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