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> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/kernel.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> | |
29 | #include <scsi/scsi_dbg.h> | |
30 | #include <scsi/scsi_device.h> | |
31 | #include <scsi/scsi_eh.h> | |
32 | #include <scsi/scsi_host.h> | |
33 | #include <scsi/scsi_ioctl.h> | |
34 | #include <scsi/scsi_request.h> | |
35 | ||
36 | #include "scsi_priv.h" | |
37 | #include "scsi_logging.h" | |
38 | ||
39 | #define SENSE_TIMEOUT (10*HZ) | |
40 | #define START_UNIT_TIMEOUT (30*HZ) | |
41 | ||
42 | /* | |
43 | * These should *probably* be handled by the host itself. | |
44 | * Since it is allowed to sleep, it probably should. | |
45 | */ | |
46 | #define BUS_RESET_SETTLE_TIME (10) | |
47 | #define HOST_RESET_SETTLE_TIME (10) | |
48 | ||
49 | /* called with shost->host_lock held */ | |
50 | void scsi_eh_wakeup(struct Scsi_Host *shost) | |
51 | { | |
52 | if (shost->host_busy == shost->host_failed) { | |
3ed7a470 | 53 | wake_up_process(shost->ehandler); |
1da177e4 LT |
54 | SCSI_LOG_ERROR_RECOVERY(5, |
55 | printk("Waking error handler thread\n")); | |
56 | } | |
57 | } | |
58 | ||
59 | /** | |
60 | * scsi_eh_scmd_add - add scsi cmd to error handling. | |
61 | * @scmd: scmd to run eh on. | |
62 | * @eh_flag: optional SCSI_EH flag. | |
63 | * | |
64 | * Return value: | |
65 | * 0 on failure. | |
66 | **/ | |
67 | int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) | |
68 | { | |
69 | struct Scsi_Host *shost = scmd->device->host; | |
70 | unsigned long flags; | |
939647ee | 71 | int ret = 0; |
1da177e4 | 72 | |
3ed7a470 | 73 | if (!shost->ehandler) |
1da177e4 LT |
74 | return 0; |
75 | ||
76 | spin_lock_irqsave(shost->host_lock, flags); | |
939647ee JB |
77 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
78 | if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) | |
79 | goto out_unlock; | |
1da177e4 | 80 | |
939647ee | 81 | ret = 1; |
3111b0d1 | 82 | scmd->eh_eflags |= eh_flag; |
1da177e4 | 83 | list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); |
1da177e4 LT |
84 | shost->host_failed++; |
85 | scsi_eh_wakeup(shost); | |
939647ee | 86 | out_unlock: |
1da177e4 | 87 | spin_unlock_irqrestore(shost->host_lock, flags); |
939647ee | 88 | return ret; |
1da177e4 LT |
89 | } |
90 | ||
91 | /** | |
92 | * scsi_add_timer - Start timeout timer for a single scsi command. | |
93 | * @scmd: scsi command that is about to start running. | |
94 | * @timeout: amount of time to allow this command to run. | |
95 | * @complete: timeout function to call if timer isn't canceled. | |
96 | * | |
97 | * Notes: | |
98 | * This should be turned into an inline function. Each scsi command | |
99 | * has its own timer, and as it is added to the queue, we set up the | |
100 | * timer. When the command completes, we cancel the timer. | |
101 | **/ | |
102 | void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, | |
103 | void (*complete)(struct scsi_cmnd *)) | |
104 | { | |
105 | ||
106 | /* | |
107 | * If the clock was already running for this command, then | |
108 | * first delete the timer. The timer handling code gets rather | |
109 | * confused if we don't do this. | |
110 | */ | |
111 | if (scmd->eh_timeout.function) | |
112 | del_timer(&scmd->eh_timeout); | |
113 | ||
114 | scmd->eh_timeout.data = (unsigned long)scmd; | |
115 | scmd->eh_timeout.expires = jiffies + timeout; | |
116 | scmd->eh_timeout.function = (void (*)(unsigned long)) complete; | |
117 | ||
118 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" | |
119 | " %d, (%p)\n", __FUNCTION__, | |
120 | scmd, timeout, complete)); | |
121 | ||
122 | add_timer(&scmd->eh_timeout); | |
123 | } | |
1da177e4 LT |
124 | |
125 | /** | |
126 | * scsi_delete_timer - Delete/cancel timer for a given function. | |
127 | * @scmd: Cmd that we are canceling timer for | |
128 | * | |
129 | * Notes: | |
130 | * This should be turned into an inline function. | |
131 | * | |
132 | * Return value: | |
133 | * 1 if we were able to detach the timer. 0 if we blew it, and the | |
134 | * timer function has already started to run. | |
135 | **/ | |
136 | int scsi_delete_timer(struct scsi_cmnd *scmd) | |
137 | { | |
138 | int rtn; | |
139 | ||
140 | rtn = del_timer(&scmd->eh_timeout); | |
141 | ||
142 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," | |
143 | " rtn: %d\n", __FUNCTION__, | |
144 | scmd, rtn)); | |
145 | ||
146 | scmd->eh_timeout.data = (unsigned long)NULL; | |
147 | scmd->eh_timeout.function = NULL; | |
148 | ||
149 | return rtn; | |
150 | } | |
1da177e4 LT |
151 | |
152 | /** | |
153 | * scsi_times_out - Timeout function for normal scsi commands. | |
154 | * @scmd: Cmd that is timing out. | |
155 | * | |
156 | * Notes: | |
157 | * We do not need to lock this. There is the potential for a race | |
158 | * only in that the normal completion handling might run, but if the | |
159 | * normal completion function determines that the timer has already | |
160 | * fired, then it mustn't do anything. | |
161 | **/ | |
162 | void scsi_times_out(struct scsi_cmnd *scmd) | |
163 | { | |
164 | scsi_log_completion(scmd, TIMEOUT_ERROR); | |
165 | ||
166 | if (scmd->device->host->hostt->eh_timed_out) | |
167 | switch (scmd->device->host->hostt->eh_timed_out(scmd)) { | |
168 | case EH_HANDLED: | |
169 | __scsi_done(scmd); | |
170 | return; | |
171 | case EH_RESET_TIMER: | |
172 | /* This allows a single retry even of a command | |
173 | * with allowed == 0 */ | |
174 | if (scmd->retries++ > scmd->allowed) | |
175 | break; | |
176 | scsi_add_timer(scmd, scmd->timeout_per_command, | |
177 | scsi_times_out); | |
178 | return; | |
179 | case EH_NOT_HANDLED: | |
180 | break; | |
181 | } | |
182 | ||
183 | if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { | |
939647ee JB |
184 | scmd->result |= DID_TIME_OUT << 16; |
185 | __scsi_done(scmd); | |
1da177e4 LT |
186 | } |
187 | } | |
188 | ||
189 | /** | |
190 | * scsi_block_when_processing_errors - Prevent cmds from being queued. | |
191 | * @sdev: Device on which we are performing recovery. | |
192 | * | |
193 | * Description: | |
194 | * We block until the host is out of error recovery, and then check to | |
195 | * see whether the host or the device is offline. | |
196 | * | |
197 | * Return value: | |
198 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. | |
199 | **/ | |
200 | int scsi_block_when_processing_errors(struct scsi_device *sdev) | |
201 | { | |
202 | int online; | |
203 | ||
939647ee | 204 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
1da177e4 LT |
205 | |
206 | online = scsi_device_online(sdev); | |
207 | ||
208 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__, | |
209 | online)); | |
210 | ||
211 | return online; | |
212 | } | |
213 | EXPORT_SYMBOL(scsi_block_when_processing_errors); | |
214 | ||
215 | #ifdef CONFIG_SCSI_LOGGING | |
216 | /** | |
217 | * scsi_eh_prt_fail_stats - Log info on failures. | |
218 | * @shost: scsi host being recovered. | |
219 | * @work_q: Queue of scsi cmds to process. | |
220 | **/ | |
221 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, | |
222 | struct list_head *work_q) | |
223 | { | |
224 | struct scsi_cmnd *scmd; | |
225 | struct scsi_device *sdev; | |
226 | int total_failures = 0; | |
227 | int cmd_failed = 0; | |
228 | int cmd_cancel = 0; | |
229 | int devices_failed = 0; | |
230 | ||
231 | shost_for_each_device(sdev, shost) { | |
232 | list_for_each_entry(scmd, work_q, eh_entry) { | |
233 | if (scmd->device == sdev) { | |
234 | ++total_failures; | |
3111b0d1 | 235 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) |
1da177e4 LT |
236 | ++cmd_cancel; |
237 | else | |
238 | ++cmd_failed; | |
239 | } | |
240 | } | |
241 | ||
242 | if (cmd_cancel || cmd_failed) { | |
243 | SCSI_LOG_ERROR_RECOVERY(3, | |
9ccfc756 JB |
244 | sdev_printk(KERN_INFO, sdev, |
245 | "%s: cmds failed: %d, cancel: %d\n", | |
246 | __FUNCTION__, cmd_failed, | |
247 | cmd_cancel)); | |
1da177e4 LT |
248 | cmd_cancel = 0; |
249 | cmd_failed = 0; | |
250 | ++devices_failed; | |
251 | } | |
252 | } | |
253 | ||
254 | SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" | |
255 | " devices require eh work\n", | |
256 | total_failures, devices_failed)); | |
257 | } | |
258 | #endif | |
259 | ||
260 | /** | |
261 | * scsi_check_sense - Examine scsi cmd sense | |
262 | * @scmd: Cmd to have sense checked. | |
263 | * | |
264 | * Return value: | |
265 | * SUCCESS or FAILED or NEEDS_RETRY | |
266 | * | |
267 | * Notes: | |
268 | * When a deferred error is detected the current command has | |
269 | * not been executed and needs retrying. | |
270 | **/ | |
271 | static int scsi_check_sense(struct scsi_cmnd *scmd) | |
272 | { | |
273 | struct scsi_sense_hdr sshdr; | |
274 | ||
275 | if (! scsi_command_normalize_sense(scmd, &sshdr)) | |
276 | return FAILED; /* no valid sense data */ | |
277 | ||
278 | if (scsi_sense_is_deferred(&sshdr)) | |
279 | return NEEDS_RETRY; | |
280 | ||
281 | /* | |
282 | * Previous logic looked for FILEMARK, EOM or ILI which are | |
283 | * mainly associated with tapes and returned SUCCESS. | |
284 | */ | |
285 | if (sshdr.response_code == 0x70) { | |
286 | /* fixed format */ | |
287 | if (scmd->sense_buffer[2] & 0xe0) | |
288 | return SUCCESS; | |
289 | } else { | |
290 | /* | |
291 | * descriptor format: look for "stream commands sense data | |
292 | * descriptor" (see SSC-3). Assume single sense data | |
293 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. | |
294 | */ | |
295 | if ((sshdr.additional_length > 3) && | |
296 | (scmd->sense_buffer[8] == 0x4) && | |
297 | (scmd->sense_buffer[11] & 0xe0)) | |
298 | return SUCCESS; | |
299 | } | |
300 | ||
301 | switch (sshdr.sense_key) { | |
302 | case NO_SENSE: | |
303 | return SUCCESS; | |
304 | case RECOVERED_ERROR: | |
305 | return /* soft_error */ SUCCESS; | |
306 | ||
307 | case ABORTED_COMMAND: | |
308 | return NEEDS_RETRY; | |
309 | case NOT_READY: | |
310 | case UNIT_ATTENTION: | |
311 | /* | |
312 | * if we are expecting a cc/ua because of a bus reset that we | |
313 | * performed, treat this just as a retry. otherwise this is | |
314 | * information that we should pass up to the upper-level driver | |
315 | * so that we can deal with it there. | |
316 | */ | |
317 | if (scmd->device->expecting_cc_ua) { | |
318 | scmd->device->expecting_cc_ua = 0; | |
319 | return NEEDS_RETRY; | |
320 | } | |
321 | /* | |
322 | * if the device is in the process of becoming ready, we | |
323 | * should retry. | |
324 | */ | |
325 | if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) | |
326 | return NEEDS_RETRY; | |
327 | /* | |
328 | * if the device is not started, we need to wake | |
329 | * the error handler to start the motor | |
330 | */ | |
331 | if (scmd->device->allow_restart && | |
332 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) | |
333 | return FAILED; | |
334 | return SUCCESS; | |
335 | ||
336 | /* these three are not supported */ | |
337 | case COPY_ABORTED: | |
338 | case VOLUME_OVERFLOW: | |
339 | case MISCOMPARE: | |
340 | return SUCCESS; | |
341 | ||
342 | case MEDIUM_ERROR: | |
343 | return NEEDS_RETRY; | |
344 | ||
345 | case HARDWARE_ERROR: | |
346 | if (scmd->device->retry_hwerror) | |
347 | return NEEDS_RETRY; | |
348 | else | |
349 | return SUCCESS; | |
350 | ||
351 | case ILLEGAL_REQUEST: | |
352 | case BLANK_CHECK: | |
353 | case DATA_PROTECT: | |
354 | default: | |
355 | return SUCCESS; | |
356 | } | |
357 | } | |
358 | ||
359 | /** | |
360 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. | |
361 | * @scmd: SCSI cmd to examine. | |
362 | * | |
363 | * Notes: | |
364 | * This is *only* called when we are examining the status of commands | |
365 | * queued during error recovery. the main difference here is that we | |
366 | * don't allow for the possibility of retries here, and we are a lot | |
367 | * more restrictive about what we consider acceptable. | |
368 | **/ | |
369 | static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) | |
370 | { | |
371 | /* | |
372 | * first check the host byte, to see if there is anything in there | |
373 | * that would indicate what we need to do. | |
374 | */ | |
375 | if (host_byte(scmd->result) == DID_RESET) { | |
376 | /* | |
377 | * rats. we are already in the error handler, so we now | |
378 | * get to try and figure out what to do next. if the sense | |
379 | * is valid, we have a pretty good idea of what to do. | |
380 | * if not, we mark it as FAILED. | |
381 | */ | |
382 | return scsi_check_sense(scmd); | |
383 | } | |
384 | if (host_byte(scmd->result) != DID_OK) | |
385 | return FAILED; | |
386 | ||
387 | /* | |
388 | * next, check the message byte. | |
389 | */ | |
390 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) | |
391 | return FAILED; | |
392 | ||
393 | /* | |
394 | * now, check the status byte to see if this indicates | |
395 | * anything special. | |
396 | */ | |
397 | switch (status_byte(scmd->result)) { | |
398 | case GOOD: | |
399 | case COMMAND_TERMINATED: | |
400 | return SUCCESS; | |
401 | case CHECK_CONDITION: | |
402 | return scsi_check_sense(scmd); | |
403 | case CONDITION_GOOD: | |
404 | case INTERMEDIATE_GOOD: | |
405 | case INTERMEDIATE_C_GOOD: | |
406 | /* | |
407 | * who knows? FIXME(eric) | |
408 | */ | |
409 | return SUCCESS; | |
410 | case BUSY: | |
411 | case QUEUE_FULL: | |
412 | case RESERVATION_CONFLICT: | |
413 | default: | |
414 | return FAILED; | |
415 | } | |
416 | return FAILED; | |
417 | } | |
418 | ||
1da177e4 LT |
419 | /** |
420 | * scsi_eh_done - Completion function for error handling. | |
421 | * @scmd: Cmd that is done. | |
422 | **/ | |
423 | static void scsi_eh_done(struct scsi_cmnd *scmd) | |
424 | { | |
7dfdc9a5 CH |
425 | SCSI_LOG_ERROR_RECOVERY(3, |
426 | printk("%s scmd: %p result: %x\n", | |
427 | __FUNCTION__, scmd, scmd->result)); | |
428 | complete(scmd->device->host->eh_action); | |
1da177e4 LT |
429 | } |
430 | ||
431 | /** | |
432 | * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery. | |
433 | * @scmd: SCSI Cmd to send. | |
434 | * @timeout: Timeout for cmd. | |
435 | * | |
1da177e4 LT |
436 | * Return value: |
437 | * SUCCESS or FAILED or NEEDS_RETRY | |
438 | **/ | |
439 | static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout) | |
440 | { | |
f59114b7 | 441 | struct scsi_device *sdev = scmd->device; |
442 | struct Scsi_Host *shost = sdev->host; | |
7dfdc9a5 CH |
443 | DECLARE_COMPLETION(done); |
444 | unsigned long timeleft; | |
1da177e4 | 445 | unsigned long flags; |
7dfdc9a5 | 446 | int rtn; |
1da177e4 | 447 | |
f59114b7 | 448 | if (sdev->scsi_level <= SCSI_2) |
1da177e4 | 449 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
f59114b7 | 450 | (sdev->lun << 5 & 0xe0); |
1da177e4 | 451 | |
7dfdc9a5 | 452 | shost->eh_action = &done; |
1da177e4 LT |
453 | scmd->request->rq_status = RQ_SCSI_BUSY; |
454 | ||
f59114b7 | 455 | spin_lock_irqsave(shost->host_lock, flags); |
1da177e4 | 456 | scsi_log_send(scmd); |
f59114b7 | 457 | shost->hostt->queuecommand(scmd, scsi_eh_done); |
458 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1da177e4 | 459 | |
7dfdc9a5 | 460 | timeleft = wait_for_completion_timeout(&done, timeout); |
1da177e4 | 461 | |
7dfdc9a5 | 462 | scmd->request->rq_status = RQ_SCSI_DONE; |
f59114b7 | 463 | shost->eh_action = NULL; |
1da177e4 | 464 | |
7dfdc9a5 | 465 | scsi_log_completion(scmd, SUCCESS); |
1da177e4 | 466 | |
7dfdc9a5 CH |
467 | SCSI_LOG_ERROR_RECOVERY(3, |
468 | printk("%s: scmd: %p, timeleft: %ld\n", | |
469 | __FUNCTION__, scmd, timeleft)); | |
1da177e4 LT |
470 | |
471 | /* | |
7dfdc9a5 CH |
472 | * If there is time left scsi_eh_done got called, and we will |
473 | * examine the actual status codes to see whether the command | |
474 | * actually did complete normally, else tell the host to forget | |
475 | * about this command. | |
1da177e4 | 476 | */ |
7dfdc9a5 | 477 | if (timeleft) { |
1da177e4 LT |
478 | rtn = scsi_eh_completed_normally(scmd); |
479 | SCSI_LOG_ERROR_RECOVERY(3, | |
480 | printk("%s: scsi_eh_completed_normally %x\n", | |
481 | __FUNCTION__, rtn)); | |
7dfdc9a5 | 482 | |
1da177e4 LT |
483 | switch (rtn) { |
484 | case SUCCESS: | |
485 | case NEEDS_RETRY: | |
486 | case FAILED: | |
487 | break; | |
488 | default: | |
489 | rtn = FAILED; | |
490 | break; | |
491 | } | |
7dfdc9a5 CH |
492 | } else { |
493 | /* | |
494 | * FIXME(eric) - we are not tracking whether we could | |
495 | * abort a timed out command or not. not sure how | |
496 | * we should treat them differently anyways. | |
497 | */ | |
498 | if (shost->hostt->eh_abort_handler) | |
499 | shost->hostt->eh_abort_handler(scmd); | |
500 | rtn = FAILED; | |
1da177e4 LT |
501 | } |
502 | ||
503 | return rtn; | |
504 | } | |
505 | ||
506 | /** | |
507 | * scsi_request_sense - Request sense data from a particular target. | |
508 | * @scmd: SCSI cmd for request sense. | |
509 | * | |
510 | * Notes: | |
511 | * Some hosts automatically obtain this information, others require | |
512 | * that we obtain it on our own. This function will *not* return until | |
513 | * the command either times out, or it completes. | |
514 | **/ | |
515 | static int scsi_request_sense(struct scsi_cmnd *scmd) | |
516 | { | |
517 | static unsigned char generic_sense[6] = | |
518 | {REQUEST_SENSE, 0, 0, 0, 252, 0}; | |
519 | unsigned char *scsi_result; | |
520 | int saved_result; | |
521 | int rtn; | |
522 | ||
523 | memcpy(scmd->cmnd, generic_sense, sizeof(generic_sense)); | |
524 | ||
bc86120a | 525 | scsi_result = kmalloc(252, GFP_ATOMIC | ((scmd->device->host->hostt->unchecked_isa_dma) ? __GFP_DMA : 0)); |
1da177e4 LT |
526 | |
527 | ||
528 | if (unlikely(!scsi_result)) { | |
529 | printk(KERN_ERR "%s: cannot allocate scsi_result.\n", | |
530 | __FUNCTION__); | |
531 | return FAILED; | |
532 | } | |
533 | ||
534 | /* | |
535 | * zero the sense buffer. some host adapters automatically always | |
536 | * request sense, so it is not a good idea that | |
537 | * scmd->request_buffer and scmd->sense_buffer point to the same | |
538 | * address (db). 0 is not a valid sense code. | |
539 | */ | |
540 | memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); | |
541 | memset(scsi_result, 0, 252); | |
542 | ||
543 | saved_result = scmd->result; | |
544 | scmd->request_buffer = scsi_result; | |
545 | scmd->request_bufflen = 252; | |
546 | scmd->use_sg = 0; | |
547 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); | |
548 | scmd->sc_data_direction = DMA_FROM_DEVICE; | |
549 | scmd->underflow = 0; | |
550 | ||
551 | rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); | |
552 | ||
553 | /* last chance to have valid sense data */ | |
554 | if(!SCSI_SENSE_VALID(scmd)) { | |
555 | memcpy(scmd->sense_buffer, scmd->request_buffer, | |
556 | sizeof(scmd->sense_buffer)); | |
557 | } | |
558 | ||
559 | kfree(scsi_result); | |
560 | ||
561 | /* | |
562 | * when we eventually call scsi_finish, we really wish to complete | |
563 | * the original request, so let's restore the original data. (db) | |
564 | */ | |
565 | scsi_setup_cmd_retry(scmd); | |
566 | scmd->result = saved_result; | |
567 | return rtn; | |
568 | } | |
569 | ||
570 | /** | |
571 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. | |
572 | * @scmd: Original SCSI cmd that eh has finished. | |
573 | * @done_q: Queue for processed commands. | |
574 | * | |
575 | * Notes: | |
576 | * We don't want to use the normal command completion while we are are | |
577 | * still handling errors - it may cause other commands to be queued, | |
578 | * and that would disturb what we are doing. thus we really want to | |
579 | * keep a list of pending commands for final completion, and once we | |
580 | * are ready to leave error handling we handle completion for real. | |
581 | **/ | |
582 | static void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, | |
583 | struct list_head *done_q) | |
584 | { | |
585 | scmd->device->host->host_failed--; | |
3111b0d1 | 586 | scmd->eh_eflags = 0; |
1da177e4 LT |
587 | |
588 | /* | |
589 | * set this back so that the upper level can correctly free up | |
590 | * things. | |
591 | */ | |
592 | scsi_setup_cmd_retry(scmd); | |
593 | list_move_tail(&scmd->eh_entry, done_q); | |
594 | } | |
595 | ||
596 | /** | |
597 | * scsi_eh_get_sense - Get device sense data. | |
598 | * @work_q: Queue of commands to process. | |
599 | * @done_q: Queue of proccessed commands.. | |
600 | * | |
601 | * Description: | |
602 | * See if we need to request sense information. if so, then get it | |
603 | * now, so we have a better idea of what to do. | |
604 | * | |
605 | * Notes: | |
606 | * This has the unfortunate side effect that if a shost adapter does | |
607 | * not automatically request sense information, that we end up shutting | |
608 | * it down before we request it. | |
609 | * | |
610 | * All drivers should request sense information internally these days, | |
611 | * so for now all I have to say is tough noogies if you end up in here. | |
612 | * | |
613 | * XXX: Long term this code should go away, but that needs an audit of | |
614 | * all LLDDs first. | |
615 | **/ | |
616 | static int scsi_eh_get_sense(struct list_head *work_q, | |
617 | struct list_head *done_q) | |
618 | { | |
937abeaa | 619 | struct scsi_cmnd *scmd, *next; |
1da177e4 LT |
620 | int rtn; |
621 | ||
937abeaa | 622 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
3111b0d1 | 623 | if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || |
1da177e4 LT |
624 | SCSI_SENSE_VALID(scmd)) |
625 | continue; | |
626 | ||
3bf743e7 JG |
627 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
628 | "%s: requesting sense\n", | |
629 | current->comm)); | |
1da177e4 LT |
630 | rtn = scsi_request_sense(scmd); |
631 | if (rtn != SUCCESS) | |
632 | continue; | |
633 | ||
634 | SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" | |
635 | " result %x\n", scmd, | |
636 | scmd->result)); | |
637 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); | |
638 | ||
639 | rtn = scsi_decide_disposition(scmd); | |
640 | ||
641 | /* | |
642 | * if the result was normal, then just pass it along to the | |
643 | * upper level. | |
644 | */ | |
645 | if (rtn == SUCCESS) | |
646 | /* we don't want this command reissued, just | |
647 | * finished with the sense data, so set | |
648 | * retries to the max allowed to ensure it | |
649 | * won't get reissued */ | |
650 | scmd->retries = scmd->allowed; | |
651 | else if (rtn != NEEDS_RETRY) | |
652 | continue; | |
653 | ||
654 | scsi_eh_finish_cmd(scmd, done_q); | |
655 | } | |
656 | ||
657 | return list_empty(work_q); | |
658 | } | |
659 | ||
660 | /** | |
661 | * scsi_try_to_abort_cmd - Ask host to abort a running command. | |
662 | * @scmd: SCSI cmd to abort from Lower Level. | |
663 | * | |
664 | * Notes: | |
665 | * This function will not return until the user's completion function | |
666 | * has been called. there is no timeout on this operation. if the | |
667 | * author of the low-level driver wishes this operation to be timed, | |
668 | * they can provide this facility themselves. helper functions in | |
669 | * scsi_error.c can be supplied to make this easier to do. | |
670 | **/ | |
671 | static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) | |
672 | { | |
1da177e4 | 673 | if (!scmd->device->host->hostt->eh_abort_handler) |
8fa728a2 | 674 | return FAILED; |
1da177e4 LT |
675 | |
676 | /* | |
677 | * scsi_done was called just after the command timed out and before | |
678 | * we had a chance to process it. (db) | |
679 | */ | |
680 | if (scmd->serial_number == 0) | |
681 | return SUCCESS; | |
8fa728a2 | 682 | return scmd->device->host->hostt->eh_abort_handler(scmd); |
1da177e4 LT |
683 | } |
684 | ||
685 | /** | |
686 | * scsi_eh_tur - Send TUR to device. | |
687 | * @scmd: Scsi cmd to send TUR | |
688 | * | |
689 | * Return value: | |
690 | * 0 - Device is ready. 1 - Device NOT ready. | |
691 | **/ | |
692 | static int scsi_eh_tur(struct scsi_cmnd *scmd) | |
693 | { | |
694 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; | |
695 | int retry_cnt = 1, rtn; | |
793698ce | 696 | int saved_result; |
1da177e4 LT |
697 | |
698 | retry_tur: | |
699 | memcpy(scmd->cmnd, tur_command, sizeof(tur_command)); | |
700 | ||
701 | /* | |
702 | * zero the sense buffer. the scsi spec mandates that any | |
703 | * untransferred sense data should be interpreted as being zero. | |
704 | */ | |
705 | memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); | |
706 | ||
793698ce | 707 | saved_result = scmd->result; |
1da177e4 LT |
708 | scmd->request_buffer = NULL; |
709 | scmd->request_bufflen = 0; | |
710 | scmd->use_sg = 0; | |
711 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); | |
712 | scmd->underflow = 0; | |
713 | scmd->sc_data_direction = DMA_NONE; | |
714 | ||
715 | rtn = scsi_send_eh_cmnd(scmd, SENSE_TIMEOUT); | |
716 | ||
717 | /* | |
718 | * when we eventually call scsi_finish, we really wish to complete | |
719 | * the original request, so let's restore the original data. (db) | |
720 | */ | |
721 | scsi_setup_cmd_retry(scmd); | |
793698ce | 722 | scmd->result = saved_result; |
1da177e4 LT |
723 | |
724 | /* | |
725 | * hey, we are done. let's look to see what happened. | |
726 | */ | |
727 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", | |
728 | __FUNCTION__, scmd, rtn)); | |
729 | if (rtn == SUCCESS) | |
730 | return 0; | |
e47373ec | 731 | else if (rtn == NEEDS_RETRY) { |
1da177e4 LT |
732 | if (retry_cnt--) |
733 | goto retry_tur; | |
e47373ec AS |
734 | return 0; |
735 | } | |
1da177e4 LT |
736 | return 1; |
737 | } | |
738 | ||
739 | /** | |
740 | * scsi_eh_abort_cmds - abort canceled commands. | |
741 | * @shost: scsi host being recovered. | |
742 | * @eh_done_q: list_head for processed commands. | |
743 | * | |
744 | * Decription: | |
745 | * Try and see whether or not it makes sense to try and abort the | |
746 | * running command. this only works out to be the case if we have one | |
747 | * command that has timed out. if the command simply failed, it makes | |
748 | * no sense to try and abort the command, since as far as the shost | |
749 | * adapter is concerned, it isn't running. | |
750 | **/ | |
751 | static int scsi_eh_abort_cmds(struct list_head *work_q, | |
752 | struct list_head *done_q) | |
753 | { | |
937abeaa | 754 | struct scsi_cmnd *scmd, *next; |
1da177e4 LT |
755 | int rtn; |
756 | ||
937abeaa | 757 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
3111b0d1 | 758 | if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) |
1da177e4 LT |
759 | continue; |
760 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" | |
761 | "0x%p\n", current->comm, | |
762 | scmd)); | |
763 | rtn = scsi_try_to_abort_cmd(scmd); | |
764 | if (rtn == SUCCESS) { | |
3111b0d1 | 765 | scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; |
1da177e4 LT |
766 | if (!scsi_device_online(scmd->device) || |
767 | !scsi_eh_tur(scmd)) { | |
768 | scsi_eh_finish_cmd(scmd, done_q); | |
769 | } | |
770 | ||
771 | } else | |
772 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" | |
773 | " cmd failed:" | |
774 | "0x%p\n", | |
775 | current->comm, | |
776 | scmd)); | |
777 | } | |
778 | ||
779 | return list_empty(work_q); | |
780 | } | |
781 | ||
782 | /** | |
783 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev | |
784 | * @scmd: SCSI cmd used to send BDR | |
785 | * | |
786 | * Notes: | |
787 | * There is no timeout for this operation. if this operation is | |
788 | * unreliable for a given host, then the host itself needs to put a | |
789 | * timer on it, and set the host back to a consistent state prior to | |
790 | * returning. | |
791 | **/ | |
792 | static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) | |
793 | { | |
94d0e7b8 | 794 | int rtn; |
1da177e4 LT |
795 | |
796 | if (!scmd->device->host->hostt->eh_device_reset_handler) | |
94d0e7b8 | 797 | return FAILED; |
1da177e4 | 798 | |
1da177e4 | 799 | rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); |
1da177e4 LT |
800 | if (rtn == SUCCESS) { |
801 | scmd->device->was_reset = 1; | |
802 | scmd->device->expecting_cc_ua = 1; | |
803 | } | |
804 | ||
805 | return rtn; | |
806 | } | |
807 | ||
808 | /** | |
809 | * scsi_eh_try_stu - Send START_UNIT to device. | |
810 | * @scmd: Scsi cmd to send START_UNIT | |
811 | * | |
812 | * Return value: | |
813 | * 0 - Device is ready. 1 - Device NOT ready. | |
814 | **/ | |
815 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) | |
816 | { | |
817 | static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; | |
818 | int rtn; | |
793698ce | 819 | int saved_result; |
1da177e4 LT |
820 | |
821 | if (!scmd->device->allow_restart) | |
822 | return 1; | |
823 | ||
824 | memcpy(scmd->cmnd, stu_command, sizeof(stu_command)); | |
825 | ||
826 | /* | |
827 | * zero the sense buffer. the scsi spec mandates that any | |
828 | * untransferred sense data should be interpreted as being zero. | |
829 | */ | |
830 | memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); | |
831 | ||
793698ce | 832 | saved_result = scmd->result; |
1da177e4 LT |
833 | scmd->request_buffer = NULL; |
834 | scmd->request_bufflen = 0; | |
835 | scmd->use_sg = 0; | |
836 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); | |
837 | scmd->underflow = 0; | |
838 | scmd->sc_data_direction = DMA_NONE; | |
839 | ||
840 | rtn = scsi_send_eh_cmnd(scmd, START_UNIT_TIMEOUT); | |
841 | ||
842 | /* | |
843 | * when we eventually call scsi_finish, we really wish to complete | |
844 | * the original request, so let's restore the original data. (db) | |
845 | */ | |
846 | scsi_setup_cmd_retry(scmd); | |
793698ce | 847 | scmd->result = saved_result; |
1da177e4 LT |
848 | |
849 | /* | |
850 | * hey, we are done. let's look to see what happened. | |
851 | */ | |
852 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", | |
853 | __FUNCTION__, scmd, rtn)); | |
854 | if (rtn == SUCCESS) | |
855 | return 0; | |
856 | return 1; | |
857 | } | |
858 | ||
859 | /** | |
860 | * scsi_eh_stu - send START_UNIT if needed | |
861 | * @shost: scsi host being recovered. | |
862 | * @eh_done_q: list_head for processed commands. | |
863 | * | |
864 | * Notes: | |
865 | * If commands are failing due to not ready, initializing command required, | |
866 | * try revalidating the device, which will end up sending a start unit. | |
867 | **/ | |
868 | static int scsi_eh_stu(struct Scsi_Host *shost, | |
869 | struct list_head *work_q, | |
870 | struct list_head *done_q) | |
871 | { | |
937abeaa | 872 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
1da177e4 LT |
873 | struct scsi_device *sdev; |
874 | ||
875 | shost_for_each_device(sdev, shost) { | |
876 | stu_scmd = NULL; | |
877 | list_for_each_entry(scmd, work_q, eh_entry) | |
878 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && | |
879 | scsi_check_sense(scmd) == FAILED ) { | |
880 | stu_scmd = scmd; | |
881 | break; | |
882 | } | |
883 | ||
884 | if (!stu_scmd) | |
885 | continue; | |
886 | ||
887 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" | |
888 | " 0x%p\n", current->comm, sdev)); | |
889 | ||
890 | if (!scsi_eh_try_stu(stu_scmd)) { | |
891 | if (!scsi_device_online(sdev) || | |
892 | !scsi_eh_tur(stu_scmd)) { | |
937abeaa CH |
893 | list_for_each_entry_safe(scmd, next, |
894 | work_q, eh_entry) { | |
1da177e4 LT |
895 | if (scmd->device == sdev) |
896 | scsi_eh_finish_cmd(scmd, done_q); | |
897 | } | |
898 | } | |
899 | } else { | |
900 | SCSI_LOG_ERROR_RECOVERY(3, | |
901 | printk("%s: START_UNIT failed to sdev:" | |
902 | " 0x%p\n", current->comm, sdev)); | |
903 | } | |
904 | } | |
905 | ||
906 | return list_empty(work_q); | |
907 | } | |
908 | ||
909 | ||
910 | /** | |
911 | * scsi_eh_bus_device_reset - send bdr if needed | |
912 | * @shost: scsi host being recovered. | |
913 | * @eh_done_q: list_head for processed commands. | |
914 | * | |
915 | * Notes: | |
916 | * Try a bus device reset. still, look to see whether we have multiple | |
917 | * devices that are jammed or not - if we have multiple devices, it | |
918 | * makes no sense to try bus_device_reset - we really would need to try | |
919 | * a bus_reset instead. | |
920 | **/ | |
921 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, | |
922 | struct list_head *work_q, | |
923 | struct list_head *done_q) | |
924 | { | |
937abeaa | 925 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
1da177e4 LT |
926 | struct scsi_device *sdev; |
927 | int rtn; | |
928 | ||
929 | shost_for_each_device(sdev, shost) { | |
930 | bdr_scmd = NULL; | |
931 | list_for_each_entry(scmd, work_q, eh_entry) | |
932 | if (scmd->device == sdev) { | |
933 | bdr_scmd = scmd; | |
934 | break; | |
935 | } | |
936 | ||
937 | if (!bdr_scmd) | |
938 | continue; | |
939 | ||
940 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" | |
941 | " 0x%p\n", current->comm, | |
942 | sdev)); | |
943 | rtn = scsi_try_bus_device_reset(bdr_scmd); | |
944 | if (rtn == SUCCESS) { | |
945 | if (!scsi_device_online(sdev) || | |
946 | !scsi_eh_tur(bdr_scmd)) { | |
937abeaa CH |
947 | list_for_each_entry_safe(scmd, next, |
948 | work_q, eh_entry) { | |
1da177e4 LT |
949 | if (scmd->device == sdev) |
950 | scsi_eh_finish_cmd(scmd, | |
951 | done_q); | |
952 | } | |
953 | } | |
954 | } else { | |
955 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" | |
956 | " failed sdev:" | |
957 | "0x%p\n", | |
958 | current->comm, | |
959 | sdev)); | |
960 | } | |
961 | } | |
962 | ||
963 | return list_empty(work_q); | |
964 | } | |
965 | ||
966 | /** | |
967 | * scsi_try_bus_reset - ask host to perform a bus reset | |
968 | * @scmd: SCSI cmd to send bus reset. | |
969 | **/ | |
970 | static int scsi_try_bus_reset(struct scsi_cmnd *scmd) | |
971 | { | |
972 | unsigned long flags; | |
973 | int rtn; | |
974 | ||
975 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", | |
976 | __FUNCTION__)); | |
1da177e4 LT |
977 | |
978 | if (!scmd->device->host->hostt->eh_bus_reset_handler) | |
979 | return FAILED; | |
980 | ||
1da177e4 | 981 | rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); |
1da177e4 LT |
982 | |
983 | if (rtn == SUCCESS) { | |
984 | if (!scmd->device->host->hostt->skip_settle_delay) | |
985 | ssleep(BUS_RESET_SETTLE_TIME); | |
986 | spin_lock_irqsave(scmd->device->host->host_lock, flags); | |
422c0d61 JG |
987 | scsi_report_bus_reset(scmd->device->host, |
988 | scmd_channel(scmd)); | |
1da177e4 LT |
989 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
990 | } | |
991 | ||
992 | return rtn; | |
993 | } | |
994 | ||
995 | /** | |
996 | * scsi_try_host_reset - ask host adapter to reset itself | |
997 | * @scmd: SCSI cmd to send hsot reset. | |
998 | **/ | |
999 | static int scsi_try_host_reset(struct scsi_cmnd *scmd) | |
1000 | { | |
1001 | unsigned long flags; | |
1002 | int rtn; | |
1003 | ||
1004 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", | |
1005 | __FUNCTION__)); | |
1da177e4 LT |
1006 | |
1007 | if (!scmd->device->host->hostt->eh_host_reset_handler) | |
1008 | return FAILED; | |
1009 | ||
1da177e4 | 1010 | rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); |
1da177e4 LT |
1011 | |
1012 | if (rtn == SUCCESS) { | |
1013 | if (!scmd->device->host->hostt->skip_settle_delay) | |
1014 | ssleep(HOST_RESET_SETTLE_TIME); | |
1015 | spin_lock_irqsave(scmd->device->host->host_lock, flags); | |
422c0d61 JG |
1016 | scsi_report_bus_reset(scmd->device->host, |
1017 | scmd_channel(scmd)); | |
1da177e4 LT |
1018 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
1019 | } | |
1020 | ||
1021 | return rtn; | |
1022 | } | |
1023 | ||
1024 | /** | |
1025 | * scsi_eh_bus_reset - send a bus reset | |
1026 | * @shost: scsi host being recovered. | |
1027 | * @eh_done_q: list_head for processed commands. | |
1028 | **/ | |
1029 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, | |
1030 | struct list_head *work_q, | |
1031 | struct list_head *done_q) | |
1032 | { | |
937abeaa | 1033 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
1da177e4 LT |
1034 | unsigned int channel; |
1035 | int rtn; | |
1036 | ||
1037 | /* | |
1038 | * we really want to loop over the various channels, and do this on | |
1039 | * a channel by channel basis. we should also check to see if any | |
1040 | * of the failed commands are on soft_reset devices, and if so, skip | |
1041 | * the reset. | |
1042 | */ | |
1043 | ||
1044 | for (channel = 0; channel <= shost->max_channel; channel++) { | |
1045 | chan_scmd = NULL; | |
1046 | list_for_each_entry(scmd, work_q, eh_entry) { | |
422c0d61 | 1047 | if (channel == scmd_channel(scmd)) { |
1da177e4 LT |
1048 | chan_scmd = scmd; |
1049 | break; | |
1050 | /* | |
1051 | * FIXME add back in some support for | |
1052 | * soft_reset devices. | |
1053 | */ | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | if (!chan_scmd) | |
1058 | continue; | |
1059 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" | |
1060 | " %d\n", current->comm, | |
1061 | channel)); | |
1062 | rtn = scsi_try_bus_reset(chan_scmd); | |
1063 | if (rtn == SUCCESS) { | |
937abeaa | 1064 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
422c0d61 | 1065 | if (channel == scmd_channel(scmd)) |
1da177e4 LT |
1066 | if (!scsi_device_online(scmd->device) || |
1067 | !scsi_eh_tur(scmd)) | |
1068 | scsi_eh_finish_cmd(scmd, | |
1069 | done_q); | |
1070 | } | |
1071 | } else { | |
1072 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" | |
1073 | " failed chan: %d\n", | |
1074 | current->comm, | |
1075 | channel)); | |
1076 | } | |
1077 | } | |
1078 | return list_empty(work_q); | |
1079 | } | |
1080 | ||
1081 | /** | |
1082 | * scsi_eh_host_reset - send a host reset | |
1083 | * @work_q: list_head for processed commands. | |
1084 | * @done_q: list_head for processed commands. | |
1085 | **/ | |
1086 | static int scsi_eh_host_reset(struct list_head *work_q, | |
1087 | struct list_head *done_q) | |
1088 | { | |
937abeaa | 1089 | struct scsi_cmnd *scmd, *next; |
1da177e4 | 1090 | int rtn; |
1da177e4 LT |
1091 | |
1092 | if (!list_empty(work_q)) { | |
1093 | scmd = list_entry(work_q->next, | |
1094 | struct scsi_cmnd, eh_entry); | |
1095 | ||
1096 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" | |
1097 | , current->comm)); | |
1098 | ||
1099 | rtn = scsi_try_host_reset(scmd); | |
1100 | if (rtn == SUCCESS) { | |
937abeaa | 1101 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
1da177e4 LT |
1102 | if (!scsi_device_online(scmd->device) || |
1103 | (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || | |
1104 | !scsi_eh_tur(scmd)) | |
1105 | scsi_eh_finish_cmd(scmd, done_q); | |
1106 | } | |
1107 | } else { | |
1108 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" | |
1109 | " failed\n", | |
1110 | current->comm)); | |
1111 | } | |
1112 | } | |
1113 | return list_empty(work_q); | |
1114 | } | |
1115 | ||
1116 | /** | |
1117 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover | |
1118 | * @work_q: list_head for processed commands. | |
1119 | * @done_q: list_head for processed commands. | |
1120 | * | |
1121 | **/ | |
1122 | static void scsi_eh_offline_sdevs(struct list_head *work_q, | |
1123 | struct list_head *done_q) | |
1124 | { | |
937abeaa | 1125 | struct scsi_cmnd *scmd, *next; |
1da177e4 | 1126 | |
937abeaa | 1127 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
9ccfc756 JB |
1128 | sdev_printk(KERN_INFO, scmd->device, |
1129 | "scsi: Device offlined - not" | |
1130 | " ready after error recovery\n"); | |
1da177e4 | 1131 | scsi_device_set_state(scmd->device, SDEV_OFFLINE); |
3111b0d1 | 1132 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { |
1da177e4 LT |
1133 | /* |
1134 | * FIXME: Handle lost cmds. | |
1135 | */ | |
1136 | } | |
1137 | scsi_eh_finish_cmd(scmd, done_q); | |
1138 | } | |
1139 | return; | |
1140 | } | |
1141 | ||
1142 | /** | |
1143 | * scsi_decide_disposition - Disposition a cmd on return from LLD. | |
1144 | * @scmd: SCSI cmd to examine. | |
1145 | * | |
1146 | * Notes: | |
1147 | * This is *only* called when we are examining the status after sending | |
1148 | * out the actual data command. any commands that are queued for error | |
1149 | * recovery (e.g. test_unit_ready) do *not* come through here. | |
1150 | * | |
1151 | * When this routine returns failed, it means the error handler thread | |
1152 | * is woken. In cases where the error code indicates an error that | |
1153 | * doesn't require the error handler read (i.e. we don't need to | |
1154 | * abort/reset), this function should return SUCCESS. | |
1155 | **/ | |
1156 | int scsi_decide_disposition(struct scsi_cmnd *scmd) | |
1157 | { | |
1158 | int rtn; | |
1159 | ||
1160 | /* | |
1161 | * if the device is offline, then we clearly just pass the result back | |
1162 | * up to the top level. | |
1163 | */ | |
1164 | if (!scsi_device_online(scmd->device)) { | |
1165 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" | |
1166 | " as SUCCESS\n", | |
1167 | __FUNCTION__)); | |
1168 | return SUCCESS; | |
1169 | } | |
1170 | ||
1171 | /* | |
1172 | * first check the host byte, to see if there is anything in there | |
1173 | * that would indicate what we need to do. | |
1174 | */ | |
1175 | switch (host_byte(scmd->result)) { | |
1176 | case DID_PASSTHROUGH: | |
1177 | /* | |
1178 | * no matter what, pass this through to the upper layer. | |
1179 | * nuke this special code so that it looks like we are saying | |
1180 | * did_ok. | |
1181 | */ | |
1182 | scmd->result &= 0xff00ffff; | |
1183 | return SUCCESS; | |
1184 | case DID_OK: | |
1185 | /* | |
1186 | * looks good. drop through, and check the next byte. | |
1187 | */ | |
1188 | break; | |
1189 | case DID_NO_CONNECT: | |
1190 | case DID_BAD_TARGET: | |
1191 | case DID_ABORT: | |
1192 | /* | |
1193 | * note - this means that we just report the status back | |
1194 | * to the top level driver, not that we actually think | |
1195 | * that it indicates SUCCESS. | |
1196 | */ | |
1197 | return SUCCESS; | |
1198 | /* | |
1199 | * when the low level driver returns did_soft_error, | |
1200 | * it is responsible for keeping an internal retry counter | |
1201 | * in order to avoid endless loops (db) | |
1202 | * | |
1203 | * actually this is a bug in this function here. we should | |
1204 | * be mindful of the maximum number of retries specified | |
1205 | * and not get stuck in a loop. | |
1206 | */ | |
1207 | case DID_SOFT_ERROR: | |
1208 | goto maybe_retry; | |
1209 | case DID_IMM_RETRY: | |
1210 | return NEEDS_RETRY; | |
1211 | ||
bf341919 | 1212 | case DID_REQUEUE: |
1213 | return ADD_TO_MLQUEUE; | |
1214 | ||
1da177e4 LT |
1215 | case DID_ERROR: |
1216 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && | |
1217 | status_byte(scmd->result) == RESERVATION_CONFLICT) | |
1218 | /* | |
1219 | * execute reservation conflict processing code | |
1220 | * lower down | |
1221 | */ | |
1222 | break; | |
1223 | /* fallthrough */ | |
1224 | ||
1225 | case DID_BUS_BUSY: | |
1226 | case DID_PARITY: | |
1227 | goto maybe_retry; | |
1228 | case DID_TIME_OUT: | |
1229 | /* | |
1230 | * when we scan the bus, we get timeout messages for | |
1231 | * these commands if there is no device available. | |
1232 | * other hosts report did_no_connect for the same thing. | |
1233 | */ | |
1234 | if ((scmd->cmnd[0] == TEST_UNIT_READY || | |
1235 | scmd->cmnd[0] == INQUIRY)) { | |
1236 | return SUCCESS; | |
1237 | } else { | |
1238 | return FAILED; | |
1239 | } | |
1240 | case DID_RESET: | |
1241 | return SUCCESS; | |
1242 | default: | |
1243 | return FAILED; | |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * next, check the message byte. | |
1248 | */ | |
1249 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) | |
1250 | return FAILED; | |
1251 | ||
1252 | /* | |
1253 | * check the status byte to see if this indicates anything special. | |
1254 | */ | |
1255 | switch (status_byte(scmd->result)) { | |
1256 | case QUEUE_FULL: | |
1257 | /* | |
1258 | * the case of trying to send too many commands to a | |
1259 | * tagged queueing device. | |
1260 | */ | |
1261 | case BUSY: | |
1262 | /* | |
1263 | * device can't talk to us at the moment. Should only | |
1264 | * occur (SAM-3) when the task queue is empty, so will cause | |
1265 | * the empty queue handling to trigger a stall in the | |
1266 | * device. | |
1267 | */ | |
1268 | return ADD_TO_MLQUEUE; | |
1269 | case GOOD: | |
1270 | case COMMAND_TERMINATED: | |
1271 | case TASK_ABORTED: | |
1272 | return SUCCESS; | |
1273 | case CHECK_CONDITION: | |
1274 | rtn = scsi_check_sense(scmd); | |
1275 | if (rtn == NEEDS_RETRY) | |
1276 | goto maybe_retry; | |
1277 | /* if rtn == FAILED, we have no sense information; | |
1278 | * returning FAILED will wake the error handler thread | |
1279 | * to collect the sense and redo the decide | |
1280 | * disposition */ | |
1281 | return rtn; | |
1282 | case CONDITION_GOOD: | |
1283 | case INTERMEDIATE_GOOD: | |
1284 | case INTERMEDIATE_C_GOOD: | |
1285 | case ACA_ACTIVE: | |
1286 | /* | |
1287 | * who knows? FIXME(eric) | |
1288 | */ | |
1289 | return SUCCESS; | |
1290 | ||
1291 | case RESERVATION_CONFLICT: | |
9ccfc756 JB |
1292 | sdev_printk(KERN_INFO, scmd->device, |
1293 | "reservation conflict\n"); | |
1da177e4 LT |
1294 | return SUCCESS; /* causes immediate i/o error */ |
1295 | default: | |
1296 | return FAILED; | |
1297 | } | |
1298 | return FAILED; | |
1299 | ||
1300 | maybe_retry: | |
1301 | ||
1302 | /* we requeue for retry because the error was retryable, and | |
1303 | * the request was not marked fast fail. Note that above, | |
1304 | * even if the request is marked fast fail, we still requeue | |
1305 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ | |
1306 | if ((++scmd->retries) < scmd->allowed | |
1307 | && !blk_noretry_request(scmd->request)) { | |
1308 | return NEEDS_RETRY; | |
1309 | } else { | |
1310 | /* | |
1311 | * no more retries - report this one back to upper level. | |
1312 | */ | |
1313 | return SUCCESS; | |
1314 | } | |
1315 | } | |
1316 | ||
1317 | /** | |
1318 | * scsi_eh_lock_done - done function for eh door lock request | |
1319 | * @scmd: SCSI command block for the door lock request | |
1320 | * | |
1321 | * Notes: | |
1322 | * We completed the asynchronous door lock request, and it has either | |
1323 | * locked the door or failed. We must free the command structures | |
1324 | * associated with this request. | |
1325 | **/ | |
1326 | static void scsi_eh_lock_done(struct scsi_cmnd *scmd) | |
1327 | { | |
1328 | struct scsi_request *sreq = scmd->sc_request; | |
1329 | ||
1330 | scsi_release_request(sreq); | |
1331 | } | |
1332 | ||
1333 | ||
1334 | /** | |
1335 | * scsi_eh_lock_door - Prevent medium removal for the specified device | |
1336 | * @sdev: SCSI device to prevent medium removal | |
1337 | * | |
1338 | * Locking: | |
1339 | * We must be called from process context; scsi_allocate_request() | |
1340 | * may sleep. | |
1341 | * | |
1342 | * Notes: | |
1343 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the | |
1344 | * head of the devices request queue, and continue. | |
1345 | * | |
1346 | * Bugs: | |
1347 | * scsi_allocate_request() may sleep waiting for existing requests to | |
1348 | * be processed. However, since we haven't kicked off any request | |
1349 | * processing for this host, this may deadlock. | |
1350 | * | |
1351 | * If scsi_allocate_request() fails for what ever reason, we | |
1352 | * completely forget to lock the door. | |
1353 | **/ | |
1354 | static void scsi_eh_lock_door(struct scsi_device *sdev) | |
1355 | { | |
1356 | struct scsi_request *sreq = scsi_allocate_request(sdev, GFP_KERNEL); | |
1357 | ||
1358 | if (unlikely(!sreq)) { | |
1359 | printk(KERN_ERR "%s: request allocate failed," | |
1360 | "prevent media removal cmd not sent\n", __FUNCTION__); | |
1361 | return; | |
1362 | } | |
1363 | ||
1364 | sreq->sr_cmnd[0] = ALLOW_MEDIUM_REMOVAL; | |
1365 | sreq->sr_cmnd[1] = 0; | |
1366 | sreq->sr_cmnd[2] = 0; | |
1367 | sreq->sr_cmnd[3] = 0; | |
1368 | sreq->sr_cmnd[4] = SCSI_REMOVAL_PREVENT; | |
1369 | sreq->sr_cmnd[5] = 0; | |
1370 | sreq->sr_data_direction = DMA_NONE; | |
1371 | sreq->sr_bufflen = 0; | |
1372 | sreq->sr_buffer = NULL; | |
1373 | sreq->sr_allowed = 5; | |
1374 | sreq->sr_done = scsi_eh_lock_done; | |
1375 | sreq->sr_timeout_per_command = 10 * HZ; | |
1376 | sreq->sr_cmd_len = COMMAND_SIZE(sreq->sr_cmnd[0]); | |
1377 | ||
1378 | scsi_insert_special_req(sreq, 1); | |
1379 | } | |
1380 | ||
1381 | ||
1382 | /** | |
1383 | * scsi_restart_operations - restart io operations to the specified host. | |
1384 | * @shost: Host we are restarting. | |
1385 | * | |
1386 | * Notes: | |
1387 | * When we entered the error handler, we blocked all further i/o to | |
1388 | * this device. we need to 'reverse' this process. | |
1389 | **/ | |
1390 | static void scsi_restart_operations(struct Scsi_Host *shost) | |
1391 | { | |
1392 | struct scsi_device *sdev; | |
939647ee | 1393 | unsigned long flags; |
1da177e4 LT |
1394 | |
1395 | /* | |
1396 | * If the door was locked, we need to insert a door lock request | |
1397 | * onto the head of the SCSI request queue for the device. There | |
1398 | * is no point trying to lock the door of an off-line device. | |
1399 | */ | |
1400 | shost_for_each_device(sdev, shost) { | |
1401 | if (scsi_device_online(sdev) && sdev->locked) | |
1402 | scsi_eh_lock_door(sdev); | |
1403 | } | |
1404 | ||
1405 | /* | |
1406 | * next free up anything directly waiting upon the host. this | |
1407 | * will be requests for character device operations, and also for | |
1408 | * ioctls to queued block devices. | |
1409 | */ | |
1410 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", | |
1411 | __FUNCTION__)); | |
1412 | ||
939647ee JB |
1413 | spin_lock_irqsave(shost->host_lock, flags); |
1414 | if (scsi_host_set_state(shost, SHOST_RUNNING)) | |
1415 | if (scsi_host_set_state(shost, SHOST_CANCEL)) | |
1416 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); | |
1417 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1da177e4 LT |
1418 | |
1419 | wake_up(&shost->host_wait); | |
1420 | ||
1421 | /* | |
1422 | * finally we need to re-initiate requests that may be pending. we will | |
1423 | * have had everything blocked while error handling is taking place, and | |
1424 | * now that error recovery is done, we will need to ensure that these | |
1425 | * requests are started. | |
1426 | */ | |
1427 | scsi_run_host_queues(shost); | |
1428 | } | |
1429 | ||
1430 | /** | |
1431 | * scsi_eh_ready_devs - check device ready state and recover if not. | |
1432 | * @shost: host to be recovered. | |
1433 | * @eh_done_q: list_head for processed commands. | |
1434 | * | |
1435 | **/ | |
1436 | static void scsi_eh_ready_devs(struct Scsi_Host *shost, | |
1437 | struct list_head *work_q, | |
1438 | struct list_head *done_q) | |
1439 | { | |
1440 | if (!scsi_eh_stu(shost, work_q, done_q)) | |
1441 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) | |
1442 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) | |
1443 | if (!scsi_eh_host_reset(work_q, done_q)) | |
1444 | scsi_eh_offline_sdevs(work_q, done_q); | |
1445 | } | |
1446 | ||
1447 | /** | |
1448 | * scsi_eh_flush_done_q - finish processed commands or retry them. | |
1449 | * @done_q: list_head of processed commands. | |
1450 | * | |
1451 | **/ | |
1452 | static void scsi_eh_flush_done_q(struct list_head *done_q) | |
1453 | { | |
937abeaa | 1454 | struct scsi_cmnd *scmd, *next; |
1da177e4 | 1455 | |
937abeaa CH |
1456 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
1457 | list_del_init(&scmd->eh_entry); | |
1da177e4 LT |
1458 | if (scsi_device_online(scmd->device) && |
1459 | !blk_noretry_request(scmd->request) && | |
1460 | (++scmd->retries < scmd->allowed)) { | |
1461 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" | |
1462 | " retry cmd: %p\n", | |
1463 | current->comm, | |
1464 | scmd)); | |
1465 | scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); | |
1466 | } else { | |
793698ce PM |
1467 | /* |
1468 | * If just we got sense for the device (called | |
1469 | * scsi_eh_get_sense), scmd->result is already | |
1470 | * set, do not set DRIVER_TIMEOUT. | |
1471 | */ | |
1da177e4 LT |
1472 | if (!scmd->result) |
1473 | scmd->result |= (DRIVER_TIMEOUT << 24); | |
1474 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" | |
1475 | " cmd: %p\n", | |
1476 | current->comm, scmd)); | |
1477 | scsi_finish_command(scmd); | |
1478 | } | |
1479 | } | |
1480 | } | |
1481 | ||
1482 | /** | |
1483 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. | |
1484 | * @shost: Host to unjam. | |
1485 | * | |
1486 | * Notes: | |
1487 | * When we come in here, we *know* that all commands on the bus have | |
1488 | * either completed, failed or timed out. we also know that no further | |
1489 | * commands are being sent to the host, so things are relatively quiet | |
1490 | * and we have freedom to fiddle with things as we wish. | |
1491 | * | |
1492 | * This is only the *default* implementation. it is possible for | |
1493 | * individual drivers to supply their own version of this function, and | |
1494 | * if the maintainer wishes to do this, it is strongly suggested that | |
1495 | * this function be taken as a template and modified. this function | |
1496 | * was designed to correctly handle problems for about 95% of the | |
1497 | * different cases out there, and it should always provide at least a | |
1498 | * reasonable amount of error recovery. | |
1499 | * | |
1500 | * Any command marked 'failed' or 'timeout' must eventually have | |
1501 | * scsi_finish_cmd() called for it. we do all of the retry stuff | |
1502 | * here, so when we restart the host after we return it should have an | |
1503 | * empty queue. | |
1504 | **/ | |
1505 | static void scsi_unjam_host(struct Scsi_Host *shost) | |
1506 | { | |
1507 | unsigned long flags; | |
1508 | LIST_HEAD(eh_work_q); | |
1509 | LIST_HEAD(eh_done_q); | |
1510 | ||
1511 | spin_lock_irqsave(shost->host_lock, flags); | |
1512 | list_splice_init(&shost->eh_cmd_q, &eh_work_q); | |
1513 | spin_unlock_irqrestore(shost->host_lock, flags); | |
1514 | ||
1515 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); | |
1516 | ||
1517 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) | |
1518 | if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) | |
1519 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); | |
1520 | ||
1521 | scsi_eh_flush_done_q(&eh_done_q); | |
1522 | } | |
1523 | ||
1524 | /** | |
ad42eb1b | 1525 | * scsi_error_handler - SCSI error handler thread |
1da177e4 LT |
1526 | * @data: Host for which we are running. |
1527 | * | |
1528 | * Notes: | |
ad42eb1b CH |
1529 | * This is the main error handling loop. This is run as a kernel thread |
1530 | * for every SCSI host and handles all error handling activity. | |
1da177e4 LT |
1531 | **/ |
1532 | int scsi_error_handler(void *data) | |
1533 | { | |
ad42eb1b | 1534 | struct Scsi_Host *shost = data; |
1da177e4 | 1535 | |
1da177e4 | 1536 | current->flags |= PF_NOFREEZE; |
1da177e4 LT |
1537 | |
1538 | /* | |
ad42eb1b CH |
1539 | * We use TASK_INTERRUPTIBLE so that the thread is not |
1540 | * counted against the load average as a running process. | |
1541 | * We never actually get interrupted because kthread_run | |
1542 | * disables singal delivery for the created thread. | |
1da177e4 | 1543 | */ |
3ed7a470 JB |
1544 | set_current_state(TASK_INTERRUPTIBLE); |
1545 | while (!kthread_should_stop()) { | |
1546 | if (shost->host_failed == 0 || | |
1547 | shost->host_failed != shost->host_busy) { | |
ad42eb1b CH |
1548 | SCSI_LOG_ERROR_RECOVERY(1, |
1549 | printk("Error handler scsi_eh_%d sleeping\n", | |
1550 | shost->host_no)); | |
3ed7a470 JB |
1551 | schedule(); |
1552 | set_current_state(TASK_INTERRUPTIBLE); | |
1553 | continue; | |
1554 | } | |
1da177e4 | 1555 | |
3ed7a470 | 1556 | __set_current_state(TASK_RUNNING); |
ad42eb1b CH |
1557 | SCSI_LOG_ERROR_RECOVERY(1, |
1558 | printk("Error handler scsi_eh_%d waking up\n", | |
1559 | shost->host_no)); | |
1da177e4 | 1560 | |
1da177e4 LT |
1561 | /* |
1562 | * We have a host that is failing for some reason. Figure out | |
1563 | * what we need to do to get it up and online again (if we can). | |
1564 | * If we fail, we end up taking the thing offline. | |
1565 | */ | |
1566 | if (shost->hostt->eh_strategy_handler) | |
ad42eb1b | 1567 | shost->hostt->eh_strategy_handler(shost); |
1da177e4 LT |
1568 | else |
1569 | scsi_unjam_host(shost); | |
1570 | ||
1da177e4 LT |
1571 | /* |
1572 | * Note - if the above fails completely, the action is to take | |
1573 | * individual devices offline and flush the queue of any | |
1574 | * outstanding requests that may have been pending. When we | |
1575 | * restart, we restart any I/O to any other devices on the bus | |
1576 | * which are still online. | |
1577 | */ | |
1578 | scsi_restart_operations(shost); | |
3ed7a470 | 1579 | set_current_state(TASK_INTERRUPTIBLE); |
1da177e4 | 1580 | } |
461a0ffb SR |
1581 | __set_current_state(TASK_RUNNING); |
1582 | ||
ad42eb1b CH |
1583 | SCSI_LOG_ERROR_RECOVERY(1, |
1584 | printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); | |
3ed7a470 | 1585 | shost->ehandler = NULL; |
1da177e4 LT |
1586 | return 0; |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * Function: scsi_report_bus_reset() | |
1591 | * | |
1592 | * Purpose: Utility function used by low-level drivers to report that | |
1593 | * they have observed a bus reset on the bus being handled. | |
1594 | * | |
1595 | * Arguments: shost - Host in question | |
1596 | * channel - channel on which reset was observed. | |
1597 | * | |
1598 | * Returns: Nothing | |
1599 | * | |
1600 | * Lock status: Host lock must be held. | |
1601 | * | |
1602 | * Notes: This only needs to be called if the reset is one which | |
1603 | * originates from an unknown location. Resets originated | |
1604 | * by the mid-level itself don't need to call this, but there | |
1605 | * should be no harm. | |
1606 | * | |
1607 | * The main purpose of this is to make sure that a CHECK_CONDITION | |
1608 | * is properly treated. | |
1609 | */ | |
1610 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) | |
1611 | { | |
1612 | struct scsi_device *sdev; | |
1613 | ||
1614 | __shost_for_each_device(sdev, shost) { | |
422c0d61 | 1615 | if (channel == sdev_channel(sdev)) { |
1da177e4 LT |
1616 | sdev->was_reset = 1; |
1617 | sdev->expecting_cc_ua = 1; | |
1618 | } | |
1619 | } | |
1620 | } | |
1621 | EXPORT_SYMBOL(scsi_report_bus_reset); | |
1622 | ||
1623 | /* | |
1624 | * Function: scsi_report_device_reset() | |
1625 | * | |
1626 | * Purpose: Utility function used by low-level drivers to report that | |
1627 | * they have observed a device reset on the device being handled. | |
1628 | * | |
1629 | * Arguments: shost - Host in question | |
1630 | * channel - channel on which reset was observed | |
1631 | * target - target on which reset was observed | |
1632 | * | |
1633 | * Returns: Nothing | |
1634 | * | |
1635 | * Lock status: Host lock must be held | |
1636 | * | |
1637 | * Notes: This only needs to be called if the reset is one which | |
1638 | * originates from an unknown location. Resets originated | |
1639 | * by the mid-level itself don't need to call this, but there | |
1640 | * should be no harm. | |
1641 | * | |
1642 | * The main purpose of this is to make sure that a CHECK_CONDITION | |
1643 | * is properly treated. | |
1644 | */ | |
1645 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) | |
1646 | { | |
1647 | struct scsi_device *sdev; | |
1648 | ||
1649 | __shost_for_each_device(sdev, shost) { | |
422c0d61 JG |
1650 | if (channel == sdev_channel(sdev) && |
1651 | target == sdev_id(sdev)) { | |
1da177e4 LT |
1652 | sdev->was_reset = 1; |
1653 | sdev->expecting_cc_ua = 1; | |
1654 | } | |
1655 | } | |
1656 | } | |
1657 | EXPORT_SYMBOL(scsi_report_device_reset); | |
1658 | ||
1659 | static void | |
1660 | scsi_reset_provider_done_command(struct scsi_cmnd *scmd) | |
1661 | { | |
1662 | } | |
1663 | ||
1664 | /* | |
1665 | * Function: scsi_reset_provider | |
1666 | * | |
1667 | * Purpose: Send requested reset to a bus or device at any phase. | |
1668 | * | |
1669 | * Arguments: device - device to send reset to | |
1670 | * flag - reset type (see scsi.h) | |
1671 | * | |
1672 | * Returns: SUCCESS/FAILURE. | |
1673 | * | |
1674 | * Notes: This is used by the SCSI Generic driver to provide | |
1675 | * Bus/Device reset capability. | |
1676 | */ | |
1677 | int | |
1678 | scsi_reset_provider(struct scsi_device *dev, int flag) | |
1679 | { | |
1680 | struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); | |
1681 | struct request req; | |
1682 | int rtn; | |
1683 | ||
1684 | scmd->request = &req; | |
1685 | memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); | |
1686 | scmd->request->rq_status = RQ_SCSI_BUSY; | |
b4edcbca | 1687 | |
1da177e4 LT |
1688 | memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd)); |
1689 | ||
1690 | scmd->scsi_done = scsi_reset_provider_done_command; | |
1691 | scmd->done = NULL; | |
1692 | scmd->buffer = NULL; | |
1693 | scmd->bufflen = 0; | |
1694 | scmd->request_buffer = NULL; | |
1695 | scmd->request_bufflen = 0; | |
1da177e4 LT |
1696 | |
1697 | scmd->cmd_len = 0; | |
1698 | ||
1699 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; | |
1700 | scmd->sc_request = NULL; | |
1701 | scmd->sc_magic = SCSI_CMND_MAGIC; | |
1702 | ||
1703 | init_timer(&scmd->eh_timeout); | |
1704 | ||
1705 | /* | |
1706 | * Sometimes the command can get back into the timer chain, | |
1707 | * so use the pid as an identifier. | |
1708 | */ | |
1709 | scmd->pid = 0; | |
1710 | ||
1711 | switch (flag) { | |
1712 | case SCSI_TRY_RESET_DEVICE: | |
1713 | rtn = scsi_try_bus_device_reset(scmd); | |
1714 | if (rtn == SUCCESS) | |
1715 | break; | |
1716 | /* FALLTHROUGH */ | |
1717 | case SCSI_TRY_RESET_BUS: | |
1718 | rtn = scsi_try_bus_reset(scmd); | |
1719 | if (rtn == SUCCESS) | |
1720 | break; | |
1721 | /* FALLTHROUGH */ | |
1722 | case SCSI_TRY_RESET_HOST: | |
1723 | rtn = scsi_try_host_reset(scmd); | |
1724 | break; | |
1725 | default: | |
1726 | rtn = FAILED; | |
1727 | } | |
1728 | ||
1da177e4 LT |
1729 | scsi_next_command(scmd); |
1730 | return rtn; | |
1731 | } | |
1732 | EXPORT_SYMBOL(scsi_reset_provider); | |
1733 | ||
1734 | /** | |
1735 | * scsi_normalize_sense - normalize main elements from either fixed or | |
1736 | * descriptor sense data format into a common format. | |
1737 | * | |
1738 | * @sense_buffer: byte array containing sense data returned by device | |
1739 | * @sb_len: number of valid bytes in sense_buffer | |
1740 | * @sshdr: pointer to instance of structure that common | |
1741 | * elements are written to. | |
1742 | * | |
1743 | * Notes: | |
1744 | * The "main elements" from sense data are: response_code, sense_key, | |
1745 | * asc, ascq and additional_length (only for descriptor format). | |
1746 | * | |
1747 | * Typically this function can be called after a device has | |
1748 | * responded to a SCSI command with the CHECK_CONDITION status. | |
1749 | * | |
1750 | * Return value: | |
1751 | * 1 if valid sense data information found, else 0; | |
1752 | **/ | |
1753 | int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, | |
1754 | struct scsi_sense_hdr *sshdr) | |
1755 | { | |
33aa687d | 1756 | if (!sense_buffer || !sb_len) |
1da177e4 LT |
1757 | return 0; |
1758 | ||
1759 | memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); | |
1760 | ||
1761 | sshdr->response_code = (sense_buffer[0] & 0x7f); | |
33aa687d JB |
1762 | |
1763 | if (!scsi_sense_valid(sshdr)) | |
1764 | return 0; | |
1765 | ||
1da177e4 LT |
1766 | if (sshdr->response_code >= 0x72) { |
1767 | /* | |
1768 | * descriptor format | |
1769 | */ | |
1770 | if (sb_len > 1) | |
1771 | sshdr->sense_key = (sense_buffer[1] & 0xf); | |
1772 | if (sb_len > 2) | |
1773 | sshdr->asc = sense_buffer[2]; | |
1774 | if (sb_len > 3) | |
1775 | sshdr->ascq = sense_buffer[3]; | |
1776 | if (sb_len > 7) | |
1777 | sshdr->additional_length = sense_buffer[7]; | |
1778 | } else { | |
1779 | /* | |
1780 | * fixed format | |
1781 | */ | |
1782 | if (sb_len > 2) | |
1783 | sshdr->sense_key = (sense_buffer[2] & 0xf); | |
1784 | if (sb_len > 7) { | |
1785 | sb_len = (sb_len < (sense_buffer[7] + 8)) ? | |
1786 | sb_len : (sense_buffer[7] + 8); | |
1787 | if (sb_len > 12) | |
1788 | sshdr->asc = sense_buffer[12]; | |
1789 | if (sb_len > 13) | |
1790 | sshdr->ascq = sense_buffer[13]; | |
1791 | } | |
1792 | } | |
1793 | ||
1794 | return 1; | |
1795 | } | |
1796 | EXPORT_SYMBOL(scsi_normalize_sense); | |
1797 | ||
1798 | int scsi_request_normalize_sense(struct scsi_request *sreq, | |
1799 | struct scsi_sense_hdr *sshdr) | |
1800 | { | |
1801 | return scsi_normalize_sense(sreq->sr_sense_buffer, | |
1802 | sizeof(sreq->sr_sense_buffer), sshdr); | |
1803 | } | |
1804 | EXPORT_SYMBOL(scsi_request_normalize_sense); | |
1805 | ||
1806 | int scsi_command_normalize_sense(struct scsi_cmnd *cmd, | |
1807 | struct scsi_sense_hdr *sshdr) | |
1808 | { | |
1809 | return scsi_normalize_sense(cmd->sense_buffer, | |
1810 | sizeof(cmd->sense_buffer), sshdr); | |
1811 | } | |
1812 | EXPORT_SYMBOL(scsi_command_normalize_sense); | |
1813 | ||
1814 | /** | |
1815 | * scsi_sense_desc_find - search for a given descriptor type in | |
1816 | * descriptor sense data format. | |
1817 | * | |
1818 | * @sense_buffer: byte array of descriptor format sense data | |
1819 | * @sb_len: number of valid bytes in sense_buffer | |
1820 | * @desc_type: value of descriptor type to find | |
1821 | * (e.g. 0 -> information) | |
1822 | * | |
1823 | * Notes: | |
1824 | * only valid when sense data is in descriptor format | |
1825 | * | |
1826 | * Return value: | |
1827 | * pointer to start of (first) descriptor if found else NULL | |
1828 | **/ | |
1829 | const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, | |
1830 | int desc_type) | |
1831 | { | |
1832 | int add_sen_len, add_len, desc_len, k; | |
1833 | const u8 * descp; | |
1834 | ||
1835 | if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) | |
1836 | return NULL; | |
1837 | if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) | |
1838 | return NULL; | |
1839 | add_sen_len = (add_sen_len < (sb_len - 8)) ? | |
1840 | add_sen_len : (sb_len - 8); | |
1841 | descp = &sense_buffer[8]; | |
1842 | for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { | |
1843 | descp += desc_len; | |
1844 | add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; | |
1845 | desc_len = add_len + 2; | |
1846 | if (descp[0] == desc_type) | |
1847 | return descp; | |
1848 | if (add_len < 0) // short descriptor ?? | |
1849 | break; | |
1850 | } | |
1851 | return NULL; | |
1852 | } | |
1853 | EXPORT_SYMBOL(scsi_sense_desc_find); | |
1854 | ||
1855 | /** | |
1856 | * scsi_get_sense_info_fld - attempts to get information field from | |
1857 | * sense data (either fixed or descriptor format) | |
1858 | * | |
1859 | * @sense_buffer: byte array of sense data | |
1860 | * @sb_len: number of valid bytes in sense_buffer | |
1861 | * @info_out: pointer to 64 integer where 8 or 4 byte information | |
1862 | * field will be placed if found. | |
1863 | * | |
1864 | * Return value: | |
1865 | * 1 if information field found, 0 if not found. | |
1866 | **/ | |
1867 | int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, | |
1868 | u64 * info_out) | |
1869 | { | |
1870 | int j; | |
1871 | const u8 * ucp; | |
1872 | u64 ull; | |
1873 | ||
1874 | if (sb_len < 7) | |
1875 | return 0; | |
1876 | switch (sense_buffer[0] & 0x7f) { | |
1877 | case 0x70: | |
1878 | case 0x71: | |
1879 | if (sense_buffer[0] & 0x80) { | |
1880 | *info_out = (sense_buffer[3] << 24) + | |
1881 | (sense_buffer[4] << 16) + | |
1882 | (sense_buffer[5] << 8) + sense_buffer[6]; | |
1883 | return 1; | |
1884 | } else | |
1885 | return 0; | |
1886 | case 0x72: | |
1887 | case 0x73: | |
1888 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, | |
1889 | 0 /* info desc */); | |
1890 | if (ucp && (0xa == ucp[1])) { | |
1891 | ull = 0; | |
1892 | for (j = 0; j < 8; ++j) { | |
1893 | if (j > 0) | |
1894 | ull <<= 8; | |
1895 | ull |= ucp[4 + j]; | |
1896 | } | |
1897 | *info_out = ull; | |
1898 | return 1; | |
1899 | } else | |
1900 | return 0; | |
1901 | default: | |
1902 | return 0; | |
1903 | } | |
1904 | } | |
1905 | EXPORT_SYMBOL(scsi_get_sense_info_fld); |