[PATCH] libata-hp-prep: add prereset() method and implement ata_std_prereset()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / scsi / 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
35#include <linux/config.h>
36#include <linux/kernel.h>
37#include <scsi/scsi.h>
38#include <scsi/scsi_host.h>
39#include <scsi/scsi_eh.h>
40#include <scsi/scsi_device.h>
41#include <scsi/scsi_cmnd.h>
f8bbfc24 42#include "scsi_transport_api.h"
ece1d636
TH
43
44#include <linux/libata.h>
45
46#include "libata.h"
47
ad9e2762
TH
48static void __ata_port_freeze(struct ata_port *ap);
49
0c247c55
TH
50static void ata_ering_record(struct ata_ering *ering, int is_io,
51 unsigned int err_mask)
52{
53 struct ata_ering_entry *ent;
54
55 WARN_ON(!err_mask);
56
57 ering->cursor++;
58 ering->cursor %= ATA_ERING_SIZE;
59
60 ent = &ering->ring[ering->cursor];
61 ent->is_io = is_io;
62 ent->err_mask = err_mask;
63 ent->timestamp = get_jiffies_64();
64}
65
66static struct ata_ering_entry * ata_ering_top(struct ata_ering *ering)
67{
68 struct ata_ering_entry *ent = &ering->ring[ering->cursor];
69 if (!ent->err_mask)
70 return NULL;
71 return ent;
72}
73
74static int ata_ering_map(struct ata_ering *ering,
75 int (*map_fn)(struct ata_ering_entry *, void *),
76 void *arg)
77{
78 int idx, rc = 0;
79 struct ata_ering_entry *ent;
80
81 idx = ering->cursor;
82 do {
83 ent = &ering->ring[idx];
84 if (!ent->err_mask)
85 break;
86 rc = map_fn(ent, arg);
87 if (rc)
88 break;
89 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
90 } while (idx != ering->cursor);
91
92 return rc;
93}
94
ece1d636
TH
95/**
96 * ata_scsi_timed_out - SCSI layer time out callback
97 * @cmd: timed out SCSI command
98 *
99 * Handles SCSI layer timeout. We race with normal completion of
100 * the qc for @cmd. If the qc is already gone, we lose and let
101 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
102 * timed out and EH should be invoked. Prevent ata_qc_complete()
103 * from finishing it by setting EH_SCHEDULED and return
104 * EH_NOT_HANDLED.
105 *
ad9e2762
TH
106 * TODO: kill this function once old EH is gone.
107 *
ece1d636
TH
108 * LOCKING:
109 * Called from timer context
110 *
111 * RETURNS:
112 * EH_HANDLED or EH_NOT_HANDLED
113 */
114enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
115{
116 struct Scsi_Host *host = cmd->device->host;
35bb94b1 117 struct ata_port *ap = ata_shost_to_port(host);
ece1d636
TH
118 unsigned long flags;
119 struct ata_queued_cmd *qc;
ad9e2762 120 enum scsi_eh_timer_return ret;
ece1d636
TH
121
122 DPRINTK("ENTER\n");
123
ad9e2762
TH
124 if (ap->ops->error_handler) {
125 ret = EH_NOT_HANDLED;
126 goto out;
127 }
128
129 ret = EH_HANDLED;
ece1d636
TH
130 spin_lock_irqsave(&ap->host_set->lock, flags);
131 qc = ata_qc_from_tag(ap, ap->active_tag);
132 if (qc) {
133 WARN_ON(qc->scsicmd != cmd);
134 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
135 qc->err_mask |= AC_ERR_TIMEOUT;
136 ret = EH_NOT_HANDLED;
137 }
138 spin_unlock_irqrestore(&ap->host_set->lock, flags);
139
ad9e2762 140 out:
ece1d636
TH
141 DPRINTK("EXIT, ret=%d\n", ret);
142 return ret;
143}
144
145/**
146 * ata_scsi_error - SCSI layer error handler callback
147 * @host: SCSI host on which error occurred
148 *
149 * Handles SCSI-layer-thrown error events.
150 *
151 * LOCKING:
152 * Inherited from SCSI layer (none, can sleep)
153 *
154 * RETURNS:
155 * Zero.
156 */
381544bb 157void ata_scsi_error(struct Scsi_Host *host)
ece1d636 158{
35bb94b1 159 struct ata_port *ap = ata_shost_to_port(host);
ad9e2762
TH
160 spinlock_t *hs_lock = &ap->host_set->lock;
161 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
162 unsigned long flags;
ece1d636
TH
163
164 DPRINTK("ENTER\n");
165
ad9e2762 166 /* synchronize with port task */
ece1d636
TH
167 ata_port_flush_task(ap);
168
ad9e2762
TH
169 /* synchronize with host_set lock and sort out timeouts */
170
171 /* For new EH, all qcs are finished in one of three ways -
172 * normal completion, error completion, and SCSI timeout.
173 * Both cmpletions can race against SCSI timeout. When normal
174 * completion wins, the qc never reaches EH. When error
175 * completion wins, the qc has ATA_QCFLAG_FAILED set.
176 *
177 * When SCSI timeout wins, things are a bit more complex.
178 * Normal or error completion can occur after the timeout but
179 * before this point. In such cases, both types of
180 * completions are honored. A scmd is determined to have
181 * timed out iff its associated qc is active and not failed.
182 */
183 if (ap->ops->error_handler) {
184 struct scsi_cmnd *scmd, *tmp;
185 int nr_timedout = 0;
186
187 spin_lock_irqsave(hs_lock, flags);
188
189 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
190 struct ata_queued_cmd *qc;
191
192 for (i = 0; i < ATA_MAX_QUEUE; i++) {
193 qc = __ata_qc_from_tag(ap, i);
194 if (qc->flags & ATA_QCFLAG_ACTIVE &&
195 qc->scsicmd == scmd)
196 break;
197 }
198
199 if (i < ATA_MAX_QUEUE) {
200 /* the scmd has an associated qc */
201 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
202 /* which hasn't failed yet, timeout */
203 qc->err_mask |= AC_ERR_TIMEOUT;
204 qc->flags |= ATA_QCFLAG_FAILED;
205 nr_timedout++;
206 }
207 } else {
208 /* Normal completion occurred after
209 * SCSI timeout but before this point.
210 * Successfully complete it.
211 */
212 scmd->retries = scmd->allowed;
213 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
214 }
215 }
216
217 /* If we have timed out qcs. They belong to EH from
218 * this point but the state of the controller is
219 * unknown. Freeze the port to make sure the IRQ
220 * handler doesn't diddle with those qcs. This must
221 * be done atomically w.r.t. setting QCFLAG_FAILED.
222 */
223 if (nr_timedout)
224 __ata_port_freeze(ap);
225
226 spin_unlock_irqrestore(hs_lock, flags);
227 } else
228 spin_unlock_wait(hs_lock);
229
230 repeat:
231 /* invoke error handler */
232 if (ap->ops->error_handler) {
f3e81b19 233 /* fetch & clear EH info */
ad9e2762 234 spin_lock_irqsave(hs_lock, flags);
f3e81b19
TH
235
236 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
237 ap->eh_context.i = ap->eh_info;
238 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
239
c6cf9e99 240 ap->flags |= ATA_FLAG_EH_IN_PROGRESS;
ad9e2762 241 ap->flags &= ~ATA_FLAG_EH_PENDING;
f3e81b19 242
ad9e2762
TH
243 spin_unlock_irqrestore(hs_lock, flags);
244
245 /* invoke EH */
246 ap->ops->error_handler(ap);
247
248 /* Exception might have happend after ->error_handler
249 * recovered the port but before this point. Repeat
250 * EH in such case.
251 */
252 spin_lock_irqsave(hs_lock, flags);
253
254 if (ap->flags & ATA_FLAG_EH_PENDING) {
255 if (--repeat_cnt) {
256 ata_port_printk(ap, KERN_INFO,
257 "EH pending after completion, "
258 "repeating EH (cnt=%d)\n", repeat_cnt);
259 spin_unlock_irqrestore(hs_lock, flags);
260 goto repeat;
261 }
262 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
263 "tries, giving up\n", ATA_EH_MAX_REPEAT);
264 }
265
f3e81b19
TH
266 /* this run is complete, make sure EH info is clear */
267 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
268
ad9e2762
TH
269 /* Clear host_eh_scheduled while holding hs_lock such
270 * that if exception occurs after this point but
271 * before EH completion, SCSI midlayer will
272 * re-initiate EH.
273 */
274 host->host_eh_scheduled = 0;
275
276 spin_unlock_irqrestore(hs_lock, flags);
277 } else {
278 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
279 ap->ops->eng_timeout(ap);
280 }
ece1d636 281
ad9e2762 282 /* finish or retry handled scmd's and clean up */
ece1d636
TH
283 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
284
285 scsi_eh_flush_done_q(&ap->eh_done_q);
286
ad9e2762
TH
287 /* clean up */
288 spin_lock_irqsave(hs_lock, flags);
289
290 if (ap->flags & ATA_FLAG_RECOVERED)
291 ata_port_printk(ap, KERN_INFO, "EH complete\n");
292 ap->flags &= ~ATA_FLAG_RECOVERED;
293
c6cf9e99
TH
294 /* tell wait_eh that we're done */
295 ap->flags &= ~ATA_FLAG_EH_IN_PROGRESS;
296 wake_up_all(&ap->eh_wait_q);
297
ad9e2762
TH
298 spin_unlock_irqrestore(hs_lock, flags);
299
ece1d636 300 DPRINTK("EXIT\n");
ece1d636
TH
301}
302
c6cf9e99
TH
303/**
304 * ata_port_wait_eh - Wait for the currently pending EH to complete
305 * @ap: Port to wait EH for
306 *
307 * Wait until the currently pending EH is complete.
308 *
309 * LOCKING:
310 * Kernel thread context (may sleep).
311 */
312void ata_port_wait_eh(struct ata_port *ap)
313{
314 unsigned long flags;
315 DEFINE_WAIT(wait);
316
317 retry:
318 spin_lock_irqsave(&ap->host_set->lock, flags);
319
320 while (ap->flags & (ATA_FLAG_EH_PENDING | ATA_FLAG_EH_IN_PROGRESS)) {
321 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
322 spin_unlock_irqrestore(&ap->host_set->lock, flags);
323 schedule();
324 spin_lock_irqsave(&ap->host_set->lock, flags);
325 }
326
327 spin_unlock_irqrestore(&ap->host_set->lock, flags);
328
329 /* make sure SCSI EH is complete */
330 if (scsi_host_in_recovery(ap->host)) {
331 msleep(10);
332 goto retry;
333 }
334}
335
ece1d636
TH
336/**
337 * ata_qc_timeout - Handle timeout of queued command
338 * @qc: Command that timed out
339 *
340 * Some part of the kernel (currently, only the SCSI layer)
341 * has noticed that the active command on port @ap has not
342 * completed after a specified length of time. Handle this
343 * condition by disabling DMA (if necessary) and completing
344 * transactions, with error if necessary.
345 *
346 * This also handles the case of the "lost interrupt", where
347 * for some reason (possibly hardware bug, possibly driver bug)
348 * an interrupt was not delivered to the driver, even though the
349 * transaction completed successfully.
350 *
ad9e2762
TH
351 * TODO: kill this function once old EH is gone.
352 *
ece1d636
TH
353 * LOCKING:
354 * Inherited from SCSI layer (none, can sleep)
355 */
356static void ata_qc_timeout(struct ata_queued_cmd *qc)
357{
358 struct ata_port *ap = qc->ap;
359 struct ata_host_set *host_set = ap->host_set;
360 u8 host_stat = 0, drv_stat;
361 unsigned long flags;
362
363 DPRINTK("ENTER\n");
364
365 ap->hsm_task_state = HSM_ST_IDLE;
366
367 spin_lock_irqsave(&host_set->lock, flags);
368
369 switch (qc->tf.protocol) {
370
371 case ATA_PROT_DMA:
372 case ATA_PROT_ATAPI_DMA:
373 host_stat = ap->ops->bmdma_status(ap);
374
375 /* before we do anything else, clear DMA-Start bit */
376 ap->ops->bmdma_stop(qc);
377
378 /* fall through */
379
380 default:
381 ata_altstatus(ap);
382 drv_stat = ata_chk_status(ap);
383
384 /* ack bmdma irq events */
385 ap->ops->irq_clear(ap);
386
f15a1daf
TH
387 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
388 "stat 0x%x host_stat 0x%x\n",
389 qc->tf.command, drv_stat, host_stat);
ece1d636
TH
390
391 /* complete taskfile transaction */
c13b56a1 392 qc->err_mask |= AC_ERR_TIMEOUT;
ece1d636
TH
393 break;
394 }
395
396 spin_unlock_irqrestore(&host_set->lock, flags);
397
398 ata_eh_qc_complete(qc);
399
400 DPRINTK("EXIT\n");
401}
402
403/**
404 * ata_eng_timeout - Handle timeout of queued command
405 * @ap: Port on which timed-out command is active
406 *
407 * Some part of the kernel (currently, only the SCSI layer)
408 * has noticed that the active command on port @ap has not
409 * completed after a specified length of time. Handle this
410 * condition by disabling DMA (if necessary) and completing
411 * transactions, with error if necessary.
412 *
413 * This also handles the case of the "lost interrupt", where
414 * for some reason (possibly hardware bug, possibly driver bug)
415 * an interrupt was not delivered to the driver, even though the
416 * transaction completed successfully.
417 *
ad9e2762
TH
418 * TODO: kill this function once old EH is gone.
419 *
ece1d636
TH
420 * LOCKING:
421 * Inherited from SCSI layer (none, can sleep)
422 */
423void ata_eng_timeout(struct ata_port *ap)
424{
425 DPRINTK("ENTER\n");
426
427 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
428
429 DPRINTK("EXIT\n");
430}
431
f686bcb8
TH
432/**
433 * ata_qc_schedule_eh - schedule qc for error handling
434 * @qc: command to schedule error handling for
435 *
436 * Schedule error handling for @qc. EH will kick in as soon as
437 * other commands are drained.
438 *
439 * LOCKING:
440 * spin_lock_irqsave(host_set lock)
441 */
442void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
443{
444 struct ata_port *ap = qc->ap;
445
446 WARN_ON(!ap->ops->error_handler);
447
448 qc->flags |= ATA_QCFLAG_FAILED;
449 qc->ap->flags |= ATA_FLAG_EH_PENDING;
450
451 /* The following will fail if timeout has already expired.
452 * ata_scsi_error() takes care of such scmds on EH entry.
453 * Note that ATA_QCFLAG_FAILED is unconditionally set after
454 * this function completes.
455 */
456 scsi_req_abort_cmd(qc->scsicmd);
457}
458
7b70fc03
TH
459/**
460 * ata_port_schedule_eh - schedule error handling without a qc
461 * @ap: ATA port to schedule EH for
462 *
463 * Schedule error handling for @ap. EH will kick in as soon as
464 * all commands are drained.
465 *
466 * LOCKING:
467 * spin_lock_irqsave(host_set lock)
468 */
469void ata_port_schedule_eh(struct ata_port *ap)
470{
471 WARN_ON(!ap->ops->error_handler);
472
473 ap->flags |= ATA_FLAG_EH_PENDING;
f8bbfc24 474 scsi_schedule_eh(ap->host);
7b70fc03
TH
475
476 DPRINTK("port EH scheduled\n");
477}
478
479/**
480 * ata_port_abort - abort all qc's on the port
481 * @ap: ATA port to abort qc's for
482 *
483 * Abort all active qc's of @ap and schedule EH.
484 *
485 * LOCKING:
486 * spin_lock_irqsave(host_set lock)
487 *
488 * RETURNS:
489 * Number of aborted qc's.
490 */
491int ata_port_abort(struct ata_port *ap)
492{
493 int tag, nr_aborted = 0;
494
495 WARN_ON(!ap->ops->error_handler);
496
497 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
498 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
499
500 if (qc) {
501 qc->flags |= ATA_QCFLAG_FAILED;
502 ata_qc_complete(qc);
503 nr_aborted++;
504 }
505 }
506
507 if (!nr_aborted)
508 ata_port_schedule_eh(ap);
509
510 return nr_aborted;
511}
512
e3180499
TH
513/**
514 * __ata_port_freeze - freeze port
515 * @ap: ATA port to freeze
516 *
517 * This function is called when HSM violation or some other
518 * condition disrupts normal operation of the port. Frozen port
519 * is not allowed to perform any operation until the port is
520 * thawed, which usually follows a successful reset.
521 *
522 * ap->ops->freeze() callback can be used for freezing the port
523 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
524 * port cannot be frozen hardware-wise, the interrupt handler
525 * must ack and clear interrupts unconditionally while the port
526 * is frozen.
527 *
528 * LOCKING:
529 * spin_lock_irqsave(host_set lock)
530 */
531static void __ata_port_freeze(struct ata_port *ap)
532{
533 WARN_ON(!ap->ops->error_handler);
534
535 if (ap->ops->freeze)
536 ap->ops->freeze(ap);
537
538 ap->flags |= ATA_FLAG_FROZEN;
539
540 DPRINTK("ata%u port frozen\n", ap->id);
541}
542
543/**
544 * ata_port_freeze - abort & freeze port
545 * @ap: ATA port to freeze
546 *
547 * Abort and freeze @ap.
548 *
549 * LOCKING:
550 * spin_lock_irqsave(host_set lock)
551 *
552 * RETURNS:
553 * Number of aborted commands.
554 */
555int ata_port_freeze(struct ata_port *ap)
556{
557 int nr_aborted;
558
559 WARN_ON(!ap->ops->error_handler);
560
561 nr_aborted = ata_port_abort(ap);
562 __ata_port_freeze(ap);
563
564 return nr_aborted;
565}
566
567/**
568 * ata_eh_freeze_port - EH helper to freeze port
569 * @ap: ATA port to freeze
570 *
571 * Freeze @ap.
572 *
573 * LOCKING:
574 * None.
575 */
576void ata_eh_freeze_port(struct ata_port *ap)
577{
578 unsigned long flags;
579
580 if (!ap->ops->error_handler)
581 return;
582
583 spin_lock_irqsave(&ap->host_set->lock, flags);
584 __ata_port_freeze(ap);
585 spin_unlock_irqrestore(&ap->host_set->lock, flags);
586}
587
588/**
589 * ata_port_thaw_port - EH helper to thaw port
590 * @ap: ATA port to thaw
591 *
592 * Thaw frozen port @ap.
593 *
594 * LOCKING:
595 * None.
596 */
597void ata_eh_thaw_port(struct ata_port *ap)
598{
599 unsigned long flags;
600
601 if (!ap->ops->error_handler)
602 return;
603
604 spin_lock_irqsave(&ap->host_set->lock, flags);
605
606 ap->flags &= ~ATA_FLAG_FROZEN;
607
608 if (ap->ops->thaw)
609 ap->ops->thaw(ap);
610
611 spin_unlock_irqrestore(&ap->host_set->lock, flags);
612
613 DPRINTK("ata%u port thawed\n", ap->id);
614}
615
ece1d636
TH
616static void ata_eh_scsidone(struct scsi_cmnd *scmd)
617{
618 /* nada */
619}
620
621static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
622{
623 struct ata_port *ap = qc->ap;
624 struct scsi_cmnd *scmd = qc->scsicmd;
625 unsigned long flags;
626
627 spin_lock_irqsave(&ap->host_set->lock, flags);
628 qc->scsidone = ata_eh_scsidone;
629 __ata_qc_complete(qc);
630 WARN_ON(ata_tag_valid(qc->tag));
631 spin_unlock_irqrestore(&ap->host_set->lock, flags);
632
633 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
634}
635
636/**
637 * ata_eh_qc_complete - Complete an active ATA command from EH
638 * @qc: Command to complete
639 *
640 * Indicate to the mid and upper layers that an ATA command has
641 * completed. To be used from EH.
642 */
643void ata_eh_qc_complete(struct ata_queued_cmd *qc)
644{
645 struct scsi_cmnd *scmd = qc->scsicmd;
646 scmd->retries = scmd->allowed;
647 __ata_eh_qc_complete(qc);
648}
649
650/**
651 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
652 * @qc: Command to retry
653 *
654 * Indicate to the mid and upper layers that an ATA command
655 * should be retried. To be used from EH.
656 *
657 * SCSI midlayer limits the number of retries to scmd->allowed.
658 * scmd->retries is decremented for commands which get retried
659 * due to unrelated failures (qc->err_mask is zero).
660 */
661void ata_eh_qc_retry(struct ata_queued_cmd *qc)
662{
663 struct scsi_cmnd *scmd = qc->scsicmd;
664 if (!qc->err_mask && scmd->retries)
665 scmd->retries--;
666 __ata_eh_qc_complete(qc);
667}
022bdb07
TH
668
669/**
670 * ata_eh_about_to_do - about to perform eh_action
671 * @ap: target ATA port
672 * @action: action about to be performed
673 *
674 * Called just before performing EH actions to clear related bits
675 * in @ap->eh_info such that eh actions are not unnecessarily
676 * repeated.
677 *
678 * LOCKING:
679 * None.
680 */
681static void ata_eh_about_to_do(struct ata_port *ap, unsigned int action)
682{
683 unsigned long flags;
684
685 spin_lock_irqsave(&ap->host_set->lock, flags);
686 ap->eh_info.action &= ~action;
687 ap->flags |= ATA_FLAG_RECOVERED;
688 spin_unlock_irqrestore(&ap->host_set->lock, flags);
689}
690
691/**
692 * ata_err_string - convert err_mask to descriptive string
693 * @err_mask: error mask to convert to string
694 *
695 * Convert @err_mask to descriptive string. Errors are
696 * prioritized according to severity and only the most severe
697 * error is reported.
698 *
699 * LOCKING:
700 * None.
701 *
702 * RETURNS:
703 * Descriptive string for @err_mask
704 */
705static const char * ata_err_string(unsigned int err_mask)
706{
707 if (err_mask & AC_ERR_HOST_BUS)
708 return "host bus error";
709 if (err_mask & AC_ERR_ATA_BUS)
710 return "ATA bus error";
711 if (err_mask & AC_ERR_TIMEOUT)
712 return "timeout";
713 if (err_mask & AC_ERR_HSM)
714 return "HSM violation";
715 if (err_mask & AC_ERR_SYSTEM)
716 return "internal error";
717 if (err_mask & AC_ERR_MEDIA)
718 return "media error";
719 if (err_mask & AC_ERR_INVALID)
720 return "invalid argument";
721 if (err_mask & AC_ERR_DEV)
722 return "device error";
723 return "unknown error";
724}
725
e8ee8451
TH
726/**
727 * ata_read_log_page - read a specific log page
728 * @dev: target device
729 * @page: page to read
730 * @buf: buffer to store read page
731 * @sectors: number of sectors to read
732 *
733 * Read log page using READ_LOG_EXT command.
734 *
735 * LOCKING:
736 * Kernel thread context (may sleep).
737 *
738 * RETURNS:
739 * 0 on success, AC_ERR_* mask otherwise.
740 */
741static unsigned int ata_read_log_page(struct ata_device *dev,
742 u8 page, void *buf, unsigned int sectors)
743{
744 struct ata_taskfile tf;
745 unsigned int err_mask;
746
747 DPRINTK("read log page - page %d\n", page);
748
749 ata_tf_init(dev, &tf);
750 tf.command = ATA_CMD_READ_LOG_EXT;
751 tf.lbal = page;
752 tf.nsect = sectors;
753 tf.hob_nsect = sectors >> 8;
754 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
755 tf.protocol = ATA_PROT_PIO;
756
757 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
758 buf, sectors * ATA_SECT_SIZE);
759
760 DPRINTK("EXIT, err_mask=%x\n", err_mask);
761 return err_mask;
762}
763
764/**
765 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
766 * @dev: Device to read log page 10h from
767 * @tag: Resulting tag of the failed command
768 * @tf: Resulting taskfile registers of the failed command
769 *
770 * Read log page 10h to obtain NCQ error details and clear error
771 * condition.
772 *
773 * LOCKING:
774 * Kernel thread context (may sleep).
775 *
776 * RETURNS:
777 * 0 on success, -errno otherwise.
778 */
779static int ata_eh_read_log_10h(struct ata_device *dev,
780 int *tag, struct ata_taskfile *tf)
781{
782 u8 *buf = dev->ap->sector_buf;
783 unsigned int err_mask;
784 u8 csum;
785 int i;
786
787 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
788 if (err_mask)
789 return -EIO;
790
791 csum = 0;
792 for (i = 0; i < ATA_SECT_SIZE; i++)
793 csum += buf[i];
794 if (csum)
795 ata_dev_printk(dev, KERN_WARNING,
796 "invalid checksum 0x%x on log page 10h\n", csum);
797
798 if (buf[0] & 0x80)
799 return -ENOENT;
800
801 *tag = buf[0] & 0x1f;
802
803 tf->command = buf[2];
804 tf->feature = buf[3];
805 tf->lbal = buf[4];
806 tf->lbam = buf[5];
807 tf->lbah = buf[6];
808 tf->device = buf[7];
809 tf->hob_lbal = buf[8];
810 tf->hob_lbam = buf[9];
811 tf->hob_lbah = buf[10];
812 tf->nsect = buf[12];
813 tf->hob_nsect = buf[13];
814
815 return 0;
816}
817
022bdb07
TH
818/**
819 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
820 * @dev: device to perform REQUEST_SENSE to
821 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
822 *
823 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
824 * SENSE. This function is EH helper.
825 *
826 * LOCKING:
827 * Kernel thread context (may sleep).
828 *
829 * RETURNS:
830 * 0 on success, AC_ERR_* mask on failure
831 */
832static unsigned int atapi_eh_request_sense(struct ata_device *dev,
833 unsigned char *sense_buf)
834{
835 struct ata_port *ap = dev->ap;
836 struct ata_taskfile tf;
837 u8 cdb[ATAPI_CDB_LEN];
838
839 DPRINTK("ATAPI request sense\n");
840
841 ata_tf_init(dev, &tf);
842
843 /* FIXME: is this needed? */
844 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
845
846 /* XXX: why tf_read here? */
847 ap->ops->tf_read(ap, &tf);
848
849 /* fill these in, for the case where they are -not- overwritten */
850 sense_buf[0] = 0x70;
851 sense_buf[2] = tf.feature >> 4;
852
853 memset(cdb, 0, ATAPI_CDB_LEN);
854 cdb[0] = REQUEST_SENSE;
855 cdb[4] = SCSI_SENSE_BUFFERSIZE;
856
857 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
858 tf.command = ATA_CMD_PACKET;
859
860 /* is it pointless to prefer PIO for "safety reasons"? */
861 if (ap->flags & ATA_FLAG_PIO_DMA) {
862 tf.protocol = ATA_PROT_ATAPI_DMA;
863 tf.feature |= ATAPI_PKT_DMA;
864 } else {
865 tf.protocol = ATA_PROT_ATAPI;
866 tf.lbam = (8 * 1024) & 0xff;
867 tf.lbah = (8 * 1024) >> 8;
868 }
869
870 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
871 sense_buf, SCSI_SENSE_BUFFERSIZE);
872}
873
874/**
875 * ata_eh_analyze_serror - analyze SError for a failed port
876 * @ap: ATA port to analyze SError for
877 *
878 * Analyze SError if available and further determine cause of
879 * failure.
880 *
881 * LOCKING:
882 * None.
883 */
884static void ata_eh_analyze_serror(struct ata_port *ap)
885{
886 struct ata_eh_context *ehc = &ap->eh_context;
887 u32 serror = ehc->i.serror;
888 unsigned int err_mask = 0, action = 0;
889
890 if (serror & SERR_PERSISTENT) {
891 err_mask |= AC_ERR_ATA_BUS;
892 action |= ATA_EH_HARDRESET;
893 }
894 if (serror &
895 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
896 err_mask |= AC_ERR_ATA_BUS;
897 action |= ATA_EH_SOFTRESET;
898 }
899 if (serror & SERR_PROTOCOL) {
900 err_mask |= AC_ERR_HSM;
901 action |= ATA_EH_SOFTRESET;
902 }
903 if (serror & SERR_INTERNAL) {
904 err_mask |= AC_ERR_SYSTEM;
905 action |= ATA_EH_SOFTRESET;
906 }
907 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG)) {
908 err_mask |= AC_ERR_ATA_BUS;
909 action |= ATA_EH_HARDRESET;
910 }
911
912 ehc->i.err_mask |= err_mask;
913 ehc->i.action |= action;
914}
915
e8ee8451
TH
916/**
917 * ata_eh_analyze_ncq_error - analyze NCQ error
918 * @ap: ATA port to analyze NCQ error for
919 *
920 * Read log page 10h, determine the offending qc and acquire
921 * error status TF. For NCQ device errors, all LLDDs have to do
922 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
923 * care of the rest.
924 *
925 * LOCKING:
926 * Kernel thread context (may sleep).
927 */
928static void ata_eh_analyze_ncq_error(struct ata_port *ap)
929{
930 struct ata_eh_context *ehc = &ap->eh_context;
931 struct ata_device *dev = ap->device;
932 struct ata_queued_cmd *qc;
933 struct ata_taskfile tf;
934 int tag, rc;
935
936 /* if frozen, we can't do much */
937 if (ap->flags & ATA_FLAG_FROZEN)
938 return;
939
940 /* is it NCQ device error? */
941 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
942 return;
943
944 /* has LLDD analyzed already? */
945 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
946 qc = __ata_qc_from_tag(ap, tag);
947
948 if (!(qc->flags & ATA_QCFLAG_FAILED))
949 continue;
950
951 if (qc->err_mask)
952 return;
953 }
954
955 /* okay, this error is ours */
956 rc = ata_eh_read_log_10h(dev, &tag, &tf);
957 if (rc) {
958 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
959 "(errno=%d)\n", rc);
960 return;
961 }
962
963 if (!(ap->sactive & (1 << tag))) {
964 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
965 "inactive tag %d\n", tag);
966 return;
967 }
968
969 /* we've got the perpetrator, condemn it */
970 qc = __ata_qc_from_tag(ap, tag);
971 memcpy(&qc->result_tf, &tf, sizeof(tf));
972 qc->err_mask |= AC_ERR_DEV;
973 ehc->i.err_mask &= ~AC_ERR_DEV;
974}
975
022bdb07
TH
976/**
977 * ata_eh_analyze_tf - analyze taskfile of a failed qc
978 * @qc: qc to analyze
979 * @tf: Taskfile registers to analyze
980 *
981 * Analyze taskfile of @qc and further determine cause of
982 * failure. This function also requests ATAPI sense data if
983 * avaliable.
984 *
985 * LOCKING:
986 * Kernel thread context (may sleep).
987 *
988 * RETURNS:
989 * Determined recovery action
990 */
991static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
992 const struct ata_taskfile *tf)
993{
994 unsigned int tmp, action = 0;
995 u8 stat = tf->command, err = tf->feature;
996
997 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
998 qc->err_mask |= AC_ERR_HSM;
999 return ATA_EH_SOFTRESET;
1000 }
1001
1002 if (!(qc->err_mask & AC_ERR_DEV))
1003 return 0;
1004
1005 switch (qc->dev->class) {
1006 case ATA_DEV_ATA:
1007 if (err & ATA_ICRC)
1008 qc->err_mask |= AC_ERR_ATA_BUS;
1009 if (err & ATA_UNC)
1010 qc->err_mask |= AC_ERR_MEDIA;
1011 if (err & ATA_IDNF)
1012 qc->err_mask |= AC_ERR_INVALID;
1013 break;
1014
1015 case ATA_DEV_ATAPI:
1016 tmp = atapi_eh_request_sense(qc->dev,
1017 qc->scsicmd->sense_buffer);
1018 if (!tmp) {
1019 /* ATA_QCFLAG_SENSE_VALID is used to tell
1020 * atapi_qc_complete() that sense data is
1021 * already valid.
1022 *
1023 * TODO: interpret sense data and set
1024 * appropriate err_mask.
1025 */
1026 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1027 } else
1028 qc->err_mask |= tmp;
1029 }
1030
1031 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1032 action |= ATA_EH_SOFTRESET;
1033
1034 return action;
1035}
1036
1037static int ata_eh_categorize_ering_entry(struct ata_ering_entry *ent)
1038{
1039 if (ent->err_mask & (AC_ERR_ATA_BUS | AC_ERR_TIMEOUT))
1040 return 1;
1041
1042 if (ent->is_io) {
1043 if (ent->err_mask & AC_ERR_HSM)
1044 return 1;
1045 if ((ent->err_mask &
1046 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1047 return 2;
1048 }
1049
1050 return 0;
1051}
1052
1053struct speed_down_needed_arg {
1054 u64 since;
1055 int nr_errors[3];
1056};
1057
1058static int speed_down_needed_cb(struct ata_ering_entry *ent, void *void_arg)
1059{
1060 struct speed_down_needed_arg *arg = void_arg;
1061
1062 if (ent->timestamp < arg->since)
1063 return -1;
1064
1065 arg->nr_errors[ata_eh_categorize_ering_entry(ent)]++;
1066 return 0;
1067}
1068
1069/**
1070 * ata_eh_speed_down_needed - Determine wheter speed down is necessary
1071 * @dev: Device of interest
1072 *
1073 * This function examines error ring of @dev and determines
1074 * whether speed down is necessary. Speed down is necessary if
1075 * there have been more than 3 of Cat-1 errors or 10 of Cat-2
1076 * errors during last 15 minutes.
1077 *
1078 * Cat-1 errors are ATA_BUS, TIMEOUT for any command and HSM
1079 * violation for known supported commands.
1080 *
1081 * Cat-2 errors are unclassified DEV error for known supported
1082 * command.
1083 *
1084 * LOCKING:
1085 * Inherited from caller.
1086 *
1087 * RETURNS:
1088 * 1 if speed down is necessary, 0 otherwise
1089 */
1090static int ata_eh_speed_down_needed(struct ata_device *dev)
1091{
1092 const u64 interval = 15LLU * 60 * HZ;
1093 static const int err_limits[3] = { -1, 3, 10 };
1094 struct speed_down_needed_arg arg;
1095 struct ata_ering_entry *ent;
1096 int err_cat;
1097 u64 j64;
1098
1099 ent = ata_ering_top(&dev->ering);
1100 if (!ent)
1101 return 0;
1102
1103 err_cat = ata_eh_categorize_ering_entry(ent);
1104 if (err_cat == 0)
1105 return 0;
1106
1107 memset(&arg, 0, sizeof(arg));
1108
1109 j64 = get_jiffies_64();
1110 if (j64 >= interval)
1111 arg.since = j64 - interval;
1112 else
1113 arg.since = 0;
1114
1115 ata_ering_map(&dev->ering, speed_down_needed_cb, &arg);
1116
1117 return arg.nr_errors[err_cat] > err_limits[err_cat];
1118}
1119
1120/**
1121 * ata_eh_speed_down - record error and speed down if necessary
1122 * @dev: Failed device
1123 * @is_io: Did the device fail during normal IO?
1124 * @err_mask: err_mask of the error
1125 *
1126 * Record error and examine error history to determine whether
1127 * adjusting transmission speed is necessary. It also sets
1128 * transmission limits appropriately if such adjustment is
1129 * necessary.
1130 *
1131 * LOCKING:
1132 * Kernel thread context (may sleep).
1133 *
1134 * RETURNS:
1135 * 0 on success, -errno otherwise
1136 */
1137static int ata_eh_speed_down(struct ata_device *dev, int is_io,
1138 unsigned int err_mask)
1139{
1140 if (!err_mask)
1141 return 0;
1142
1143 /* record error and determine whether speed down is necessary */
1144 ata_ering_record(&dev->ering, is_io, err_mask);
1145
1146 if (!ata_eh_speed_down_needed(dev))
1147 return 0;
1148
1149 /* speed down SATA link speed if possible */
1150 if (sata_down_spd_limit(dev->ap) == 0)
1151 return ATA_EH_HARDRESET;
1152
1153 /* lower transfer mode */
1154 if (ata_down_xfermask_limit(dev, 0) == 0)
1155 return ATA_EH_SOFTRESET;
1156
1157 ata_dev_printk(dev, KERN_ERR,
1158 "speed down requested but no transfer mode left\n");
1159 return 0;
1160}
1161
1162/**
1163 * ata_eh_autopsy - analyze error and determine recovery action
1164 * @ap: ATA port to perform autopsy on
1165 *
1166 * Analyze why @ap failed and determine which recovery action is
1167 * needed. This function also sets more detailed AC_ERR_* values
1168 * and fills sense data for ATAPI CHECK SENSE.
1169 *
1170 * LOCKING:
1171 * Kernel thread context (may sleep).
1172 */
1173static void ata_eh_autopsy(struct ata_port *ap)
1174{
1175 struct ata_eh_context *ehc = &ap->eh_context;
1176 unsigned int action = ehc->i.action;
1177 struct ata_device *failed_dev = NULL;
1178 unsigned int all_err_mask = 0;
1179 int tag, is_io = 0;
1180 u32 serror;
1181 int rc;
1182
1183 DPRINTK("ENTER\n");
1184
1185 /* obtain and analyze SError */
1186 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1187 if (rc == 0) {
1188 ehc->i.serror |= serror;
1189 ata_eh_analyze_serror(ap);
1190 } else if (rc != -EOPNOTSUPP)
1191 action |= ATA_EH_HARDRESET;
1192
e8ee8451
TH
1193 /* analyze NCQ failure */
1194 ata_eh_analyze_ncq_error(ap);
1195
022bdb07
TH
1196 /* any real error trumps AC_ERR_OTHER */
1197 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1198 ehc->i.err_mask &= ~AC_ERR_OTHER;
1199
1200 all_err_mask |= ehc->i.err_mask;
1201
1202 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1203 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1204
1205 if (!(qc->flags & ATA_QCFLAG_FAILED))
1206 continue;
1207
1208 /* inherit upper level err_mask */
1209 qc->err_mask |= ehc->i.err_mask;
1210
022bdb07
TH
1211 /* analyze TF */
1212 action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1213
1214 /* DEV errors are probably spurious in case of ATA_BUS error */
1215 if (qc->err_mask & AC_ERR_ATA_BUS)
1216 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1217 AC_ERR_INVALID);
1218
1219 /* any real error trumps unknown error */
1220 if (qc->err_mask & ~AC_ERR_OTHER)
1221 qc->err_mask &= ~AC_ERR_OTHER;
1222
1223 /* SENSE_VALID trumps dev/unknown error and revalidation */
1224 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1225 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1226 action &= ~ATA_EH_REVALIDATE;
1227 }
1228
1229 /* accumulate error info */
1230 failed_dev = qc->dev;
1231 all_err_mask |= qc->err_mask;
1232 if (qc->flags & ATA_QCFLAG_IO)
1233 is_io = 1;
1234 }
1235
1236 /* speed down iff command was in progress */
1237 if (failed_dev)
1238 action |= ata_eh_speed_down(failed_dev, is_io, all_err_mask);
1239
a20f33ff
TH
1240 /* enforce default EH actions */
1241 if (ap->flags & ATA_FLAG_FROZEN ||
1242 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1243 action |= ATA_EH_SOFTRESET;
1244 else if (all_err_mask)
022bdb07
TH
1245 action |= ATA_EH_REVALIDATE;
1246
a20f33ff 1247 /* record autopsy result */
022bdb07
TH
1248 ehc->i.dev = failed_dev;
1249 ehc->i.action = action;
1250
1251 DPRINTK("EXIT\n");
1252}
1253
1254/**
1255 * ata_eh_report - report error handling to user
1256 * @ap: ATA port EH is going on
1257 *
1258 * Report EH to user.
1259 *
1260 * LOCKING:
1261 * None.
1262 */
1263static void ata_eh_report(struct ata_port *ap)
1264{
1265 struct ata_eh_context *ehc = &ap->eh_context;
1266 const char *frozen, *desc;
1267 int tag, nr_failed = 0;
1268
1269 desc = NULL;
1270 if (ehc->i.desc[0] != '\0')
1271 desc = ehc->i.desc;
1272
1273 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1274 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1275
1276 if (!(qc->flags & ATA_QCFLAG_FAILED))
1277 continue;
1278 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1279 continue;
1280
1281 nr_failed++;
1282 }
1283
1284 if (!nr_failed && !ehc->i.err_mask)
1285 return;
1286
1287 frozen = "";
1288 if (ap->flags & ATA_FLAG_FROZEN)
1289 frozen = " frozen";
1290
1291 if (ehc->i.dev) {
e8ee8451
TH
1292 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1293 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1294 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1295 ehc->i.action, frozen);
022bdb07
TH
1296 if (desc)
1297 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1298 } else {
e8ee8451
TH
1299 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1300 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1301 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1302 ehc->i.action, frozen);
022bdb07
TH
1303 if (desc)
1304 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1305 }
1306
1307 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1308 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1309
1310 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1311 continue;
1312
1313 ata_dev_printk(qc->dev, KERN_ERR, "tag %d cmd 0x%x "
1314 "Emask 0x%x stat 0x%x err 0x%x (%s)\n",
1315 qc->tag, qc->tf.command, qc->err_mask,
1316 qc->result_tf.command, qc->result_tf.feature,
1317 ata_err_string(qc->err_mask));
1318 }
1319}
1320
f5914a46
TH
1321static int ata_eh_reset(struct ata_port *ap,
1322 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
022bdb07
TH
1323 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1324{
1325 struct ata_eh_context *ehc = &ap->eh_context;
1326 unsigned int classes[ATA_MAX_DEVICES];
1327 int tries = ATA_EH_RESET_TRIES;
f5914a46 1328 unsigned int action;
022bdb07 1329 ata_reset_fn_t reset;
20952b69 1330 int i, rc;
022bdb07 1331
f5914a46
TH
1332 /* Determine which reset to use and record in ehc->i.action.
1333 * prereset() may examine and modify it.
1334 */
1335 action = ehc->i.action;
1336 ehc->i.action &= ~ATA_EH_RESET_MASK;
022bdb07 1337 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
f5914a46
TH
1338 !(action & ATA_EH_HARDRESET))))
1339 ehc->i.action |= ATA_EH_SOFTRESET;
022bdb07 1340 else
f5914a46
TH
1341 ehc->i.action |= ATA_EH_HARDRESET;
1342
1343 if (prereset) {
1344 rc = prereset(ap);
1345 if (rc) {
1346 ata_port_printk(ap, KERN_ERR,
1347 "prereset failed (errno=%d)\n", rc);
1348 return rc;
1349 }
1350 }
1351
1352 /* prereset() might have modified ehc->i.action */
1353 if (ehc->i.action & ATA_EH_HARDRESET)
022bdb07 1354 reset = hardreset;
f5914a46
TH
1355 else if (ehc->i.action & ATA_EH_SOFTRESET)
1356 reset = softreset;
1357 else {
1358 /* prereset told us not to reset, bang classes and return */
1359 for (i = 0; i < ATA_MAX_DEVICES; i++)
1360 classes[i] = ATA_DEV_NONE;
1361 return 0;
1362 }
1363
1364 /* did prereset() screw up? if so, fix up to avoid oopsing */
1365 if (!reset) {
1366 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1367 "invalid reset type\n");
1368 if (softreset)
1369 reset = softreset;
1370 else
1371 reset = hardreset;
1372 }
022bdb07
TH
1373
1374 retry:
1375 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1376 reset == softreset ? "soft" : "hard");
1377
1378 /* reset */
1379 ata_eh_about_to_do(ap, ATA_EH_RESET_MASK);
1380 ehc->i.flags |= ATA_EHI_DID_RESET;
1381
1382 rc = ata_do_reset(ap, reset, classes);
1383
1384 if (rc && --tries) {
1385 ata_port_printk(ap, KERN_WARNING,
1386 "%sreset failed, retrying in 5 secs\n",
1387 reset == softreset ? "soft" : "hard");
1388 ssleep(5);
1389
1390 if (reset == hardreset)
1391 sata_down_spd_limit(ap);
1392 if (hardreset)
1393 reset = hardreset;
1394 goto retry;
1395 }
1396
1397 if (rc == 0) {
20952b69
TH
1398 /* After the reset, the device state is PIO 0 and the
1399 * controller state is undefined. Record the mode.
1400 */
1401 for (i = 0; i < ATA_MAX_DEVICES; i++)
1402 ap->device[i].pio_mode = XFER_PIO_0;
1403
022bdb07
TH
1404 if (postreset)
1405 postreset(ap, classes);
1406
1407 /* reset successful, schedule revalidation */
1408 ehc->i.dev = NULL;
1409 ehc->i.action &= ~ATA_EH_RESET_MASK;
1410 ehc->i.action |= ATA_EH_REVALIDATE;
1411 }
1412
1413 return rc;
1414}
1415
1416static int ata_eh_revalidate(struct ata_port *ap,
1417 struct ata_device **r_failed_dev)
1418{
1419 struct ata_eh_context *ehc = &ap->eh_context;
1420 struct ata_device *dev;
1421 int i, rc = 0;
1422
1423 DPRINTK("ENTER\n");
1424
1425 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1426 dev = &ap->device[i];
1427
1428 if (ehc->i.action & ATA_EH_REVALIDATE && ata_dev_enabled(dev) &&
1429 (!ehc->i.dev || ehc->i.dev == dev)) {
1430 if (ata_port_offline(ap)) {
1431 rc = -EIO;
1432 break;
1433 }
1434
1435 ata_eh_about_to_do(ap, ATA_EH_REVALIDATE);
1436 rc = ata_dev_revalidate(dev,
1437 ehc->i.flags & ATA_EHI_DID_RESET);
1438 if (rc)
1439 break;
1440
1441 ehc->i.action &= ~ATA_EH_REVALIDATE;
1442 }
1443 }
1444
1445 if (rc)
1446 *r_failed_dev = dev;
1447
1448 DPRINTK("EXIT\n");
1449 return rc;
1450}
1451
1452static int ata_port_nr_enabled(struct ata_port *ap)
1453{
1454 int i, cnt = 0;
1455
1456 for (i = 0; i < ATA_MAX_DEVICES; i++)
1457 if (ata_dev_enabled(&ap->device[i]))
1458 cnt++;
1459 return cnt;
1460}
1461
1462/**
1463 * ata_eh_recover - recover host port after error
1464 * @ap: host port to recover
f5914a46 1465 * @prereset: prereset method (can be NULL)
022bdb07
TH
1466 * @softreset: softreset method (can be NULL)
1467 * @hardreset: hardreset method (can be NULL)
1468 * @postreset: postreset method (can be NULL)
1469 *
1470 * This is the alpha and omega, eum and yang, heart and soul of
1471 * libata exception handling. On entry, actions required to
1472 * recover each devices are recorded in eh_context. This
1473 * function executes all the operations with appropriate retrials
1474 * and fallbacks to resurrect failed devices.
1475 *
1476 * LOCKING:
1477 * Kernel thread context (may sleep).
1478 *
1479 * RETURNS:
1480 * 0 on success, -errno on failure.
1481 */
f5914a46
TH
1482static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
1483 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
022bdb07
TH
1484 ata_postreset_fn_t postreset)
1485{
1486 struct ata_eh_context *ehc = &ap->eh_context;
1487 struct ata_device *dev;
1488 int down_xfermask, i, rc;
1489
1490 DPRINTK("ENTER\n");
1491
1492 /* prep for recovery */
1493 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1494 dev = &ap->device[i];
1495
1496 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
1497 }
1498
1499 retry:
1500 down_xfermask = 0;
1501 rc = 0;
1502
1503 /* skip EH if possible. */
1504 if (!ata_port_nr_enabled(ap) && !(ap->flags & ATA_FLAG_FROZEN))
1505 ehc->i.action = 0;
1506
1507 /* reset */
1508 if (ehc->i.action & ATA_EH_RESET_MASK) {
1509 ata_eh_freeze_port(ap);
1510
f5914a46
TH
1511 rc = ata_eh_reset(ap, prereset, softreset, hardreset,
1512 postreset);
022bdb07
TH
1513 if (rc) {
1514 ata_port_printk(ap, KERN_ERR,
1515 "reset failed, giving up\n");
1516 goto out;
1517 }
1518
1519 ata_eh_thaw_port(ap);
1520 }
1521
1522 /* revalidate existing devices */
1523 rc = ata_eh_revalidate(ap, &dev);
1524 if (rc)
1525 goto dev_fail;
1526
1527 /* configure transfer mode if the port has been reset */
1528 if (ehc->i.flags & ATA_EHI_DID_RESET) {
1529 rc = ata_set_mode(ap, &dev);
1530 if (rc) {
1531 down_xfermask = 1;
1532 goto dev_fail;
1533 }
1534 }
1535
1536 goto out;
1537
1538 dev_fail:
1539 switch (rc) {
1540 case -ENODEV:
1541 case -EINVAL:
1542 ehc->tries[dev->devno] = 0;
1543 break;
1544 case -EIO:
1545 sata_down_spd_limit(ap);
1546 default:
1547 ehc->tries[dev->devno]--;
1548 if (down_xfermask &&
1549 ata_down_xfermask_limit(dev, ehc->tries[dev->devno] == 1))
1550 ehc->tries[dev->devno] = 0;
1551 }
1552
1553 /* disable device if it has used up all its chances */
1554 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno])
1555 ata_dev_disable(dev);
1556
1557 /* soft didn't work? be haaaaard */
1558 if (ehc->i.flags & ATA_EHI_DID_RESET)
1559 ehc->i.action |= ATA_EH_HARDRESET;
1560 else
1561 ehc->i.action |= ATA_EH_SOFTRESET;
1562
1563 if (ata_port_nr_enabled(ap)) {
1564 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
1565 "devices, retrying in 5 secs\n");
1566 ssleep(5);
1567 } else {
1568 /* no device left, repeat fast */
1569 msleep(500);
1570 }
1571
1572 goto retry;
1573
1574 out:
1575 if (rc) {
1576 for (i = 0; i < ATA_MAX_DEVICES; i++)
1577 ata_dev_disable(&ap->device[i]);
1578 }
1579
1580 DPRINTK("EXIT, rc=%d\n", rc);
1581 return rc;
1582}
1583
1584/**
1585 * ata_eh_finish - finish up EH
1586 * @ap: host port to finish EH for
1587 *
1588 * Recovery is complete. Clean up EH states and retry or finish
1589 * failed qcs.
1590 *
1591 * LOCKING:
1592 * None.
1593 */
1594static void ata_eh_finish(struct ata_port *ap)
1595{
1596 int tag;
1597
1598 /* retry or finish qcs */
1599 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1600 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1601
1602 if (!(qc->flags & ATA_QCFLAG_FAILED))
1603 continue;
1604
1605 if (qc->err_mask) {
1606 /* FIXME: Once EH migration is complete,
1607 * generate sense data in this function,
1608 * considering both err_mask and tf.
1609 */
1610 if (qc->err_mask & AC_ERR_INVALID)
1611 ata_eh_qc_complete(qc);
1612 else
1613 ata_eh_qc_retry(qc);
1614 } else {
1615 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1616 ata_eh_qc_complete(qc);
1617 } else {
1618 /* feed zero TF to sense generation */
1619 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
1620 ata_eh_qc_retry(qc);
1621 }
1622 }
1623 }
1624}
1625
1626/**
1627 * ata_do_eh - do standard error handling
1628 * @ap: host port to handle error for
f5914a46 1629 * @prereset: prereset method (can be NULL)
022bdb07
TH
1630 * @softreset: softreset method (can be NULL)
1631 * @hardreset: hardreset method (can be NULL)
1632 * @postreset: postreset method (can be NULL)
1633 *
1634 * Perform standard error handling sequence.
1635 *
1636 * LOCKING:
1637 * Kernel thread context (may sleep).
1638 */
f5914a46
TH
1639void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
1640 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
1641 ata_postreset_fn_t postreset)
022bdb07
TH
1642{
1643 ata_eh_autopsy(ap);
1644 ata_eh_report(ap);
f5914a46 1645 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
022bdb07
TH
1646 ata_eh_finish(ap);
1647}