]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * scsi_error.c Copyright (C) 1997 Eric Youngdale | |
3 | * | |
4 | * SCSI error/timeout handling | |
5 | * Initial versions: Eric Youngdale. Based upon conversations with | |
6 | * Leonard Zubkoff and David Miller at Linux Expo, | |
7 | * ideas originating from all over the place. | |
8 | * | |
9 | * Restructured scsi_unjam_host and associated functions. | |
10 | * September 04, 2002 Mike Anderson (andmike@us.ibm.com) | |
11 | * | |
12 | * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and | |
13 | * minor cleanups. | |
14 | * September 30, 2002 Mike Anderson (andmike@us.ibm.com) | |
15 | */ | |
16 | ||
17 | #include <linux/module.h> | |
18 | #include <linux/sched.h> | |
19 | #include <linux/timer.h> | |
20 | #include <linux/string.h> | |
1da177e4 | 21 | #include <linux/kernel.h> |
83144186 | 22 | #include <linux/freezer.h> |
c5478def | 23 | #include <linux/kthread.h> |
1da177e4 LT |
24 | #include <linux/interrupt.h> |
25 | #include <linux/blkdev.h> | |
26 | #include <linux/delay.h> | |
27 | ||
28 | #include <scsi/scsi.h> | |
beb40487 | 29 | #include <scsi/scsi_cmnd.h> |
1da177e4 LT |
30 | #include <scsi/scsi_dbg.h> |
31 | #include <scsi/scsi_device.h> | |
32 | #include <scsi/scsi_eh.h> | |
c829c394 | 33 | #include <scsi/scsi_transport.h> |
1da177e4 LT |
34 | #include <scsi/scsi_host.h> |
35 | #include <scsi/scsi_ioctl.h> | |
1da177e4 LT |
36 | |
37 | #include "scsi_priv.h" | |
38 | #include "scsi_logging.h" | |
79ee8304 | 39 | #include "scsi_transport_api.h" |
1da177e4 LT |
40 | |
41 | #define SENSE_TIMEOUT (10*HZ) | |
1da177e4 LT |
42 | |
43 | /* | |
44 | * These should *probably* be handled by the host itself. | |
45 | * Since it is allowed to sleep, it probably should. | |
46 | */ | |
47 | #define BUS_RESET_SETTLE_TIME (10) | |
48 | #define HOST_RESET_SETTLE_TIME (10) | |
49 | ||
50 | /* called with shost->host_lock held */ | |
51 | void scsi_eh_wakeup(struct Scsi_Host *shost) | |
52 | { | |
53 | if (shost->host_busy == shost->host_failed) { | |
3ed7a470 | 54 | wake_up_process(shost->ehandler); |
1da177e4 LT |
55 | SCSI_LOG_ERROR_RECOVERY(5, |
56 | printk("Waking error handler thread\n")); | |
57 | } | |
58 | } | |
f8bbfc24 TH |
59 | |
60 | /** | |
61 | * scsi_schedule_eh - schedule EH for SCSI host | |
62 | * @shost: SCSI host to invoke error handling on. | |
63 | * | |
64 | * Schedule SCSI EH without scmd. | |
dc8875e1 | 65 | */ |
f8bbfc24 TH |
66 | void scsi_schedule_eh(struct Scsi_Host *shost) |
67 | { | |
68 | unsigned long flags; | |
69 | ||
70 | spin_lock_irqsave(shost->host_lock, flags); | |
71 | ||
72 | if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || | |
73 | scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { | |
74 | shost->host_eh_scheduled++; | |
75 | scsi_eh_wakeup(shost); | |
76 | } | |
77 | ||
78 | spin_unlock_irqrestore(shost->host_lock, flags); | |
79 | } | |
80 | EXPORT_SYMBOL_GPL(scsi_schedule_eh); | |
1da177e4 LT |
81 | |
82 | /** | |
83 | * scsi_eh_scmd_add - add scsi cmd to error handling. | |
84 | * @scmd: scmd to run eh on. | |
85 | * @eh_flag: optional SCSI_EH flag. | |
86 | * | |
87 | * Return value: | |
88 | * 0 on failure. | |
dc8875e1 | 89 | */ |
1da177e4 LT |
90 | int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) |
91 | { | |
92 | struct Scsi_Host *shost = scmd->device->host; | |
93 | unsigned long flags; | |
939647ee | 94 | int ret = 0; |
1da177e4 | 95 | |
3ed7a470 | 96 | if (!shost->ehandler) |
1da177e4 LT |
97 | return 0; |
98 | ||
99 | spin_lock_irqsave(shost->host_lock, flags); | |
939647ee JB |
100 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
101 | if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) | |
102 | goto out_unlock; | |
1da177e4 | 103 | |
939647ee | 104 | ret = 1; |
3111b0d1 | 105 | scmd->eh_eflags |= eh_flag; |
1da177e4 | 106 | list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); |
1da177e4 LT |
107 | shost->host_failed++; |
108 | scsi_eh_wakeup(shost); | |
939647ee | 109 | out_unlock: |
1da177e4 | 110 | spin_unlock_irqrestore(shost->host_lock, flags); |
939647ee | 111 | return ret; |
1da177e4 LT |
112 | } |
113 | ||
114 | /** | |
115 | * scsi_add_timer - Start timeout timer for a single scsi command. | |
116 | * @scmd: scsi command that is about to start running. | |
117 | * @timeout: amount of time to allow this command to run. | |
118 | * @complete: timeout function to call if timer isn't canceled. | |
119 | * | |
120 | * Notes: | |
121 | * This should be turned into an inline function. Each scsi command | |
122 | * has its own timer, and as it is added to the queue, we set up the | |
123 | * timer. When the command completes, we cancel the timer. | |
dc8875e1 | 124 | */ |
1da177e4 LT |
125 | void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, |
126 | void (*complete)(struct scsi_cmnd *)) | |
127 | { | |
128 | ||
129 | /* | |
130 | * If the clock was already running for this command, then | |
131 | * first delete the timer. The timer handling code gets rather | |
132 | * confused if we don't do this. | |
133 | */ | |
134 | if (scmd->eh_timeout.function) | |
135 | del_timer(&scmd->eh_timeout); | |
136 | ||
137 | scmd->eh_timeout.data = (unsigned long)scmd; | |
138 | scmd->eh_timeout.expires = jiffies + timeout; | |
139 | scmd->eh_timeout.function = (void (*)(unsigned long)) complete; | |
140 | ||
141 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" | |
cadbd4a5 | 142 | " %d, (%p)\n", __func__, |
1da177e4 LT |
143 | scmd, timeout, complete)); |
144 | ||
145 | add_timer(&scmd->eh_timeout); | |
146 | } | |
1da177e4 LT |
147 | |
148 | /** | |
149 | * scsi_delete_timer - Delete/cancel timer for a given function. | |
150 | * @scmd: Cmd that we are canceling timer for | |
151 | * | |
152 | * Notes: | |
153 | * This should be turned into an inline function. | |
154 | * | |
155 | * Return value: | |
156 | * 1 if we were able to detach the timer. 0 if we blew it, and the | |
157 | * timer function has already started to run. | |
dc8875e1 | 158 | */ |
1da177e4 LT |
159 | int scsi_delete_timer(struct scsi_cmnd *scmd) |
160 | { | |
161 | int rtn; | |
162 | ||
163 | rtn = del_timer(&scmd->eh_timeout); | |
164 | ||
165 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," | |
cadbd4a5 | 166 | " rtn: %d\n", __func__, |
1da177e4 LT |
167 | scmd, rtn)); |
168 | ||
169 | scmd->eh_timeout.data = (unsigned long)NULL; | |
170 | scmd->eh_timeout.function = NULL; | |
171 | ||
172 | return rtn; | |
173 | } | |
1da177e4 LT |
174 | |
175 | /** | |
176 | * scsi_times_out - Timeout function for normal scsi commands. | |
177 | * @scmd: Cmd that is timing out. | |
178 | * | |
179 | * Notes: | |
180 | * We do not need to lock this. There is the potential for a race | |
181 | * only in that the normal completion handling might run, but if the | |
182 | * normal completion function determines that the timer has already | |
183 | * fired, then it mustn't do anything. | |
dc8875e1 | 184 | */ |
1da177e4 LT |
185 | void scsi_times_out(struct scsi_cmnd *scmd) |
186 | { | |
6c5f8ce1 JB |
187 | enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *); |
188 | ||
1da177e4 LT |
189 | scsi_log_completion(scmd, TIMEOUT_ERROR); |
190 | ||
c829c394 | 191 | if (scmd->device->host->transportt->eh_timed_out) |
6c5f8ce1 JB |
192 | eh_timed_out = scmd->device->host->transportt->eh_timed_out; |
193 | else if (scmd->device->host->hostt->eh_timed_out) | |
194 | eh_timed_out = scmd->device->host->hostt->eh_timed_out; | |
195 | else | |
196 | eh_timed_out = NULL; | |
197 | ||
198 | if (eh_timed_out) | |
199 | switch (eh_timed_out(scmd)) { | |
1da177e4 LT |
200 | case EH_HANDLED: |
201 | __scsi_done(scmd); | |
202 | return; | |
203 | case EH_RESET_TIMER: | |
1da177e4 LT |
204 | scsi_add_timer(scmd, scmd->timeout_per_command, |
205 | scsi_times_out); | |
206 | return; | |
207 | case EH_NOT_HANDLED: | |
208 | break; | |
209 | } | |
210 | ||
211 | if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { | |
939647ee JB |
212 | scmd->result |= DID_TIME_OUT << 16; |
213 | __scsi_done(scmd); | |
1da177e4 LT |
214 | } |
215 | } | |
216 | ||
217 | /** | |
218 | * scsi_block_when_processing_errors - Prevent cmds from being queued. | |
219 | * @sdev: Device on which we are performing recovery. | |
220 | * | |
221 | * Description: | |
222 | * We block until the host is out of error recovery, and then check to | |
223 | * see whether the host or the device is offline. | |
224 | * | |
225 | * Return value: | |
226 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. | |
dc8875e1 | 227 | */ |
1da177e4 LT |
228 | int scsi_block_when_processing_errors(struct scsi_device *sdev) |
229 | { | |
230 | int online; | |
231 | ||
939647ee | 232 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
1da177e4 LT |
233 | |
234 | online = scsi_device_online(sdev); | |
235 | ||
cadbd4a5 | 236 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__, |
1da177e4 LT |
237 | online)); |
238 | ||
239 | return online; | |
240 | } | |
241 | EXPORT_SYMBOL(scsi_block_when_processing_errors); | |
242 | ||
243 | #ifdef CONFIG_SCSI_LOGGING | |
244 | /** | |
245 | * scsi_eh_prt_fail_stats - Log info on failures. | |
246 | * @shost: scsi host being recovered. | |
247 | * @work_q: Queue of scsi cmds to process. | |
dc8875e1 | 248 | */ |
1da177e4 LT |
249 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, |
250 | struct list_head *work_q) | |
251 | { | |
252 | struct scsi_cmnd *scmd; | |
253 | struct scsi_device *sdev; | |
254 | int total_failures = 0; | |
255 | int cmd_failed = 0; | |
256 | int cmd_cancel = 0; | |
257 | int devices_failed = 0; | |
258 | ||
259 | shost_for_each_device(sdev, shost) { | |
260 | list_for_each_entry(scmd, work_q, eh_entry) { | |
261 | if (scmd->device == sdev) { | |
262 | ++total_failures; | |
3111b0d1 | 263 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) |
1da177e4 LT |
264 | ++cmd_cancel; |
265 | else | |
266 | ++cmd_failed; | |
267 | } | |
268 | } | |
269 | ||
270 | if (cmd_cancel || cmd_failed) { | |
271 | SCSI_LOG_ERROR_RECOVERY(3, | |
9ccfc756 JB |
272 | sdev_printk(KERN_INFO, sdev, |
273 | "%s: cmds failed: %d, cancel: %d\n", | |
cadbd4a5 | 274 | __func__, cmd_failed, |
9ccfc756 | 275 | cmd_cancel)); |
1da177e4 LT |
276 | cmd_cancel = 0; |
277 | cmd_failed = 0; | |
278 | ++devices_failed; | |
279 | } | |
280 | } | |
281 | ||
282 | SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" | |
283 | " devices require eh work\n", | |
284 | total_failures, devices_failed)); | |
285 | } | |
286 | #endif | |
287 | ||
288 | /** | |
289 | * scsi_check_sense - Examine scsi cmd sense | |
290 | * @scmd: Cmd to have sense checked. | |
291 | * | |
292 | * Return value: | |
293 | * SUCCESS or FAILED or NEEDS_RETRY | |
294 | * | |
295 | * Notes: | |
296 | * When a deferred error is detected the current command has | |
297 | * not been executed and needs retrying. | |
dc8875e1 | 298 | */ |
1da177e4 LT |
299 | static int scsi_check_sense(struct scsi_cmnd *scmd) |
300 | { | |
a6a8d9f8 | 301 | struct scsi_device *sdev = scmd->device; |
1da177e4 LT |
302 | struct scsi_sense_hdr sshdr; |
303 | ||
304 | if (! scsi_command_normalize_sense(scmd, &sshdr)) | |
305 | return FAILED; /* no valid sense data */ | |
306 | ||
307 | if (scsi_sense_is_deferred(&sshdr)) | |
308 | return NEEDS_RETRY; | |
309 | ||
a6a8d9f8 CS |
310 | if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh && |
311 | sdev->scsi_dh_data->scsi_dh->check_sense) { | |
312 | int rc; | |
313 | ||
314 | rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr); | |
315 | if (rc != SCSI_RETURN_NOT_HANDLED) | |
316 | return rc; | |
317 | /* handler does not care. Drop down to default handling */ | |
318 | } | |
319 | ||
1da177e4 LT |
320 | /* |
321 | * Previous logic looked for FILEMARK, EOM or ILI which are | |
322 | * mainly associated with tapes and returned SUCCESS. | |
323 | */ | |
324 | if (sshdr.response_code == 0x70) { | |
325 | /* fixed format */ | |
326 | if (scmd->sense_buffer[2] & 0xe0) | |
327 | return SUCCESS; | |
328 | } else { | |
329 | /* | |
330 | * descriptor format: look for "stream commands sense data | |
331 | * descriptor" (see SSC-3). Assume single sense data | |
332 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. | |
333 | */ | |
334 | if ((sshdr.additional_length > 3) && | |
335 | (scmd->sense_buffer[8] == 0x4) && | |
336 | (scmd->sense_buffer[11] & 0xe0)) | |
337 | return SUCCESS; | |
338 | } | |
339 | ||
340 | switch (sshdr.sense_key) { | |
341 | case NO_SENSE: | |
342 | return SUCCESS; | |
343 | case RECOVERED_ERROR: | |
344 | return /* soft_error */ SUCCESS; | |
345 | ||
346 | case ABORTED_COMMAND: | |
511e44f4 MP |
347 | if (sshdr.asc == 0x10) /* DIF */ |
348 | return SUCCESS; | |
349 | ||
1da177e4 LT |
350 | return NEEDS_RETRY; |
351 | case NOT_READY: | |
352 | case UNIT_ATTENTION: | |
353 | /* | |
354 | * if we are expecting a cc/ua because of a bus reset that we | |
355 | * performed, treat this just as a retry. otherwise this is | |
356 | * information that we should pass up to the upper-level driver | |
357 | * so that we can deal with it there. | |
358 | */ | |
359 | if (scmd->device->expecting_cc_ua) { | |
360 | scmd->device->expecting_cc_ua = 0; | |
361 | return NEEDS_RETRY; | |
362 | } | |
363 | /* | |
364 | * if the device is in the process of becoming ready, we | |
365 | * should retry. | |
366 | */ | |
367 | if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) | |
368 | return NEEDS_RETRY; | |
369 | /* | |
370 | * if the device is not started, we need to wake | |
371 | * the error handler to start the motor | |
372 | */ | |
373 | if (scmd->device->allow_restart && | |
374 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) | |
375 | return FAILED; | |
376 | return SUCCESS; | |
377 | ||
378 | /* these three are not supported */ | |
379 | case COPY_ABORTED: | |
380 | case VOLUME_OVERFLOW: | |
381 | case MISCOMPARE: | |
382 | return SUCCESS; | |
383 | ||
384 | case MEDIUM_ERROR: | |
fd1b494d LT |
385 | if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ |
386 | sshdr.asc == 0x13 || /* AMNF DATA FIELD */ | |
387 | sshdr.asc == 0x14) { /* RECORD NOT FOUND */ | |
388 | return SUCCESS; | |
389 | } | |
1da177e4 LT |
390 | return NEEDS_RETRY; |
391 | ||
392 | case HARDWARE_ERROR: | |
393 | if (scmd->device->retry_hwerror) | |
bb0003c1 | 394 | return ADD_TO_MLQUEUE; |
1da177e4 LT |
395 | else |
396 | return SUCCESS; | |
397 | ||
398 | case ILLEGAL_REQUEST: | |
399 | case BLANK_CHECK: | |
400 | case DATA_PROTECT: | |
401 | default: | |
402 | return SUCCESS; | |
403 | } | |
404 | } | |
405 | ||
406 | /** | |
407 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. | |
408 | * @scmd: SCSI cmd to examine. | |
409 | * | |
410 | * Notes: | |
411 | * This is *only* called when we are examining the status of commands | |
412 | * queued during error recovery. the main difference here is that we | |
413 | * don't allow for the possibility of retries here, and we are a lot | |
414 | * more restrictive about what we consider acceptable. | |
dc8875e1 | 415 | */ |
1da177e4 LT |
416 | static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) |
417 | { | |
418 | /* | |
419 | * first check the host byte, to see if there is anything in there | |
420 | * that would indicate what we need to do. | |
421 | */ | |
422 | if (host_byte(scmd->result) == DID_RESET) { | |
423 | /* | |
424 | * rats. we are already in the error handler, so we now | |
425 | * get to try and figure out what to do next. if the sense | |
426 | * is valid, we have a pretty good idea of what to do. | |
427 | * if not, we mark it as FAILED. | |
428 | */ | |
429 | return scsi_check_sense(scmd); | |
430 | } | |
431 | if (host_byte(scmd->result) != DID_OK) | |
432 | return FAILED; | |
433 | ||
434 | /* | |
435 | * next, check the message byte. | |
436 | */ | |
437 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) | |
438 | return FAILED; | |
439 | ||
440 | /* | |
441 | * now, check the status byte to see if this indicates | |
442 | * anything special. | |
443 | */ | |
444 | switch (status_byte(scmd->result)) { | |
445 | case GOOD: | |
446 | case COMMAND_TERMINATED: | |
447 | return SUCCESS; | |
448 | case CHECK_CONDITION: | |
449 | return scsi_check_sense(scmd); | |
450 | case CONDITION_GOOD: | |
451 | case INTERMEDIATE_GOOD: | |
452 | case INTERMEDIATE_C_GOOD: | |
453 | /* | |
454 | * who knows? FIXME(eric) | |
455 | */ | |
456 | return SUCCESS; | |
457 | case BUSY: | |
458 | case QUEUE_FULL: | |
459 | case RESERVATION_CONFLICT: | |
460 | default: | |
461 | return FAILED; | |
462 | } | |
463 | return FAILED; | |
464 | } | |
465 | ||
1da177e4 LT |
466 | /** |
467 | * scsi_eh_done - Completion function for error handling. | |
468 | * @scmd: Cmd that is done. | |
dc8875e1 | 469 | */ |
1da177e4 LT |
470 | static void scsi_eh_done(struct scsi_cmnd *scmd) |
471 | { | |
85631672 MR |
472 | struct completion *eh_action; |
473 | ||
7dfdc9a5 CH |
474 | SCSI_LOG_ERROR_RECOVERY(3, |
475 | printk("%s scmd: %p result: %x\n", | |
cadbd4a5 | 476 | __func__, scmd, scmd->result)); |
85631672 MR |
477 | |
478 | eh_action = scmd->device->host->eh_action; | |
479 | if (eh_action) | |
480 | complete(eh_action); | |
1da177e4 LT |
481 | } |
482 | ||
292148f8 BK |
483 | /** |
484 | * scsi_try_host_reset - ask host adapter to reset itself | |
485 | * @scmd: SCSI cmd to send hsot reset. | |
dc8875e1 | 486 | */ |
292148f8 BK |
487 | static int scsi_try_host_reset(struct scsi_cmnd *scmd) |
488 | { | |
489 | unsigned long flags; | |
490 | int rtn; | |
491 | ||
492 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", | |
cadbd4a5 | 493 | __func__)); |
292148f8 BK |
494 | |
495 | if (!scmd->device->host->hostt->eh_host_reset_handler) | |
496 | return FAILED; | |
497 | ||
498 | rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); | |
499 | ||
500 | if (rtn == SUCCESS) { | |
501 | if (!scmd->device->host->hostt->skip_settle_delay) | |
502 | ssleep(HOST_RESET_SETTLE_TIME); | |
503 | spin_lock_irqsave(scmd->device->host->host_lock, flags); | |
504 | scsi_report_bus_reset(scmd->device->host, | |
505 | scmd_channel(scmd)); | |
506 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); | |
507 | } | |
508 | ||
509 | return rtn; | |
510 | } | |
511 | ||
512 | /** | |
513 | * scsi_try_bus_reset - ask host to perform a bus reset | |
514 | * @scmd: SCSI cmd to send bus reset. | |
dc8875e1 | 515 | */ |
292148f8 BK |
516 | static int scsi_try_bus_reset(struct scsi_cmnd *scmd) |
517 | { | |
518 | unsigned long flags; | |
519 | int rtn; | |
520 | ||
521 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", | |
cadbd4a5 | 522 | __func__)); |
292148f8 BK |
523 | |
524 | if (!scmd->device->host->hostt->eh_bus_reset_handler) | |
525 | return FAILED; | |
526 | ||
527 | rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); | |
528 | ||
529 | if (rtn == SUCCESS) { | |
530 | if (!scmd->device->host->hostt->skip_settle_delay) | |
531 | ssleep(BUS_RESET_SETTLE_TIME); | |
532 | spin_lock_irqsave(scmd->device->host->host_lock, flags); | |
533 | scsi_report_bus_reset(scmd->device->host, | |
534 | scmd_channel(scmd)); | |
535 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); | |
536 | } | |
537 | ||
538 | return rtn; | |
539 | } | |
540 | ||
30bd7df8 MC |
541 | static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) |
542 | { | |
543 | sdev->was_reset = 1; | |
544 | sdev->expecting_cc_ua = 1; | |
545 | } | |
546 | ||
547 | /** | |
548 | * scsi_try_target_reset - Ask host to perform a target reset | |
549 | * @scmd: SCSI cmd used to send a target reset | |
550 | * | |
551 | * Notes: | |
552 | * There is no timeout for this operation. if this operation is | |
553 | * unreliable for a given host, then the host itself needs to put a | |
554 | * timer on it, and set the host back to a consistent state prior to | |
555 | * returning. | |
556 | */ | |
557 | static int scsi_try_target_reset(struct scsi_cmnd *scmd) | |
558 | { | |
559 | unsigned long flags; | |
560 | int rtn; | |
561 | ||
562 | if (!scmd->device->host->hostt->eh_target_reset_handler) | |
563 | return FAILED; | |
564 | ||
565 | rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd); | |
566 | if (rtn == SUCCESS) { | |
567 | spin_lock_irqsave(scmd->device->host->host_lock, flags); | |
568 | __starget_for_each_device(scsi_target(scmd->device), NULL, | |
569 | __scsi_report_device_reset); | |
570 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); | |
571 | } | |
572 | ||
573 | return rtn; | |
574 | } | |
575 | ||
292148f8 BK |
576 | /** |
577 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev | |
578 | * @scmd: SCSI cmd used to send BDR | |
579 | * | |
580 | * Notes: | |
581 | * There is no timeout for this operation. if this operation is | |
582 | * unreliable for a given host, then the host itself needs to put a | |
583 | * timer on it, and set the host back to a consistent state prior to | |
584 | * returning. | |
dc8875e1 | 585 | */ |
292148f8 BK |
586 | static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) |
587 | { | |
588 | int rtn; | |
589 | ||
590 | if (!scmd->device->host->hostt->eh_device_reset_handler) | |
591 | return FAILED; | |
592 | ||
593 | rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); | |
30bd7df8 MC |
594 | if (rtn == SUCCESS) |
595 | __scsi_report_device_reset(scmd->device, NULL); | |
292148f8 BK |
596 | return rtn; |
597 | } | |
598 | ||
599 | static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) | |
600 | { | |
601 | if (!scmd->device->host->hostt->eh_abort_handler) | |
602 | return FAILED; | |
603 | ||
604 | return scmd->device->host->hostt->eh_abort_handler(scmd); | |
605 | } | |
606 | ||
607 | /** | |
608 | * scsi_try_to_abort_cmd - Ask host to abort a running command. | |
609 | * @scmd: SCSI cmd to abort from Lower Level. | |
610 | * | |
611 | * Notes: | |
612 | * This function will not return until the user's completion function | |
613 | * has been called. there is no timeout on this operation. if the | |
614 | * author of the low-level driver wishes this operation to be timed, | |
615 | * they can provide this facility themselves. helper functions in | |
616 | * scsi_error.c can be supplied to make this easier to do. | |
dc8875e1 | 617 | */ |
292148f8 BK |
618 | static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) |
619 | { | |
620 | /* | |
621 | * scsi_done was called just after the command timed out and before | |
622 | * we had a chance to process it. (db) | |
623 | */ | |
624 | if (scmd->serial_number == 0) | |
625 | return SUCCESS; | |
626 | return __scsi_try_to_abort_cmd(scmd); | |
627 | } | |
628 | ||
629 | static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) | |
630 | { | |
631 | if (__scsi_try_to_abort_cmd(scmd) != SUCCESS) | |
632 | if (scsi_try_bus_device_reset(scmd) != SUCCESS) | |
30bd7df8 MC |
633 | if (scsi_try_target_reset(scmd) != SUCCESS) |
634 | if (scsi_try_bus_reset(scmd) != SUCCESS) | |
635 | scsi_try_host_reset(scmd); | |
292148f8 BK |
636 | } |
637 | ||
1da177e4 | 638 | /** |
e1c23468 | 639 | * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory |
2dc611de | 640 | * @scmd: SCSI command structure to hijack |
e1c23468 | 641 | * @ses: structure to save restore information |
55db6c1b | 642 | * @cmnd: CDB to send. Can be NULL if no new cmnd is needed |
64a87b24 | 643 | * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) |
55db6c1b | 644 | * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) |
2dc611de | 645 | * |
e1c23468 | 646 | * This function is used to save a scsi command information before re-execution |
55db6c1b BH |
647 | * as part of the error recovery process. If @sense_bytes is 0 the command |
648 | * sent must be one that does not transfer any data. If @sense_bytes != 0 | |
649 | * @cmnd is ignored and this functions sets up a REQUEST_SENSE command | |
650 | * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. | |
dc8875e1 | 651 | */ |
e1c23468 BH |
652 | void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, |
653 | unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) | |
1da177e4 | 654 | { |
f59114b7 | 655 | struct scsi_device *sdev = scmd->device; |
1da177e4 | 656 | |
631c228c CH |
657 | /* |
658 | * We need saved copies of a number of fields - this is because | |
659 | * error handling may need to overwrite these with different values | |
660 | * to run different commands, and once error handling is complete, | |
661 | * we will need to restore these values prior to running the actual | |
662 | * command. | |
663 | */ | |
e1c23468 | 664 | ses->cmd_len = scmd->cmd_len; |
64a87b24 | 665 | ses->cmnd = scmd->cmnd; |
e1c23468 | 666 | ses->data_direction = scmd->sc_data_direction; |
30b0c37b | 667 | ses->sdb = scmd->sdb; |
6f9a35e2 | 668 | ses->next_rq = scmd->request->next_rq; |
e1c23468 | 669 | ses->result = scmd->result; |
12265709 | 670 | ses->underflow = scmd->underflow; |
db007fc5 | 671 | ses->prot_op = scmd->prot_op; |
631c228c | 672 | |
db007fc5 | 673 | scmd->prot_op = SCSI_PROT_NORMAL; |
64a87b24 BH |
674 | scmd->cmnd = ses->eh_cmnd; |
675 | memset(scmd->cmnd, 0, BLK_MAX_CDB); | |
30b0c37b | 676 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
6f9a35e2 | 677 | scmd->request->next_rq = NULL; |
30b0c37b | 678 | |
55db6c1b | 679 | if (sense_bytes) { |
30b0c37b BH |
680 | scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, |
681 | sense_bytes); | |
e1c23468 | 682 | sg_init_one(&ses->sense_sgl, scmd->sense_buffer, |
30b0c37b BH |
683 | scmd->sdb.length); |
684 | scmd->sdb.table.sgl = &ses->sense_sgl; | |
55db6c1b | 685 | scmd->sc_data_direction = DMA_FROM_DEVICE; |
30b0c37b | 686 | scmd->sdb.table.nents = 1; |
55db6c1b | 687 | scmd->cmnd[0] = REQUEST_SENSE; |
30b0c37b | 688 | scmd->cmnd[4] = scmd->sdb.length; |
55db6c1b | 689 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
631c228c | 690 | } else { |
631c228c | 691 | scmd->sc_data_direction = DMA_NONE; |
55db6c1b | 692 | if (cmnd) { |
64a87b24 | 693 | BUG_ON(cmnd_size > BLK_MAX_CDB); |
55db6c1b BH |
694 | memcpy(scmd->cmnd, cmnd, cmnd_size); |
695 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); | |
696 | } | |
631c228c CH |
697 | } |
698 | ||
699 | scmd->underflow = 0; | |
631c228c | 700 | |
55db6c1b | 701 | if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) |
1da177e4 | 702 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
f59114b7 | 703 | (sdev->lun << 5 & 0xe0); |
1da177e4 | 704 | |
631c228c CH |
705 | /* |
706 | * Zero the sense buffer. The scsi spec mandates that any | |
707 | * untransferred sense data should be interpreted as being zero. | |
708 | */ | |
b80ca4f7 | 709 | memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); |
e1c23468 BH |
710 | } |
711 | EXPORT_SYMBOL(scsi_eh_prep_cmnd); | |
712 | ||
713 | /** | |
714 | * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory | |
715 | * @scmd: SCSI command structure to restore | |
716 | * @ses: saved information from a coresponding call to scsi_prep_eh_cmnd | |
717 | * | |
718 | * Undo any damage done by above scsi_prep_eh_cmnd(). | |
dc8875e1 | 719 | */ |
e1c23468 BH |
720 | void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) |
721 | { | |
722 | /* | |
723 | * Restore original data | |
724 | */ | |
725 | scmd->cmd_len = ses->cmd_len; | |
64a87b24 | 726 | scmd->cmnd = ses->cmnd; |
e1c23468 | 727 | scmd->sc_data_direction = ses->data_direction; |
30b0c37b | 728 | scmd->sdb = ses->sdb; |
6f9a35e2 | 729 | scmd->request->next_rq = ses->next_rq; |
e1c23468 | 730 | scmd->result = ses->result; |
12265709 | 731 | scmd->underflow = ses->underflow; |
db007fc5 | 732 | scmd->prot_op = ses->prot_op; |
e1c23468 BH |
733 | } |
734 | EXPORT_SYMBOL(scsi_eh_restore_cmnd); | |
631c228c | 735 | |
e1c23468 BH |
736 | /** |
737 | * scsi_send_eh_cmnd - submit a scsi command as part of error recory | |
738 | * @scmd: SCSI command structure to hijack | |
739 | * @cmnd: CDB to send | |
740 | * @cmnd_size: size in bytes of @cmnd | |
741 | * @timeout: timeout for this request | |
742 | * @sense_bytes: size of sense data to copy or 0 | |
743 | * | |
744 | * This function is used to send a scsi command down to a target device | |
745 | * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. | |
746 | * | |
747 | * Return value: | |
748 | * SUCCESS or FAILED or NEEDS_RETRY | |
dc8875e1 | 749 | */ |
e1c23468 BH |
750 | static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, |
751 | int cmnd_size, int timeout, unsigned sense_bytes) | |
752 | { | |
753 | struct scsi_device *sdev = scmd->device; | |
754 | struct Scsi_Host *shost = sdev->host; | |
755 | DECLARE_COMPLETION_ONSTACK(done); | |
756 | unsigned long timeleft; | |
757 | unsigned long flags; | |
758 | struct scsi_eh_save ses; | |
759 | int rtn; | |
760 | ||
761 | scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); | |
7dfdc9a5 | 762 | shost->eh_action = &done; |
1da177e4 | 763 | |
f59114b7 | 764 | spin_lock_irqsave(shost->host_lock, flags); |
1da177e4 | 765 | scsi_log_send(scmd); |
f59114b7 TH |
766 | shost->hostt->queuecommand(scmd, scsi_eh_done); |
767 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1da177e4 | 768 | |
7dfdc9a5 | 769 | timeleft = wait_for_completion_timeout(&done, timeout); |
1da177e4 | 770 | |
f59114b7 | 771 | shost->eh_action = NULL; |
1da177e4 | 772 | |
7dfdc9a5 | 773 | scsi_log_completion(scmd, SUCCESS); |
1da177e4 | 774 | |
7dfdc9a5 CH |
775 | SCSI_LOG_ERROR_RECOVERY(3, |
776 | printk("%s: scmd: %p, timeleft: %ld\n", | |
cadbd4a5 | 777 | __func__, scmd, timeleft)); |
1da177e4 LT |
778 | |
779 | /* | |
7dfdc9a5 CH |
780 | * If there is time left scsi_eh_done got called, and we will |
781 | * examine the actual status codes to see whether the command | |
782 | * actually did complete normally, else tell the host to forget | |
783 | * about this command. | |
1da177e4 | 784 | */ |
7dfdc9a5 | 785 | if (timeleft) { |
1da177e4 LT |
786 | rtn = scsi_eh_completed_normally(scmd); |
787 | SCSI_LOG_ERROR_RECOVERY(3, | |
788 | printk("%s: scsi_eh_completed_normally %x\n", | |
cadbd4a5 | 789 | __func__, rtn)); |
7dfdc9a5 | 790 | |
1da177e4 LT |
791 | switch (rtn) { |
792 | case SUCCESS: | |
793 | case NEEDS_RETRY: | |
794 | case FAILED: | |
795 | break; | |
796 | default: | |
797 | rtn = FAILED; | |
798 | break; | |
799 | } | |
7dfdc9a5 | 800 | } else { |
292148f8 | 801 | scsi_abort_eh_cmnd(scmd); |
7dfdc9a5 | 802 | rtn = FAILED; |
1da177e4 LT |
803 | } |
804 | ||
e1c23468 | 805 | scsi_eh_restore_cmnd(scmd, &ses); |
1da177e4 LT |
806 | return rtn; |
807 | } | |
808 | ||
809 | /** | |
810 | * scsi_request_sense - Request sense data from a particular target. | |
811 | * @scmd: SCSI cmd for request sense. | |
812 | * | |
813 | * Notes: | |
814 | * Some hosts automatically obtain this information, others require | |
815 | * that we obtain it on our own. This function will *not* return until | |
816 | * the command either times out, or it completes. | |
dc8875e1 | 817 | */ |
1da177e4 LT |
818 | static int scsi_request_sense(struct scsi_cmnd *scmd) |
819 | { | |
55db6c1b | 820 | return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0); |
1da177e4 LT |
821 | } |
822 | ||
823 | /** | |
824 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. | |
825 | * @scmd: Original SCSI cmd that eh has finished. | |
826 | * @done_q: Queue for processed commands. | |
827 | * | |
828 | * Notes: | |
829 | * We don't want to use the normal command completion while we are are | |
830 | * still handling errors - it may cause other commands to be queued, | |
eb44820c | 831 | * and that would disturb what we are doing. Thus we really want to |
1da177e4 LT |
832 | * keep a list of pending commands for final completion, and once we |
833 | * are ready to leave error handling we handle completion for real. | |
dc8875e1 | 834 | */ |
041c5fc3 | 835 | void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) |
1da177e4 LT |
836 | { |
837 | scmd->device->host->host_failed--; | |
3111b0d1 | 838 | scmd->eh_eflags = 0; |
1da177e4 LT |
839 | list_move_tail(&scmd->eh_entry, done_q); |
840 | } | |
041c5fc3 | 841 | EXPORT_SYMBOL(scsi_eh_finish_cmd); |
1da177e4 LT |
842 | |
843 | /** | |
844 | * scsi_eh_get_sense - Get device sense data. | |
845 | * @work_q: Queue of commands to process. | |
eb44820c | 846 | * @done_q: Queue of processed commands. |
1da177e4 LT |
847 | * |
848 | * Description: | |
849 | * See if we need to request sense information. if so, then get it | |
850 | * now, so we have a better idea of what to do. | |
851 | * | |
852 | * Notes: | |
853 | * This has the unfortunate side effect that if a shost adapter does | |
eb44820c | 854 | * not automatically request sense information, we end up shutting |
1da177e4 LT |
855 | * it down before we request it. |
856 | * | |
857 | * All drivers should request sense information internally these days, | |
858 | * so for now all I have to say is tough noogies if you end up in here. | |
859 | * | |
860 | * XXX: Long term this code should go away, but that needs an audit of | |
861 | * all LLDDs first. | |
dc8875e1 | 862 | */ |
dca84e46 DW |
863 | int scsi_eh_get_sense(struct list_head *work_q, |
864 | struct list_head *done_q) | |
1da177e4 | 865 | { |
937abeaa | 866 | struct scsi_cmnd *scmd, *next; |
1da177e4 LT |
867 | int rtn; |
868 | ||
937abeaa | 869 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
3111b0d1 | 870 | if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || |
1da177e4 LT |
871 | SCSI_SENSE_VALID(scmd)) |
872 | continue; | |
873 | ||
3bf743e7 JG |
874 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
875 | "%s: requesting sense\n", | |
876 | current->comm)); | |
1da177e4 LT |
877 | rtn = scsi_request_sense(scmd); |
878 | if (rtn != SUCCESS) | |
879 | continue; | |
880 | ||
881 | SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" | |
882 | " result %x\n", scmd, | |
883 | scmd->result)); | |
884 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); | |
885 | ||
886 | rtn = scsi_decide_disposition(scmd); | |
887 | ||
888 | /* | |
889 | * if the result was normal, then just pass it along to the | |
890 | * upper level. | |
891 | */ | |
892 | if (rtn == SUCCESS) | |
893 | /* we don't want this command reissued, just | |
894 | * finished with the sense data, so set | |
895 | * retries to the max allowed to ensure it | |
896 | * won't get reissued */ | |
897 | scmd->retries = scmd->allowed; | |
898 | else if (rtn != NEEDS_RETRY) | |
899 | continue; | |
900 | ||
901 | scsi_eh_finish_cmd(scmd, done_q); | |
902 | } | |
903 | ||
904 | return list_empty(work_q); | |
905 | } | |
dca84e46 | 906 | EXPORT_SYMBOL_GPL(scsi_eh_get_sense); |
1da177e4 | 907 | |
1da177e4 LT |
908 | /** |
909 | * scsi_eh_tur - Send TUR to device. | |
eb44820c | 910 | * @scmd: &scsi_cmnd to send TUR |
1da177e4 LT |
911 | * |
912 | * Return value: | |
913 | * 0 - Device is ready. 1 - Device NOT ready. | |
dc8875e1 | 914 | */ |
1da177e4 LT |
915 | static int scsi_eh_tur(struct scsi_cmnd *scmd) |
916 | { | |
917 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; | |
918 | int retry_cnt = 1, rtn; | |
919 | ||
920 | retry_tur: | |
0db99e33 | 921 | rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0); |
1da177e4 | 922 | |
1da177e4 | 923 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", |
cadbd4a5 | 924 | __func__, scmd, rtn)); |
631c228c CH |
925 | |
926 | switch (rtn) { | |
927 | case NEEDS_RETRY: | |
1da177e4 LT |
928 | if (retry_cnt--) |
929 | goto retry_tur; | |
631c228c CH |
930 | /*FALLTHRU*/ |
931 | case SUCCESS: | |
e47373ec | 932 | return 0; |
631c228c CH |
933 | default: |
934 | return 1; | |
e47373ec | 935 | } |
1da177e4 LT |
936 | } |
937 | ||
938 | /** | |
eb44820c RL |
939 | * scsi_eh_abort_cmds - abort pending commands. |
940 | * @work_q: &list_head for pending commands. | |
941 | * @done_q: &list_head for processed commands. | |
1da177e4 LT |
942 | * |
943 | * Decription: | |
944 | * Try and see whether or not it makes sense to try and abort the | |
eb44820c RL |
945 | * running command. This only works out to be the case if we have one |
946 | * command that has timed out. If the command simply failed, it makes | |
1da177e4 LT |
947 | * no sense to try and abort the command, since as far as the shost |
948 | * adapter is concerned, it isn't running. | |
dc8875e1 | 949 | */ |
1da177e4 LT |
950 | static int scsi_eh_abort_cmds(struct list_head *work_q, |
951 | struct list_head *done_q) | |
952 | { | |
937abeaa | 953 | struct scsi_cmnd *scmd, *next; |
1da177e4 LT |
954 | int rtn; |
955 | ||
937abeaa | 956 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
3111b0d1 | 957 | if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) |
1da177e4 LT |
958 | continue; |
959 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" | |
960 | "0x%p\n", current->comm, | |
961 | scmd)); | |
962 | rtn = scsi_try_to_abort_cmd(scmd); | |
963 | if (rtn == SUCCESS) { | |
3111b0d1 | 964 | scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; |
1da177e4 LT |
965 | if (!scsi_device_online(scmd->device) || |
966 | !scsi_eh_tur(scmd)) { | |
967 | scsi_eh_finish_cmd(scmd, done_q); | |
968 | } | |
969 | ||
970 | } else | |
971 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" | |
972 | " cmd failed:" | |
973 | "0x%p\n", | |
974 | current->comm, | |
975 | scmd)); | |
976 | } | |
977 | ||
978 | return list_empty(work_q); | |
979 | } | |
980 | ||
1da177e4 LT |
981 | /** |
982 | * scsi_eh_try_stu - Send START_UNIT to device. | |
eb44820c | 983 | * @scmd: &scsi_cmnd to send START_UNIT |
1da177e4 LT |
984 | * |
985 | * Return value: | |
986 | * 0 - Device is ready. 1 - Device NOT ready. | |
dc8875e1 | 987 | */ |
1da177e4 LT |
988 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) |
989 | { | |
990 | static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; | |
1da177e4 | 991 | |
631c228c | 992 | if (scmd->device->allow_restart) { |
ed773e66 BK |
993 | int i, rtn = NEEDS_RETRY; |
994 | ||
995 | for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) | |
996 | rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, | |
e555db93 | 997 | scmd->device->timeout, 0); |
1da177e4 | 998 | |
631c228c CH |
999 | if (rtn == SUCCESS) |
1000 | return 0; | |
1001 | } | |
1da177e4 | 1002 | |
1da177e4 LT |
1003 | return 1; |
1004 | } | |
1005 | ||
1006 | /** | |
1007 | * scsi_eh_stu - send START_UNIT if needed | |
eb44820c RL |
1008 | * @shost: &scsi host being recovered. |
1009 | * @work_q: &list_head for pending commands. | |
1010 | * @done_q: &list_head for processed commands. | |
1da177e4 LT |
1011 | * |
1012 | * Notes: | |
1013 | * If commands are failing due to not ready, initializing command required, | |
1014 | * try revalidating the device, which will end up sending a start unit. | |
dc8875e1 | 1015 | */ |
1da177e4 LT |
1016 | static int scsi_eh_stu(struct Scsi_Host *shost, |
1017 | struct list_head *work_q, | |
1018 | struct list_head *done_q) | |
1019 | { | |
937abeaa | 1020 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
1da177e4 LT |
1021 | struct scsi_device *sdev; |
1022 | ||
1023 | shost_for_each_device(sdev, shost) { | |
1024 | stu_scmd = NULL; | |
1025 | list_for_each_entry(scmd, work_q, eh_entry) | |
1026 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && | |
1027 | scsi_check_sense(scmd) == FAILED ) { | |
1028 | stu_scmd = scmd; | |
1029 | break; | |
1030 | } | |
1031 | ||
1032 | if (!stu_scmd) | |
1033 | continue; | |
1034 | ||
1035 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" | |
1036 | " 0x%p\n", current->comm, sdev)); | |
1037 | ||
1038 | if (!scsi_eh_try_stu(stu_scmd)) { | |
1039 | if (!scsi_device_online(sdev) || | |
1040 | !scsi_eh_tur(stu_scmd)) { | |
937abeaa CH |
1041 | list_for_each_entry_safe(scmd, next, |
1042 | work_q, eh_entry) { | |
1da177e4 LT |
1043 | if (scmd->device == sdev) |
1044 | scsi_eh_finish_cmd(scmd, done_q); | |
1045 | } | |
1046 | } | |
1047 | } else { | |
1048 | SCSI_LOG_ERROR_RECOVERY(3, | |
1049 | printk("%s: START_UNIT failed to sdev:" | |
1050 | " 0x%p\n", current->comm, sdev)); | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | return list_empty(work_q); | |
1055 | } | |
1056 | ||
1057 | ||
1058 | /** | |
1059 | * scsi_eh_bus_device_reset - send bdr if needed | |
1060 | * @shost: scsi host being recovered. | |
eb44820c RL |
1061 | * @work_q: &list_head for pending commands. |
1062 | * @done_q: &list_head for processed commands. | |
1da177e4 LT |
1063 | * |
1064 | * Notes: | |
eb44820c | 1065 | * Try a bus device reset. Still, look to see whether we have multiple |
1da177e4 LT |
1066 | * devices that are jammed or not - if we have multiple devices, it |
1067 | * makes no sense to try bus_device_reset - we really would need to try | |
1068 | * a bus_reset instead. | |
dc8875e1 | 1069 | */ |
1da177e4 LT |
1070 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, |
1071 | struct list_head *work_q, | |
1072 | struct list_head *done_q) | |
1073 | { | |
937abeaa | 1074 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
1da177e4 LT |
1075 | struct scsi_device *sdev; |
1076 | int rtn; | |
1077 | ||
1078 | shost_for_each_device(sdev, shost) { | |
1079 | bdr_scmd = NULL; | |
1080 | list_for_each_entry(scmd, work_q, eh_entry) | |
1081 | if (scmd->device == sdev) { | |
1082 | bdr_scmd = scmd; | |
1083 | break; | |
1084 | } | |
1085 | ||
1086 | if (!bdr_scmd) | |
1087 | continue; | |
1088 | ||
1089 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" | |
1090 | " 0x%p\n", current->comm, | |
1091 | sdev)); | |
1092 | rtn = scsi_try_bus_device_reset(bdr_scmd); | |
1093 | if (rtn == SUCCESS) { | |
1094 | if (!scsi_device_online(sdev) || | |
1095 | !scsi_eh_tur(bdr_scmd)) { | |
937abeaa CH |
1096 | list_for_each_entry_safe(scmd, next, |
1097 | work_q, eh_entry) { | |
1da177e4 LT |
1098 | if (scmd->device == sdev) |
1099 | scsi_eh_finish_cmd(scmd, | |
1100 | done_q); | |
1101 | } | |
1102 | } | |
1103 | } else { | |
1104 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" | |
1105 | " failed sdev:" | |
1106 | "0x%p\n", | |
1107 | current->comm, | |
1108 | sdev)); | |
1109 | } | |
1110 | } | |
1111 | ||
1112 | return list_empty(work_q); | |
1113 | } | |
1114 | ||
30bd7df8 MC |
1115 | /** |
1116 | * scsi_eh_target_reset - send target reset if needed | |
1117 | * @shost: scsi host being recovered. | |
1118 | * @work_q: &list_head for pending commands. | |
1119 | * @done_q: &list_head for processed commands. | |
1120 | * | |
1121 | * Notes: | |
1122 | * Try a target reset. | |
1123 | */ | |
1124 | static int scsi_eh_target_reset(struct Scsi_Host *shost, | |
1125 | struct list_head *work_q, | |
1126 | struct list_head *done_q) | |
1127 | { | |
1128 | struct scsi_cmnd *scmd, *tgtr_scmd, *next; | |
1129 | unsigned int id; | |
1130 | int rtn; | |
1131 | ||
1132 | for (id = 0; id <= shost->max_id; id++) { | |
1133 | tgtr_scmd = NULL; | |
1134 | list_for_each_entry(scmd, work_q, eh_entry) { | |
1135 | if (id == scmd_id(scmd)) { | |
1136 | tgtr_scmd = scmd; | |
1137 | break; | |
1138 | } | |
1139 | } | |
1140 | if (!tgtr_scmd) | |
1141 | continue; | |
1142 | ||
1143 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset " | |
1144 | "to target %d\n", | |
1145 | current->comm, id)); | |
1146 | rtn = scsi_try_target_reset(tgtr_scmd); | |
1147 | if (rtn == SUCCESS) { | |
1148 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { | |
1149 | if (id == scmd_id(scmd)) | |
1150 | if (!scsi_device_online(scmd->device) || | |
1151 | !scsi_eh_tur(tgtr_scmd)) | |
1152 | scsi_eh_finish_cmd(scmd, | |
1153 | done_q); | |
1154 | } | |
1155 | } else | |
1156 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset" | |
1157 | " failed target: " | |
1158 | "%d\n", | |
1159 | current->comm, id)); | |
1160 | } | |
1161 | ||
1162 | return list_empty(work_q); | |
1163 | } | |
1164 | ||
1da177e4 LT |
1165 | /** |
1166 | * scsi_eh_bus_reset - send a bus reset | |
eb44820c RL |
1167 | * @shost: &scsi host being recovered. |
1168 | * @work_q: &list_head for pending commands. | |
1169 | * @done_q: &list_head for processed commands. | |
dc8875e1 | 1170 | */ |
1da177e4 LT |
1171 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, |
1172 | struct list_head *work_q, | |
1173 | struct list_head *done_q) | |
1174 | { | |
937abeaa | 1175 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
1da177e4 LT |
1176 | unsigned int channel; |
1177 | int rtn; | |
1178 | ||
1179 | /* | |
1180 | * we really want to loop over the various channels, and do this on | |
1181 | * a channel by channel basis. we should also check to see if any | |
1182 | * of the failed commands are on soft_reset devices, and if so, skip | |
1183 | * the reset. | |
1184 | */ | |
1185 | ||
1186 | for (channel = 0; channel <= shost->max_channel; channel++) { | |
1187 | chan_scmd = NULL; | |
1188 | list_for_each_entry(scmd, work_q, eh_entry) { | |
422c0d61 | 1189 | if (channel == scmd_channel(scmd)) { |
1da177e4 LT |
1190 | chan_scmd = scmd; |
1191 | break; | |
1192 | /* | |
1193 | * FIXME add back in some support for | |
1194 | * soft_reset devices. | |
1195 | */ | |
1196 | } | |
1197 | } | |
1198 | ||
1199 | if (!chan_scmd) | |
1200 | continue; | |
1201 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" | |
1202 | " %d\n", current->comm, | |
1203 | channel)); | |
1204 | rtn = scsi_try_bus_reset(chan_scmd); | |
1205 | if (rtn == SUCCESS) { | |
937abeaa | 1206 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
422c0d61 | 1207 | if (channel == scmd_channel(scmd)) |
1da177e4 LT |
1208 | if (!scsi_device_online(scmd->device) || |
1209 | !scsi_eh_tur(scmd)) | |
1210 | scsi_eh_finish_cmd(scmd, | |
1211 | done_q); | |
1212 | } | |
1213 | } else { | |
1214 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" | |
1215 | " failed chan: %d\n", | |
1216 | current->comm, | |
1217 | channel)); | |
1218 | } | |
1219 | } | |
1220 | return list_empty(work_q); | |
1221 | } | |
1222 | ||
1223 | /** | |
1224 | * scsi_eh_host_reset - send a host reset | |
1225 | * @work_q: list_head for processed commands. | |
1226 | * @done_q: list_head for processed commands. | |
dc8875e1 | 1227 | */ |
1da177e4 LT |
1228 | static int scsi_eh_host_reset(struct list_head *work_q, |
1229 | struct list_head *done_q) | |
1230 | { | |
937abeaa | 1231 | struct scsi_cmnd *scmd, *next; |
1da177e4 | 1232 | int rtn; |
1da177e4 LT |
1233 | |
1234 | if (!list_empty(work_q)) { | |
1235 | scmd = list_entry(work_q->next, | |
1236 | struct scsi_cmnd, eh_entry); | |
1237 | ||
1238 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" | |
1239 | , current->comm)); | |
1240 | ||
1241 | rtn = scsi_try_host_reset(scmd); | |
1242 | if (rtn == SUCCESS) { | |
937abeaa | 1243 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
1da177e4 LT |
1244 | if (!scsi_device_online(scmd->device) || |
1245 | (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || | |
1246 | !scsi_eh_tur(scmd)) | |
1247 | scsi_eh_finish_cmd(scmd, done_q); | |
1248 | } | |
1249 | } else { | |
1250 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" | |
1251 | " failed\n", | |
1252 | current->comm)); | |
1253 | } | |
1254 | } | |
1255 | return list_empty(work_q); | |
1256 | } | |
1257 | ||
1258 | /** | |
1259 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover | |
1260 | * @work_q: list_head for processed commands. | |
1261 | * @done_q: list_head for processed commands. | |
dc8875e1 | 1262 | */ |
1da177e4 LT |
1263 | static void scsi_eh_offline_sdevs(struct list_head *work_q, |
1264 | struct list_head *done_q) | |
1265 | { | |
937abeaa | 1266 | struct scsi_cmnd *scmd, *next; |
1da177e4 | 1267 | |
937abeaa | 1268 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
31765d7d MW |
1269 | sdev_printk(KERN_INFO, scmd->device, "Device offlined - " |
1270 | "not ready after error recovery\n"); | |
1da177e4 | 1271 | scsi_device_set_state(scmd->device, SDEV_OFFLINE); |
3111b0d1 | 1272 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { |
1da177e4 LT |
1273 | /* |
1274 | * FIXME: Handle lost cmds. | |
1275 | */ | |
1276 | } | |
1277 | scsi_eh_finish_cmd(scmd, done_q); | |
1278 | } | |
1279 | return; | |
1280 | } | |
1281 | ||
1282 | /** | |
1283 | * scsi_decide_disposition - Disposition a cmd on return from LLD. | |
1284 | * @scmd: SCSI cmd to examine. | |
1285 | * | |
1286 | * Notes: | |
1287 | * This is *only* called when we are examining the status after sending | |
1288 | * out the actual data command. any commands that are queued for error | |
1289 | * recovery (e.g. test_unit_ready) do *not* come through here. | |
1290 | * | |
1291 | * When this routine returns failed, it means the error handler thread | |
1292 | * is woken. In cases where the error code indicates an error that | |
1293 | * doesn't require the error handler read (i.e. we don't need to | |
1294 | * abort/reset), this function should return SUCCESS. | |
dc8875e1 | 1295 | */ |
1da177e4 LT |
1296 | int scsi_decide_disposition(struct scsi_cmnd *scmd) |
1297 | { | |
1298 | int rtn; | |
1299 | ||
1300 | /* | |
1301 | * if the device is offline, then we clearly just pass the result back | |
1302 | * up to the top level. | |
1303 | */ | |
1304 | if (!scsi_device_online(scmd->device)) { | |
1305 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" | |
1306 | " as SUCCESS\n", | |
cadbd4a5 | 1307 | __func__)); |
1da177e4 LT |
1308 | return SUCCESS; |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * first check the host byte, to see if there is anything in there | |
1313 | * that would indicate what we need to do. | |
1314 | */ | |
1315 | switch (host_byte(scmd->result)) { | |
1316 | case DID_PASSTHROUGH: | |
1317 | /* | |
1318 | * no matter what, pass this through to the upper layer. | |
1319 | * nuke this special code so that it looks like we are saying | |
1320 | * did_ok. | |
1321 | */ | |
1322 | scmd->result &= 0xff00ffff; | |
1323 | return SUCCESS; | |
1324 | case DID_OK: | |
1325 | /* | |
1326 | * looks good. drop through, and check the next byte. | |
1327 | */ | |
1328 | break; | |
1329 | case DID_NO_CONNECT: | |
1330 | case DID_BAD_TARGET: | |
1331 | case DID_ABORT: | |
1332 | /* | |
1333 | * note - this means that we just report the status back | |
1334 | * to the top level driver, not that we actually think | |
1335 | * that it indicates SUCCESS. | |
1336 | */ | |
1337 | return SUCCESS; | |
1338 | /* | |
1339 | * when the low level driver returns did_soft_error, | |
1340 | * it is responsible for keeping an internal retry counter | |
1341 | * in order to avoid endless loops (db) | |
1342 | * | |
1343 | * actually this is a bug in this function here. we should | |
1344 | * be mindful of the maximum number of retries specified | |
1345 | * and not get stuck in a loop. | |
1346 | */ | |
1347 | case DID_SOFT_ERROR: | |
1348 | goto maybe_retry; | |
1349 | case DID_IMM_RETRY: | |
1350 | return NEEDS_RETRY; | |
1351 | ||
bf341919 JB |
1352 | case DID_REQUEUE: |
1353 | return ADD_TO_MLQUEUE; | |
1354 | ||
1da177e4 LT |
1355 | case DID_ERROR: |
1356 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && | |
1357 | status_byte(scmd->result) == RESERVATION_CONFLICT) | |
1358 | /* | |
1359 | * execute reservation conflict processing code | |
1360 | * lower down | |
1361 | */ | |
1362 | break; | |
1363 | /* fallthrough */ | |
1364 | ||
1365 | case DID_BUS_BUSY: | |
1366 | case DID_PARITY: | |
1367 | goto maybe_retry; | |
1368 | case DID_TIME_OUT: | |
1369 | /* | |
1370 | * when we scan the bus, we get timeout messages for | |
1371 | * these commands if there is no device available. | |
1372 | * other hosts report did_no_connect for the same thing. | |
1373 | */ | |
1374 | if ((scmd->cmnd[0] == TEST_UNIT_READY || | |
1375 | scmd->cmnd[0] == INQUIRY)) { | |
1376 | return SUCCESS; | |
1377 | } else { | |
1378 | return FAILED; | |
1379 | } | |
1380 | case DID_RESET: | |
1381 | return SUCCESS; | |
1382 | default: | |
1383 | return FAILED; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * next, check the message byte. | |
1388 | */ | |
1389 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) | |
1390 | return FAILED; | |
1391 | ||
1392 | /* | |
1393 | * check the status byte to see if this indicates anything special. | |
1394 | */ | |
1395 | switch (status_byte(scmd->result)) { | |
1396 | case QUEUE_FULL: | |
1397 | /* | |
1398 | * the case of trying to send too many commands to a | |
1399 | * tagged queueing device. | |
1400 | */ | |
1401 | case BUSY: | |
1402 | /* | |
1403 | * device can't talk to us at the moment. Should only | |
1404 | * occur (SAM-3) when the task queue is empty, so will cause | |
1405 | * the empty queue handling to trigger a stall in the | |
1406 | * device. | |
1407 | */ | |
1408 | return ADD_TO_MLQUEUE; | |
1409 | case GOOD: | |
1410 | case COMMAND_TERMINATED: | |
1411 | case TASK_ABORTED: | |
1412 | return SUCCESS; | |
1413 | case CHECK_CONDITION: | |
1414 | rtn = scsi_check_sense(scmd); | |
1415 | if (rtn == NEEDS_RETRY) | |
1416 | goto maybe_retry; | |
1417 | /* if rtn == FAILED, we have no sense information; | |
1418 | * returning FAILED will wake the error handler thread | |
1419 | * to collect the sense and redo the decide | |
1420 | * disposition */ | |
1421 | return rtn; | |
1422 | case CONDITION_GOOD: | |
1423 | case INTERMEDIATE_GOOD: | |
1424 | case INTERMEDIATE_C_GOOD: | |
1425 | case ACA_ACTIVE: | |
1426 | /* | |
1427 | * who knows? FIXME(eric) | |
1428 | */ | |
1429 | return SUCCESS; | |
1430 | ||
1431 | case RESERVATION_CONFLICT: | |
9ccfc756 JB |
1432 | sdev_printk(KERN_INFO, scmd->device, |
1433 | "reservation conflict\n"); | |
1da177e4 LT |
1434 | return SUCCESS; /* causes immediate i/o error */ |
1435 | default: | |
1436 | return FAILED; | |
1437 | } | |
1438 | return FAILED; | |
1439 | ||
1440 | maybe_retry: | |
1441 | ||
1442 | /* we requeue for retry because the error was retryable, and | |
1443 | * the request was not marked fast fail. Note that above, | |
1444 | * even if the request is marked fast fail, we still requeue | |
1445 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ | |
8884efab | 1446 | if ((++scmd->retries) <= scmd->allowed |
1da177e4 LT |
1447 | && !blk_noretry_request(scmd->request)) { |
1448 | return NEEDS_RETRY; | |
1449 | } else { | |
1450 | /* | |
1451 | * no more retries - report this one back to upper level. | |
1452 | */ | |
1453 | return SUCCESS; | |
1454 | } | |
1455 | } | |
1456 | ||
1da177e4 LT |
1457 | /** |
1458 | * scsi_eh_lock_door - Prevent medium removal for the specified device | |
1459 | * @sdev: SCSI device to prevent medium removal | |
1460 | * | |
1461 | * Locking: | |
1462 | * We must be called from process context; scsi_allocate_request() | |
1463 | * may sleep. | |
1464 | * | |
1465 | * Notes: | |
1466 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the | |
1467 | * head of the devices request queue, and continue. | |
1468 | * | |
1469 | * Bugs: | |
1470 | * scsi_allocate_request() may sleep waiting for existing requests to | |
1471 | * be processed. However, since we haven't kicked off any request | |
1472 | * processing for this host, this may deadlock. | |
1473 | * | |
1474 | * If scsi_allocate_request() fails for what ever reason, we | |
1475 | * completely forget to lock the door. | |
dc8875e1 | 1476 | */ |
1da177e4 LT |
1477 | static void scsi_eh_lock_door(struct scsi_device *sdev) |
1478 | { | |
6e68af66 | 1479 | unsigned char cmnd[MAX_COMMAND_SIZE]; |
1da177e4 | 1480 | |
6e68af66 MC |
1481 | cmnd[0] = ALLOW_MEDIUM_REMOVAL; |
1482 | cmnd[1] = 0; | |
1483 | cmnd[2] = 0; | |
1484 | cmnd[3] = 0; | |
1485 | cmnd[4] = SCSI_REMOVAL_PREVENT; | |
1486 | cmnd[5] = 0; | |
1da177e4 | 1487 | |
bb1d1073 | 1488 | scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ, |
6e68af66 | 1489 | 5, NULL, NULL, GFP_KERNEL); |
1da177e4 LT |
1490 | } |
1491 | ||
1492 | ||
1493 | /** | |
1494 | * scsi_restart_operations - restart io operations to the specified host. | |
1495 | * @shost: Host we are restarting. | |
1496 | * | |
1497 | * Notes: | |
1498 | * When we entered the error handler, we blocked all further i/o to | |
1499 | * this device. we need to 'reverse' this process. | |
dc8875e1 | 1500 | */ |
1da177e4 LT |
1501 | static void scsi_restart_operations(struct Scsi_Host *shost) |
1502 | { | |
1503 | struct scsi_device *sdev; | |
939647ee | 1504 | unsigned long flags; |
1da177e4 LT |
1505 | |
1506 | /* | |
1507 | * If the door was locked, we need to insert a door lock request | |
1508 | * onto the head of the SCSI request queue for the device. There | |
1509 | * is no point trying to lock the door of an off-line device. | |
1510 | */ | |
1511 | shost_for_each_device(sdev, shost) { | |
1512 | if (scsi_device_online(sdev) && sdev->locked) | |
1513 | scsi_eh_lock_door(sdev); | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * next free up anything directly waiting upon the host. this | |
1518 | * will be requests for character device operations, and also for | |
1519 | * ioctls to queued block devices. | |
1520 | */ | |
1521 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", | |
cadbd4a5 | 1522 | __func__)); |
1da177e4 | 1523 | |
939647ee JB |
1524 | spin_lock_irqsave(shost->host_lock, flags); |
1525 | if (scsi_host_set_state(shost, SHOST_RUNNING)) | |
1526 | if (scsi_host_set_state(shost, SHOST_CANCEL)) | |
1527 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); | |
1528 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1da177e4 LT |
1529 | |
1530 | wake_up(&shost->host_wait); | |
1531 | ||
1532 | /* | |
1533 | * finally we need to re-initiate requests that may be pending. we will | |
1534 | * have had everything blocked while error handling is taking place, and | |
1535 | * now that error recovery is done, we will need to ensure that these | |
1536 | * requests are started. | |
1537 | */ | |
1538 | scsi_run_host_queues(shost); | |
1539 | } | |
1540 | ||
1541 | /** | |
1542 | * scsi_eh_ready_devs - check device ready state and recover if not. | |
1543 | * @shost: host to be recovered. | |
eb44820c RL |
1544 | * @work_q: &list_head for pending commands. |
1545 | * @done_q: &list_head for processed commands. | |
dc8875e1 | 1546 | */ |
dca84e46 DW |
1547 | void scsi_eh_ready_devs(struct Scsi_Host *shost, |
1548 | struct list_head *work_q, | |
1549 | struct list_head *done_q) | |
1da177e4 LT |
1550 | { |
1551 | if (!scsi_eh_stu(shost, work_q, done_q)) | |
1552 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) | |
30bd7df8 MC |
1553 | if (!scsi_eh_target_reset(shost, work_q, done_q)) |
1554 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) | |
1555 | if (!scsi_eh_host_reset(work_q, done_q)) | |
1556 | scsi_eh_offline_sdevs(work_q, | |
1557 | done_q); | |
1da177e4 | 1558 | } |
dca84e46 | 1559 | EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); |
1da177e4 LT |
1560 | |
1561 | /** | |
1562 | * scsi_eh_flush_done_q - finish processed commands or retry them. | |
1563 | * @done_q: list_head of processed commands. | |
dc8875e1 | 1564 | */ |
041c5fc3 | 1565 | void scsi_eh_flush_done_q(struct list_head *done_q) |
1da177e4 | 1566 | { |
937abeaa | 1567 | struct scsi_cmnd *scmd, *next; |
1da177e4 | 1568 | |
937abeaa CH |
1569 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
1570 | list_del_init(&scmd->eh_entry); | |
1da177e4 LT |
1571 | if (scsi_device_online(scmd->device) && |
1572 | !blk_noretry_request(scmd->request) && | |
8884efab | 1573 | (++scmd->retries <= scmd->allowed)) { |
1da177e4 LT |
1574 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" |
1575 | " retry cmd: %p\n", | |
1576 | current->comm, | |
1577 | scmd)); | |
1578 | scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); | |
1579 | } else { | |
793698ce PM |
1580 | /* |
1581 | * If just we got sense for the device (called | |
1582 | * scsi_eh_get_sense), scmd->result is already | |
1583 | * set, do not set DRIVER_TIMEOUT. | |
1584 | */ | |
1da177e4 LT |
1585 | if (!scmd->result) |
1586 | scmd->result |= (DRIVER_TIMEOUT << 24); | |
1587 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" | |
1588 | " cmd: %p\n", | |
1589 | current->comm, scmd)); | |
1590 | scsi_finish_command(scmd); | |
1591 | } | |
1592 | } | |
1593 | } | |
041c5fc3 | 1594 | EXPORT_SYMBOL(scsi_eh_flush_done_q); |
1da177e4 LT |
1595 | |
1596 | /** | |
1597 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. | |
1598 | * @shost: Host to unjam. | |
1599 | * | |
1600 | * Notes: | |
1601 | * When we come in here, we *know* that all commands on the bus have | |
1602 | * either completed, failed or timed out. we also know that no further | |
1603 | * commands are being sent to the host, so things are relatively quiet | |
1604 | * and we have freedom to fiddle with things as we wish. | |
1605 | * | |
1606 | * This is only the *default* implementation. it is possible for | |
1607 | * individual drivers to supply their own version of this function, and | |
1608 | * if the maintainer wishes to do this, it is strongly suggested that | |
1609 | * this function be taken as a template and modified. this function | |
1610 | * was designed to correctly handle problems for about 95% of the | |
1611 | * different cases out there, and it should always provide at least a | |
1612 | * reasonable amount of error recovery. | |
1613 | * | |
1614 | * Any command marked 'failed' or 'timeout' must eventually have | |
1615 | * scsi_finish_cmd() called for it. we do all of the retry stuff | |
1616 | * here, so when we restart the host after we return it should have an | |
1617 | * empty queue. | |
dc8875e1 | 1618 | */ |
1da177e4 LT |
1619 | static void scsi_unjam_host(struct Scsi_Host *shost) |
1620 | { | |
1621 | unsigned long flags; | |
1622 | LIST_HEAD(eh_work_q); | |
1623 | LIST_HEAD(eh_done_q); | |
1624 | ||
1625 | spin_lock_irqsave(shost->host_lock, flags); | |
1626 | list_splice_init(&shost->eh_cmd_q, &eh_work_q); | |
1627 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1628 | ||
1629 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); | |
1630 | ||
1631 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) | |
1632 | if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) | |
1633 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); | |
1634 | ||
1635 | scsi_eh_flush_done_q(&eh_done_q); | |
1636 | } | |
1637 | ||
1638 | /** | |
ad42eb1b | 1639 | * scsi_error_handler - SCSI error handler thread |
1da177e4 LT |
1640 | * @data: Host for which we are running. |
1641 | * | |
1642 | * Notes: | |
ad42eb1b CH |
1643 | * This is the main error handling loop. This is run as a kernel thread |
1644 | * for every SCSI host and handles all error handling activity. | |
dc8875e1 | 1645 | */ |
1da177e4 LT |
1646 | int scsi_error_handler(void *data) |
1647 | { | |
ad42eb1b | 1648 | struct Scsi_Host *shost = data; |
1da177e4 | 1649 | |
1da177e4 | 1650 | /* |
ad42eb1b CH |
1651 | * We use TASK_INTERRUPTIBLE so that the thread is not |
1652 | * counted against the load average as a running process. | |
1653 | * We never actually get interrupted because kthread_run | |
1654 | * disables singal delivery for the created thread. | |
1da177e4 | 1655 | */ |
3ed7a470 JB |
1656 | set_current_state(TASK_INTERRUPTIBLE); |
1657 | while (!kthread_should_stop()) { | |
ee7863bc | 1658 | if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || |
3ed7a470 | 1659 | shost->host_failed != shost->host_busy) { |
ad42eb1b CH |
1660 | SCSI_LOG_ERROR_RECOVERY(1, |
1661 | printk("Error handler scsi_eh_%d sleeping\n", | |
1662 | shost->host_no)); | |
3ed7a470 JB |
1663 | schedule(); |
1664 | set_current_state(TASK_INTERRUPTIBLE); | |
1665 | continue; | |
1666 | } | |
1da177e4 | 1667 | |
3ed7a470 | 1668 | __set_current_state(TASK_RUNNING); |
ad42eb1b CH |
1669 | SCSI_LOG_ERROR_RECOVERY(1, |
1670 | printk("Error handler scsi_eh_%d waking up\n", | |
1671 | shost->host_no)); | |
1da177e4 | 1672 | |
1da177e4 LT |
1673 | /* |
1674 | * We have a host that is failing for some reason. Figure out | |
1675 | * what we need to do to get it up and online again (if we can). | |
1676 | * If we fail, we end up taking the thing offline. | |
1677 | */ | |
9227c33d CH |
1678 | if (shost->transportt->eh_strategy_handler) |
1679 | shost->transportt->eh_strategy_handler(shost); | |
1da177e4 LT |
1680 | else |
1681 | scsi_unjam_host(shost); | |
1682 | ||
1da177e4 LT |
1683 | /* |
1684 | * Note - if the above fails completely, the action is to take | |
1685 | * individual devices offline and flush the queue of any | |
1686 | * outstanding requests that may have been pending. When we | |
1687 | * restart, we restart any I/O to any other devices on the bus | |
1688 | * which are still online. | |
1689 | */ | |
1690 | scsi_restart_operations(shost); | |
3ed7a470 | 1691 | set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 | 1692 | } |
461a0ffb SR |
1693 | __set_current_state(TASK_RUNNING); |
1694 | ||
ad42eb1b CH |
1695 | SCSI_LOG_ERROR_RECOVERY(1, |
1696 | printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); | |
3ed7a470 | 1697 | shost->ehandler = NULL; |
1da177e4 LT |
1698 | return 0; |
1699 | } | |
1700 | ||
1701 | /* | |
1702 | * Function: scsi_report_bus_reset() | |
1703 | * | |
1704 | * Purpose: Utility function used by low-level drivers to report that | |
1705 | * they have observed a bus reset on the bus being handled. | |
1706 | * | |
1707 | * Arguments: shost - Host in question | |
1708 | * channel - channel on which reset was observed. | |
1709 | * | |
1710 | * Returns: Nothing | |
1711 | * | |
1712 | * Lock status: Host lock must be held. | |
1713 | * | |
1714 | * Notes: This only needs to be called if the reset is one which | |
1715 | * originates from an unknown location. Resets originated | |
1716 | * by the mid-level itself don't need to call this, but there | |
1717 | * should be no harm. | |
1718 | * | |
1719 | * The main purpose of this is to make sure that a CHECK_CONDITION | |
1720 | * is properly treated. | |
1721 | */ | |
1722 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) | |
1723 | { | |
1724 | struct scsi_device *sdev; | |
1725 | ||
1726 | __shost_for_each_device(sdev, shost) { | |
30bd7df8 MC |
1727 | if (channel == sdev_channel(sdev)) |
1728 | __scsi_report_device_reset(sdev, NULL); | |
1da177e4 LT |
1729 | } |
1730 | } | |
1731 | EXPORT_SYMBOL(scsi_report_bus_reset); | |
1732 | ||
1733 | /* | |
1734 | * Function: scsi_report_device_reset() | |
1735 | * | |
1736 | * Purpose: Utility function used by low-level drivers to report that | |
1737 | * they have observed a device reset on the device being handled. | |
1738 | * | |
1739 | * Arguments: shost - Host in question | |
1740 | * channel - channel on which reset was observed | |
1741 | * target - target on which reset was observed | |
1742 | * | |
1743 | * Returns: Nothing | |
1744 | * | |
1745 | * Lock status: Host lock must be held | |
1746 | * | |
1747 | * Notes: This only needs to be called if the reset is one which | |
1748 | * originates from an unknown location. Resets originated | |
1749 | * by the mid-level itself don't need to call this, but there | |
1750 | * should be no harm. | |
1751 | * | |
1752 | * The main purpose of this is to make sure that a CHECK_CONDITION | |
1753 | * is properly treated. | |
1754 | */ | |
1755 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) | |
1756 | { | |
1757 | struct scsi_device *sdev; | |
1758 | ||
1759 | __shost_for_each_device(sdev, shost) { | |
422c0d61 | 1760 | if (channel == sdev_channel(sdev) && |
30bd7df8 MC |
1761 | target == sdev_id(sdev)) |
1762 | __scsi_report_device_reset(sdev, NULL); | |
1da177e4 LT |
1763 | } |
1764 | } | |
1765 | EXPORT_SYMBOL(scsi_report_device_reset); | |
1766 | ||
1767 | static void | |
1768 | scsi_reset_provider_done_command(struct scsi_cmnd *scmd) | |
1769 | { | |
1770 | } | |
1771 | ||
1772 | /* | |
1773 | * Function: scsi_reset_provider | |
1774 | * | |
1775 | * Purpose: Send requested reset to a bus or device at any phase. | |
1776 | * | |
1777 | * Arguments: device - device to send reset to | |
1778 | * flag - reset type (see scsi.h) | |
1779 | * | |
1780 | * Returns: SUCCESS/FAILURE. | |
1781 | * | |
1782 | * Notes: This is used by the SCSI Generic driver to provide | |
1783 | * Bus/Device reset capability. | |
1784 | */ | |
1785 | int | |
1786 | scsi_reset_provider(struct scsi_device *dev, int flag) | |
1787 | { | |
1788 | struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); | |
d7a1bb0a | 1789 | struct Scsi_Host *shost = dev->host; |
1da177e4 | 1790 | struct request req; |
d7a1bb0a | 1791 | unsigned long flags; |
1da177e4 LT |
1792 | int rtn; |
1793 | ||
4f54eec8 | 1794 | blk_rq_init(NULL, &req); |
1da177e4 LT |
1795 | scmd->request = &req; |
1796 | memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); | |
b4edcbca | 1797 | |
64a87b24 BH |
1798 | scmd->cmnd = req.cmd; |
1799 | ||
1da177e4 | 1800 | scmd->scsi_done = scsi_reset_provider_done_command; |
30b0c37b | 1801 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
1da177e4 LT |
1802 | |
1803 | scmd->cmd_len = 0; | |
1804 | ||
1805 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; | |
1da177e4 LT |
1806 | |
1807 | init_timer(&scmd->eh_timeout); | |
1808 | ||
d7a1bb0a JS |
1809 | spin_lock_irqsave(shost->host_lock, flags); |
1810 | shost->tmf_in_progress = 1; | |
1811 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1812 | ||
1da177e4 LT |
1813 | switch (flag) { |
1814 | case SCSI_TRY_RESET_DEVICE: | |
1815 | rtn = scsi_try_bus_device_reset(scmd); | |
1816 | if (rtn == SUCCESS) | |
1817 | break; | |
1818 | /* FALLTHROUGH */ | |
30bd7df8 MC |
1819 | case SCSI_TRY_RESET_TARGET: |
1820 | rtn = scsi_try_target_reset(scmd); | |
1821 | if (rtn == SUCCESS) | |
1822 | break; | |
1823 | /* FALLTHROUGH */ | |
1da177e4 LT |
1824 | case SCSI_TRY_RESET_BUS: |
1825 | rtn = scsi_try_bus_reset(scmd); | |
1826 | if (rtn == SUCCESS) | |
1827 | break; | |
1828 | /* FALLTHROUGH */ | |
1829 | case SCSI_TRY_RESET_HOST: | |
1830 | rtn = scsi_try_host_reset(scmd); | |
1831 | break; | |
1832 | default: | |
1833 | rtn = FAILED; | |
1834 | } | |
1835 | ||
d7a1bb0a JS |
1836 | spin_lock_irqsave(shost->host_lock, flags); |
1837 | shost->tmf_in_progress = 0; | |
1838 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1839 | ||
1840 | /* | |
1841 | * be sure to wake up anyone who was sleeping or had their queue | |
1842 | * suspended while we performed the TMF. | |
1843 | */ | |
1844 | SCSI_LOG_ERROR_RECOVERY(3, | |
1845 | printk("%s: waking up host to restart after TMF\n", | |
cadbd4a5 | 1846 | __func__)); |
d7a1bb0a JS |
1847 | |
1848 | wake_up(&shost->host_wait); | |
1849 | ||
1850 | scsi_run_host_queues(shost); | |
1851 | ||
1da177e4 LT |
1852 | scsi_next_command(scmd); |
1853 | return rtn; | |
1854 | } | |
1855 | EXPORT_SYMBOL(scsi_reset_provider); | |
1856 | ||
1857 | /** | |
1858 | * scsi_normalize_sense - normalize main elements from either fixed or | |
1859 | * descriptor sense data format into a common format. | |
1860 | * | |
1861 | * @sense_buffer: byte array containing sense data returned by device | |
1862 | * @sb_len: number of valid bytes in sense_buffer | |
1863 | * @sshdr: pointer to instance of structure that common | |
1864 | * elements are written to. | |
1865 | * | |
1866 | * Notes: | |
1867 | * The "main elements" from sense data are: response_code, sense_key, | |
1868 | * asc, ascq and additional_length (only for descriptor format). | |
1869 | * | |
1870 | * Typically this function can be called after a device has | |
1871 | * responded to a SCSI command with the CHECK_CONDITION status. | |
1872 | * | |
1873 | * Return value: | |
1874 | * 1 if valid sense data information found, else 0; | |
dc8875e1 | 1875 | */ |
1da177e4 LT |
1876 | int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, |
1877 | struct scsi_sense_hdr *sshdr) | |
1878 | { | |
33aa687d | 1879 | if (!sense_buffer || !sb_len) |
1da177e4 LT |
1880 | return 0; |
1881 | ||
1882 | memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); | |
1883 | ||
1884 | sshdr->response_code = (sense_buffer[0] & 0x7f); | |
33aa687d JB |
1885 | |
1886 | if (!scsi_sense_valid(sshdr)) | |
1887 | return 0; | |
1888 | ||
1da177e4 LT |
1889 | if (sshdr->response_code >= 0x72) { |
1890 | /* | |
1891 | * descriptor format | |
1892 | */ | |
1893 | if (sb_len > 1) | |
1894 | sshdr->sense_key = (sense_buffer[1] & 0xf); | |
1895 | if (sb_len > 2) | |
1896 | sshdr->asc = sense_buffer[2]; | |
1897 | if (sb_len > 3) | |
1898 | sshdr->ascq = sense_buffer[3]; | |
1899 | if (sb_len > 7) | |
1900 | sshdr->additional_length = sense_buffer[7]; | |
1901 | } else { | |
1902 | /* | |
1903 | * fixed format | |
1904 | */ | |
1905 | if (sb_len > 2) | |
1906 | sshdr->sense_key = (sense_buffer[2] & 0xf); | |
1907 | if (sb_len > 7) { | |
1908 | sb_len = (sb_len < (sense_buffer[7] + 8)) ? | |
1909 | sb_len : (sense_buffer[7] + 8); | |
1910 | if (sb_len > 12) | |
1911 | sshdr->asc = sense_buffer[12]; | |
1912 | if (sb_len > 13) | |
1913 | sshdr->ascq = sense_buffer[13]; | |
1914 | } | |
1915 | } | |
1916 | ||
1917 | return 1; | |
1918 | } | |
1919 | EXPORT_SYMBOL(scsi_normalize_sense); | |
1920 | ||
1da177e4 LT |
1921 | int scsi_command_normalize_sense(struct scsi_cmnd *cmd, |
1922 | struct scsi_sense_hdr *sshdr) | |
1923 | { | |
1924 | return scsi_normalize_sense(cmd->sense_buffer, | |
b80ca4f7 | 1925 | SCSI_SENSE_BUFFERSIZE, sshdr); |
1da177e4 LT |
1926 | } |
1927 | EXPORT_SYMBOL(scsi_command_normalize_sense); | |
1928 | ||
1929 | /** | |
eb44820c | 1930 | * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format. |
1da177e4 LT |
1931 | * @sense_buffer: byte array of descriptor format sense data |
1932 | * @sb_len: number of valid bytes in sense_buffer | |
1933 | * @desc_type: value of descriptor type to find | |
1934 | * (e.g. 0 -> information) | |
1935 | * | |
1936 | * Notes: | |
1937 | * only valid when sense data is in descriptor format | |
1938 | * | |
1939 | * Return value: | |
1940 | * pointer to start of (first) descriptor if found else NULL | |
dc8875e1 | 1941 | */ |
1da177e4 LT |
1942 | const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, |
1943 | int desc_type) | |
1944 | { | |
1945 | int add_sen_len, add_len, desc_len, k; | |
1946 | const u8 * descp; | |
1947 | ||
1948 | if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) | |
1949 | return NULL; | |
1950 | if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) | |
1951 | return NULL; | |
1952 | add_sen_len = (add_sen_len < (sb_len - 8)) ? | |
1953 | add_sen_len : (sb_len - 8); | |
1954 | descp = &sense_buffer[8]; | |
1955 | for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { | |
1956 | descp += desc_len; | |
1957 | add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; | |
1958 | desc_len = add_len + 2; | |
1959 | if (descp[0] == desc_type) | |
1960 | return descp; | |
1961 | if (add_len < 0) // short descriptor ?? | |
1962 | break; | |
1963 | } | |
1964 | return NULL; | |
1965 | } | |
1966 | EXPORT_SYMBOL(scsi_sense_desc_find); | |
1967 | ||
1968 | /** | |
eb44820c | 1969 | * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) |
1da177e4 LT |
1970 | * @sense_buffer: byte array of sense data |
1971 | * @sb_len: number of valid bytes in sense_buffer | |
1972 | * @info_out: pointer to 64 integer where 8 or 4 byte information | |
1973 | * field will be placed if found. | |
1974 | * | |
1975 | * Return value: | |
1976 | * 1 if information field found, 0 if not found. | |
dc8875e1 | 1977 | */ |
1da177e4 LT |
1978 | int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, |
1979 | u64 * info_out) | |
1980 | { | |
1981 | int j; | |
1982 | const u8 * ucp; | |
1983 | u64 ull; | |
1984 | ||
1985 | if (sb_len < 7) | |
1986 | return 0; | |
1987 | switch (sense_buffer[0] & 0x7f) { | |
1988 | case 0x70: | |
1989 | case 0x71: | |
1990 | if (sense_buffer[0] & 0x80) { | |
1991 | *info_out = (sense_buffer[3] << 24) + | |
1992 | (sense_buffer[4] << 16) + | |
1993 | (sense_buffer[5] << 8) + sense_buffer[6]; | |
1994 | return 1; | |
1995 | } else | |
1996 | return 0; | |
1997 | case 0x72: | |
1998 | case 0x73: | |
1999 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, | |
2000 | 0 /* info desc */); | |
2001 | if (ucp && (0xa == ucp[1])) { | |
2002 | ull = 0; | |
2003 | for (j = 0; j < 8; ++j) { | |
2004 | if (j > 0) | |
2005 | ull <<= 8; | |
2006 | ull |= ucp[4 + j]; | |
2007 | } | |
2008 | *info_out = ull; | |
2009 | return 1; | |
2010 | } else | |
2011 | return 0; | |
2012 | default: | |
2013 | return 0; | |
2014 | } | |
2015 | } | |
2016 | EXPORT_SYMBOL(scsi_get_sense_info_fld); | |
3bc6a261 FT |
2017 | |
2018 | /** | |
2019 | * scsi_build_sense_buffer - build sense data in a buffer | |
2020 | * @desc: Sense format (non zero == descriptor format, | |
2021 | * 0 == fixed format) | |
2022 | * @buf: Where to build sense data | |
2023 | * @key: Sense key | |
2024 | * @asc: Additional sense code | |
2025 | * @ascq: Additional sense code qualifier | |
2026 | * | |
2027 | **/ | |
2028 | void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq) | |
2029 | { | |
2030 | if (desc) { | |
2031 | buf[0] = 0x72; /* descriptor, current */ | |
2032 | buf[1] = key; | |
2033 | buf[2] = asc; | |
2034 | buf[3] = ascq; | |
2035 | buf[7] = 0; | |
2036 | } else { | |
2037 | buf[0] = 0x70; /* fixed, current */ | |
2038 | buf[2] = key; | |
2039 | buf[7] = 0xa; | |
2040 | buf[12] = asc; | |
2041 | buf[13] = ascq; | |
2042 | } | |
2043 | } | |
2044 | EXPORT_SYMBOL(scsi_build_sense_buffer); |