va_list args;
129
130 if (ehi->desc_len)
131 __ata_ehi_push_desc(ehi, ", ");
132
133 va_start(args, fmt);
134 __ata_ehi_pushv_desc(ehi, fmt, args);
135 va_end(args);
136}
137
138/**
139 * ata_ehi_clear_desc - clean error description
140 * @ehi: target EHI
141 *
142 * Clear @ehi->desc.
143 *
144 * LOCKING:
145 * spin_lock_irqsave(host lock)
146 */
147void ata_ehi_clear_desc(struct ata_eh_info *ehi)
148{
149 ehi->desc[0] = '\0';
150 ehi->desc_len = 0;
151}
152
0c247c55
TH
153static void ata_ering_record(struct ata_ering *ering, int is_io,
154 unsigned int err_mask)
155{
156 struct ata_ering_entry *ent;
157
158 WARN_ON(!err_mask);
159
160 ering->cursor++;
161 ering->cursor %= ATA_ERING_SIZE;
162
163 ent = &ering->ring[ering->cursor];
164 ent->is_io = is_io;
165 ent->err_mask = err_mask;
166 ent->timestamp = get_jiffies_64();
167}
168
7d47e8d4 169static void ata_ering_clear(struct ata_ering *ering)
0c247c55 170{
7d47e8d4 171 memset(ering, 0, sizeof(*ering));
0c247c55
TH
172}
173
174static int ata_ering_map(struct ata_ering *ering,
175 int (*map_fn)(struct ata_ering_entry *, void *),
176 void *arg)
177{
178 int idx, rc = 0;
179 struct ata_ering_entry *ent;
180
181 idx = ering->cursor;
182 do {
183 ent = &ering->ring[idx];
184 if (!ent->err_mask)
185 break;
186 rc = map_fn(ent, arg);
187 if (rc)
188 break;
189 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
190 } while (idx != ering->cursor);
191
192 return rc;
193}
194
64f65ca6
TH
195static unsigned int ata_eh_dev_action(struct ata_device *dev)
196{
197 struct ata_eh_context *ehc = &dev->ap->eh_context;
198
199 return ehc->i.action | ehc->i.dev_action[dev->devno];
200}
201
af181c2d
TH
202static void ata_eh_clear_action(struct ata_device *dev,
203 struct ata_eh_info *ehi, unsigned int action)
204{
205 int i;
206
207 if (!dev) {
208 ehi->action &= ~action;
209 for (i = 0; i < ATA_MAX_DEVICES; i++)
210 ehi->dev_action[i] &= ~action;
211 } else {
212 /* doesn't make sense for port-wide EH actions */
213 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
214
215 /* break ehi->action into ehi->dev_action */
216 if (ehi->action & action) {
217 for (i = 0; i < ATA_MAX_DEVICES; i++)
218 ehi->dev_action[i] |= ehi->action & action;
219 ehi->action &= ~action;
220 }
221
222 /* turn off the specified per-dev action */
223 ehi->dev_action[dev->devno] &= ~action;
224 }
225}
226
ece1d636
TH
227/**
228 * ata_scsi_timed_out - SCSI layer time out callback
229 * @cmd: timed out SCSI command
230 *
231 * Handles SCSI layer timeout. We race with normal completion of
232 * the qc for @cmd. If the qc is already gone, we lose and let
233 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
234 * timed out and EH should be invoked. Prevent ata_qc_complete()
235 * from finishing it by setting EH_SCHEDULED and return
236 * EH_NOT_HANDLED.
237 *
ad9e2762
TH
238 * TODO: kill this function once old EH is gone.
239 *
ece1d636
TH
240 * LOCKING:
241 * Called from timer context
242 *
243 * RETURNS:
244 * EH_HANDLED or EH_NOT_HANDLED
245 */
246enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
247{
248 struct Scsi_Host *host = cmd->device->host;
35bb94b1 249 struct ata_port *ap = ata_shost_to_port(host);
ece1d636
TH
250 unsigned long flags;
251 struct ata_queued_cmd *qc;
ad9e2762 252 enum scsi_eh_timer_return ret;
ece1d636
TH
253
254 DPRINTK("ENTER\n");
255
ad9e2762
TH
256 if (ap->ops->error_handler) {
257 ret = EH_NOT_HANDLED;
258 goto out;
259 }
260
261 ret = EH_HANDLED;
ba6a1308 262 spin_lock_irqsave(ap->lock, flags);
ece1d636
TH
263 qc = ata_qc_from_tag(ap, ap->active_tag);
264 if (qc) {
265 WARN_ON(qc->scsicmd != cmd);
266 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
267 qc->err_mask |= AC_ERR_TIMEOUT;
268 ret = EH_NOT_HANDLED;
269 }
ba6a1308 270 spin_unlock_irqrestore(ap->lock, flags);
ece1d636 271
ad9e2762 272 out:
ece1d636
TH
273 DPRINTK("EXIT, ret=%d\n", ret);
274 return ret;
275}
276
277/**
278 * ata_scsi_error - SCSI layer error handler callback
279 * @host: SCSI host on which error occurred
280 *
281 * Handles SCSI-layer-thrown error events.
282 *
283 * LOCKING:
284 * Inherited from SCSI layer (none, can sleep)
285 *
286 * RETURNS:
287 * Zero.
288 */
381544bb 289void ata_scsi_error(struct Scsi_Host *host)
ece1d636 290{
35bb94b1 291 struct ata_port *ap = ata_shost_to_port(host);
ad9e2762
TH
292 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
293 unsigned long flags;
ece1d636
TH
294
295 DPRINTK("ENTER\n");
296
ad9e2762 297 /* synchronize with port task */
ece1d636
TH
298 ata_port_flush_task(ap);
299
cca3974e 300 /* synchronize with host lock and sort out timeouts */
ad9e2762
TH
301
302 /* For new EH, all qcs are finished in one of three ways -
303 * normal completion, error completion, and SCSI timeout.
304 * Both cmpletions can race against SCSI timeout. When normal
305 * completion wins, the qc never reaches EH. When error
306 * completion wins, the qc has ATA_QCFLAG_FAILED set.
307 *
308 * When SCSI timeout wins, things are a bit more complex.
309 * Normal or error completion can occur after the timeout but
310 * before this point. In such cases, both types of
311 * completions are honored. A scmd is determined to have
312 * timed out iff its associated qc is active and not failed.
313 */
314 if (ap->ops->error_handler) {
315 struct scsi_cmnd *scmd, *tmp;
316 int nr_timedout = 0;
317
e30349d2 318 spin_lock_irqsave(ap->lock, flags);
ad9e2762
TH
319
320 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
321 struct ata_queued_cmd *qc;
322
323 for (i = 0; i < ATA_MAX_QUEUE; i++) {
324 qc = __ata_qc_from_tag(ap, i);
325 if (qc->flags & ATA_QCFLAG_ACTIVE &&
326 qc->scsicmd == scmd)
327 break;
328 }
329
330 if (i < ATA_MAX_QUEUE) {
331 /* the scmd has an associated qc */
332 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
333 /* which hasn't failed yet, timeout */
334 qc->err_mask |= AC_ERR_TIMEOUT;
335 qc->flags |= ATA_QCFLAG_FAILED;
336 nr_timedout++;
337 }
338 } else {
339 /* Normal completion occurred after
340 * SCSI timeout but before this point.
341 * Successfully complete it.
342 */
343 scmd->retries = scmd->allowed;
344 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
345 }
346 }
347
348 /* If we have timed out qcs. They belong to EH from
349 * this point but the state of the controller is
350 * unknown. Freeze the port to make sure the IRQ
351 * handler doesn't diddle with those qcs. This must
352 * be done atomically w.r.t. setting QCFLAG_FAILED.
353 */
354 if (nr_timedout)
355 __ata_port_freeze(ap);
356
e30349d2 357 spin_unlock_irqrestore(ap->lock, flags);
ad9e2762 358 } else
e30349d2 359 spin_unlock_wait(ap->lock);
ad9e2762
TH
360
361 repeat:
362 /* invoke error handler */
363 if (ap->ops->error_handler) {
500530f6
TH
364 /* process port resume request */
365 ata_eh_handle_port_resume(ap);
366
f3e81b19 367 /* fetch & clear EH info */
e30349d2 368 spin_lock_irqsave(ap->lock, flags);
f3e81b19
TH
369
370 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
371 ap->eh_context.i = ap->eh_info;
372 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
373
b51e9e5d
TH
374 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
375 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
f3e81b19 376
e30349d2 377 spin_unlock_irqrestore(ap->lock, flags);
ad9e2762 378
500530f6
TH
379 /* invoke EH, skip if unloading or suspended */
380 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
720ba126
TH
381 ap->ops->error_handler(ap);
382 else
383 ata_eh_finish(ap);
ad9e2762 384
500530f6
TH
385 /* process port suspend request */
386 ata_eh_handle_port_suspend(ap);
387
ad9e2762
TH
388 /* Exception might have happend after ->error_handler
389 * recovered the port but before this point. Repeat
390 * EH in such case.
391 */
e30349d2 392 spin_lock_irqsave(ap->lock, flags);
ad9e2762 393
b51e9e5d 394 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
ad9e2762
TH
395 if (--repeat_cnt) {
396 ata_port_printk(ap, KERN_INFO,
397 "EH pending after completion, "
398 "repeating EH (cnt=%d)\n", repeat_cnt);
e30349d2 399 spin_unlock_irqrestore(ap->lock, flags);
ad9e2762
TH
400 goto repeat;
401 }
402 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
403 "tries, giving up\n", ATA_EH_MAX_REPEAT);
914616a3 404 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
ad9e2762
TH
405 }
406
f3e81b19
TH
407 /* this run is complete, make sure EH info is clear */
408 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
409
e30349d2 410 /* Clear host_eh_scheduled while holding ap->lock such
ad9e2762
TH
411 * that if exception occurs after this point but
412 * before EH completion, SCSI midlayer will
413 * re-initiate EH.
414 */
415 host->host_eh_scheduled = 0;
416
e30349d2 417 spin_unlock_irqrestore(ap->lock, flags);
ad9e2762
TH
418 } else {
419 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
420 ap->ops->eng_timeout(ap);
421 }
ece1d636 422
ad9e2762 423 /* finish or retry handled scmd's and clean up */
ece1d636
TH
424 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
425
426 scsi_eh_flush_done_q(&ap->eh_done_q);
427
ad9e2762 428 /* clean up */
e30349d2 429 spin_lock_irqsave(ap->lock, flags);
ad9e2762 430
1cdaf534 431 if (ap->pflags & ATA_PFLAG_LOADING)
b51e9e5d 432 ap->pflags &= ~ATA_PFLAG_LOADING;
1cdaf534 433 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
52bad64d 434 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
1cdaf534
TH
435
436 if (ap->pflags & ATA_PFLAG_RECOVERED)
437 ata_port_printk(ap, KERN_INFO, "EH complete\n");
580b2102 438
b51e9e5d 439 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
ad9e2762 440
c6cf9e99 441 /* tell wait_eh that we're done */
b51e9e5d 442 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
c6cf9e99
TH
443 wake_up_all(&ap->eh_wait_q);
444
e30349d2 445 spin_unlock_irqrestore(ap->lock, flags);
ad9e2762 446
ece1d636 447 DPRINTK("EXIT\n");
ece1d636
TH
448}
449
c6cf9e99
TH
450/**
451 * ata_port_wait_eh - Wait for the currently pending EH to complete
452 * @ap: Port to wait EH for
453 *
454 * Wait until the currently pending EH is complete.
455 *
456 * LOCKING:
457 * Kernel thread context (may sleep).
458 */
459void ata_port_wait_eh(struct ata_port *ap)
460{
461 unsigned long flags;
462 DEFINE_WAIT(wait);
463
464 retry:
ba6a1308 465 spin_lock_irqsave(ap->lock, flags);
c6cf9e99 466
b51e9e5d 467 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
c6cf9e99 468 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
ba6a1308 469 spin_unlock_irqrestore(ap->lock, flags);
c6cf9e99 470 schedule();
ba6a1308 471 spin_lock_irqsave(ap->lock, flags);
c6cf9e99 472 }
0a1b622e 473 finish_wait(&ap->eh_wait_q, &wait);
c6cf9e99 474
ba6a1308 475 spin_unlock_irqrestore(ap->lock, flags);
c6cf9e99
TH
476
477 /* make sure SCSI EH is complete */
cca3974e 478 if (scsi_host_in_recovery(ap->scsi_host)) {
c6cf9e99
TH
479 msleep(10);
480 goto retry;
481 }
482}
483
ece1d636
TH
484/**
485 * ata_qc_timeout - Handle timeout of queued command
486 * @qc: Command that timed out
487 *
488 * Some part of the kernel (currently, only the SCSI layer)
489 * has noticed that the active command on port @ap has not
490 * completed after a specified length of time. Handle this
491 * condition by disabling DMA (if necessary) and completing
492 * transactions, with error if necessary.
493 *
494 * This also handles the case of the "lost interrupt", where
495 * for some reason (possibly hardware bug, possibly driver bug)
496 * an interrupt was not delivered to the driver, even though the
497 * transaction completed successfully.
498 *
ad9e2762
TH
499 * TODO: kill this function once old EH is gone.
500 *
ece1d636
TH
501 * LOCKING:
502 * Inherited from SCSI layer (none, can sleep)
503 */
504static void ata_qc_timeout(struct ata_queued_cmd *qc)
505{
506 struct ata_port *ap = qc->ap;
ece1d636
TH
507 u8 host_stat = 0, drv_stat;
508 unsigned long flags;
509
510 DPRINTK("ENTER\n");
511
512 ap->hsm_task_state = HSM_ST_IDLE;
513
ba6a1308 514 spin_lock_irqsave(ap->lock, flags);
ece1d636
TH
515
516 switch (qc->tf.protocol) {
517
518 case ATA_PROT_DMA:
519 case ATA_PROT_ATAPI_DMA:
520 host_stat = ap->ops->bmdma_status(ap);
521
522 /* before we do anything else, clear DMA-Start bit */
523 ap->ops->bmdma_stop(qc);
524
525 /* fall through */
526
527 default:
528 ata_altstatus(ap);
529 drv_stat = ata_chk_status(ap);
530
531 /* ack bmdma irq events */
532 ap->ops->irq_clear(ap);
533
f15a1daf
TH
534 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
535 "stat 0x%x host_stat 0x%x\n",
536 qc->tf.command, drv_stat, host_stat);
ece1d636
TH
537
538 /* complete taskfile transaction */
c13b56a1 539 qc->err_mask |= AC_ERR_TIMEOUT;
ece1d636
TH
540 break;
541 }
542
ba6a1308 543 spin_unlock_irqrestore(ap->lock, flags);
ece1d636
TH
544
545 ata_eh_qc_complete(qc);
546
547 DPRINTK("EXIT\n");
548}
549
550/**
551 * ata_eng_timeout - Handle timeout of queued command
552 * @ap: Port on which timed-out command is active
553 *
554 * Some part of the kernel (currently, only the SCSI layer)
555 * has noticed that the active command on port @ap has not
556 * completed after a specified length of time. Handle this
557 * condition by disabling DMA (if necessary) and completing
558 * transactions, with error if necessary.
559 *
560 * This also handles the case of the "lost interrupt", where
561 * for some reason (possibly hardware bug, possibly driver bug)
562 * an interrupt was not delivered to the driver, even though the
563 * transaction completed successfully.
564 *
ad9e2762
TH
565 * TODO: kill this function once old EH is gone.
566 *
ece1d636
TH
567 * LOCKING:
568 * Inherited from SCSI layer (none, can sleep)
569 */
570void ata_eng_timeout(struct ata_port *ap)
571{
572 DPRINTK("ENTER\n");
573
574 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
575
576 DPRINTK("EXIT\n");
577}
578
f686bcb8
TH
579/**
580 * ata_qc_schedule_eh - schedule qc for error handling
581 * @qc: command to schedule error handling for
582 *
583 * Schedule error handling for @qc. EH will kick in as soon as
584 * other commands are drained.
585 *
586 * LOCKING:
cca3974e 587 * spin_lock_irqsave(host lock)
f686bcb8
TH
588 */
589void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
590{
591 struct ata_port *ap = qc->ap;
592
593 WARN_ON(!ap->ops->error_handler);
594
595 qc->flags |= ATA_QCFLAG_FAILED;
b51e9e5d 596 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
f686bcb8
TH
597
598 /* The following will fail if timeout has already expired.
599 * ata_scsi_error() takes care of such scmds on EH entry.
600 * Note that ATA_QCFLAG_FAILED is unconditionally set after
601 * this function completes.
602 */
603 scsi_req_abort_cmd(qc->scsicmd);
604}
605
7b70fc03
TH
606/**
607 * ata_port_schedule_eh - schedule error handling without a qc
608 * @ap: ATA port to schedule EH for
609 *
610 * Schedule error handling for @ap. EH will kick in as soon as
611 * all commands are drained.
612 *
613 * LOCKING:
cca3974e 614 * spin_lock_irqsave(host lock)
7b70fc03
TH
615 */
616void ata_port_schedule_eh(struct ata_port *ap)
617{
618 WARN_ON(!ap->ops->error_handler);
619
f4d6d004
TH
620 if (ap->pflags & ATA_PFLAG_INITIALIZING)
621 return;
622
b51e9e5d 623 ap->pflags |= ATA_PFLAG_EH_PENDING;
cca3974e 624 scsi_schedule_eh(ap->scsi_host);
7b70fc03
TH
625
626 DPRINTK("port EH scheduled\n");
627}
628
629/**
630 * ata_port_abort - abort all qc's on the port
631 * @ap: ATA port to abort qc's for
632 *
633 * Abort all active qc's of @ap and schedule EH.
634 *
635 * LOCKING:
cca3974e 636 * spin_lock_irqsave(host lock)
7b70fc03
TH
637 *
638 * RETURNS:
639 * Number of aborted qc's.
640 */
641int ata_port_abort(struct ata_port *ap)
642{
643 int tag, nr_aborted = 0;
644
645 WARN_ON(!ap->ops->error_handler);
646
647 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
648 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
649
650 if (qc) {
651 qc->flags |= ATA_QCFLAG_FAILED;
652 ata_qc_complete(qc);
653 nr_aborted++;
654 }
655 }
656
657 if (!nr_aborted)
658 ata_port_schedule_eh(ap);
659
660 return nr_aborted;
661}
662
e3180499
TH
663/**
664 * __ata_port_freeze - freeze port
665 * @ap: ATA port to freeze
666 *
667 * This function is called when HSM violation or some other
668 * condition disrupts normal operation of the port. Frozen port
669 * is not allowed to perform any operation until the port is
670 * thawed, which usually follows a successful reset.
671 *
672 * ap->ops->freeze() callback can be used for freezing the port
673 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
674 * port cannot be frozen hardware-wise, the interrupt handler
675 * must ack and clear interrupts unconditionally while the port
676 * is frozen.
677 *
678 * LOCKING:
cca3974e 679 * spin_lock_irqsave(host lock)
e3180499
TH
680 */
681static void __ata_port_freeze(struct ata_port *ap)
682{
683 WARN_ON(!ap->ops->error_handler);
684
685 if (ap->ops->freeze)
686 ap->ops->freeze(ap);
687
b51e9e5d 688 ap->pflags |= ATA_PFLAG_FROZEN;
e3180499 689
44877b4e 690 DPRINTK("ata%u port frozen\n", ap->print_id);
e3180499
TH
691}
692
693/**
694 * ata_port_freeze - abort & freeze port
695 * @ap: ATA port to freeze
696 *
697 * Abort and freeze @ap.
698 *
699 * LOCKING:
cca3974e 700 * spin_lock_irqsave(host lock)
e3180499
TH
701 *
702 * RETURNS:
703 * Number of aborted commands.
704 */
705int ata_port_freeze(struct ata_port *ap)
706{
707 int nr_aborted;
708
709 WARN_ON(!ap->ops->error_handler);
710
711 nr_aborted = ata_port_abort(ap);
712 __ata_port_freeze(ap);
713
714 return nr_aborted;
715}
716
717/**
718 * ata_eh_freeze_port - EH helper to freeze port
719 * @ap: ATA port to freeze
720 *
721 * Freeze @ap.
722 *
723 * LOCKING:
724 * None.
725 */
726void ata_eh_freeze_port(struct ata_port *ap)
727{
728 unsigned long flags;
729
730 if (!ap->ops->error_handler)
731 return;
732
ba6a1308 733 spin_lock_irqsave(ap->lock, flags);
e3180499 734 __ata_port_freeze(ap);
ba6a1308 735 spin_unlock_irqrestore(ap->lock, flags);
e3180499
TH
736}
737
738/**
739 * ata_port_thaw_port - EH helper to thaw port
740 * @ap: ATA port to thaw
741 *
742 * Thaw frozen port @ap.
743 *
744 * LOCKING:
745 * None.
746 */
747void ata_eh_thaw_port(struct ata_port *ap)
748{
749 unsigned long flags;
750
751 if (!ap->ops->error_handler)
752 return;
753
ba6a1308 754 spin_lock_irqsave(ap->lock, flags);
e3180499 755
b51e9e5d 756 ap->pflags &= ~ATA_PFLAG_FROZEN;
e3180499
TH
757
758 if (ap->ops->thaw)
759 ap->ops->thaw(ap);
760
ba6a1308 761 spin_unlock_irqrestore(ap->lock, flags);
e3180499 762
44877b4e 763 DPRINTK("ata%u port thawed\n", ap->print_id);
e3180499
TH
764}
765
ece1d636
TH
766static void ata_eh_scsidone(struct scsi_cmnd *scmd)
767{
768 /* nada */
769}
770
771static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
772{
773 struct ata_port *ap = qc->ap;
774 struct scsi_cmnd *scmd = qc->scsicmd;
775 unsigned long flags;
776
ba6a1308 777 spin_lock_irqsave(ap->lock, flags);
ece1d636
TH
778 qc->scsidone = ata_eh_scsidone;
779 __ata_qc_complete(qc);
780 WARN_ON(ata_tag_valid(qc->tag));
ba6a1308 781 spin_unlock_irqrestore(ap->lock, flags);
ece1d636
TH
782
783 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
784}
785
786/**
787 * ata_eh_qc_complete - Complete an active ATA command from EH
788 * @qc: Command to complete
789 *
790 * Indicate to the mid and upper layers that an ATA command has
791 * completed. To be used from EH.
792 */
793void ata_eh_qc_complete(struct ata_queued_cmd *qc)
794{
795 struct scsi_cmnd *scmd = qc->scsicmd;
796 scmd->retries = scmd->allowed;
797 __ata_eh_qc_complete(qc);
798}
799
800/**
801 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
802 * @qc: Command to retry
803 *
804 * Indicate to the mid and upper layers that an ATA command
805 * should be retried. To be used from EH.
806 *
807 * SCSI midlayer limits the number of retries to scmd->allowed.
808 * scmd->retries is decremented for commands which get retried
809 * due to unrelated failures (qc->err_mask is zero).
810 */
811void ata_eh_qc_retry(struct ata_queued_cmd *qc)
812{
813 struct scsi_cmnd *scmd = qc->scsicmd;
814 if (!qc->err_mask && scmd->retries)
815 scmd->retries--;
816 __ata_eh_qc_complete(qc);
817}
022bdb07 818
0ea035a3
TH
819/**
820 * ata_eh_detach_dev - detach ATA device
821 * @dev: ATA device to detach
822 *
823 * Detach @dev.
824 *
825 * LOCKING:
826 * None.
827 */
828static void ata_eh_detach_dev(struct ata_device *dev)
829{
830 struct ata_port *ap = dev->ap;
831 unsigned long flags;
832
833 ata_dev_disable(dev);
834
ba6a1308 835 spin_lock_irqsave(ap->lock, flags);
0ea035a3
TH
836
837 dev->flags &= ~ATA_DFLAG_DETACH;
838
839 if (ata_scsi_offline_dev(dev)) {
840 dev->flags |= ATA_DFLAG_DETACHED;
b51e9e5d 841 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
0ea035a3
TH
842 }
843
beb07c1a
TH
844 /* clear per-dev EH actions */
845 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
846 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
847
ba6a1308 848 spin_unlock_irqrestore(ap->lock, flags);
0ea035a3
TH
849}
850
022bdb07
TH
851/**
852 * ata_eh_about_to_do - about to perform eh_action
853 * @ap: target ATA port
47005f25 854 * @dev: target ATA dev for per-dev action (can be NULL)
022bdb07
TH
855 * @action: action about to be performed
856 *
857 * Called just before performing EH actions to clear related bits
858 * in @ap->eh_info such that eh actions are not unnecessarily
859 * repeated.
860 *
861 * LOCKING:
862 * None.
863 */
47005f25
TH
864static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
865 unsigned int action)
022bdb07
TH
866{
867 unsigned long flags;
13abf50d
TH
868 struct ata_eh_info *ehi = &ap->eh_info;
869 struct ata_eh_context *ehc = &ap->eh_context;
022bdb07 870
ba6a1308 871 spin_lock_irqsave(ap->lock, flags);
1cdaf534 872
13abf50d
TH
873 /* Reset is represented by combination of actions and EHI
874 * flags. Suck in all related bits before clearing eh_info to
875 * avoid losing requested action.
876 */
877 if (action & ATA_EH_RESET_MASK) {
878 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
879 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
880
881 /* make sure all reset actions are cleared & clear EHI flags */
882 action |= ATA_EH_RESET_MASK;
883 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
884 }
885
886 ata_eh_clear_action(dev, ehi, action);
1cdaf534 887
13abf50d 888 if (!(ehc->i.flags & ATA_EHI_QUIET))
1cdaf534
TH
889 ap->pflags |= ATA_PFLAG_RECOVERED;
890
ba6a1308 891 spin_unlock_irqrestore(ap->lock, flags);
022bdb07
TH
892}
893
47005f25
TH
894/**
895 * ata_eh_done - EH action complete
896 * @ap: target ATA port
897 * @dev: target ATA dev for per-dev action (can be NULL)
898 * @action: action just completed
899 *
900 * Called right after performing EH actions to clear related bits
901 * in @ap->eh_context.
902 *
903 * LOCKING:
904 * None.
905 */
906static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
907 unsigned int action)
908{
13abf50d
TH
909 /* if reset is complete, clear all reset actions & reset modifier */
910 if (action & ATA_EH_RESET_MASK) {
911 action |= ATA_EH_RESET_MASK;
912 ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
913 }
914
47005f25
TH
915 ata_eh_clear_action(dev, &ap->eh_context.i, action);
916}
917
022bdb07
TH
918/**
919 * ata_err_string - convert err_mask to descriptive string
920 * @err_mask: error mask to convert to string
921 *
922 * Convert @err_mask to descriptive string. Errors are
923 * prioritized according to severity and only the most severe
924 * error is reported.
925 *
926 * LOCKING:
927 * None.
928 *
929 * RETURNS:
930 * Descriptive string for @err_mask
931 */
932static const char * ata_err_string(unsigned int err_mask)
933{
934 if (err_mask & AC_ERR_HOST_BUS)
935 return "host bus error";
936 if (err_mask & AC_ERR_ATA_BUS)
937 return "ATA bus error";
938 if (err_mask & AC_ERR_TIMEOUT)
939 return "timeout";
940 if (err_mask & AC_ERR_HSM)
941 return "HSM violation";
942 if (err_mask & AC_ERR_SYSTEM)
943 return "internal error";
944 if (err_mask & AC_ERR_MEDIA)
945 return "media error";
946 if (err_mask & AC_ERR_INVALID)
947 return "invalid argument";
948 if (err_mask & AC_ERR_DEV)
949 return "device error";
950 return "unknown error";
951}
952
e8ee8451
TH
953/**
954 * ata_read_log_page - read a specific log page
955 * @dev: target device
956 * @page: page to read
957 * @buf: buffer to store read page
958 * @sectors: number of sectors to read
959 *
960 * Read log page using READ_LOG_EXT command.
961 *
962 * LOCKING:
963 * Kernel thread context (may sleep).
964 *
965 * RETURNS:
966 * 0 on success, AC_ERR_* mask otherwise.
967 */
968static unsigned int ata_read_log_page(struct ata_device *dev,
969 u8 page, void *buf, unsigned int sectors)
970{
971 struct ata_taskfile tf;
972 unsigned int err_mask;
973
974 DPRINTK("read log page - page %d\n", page);
975
976 ata_tf_init(dev, &tf);
977 tf.command = ATA_CMD_READ_LOG_EXT;
978 tf.lbal = page;
979 tf.nsect = sectors;
980 tf.hob_nsect = sectors >> 8;
981 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
982 tf.protocol = ATA_PROT_PIO;
983
984 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
985 buf, sectors * ATA_SECT_SIZE);
986
987 DPRINTK("EXIT, err_mask=%x\n", err_mask);
988 return err_mask;
989}
990
991/**
992 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
993 * @dev: Device to read log page 10h from
994 * @tag: Resulting tag of the failed command
995 * @tf: Resulting taskfile registers of the failed command
996 *
997 * Read log page 10h to obtain NCQ error details and clear error
998 * condition.
999 *
1000 * LOCKING:
1001 * Kernel thread context (may sleep).
1002 *
1003 * RETURNS:
1004 * 0 on success, -errno otherwise.
1005 */
1006static int ata_eh_read_log_10h(struct ata_device *dev,
1007 int *tag, struct ata_taskfile *tf)
1008{
1009 u8 *buf = dev->ap->sector_buf;
1010 unsigned int err_mask;
1011 u8 csum;
1012 int i;
1013
1014 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
1015 if (err_mask)
1016 return -EIO;
1017
1018 csum = 0;
1019 for (i = 0; i < ATA_SECT_SIZE; i++)
1020 csum += buf[i];
1021 if (csum)
1022 ata_dev_printk(dev, KERN_WARNING,
1023 "invalid checksum 0x%x on log page 10h\n", csum);
1024
1025 if (buf[0] & 0x80)
1026 return -ENOENT;
1027
1028 *tag = buf[0] & 0x1f;
1029
1030 tf->command = buf[2];
1031 tf->feature = buf[3];
1032 tf->lbal = buf[4];
1033 tf->lbam = buf[5];
1034 tf->lbah = buf[6];
1035 tf->device = buf[7];
1036 tf->hob_lbal = buf[8];
1037 tf->hob_lbam = buf[9];
1038 tf->hob_lbah = buf[10];
1039 tf->nsect = buf[12];
1040 tf->hob_nsect = buf[13];
1041
1042 return 0;
1043}
1044
022bdb07
TH
1045/**
1046 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
1047 * @dev: device to perform REQUEST_SENSE to
1048 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
1049 *
1050 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
1051 * SENSE. This function is EH helper.
1052 *
1053 * LOCKING:
1054 * Kernel thread context (may sleep).
1055 *
1056 * RETURNS:
1057 * 0 on success, AC_ERR_* mask on failure
1058 */
56287768 1059static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
022bdb07 1060{
56287768
AL
1061 struct ata_device *dev = qc->dev;
1062 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
022bdb07
TH
1063 struct ata_port *ap = dev->ap;
1064 struct ata_taskfile tf;
1065 u8 cdb[ATAPI_CDB_LEN];
1066
1067 DPRINTK("ATAPI request sense\n");
1068
022bdb07
TH
1069 /* FIXME: is this needed? */
1070 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1071
56287768
AL
1072 /* initialize sense_buf with the error register,
1073 * for the case where they are -not- overwritten
1074 */
022bdb07 1075 sense_buf[0] = 0x70;
56287768
AL
1076 sense_buf[2] = qc->result_tf.feature >> 4;
1077
a617c09f 1078 /* some devices time out if garbage left in tf */
56287768 1079 ata_tf_init(dev, &tf);
022bdb07
TH
1080
1081 memset(cdb, 0, ATAPI_CDB_LEN);
1082 cdb[0] = REQUEST_SENSE;
1083 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1084
1085 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1086 tf.command = ATA_CMD_PACKET;
1087
1088 /* is it pointless to prefer PIO for "safety reasons"? */
1089 if (ap->flags & ATA_FLAG_PIO_DMA) {
1090 tf.protocol = ATA_PROT_ATAPI_DMA;
1091 tf.feature |= ATAPI_PKT_DMA;
1092 } else {
1093 tf.protocol = ATA_PROT_ATAPI;
1094 tf.lbam = (8 * 1024) & 0xff;
1095 tf.lbah = (8 * 1024) >> 8;
1096 }
1097
1098 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1099 sense_buf, SCSI_SENSE_BUFFERSIZE);
1100}
1101
1102/**
1103 * ata_eh_analyze_serror - analyze SError for a failed port
1104 * @ap: ATA port to analyze SError for
1105 *
1106 * Analyze SError if available and further determine cause of
1107 * failure.
1108 *
1109 * LOCKING:
1110 * None.
1111 */
1112static void ata_eh_analyze_serror(struct ata_port *ap)
1113{
1114 struct ata_eh_context *ehc = &ap->eh_context;
1115 u32 serror = ehc->i.serror;
1116 unsigned int err_mask = 0, action = 0;
1117
1118 if (serror & SERR_PERSISTENT) {
1119 err_mask |= AC_ERR_ATA_BUS;
1120 action |= ATA_EH_HARDRESET;
1121 }
1122 if (serror &
1123 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1124 err_mask |= AC_ERR_ATA_BUS;
1125 action |= ATA_EH_SOFTRESET;
1126 }
1127 if (serror & SERR_PROTOCOL) {
1128 err_mask |= AC_ERR_HSM;
1129 action |= ATA_EH_SOFTRESET;
1130 }
1131 if (serror & SERR_INTERNAL) {
1132 err_mask |= AC_ERR_SYSTEM;
771b8dad 1133 action |= ATA_EH_HARDRESET;
022bdb07 1134 }
084fe639
TH
1135 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1136 ata_ehi_hotplugged(&ehc->i);
022bdb07
TH
1137
1138 ehc->i.err_mask |= err_mask;
1139 ehc->i.action |= action;
1140}
1141
e8ee8451
TH
1142/**
1143 * ata_eh_analyze_ncq_error - analyze NCQ error
1144 * @ap: ATA port to analyze NCQ error for
1145 *
1146 * Read log page 10h, determine the offending qc and acquire
1147 * error status TF. For NCQ device errors, all LLDDs have to do
1148 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1149 * care of the rest.
1150 *
1151 * LOCKING:
1152 * Kernel thread context (may sleep).
1153 */
1154static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1155{
1156 struct ata_eh_context *ehc = &ap->eh_context;
1157 struct ata_device *dev = ap->device;
1158 struct ata_queued_cmd *qc;
1159 struct ata_taskfile tf;
1160 int tag, rc;
1161
1162 /* if frozen, we can't do much */
b51e9e5d 1163 if (ap->pflags & ATA_PFLAG_FROZEN)
e8ee8451
TH
1164 return;
1165
1166 /* is it NCQ device error? */
1167 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1168 return;
1169
1170 /* has LLDD analyzed already? */
1171 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1172 qc = __ata_qc_from_tag(ap, tag);
1173
1174 if (!(qc->flags & ATA_QCFLAG_FAILED))
1175 continue;
1176
1177 if (qc->err_mask)
1178 return;
1179 }
1180
1181 /* okay, this error is ours */
1182 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1183 if (rc) {
1184 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1185 "(errno=%d)\n", rc);
1186 return;
1187 }
1188
1189 if (!(ap->sactive & (1 << tag))) {
1190 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1191 "inactive tag %d\n", tag);
1192 return;
1193 }
1194
1195 /* we've got the perpetrator, condemn it */
1196 qc = __ata_qc_from_tag(ap, tag);
1197 memcpy(&qc->result_tf, &tf, sizeof(tf));
1198 qc->err_mask |= AC_ERR_DEV;
1199 ehc->i.err_mask &= ~AC_ERR_DEV;
1200}
1201
022bdb07
TH
1202/**
1203 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1204 * @qc: qc to analyze
1205 * @tf: Taskfile registers to analyze
1206 *
1207 * Analyze taskfile of @qc and further determine cause of
1208 * failure. This function also requests ATAPI sense data if
1209 * avaliable.
1210 *
1211 * LOCKING:
1212 * Kernel thread context (may sleep).
1213 *
1214 * RETURNS:
1215 * Determined recovery action
1216 */
1217static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1218 const struct ata_taskfile *tf)
1219{
1220 unsigned int tmp, action = 0;
1221 u8 stat = tf->command, err = tf->feature;
1222
1223 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1224 qc->err_mask |= AC_ERR_HSM;
1225 return ATA_EH_SOFTRESET;
1226 }
1227
a51d644a
TH
1228 if (stat & (ATA_ERR | ATA_DF))
1229 qc->err_mask |= AC_ERR_DEV;
1230 else
022bdb07
TH
1231 return 0;
1232
1233 switch (qc->dev->class) {
1234 case ATA_DEV_ATA:
1235 if (err & ATA_ICRC)
1236 qc->err_mask |= AC_ERR_ATA_BUS;
1237 if (err & ATA_UNC)
1238 qc->err_mask |= AC_ERR_MEDIA;
1239 if (err & ATA_IDNF)
1240 qc->err_mask |= AC_ERR_INVALID;
1241 break;
1242
1243 case ATA_DEV_ATAPI:
a569a30d 1244 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
56287768 1245 tmp = atapi_eh_request_sense(qc);
a569a30d
TH
1246 if (!tmp) {
1247 /* ATA_QCFLAG_SENSE_VALID is used to
1248 * tell atapi_qc_complete() that sense
1249 * data is already valid.
1250 *
1251 * TODO: interpret sense data and set
1252 * appropriate err_mask.
1253 */
1254 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1255 } else
1256 qc->err_mask |= tmp;
1257 }
022bdb07
TH
1258 }
1259
1260 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1261 action |= ATA_EH_SOFTRESET;
1262
1263 return action;
1264}
1265
7d47e8d4 1266static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
022bdb07 1267{
7d47e8d4 1268 if (err_mask & AC_ERR_ATA_BUS)
022bdb07
TH
1269 return 1;
1270
7d47e8d4
TH
1271 if (err_mask & AC_ERR_TIMEOUT)
1272 return 2;
1273
1274 if (is_io) {
1275 if (err_mask & AC_ERR_HSM)
022bdb07 1276 return 2;
7d47e8d4
TH
1277 if ((err_mask &
1278 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1279 return 3;
022bdb07
TH
1280 }
1281
1282 return 0;
1283}
1284
7d47e8d4 1285struct speed_down_verdict_arg {
022bdb07 1286 u64 since;
7d47e8d4 1287 int nr_errors[4];
022bdb07
TH
1288};
1289
7d47e8d4 1290static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
022bdb07 1291{
7d47e8d4
TH
1292 struct speed_down_verdict_arg *arg = void_arg;
1293 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
022bdb07
TH
1294
1295 if (ent->timestamp < arg->since)
1296 return -1;
1297
7d47e8d4 1298 arg->nr_errors[cat]++;
022bdb07
TH
1299 return 0;
1300}
1301
1302/**
7d47e8d4 1303 * ata_eh_speed_down_verdict - Determine speed down verdict
022bdb07
TH
1304 * @dev: Device of interest
1305 *
1306 * This function examines error ring of @dev and determines
7d47e8d4
TH
1307 * whether NCQ needs to be turned off, transfer speed should be
1308 * stepped down, or falling back to PIO is necessary.
022bdb07 1309 *
7d47e8d4 1310 * Cat-1 is ATA_BUS error for any command.
022bdb07 1311 *
7d47e8d4
TH
1312 * Cat-2 is TIMEOUT for any command or HSM violation for known
1313 * supported commands.
1314 *
1315 * Cat-3 is is unclassified DEV error for known supported
022bdb07
TH
1316 * command.
1317 *
7d47e8d4
TH
1318 * NCQ needs to be turned off if there have been more than 3
1319 * Cat-2 + Cat-3 errors during last 10 minutes.
1320 *
1321 * Speed down is necessary if there have been more than 3 Cat-1 +
1322 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1323 *
1324 * Falling back to PIO mode is necessary if there have been more
1325 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1326 *
022bdb07
TH
1327 * LOCKING:
1328 * Inherited from caller.
1329 *
1330 * RETURNS:
7d47e8d4 1331 * OR of ATA_EH_SPDN_* flags.
022bdb07 1332 */
7d47e8d4 1333static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
022bdb07 1334{
7d47e8d4
TH
1335 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1336 u64 j64 = get_jiffies_64();
1337 struct speed_down_verdict_arg arg;
1338 unsigned int verdict = 0;
022bdb07 1339
7d47e8d4
TH
1340 /* scan past 10 mins of error history */
1341 memset(&arg, 0, sizeof(arg));
1342 arg.since = j64 - min(j64, j10mins);
1343 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
022bdb07 1344
7d47e8d4
TH
1345 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1346 verdict |= ATA_EH_SPDN_NCQ_OFF;
1347 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1348 verdict |= ATA_EH_SPDN_SPEED_DOWN;
022bdb07 1349
7d47e8d4 1350 /* scan past 3 mins of error history */
022bdb07 1351 memset(&arg, 0, sizeof(arg));
7d47e8d4
TH
1352 arg.since = j64 - min(j64, j5mins);
1353 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
022bdb07 1354
7d47e8d4
TH
1355 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1356 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
022bdb07 1357
7d47e8d4 1358 return verdict;
022bdb07
TH
1359}
1360
1361/**
1362 * ata_eh_speed_down - record error and speed down if necessary
1363 * @dev: Failed device
1364 * @is_io: Did the device fail during normal IO?
1365 * @err_mask: err_mask of the error
1366 *
1367 * Record error and examine error history to determine whether
1368 * adjusting transmission speed is necessary. It also sets
1369 * transmission limits appropriately if such adjustment is
1370 * necessary.
1371 *
1372 * LOCKING:
1373 * Kernel thread context (may sleep).
1374 *
1375 * RETURNS:
7d47e8d4 1376 * Determined recovery action.
022bdb07 1377 */
7d47e8d4
TH
1378static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1379 unsigned int err_mask)
022bdb07 1380{
7d47e8d4
TH
1381 unsigned int verdict;
1382 unsigned int action = 0;
1383
1384 /* don't bother if Cat-0 error */
1385 if (ata_eh_categorize_error(is_io, err_mask) == 0)
022bdb07
TH
1386 return 0;
1387
1388 /* record error and determine whether speed down is necessary */
1389 ata_ering_record(&dev->ering, is_io, err_mask);
7d47e8d4 1390 verdict = ata_eh_speed_down_verdict(dev);
022bdb07 1391
7d47e8d4
TH
1392 /* turn off NCQ? */
1393 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1394 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1395 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1396 dev->flags |= ATA_DFLAG_NCQ_OFF;
1397 ata_dev_printk(dev, KERN_WARNING,
1398 "NCQ disabled due to excessive errors\n");
1399 goto done;
1400 }
022bdb07 1401
7d47e8d4
TH
1402 /* speed down? */
1403 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1404 /* speed down SATA link speed if possible */
1405 if (sata_down_spd_limit(dev->ap) == 0) {
1406 action |= ATA_EH_HARDRESET;
1407 goto done;
1408 }
022bdb07 1409
7d47e8d4
TH
1410 /* lower transfer mode */
1411 if (dev->spdn_cnt < 2) {
1412 static const int dma_dnxfer_sel[] =
1413 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1414 static const int pio_dnxfer_sel[] =
1415 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1416 int sel;
1417
1418 if (dev->xfer_shift != ATA_SHIFT_PIO)
1419 sel = dma_dnxfer_sel[dev->spdn_cnt];
1420 else
1421 sel = pio_dnxfer_sel[dev->spdn_cnt];
1422
1423 dev->spdn_cnt++;
1424
1425 if (ata_down_xfermask_limit(dev, sel) == 0) {
1426 action |= ATA_EH_SOFTRESET;
1427 goto done;
1428 }
1429 }
1430 }
1431
1432 /* Fall back to PIO? Slowing down to PIO is meaningless for
1433 * SATA. Consider it only for PATA.
1434 */
1435 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1436 (dev->ap->cbl != ATA_CBL_SATA) &&
1437 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1438 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1439 dev->spdn_cnt = 0;
1440 action |= ATA_EH_SOFTRESET;
1441 goto done;
1442 }
1443 }
022bdb07 1444
022bdb07 1445 return 0;
7d47e8d4
TH
1446 done:
1447 /* device has been slowed down, blow error history */
1448 ata_ering_clear(&dev->ering);
1449 return action;
022bdb07
TH
1450}
1451
1452/**
1453 * ata_eh_autopsy - analyze error and determine recovery action
1454 * @ap: ATA port to perform autopsy on
1455 *
1456 * Analyze why @ap failed and determine which recovery action is
1457 * needed. This function also sets more detailed AC_ERR_* values
1458 * and fills sense data for ATAPI CHECK SENSE.
1459 *
1460 * LOCKING:
1461 * Kernel thread context (may sleep).
1462 */
1463static void ata_eh_autopsy(struct ata_port *ap)
1464{
1465 struct ata_eh_context *ehc = &ap->eh_context;
022bdb07
TH
1466 unsigned int all_err_mask = 0;
1467 int tag, is_io = 0;
1468 u32 serror;
1469 int rc;
1470
1471 DPRINTK("ENTER\n");
1472
1cdaf534
TH
1473 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1474 return;
1475
022bdb07
TH
1476 /* obtain and analyze SError */
1477 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1478 if (rc == 0) {
1479 ehc->i.serror |= serror;
1480 ata_eh_analyze_serror(ap);
1481 } else if (rc != -EOPNOTSUPP)
4528e4da 1482 ehc->i.action |= ATA_EH_HARDRESET;
022bdb07 1483
e8ee8451
TH
1484 /* analyze NCQ failure */
1485 ata_eh_analyze_ncq_error(ap);
1486
022bdb07
TH
1487 /* any real error trumps AC_ERR_OTHER */
1488 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1489 ehc->i.err_mask &= ~AC_ERR_OTHER;
1490
1491 all_err_mask |= ehc->i.err_mask;
1492
1493 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1494 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1495
1496 if (!(qc->flags & ATA_QCFLAG_FAILED))
1497 continue;
1498
1499 /* inherit upper level err_mask */
1500 qc->err_mask |= ehc->i.err_mask;
1501
022bdb07 1502 /* analyze TF */
4528e4da 1503 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
022bdb07
TH
1504
1505 /* DEV errors are probably spurious in case of ATA_BUS error */
1506 if (qc->err_mask & AC_ERR_ATA_BUS)
1507 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1508 AC_ERR_INVALID);
1509
1510 /* any real error trumps unknown error */
1511 if (qc->err_mask & ~AC_ERR_OTHER)
1512 qc->err_mask &= ~AC_ERR_OTHER;
1513
1514 /* SENSE_VALID trumps dev/unknown error and revalidation */
1515 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1516 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
4528e4da 1517 ehc->i.action &= ~ATA_EH_REVALIDATE;
022bdb07
TH
1518 }
1519
1520 /* accumulate error info */
4528e4da 1521 ehc->i.dev = qc->dev;
022bdb07
TH
1522 all_err_mask |= qc->err_mask;
1523 if (qc->flags & ATA_QCFLAG_IO)
1524 is_io = 1;
1525 }
1526
a20f33ff 1527 /* enforce default EH actions */
b51e9e5d 1528 if (ap->pflags & ATA_PFLAG_FROZEN ||
a20f33ff 1529 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
4528e4da 1530 ehc->i.action |= ATA_EH_SOFTRESET;
a20f33ff 1531 else if (all_err_mask)
4528e4da 1532 ehc->i.action |= ATA_EH_REVALIDATE;
022bdb07 1533
47005f25 1534 /* if we have offending qcs and the associated failed device */
4528e4da 1535 if (ehc->i.dev) {
47005f25 1536 /* speed down */
4528e4da
TH
1537 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1538 all_err_mask);
47005f25
TH
1539
1540 /* perform per-dev EH action only on the offending device */
4528e4da
TH
1541 ehc->i.dev_action[ehc->i.dev->devno] |=
1542 ehc->i.action & ATA_EH_PERDEV_MASK;
1543 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
47005f25
TH
1544 }
1545
022bdb07
TH
1546 DPRINTK("EXIT\n");
1547}
1548
1549/**
1550 * ata_eh_report - report error handling to user
1551 * @ap: ATA port EH is going on
1552 *
1553 * Report EH to user.
1554 *
1555 * LOCKING:
1556 * None.
1557 */
1558static void ata_eh_report(struct ata_port *ap)
1559{
1560 struct ata_eh_context *ehc = &ap->eh_context;
1561 const char *frozen, *desc;
1562 int tag, nr_failed = 0;
1563
1564 desc = NULL;
1565 if (ehc->i.desc[0] != '\0')
1566 desc = ehc->i.desc;
1567
1568 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1569 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1570
1571 if (!(qc->flags & ATA_QCFLAG_FAILED))
1572 continue;
1573 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1574 continue;
1575
1576 nr_failed++;
1577 }
1578
1579 if (!nr_failed && !ehc->i.err_mask)
1580 return;
1581
1582 frozen = "";
b51e9e5d 1583 if (ap->pflags & ATA_PFLAG_FROZEN)
022bdb07
TH
1584 frozen = " frozen";
1585
1586 if (ehc->i.dev) {
e8ee8451
TH
1587 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1588 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1589 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1590 ehc->i.action, frozen);
022bdb07 1591 if (desc)
b64bbc39 1592 ata_dev_printk(ehc->i.dev, KERN_ERR, "%s\n", desc);
022bdb07 1593 } else {
e8ee8451
TH
1594 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1595 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1596 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1597 ehc->i.action, frozen);
022bdb07 1598 if (desc)
b64bbc39 1599 ata_port_printk(ap, KERN_ERR, "%s\n", desc);
022bdb07
TH
1600 }
1601
1602 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
8a937581
TH
1603 static const char *dma_str[] = {
1604 [DMA_BIDIRECTIONAL] = "bidi",
1605 [DMA_TO_DEVICE] = "out",
1606 [DMA_FROM_DEVICE] = "in",
1607 [DMA_NONE] = "",
1608 };
022bdb07 1609 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
8a937581 1610 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
022bdb07
TH
1611
1612 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1613 continue;
1614
8a937581
TH
1615 ata_dev_printk(qc->dev, KERN_ERR,
1616 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
664e8503 1617 "tag %d cdb 0x%x data %u %s\n "
8a937581
TH
1618 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1619 "Emask 0x%x (%s)\n",
1620 cmd->command, cmd->feature, cmd->nsect,
1621 cmd->lbal, cmd->lbam, cmd->lbah,
1622 cmd->hob_feature, cmd->hob_nsect,
1623 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
726f0785 1624 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
664e8503 1625 dma_str[qc->dma_dir],
8a937581
TH
1626 res->command, res->feature, res->nsect,
1627 res->lbal, res->lbam, res->lbah,
1628 res->hob_feature, res->hob_nsect,
1629 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1630 res->device, qc->err_mask, ata_err_string(qc->err_mask));
022bdb07
TH
1631 }
1632}
1633
d87fa38e 1634static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
d4b2bab4 1635 unsigned int *classes, unsigned long deadline)
d87fa38e
TH
1636{
1637 int i, rc;
1638
1639 for (i = 0; i < ATA_MAX_DEVICES; i++)
1640 classes[i] = ATA_DEV_UNKNOWN;
1641
d4b2bab4 1642 rc = reset(ap, classes, deadline);
d87fa38e
TH
1643 if (rc)
1644 return rc;
1645
1646 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1647 * is complete and convert all ATA_DEV_UNKNOWN to
1648 * ATA_DEV_NONE.
1649 */
1650 for (i = 0; i < ATA_MAX_DEVICES; i++)
1651 if (classes[i] != ATA_DEV_UNKNOWN)
1652 break;
1653
1654 if (i < ATA_MAX_DEVICES)
1655 for (i = 0; i < ATA_MAX_DEVICES; i++)
1656 if (classes[i] == ATA_DEV_UNKNOWN)
1657 classes[i] = ATA_DEV_NONE;
1658
1659 return 0;
1660}
1661
664faf09
TH
1662static int ata_eh_followup_srst_needed(int rc, int classify,
1663 const unsigned int *classes)
1664{
1665 if (rc == -EAGAIN)
1666 return 1;
1667 if (rc != 0)
1668 return 0;
1669 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1670 return 1;
1671 return 0;
1672}
1673
1674static int ata_eh_reset(struct ata_port *ap, int classify,
f5914a46 1675 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
022bdb07
TH
1676 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1677{
1678 struct ata_eh_context *ehc = &ap->eh_context;
664faf09 1679 unsigned int *classes = ehc->classes;
1cdaf534 1680 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
31daabda
TH
1681 int try = 0;
1682 unsigned long deadline;
f5914a46 1683 unsigned int action;
022bdb07 1684 ata_reset_fn_t reset;
8b5bb2fa 1685 int i, rc;
022bdb07 1686
13abf50d
TH
1687 /* about to reset */
1688 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1689
f5914a46
TH
1690 /* Determine which reset to use and record in ehc->i.action.
1691 * prereset() may examine and modify it.
1692 */
1693 action = ehc->i.action;
1694 ehc->i.action &= ~ATA_EH_RESET_MASK;
022bdb07 1695 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
f5914a46
TH
1696 !(action & ATA_EH_HARDRESET))))
1697 ehc->i.action |= ATA_EH_SOFTRESET;
022bdb07 1698 else
f5914a46
TH
1699 ehc->i.action |= ATA_EH_HARDRESET;
1700
1701 if (prereset) {
31daabda 1702 rc = prereset(ap, jiffies + ATA_EH_PRERESET_TIMEOUT);
f5914a46 1703 if (rc) {
c961922b 1704 if (rc == -ENOENT) {
4aa9ab67
TH
1705 ata_port_printk(ap, KERN_DEBUG,
1706 "port disabled. ignoring.\n");
c961922b 1707 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
4aa9ab67
TH
1708
1709 for (i = 0; i < ATA_MAX_DEVICES; i++)
1710 classes[i] = ATA_DEV_NONE;
1711
1712 rc = 0;
c961922b
AC
1713 } else
1714 ata_port_printk(ap, KERN_ERR,
f5914a46
TH
1715 "prereset failed (errno=%d)\n", rc);
1716 return rc;
1717 }
1718 }
1719
1720 /* prereset() might have modified ehc->i.action */
1721 if (ehc->i.action & ATA_EH_HARDRESET)
022bdb07 1722 reset = hardreset;
f5914a46
TH
1723 else if (ehc->i.action & ATA_EH_SOFTRESET)
1724 reset = softreset;
1725 else {
1726 /* prereset told us not to reset, bang classes and return */
1727 for (i = 0; i < ATA_MAX_DEVICES; i++)
1728 classes[i] = ATA_DEV_NONE;
1729 return 0;
1730 }
1731
1732 /* did prereset() screw up? if so, fix up to avoid oopsing */
1733 if (!reset) {
f5914a46
TH
1734 if (softreset)
1735 reset = softreset;
1736 else
1737 reset = hardreset;
1738 }
022bdb07
TH
1739
1740 retry:
31daabda
TH
1741 deadline = jiffies + ata_eh_reset_timeouts[try++];
1742
3e706399
TH
1743 /* shut up during boot probing */
1744 if (verbose)
1745 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1746 reset == softreset ? "soft" : "hard");
022bdb07 1747
13abf50d 1748 /* mark that this EH session started with reset */
0d64a233
TH
1749 if (reset == hardreset)
1750 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1751 else
1752 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
022bdb07 1753
31daabda 1754 rc = ata_do_reset(ap, reset, classes, deadline);
022bdb07 1755
664faf09
TH
1756 if (reset == hardreset &&
1757 ata_eh_followup_srst_needed(rc, classify, classes)) {
1758 /* okay, let's do follow-up softreset */
664faf09
TH
1759 reset = softreset;
1760
1761 if (!reset) {
1762 ata_port_printk(ap, KERN_ERR,
1763 "follow-up softreset required "
1764 "but no softreset avaliable\n");
1765 return -EINVAL;
1766 }
1767
47005f25 1768 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
31daabda 1769 rc = ata_do_reset(ap, reset, classes, deadline);
664faf09
TH
1770
1771 if (rc == 0 && classify &&
1772 classes[0] == ATA_DEV_UNKNOWN) {
1773 ata_port_printk(ap, KERN_ERR,
1774 "classification failed\n");
1775 return -EINVAL;
1776 }
1777 }
1778
31daabda
TH
1779 if (rc && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
1780 unsigned long now = jiffies;
664faf09 1781
31daabda
TH
1782 if (time_before(now, deadline)) {
1783 unsigned long delta = deadline - jiffies;
664faf09 1784
31daabda
TH
1785 ata_port_printk(ap, KERN_WARNING, "reset failed "
1786 "(errno=%d), retrying in %u secs\n",
1787 rc, (jiffies_to_msecs(delta) + 999) / 1000);
1788
1789 schedule_timeout_uninterruptible(delta);
1790 }
022bdb07 1791
31daabda
TH
1792 if (reset == hardreset &&
1793 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
022bdb07
TH
1794 sata_down_spd_limit(ap);
1795 if (hardreset)
1796 reset = hardreset;
1797 goto retry;
1798 }
1799
1800 if (rc == 0) {
20952b69
TH
1801 /* After the reset, the device state is PIO 0 and the
1802 * controller state is undefined. Record the mode.
1803 */
1804 for (i = 0; i < ATA_MAX_DEVICES; i++)
1805 ap->device[i].pio_mode = XFER_PIO_0;
1806
022bdb07
TH
1807 if (postreset)
1808 postreset(ap, classes);
1809
1810 /* reset successful, schedule revalidation */
13abf50d 1811 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
022bdb07
TH
1812 ehc->i.action |= ATA_EH_REVALIDATE;
1813 }
1814
1815 return rc;
1816}
1817
084fe639
TH
1818static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1819 struct ata_device **r_failed_dev)
022bdb07
TH
1820{
1821 struct ata_eh_context *ehc = &ap->eh_context;
1822 struct ata_device *dev;
8c3c52a8 1823 unsigned int new_mask = 0;
084fe639 1824 unsigned long flags;
022bdb07
TH
1825 int i, rc = 0;
1826
1827 DPRINTK("ENTER\n");
1828
8c3c52a8
TH
1829 /* For PATA drive side cable detection to work, IDENTIFY must
1830 * be done backwards such that PDIAG- is released by the slave
1831 * device before the master device is identified.
1832 */
1833 for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) {
bff04647 1834 unsigned int action, readid_flags = 0;
47005f25 1835
022bdb07 1836 dev = &ap->device[i];
64f65ca6 1837 action = ata_eh_dev_action(dev);
022bdb07 1838
bff04647
TH
1839 if (ehc->i.flags & ATA_EHI_DID_RESET)
1840 readid_flags |= ATA_READID_POSTRESET;
1841
9666f400 1842 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
022bdb07
TH
1843 if (ata_port_offline(ap)) {
1844 rc = -EIO;
8c3c52a8 1845 goto err;
022bdb07
TH
1846 }
1847
47005f25 1848 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
bff04647 1849 rc = ata_dev_revalidate(dev, readid_flags);
022bdb07 1850 if (rc)
8c3c52a8 1851 goto err;
022bdb07 1852
47005f25
TH
1853 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1854
baa1e78a
TH
1855 /* Configuration may have changed, reconfigure
1856 * transfer mode.
1857 */
1858 ehc->i.flags |= ATA_EHI_SETMODE;
1859
3057ac3c 1860 /* schedule the scsi_rescan_device() here */
1861 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
084fe639
TH
1862 } else if (dev->class == ATA_DEV_UNKNOWN &&
1863 ehc->tries[dev->devno] &&
1864 ata_class_enabled(ehc->classes[dev->devno])) {
1865 dev->class = ehc->classes[dev->devno];
1866
bff04647
TH
1867 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1868 dev->id);
8c3c52a8
TH
1869 switch (rc) {
1870 case 0:
1871 new_mask |= 1 << i;
1872 break;
1873 case -ENOENT:
55a8e2c8
TH
1874 /* IDENTIFY was issued to non-existent
1875 * device. No need to reset. Just
1876 * thaw and kill the device.
1877 */
1878 ata_eh_thaw_port(ap);
084fe639
TH
1879 dev->class = ATA_DEV_UNKNOWN;
1880 break;
8c3c52a8
TH
1881 default:
1882 dev->class = ATA_DEV_UNKNOWN;
1883 goto err;
084fe639 1884 }
8c3c52a8
TH
1885 }
1886 }
084fe639 1887
c1c4e8d5
TH
1888 /* PDIAG- should have been released, ask cable type if post-reset */
1889 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
1890 ap->cbl = ap->ops->cable_detect(ap);
1891
8c3c52a8
TH
1892 /* Configure new devices forward such that user doesn't see
1893 * device detection messages backwards.
1894 */
1895 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1896 dev = &ap->device[i];
baa1e78a 1897
8c3c52a8
TH
1898 if (!(new_mask & (1 << i)))
1899 continue;
1900
1901 ehc->i.flags |= ATA_EHI_PRINTINFO;
1902 rc = ata_dev_configure(dev);
1903 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1904 if (rc)
1905 goto err;
1906
1907 spin_lock_irqsave(ap->lock, flags);
1908 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1909 spin_unlock_irqrestore(ap->lock, flags);
1910
1911 /* new device discovered, configure xfermode */
1912 ehc->i.flags |= ATA_EHI_SETMODE;
022bdb07
TH
1913 }
1914
8c3c52a8 1915 return 0;
022bdb07 1916
8c3c52a8
TH
1917 err:
1918 *r_failed_dev = dev;
1919 DPRINTK("EXIT rc=%d\n", rc);
022bdb07
TH
1920 return rc;
1921}
1922
1923static int ata_port_nr_enabled(struct ata_port *ap)
1924{
1925 int i, cnt = 0;
1926
1927 for (i = 0; i < ATA_MAX_DEVICES; i++)
1928 if (ata_dev_enabled(&ap->device[i]))
1929 cnt++;
1930 return cnt;
1931}
1932
084fe639
TH
1933static int ata_port_nr_vacant(struct ata_port *ap)
1934{
1935 int i, cnt = 0;
1936
1937 for (i = 0; i < ATA_MAX_DEVICES; i++)
1938 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1939 cnt++;
1940 return cnt;
1941}
1942
1943static int ata_eh_skip_recovery(struct ata_port *ap)
1944{
1945 struct ata_eh_context *ehc = &ap->eh_context;
1946 int i;
1947
7c8c2cff
TH
1948 /* thaw frozen port, resume link and recover failed devices */
1949 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
1950 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
084fe639
TH
1951 return 0;
1952
1953 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1954 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1955 struct ata_device *dev = &ap->device[i];
1956
1957 if (dev->class == ATA_DEV_UNKNOWN &&
1958 ehc->classes[dev->devno] != ATA_DEV_NONE)
1959 return 0;
1960 }
1961
1962 return 1;
1963}
1964
fee7ca72
TH
1965static void ata_eh_handle_dev_fail(struct ata_device *dev, int err)
1966{
1967 struct ata_port *ap = dev->ap;
1968 struct ata_eh_context *ehc = &ap->eh_context;
1969
1970 ehc->tries[dev->devno]--;
1971
1972 switch (err) {
1973 case -ENODEV:
1974 /* device missing or wrong IDENTIFY data, schedule probing */
1975 ehc->i.probe_mask |= (1 << dev->devno);
1976 case -EINVAL:
1977 /* give it just one more chance */
1978 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
1979 case -EIO:
1980 if (ehc->tries[dev->devno] == 1) {
1981 /* This is the last chance, better to slow
1982 * down than lose it.
1983 */
1984 sata_down_spd_limit(ap);
1985 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
1986 }
1987 }
1988
1989 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
1990 /* disable device if it has used up all its chances */
1991 ata_dev_disable(dev);
1992
1993 /* detach if offline */
1994 if (ata_port_offline(ap))
1995 ata_eh_detach_dev(dev);
1996
1997 /* probe if requested */
1998 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
1999 !(ehc->did_probe_mask & (1 << dev->devno))) {
2000 ata_eh_detach_dev(dev);
2001 ata_dev_init(dev);
2002
2003 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2004 ehc->did_probe_mask |= (1 << dev->devno);
2005 ehc->i.action |= ATA_EH_SOFTRESET;
2006 }
2007 } else {
2008 /* soft didn't work? be haaaaard */
2009 if (ehc->i.flags & ATA_EHI_DID_RESET)
2010 ehc->i.action |= ATA_EH_HARDRESET;
2011 else
2012 ehc->i.action |= ATA_EH_SOFTRESET;
2013 }
2014}
2015
022bdb07
TH
2016/**
2017 * ata_eh_recover - recover host port after error
2018 * @ap: host port to recover
f5914a46 2019 * @prereset: prereset method (can be NULL)
022bdb07
TH
2020 * @softreset: softreset method (can be NULL)
2021 * @hardreset: hardreset method (can be NULL)
2022 * @postreset: postreset method (can be NULL)
2023 *
2024 * This is the alpha and omega, eum and yang, heart and soul of
2025 * libata exception handling. On entry, actions required to
084fe639
TH
2026 * recover the port and hotplug requests are recorded in
2027 * eh_context. This function executes all the operations with
2028 * appropriate retrials and fallbacks to resurrect failed
2029 * devices, detach goners and greet newcomers.
022bdb07
TH
2030 *
2031 * LOCKING:
2032 * Kernel thread context (may sleep).
2033 *
2034 * RETURNS:
2035 * 0 on success, -errno on failure.
2036 */
f5914a46
TH
2037static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
2038 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
022bdb07
TH
2039 ata_postreset_fn_t postreset)
2040{
2041 struct ata_eh_context *ehc = &ap->eh_context;
2042 struct ata_device *dev;
4ae72a1e 2043 int i, rc;
022bdb07
TH
2044
2045 DPRINTK("ENTER\n");
2046
2047 /* prep for recovery */
2048 for (i = 0; i < ATA_MAX_DEVICES; i++) {
2049 dev = &ap->device[i];
2050
2051 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
084fe639 2052
79a55b72
TH
2053 /* collect port action mask recorded in dev actions */
2054 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
2055 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
2056
084fe639
TH
2057 /* process hotplug request */
2058 if (dev->flags & ATA_DFLAG_DETACH)
2059 ata_eh_detach_dev(dev);
2060
2061 if (!ata_dev_enabled(dev) &&
2062 ((ehc->i.probe_mask & (1 << dev->devno)) &&
2063 !(ehc->did_probe_mask & (1 << dev->devno)))) {
2064 ata_eh_detach_dev(dev);
2065 ata_dev_init(dev);
2066 ehc->did_probe_mask |= (1 << dev->devno);
2067 ehc->i.action |= ATA_EH_SOFTRESET;
2068 }
022bdb07
TH
2069 }
2070
2071 retry:
022bdb07
TH
2072 rc = 0;
2073
aeb2ecd6 2074 /* if UNLOADING, finish immediately */
b51e9e5d 2075 if (ap->pflags & ATA_PFLAG_UNLOADING)
aeb2ecd6
TH
2076 goto out;
2077
022bdb07 2078 /* skip EH if possible. */
084fe639 2079 if (ata_eh_skip_recovery(ap))
022bdb07
TH
2080 ehc->i.action = 0;
2081
084fe639
TH
2082 for (i = 0; i < ATA_MAX_DEVICES; i++)
2083 ehc->classes[i] = ATA_DEV_UNKNOWN;
2084
022bdb07
TH
2085 /* reset */
2086 if (ehc->i.action & ATA_EH_RESET_MASK) {
2087 ata_eh_freeze_port(ap);
2088
084fe639
TH
2089 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
2090 softreset, hardreset, postreset);
022bdb07
TH
2091 if (rc) {
2092 ata_port_printk(ap, KERN_ERR,
2093 "reset failed, giving up\n");
2094 goto out;
2095 }
2096
2097 ata_eh_thaw_port(ap);
2098 }
2099
084fe639
TH
2100 /* revalidate existing devices and attach new ones */
2101 rc = ata_eh_revalidate_and_attach(ap, &dev);
022bdb07
TH
2102 if (rc)
2103 goto dev_fail;
2104
baa1e78a
TH
2105 /* configure transfer mode if necessary */
2106 if (ehc->i.flags & ATA_EHI_SETMODE) {
022bdb07 2107 rc = ata_set_mode(ap, &dev);
4ae72a1e 2108 if (rc)
022bdb07 2109 goto dev_fail;
baa1e78a 2110 ehc->i.flags &= ~ATA_EHI_SETMODE;
022bdb07
TH
2111 }
2112
2113 goto out;
2114
2115 dev_fail:
fee7ca72 2116 ata_eh_handle_dev_fail(dev, rc);
022bdb07
TH
2117
2118 if (ata_port_nr_enabled(ap)) {
2119 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2120 "devices, retrying in 5 secs\n");
2121 ssleep(5);
2122 } else {
2123 /* no device left, repeat fast */
2124 msleep(500);
2125 }
2126
2127 goto retry;
2128
2129 out:
2130 if (rc) {
2131 for (i = 0; i < ATA_MAX_DEVICES; i++)
2132 ata_dev_disable(&ap->device[i]);
2133 }
2134
2135 DPRINTK("EXIT, rc=%d\n", rc);
2136 return rc;
2137}
2138
2139/**
2140 * ata_eh_finish - finish up EH
2141 * @ap: host port to finish EH for
2142 *
2143 * Recovery is complete. Clean up EH states and retry or finish
2144 * failed qcs.
2145 *
2146 * LOCKING:
2147 * None.
2148 */
2149static void ata_eh_finish(struct ata_port *ap)
2150{
2151 int tag;
2152
2153 /* retry or finish qcs */
2154 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2155 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2156
2157 if (!(qc->flags & ATA_QCFLAG_FAILED))
2158 continue;
2159
2160 if (qc->err_mask) {
2161 /* FIXME: Once EH migration is complete,
2162 * generate sense data in this function,
2163 * considering both err_mask and tf.
2164 */
2165 if (qc->err_mask & AC_ERR_INVALID)
2166 ata_eh_qc_complete(qc);
2167 else
2168 ata_eh_qc_retry(qc);
2169 } else {
2170 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2171 ata_eh_qc_complete(qc);
2172 } else {
2173 /* feed zero TF to sense generation */
2174 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2175 ata_eh_qc_retry(qc);
2176 }
2177 }
2178 }
2179}
2180
2181/**
2182 * ata_do_eh - do standard error handling
2183 * @ap: host port to handle error for
f5914a46 2184 * @prereset: prereset method (can be NULL)
022bdb07
TH
2185 * @softreset: softreset method (can be NULL)
2186 * @hardreset: hardreset method (can be NULL)
2187 * @postreset: postreset method (can be NULL)
2188 *
2189 * Perform standard error handling sequence.
2190 *
2191 * LOCKING:
2192 * Kernel thread context (may sleep).
2193 */
f5914a46
TH
2194void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2195 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2196 ata_postreset_fn_t postreset)
022bdb07 2197{
1cdaf534
TH
2198 ata_eh_autopsy(ap);
2199 ata_eh_report(ap);
f5914a46 2200 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
022bdb07
TH
2201 ata_eh_finish(ap);
2202}
500530f6 2203
6ffa01d8 2204#ifdef CONFIG_PM
500530f6
TH
2205/**
2206 * ata_eh_handle_port_suspend - perform port suspend operation
2207 * @ap: port to suspend
2208 *
2209 * Suspend @ap.
2210 *
2211 * LOCKING:
2212 * Kernel thread context (may sleep).
2213 */
2214static void ata_eh_handle_port_suspend(struct ata_port *ap)
2215{
2216 unsigned long flags;
2217 int rc = 0;
2218
2219 /* are we suspending? */
2220 spin_lock_irqsave(ap->lock, flags);
2221 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2222 ap->pm_mesg.event == PM_EVENT_ON) {
2223 spin_unlock_irqrestore(ap->lock, flags);
2224 return;
2225 }
2226 spin_unlock_irqrestore(ap->lock, flags);
2227
2228 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2229
64578a3d
TH
2230 /* tell ACPI we're suspending */
2231 rc = ata_acpi_on_suspend(ap);
2232 if (rc)
2233 goto out;
2234
500530f6
TH
2235 /* suspend */
2236 ata_eh_freeze_port(ap);
2237
2238 if (ap->ops->port_suspend)
2239 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2240
64578a3d 2241 out:
500530f6
TH
2242 /* report result */
2243 spin_lock_irqsave(ap->lock, flags);
2244
2245 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2246 if (rc == 0)
2247 ap->pflags |= ATA_PFLAG_SUSPENDED;
64578a3d 2248 else if (ap->pflags & ATA_PFLAG_FROZEN)
500530f6
TH
2249 ata_port_schedule_eh(ap);
2250
2251 if (ap->pm_result) {
2252 *ap->pm_result = rc;
2253 ap->pm_result = NULL;
2254 }
2255
2256 spin_unlock_irqrestore(ap->lock, flags);
2257
2258 return;
2259}
2260
2261/**
2262 * ata_eh_handle_port_resume - perform port resume operation
2263 * @ap: port to resume
2264 *
2265 * Resume @ap.
2266 *
500530f6
TH
2267 * LOCKING:
2268 * Kernel thread context (may sleep).
2269 */
2270static void ata_eh_handle_port_resume(struct ata_port *ap)
2271{
500530f6 2272 unsigned long flags;
9666f400 2273 int rc = 0;
500530f6
TH
2274
2275 /* are we resuming? */
2276 spin_lock_irqsave(ap->lock, flags);
2277 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2278 ap->pm_mesg.event != PM_EVENT_ON) {
2279 spin_unlock_irqrestore(ap->lock, flags);
2280 return;
2281 }
2282 spin_unlock_irqrestore(ap->lock, flags);
2283
9666f400 2284 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
500530f6
TH
2285
2286 if (ap->ops->port_resume)
2287 rc = ap->ops->port_resume(ap);
2288
6746544c
TH
2289 /* tell ACPI that we're resuming */
2290 ata_acpi_on_resume(ap);
2291
9666f400 2292 /* report result */
500530f6
TH
2293 spin_lock_irqsave(ap->lock, flags);
2294 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2295 if (ap->pm_result) {
2296 *ap->pm_result = rc;
2297 ap->pm_result = NULL;
2298 }
2299 spin_unlock_irqrestore(ap->lock, flags);
2300}
6ffa01d8 2301#endif /* CONFIG_PM */