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 | ||
ece1d636 TH |
35 | #include <linux/kernel.h> |
36 | #include <scsi/scsi.h> | |
37 | #include <scsi/scsi_host.h> | |
38 | #include <scsi/scsi_eh.h> | |
39 | #include <scsi/scsi_device.h> | |
40 | #include <scsi/scsi_cmnd.h> | |
c6fd2807 | 41 | #include "../scsi/scsi_transport_api.h" |
ece1d636 TH |
42 | |
43 | #include <linux/libata.h> | |
44 | ||
45 | #include "libata.h" | |
46 | ||
7d47e8d4 TH |
47 | enum { |
48 | ATA_EH_SPDN_NCQ_OFF = (1 << 0), | |
49 | ATA_EH_SPDN_SPEED_DOWN = (1 << 1), | |
50 | ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2), | |
51 | }; | |
52 | ||
ad9e2762 | 53 | static void __ata_port_freeze(struct ata_port *ap); |
720ba126 | 54 | static void ata_eh_finish(struct ata_port *ap); |
6ffa01d8 | 55 | #ifdef CONFIG_PM |
500530f6 TH |
56 | static void ata_eh_handle_port_suspend(struct ata_port *ap); |
57 | static void ata_eh_handle_port_resume(struct ata_port *ap); | |
6ffa01d8 TH |
58 | static int ata_eh_suspend(struct ata_port *ap, |
59 | struct ata_device **r_failed_dev); | |
60 | static void ata_eh_prep_resume(struct ata_port *ap); | |
61 | static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev); | |
62 | #else /* CONFIG_PM */ | |
63 | static void ata_eh_handle_port_suspend(struct ata_port *ap) | |
64 | { } | |
65 | ||
66 | static void ata_eh_handle_port_resume(struct ata_port *ap) | |
67 | { } | |
68 | ||
69 | static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev) | |
70 | { | |
71 | return 0; | |
72 | } | |
73 | ||
74 | static void ata_eh_prep_resume(struct ata_port *ap) | |
75 | { } | |
76 | ||
77 | static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev) | |
78 | { | |
79 | return 0; | |
80 | } | |
81 | #endif /* CONFIG_PM */ | |
ad9e2762 | 82 | |
0c247c55 TH |
83 | static void ata_ering_record(struct ata_ering *ering, int is_io, |
84 | unsigned int err_mask) | |
85 | { | |
86 | struct ata_ering_entry *ent; | |
87 | ||
88 | WARN_ON(!err_mask); | |
89 | ||
90 | ering->cursor++; | |
91 | ering->cursor %= ATA_ERING_SIZE; | |
92 | ||
93 | ent = &ering->ring[ering->cursor]; | |
94 | ent->is_io = is_io; | |
95 | ent->err_mask = err_mask; | |
96 | ent->timestamp = get_jiffies_64(); | |
97 | } | |
98 | ||
7d47e8d4 | 99 | static void ata_ering_clear(struct ata_ering *ering) |
0c247c55 | 100 | { |
7d47e8d4 | 101 | memset(ering, 0, sizeof(*ering)); |
0c247c55 TH |
102 | } |
103 | ||
104 | static int ata_ering_map(struct ata_ering *ering, | |
105 | int (*map_fn)(struct ata_ering_entry *, void *), | |
106 | void *arg) | |
107 | { | |
108 | int idx, rc = 0; | |
109 | struct ata_ering_entry *ent; | |
110 | ||
111 | idx = ering->cursor; | |
112 | do { | |
113 | ent = &ering->ring[idx]; | |
114 | if (!ent->err_mask) | |
115 | break; | |
116 | rc = map_fn(ent, arg); | |
117 | if (rc) | |
118 | break; | |
119 | idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE; | |
120 | } while (idx != ering->cursor); | |
121 | ||
122 | return rc; | |
123 | } | |
124 | ||
64f65ca6 TH |
125 | static unsigned int ata_eh_dev_action(struct ata_device *dev) |
126 | { | |
127 | struct ata_eh_context *ehc = &dev->ap->eh_context; | |
128 | ||
129 | return ehc->i.action | ehc->i.dev_action[dev->devno]; | |
130 | } | |
131 | ||
af181c2d TH |
132 | static void ata_eh_clear_action(struct ata_device *dev, |
133 | struct ata_eh_info *ehi, unsigned int action) | |
134 | { | |
135 | int i; | |
136 | ||
137 | if (!dev) { | |
138 | ehi->action &= ~action; | |
139 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
140 | ehi->dev_action[i] &= ~action; | |
141 | } else { | |
142 | /* doesn't make sense for port-wide EH actions */ | |
143 | WARN_ON(!(action & ATA_EH_PERDEV_MASK)); | |
144 | ||
145 | /* break ehi->action into ehi->dev_action */ | |
146 | if (ehi->action & action) { | |
147 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
148 | ehi->dev_action[i] |= ehi->action & action; | |
149 | ehi->action &= ~action; | |
150 | } | |
151 | ||
152 | /* turn off the specified per-dev action */ | |
153 | ehi->dev_action[dev->devno] &= ~action; | |
154 | } | |
155 | } | |
156 | ||
ece1d636 TH |
157 | /** |
158 | * ata_scsi_timed_out - SCSI layer time out callback | |
159 | * @cmd: timed out SCSI command | |
160 | * | |
161 | * Handles SCSI layer timeout. We race with normal completion of | |
162 | * the qc for @cmd. If the qc is already gone, we lose and let | |
163 | * the scsi command finish (EH_HANDLED). Otherwise, the qc has | |
164 | * timed out and EH should be invoked. Prevent ata_qc_complete() | |
165 | * from finishing it by setting EH_SCHEDULED and return | |
166 | * EH_NOT_HANDLED. | |
167 | * | |
ad9e2762 TH |
168 | * TODO: kill this function once old EH is gone. |
169 | * | |
ece1d636 TH |
170 | * LOCKING: |
171 | * Called from timer context | |
172 | * | |
173 | * RETURNS: | |
174 | * EH_HANDLED or EH_NOT_HANDLED | |
175 | */ | |
176 | enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd) | |
177 | { | |
178 | struct Scsi_Host *host = cmd->device->host; | |
35bb94b1 | 179 | struct ata_port *ap = ata_shost_to_port(host); |
ece1d636 TH |
180 | unsigned long flags; |
181 | struct ata_queued_cmd *qc; | |
ad9e2762 | 182 | enum scsi_eh_timer_return ret; |
ece1d636 TH |
183 | |
184 | DPRINTK("ENTER\n"); | |
185 | ||
ad9e2762 TH |
186 | if (ap->ops->error_handler) { |
187 | ret = EH_NOT_HANDLED; | |
188 | goto out; | |
189 | } | |
190 | ||
191 | ret = EH_HANDLED; | |
ba6a1308 | 192 | spin_lock_irqsave(ap->lock, flags); |
ece1d636 TH |
193 | qc = ata_qc_from_tag(ap, ap->active_tag); |
194 | if (qc) { | |
195 | WARN_ON(qc->scsicmd != cmd); | |
196 | qc->flags |= ATA_QCFLAG_EH_SCHEDULED; | |
197 | qc->err_mask |= AC_ERR_TIMEOUT; | |
198 | ret = EH_NOT_HANDLED; | |
199 | } | |
ba6a1308 | 200 | spin_unlock_irqrestore(ap->lock, flags); |
ece1d636 | 201 | |
ad9e2762 | 202 | out: |
ece1d636 TH |
203 | DPRINTK("EXIT, ret=%d\n", ret); |
204 | return ret; | |
205 | } | |
206 | ||
207 | /** | |
208 | * ata_scsi_error - SCSI layer error handler callback | |
209 | * @host: SCSI host on which error occurred | |
210 | * | |
211 | * Handles SCSI-layer-thrown error events. | |
212 | * | |
213 | * LOCKING: | |
214 | * Inherited from SCSI layer (none, can sleep) | |
215 | * | |
216 | * RETURNS: | |
217 | * Zero. | |
218 | */ | |
381544bb | 219 | void ata_scsi_error(struct Scsi_Host *host) |
ece1d636 | 220 | { |
35bb94b1 | 221 | struct ata_port *ap = ata_shost_to_port(host); |
ad9e2762 TH |
222 | int i, repeat_cnt = ATA_EH_MAX_REPEAT; |
223 | unsigned long flags; | |
ece1d636 TH |
224 | |
225 | DPRINTK("ENTER\n"); | |
226 | ||
ad9e2762 | 227 | /* synchronize with port task */ |
ece1d636 TH |
228 | ata_port_flush_task(ap); |
229 | ||
cca3974e | 230 | /* synchronize with host lock and sort out timeouts */ |
ad9e2762 TH |
231 | |
232 | /* For new EH, all qcs are finished in one of three ways - | |
233 | * normal completion, error completion, and SCSI timeout. | |
234 | * Both cmpletions can race against SCSI timeout. When normal | |
235 | * completion wins, the qc never reaches EH. When error | |
236 | * completion wins, the qc has ATA_QCFLAG_FAILED set. | |
237 | * | |
238 | * When SCSI timeout wins, things are a bit more complex. | |
239 | * Normal or error completion can occur after the timeout but | |
240 | * before this point. In such cases, both types of | |
241 | * completions are honored. A scmd is determined to have | |
242 | * timed out iff its associated qc is active and not failed. | |
243 | */ | |
244 | if (ap->ops->error_handler) { | |
245 | struct scsi_cmnd *scmd, *tmp; | |
246 | int nr_timedout = 0; | |
247 | ||
e30349d2 | 248 | spin_lock_irqsave(ap->lock, flags); |
ad9e2762 TH |
249 | |
250 | list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) { | |
251 | struct ata_queued_cmd *qc; | |
252 | ||
253 | for (i = 0; i < ATA_MAX_QUEUE; i++) { | |
254 | qc = __ata_qc_from_tag(ap, i); | |
255 | if (qc->flags & ATA_QCFLAG_ACTIVE && | |
256 | qc->scsicmd == scmd) | |
257 | break; | |
258 | } | |
259 | ||
260 | if (i < ATA_MAX_QUEUE) { | |
261 | /* the scmd has an associated qc */ | |
262 | if (!(qc->flags & ATA_QCFLAG_FAILED)) { | |
263 | /* which hasn't failed yet, timeout */ | |
264 | qc->err_mask |= AC_ERR_TIMEOUT; | |
265 | qc->flags |= ATA_QCFLAG_FAILED; | |
266 | nr_timedout++; | |
267 | } | |
268 | } else { | |
269 | /* Normal completion occurred after | |
270 | * SCSI timeout but before this point. | |
271 | * Successfully complete it. | |
272 | */ | |
273 | scmd->retries = scmd->allowed; | |
274 | scsi_eh_finish_cmd(scmd, &ap->eh_done_q); | |
275 | } | |
276 | } | |
277 | ||
278 | /* If we have timed out qcs. They belong to EH from | |
279 | * this point but the state of the controller is | |
280 | * unknown. Freeze the port to make sure the IRQ | |
281 | * handler doesn't diddle with those qcs. This must | |
282 | * be done atomically w.r.t. setting QCFLAG_FAILED. | |
283 | */ | |
284 | if (nr_timedout) | |
285 | __ata_port_freeze(ap); | |
286 | ||
e30349d2 | 287 | spin_unlock_irqrestore(ap->lock, flags); |
ad9e2762 | 288 | } else |
e30349d2 | 289 | spin_unlock_wait(ap->lock); |
ad9e2762 TH |
290 | |
291 | repeat: | |
292 | /* invoke error handler */ | |
293 | if (ap->ops->error_handler) { | |
500530f6 TH |
294 | /* process port resume request */ |
295 | ata_eh_handle_port_resume(ap); | |
296 | ||
f3e81b19 | 297 | /* fetch & clear EH info */ |
e30349d2 | 298 | spin_lock_irqsave(ap->lock, flags); |
f3e81b19 TH |
299 | |
300 | memset(&ap->eh_context, 0, sizeof(ap->eh_context)); | |
301 | ap->eh_context.i = ap->eh_info; | |
302 | memset(&ap->eh_info, 0, sizeof(ap->eh_info)); | |
303 | ||
b51e9e5d TH |
304 | ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS; |
305 | ap->pflags &= ~ATA_PFLAG_EH_PENDING; | |
f3e81b19 | 306 | |
e30349d2 | 307 | spin_unlock_irqrestore(ap->lock, flags); |
ad9e2762 | 308 | |
500530f6 TH |
309 | /* invoke EH, skip if unloading or suspended */ |
310 | if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED))) | |
720ba126 TH |
311 | ap->ops->error_handler(ap); |
312 | else | |
313 | ata_eh_finish(ap); | |
ad9e2762 | 314 | |
500530f6 TH |
315 | /* process port suspend request */ |
316 | ata_eh_handle_port_suspend(ap); | |
317 | ||
ad9e2762 TH |
318 | /* Exception might have happend after ->error_handler |
319 | * recovered the port but before this point. Repeat | |
320 | * EH in such case. | |
321 | */ | |
e30349d2 | 322 | spin_lock_irqsave(ap->lock, flags); |
ad9e2762 | 323 | |
b51e9e5d | 324 | if (ap->pflags & ATA_PFLAG_EH_PENDING) { |
ad9e2762 TH |
325 | if (--repeat_cnt) { |
326 | ata_port_printk(ap, KERN_INFO, | |
327 | "EH pending after completion, " | |
328 | "repeating EH (cnt=%d)\n", repeat_cnt); | |
e30349d2 | 329 | spin_unlock_irqrestore(ap->lock, flags); |
ad9e2762 TH |
330 | goto repeat; |
331 | } | |
332 | ata_port_printk(ap, KERN_ERR, "EH pending after %d " | |
333 | "tries, giving up\n", ATA_EH_MAX_REPEAT); | |
334 | } | |
335 | ||
f3e81b19 TH |
336 | /* this run is complete, make sure EH info is clear */ |
337 | memset(&ap->eh_info, 0, sizeof(ap->eh_info)); | |
338 | ||
e30349d2 | 339 | /* Clear host_eh_scheduled while holding ap->lock such |
ad9e2762 TH |
340 | * that if exception occurs after this point but |
341 | * before EH completion, SCSI midlayer will | |
342 | * re-initiate EH. | |
343 | */ | |
344 | host->host_eh_scheduled = 0; | |
345 | ||
e30349d2 | 346 | spin_unlock_irqrestore(ap->lock, flags); |
ad9e2762 TH |
347 | } else { |
348 | WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL); | |
349 | ap->ops->eng_timeout(ap); | |
350 | } | |
ece1d636 | 351 | |
ad9e2762 | 352 | /* finish or retry handled scmd's and clean up */ |
ece1d636 TH |
353 | WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q)); |
354 | ||
355 | scsi_eh_flush_done_q(&ap->eh_done_q); | |
356 | ||
ad9e2762 | 357 | /* clean up */ |
e30349d2 | 358 | spin_lock_irqsave(ap->lock, flags); |
ad9e2762 | 359 | |
1cdaf534 | 360 | if (ap->pflags & ATA_PFLAG_LOADING) |
b51e9e5d | 361 | ap->pflags &= ~ATA_PFLAG_LOADING; |
1cdaf534 | 362 | else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG) |
52bad64d | 363 | queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0); |
1cdaf534 TH |
364 | |
365 | if (ap->pflags & ATA_PFLAG_RECOVERED) | |
366 | ata_port_printk(ap, KERN_INFO, "EH complete\n"); | |
580b2102 | 367 | |
b51e9e5d | 368 | ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED); |
ad9e2762 | 369 | |
c6cf9e99 | 370 | /* tell wait_eh that we're done */ |
b51e9e5d | 371 | ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS; |
c6cf9e99 TH |
372 | wake_up_all(&ap->eh_wait_q); |
373 | ||
e30349d2 | 374 | spin_unlock_irqrestore(ap->lock, flags); |
ad9e2762 | 375 | |
ece1d636 | 376 | DPRINTK("EXIT\n"); |
ece1d636 TH |
377 | } |
378 | ||
c6cf9e99 TH |
379 | /** |
380 | * ata_port_wait_eh - Wait for the currently pending EH to complete | |
381 | * @ap: Port to wait EH for | |
382 | * | |
383 | * Wait until the currently pending EH is complete. | |
384 | * | |
385 | * LOCKING: | |
386 | * Kernel thread context (may sleep). | |
387 | */ | |
388 | void ata_port_wait_eh(struct ata_port *ap) | |
389 | { | |
390 | unsigned long flags; | |
391 | DEFINE_WAIT(wait); | |
392 | ||
393 | retry: | |
ba6a1308 | 394 | spin_lock_irqsave(ap->lock, flags); |
c6cf9e99 | 395 | |
b51e9e5d | 396 | while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) { |
c6cf9e99 | 397 | prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE); |
ba6a1308 | 398 | spin_unlock_irqrestore(ap->lock, flags); |
c6cf9e99 | 399 | schedule(); |
ba6a1308 | 400 | spin_lock_irqsave(ap->lock, flags); |
c6cf9e99 | 401 | } |
0a1b622e | 402 | finish_wait(&ap->eh_wait_q, &wait); |
c6cf9e99 | 403 | |
ba6a1308 | 404 | spin_unlock_irqrestore(ap->lock, flags); |
c6cf9e99 TH |
405 | |
406 | /* make sure SCSI EH is complete */ | |
cca3974e | 407 | if (scsi_host_in_recovery(ap->scsi_host)) { |
c6cf9e99 TH |
408 | msleep(10); |
409 | goto retry; | |
410 | } | |
411 | } | |
412 | ||
ece1d636 TH |
413 | /** |
414 | * ata_qc_timeout - Handle timeout of queued command | |
415 | * @qc: Command that timed out | |
416 | * | |
417 | * Some part of the kernel (currently, only the SCSI layer) | |
418 | * has noticed that the active command on port @ap has not | |
419 | * completed after a specified length of time. Handle this | |
420 | * condition by disabling DMA (if necessary) and completing | |
421 | * transactions, with error if necessary. | |
422 | * | |
423 | * This also handles the case of the "lost interrupt", where | |
424 | * for some reason (possibly hardware bug, possibly driver bug) | |
425 | * an interrupt was not delivered to the driver, even though the | |
426 | * transaction completed successfully. | |
427 | * | |
ad9e2762 TH |
428 | * TODO: kill this function once old EH is gone. |
429 | * | |
ece1d636 TH |
430 | * LOCKING: |
431 | * Inherited from SCSI layer (none, can sleep) | |
432 | */ | |
433 | static void ata_qc_timeout(struct ata_queued_cmd *qc) | |
434 | { | |
435 | struct ata_port *ap = qc->ap; | |
ece1d636 TH |
436 | u8 host_stat = 0, drv_stat; |
437 | unsigned long flags; | |
438 | ||
439 | DPRINTK("ENTER\n"); | |
440 | ||
441 | ap->hsm_task_state = HSM_ST_IDLE; | |
442 | ||
ba6a1308 | 443 | spin_lock_irqsave(ap->lock, flags); |
ece1d636 TH |
444 | |
445 | switch (qc->tf.protocol) { | |
446 | ||
447 | case ATA_PROT_DMA: | |
448 | case ATA_PROT_ATAPI_DMA: | |
449 | host_stat = ap->ops->bmdma_status(ap); | |
450 | ||
451 | /* before we do anything else, clear DMA-Start bit */ | |
452 | ap->ops->bmdma_stop(qc); | |
453 | ||
454 | /* fall through */ | |
455 | ||
456 | default: | |
457 | ata_altstatus(ap); | |
458 | drv_stat = ata_chk_status(ap); | |
459 | ||
460 | /* ack bmdma irq events */ | |
461 | ap->ops->irq_clear(ap); | |
462 | ||
f15a1daf TH |
463 | ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, " |
464 | "stat 0x%x host_stat 0x%x\n", | |
465 | qc->tf.command, drv_stat, host_stat); | |
ece1d636 TH |
466 | |
467 | /* complete taskfile transaction */ | |
c13b56a1 | 468 | qc->err_mask |= AC_ERR_TIMEOUT; |
ece1d636 TH |
469 | break; |
470 | } | |
471 | ||
ba6a1308 | 472 | spin_unlock_irqrestore(ap->lock, flags); |
ece1d636 TH |
473 | |
474 | ata_eh_qc_complete(qc); | |
475 | ||
476 | DPRINTK("EXIT\n"); | |
477 | } | |
478 | ||
479 | /** | |
480 | * ata_eng_timeout - Handle timeout of queued command | |
481 | * @ap: Port on which timed-out command is active | |
482 | * | |
483 | * Some part of the kernel (currently, only the SCSI layer) | |
484 | * has noticed that the active command on port @ap has not | |
485 | * completed after a specified length of time. Handle this | |
486 | * condition by disabling DMA (if necessary) and completing | |
487 | * transactions, with error if necessary. | |
488 | * | |
489 | * This also handles the case of the "lost interrupt", where | |
490 | * for some reason (possibly hardware bug, possibly driver bug) | |
491 | * an interrupt was not delivered to the driver, even though the | |
492 | * transaction completed successfully. | |
493 | * | |
ad9e2762 TH |
494 | * TODO: kill this function once old EH is gone. |
495 | * | |
ece1d636 TH |
496 | * LOCKING: |
497 | * Inherited from SCSI layer (none, can sleep) | |
498 | */ | |
499 | void ata_eng_timeout(struct ata_port *ap) | |
500 | { | |
501 | DPRINTK("ENTER\n"); | |
502 | ||
503 | ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag)); | |
504 | ||
505 | DPRINTK("EXIT\n"); | |
506 | } | |
507 | ||
f686bcb8 TH |
508 | /** |
509 | * ata_qc_schedule_eh - schedule qc for error handling | |
510 | * @qc: command to schedule error handling for | |
511 | * | |
512 | * Schedule error handling for @qc. EH will kick in as soon as | |
513 | * other commands are drained. | |
514 | * | |
515 | * LOCKING: | |
cca3974e | 516 | * spin_lock_irqsave(host lock) |
f686bcb8 TH |
517 | */ |
518 | void ata_qc_schedule_eh(struct ata_queued_cmd *qc) | |
519 | { | |
520 | struct ata_port *ap = qc->ap; | |
521 | ||
522 | WARN_ON(!ap->ops->error_handler); | |
523 | ||
524 | qc->flags |= ATA_QCFLAG_FAILED; | |
b51e9e5d | 525 | qc->ap->pflags |= ATA_PFLAG_EH_PENDING; |
f686bcb8 TH |
526 | |
527 | /* The following will fail if timeout has already expired. | |
528 | * ata_scsi_error() takes care of such scmds on EH entry. | |
529 | * Note that ATA_QCFLAG_FAILED is unconditionally set after | |
530 | * this function completes. | |
531 | */ | |
532 | scsi_req_abort_cmd(qc->scsicmd); | |
533 | } | |
534 | ||
7b70fc03 TH |
535 | /** |
536 | * ata_port_schedule_eh - schedule error handling without a qc | |
537 | * @ap: ATA port to schedule EH for | |
538 | * | |
539 | * Schedule error handling for @ap. EH will kick in as soon as | |
540 | * all commands are drained. | |
541 | * | |
542 | * LOCKING: | |
cca3974e | 543 | * spin_lock_irqsave(host lock) |
7b70fc03 TH |
544 | */ |
545 | void ata_port_schedule_eh(struct ata_port *ap) | |
546 | { | |
547 | WARN_ON(!ap->ops->error_handler); | |
548 | ||
b51e9e5d | 549 | ap->pflags |= ATA_PFLAG_EH_PENDING; |
cca3974e | 550 | scsi_schedule_eh(ap->scsi_host); |
7b70fc03 TH |
551 | |
552 | DPRINTK("port EH scheduled\n"); | |
553 | } | |
554 | ||
555 | /** | |
556 | * ata_port_abort - abort all qc's on the port | |
557 | * @ap: ATA port to abort qc's for | |
558 | * | |
559 | * Abort all active qc's of @ap and schedule EH. | |
560 | * | |
561 | * LOCKING: | |
cca3974e | 562 | * spin_lock_irqsave(host lock) |
7b70fc03 TH |
563 | * |
564 | * RETURNS: | |
565 | * Number of aborted qc's. | |
566 | */ | |
567 | int ata_port_abort(struct ata_port *ap) | |
568 | { | |
569 | int tag, nr_aborted = 0; | |
570 | ||
571 | WARN_ON(!ap->ops->error_handler); | |
572 | ||
573 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { | |
574 | struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag); | |
575 | ||
576 | if (qc) { | |
577 | qc->flags |= ATA_QCFLAG_FAILED; | |
578 | ata_qc_complete(qc); | |
579 | nr_aborted++; | |
580 | } | |
581 | } | |
582 | ||
583 | if (!nr_aborted) | |
584 | ata_port_schedule_eh(ap); | |
585 | ||
586 | return nr_aborted; | |
587 | } | |
588 | ||
e3180499 TH |
589 | /** |
590 | * __ata_port_freeze - freeze port | |
591 | * @ap: ATA port to freeze | |
592 | * | |
593 | * This function is called when HSM violation or some other | |
594 | * condition disrupts normal operation of the port. Frozen port | |
595 | * is not allowed to perform any operation until the port is | |
596 | * thawed, which usually follows a successful reset. | |
597 | * | |
598 | * ap->ops->freeze() callback can be used for freezing the port | |
599 | * hardware-wise (e.g. mask interrupt and stop DMA engine). If a | |
600 | * port cannot be frozen hardware-wise, the interrupt handler | |
601 | * must ack and clear interrupts unconditionally while the port | |
602 | * is frozen. | |
603 | * | |
604 | * LOCKING: | |
cca3974e | 605 | * spin_lock_irqsave(host lock) |
e3180499 TH |
606 | */ |
607 | static void __ata_port_freeze(struct ata_port *ap) | |
608 | { | |
609 | WARN_ON(!ap->ops->error_handler); | |
610 | ||
611 | if (ap->ops->freeze) | |
612 | ap->ops->freeze(ap); | |
613 | ||
b51e9e5d | 614 | ap->pflags |= ATA_PFLAG_FROZEN; |
e3180499 | 615 | |
44877b4e | 616 | DPRINTK("ata%u port frozen\n", ap->print_id); |
e3180499 TH |
617 | } |
618 | ||
619 | /** | |
620 | * ata_port_freeze - abort & freeze port | |
621 | * @ap: ATA port to freeze | |
622 | * | |
623 | * Abort and freeze @ap. | |
624 | * | |
625 | * LOCKING: | |
cca3974e | 626 | * spin_lock_irqsave(host lock) |
e3180499 TH |
627 | * |
628 | * RETURNS: | |
629 | * Number of aborted commands. | |
630 | */ | |
631 | int ata_port_freeze(struct ata_port *ap) | |
632 | { | |
633 | int nr_aborted; | |
634 | ||
635 | WARN_ON(!ap->ops->error_handler); | |
636 | ||
637 | nr_aborted = ata_port_abort(ap); | |
638 | __ata_port_freeze(ap); | |
639 | ||
640 | return nr_aborted; | |
641 | } | |
642 | ||
643 | /** | |
644 | * ata_eh_freeze_port - EH helper to freeze port | |
645 | * @ap: ATA port to freeze | |
646 | * | |
647 | * Freeze @ap. | |
648 | * | |
649 | * LOCKING: | |
650 | * None. | |
651 | */ | |
652 | void ata_eh_freeze_port(struct ata_port *ap) | |
653 | { | |
654 | unsigned long flags; | |
655 | ||
656 | if (!ap->ops->error_handler) | |
657 | return; | |
658 | ||
ba6a1308 | 659 | spin_lock_irqsave(ap->lock, flags); |
e3180499 | 660 | __ata_port_freeze(ap); |
ba6a1308 | 661 | spin_unlock_irqrestore(ap->lock, flags); |
e3180499 TH |
662 | } |
663 | ||
664 | /** | |
665 | * ata_port_thaw_port - EH helper to thaw port | |
666 | * @ap: ATA port to thaw | |
667 | * | |
668 | * Thaw frozen port @ap. | |
669 | * | |
670 | * LOCKING: | |
671 | * None. | |
672 | */ | |
673 | void ata_eh_thaw_port(struct ata_port *ap) | |
674 | { | |
675 | unsigned long flags; | |
676 | ||
677 | if (!ap->ops->error_handler) | |
678 | return; | |
679 | ||
ba6a1308 | 680 | spin_lock_irqsave(ap->lock, flags); |
e3180499 | 681 | |
b51e9e5d | 682 | ap->pflags &= ~ATA_PFLAG_FROZEN; |
e3180499 TH |
683 | |
684 | if (ap->ops->thaw) | |
685 | ap->ops->thaw(ap); | |
686 | ||
ba6a1308 | 687 | spin_unlock_irqrestore(ap->lock, flags); |
e3180499 | 688 | |
44877b4e | 689 | DPRINTK("ata%u port thawed\n", ap->print_id); |
e3180499 TH |
690 | } |
691 | ||
ece1d636 TH |
692 | static void ata_eh_scsidone(struct scsi_cmnd *scmd) |
693 | { | |
694 | /* nada */ | |
695 | } | |
696 | ||
697 | static void __ata_eh_qc_complete(struct ata_queued_cmd *qc) | |
698 | { | |
699 | struct ata_port *ap = qc->ap; | |
700 | struct scsi_cmnd *scmd = qc->scsicmd; | |
701 | unsigned long flags; | |
702 | ||
ba6a1308 | 703 | spin_lock_irqsave(ap->lock, flags); |
ece1d636 TH |
704 | qc->scsidone = ata_eh_scsidone; |
705 | __ata_qc_complete(qc); | |
706 | WARN_ON(ata_tag_valid(qc->tag)); | |
ba6a1308 | 707 | spin_unlock_irqrestore(ap->lock, flags); |
ece1d636 TH |
708 | |
709 | scsi_eh_finish_cmd(scmd, &ap->eh_done_q); | |
710 | } | |
711 | ||
712 | /** | |
713 | * ata_eh_qc_complete - Complete an active ATA command from EH | |
714 | * @qc: Command to complete | |
715 | * | |
716 | * Indicate to the mid and upper layers that an ATA command has | |
717 | * completed. To be used from EH. | |
718 | */ | |
719 | void ata_eh_qc_complete(struct ata_queued_cmd *qc) | |
720 | { | |
721 | struct scsi_cmnd *scmd = qc->scsicmd; | |
722 | scmd->retries = scmd->allowed; | |
723 | __ata_eh_qc_complete(qc); | |
724 | } | |
725 | ||
726 | /** | |
727 | * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH | |
728 | * @qc: Command to retry | |
729 | * | |
730 | * Indicate to the mid and upper layers that an ATA command | |
731 | * should be retried. To be used from EH. | |
732 | * | |
733 | * SCSI midlayer limits the number of retries to scmd->allowed. | |
734 | * scmd->retries is decremented for commands which get retried | |
735 | * due to unrelated failures (qc->err_mask is zero). | |
736 | */ | |
737 | void ata_eh_qc_retry(struct ata_queued_cmd *qc) | |
738 | { | |
739 | struct scsi_cmnd *scmd = qc->scsicmd; | |
740 | if (!qc->err_mask && scmd->retries) | |
741 | scmd->retries--; | |
742 | __ata_eh_qc_complete(qc); | |
743 | } | |
022bdb07 | 744 | |
0ea035a3 TH |
745 | /** |
746 | * ata_eh_detach_dev - detach ATA device | |
747 | * @dev: ATA device to detach | |
748 | * | |
749 | * Detach @dev. | |
750 | * | |
751 | * LOCKING: | |
752 | * None. | |
753 | */ | |
754 | static void ata_eh_detach_dev(struct ata_device *dev) | |
755 | { | |
756 | struct ata_port *ap = dev->ap; | |
757 | unsigned long flags; | |
758 | ||
759 | ata_dev_disable(dev); | |
760 | ||
ba6a1308 | 761 | spin_lock_irqsave(ap->lock, flags); |
0ea035a3 TH |
762 | |
763 | dev->flags &= ~ATA_DFLAG_DETACH; | |
764 | ||
765 | if (ata_scsi_offline_dev(dev)) { | |
766 | dev->flags |= ATA_DFLAG_DETACHED; | |
b51e9e5d | 767 | ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; |
0ea035a3 TH |
768 | } |
769 | ||
beb07c1a TH |
770 | /* clear per-dev EH actions */ |
771 | ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK); | |
772 | ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK); | |
773 | ||
ba6a1308 | 774 | spin_unlock_irqrestore(ap->lock, flags); |
0ea035a3 TH |
775 | } |
776 | ||
022bdb07 TH |
777 | /** |
778 | * ata_eh_about_to_do - about to perform eh_action | |
779 | * @ap: target ATA port | |
47005f25 | 780 | * @dev: target ATA dev for per-dev action (can be NULL) |
022bdb07 TH |
781 | * @action: action about to be performed |
782 | * | |
783 | * Called just before performing EH actions to clear related bits | |
784 | * in @ap->eh_info such that eh actions are not unnecessarily | |
785 | * repeated. | |
786 | * | |
787 | * LOCKING: | |
788 | * None. | |
789 | */ | |
47005f25 TH |
790 | static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev, |
791 | unsigned int action) | |
022bdb07 TH |
792 | { |
793 | unsigned long flags; | |
13abf50d TH |
794 | struct ata_eh_info *ehi = &ap->eh_info; |
795 | struct ata_eh_context *ehc = &ap->eh_context; | |
022bdb07 | 796 | |
ba6a1308 | 797 | spin_lock_irqsave(ap->lock, flags); |
1cdaf534 | 798 | |
13abf50d TH |
799 | /* Reset is represented by combination of actions and EHI |
800 | * flags. Suck in all related bits before clearing eh_info to | |
801 | * avoid losing requested action. | |
802 | */ | |
803 | if (action & ATA_EH_RESET_MASK) { | |
804 | ehc->i.action |= ehi->action & ATA_EH_RESET_MASK; | |
805 | ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK; | |
806 | ||
807 | /* make sure all reset actions are cleared & clear EHI flags */ | |
808 | action |= ATA_EH_RESET_MASK; | |
809 | ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK; | |
810 | } | |
811 | ||
812 | ata_eh_clear_action(dev, ehi, action); | |
1cdaf534 | 813 | |
13abf50d | 814 | if (!(ehc->i.flags & ATA_EHI_QUIET)) |
1cdaf534 TH |
815 | ap->pflags |= ATA_PFLAG_RECOVERED; |
816 | ||
ba6a1308 | 817 | spin_unlock_irqrestore(ap->lock, flags); |
022bdb07 TH |
818 | } |
819 | ||
47005f25 TH |
820 | /** |
821 | * ata_eh_done - EH action complete | |
822 | * @ap: target ATA port | |
823 | * @dev: target ATA dev for per-dev action (can be NULL) | |
824 | * @action: action just completed | |
825 | * | |
826 | * Called right after performing EH actions to clear related bits | |
827 | * in @ap->eh_context. | |
828 | * | |
829 | * LOCKING: | |
830 | * None. | |
831 | */ | |
832 | static void ata_eh_done(struct ata_port *ap, struct ata_device *dev, | |
833 | unsigned int action) | |
834 | { | |
13abf50d TH |
835 | /* if reset is complete, clear all reset actions & reset modifier */ |
836 | if (action & ATA_EH_RESET_MASK) { | |
837 | action |= ATA_EH_RESET_MASK; | |
838 | ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK; | |
839 | } | |
840 | ||
47005f25 TH |
841 | ata_eh_clear_action(dev, &ap->eh_context.i, action); |
842 | } | |
843 | ||
022bdb07 TH |
844 | /** |
845 | * ata_err_string - convert err_mask to descriptive string | |
846 | * @err_mask: error mask to convert to string | |
847 | * | |
848 | * Convert @err_mask to descriptive string. Errors are | |
849 | * prioritized according to severity and only the most severe | |
850 | * error is reported. | |
851 | * | |
852 | * LOCKING: | |
853 | * None. | |
854 | * | |
855 | * RETURNS: | |
856 | * Descriptive string for @err_mask | |
857 | */ | |
858 | static const char * ata_err_string(unsigned int err_mask) | |
859 | { | |
860 | if (err_mask & AC_ERR_HOST_BUS) | |
861 | return "host bus error"; | |
862 | if (err_mask & AC_ERR_ATA_BUS) | |
863 | return "ATA bus error"; | |
864 | if (err_mask & AC_ERR_TIMEOUT) | |
865 | return "timeout"; | |
866 | if (err_mask & AC_ERR_HSM) | |
867 | return "HSM violation"; | |
868 | if (err_mask & AC_ERR_SYSTEM) | |
869 | return "internal error"; | |
870 | if (err_mask & AC_ERR_MEDIA) | |
871 | return "media error"; | |
872 | if (err_mask & AC_ERR_INVALID) | |
873 | return "invalid argument"; | |
874 | if (err_mask & AC_ERR_DEV) | |
875 | return "device error"; | |
876 | return "unknown error"; | |
877 | } | |
878 | ||
e8ee8451 TH |
879 | /** |
880 | * ata_read_log_page - read a specific log page | |
881 | * @dev: target device | |
882 | * @page: page to read | |
883 | * @buf: buffer to store read page | |
884 | * @sectors: number of sectors to read | |
885 | * | |
886 | * Read log page using READ_LOG_EXT command. | |
887 | * | |
888 | * LOCKING: | |
889 | * Kernel thread context (may sleep). | |
890 | * | |
891 | * RETURNS: | |
892 | * 0 on success, AC_ERR_* mask otherwise. | |
893 | */ | |
894 | static unsigned int ata_read_log_page(struct ata_device *dev, | |
895 | u8 page, void *buf, unsigned int sectors) | |
896 | { | |
897 | struct ata_taskfile tf; | |
898 | unsigned int err_mask; | |
899 | ||
900 | DPRINTK("read log page - page %d\n", page); | |
901 | ||
902 | ata_tf_init(dev, &tf); | |
903 | tf.command = ATA_CMD_READ_LOG_EXT; | |
904 | tf.lbal = page; | |
905 | tf.nsect = sectors; | |
906 | tf.hob_nsect = sectors >> 8; | |
907 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE; | |
908 | tf.protocol = ATA_PROT_PIO; | |
909 | ||
910 | err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE, | |
911 | buf, sectors * ATA_SECT_SIZE); | |
912 | ||
913 | DPRINTK("EXIT, err_mask=%x\n", err_mask); | |
914 | return err_mask; | |
915 | } | |
916 | ||
917 | /** | |
918 | * ata_eh_read_log_10h - Read log page 10h for NCQ error details | |
919 | * @dev: Device to read log page 10h from | |
920 | * @tag: Resulting tag of the failed command | |
921 | * @tf: Resulting taskfile registers of the failed command | |
922 | * | |
923 | * Read log page 10h to obtain NCQ error details and clear error | |
924 | * condition. | |
925 | * | |
926 | * LOCKING: | |
927 | * Kernel thread context (may sleep). | |
928 | * | |
929 | * RETURNS: | |
930 | * 0 on success, -errno otherwise. | |
931 | */ | |
932 | static int ata_eh_read_log_10h(struct ata_device *dev, | |
933 | int *tag, struct ata_taskfile *tf) | |
934 | { | |
935 | u8 *buf = dev->ap->sector_buf; | |
936 | unsigned int err_mask; | |
937 | u8 csum; | |
938 | int i; | |
939 | ||
940 | err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1); | |
941 | if (err_mask) | |
942 | return -EIO; | |
943 | ||
944 | csum = 0; | |
945 | for (i = 0; i < ATA_SECT_SIZE; i++) | |
946 | csum += buf[i]; | |
947 | if (csum) | |
948 | ata_dev_printk(dev, KERN_WARNING, | |
949 | "invalid checksum 0x%x on log page 10h\n", csum); | |
950 | ||
951 | if (buf[0] & 0x80) | |
952 | return -ENOENT; | |
953 | ||
954 | *tag = buf[0] & 0x1f; | |
955 | ||
956 | tf->command = buf[2]; | |
957 | tf->feature = buf[3]; | |
958 | tf->lbal = buf[4]; | |
959 | tf->lbam = buf[5]; | |
960 | tf->lbah = buf[6]; | |
961 | tf->device = buf[7]; | |
962 | tf->hob_lbal = buf[8]; | |
963 | tf->hob_lbam = buf[9]; | |
964 | tf->hob_lbah = buf[10]; | |
965 | tf->nsect = buf[12]; | |
966 | tf->hob_nsect = buf[13]; | |
967 | ||
968 | return 0; | |
969 | } | |
970 | ||
022bdb07 TH |
971 | /** |
972 | * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE | |
973 | * @dev: device to perform REQUEST_SENSE to | |
974 | * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long) | |
975 | * | |
976 | * Perform ATAPI REQUEST_SENSE after the device reported CHECK | |
977 | * SENSE. This function is EH helper. | |
978 | * | |
979 | * LOCKING: | |
980 | * Kernel thread context (may sleep). | |
981 | * | |
982 | * RETURNS: | |
983 | * 0 on success, AC_ERR_* mask on failure | |
984 | */ | |
56287768 | 985 | static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc) |
022bdb07 | 986 | { |
56287768 AL |
987 | struct ata_device *dev = qc->dev; |
988 | unsigned char *sense_buf = qc->scsicmd->sense_buffer; | |
022bdb07 TH |
989 | struct ata_port *ap = dev->ap; |
990 | struct ata_taskfile tf; | |
991 | u8 cdb[ATAPI_CDB_LEN]; | |
992 | ||
993 | DPRINTK("ATAPI request sense\n"); | |
994 | ||
022bdb07 TH |
995 | /* FIXME: is this needed? */ |
996 | memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE); | |
997 | ||
56287768 AL |
998 | /* initialize sense_buf with the error register, |
999 | * for the case where they are -not- overwritten | |
1000 | */ | |
022bdb07 | 1001 | sense_buf[0] = 0x70; |
56287768 AL |
1002 | sense_buf[2] = qc->result_tf.feature >> 4; |
1003 | ||
1004 | /* some devices time out if garbage left in tf */ | |
1005 | ata_tf_init(dev, &tf); | |
022bdb07 TH |
1006 | |
1007 | memset(cdb, 0, ATAPI_CDB_LEN); | |
1008 | cdb[0] = REQUEST_SENSE; | |
1009 | cdb[4] = SCSI_SENSE_BUFFERSIZE; | |
1010 | ||
1011 | tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE; | |
1012 | tf.command = ATA_CMD_PACKET; | |
1013 | ||
1014 | /* is it pointless to prefer PIO for "safety reasons"? */ | |
1015 | if (ap->flags & ATA_FLAG_PIO_DMA) { | |
1016 | tf.protocol = ATA_PROT_ATAPI_DMA; | |
1017 | tf.feature |= ATAPI_PKT_DMA; | |
1018 | } else { | |
1019 | tf.protocol = ATA_PROT_ATAPI; | |
1020 | tf.lbam = (8 * 1024) & 0xff; | |
1021 | tf.lbah = (8 * 1024) >> 8; | |
1022 | } | |
1023 | ||
1024 | return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE, | |
1025 | sense_buf, SCSI_SENSE_BUFFERSIZE); | |
1026 | } | |
1027 | ||
1028 | /** | |
1029 | * ata_eh_analyze_serror - analyze SError for a failed port | |
1030 | * @ap: ATA port to analyze SError for | |
1031 | * | |
1032 | * Analyze SError if available and further determine cause of | |
1033 | * failure. | |
1034 | * | |
1035 | * LOCKING: | |
1036 | * None. | |
1037 | */ | |
1038 | static void ata_eh_analyze_serror(struct ata_port *ap) | |
1039 | { | |
1040 | struct ata_eh_context *ehc = &ap->eh_context; | |
1041 | u32 serror = ehc->i.serror; | |
1042 | unsigned int err_mask = 0, action = 0; | |
1043 | ||
1044 | if (serror & SERR_PERSISTENT) { | |
1045 | err_mask |= AC_ERR_ATA_BUS; | |
1046 | action |= ATA_EH_HARDRESET; | |
1047 | } | |
1048 | if (serror & | |
1049 | (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) { | |
1050 | err_mask |= AC_ERR_ATA_BUS; | |
1051 | action |= ATA_EH_SOFTRESET; | |
1052 | } | |
1053 | if (serror & SERR_PROTOCOL) { | |
1054 | err_mask |= AC_ERR_HSM; | |
1055 | action |= ATA_EH_SOFTRESET; | |
1056 | } | |
1057 | if (serror & SERR_INTERNAL) { | |
1058 | err_mask |= AC_ERR_SYSTEM; | |
1059 | action |= ATA_EH_SOFTRESET; | |
1060 | } | |
084fe639 TH |
1061 | if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG)) |
1062 | ata_ehi_hotplugged(&ehc->i); | |
022bdb07 TH |
1063 | |
1064 | ehc->i.err_mask |= err_mask; | |
1065 | ehc->i.action |= action; | |
1066 | } | |
1067 | ||
e8ee8451 TH |
1068 | /** |
1069 | * ata_eh_analyze_ncq_error - analyze NCQ error | |
1070 | * @ap: ATA port to analyze NCQ error for | |
1071 | * | |
1072 | * Read log page 10h, determine the offending qc and acquire | |
1073 | * error status TF. For NCQ device errors, all LLDDs have to do | |
1074 | * is setting AC_ERR_DEV in ehi->err_mask. This function takes | |
1075 | * care of the rest. | |
1076 | * | |
1077 | * LOCKING: | |
1078 | * Kernel thread context (may sleep). | |
1079 | */ | |
1080 | static void ata_eh_analyze_ncq_error(struct ata_port *ap) | |
1081 | { | |
1082 | struct ata_eh_context *ehc = &ap->eh_context; | |
1083 | struct ata_device *dev = ap->device; | |
1084 | struct ata_queued_cmd *qc; | |
1085 | struct ata_taskfile tf; | |
1086 | int tag, rc; | |
1087 | ||
1088 | /* if frozen, we can't do much */ | |
b51e9e5d | 1089 | if (ap->pflags & ATA_PFLAG_FROZEN) |
e8ee8451 TH |
1090 | return; |
1091 | ||
1092 | /* is it NCQ device error? */ | |
1093 | if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV)) | |
1094 | return; | |
1095 | ||
1096 | /* has LLDD analyzed already? */ | |
1097 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { | |
1098 | qc = __ata_qc_from_tag(ap, tag); | |
1099 | ||
1100 | if (!(qc->flags & ATA_QCFLAG_FAILED)) | |
1101 | continue; | |
1102 | ||
1103 | if (qc->err_mask) | |
1104 | return; | |
1105 | } | |
1106 | ||
1107 | /* okay, this error is ours */ | |
1108 | rc = ata_eh_read_log_10h(dev, &tag, &tf); | |
1109 | if (rc) { | |
1110 | ata_port_printk(ap, KERN_ERR, "failed to read log page 10h " | |
1111 | "(errno=%d)\n", rc); | |
1112 | return; | |
1113 | } | |
1114 | ||
1115 | if (!(ap->sactive & (1 << tag))) { | |
1116 | ata_port_printk(ap, KERN_ERR, "log page 10h reported " | |
1117 | "inactive tag %d\n", tag); | |
1118 | return; | |
1119 | } | |
1120 | ||
1121 | /* we've got the perpetrator, condemn it */ | |
1122 | qc = __ata_qc_from_tag(ap, tag); | |
1123 | memcpy(&qc->result_tf, &tf, sizeof(tf)); | |
1124 | qc->err_mask |= AC_ERR_DEV; | |
1125 | ehc->i.err_mask &= ~AC_ERR_DEV; | |
1126 | } | |
1127 | ||
022bdb07 TH |
1128 | /** |
1129 | * ata_eh_analyze_tf - analyze taskfile of a failed qc | |
1130 | * @qc: qc to analyze | |
1131 | * @tf: Taskfile registers to analyze | |
1132 | * | |
1133 | * Analyze taskfile of @qc and further determine cause of | |
1134 | * failure. This function also requests ATAPI sense data if | |
1135 | * avaliable. | |
1136 | * | |
1137 | * LOCKING: | |
1138 | * Kernel thread context (may sleep). | |
1139 | * | |
1140 | * RETURNS: | |
1141 | * Determined recovery action | |
1142 | */ | |
1143 | static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc, | |
1144 | const struct ata_taskfile *tf) | |
1145 | { | |
1146 | unsigned int tmp, action = 0; | |
1147 | u8 stat = tf->command, err = tf->feature; | |
1148 | ||
1149 | if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) { | |
1150 | qc->err_mask |= AC_ERR_HSM; | |
1151 | return ATA_EH_SOFTRESET; | |
1152 | } | |
1153 | ||
1154 | if (!(qc->err_mask & AC_ERR_DEV)) | |
1155 | return 0; | |
1156 | ||
1157 | switch (qc->dev->class) { | |
1158 | case ATA_DEV_ATA: | |
1159 | if (err & ATA_ICRC) | |
1160 | qc->err_mask |= AC_ERR_ATA_BUS; | |
1161 | if (err & ATA_UNC) | |
1162 | qc->err_mask |= AC_ERR_MEDIA; | |
1163 | if (err & ATA_IDNF) | |
1164 | qc->err_mask |= AC_ERR_INVALID; | |
1165 | break; | |
1166 | ||
1167 | case ATA_DEV_ATAPI: | |
a569a30d | 1168 | if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) { |
56287768 | 1169 | tmp = atapi_eh_request_sense(qc); |
a569a30d TH |
1170 | if (!tmp) { |
1171 | /* ATA_QCFLAG_SENSE_VALID is used to | |
1172 | * tell atapi_qc_complete() that sense | |
1173 | * data is already valid. | |
1174 | * | |
1175 | * TODO: interpret sense data and set | |
1176 | * appropriate err_mask. | |
1177 | */ | |
1178 | qc->flags |= ATA_QCFLAG_SENSE_VALID; | |
1179 | } else | |
1180 | qc->err_mask |= tmp; | |
1181 | } | |
022bdb07 TH |
1182 | } |
1183 | ||
1184 | if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS)) | |
1185 | action |= ATA_EH_SOFTRESET; | |
1186 | ||
1187 | return action; | |
1188 | } | |
1189 | ||
7d47e8d4 | 1190 | static int ata_eh_categorize_error(int is_io, unsigned int err_mask) |
022bdb07 | 1191 | { |
7d47e8d4 | 1192 | if (err_mask & AC_ERR_ATA_BUS) |
022bdb07 TH |
1193 | return 1; |
1194 | ||
7d47e8d4 TH |
1195 | if (err_mask & AC_ERR_TIMEOUT) |
1196 | return 2; | |
1197 | ||
1198 | if (is_io) { | |
1199 | if (err_mask & AC_ERR_HSM) | |
022bdb07 | 1200 | return 2; |
7d47e8d4 TH |
1201 | if ((err_mask & |
1202 | (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV) | |
1203 | return 3; | |
022bdb07 TH |
1204 | } |
1205 | ||
1206 | return 0; | |
1207 | } | |
1208 | ||
7d47e8d4 | 1209 | struct speed_down_verdict_arg { |
022bdb07 | 1210 | u64 since; |
7d47e8d4 | 1211 | int nr_errors[4]; |
022bdb07 TH |
1212 | }; |
1213 | ||
7d47e8d4 | 1214 | static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg) |
022bdb07 | 1215 | { |
7d47e8d4 TH |
1216 | struct speed_down_verdict_arg *arg = void_arg; |
1217 | int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask); | |
022bdb07 TH |
1218 | |
1219 | if (ent->timestamp < arg->since) | |
1220 | return -1; | |
1221 | ||
7d47e8d4 | 1222 | arg->nr_errors[cat]++; |
022bdb07 TH |
1223 | return 0; |
1224 | } | |
1225 | ||
1226 | /** | |
7d47e8d4 | 1227 | * ata_eh_speed_down_verdict - Determine speed down verdict |
022bdb07 TH |
1228 | * @dev: Device of interest |
1229 | * | |
1230 | * This function examines error ring of @dev and determines | |
7d47e8d4 TH |
1231 | * whether NCQ needs to be turned off, transfer speed should be |
1232 | * stepped down, or falling back to PIO is necessary. | |
022bdb07 | 1233 | * |
7d47e8d4 | 1234 | * Cat-1 is ATA_BUS error for any command. |
022bdb07 | 1235 | * |
7d47e8d4 TH |
1236 | * Cat-2 is TIMEOUT for any command or HSM violation for known |
1237 | * supported commands. | |
1238 | * | |
1239 | * Cat-3 is is unclassified DEV error for known supported | |
022bdb07 TH |
1240 | * command. |
1241 | * | |
7d47e8d4 TH |
1242 | * NCQ needs to be turned off if there have been more than 3 |
1243 | * Cat-2 + Cat-3 errors during last 10 minutes. | |
1244 | * | |
1245 | * Speed down is necessary if there have been more than 3 Cat-1 + | |
1246 | * Cat-2 errors or 10 Cat-3 errors during last 10 minutes. | |
1247 | * | |
1248 | * Falling back to PIO mode is necessary if there have been more | |
1249 | * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes. | |
1250 | * | |
022bdb07 TH |
1251 | * LOCKING: |
1252 | * Inherited from caller. | |
1253 | * | |
1254 | * RETURNS: | |
7d47e8d4 | 1255 | * OR of ATA_EH_SPDN_* flags. |
022bdb07 | 1256 | */ |
7d47e8d4 | 1257 | static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev) |
022bdb07 | 1258 | { |
7d47e8d4 TH |
1259 | const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ; |
1260 | u64 j64 = get_jiffies_64(); | |
1261 | struct speed_down_verdict_arg arg; | |
1262 | unsigned int verdict = 0; | |
022bdb07 | 1263 | |
7d47e8d4 TH |
1264 | /* scan past 10 mins of error history */ |
1265 | memset(&arg, 0, sizeof(arg)); | |
1266 | arg.since = j64 - min(j64, j10mins); | |
1267 | ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg); | |
022bdb07 | 1268 | |
7d47e8d4 TH |
1269 | if (arg.nr_errors[2] + arg.nr_errors[3] > 3) |
1270 | verdict |= ATA_EH_SPDN_NCQ_OFF; | |
1271 | if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10) | |
1272 | verdict |= ATA_EH_SPDN_SPEED_DOWN; | |
022bdb07 | 1273 | |
7d47e8d4 | 1274 | /* scan past 3 mins of error history */ |
022bdb07 | 1275 | memset(&arg, 0, sizeof(arg)); |
7d47e8d4 TH |
1276 | arg.since = j64 - min(j64, j5mins); |
1277 | ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg); | |
022bdb07 | 1278 | |
7d47e8d4 TH |
1279 | if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10) |
1280 | verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO; | |
022bdb07 | 1281 | |
7d47e8d4 | 1282 | return verdict; |
022bdb07 TH |
1283 | } |
1284 | ||
1285 | /** | |
1286 | * ata_eh_speed_down - record error and speed down if necessary | |
1287 | * @dev: Failed device | |
1288 | * @is_io: Did the device fail during normal IO? | |
1289 | * @err_mask: err_mask of the error | |
1290 | * | |
1291 | * Record error and examine error history to determine whether | |
1292 | * adjusting transmission speed is necessary. It also sets | |
1293 | * transmission limits appropriately if such adjustment is | |
1294 | * necessary. | |
1295 | * | |
1296 | * LOCKING: | |
1297 | * Kernel thread context (may sleep). | |
1298 | * | |
1299 | * RETURNS: | |
7d47e8d4 | 1300 | * Determined recovery action. |
022bdb07 | 1301 | */ |
7d47e8d4 TH |
1302 | static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io, |
1303 | unsigned int err_mask) | |
022bdb07 | 1304 | { |
7d47e8d4 TH |
1305 | unsigned int verdict; |
1306 | unsigned int action = 0; | |
1307 | ||
1308 | /* don't bother if Cat-0 error */ | |
1309 | if (ata_eh_categorize_error(is_io, err_mask) == 0) | |
022bdb07 TH |
1310 | return 0; |
1311 | ||
1312 | /* record error and determine whether speed down is necessary */ | |
1313 | ata_ering_record(&dev->ering, is_io, err_mask); | |
7d47e8d4 | 1314 | verdict = ata_eh_speed_down_verdict(dev); |
022bdb07 | 1315 | |
7d47e8d4 TH |
1316 | /* turn off NCQ? */ |
1317 | if ((verdict & ATA_EH_SPDN_NCQ_OFF) && | |
1318 | (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ | | |
1319 | ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) { | |
1320 | dev->flags |= ATA_DFLAG_NCQ_OFF; | |
1321 | ata_dev_printk(dev, KERN_WARNING, | |
1322 | "NCQ disabled due to excessive errors\n"); | |
1323 | goto done; | |
1324 | } | |
022bdb07 | 1325 | |
7d47e8d4 TH |
1326 | /* speed down? */ |
1327 | if (verdict & ATA_EH_SPDN_SPEED_DOWN) { | |
1328 | /* speed down SATA link speed if possible */ | |
1329 | if (sata_down_spd_limit(dev->ap) == 0) { | |
1330 | action |= ATA_EH_HARDRESET; | |
1331 | goto done; | |
1332 | } | |
022bdb07 | 1333 | |
7d47e8d4 TH |
1334 | /* lower transfer mode */ |
1335 | if (dev->spdn_cnt < 2) { | |
1336 | static const int dma_dnxfer_sel[] = | |
1337 | { ATA_DNXFER_DMA, ATA_DNXFER_40C }; | |
1338 | static const int pio_dnxfer_sel[] = | |
1339 | { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 }; | |
1340 | int sel; | |
1341 | ||
1342 | if (dev->xfer_shift != ATA_SHIFT_PIO) | |
1343 | sel = dma_dnxfer_sel[dev->spdn_cnt]; | |
1344 | else | |
1345 | sel = pio_dnxfer_sel[dev->spdn_cnt]; | |
1346 | ||
1347 | dev->spdn_cnt++; | |
1348 | ||
1349 | if (ata_down_xfermask_limit(dev, sel) == 0) { | |
1350 | action |= ATA_EH_SOFTRESET; | |
1351 | goto done; | |
1352 | } | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | /* Fall back to PIO? Slowing down to PIO is meaningless for | |
1357 | * SATA. Consider it only for PATA. | |
1358 | */ | |
1359 | if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) && | |
1360 | (dev->ap->cbl != ATA_CBL_SATA) && | |
1361 | (dev->xfer_shift != ATA_SHIFT_PIO)) { | |
1362 | if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) { | |
1363 | dev->spdn_cnt = 0; | |
1364 | action |= ATA_EH_SOFTRESET; | |
1365 | goto done; | |
1366 | } | |
1367 | } | |
022bdb07 | 1368 | |
022bdb07 | 1369 | return 0; |
7d47e8d4 TH |
1370 | done: |
1371 | /* device has been slowed down, blow error history */ | |
1372 | ata_ering_clear(&dev->ering); | |
1373 | return action; | |
022bdb07 TH |
1374 | } |
1375 | ||
1376 | /** | |
1377 | * ata_eh_autopsy - analyze error and determine recovery action | |
1378 | * @ap: ATA port to perform autopsy on | |
1379 | * | |
1380 | * Analyze why @ap failed and determine which recovery action is | |
1381 | * needed. This function also sets more detailed AC_ERR_* values | |
1382 | * and fills sense data for ATAPI CHECK SENSE. | |
1383 | * | |
1384 | * LOCKING: | |
1385 | * Kernel thread context (may sleep). | |
1386 | */ | |
1387 | static void ata_eh_autopsy(struct ata_port *ap) | |
1388 | { | |
1389 | struct ata_eh_context *ehc = &ap->eh_context; | |
022bdb07 TH |
1390 | unsigned int all_err_mask = 0; |
1391 | int tag, is_io = 0; | |
1392 | u32 serror; | |
1393 | int rc; | |
1394 | ||
1395 | DPRINTK("ENTER\n"); | |
1396 | ||
1cdaf534 TH |
1397 | if (ehc->i.flags & ATA_EHI_NO_AUTOPSY) |
1398 | return; | |
1399 | ||
022bdb07 TH |
1400 | /* obtain and analyze SError */ |
1401 | rc = sata_scr_read(ap, SCR_ERROR, &serror); | |
1402 | if (rc == 0) { | |
1403 | ehc->i.serror |= serror; | |
1404 | ata_eh_analyze_serror(ap); | |
1405 | } else if (rc != -EOPNOTSUPP) | |
4528e4da | 1406 | ehc->i.action |= ATA_EH_HARDRESET; |
022bdb07 | 1407 | |
e8ee8451 TH |
1408 | /* analyze NCQ failure */ |
1409 | ata_eh_analyze_ncq_error(ap); | |
1410 | ||
022bdb07 TH |
1411 | /* any real error trumps AC_ERR_OTHER */ |
1412 | if (ehc->i.err_mask & ~AC_ERR_OTHER) | |
1413 | ehc->i.err_mask &= ~AC_ERR_OTHER; | |
1414 | ||
1415 | all_err_mask |= ehc->i.err_mask; | |
1416 | ||
1417 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { | |
1418 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); | |
1419 | ||
1420 | if (!(qc->flags & ATA_QCFLAG_FAILED)) | |
1421 | continue; | |
1422 | ||
1423 | /* inherit upper level err_mask */ | |
1424 | qc->err_mask |= ehc->i.err_mask; | |
1425 | ||
022bdb07 | 1426 | /* analyze TF */ |
4528e4da | 1427 | ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf); |
022bdb07 TH |
1428 | |
1429 | /* DEV errors are probably spurious in case of ATA_BUS error */ | |
1430 | if (qc->err_mask & AC_ERR_ATA_BUS) | |
1431 | qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA | | |
1432 | AC_ERR_INVALID); | |
1433 | ||
1434 | /* any real error trumps unknown error */ | |
1435 | if (qc->err_mask & ~AC_ERR_OTHER) | |
1436 | qc->err_mask &= ~AC_ERR_OTHER; | |
1437 | ||
1438 | /* SENSE_VALID trumps dev/unknown error and revalidation */ | |
1439 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) { | |
1440 | qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER); | |
4528e4da | 1441 | ehc->i.action &= ~ATA_EH_REVALIDATE; |
022bdb07 TH |
1442 | } |
1443 | ||
1444 | /* accumulate error info */ | |
4528e4da | 1445 | ehc->i.dev = qc->dev; |
022bdb07 TH |
1446 | all_err_mask |= qc->err_mask; |
1447 | if (qc->flags & ATA_QCFLAG_IO) | |
1448 | is_io = 1; | |
1449 | } | |
1450 | ||
a20f33ff | 1451 | /* enforce default EH actions */ |
b51e9e5d | 1452 | if (ap->pflags & ATA_PFLAG_FROZEN || |
a20f33ff | 1453 | all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT)) |
4528e4da | 1454 | ehc->i.action |= ATA_EH_SOFTRESET; |
a20f33ff | 1455 | else if (all_err_mask) |
4528e4da | 1456 | ehc->i.action |= ATA_EH_REVALIDATE; |
022bdb07 | 1457 | |
47005f25 | 1458 | /* if we have offending qcs and the associated failed device */ |
4528e4da | 1459 | if (ehc->i.dev) { |
47005f25 | 1460 | /* speed down */ |
4528e4da TH |
1461 | ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io, |
1462 | all_err_mask); | |
47005f25 TH |
1463 | |
1464 | /* perform per-dev EH action only on the offending device */ | |
4528e4da TH |
1465 | ehc->i.dev_action[ehc->i.dev->devno] |= |
1466 | ehc->i.action & ATA_EH_PERDEV_MASK; | |
1467 | ehc->i.action &= ~ATA_EH_PERDEV_MASK; | |
47005f25 TH |
1468 | } |
1469 | ||
022bdb07 TH |
1470 | DPRINTK("EXIT\n"); |
1471 | } | |
1472 | ||
1473 | /** | |
1474 | * ata_eh_report - report error handling to user | |
1475 | * @ap: ATA port EH is going on | |
1476 | * | |
1477 | * Report EH to user. | |
1478 | * | |
1479 | * LOCKING: | |
1480 | * None. | |
1481 | */ | |
1482 | static void ata_eh_report(struct ata_port *ap) | |
1483 | { | |
1484 | struct ata_eh_context *ehc = &ap->eh_context; | |
1485 | const char *frozen, *desc; | |
1486 | int tag, nr_failed = 0; | |
1487 | ||
1488 | desc = NULL; | |
1489 | if (ehc->i.desc[0] != '\0') | |
1490 | desc = ehc->i.desc; | |
1491 | ||
1492 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { | |
1493 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); | |
1494 | ||
1495 | if (!(qc->flags & ATA_QCFLAG_FAILED)) | |
1496 | continue; | |
1497 | if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask) | |
1498 | continue; | |
1499 | ||
1500 | nr_failed++; | |
1501 | } | |
1502 | ||
1503 | if (!nr_failed && !ehc->i.err_mask) | |
1504 | return; | |
1505 | ||
1506 | frozen = ""; | |
b51e9e5d | 1507 | if (ap->pflags & ATA_PFLAG_FROZEN) |
022bdb07 TH |
1508 | frozen = " frozen"; |
1509 | ||
1510 | if (ehc->i.dev) { | |
e8ee8451 TH |
1511 | ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x " |
1512 | "SAct 0x%x SErr 0x%x action 0x%x%s\n", | |
1513 | ehc->i.err_mask, ap->sactive, ehc->i.serror, | |
1514 | ehc->i.action, frozen); | |
022bdb07 TH |
1515 | if (desc) |
1516 | ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc); | |
1517 | } else { | |
e8ee8451 TH |
1518 | ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x " |
1519 | "SAct 0x%x SErr 0x%x action 0x%x%s\n", | |
1520 | ehc->i.err_mask, ap->sactive, ehc->i.serror, | |
1521 | ehc->i.action, frozen); | |
022bdb07 TH |
1522 | if (desc) |
1523 | ata_port_printk(ap, KERN_ERR, "(%s)\n", desc); | |
1524 | } | |
1525 | ||
1526 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { | |
8a937581 TH |
1527 | static const char *dma_str[] = { |
1528 | [DMA_BIDIRECTIONAL] = "bidi", | |
1529 | [DMA_TO_DEVICE] = "out", | |
1530 | [DMA_FROM_DEVICE] = "in", | |
1531 | [DMA_NONE] = "", | |
1532 | }; | |
022bdb07 | 1533 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); |
8a937581 | 1534 | struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf; |
022bdb07 TH |
1535 | |
1536 | if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask) | |
1537 | continue; | |
1538 | ||
8a937581 TH |
1539 | ata_dev_printk(qc->dev, KERN_ERR, |
1540 | "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " | |
664e8503 | 1541 | "tag %d cdb 0x%x data %u %s\n " |
8a937581 TH |
1542 | "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x " |
1543 | "Emask 0x%x (%s)\n", | |
1544 | cmd->command, cmd->feature, cmd->nsect, | |
1545 | cmd->lbal, cmd->lbam, cmd->lbah, | |
1546 | cmd->hob_feature, cmd->hob_nsect, | |
1547 | cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah, | |
726f0785 | 1548 | cmd->device, qc->tag, qc->cdb[0], qc->nbytes, |
664e8503 | 1549 | dma_str[qc->dma_dir], |
8a937581 TH |
1550 | res->command, res->feature, res->nsect, |
1551 | res->lbal, res->lbam, res->lbah, | |
1552 | res->hob_feature, res->hob_nsect, | |
1553 | res->hob_lbal, res->hob_lbam, res->hob_lbah, | |
1554 | res->device, qc->err_mask, ata_err_string(qc->err_mask)); | |
022bdb07 TH |
1555 | } |
1556 | } | |
1557 | ||
d87fa38e TH |
1558 | static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset, |
1559 | unsigned int *classes) | |
1560 | { | |
1561 | int i, rc; | |
1562 | ||
1563 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
1564 | classes[i] = ATA_DEV_UNKNOWN; | |
1565 | ||
1566 | rc = reset(ap, classes); | |
1567 | if (rc) | |
1568 | return rc; | |
1569 | ||
1570 | /* If any class isn't ATA_DEV_UNKNOWN, consider classification | |
1571 | * is complete and convert all ATA_DEV_UNKNOWN to | |
1572 | * ATA_DEV_NONE. | |
1573 | */ | |
1574 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
1575 | if (classes[i] != ATA_DEV_UNKNOWN) | |
1576 | break; | |
1577 | ||
1578 | if (i < ATA_MAX_DEVICES) | |
1579 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
1580 | if (classes[i] == ATA_DEV_UNKNOWN) | |
1581 | classes[i] = ATA_DEV_NONE; | |
1582 | ||
1583 | return 0; | |
1584 | } | |
1585 | ||
664faf09 TH |
1586 | static int ata_eh_followup_srst_needed(int rc, int classify, |
1587 | const unsigned int *classes) | |
1588 | { | |
1589 | if (rc == -EAGAIN) | |
1590 | return 1; | |
1591 | if (rc != 0) | |
1592 | return 0; | |
1593 | if (classify && classes[0] == ATA_DEV_UNKNOWN) | |
1594 | return 1; | |
1595 | return 0; | |
1596 | } | |
1597 | ||
1598 | static int ata_eh_reset(struct ata_port *ap, int classify, | |
f5914a46 | 1599 | ata_prereset_fn_t prereset, ata_reset_fn_t softreset, |
022bdb07 TH |
1600 | ata_reset_fn_t hardreset, ata_postreset_fn_t postreset) |
1601 | { | |
1602 | struct ata_eh_context *ehc = &ap->eh_context; | |
664faf09 | 1603 | unsigned int *classes = ehc->classes; |
022bdb07 | 1604 | int tries = ATA_EH_RESET_TRIES; |
1cdaf534 | 1605 | int verbose = !(ehc->i.flags & ATA_EHI_QUIET); |
f5914a46 | 1606 | unsigned int action; |
022bdb07 | 1607 | ata_reset_fn_t reset; |
664faf09 | 1608 | int i, did_followup_srst, rc; |
022bdb07 | 1609 | |
13abf50d TH |
1610 | /* about to reset */ |
1611 | ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK); | |
1612 | ||
f5914a46 TH |
1613 | /* Determine which reset to use and record in ehc->i.action. |
1614 | * prereset() may examine and modify it. | |
1615 | */ | |
1616 | action = ehc->i.action; | |
1617 | ehc->i.action &= ~ATA_EH_RESET_MASK; | |
022bdb07 | 1618 | if (softreset && (!hardreset || (!sata_set_spd_needed(ap) && |
f5914a46 TH |
1619 | !(action & ATA_EH_HARDRESET)))) |
1620 | ehc->i.action |= ATA_EH_SOFTRESET; | |
022bdb07 | 1621 | else |
f5914a46 TH |
1622 | ehc->i.action |= ATA_EH_HARDRESET; |
1623 | ||
1624 | if (prereset) { | |
1625 | rc = prereset(ap); | |
1626 | if (rc) { | |
c961922b | 1627 | if (rc == -ENOENT) { |
4aa9ab67 TH |
1628 | ata_port_printk(ap, KERN_DEBUG, |
1629 | "port disabled. ignoring.\n"); | |
c961922b | 1630 | ap->eh_context.i.action &= ~ATA_EH_RESET_MASK; |
4aa9ab67 TH |
1631 | |
1632 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
1633 | classes[i] = ATA_DEV_NONE; | |
1634 | ||
1635 | rc = 0; | |
c961922b AC |
1636 | } else |
1637 | ata_port_printk(ap, KERN_ERR, | |
f5914a46 TH |
1638 | "prereset failed (errno=%d)\n", rc); |
1639 | return rc; | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | /* prereset() might have modified ehc->i.action */ | |
1644 | if (ehc->i.action & ATA_EH_HARDRESET) | |
022bdb07 | 1645 | reset = hardreset; |
f5914a46 TH |
1646 | else if (ehc->i.action & ATA_EH_SOFTRESET) |
1647 | reset = softreset; | |
1648 | else { | |
1649 | /* prereset told us not to reset, bang classes and return */ | |
1650 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
1651 | classes[i] = ATA_DEV_NONE; | |
1652 | return 0; | |
1653 | } | |
1654 | ||
1655 | /* did prereset() screw up? if so, fix up to avoid oopsing */ | |
1656 | if (!reset) { | |
1657 | ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested " | |
1658 | "invalid reset type\n"); | |
1659 | if (softreset) | |
1660 | reset = softreset; | |
1661 | else | |
1662 | reset = hardreset; | |
1663 | } | |
022bdb07 TH |
1664 | |
1665 | retry: | |
3e706399 TH |
1666 | /* shut up during boot probing */ |
1667 | if (verbose) | |
1668 | ata_port_printk(ap, KERN_INFO, "%s resetting port\n", | |
1669 | reset == softreset ? "soft" : "hard"); | |
022bdb07 | 1670 | |
13abf50d | 1671 | /* mark that this EH session started with reset */ |
022bdb07 TH |
1672 | ehc->i.flags |= ATA_EHI_DID_RESET; |
1673 | ||
1674 | rc = ata_do_reset(ap, reset, classes); | |
1675 | ||
664faf09 TH |
1676 | did_followup_srst = 0; |
1677 | if (reset == hardreset && | |
1678 | ata_eh_followup_srst_needed(rc, classify, classes)) { | |
1679 | /* okay, let's do follow-up softreset */ | |
1680 | did_followup_srst = 1; | |
1681 | reset = softreset; | |
1682 | ||
1683 | if (!reset) { | |
1684 | ata_port_printk(ap, KERN_ERR, | |
1685 | "follow-up softreset required " | |
1686 | "but no softreset avaliable\n"); | |
1687 | return -EINVAL; | |
1688 | } | |
1689 | ||
47005f25 | 1690 | ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK); |
664faf09 TH |
1691 | rc = ata_do_reset(ap, reset, classes); |
1692 | ||
1693 | if (rc == 0 && classify && | |
1694 | classes[0] == ATA_DEV_UNKNOWN) { | |
1695 | ata_port_printk(ap, KERN_ERR, | |
1696 | "classification failed\n"); | |
1697 | return -EINVAL; | |
1698 | } | |
1699 | } | |
1700 | ||
022bdb07 | 1701 | if (rc && --tries) { |
664faf09 TH |
1702 | const char *type; |
1703 | ||
1704 | if (reset == softreset) { | |
1705 | if (did_followup_srst) | |
1706 | type = "follow-up soft"; | |
1707 | else | |
1708 | type = "soft"; | |
1709 | } else | |
1710 | type = "hard"; | |
1711 | ||
022bdb07 | 1712 | ata_port_printk(ap, KERN_WARNING, |
664faf09 | 1713 | "%sreset failed, retrying in 5 secs\n", type); |
022bdb07 TH |
1714 | ssleep(5); |
1715 | ||
1716 | if (reset == hardreset) | |
1717 | sata_down_spd_limit(ap); | |
1718 | if (hardreset) | |
1719 | reset = hardreset; | |
1720 | goto retry; | |
1721 | } | |
1722 | ||
1723 | if (rc == 0) { | |
20952b69 TH |
1724 | /* After the reset, the device state is PIO 0 and the |
1725 | * controller state is undefined. Record the mode. | |
1726 | */ | |
1727 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
1728 | ap->device[i].pio_mode = XFER_PIO_0; | |
1729 | ||
022bdb07 TH |
1730 | if (postreset) |
1731 | postreset(ap, classes); | |
1732 | ||
1733 | /* reset successful, schedule revalidation */ | |
13abf50d | 1734 | ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK); |
022bdb07 TH |
1735 | ehc->i.action |= ATA_EH_REVALIDATE; |
1736 | } | |
1737 | ||
1738 | return rc; | |
1739 | } | |
1740 | ||
084fe639 TH |
1741 | static int ata_eh_revalidate_and_attach(struct ata_port *ap, |
1742 | struct ata_device **r_failed_dev) | |
022bdb07 TH |
1743 | { |
1744 | struct ata_eh_context *ehc = &ap->eh_context; | |
1745 | struct ata_device *dev; | |
8c3c52a8 | 1746 | unsigned int new_mask = 0; |
084fe639 | 1747 | unsigned long flags; |
022bdb07 TH |
1748 | int i, rc = 0; |
1749 | ||
1750 | DPRINTK("ENTER\n"); | |
1751 | ||
8c3c52a8 TH |
1752 | /* For PATA drive side cable detection to work, IDENTIFY must |
1753 | * be done backwards such that PDIAG- is released by the slave | |
1754 | * device before the master device is identified. | |
1755 | */ | |
1756 | for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) { | |
bff04647 | 1757 | unsigned int action, readid_flags = 0; |
47005f25 | 1758 | |
022bdb07 | 1759 | dev = &ap->device[i]; |
64f65ca6 | 1760 | action = ata_eh_dev_action(dev); |
022bdb07 | 1761 | |
bff04647 TH |
1762 | if (ehc->i.flags & ATA_EHI_DID_RESET) |
1763 | readid_flags |= ATA_READID_POSTRESET; | |
1764 | ||
02670bf3 | 1765 | if (action & ATA_EH_REVALIDATE && ata_dev_ready(dev)) { |
022bdb07 TH |
1766 | if (ata_port_offline(ap)) { |
1767 | rc = -EIO; | |
8c3c52a8 | 1768 | goto err; |
022bdb07 TH |
1769 | } |
1770 | ||
47005f25 | 1771 | ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE); |
bff04647 | 1772 | rc = ata_dev_revalidate(dev, readid_flags); |
022bdb07 | 1773 | if (rc) |
8c3c52a8 | 1774 | goto err; |
022bdb07 | 1775 | |
47005f25 TH |
1776 | ata_eh_done(ap, dev, ATA_EH_REVALIDATE); |
1777 | ||
baa1e78a TH |
1778 | /* Configuration may have changed, reconfigure |
1779 | * transfer mode. | |
1780 | */ | |
1781 | ehc->i.flags |= ATA_EHI_SETMODE; | |
1782 | ||
3057ac3c | 1783 | /* schedule the scsi_rescan_device() here */ |
1784 | queue_work(ata_aux_wq, &(ap->scsi_rescan_task)); | |
084fe639 TH |
1785 | } else if (dev->class == ATA_DEV_UNKNOWN && |
1786 | ehc->tries[dev->devno] && | |
1787 | ata_class_enabled(ehc->classes[dev->devno])) { | |
1788 | dev->class = ehc->classes[dev->devno]; | |
1789 | ||
bff04647 TH |
1790 | rc = ata_dev_read_id(dev, &dev->class, readid_flags, |
1791 | dev->id); | |
8c3c52a8 TH |
1792 | switch (rc) { |
1793 | case 0: | |
1794 | new_mask |= 1 << i; | |
1795 | break; | |
1796 | case -ENOENT: | |
55a8e2c8 TH |
1797 | /* IDENTIFY was issued to non-existent |
1798 | * device. No need to reset. Just | |
1799 | * thaw and kill the device. | |
1800 | */ | |
1801 | ata_eh_thaw_port(ap); | |
084fe639 TH |
1802 | dev->class = ATA_DEV_UNKNOWN; |
1803 | break; | |
8c3c52a8 TH |
1804 | default: |
1805 | dev->class = ATA_DEV_UNKNOWN; | |
1806 | goto err; | |
084fe639 | 1807 | } |
8c3c52a8 TH |
1808 | } |
1809 | } | |
084fe639 | 1810 | |
8c3c52a8 TH |
1811 | /* Configure new devices forward such that user doesn't see |
1812 | * device detection messages backwards. | |
1813 | */ | |
1814 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
1815 | dev = &ap->device[i]; | |
baa1e78a | 1816 | |
8c3c52a8 TH |
1817 | if (!(new_mask & (1 << i))) |
1818 | continue; | |
1819 | ||
1820 | ehc->i.flags |= ATA_EHI_PRINTINFO; | |
1821 | rc = ata_dev_configure(dev); | |
1822 | ehc->i.flags &= ~ATA_EHI_PRINTINFO; | |
1823 | if (rc) | |
1824 | goto err; | |
1825 | ||
1826 | spin_lock_irqsave(ap->lock, flags); | |
1827 | ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG; | |
1828 | spin_unlock_irqrestore(ap->lock, flags); | |
1829 | ||
1830 | /* new device discovered, configure xfermode */ | |
1831 | ehc->i.flags |= ATA_EHI_SETMODE; | |
022bdb07 TH |
1832 | } |
1833 | ||
8c3c52a8 | 1834 | return 0; |
022bdb07 | 1835 | |
8c3c52a8 TH |
1836 | err: |
1837 | *r_failed_dev = dev; | |
1838 | DPRINTK("EXIT rc=%d\n", rc); | |
022bdb07 TH |
1839 | return rc; |
1840 | } | |
1841 | ||
6ffa01d8 | 1842 | #ifdef CONFIG_PM |
02670bf3 TH |
1843 | /** |
1844 | * ata_eh_suspend - handle suspend EH action | |
1845 | * @ap: target host port | |
1846 | * @r_failed_dev: result parameter to indicate failing device | |
1847 | * | |
1848 | * Handle suspend EH action. Disk devices are spinned down and | |
1849 | * other types of devices are just marked suspended. Once | |
1850 | * suspended, no EH action to the device is allowed until it is | |
1851 | * resumed. | |
1852 | * | |
1853 | * LOCKING: | |
1854 | * Kernel thread context (may sleep). | |
1855 | * | |
1856 | * RETURNS: | |
1857 | * 0 on success, -errno otherwise | |
1858 | */ | |
1859 | static int ata_eh_suspend(struct ata_port *ap, struct ata_device **r_failed_dev) | |
1860 | { | |
1861 | struct ata_device *dev; | |
1862 | int i, rc = 0; | |
1863 | ||
1864 | DPRINTK("ENTER\n"); | |
1865 | ||
1866 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
1867 | unsigned long flags; | |
1868 | unsigned int action, err_mask; | |
1869 | ||
1870 | dev = &ap->device[i]; | |
1871 | action = ata_eh_dev_action(dev); | |
1872 | ||
1873 | if (!ata_dev_enabled(dev) || !(action & ATA_EH_SUSPEND)) | |
1874 | continue; | |
1875 | ||
1876 | WARN_ON(dev->flags & ATA_DFLAG_SUSPENDED); | |
1877 | ||
1878 | ata_eh_about_to_do(ap, dev, ATA_EH_SUSPEND); | |
1879 | ||
1880 | if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) { | |
1881 | /* flush cache */ | |
1882 | rc = ata_flush_cache(dev); | |
1883 | if (rc) | |
1884 | break; | |
1885 | ||
1886 | /* spin down */ | |
1887 | err_mask = ata_do_simple_cmd(dev, ATA_CMD_STANDBYNOW1); | |
1888 | if (err_mask) { | |
1889 | ata_dev_printk(dev, KERN_ERR, "failed to " | |
1890 | "spin down (err_mask=0x%x)\n", | |
1891 | err_mask); | |
1892 | rc = -EIO; | |
1893 | break; | |
1894 | } | |
1895 | } | |
1896 | ||
1897 | spin_lock_irqsave(ap->lock, flags); | |
1898 | dev->flags |= ATA_DFLAG_SUSPENDED; | |
1899 | spin_unlock_irqrestore(ap->lock, flags); | |
1900 | ||
1901 | ata_eh_done(ap, dev, ATA_EH_SUSPEND); | |
1902 | } | |
1903 | ||
1904 | if (rc) | |
1905 | *r_failed_dev = dev; | |
1906 | ||
1907 | DPRINTK("EXIT\n"); | |
03ee5b1c | 1908 | return rc; |
02670bf3 TH |
1909 | } |
1910 | ||
1911 | /** | |
1912 | * ata_eh_prep_resume - prep for resume EH action | |
1913 | * @ap: target host port | |
1914 | * | |
1915 | * Clear SUSPENDED in preparation for scheduled resume actions. | |
1916 | * This allows other parts of EH to access the devices being | |
1917 | * resumed. | |
1918 | * | |
1919 | * LOCKING: | |
1920 | * Kernel thread context (may sleep). | |
1921 | */ | |
1922 | static void ata_eh_prep_resume(struct ata_port *ap) | |
1923 | { | |
1924 | struct ata_device *dev; | |
1925 | unsigned long flags; | |
1926 | int i; | |
1927 | ||
1928 | DPRINTK("ENTER\n"); | |
1929 | ||
1930 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
1931 | unsigned int action; | |
1932 | ||
1933 | dev = &ap->device[i]; | |
1934 | action = ata_eh_dev_action(dev); | |
1935 | ||
1936 | if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME)) | |
1937 | continue; | |
1938 | ||
1939 | spin_lock_irqsave(ap->lock, flags); | |
1940 | dev->flags &= ~ATA_DFLAG_SUSPENDED; | |
1941 | spin_unlock_irqrestore(ap->lock, flags); | |
1942 | } | |
1943 | ||
1944 | DPRINTK("EXIT\n"); | |
1945 | } | |
1946 | ||
1947 | /** | |
1948 | * ata_eh_resume - handle resume EH action | |
1949 | * @ap: target host port | |
1950 | * @r_failed_dev: result parameter to indicate failing device | |
1951 | * | |
1952 | * Handle resume EH action. Target devices are already reset and | |
1953 | * revalidated. Spinning up is the only operation left. | |
1954 | * | |
1955 | * LOCKING: | |
1956 | * Kernel thread context (may sleep). | |
1957 | * | |
1958 | * RETURNS: | |
1959 | * 0 on success, -errno otherwise | |
1960 | */ | |
1961 | static int ata_eh_resume(struct ata_port *ap, struct ata_device **r_failed_dev) | |
1962 | { | |
1963 | struct ata_device *dev; | |
1964 | int i, rc = 0; | |
1965 | ||
1966 | DPRINTK("ENTER\n"); | |
1967 | ||
1968 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
1969 | unsigned int action, err_mask; | |
1970 | ||
1971 | dev = &ap->device[i]; | |
1972 | action = ata_eh_dev_action(dev); | |
1973 | ||
1974 | if (!ata_dev_enabled(dev) || !(action & ATA_EH_RESUME)) | |
1975 | continue; | |
1976 | ||
1977 | ata_eh_about_to_do(ap, dev, ATA_EH_RESUME); | |
1978 | ||
1979 | if (dev->class == ATA_DEV_ATA && !(action & ATA_EH_PM_FREEZE)) { | |
1980 | err_mask = ata_do_simple_cmd(dev, | |
1981 | ATA_CMD_IDLEIMMEDIATE); | |
1982 | if (err_mask) { | |
1983 | ata_dev_printk(dev, KERN_ERR, "failed to " | |
1984 | "spin up (err_mask=0x%x)\n", | |
1985 | err_mask); | |
1986 | rc = -EIO; | |
1987 | break; | |
1988 | } | |
1989 | } | |
1990 | ||
1991 | ata_eh_done(ap, dev, ATA_EH_RESUME); | |
1992 | } | |
1993 | ||
1994 | if (rc) | |
1995 | *r_failed_dev = dev; | |
1996 | ||
1997 | DPRINTK("EXIT\n"); | |
1998 | return 0; | |
1999 | } | |
6ffa01d8 | 2000 | #endif /* CONFIG_PM */ |
02670bf3 | 2001 | |
022bdb07 TH |
2002 | static int ata_port_nr_enabled(struct ata_port *ap) |
2003 | { | |
2004 | int i, cnt = 0; | |
2005 | ||
2006 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
2007 | if (ata_dev_enabled(&ap->device[i])) | |
2008 | cnt++; | |
2009 | return cnt; | |
2010 | } | |
2011 | ||
084fe639 TH |
2012 | static int ata_port_nr_vacant(struct ata_port *ap) |
2013 | { | |
2014 | int i, cnt = 0; | |
2015 | ||
2016 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
2017 | if (ap->device[i].class == ATA_DEV_UNKNOWN) | |
2018 | cnt++; | |
2019 | return cnt; | |
2020 | } | |
2021 | ||
2022 | static int ata_eh_skip_recovery(struct ata_port *ap) | |
2023 | { | |
2024 | struct ata_eh_context *ehc = &ap->eh_context; | |
2025 | int i; | |
2026 | ||
02670bf3 TH |
2027 | /* skip if all possible devices are suspended */ |
2028 | for (i = 0; i < ata_port_max_devices(ap); i++) { | |
2029 | struct ata_device *dev = &ap->device[i]; | |
2030 | ||
7c8c2cff | 2031 | if (!(dev->flags & ATA_DFLAG_SUSPENDED)) |
02670bf3 TH |
2032 | break; |
2033 | } | |
2034 | ||
2035 | if (i == ata_port_max_devices(ap)) | |
2036 | return 1; | |
2037 | ||
7c8c2cff TH |
2038 | /* thaw frozen port, resume link and recover failed devices */ |
2039 | if ((ap->pflags & ATA_PFLAG_FROZEN) || | |
2040 | (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap)) | |
084fe639 TH |
2041 | return 0; |
2042 | ||
2043 | /* skip if class codes for all vacant slots are ATA_DEV_NONE */ | |
2044 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
2045 | struct ata_device *dev = &ap->device[i]; | |
2046 | ||
2047 | if (dev->class == ATA_DEV_UNKNOWN && | |
2048 | ehc->classes[dev->devno] != ATA_DEV_NONE) | |
2049 | return 0; | |
2050 | } | |
2051 | ||
2052 | return 1; | |
2053 | } | |
2054 | ||
022bdb07 TH |
2055 | /** |
2056 | * ata_eh_recover - recover host port after error | |
2057 | * @ap: host port to recover | |
f5914a46 | 2058 | * @prereset: prereset method (can be NULL) |
022bdb07 TH |
2059 | * @softreset: softreset method (can be NULL) |
2060 | * @hardreset: hardreset method (can be NULL) | |
2061 | * @postreset: postreset method (can be NULL) | |
2062 | * | |
2063 | * This is the alpha and omega, eum and yang, heart and soul of | |
2064 | * libata exception handling. On entry, actions required to | |
084fe639 TH |
2065 | * recover the port and hotplug requests are recorded in |
2066 | * eh_context. This function executes all the operations with | |
2067 | * appropriate retrials and fallbacks to resurrect failed | |
2068 | * devices, detach goners and greet newcomers. | |
022bdb07 TH |
2069 | * |
2070 | * LOCKING: | |
2071 | * Kernel thread context (may sleep). | |
2072 | * | |
2073 | * RETURNS: | |
2074 | * 0 on success, -errno on failure. | |
2075 | */ | |
f5914a46 TH |
2076 | static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset, |
2077 | ata_reset_fn_t softreset, ata_reset_fn_t hardreset, | |
022bdb07 TH |
2078 | ata_postreset_fn_t postreset) |
2079 | { | |
2080 | struct ata_eh_context *ehc = &ap->eh_context; | |
2081 | struct ata_device *dev; | |
4ae72a1e | 2082 | int i, rc; |
022bdb07 TH |
2083 | |
2084 | DPRINTK("ENTER\n"); | |
2085 | ||
2086 | /* prep for recovery */ | |
2087 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
2088 | dev = &ap->device[i]; | |
2089 | ||
2090 | ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; | |
084fe639 | 2091 | |
79a55b72 TH |
2092 | /* collect port action mask recorded in dev actions */ |
2093 | ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK; | |
2094 | ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK; | |
2095 | ||
084fe639 TH |
2096 | /* process hotplug request */ |
2097 | if (dev->flags & ATA_DFLAG_DETACH) | |
2098 | ata_eh_detach_dev(dev); | |
2099 | ||
2100 | if (!ata_dev_enabled(dev) && | |
2101 | ((ehc->i.probe_mask & (1 << dev->devno)) && | |
2102 | !(ehc->did_probe_mask & (1 << dev->devno)))) { | |
2103 | ata_eh_detach_dev(dev); | |
2104 | ata_dev_init(dev); | |
2105 | ehc->did_probe_mask |= (1 << dev->devno); | |
2106 | ehc->i.action |= ATA_EH_SOFTRESET; | |
2107 | } | |
022bdb07 TH |
2108 | } |
2109 | ||
2110 | retry: | |
022bdb07 TH |
2111 | rc = 0; |
2112 | ||
aeb2ecd6 | 2113 | /* if UNLOADING, finish immediately */ |
b51e9e5d | 2114 | if (ap->pflags & ATA_PFLAG_UNLOADING) |
aeb2ecd6 TH |
2115 | goto out; |
2116 | ||
02670bf3 TH |
2117 | /* prep for resume */ |
2118 | ata_eh_prep_resume(ap); | |
2119 | ||
022bdb07 | 2120 | /* skip EH if possible. */ |
084fe639 | 2121 | if (ata_eh_skip_recovery(ap)) |
022bdb07 TH |
2122 | ehc->i.action = 0; |
2123 | ||
084fe639 TH |
2124 | for (i = 0; i < ATA_MAX_DEVICES; i++) |
2125 | ehc->classes[i] = ATA_DEV_UNKNOWN; | |
2126 | ||
022bdb07 TH |
2127 | /* reset */ |
2128 | if (ehc->i.action & ATA_EH_RESET_MASK) { | |
2129 | ata_eh_freeze_port(ap); | |
2130 | ||
084fe639 TH |
2131 | rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset, |
2132 | softreset, hardreset, postreset); | |
022bdb07 TH |
2133 | if (rc) { |
2134 | ata_port_printk(ap, KERN_ERR, | |
2135 | "reset failed, giving up\n"); | |
2136 | goto out; | |
2137 | } | |
2138 | ||
2139 | ata_eh_thaw_port(ap); | |
2140 | } | |
2141 | ||
084fe639 TH |
2142 | /* revalidate existing devices and attach new ones */ |
2143 | rc = ata_eh_revalidate_and_attach(ap, &dev); | |
022bdb07 TH |
2144 | if (rc) |
2145 | goto dev_fail; | |
2146 | ||
02670bf3 TH |
2147 | /* resume devices */ |
2148 | rc = ata_eh_resume(ap, &dev); | |
2149 | if (rc) | |
2150 | goto dev_fail; | |
2151 | ||
baa1e78a TH |
2152 | /* configure transfer mode if necessary */ |
2153 | if (ehc->i.flags & ATA_EHI_SETMODE) { | |
022bdb07 | 2154 | rc = ata_set_mode(ap, &dev); |
4ae72a1e | 2155 | if (rc) |
022bdb07 | 2156 | goto dev_fail; |
baa1e78a | 2157 | ehc->i.flags &= ~ATA_EHI_SETMODE; |
022bdb07 TH |
2158 | } |
2159 | ||
02670bf3 TH |
2160 | /* suspend devices */ |
2161 | rc = ata_eh_suspend(ap, &dev); | |
2162 | if (rc) | |
2163 | goto dev_fail; | |
2164 | ||
022bdb07 TH |
2165 | goto out; |
2166 | ||
2167 | dev_fail: | |
4ae72a1e TH |
2168 | ehc->tries[dev->devno]--; |
2169 | ||
022bdb07 | 2170 | switch (rc) { |
022bdb07 | 2171 | case -EINVAL: |
4ae72a1e | 2172 | /* eeek, something went very wrong, give up */ |
022bdb07 TH |
2173 | ehc->tries[dev->devno] = 0; |
2174 | break; | |
4ae72a1e TH |
2175 | |
2176 | case -ENODEV: | |
2177 | /* device missing or wrong IDENTIFY data, schedule probing */ | |
2178 | ehc->i.probe_mask |= (1 << dev->devno); | |
2179 | /* give it just one more chance */ | |
2180 | ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1); | |
022bdb07 | 2181 | case -EIO: |
4ae72a1e TH |
2182 | if (ehc->tries[dev->devno] == 1) { |
2183 | /* This is the last chance, better to slow | |
2184 | * down than lose it. | |
2185 | */ | |
2186 | sata_down_spd_limit(ap); | |
2187 | ata_down_xfermask_limit(dev, ATA_DNXFER_PIO); | |
2188 | } | |
022bdb07 TH |
2189 | } |
2190 | ||
084fe639 TH |
2191 | if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) { |
2192 | /* disable device if it has used up all its chances */ | |
022bdb07 TH |
2193 | ata_dev_disable(dev); |
2194 | ||
084fe639 TH |
2195 | /* detach if offline */ |
2196 | if (ata_port_offline(ap)) | |
2197 | ata_eh_detach_dev(dev); | |
2198 | ||
2199 | /* probe if requested */ | |
2200 | if ((ehc->i.probe_mask & (1 << dev->devno)) && | |
2201 | !(ehc->did_probe_mask & (1 << dev->devno))) { | |
2202 | ata_eh_detach_dev(dev); | |
2203 | ata_dev_init(dev); | |
2204 | ||
2205 | ehc->tries[dev->devno] = ATA_EH_DEV_TRIES; | |
2206 | ehc->did_probe_mask |= (1 << dev->devno); | |
2207 | ehc->i.action |= ATA_EH_SOFTRESET; | |
2208 | } | |
2209 | } else { | |
2210 | /* soft didn't work? be haaaaard */ | |
2211 | if (ehc->i.flags & ATA_EHI_DID_RESET) | |
2212 | ehc->i.action |= ATA_EH_HARDRESET; | |
2213 | else | |
2214 | ehc->i.action |= ATA_EH_SOFTRESET; | |
2215 | } | |
022bdb07 TH |
2216 | |
2217 | if (ata_port_nr_enabled(ap)) { | |
2218 | ata_port_printk(ap, KERN_WARNING, "failed to recover some " | |
2219 | "devices, retrying in 5 secs\n"); | |
2220 | ssleep(5); | |
2221 | } else { | |
2222 | /* no device left, repeat fast */ | |
2223 | msleep(500); | |
2224 | } | |
2225 | ||
2226 | goto retry; | |
2227 | ||
2228 | out: | |
2229 | if (rc) { | |
2230 | for (i = 0; i < ATA_MAX_DEVICES; i++) | |
2231 | ata_dev_disable(&ap->device[i]); | |
2232 | } | |
2233 | ||
2234 | DPRINTK("EXIT, rc=%d\n", rc); | |
2235 | return rc; | |
2236 | } | |
2237 | ||
2238 | /** | |
2239 | * ata_eh_finish - finish up EH | |
2240 | * @ap: host port to finish EH for | |
2241 | * | |
2242 | * Recovery is complete. Clean up EH states and retry or finish | |
2243 | * failed qcs. | |
2244 | * | |
2245 | * LOCKING: | |
2246 | * None. | |
2247 | */ | |
2248 | static void ata_eh_finish(struct ata_port *ap) | |
2249 | { | |
2250 | int tag; | |
2251 | ||
2252 | /* retry or finish qcs */ | |
2253 | for (tag = 0; tag < ATA_MAX_QUEUE; tag++) { | |
2254 | struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag); | |
2255 | ||
2256 | if (!(qc->flags & ATA_QCFLAG_FAILED)) | |
2257 | continue; | |
2258 | ||
2259 | if (qc->err_mask) { | |
2260 | /* FIXME: Once EH migration is complete, | |
2261 | * generate sense data in this function, | |
2262 | * considering both err_mask and tf. | |
2263 | */ | |
2264 | if (qc->err_mask & AC_ERR_INVALID) | |
2265 | ata_eh_qc_complete(qc); | |
2266 | else | |
2267 | ata_eh_qc_retry(qc); | |
2268 | } else { | |
2269 | if (qc->flags & ATA_QCFLAG_SENSE_VALID) { | |
2270 | ata_eh_qc_complete(qc); | |
2271 | } else { | |
2272 | /* feed zero TF to sense generation */ | |
2273 | memset(&qc->result_tf, 0, sizeof(qc->result_tf)); | |
2274 | ata_eh_qc_retry(qc); | |
2275 | } | |
2276 | } | |
2277 | } | |
2278 | } | |
2279 | ||
2280 | /** | |
2281 | * ata_do_eh - do standard error handling | |
2282 | * @ap: host port to handle error for | |
f5914a46 | 2283 | * @prereset: prereset method (can be NULL) |
022bdb07 TH |
2284 | * @softreset: softreset method (can be NULL) |
2285 | * @hardreset: hardreset method (can be NULL) | |
2286 | * @postreset: postreset method (can be NULL) | |
2287 | * | |
2288 | * Perform standard error handling sequence. | |
2289 | * | |
2290 | * LOCKING: | |
2291 | * Kernel thread context (may sleep). | |
2292 | */ | |
f5914a46 TH |
2293 | void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset, |
2294 | ata_reset_fn_t softreset, ata_reset_fn_t hardreset, | |
2295 | ata_postreset_fn_t postreset) | |
022bdb07 | 2296 | { |
1cdaf534 TH |
2297 | ata_eh_autopsy(ap); |
2298 | ata_eh_report(ap); | |
f5914a46 | 2299 | ata_eh_recover(ap, prereset, softreset, hardreset, postreset); |
022bdb07 TH |
2300 | ata_eh_finish(ap); |
2301 | } | |
500530f6 | 2302 | |
6ffa01d8 | 2303 | #ifdef CONFIG_PM |
500530f6 TH |
2304 | /** |
2305 | * ata_eh_handle_port_suspend - perform port suspend operation | |
2306 | * @ap: port to suspend | |
2307 | * | |
2308 | * Suspend @ap. | |
2309 | * | |
2310 | * LOCKING: | |
2311 | * Kernel thread context (may sleep). | |
2312 | */ | |
2313 | static void ata_eh_handle_port_suspend(struct ata_port *ap) | |
2314 | { | |
2315 | unsigned long flags; | |
2316 | int rc = 0; | |
2317 | ||
2318 | /* are we suspending? */ | |
2319 | spin_lock_irqsave(ap->lock, flags); | |
2320 | if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || | |
2321 | ap->pm_mesg.event == PM_EVENT_ON) { | |
2322 | spin_unlock_irqrestore(ap->lock, flags); | |
2323 | return; | |
2324 | } | |
2325 | spin_unlock_irqrestore(ap->lock, flags); | |
2326 | ||
2327 | WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED); | |
2328 | ||
2329 | /* suspend */ | |
2330 | ata_eh_freeze_port(ap); | |
2331 | ||
2332 | if (ap->ops->port_suspend) | |
2333 | rc = ap->ops->port_suspend(ap, ap->pm_mesg); | |
2334 | ||
2335 | /* report result */ | |
2336 | spin_lock_irqsave(ap->lock, flags); | |
2337 | ||
2338 | ap->pflags &= ~ATA_PFLAG_PM_PENDING; | |
2339 | if (rc == 0) | |
2340 | ap->pflags |= ATA_PFLAG_SUSPENDED; | |
2341 | else | |
2342 | ata_port_schedule_eh(ap); | |
2343 | ||
2344 | if (ap->pm_result) { | |
2345 | *ap->pm_result = rc; | |
2346 | ap->pm_result = NULL; | |
2347 | } | |
2348 | ||
2349 | spin_unlock_irqrestore(ap->lock, flags); | |
2350 | ||
2351 | return; | |
2352 | } | |
2353 | ||
2354 | /** | |
2355 | * ata_eh_handle_port_resume - perform port resume operation | |
2356 | * @ap: port to resume | |
2357 | * | |
2358 | * Resume @ap. | |
2359 | * | |
2360 | * This function also waits upto one second until all devices | |
2361 | * hanging off this port requests resume EH action. This is to | |
2362 | * prevent invoking EH and thus reset multiple times on resume. | |
2363 | * | |
2364 | * On DPM resume, where some of devices might not be resumed | |
2365 | * together, this may delay port resume upto one second, but such | |
2366 | * DPM resumes are rare and 1 sec delay isn't too bad. | |
2367 | * | |
2368 | * LOCKING: | |
2369 | * Kernel thread context (may sleep). | |
2370 | */ | |
2371 | static void ata_eh_handle_port_resume(struct ata_port *ap) | |
2372 | { | |
2373 | unsigned long timeout; | |
2374 | unsigned long flags; | |
2375 | int i, rc = 0; | |
2376 | ||
2377 | /* are we resuming? */ | |
2378 | spin_lock_irqsave(ap->lock, flags); | |
2379 | if (!(ap->pflags & ATA_PFLAG_PM_PENDING) || | |
2380 | ap->pm_mesg.event != PM_EVENT_ON) { | |
2381 | spin_unlock_irqrestore(ap->lock, flags); | |
2382 | return; | |
2383 | } | |
2384 | spin_unlock_irqrestore(ap->lock, flags); | |
2385 | ||
2386 | /* spurious? */ | |
2387 | if (!(ap->pflags & ATA_PFLAG_SUSPENDED)) | |
2388 | goto done; | |
2389 | ||
2390 | if (ap->ops->port_resume) | |
2391 | rc = ap->ops->port_resume(ap); | |
2392 | ||
2393 | /* give devices time to request EH */ | |
2394 | timeout = jiffies + HZ; /* 1s max */ | |
2395 | while (1) { | |
2396 | for (i = 0; i < ATA_MAX_DEVICES; i++) { | |
2397 | struct ata_device *dev = &ap->device[i]; | |
2398 | unsigned int action = ata_eh_dev_action(dev); | |
2399 | ||
2400 | if ((dev->flags & ATA_DFLAG_SUSPENDED) && | |
2401 | !(action & ATA_EH_RESUME)) | |
2402 | break; | |
2403 | } | |
2404 | ||
2405 | if (i == ATA_MAX_DEVICES || time_after(jiffies, timeout)) | |
2406 | break; | |
2407 | msleep(10); | |
2408 | } | |
2409 | ||
2410 | done: | |
2411 | spin_lock_irqsave(ap->lock, flags); | |
2412 | ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED); | |
2413 | if (ap->pm_result) { | |
2414 | *ap->pm_result = rc; | |
2415 | ap->pm_result = NULL; | |
2416 | } | |
2417 | spin_unlock_irqrestore(ap->lock, flags); | |
2418 | } | |
6ffa01d8 | 2419 | #endif /* CONFIG_PM */ |