blob: d45c69ca5737b335644fb4490bedf3fd6cf8779c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/timer.h>
21#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070023#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020024#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
28
29#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020030#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <scsi/scsi_dbg.h>
32#include <scsi/scsi_device.h>
33#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050034#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <scsi/scsi_host.h>
36#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include "scsi_priv.h"
39#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070040#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42#define SENSE_TIMEOUT (10*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44/*
45 * These should *probably* be handled by the host itself.
46 * Since it is allowed to sleep, it probably should.
47 */
48#define BUS_RESET_SETTLE_TIME (10)
49#define HOST_RESET_SETTLE_TIME (10)
50
51/* called with shost->host_lock held */
52void scsi_eh_wakeup(struct Scsi_Host *shost)
53{
54 if (shost->host_busy == shost->host_failed) {
James Bottomley3ed7a472005-09-19 09:50:04 -050055 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 SCSI_LOG_ERROR_RECOVERY(5,
57 printk("Waking error handler thread\n"));
58 }
59}
Tejun Heof8bbfc22006-05-19 21:07:05 +090060
61/**
62 * scsi_schedule_eh - schedule EH for SCSI host
63 * @shost: SCSI host to invoke error handling on.
64 *
65 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080066 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090067void scsi_schedule_eh(struct Scsi_Host *shost)
68{
69 unsigned long flags;
70
71 spin_lock_irqsave(shost->host_lock, flags);
72
73 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
74 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
75 shost->host_eh_scheduled++;
76 scsi_eh_wakeup(shost);
77 }
78
79 spin_unlock_irqrestore(shost->host_lock, flags);
80}
81EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83/**
84 * scsi_eh_scmd_add - add scsi cmd to error handling.
85 * @scmd: scmd to run eh on.
86 * @eh_flag: optional SCSI_EH flag.
87 *
88 * Return value:
89 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080090 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070091int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
92{
93 struct Scsi_Host *shost = scmd->device->host;
94 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -050095 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
James Bottomley3ed7a472005-09-19 09:50:04 -050097 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 return 0;
99
100 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500101 if (scsi_host_set_state(shost, SHOST_RECOVERY))
102 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
103 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
James Bottomley939647e2005-09-18 15:05:20 -0500105 ret = 1;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200106 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 shost->host_failed++;
109 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500110 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500112 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113}
114
115/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700117 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 *
119 * Notes:
120 * We do not need to lock this. There is the potential for a race
121 * only in that the normal completion handling might run, but if the
122 * normal completion function determines that the timer has already
123 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800124 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700125enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700127 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700128 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 scsi_log_completion(scmd, TIMEOUT_ERROR);
131
James Smartc829c392006-03-13 08:28:57 -0500132 if (scmd->device->host->transportt->eh_timed_out)
Christoph Hellwigfa990782008-11-05 15:03:15 +0100133 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500134 else if (scmd->device->host->hostt->eh_timed_out)
Christoph Hellwigfa990782008-11-05 15:03:15 +0100135 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500136
Christoph Hellwigfa990782008-11-05 15:03:15 +0100137 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
138 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
James Bottomley939647e2005-09-18 15:05:20 -0500139 scmd->result |= DID_TIME_OUT << 16;
Christoph Hellwigfa990782008-11-05 15:03:15 +0100140 rtn = BLK_EH_HANDLED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700142
Christoph Hellwigfa990782008-11-05 15:03:15 +0100143 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
146/**
147 * scsi_block_when_processing_errors - Prevent cmds from being queued.
148 * @sdev: Device on which we are performing recovery.
149 *
150 * Description:
151 * We block until the host is out of error recovery, and then check to
152 * see whether the host or the device is offline.
153 *
154 * Return value:
155 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800156 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157int scsi_block_when_processing_errors(struct scsi_device *sdev)
158{
159 int online;
160
James Bottomley939647e2005-09-18 15:05:20 -0500161 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 online = scsi_device_online(sdev);
164
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700165 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 online));
167
168 return online;
169}
170EXPORT_SYMBOL(scsi_block_when_processing_errors);
171
172#ifdef CONFIG_SCSI_LOGGING
173/**
174 * scsi_eh_prt_fail_stats - Log info on failures.
175 * @shost: scsi host being recovered.
176 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800177 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
179 struct list_head *work_q)
180{
181 struct scsi_cmnd *scmd;
182 struct scsi_device *sdev;
183 int total_failures = 0;
184 int cmd_failed = 0;
185 int cmd_cancel = 0;
186 int devices_failed = 0;
187
188 shost_for_each_device(sdev, shost) {
189 list_for_each_entry(scmd, work_q, eh_entry) {
190 if (scmd->device == sdev) {
191 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200192 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 ++cmd_cancel;
194 else
195 ++cmd_failed;
196 }
197 }
198
199 if (cmd_cancel || cmd_failed) {
200 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500201 sdev_printk(KERN_INFO, sdev,
202 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700203 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500204 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 cmd_cancel = 0;
206 cmd_failed = 0;
207 ++devices_failed;
208 }
209 }
210
211 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
212 " devices require eh work\n",
213 total_failures, devices_failed));
214}
215#endif
216
217/**
218 * scsi_check_sense - Examine scsi cmd sense
219 * @scmd: Cmd to have sense checked.
220 *
221 * Return value:
222 * SUCCESS or FAILED or NEEDS_RETRY
223 *
224 * Notes:
225 * When a deferred error is detected the current command has
226 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800227 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228static int scsi_check_sense(struct scsi_cmnd *scmd)
229{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700230 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 struct scsi_sense_hdr sshdr;
232
233 if (! scsi_command_normalize_sense(scmd, &sshdr))
234 return FAILED; /* no valid sense data */
235
236 if (scsi_sense_is_deferred(&sshdr))
237 return NEEDS_RETRY;
238
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700239 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
240 sdev->scsi_dh_data->scsi_dh->check_sense) {
241 int rc;
242
243 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
244 if (rc != SCSI_RETURN_NOT_HANDLED)
245 return rc;
246 /* handler does not care. Drop down to default handling */
247 }
248
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 /*
250 * Previous logic looked for FILEMARK, EOM or ILI which are
251 * mainly associated with tapes and returned SUCCESS.
252 */
253 if (sshdr.response_code == 0x70) {
254 /* fixed format */
255 if (scmd->sense_buffer[2] & 0xe0)
256 return SUCCESS;
257 } else {
258 /*
259 * descriptor format: look for "stream commands sense data
260 * descriptor" (see SSC-3). Assume single sense data
261 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
262 */
263 if ((sshdr.additional_length > 3) &&
264 (scmd->sense_buffer[8] == 0x4) &&
265 (scmd->sense_buffer[11] & 0xe0))
266 return SUCCESS;
267 }
268
269 switch (sshdr.sense_key) {
270 case NO_SENSE:
271 return SUCCESS;
272 case RECOVERED_ERROR:
273 return /* soft_error */ SUCCESS;
274
275 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400276 if (sshdr.asc == 0x10) /* DIF */
277 return SUCCESS;
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 return NEEDS_RETRY;
280 case NOT_READY:
281 case UNIT_ATTENTION:
282 /*
283 * if we are expecting a cc/ua because of a bus reset that we
284 * performed, treat this just as a retry. otherwise this is
285 * information that we should pass up to the upper-level driver
286 * so that we can deal with it there.
287 */
288 if (scmd->device->expecting_cc_ua) {
289 scmd->device->expecting_cc_ua = 0;
290 return NEEDS_RETRY;
291 }
292 /*
293 * if the device is in the process of becoming ready, we
294 * should retry.
295 */
296 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
297 return NEEDS_RETRY;
298 /*
299 * if the device is not started, we need to wake
300 * the error handler to start the motor
301 */
302 if (scmd->device->allow_restart &&
303 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
304 return FAILED;
305 return SUCCESS;
306
307 /* these three are not supported */
308 case COPY_ABORTED:
309 case VOLUME_OVERFLOW:
310 case MISCOMPARE:
311 return SUCCESS;
312
313 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800314 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
315 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
316 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
317 return SUCCESS;
318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 return NEEDS_RETRY;
320
321 case HARDWARE_ERROR:
322 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700323 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 else
325 return SUCCESS;
326
327 case ILLEGAL_REQUEST:
328 case BLANK_CHECK:
329 case DATA_PROTECT:
330 default:
331 return SUCCESS;
332 }
333}
334
Vasu Dev4a840672009-10-22 15:46:33 -0700335static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
336{
337 struct scsi_host_template *sht = sdev->host->hostt;
338 struct scsi_device *tmp_sdev;
339
340 if (!sht->change_queue_depth ||
341 sdev->queue_depth >= sdev->max_queue_depth)
342 return;
343
344 if (time_before(jiffies,
345 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
346 return;
347
348 if (time_before(jiffies,
349 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
350 return;
351
352 /*
353 * Walk all devices of a target and do
354 * ramp up on them.
355 */
356 shost_for_each_device(tmp_sdev, sdev->host) {
357 if (tmp_sdev->channel != sdev->channel ||
358 tmp_sdev->id != sdev->id ||
359 tmp_sdev->queue_depth == sdev->max_queue_depth)
360 continue;
361 /*
362 * call back into LLD to increase queue_depth by one
363 * with ramp up reason code.
364 */
365 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
366 SCSI_QDEPTH_RAMP_UP);
367 sdev->last_queue_ramp_up = jiffies;
368 }
369}
370
Mike Christie42a6a912009-10-15 17:46:44 -0700371static void scsi_handle_queue_full(struct scsi_device *sdev)
372{
373 struct scsi_host_template *sht = sdev->host->hostt;
374 struct scsi_device *tmp_sdev;
375
376 if (!sht->change_queue_depth)
377 return;
378
379 shost_for_each_device(tmp_sdev, sdev->host) {
380 if (tmp_sdev->channel != sdev->channel ||
381 tmp_sdev->id != sdev->id)
382 continue;
383 /*
384 * We do not know the number of commands that were at
385 * the device when we got the queue full so we start
386 * from the highest possible value and work our way down.
387 */
388 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
389 SCSI_QDEPTH_QFULL);
390 }
391}
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393/**
394 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
395 * @scmd: SCSI cmd to examine.
396 *
397 * Notes:
398 * This is *only* called when we are examining the status of commands
399 * queued during error recovery. the main difference here is that we
400 * don't allow for the possibility of retries here, and we are a lot
401 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800402 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
404{
405 /*
406 * first check the host byte, to see if there is anything in there
407 * that would indicate what we need to do.
408 */
409 if (host_byte(scmd->result) == DID_RESET) {
410 /*
411 * rats. we are already in the error handler, so we now
412 * get to try and figure out what to do next. if the sense
413 * is valid, we have a pretty good idea of what to do.
414 * if not, we mark it as FAILED.
415 */
416 return scsi_check_sense(scmd);
417 }
418 if (host_byte(scmd->result) != DID_OK)
419 return FAILED;
420
421 /*
422 * next, check the message byte.
423 */
424 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
425 return FAILED;
426
427 /*
428 * now, check the status byte to see if this indicates
429 * anything special.
430 */
431 switch (status_byte(scmd->result)) {
432 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700433 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 case COMMAND_TERMINATED:
435 return SUCCESS;
436 case CHECK_CONDITION:
437 return scsi_check_sense(scmd);
438 case CONDITION_GOOD:
439 case INTERMEDIATE_GOOD:
440 case INTERMEDIATE_C_GOOD:
441 /*
442 * who knows? FIXME(eric)
443 */
444 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500445 case RESERVATION_CONFLICT:
446 /*
447 * let issuer deal with this, it could be just fine
448 */
449 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700451 scsi_handle_queue_full(scmd->device);
452 /* fall through */
453 case BUSY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 default:
455 return FAILED;
456 }
457 return FAILED;
458}
459
460/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 * scsi_eh_done - Completion function for error handling.
462 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800463 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464static void scsi_eh_done(struct scsi_cmnd *scmd)
465{
Michael Reed85631672005-12-07 21:46:27 -0600466 struct completion *eh_action;
467
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100468 SCSI_LOG_ERROR_RECOVERY(3,
469 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700470 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600471
472 eh_action = scmd->device->host->eh_action;
473 if (eh_action)
474 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477/**
Brian King292148f2007-01-30 17:51:17 -0600478 * scsi_try_host_reset - ask host adapter to reset itself
479 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800480 */
Brian King292148f2007-01-30 17:51:17 -0600481static int scsi_try_host_reset(struct scsi_cmnd *scmd)
482{
483 unsigned long flags;
484 int rtn;
485
486 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700487 __func__));
Brian King292148f2007-01-30 17:51:17 -0600488
489 if (!scmd->device->host->hostt->eh_host_reset_handler)
490 return FAILED;
491
492 rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
493
494 if (rtn == SUCCESS) {
495 if (!scmd->device->host->hostt->skip_settle_delay)
496 ssleep(HOST_RESET_SETTLE_TIME);
497 spin_lock_irqsave(scmd->device->host->host_lock, flags);
498 scsi_report_bus_reset(scmd->device->host,
499 scmd_channel(scmd));
500 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
501 }
502
503 return rtn;
504}
505
506/**
507 * scsi_try_bus_reset - ask host to perform a bus reset
508 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800509 */
Brian King292148f2007-01-30 17:51:17 -0600510static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
511{
512 unsigned long flags;
513 int rtn;
514
515 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700516 __func__));
Brian King292148f2007-01-30 17:51:17 -0600517
518 if (!scmd->device->host->hostt->eh_bus_reset_handler)
519 return FAILED;
520
521 rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
522
523 if (rtn == SUCCESS) {
524 if (!scmd->device->host->hostt->skip_settle_delay)
525 ssleep(BUS_RESET_SETTLE_TIME);
526 spin_lock_irqsave(scmd->device->host->host_lock, flags);
527 scsi_report_bus_reset(scmd->device->host,
528 scmd_channel(scmd));
529 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
530 }
531
532 return rtn;
533}
534
Mike Christie30bd7df2008-02-29 18:25:19 -0600535static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
536{
537 sdev->was_reset = 1;
538 sdev->expecting_cc_ua = 1;
539}
540
541/**
542 * scsi_try_target_reset - Ask host to perform a target reset
543 * @scmd: SCSI cmd used to send a target reset
544 *
545 * Notes:
546 * There is no timeout for this operation. if this operation is
547 * unreliable for a given host, then the host itself needs to put a
548 * timer on it, and set the host back to a consistent state prior to
549 * returning.
550 */
551static int scsi_try_target_reset(struct scsi_cmnd *scmd)
552{
553 unsigned long flags;
554 int rtn;
555
556 if (!scmd->device->host->hostt->eh_target_reset_handler)
557 return FAILED;
558
559 rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
560 if (rtn == SUCCESS) {
561 spin_lock_irqsave(scmd->device->host->host_lock, flags);
562 __starget_for_each_device(scsi_target(scmd->device), NULL,
563 __scsi_report_device_reset);
564 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
565 }
566
567 return rtn;
568}
569
Brian King292148f2007-01-30 17:51:17 -0600570/**
571 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
572 * @scmd: SCSI cmd used to send BDR
573 *
574 * Notes:
575 * There is no timeout for this operation. if this operation is
576 * unreliable for a given host, then the host itself needs to put a
577 * timer on it, and set the host back to a consistent state prior to
578 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800579 */
Brian King292148f2007-01-30 17:51:17 -0600580static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
581{
582 int rtn;
583
584 if (!scmd->device->host->hostt->eh_device_reset_handler)
585 return FAILED;
586
587 rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600588 if (rtn == SUCCESS)
589 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600590 return rtn;
591}
592
593static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
594{
595 if (!scmd->device->host->hostt->eh_abort_handler)
596 return FAILED;
597
598 return scmd->device->host->hostt->eh_abort_handler(scmd);
599}
600
601/**
602 * scsi_try_to_abort_cmd - Ask host to abort a running command.
603 * @scmd: SCSI cmd to abort from Lower Level.
604 *
605 * Notes:
606 * This function will not return until the user's completion function
607 * has been called. there is no timeout on this operation. if the
608 * author of the low-level driver wishes this operation to be timed,
609 * they can provide this facility themselves. helper functions in
610 * scsi_error.c can be supplied to make this easier to do.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800611 */
Brian King292148f2007-01-30 17:51:17 -0600612static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
613{
614 /*
615 * scsi_done was called just after the command timed out and before
616 * we had a chance to process it. (db)
617 */
618 if (scmd->serial_number == 0)
619 return SUCCESS;
620 return __scsi_try_to_abort_cmd(scmd);
621}
622
623static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
624{
625 if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
626 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600627 if (scsi_try_target_reset(scmd) != SUCCESS)
628 if (scsi_try_bus_reset(scmd) != SUCCESS)
629 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600630}
631
632/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200633 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100634 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200635 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200636 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300637 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200638 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100639 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200640 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200641 * as part of the error recovery process. If @sense_bytes is 0 the command
642 * sent must be one that does not transfer any data. If @sense_bytes != 0
643 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
644 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800645 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200646void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
647 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
f59114b2005-04-17 15:00:23 -0500649 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Christoph Hellwig631c2282006-07-08 20:42:15 +0200651 /*
652 * We need saved copies of a number of fields - this is because
653 * error handling may need to overwrite these with different values
654 * to run different commands, and once error handling is complete,
655 * we will need to restore these values prior to running the actual
656 * command.
657 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200658 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300659 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200660 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200661 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200662 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200663 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400664 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400665 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200666
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400667 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300668 scmd->cmnd = ses->eh_cmnd;
669 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200670 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200671 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200672
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200673 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200674 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
675 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200676 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200677 scmd->sdb.length);
678 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200679 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200680 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200681 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200682 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200683 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200684 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200685 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200686 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300687 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200688 memcpy(scmd->cmnd, cmnd, cmnd_size);
689 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
690 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200691 }
692
693 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200694
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200695 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500697 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Christoph Hellwig631c2282006-07-08 20:42:15 +0200699 /*
700 * Zero the sense buffer. The scsi spec mandates that any
701 * untransferred sense data should be interpreted as being zero.
702 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900703 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200704}
705EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200706
Boaz Harroshe1c23462007-10-08 16:36:45 +0200707/**
708 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
709 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200710 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200711 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200712 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800713 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200714void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
715{
716 /*
717 * Restore original data
718 */
719 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300720 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200721 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200722 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200723 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200724 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400725 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400726 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200727}
728EXPORT_SYMBOL(scsi_eh_restore_cmnd);
729
730/**
731 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
732 * @scmd: SCSI command structure to hijack
733 * @cmnd: CDB to send
734 * @cmnd_size: size in bytes of @cmnd
735 * @timeout: timeout for this request
736 * @sense_bytes: size of sense data to copy or 0
737 *
738 * This function is used to send a scsi command down to a target device
739 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
740 *
741 * Return value:
742 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800743 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200744static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
745 int cmnd_size, int timeout, unsigned sense_bytes)
746{
747 struct scsi_device *sdev = scmd->device;
748 struct Scsi_Host *shost = sdev->host;
749 DECLARE_COMPLETION_ONSTACK(done);
750 unsigned long timeleft;
751 unsigned long flags;
752 struct scsi_eh_save ses;
753 int rtn;
754
755 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100756 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
f59114b2005-04-17 15:00:23 -0500758 spin_lock_irqsave(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 scsi_log_send(scmd);
f59114b2005-04-17 15:00:23 -0500760 shost->hostt->queuecommand(scmd, scsi_eh_done);
761 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100763 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
f59114b2005-04-17 15:00:23 -0500765 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100767 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100769 SCSI_LOG_ERROR_RECOVERY(3,
770 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700771 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
773 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100774 * If there is time left scsi_eh_done got called, and we will
775 * examine the actual status codes to see whether the command
776 * actually did complete normally, else tell the host to forget
777 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100779 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 rtn = scsi_eh_completed_normally(scmd);
781 SCSI_LOG_ERROR_RECOVERY(3,
782 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700783 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100784
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 switch (rtn) {
786 case SUCCESS:
787 case NEEDS_RETRY:
788 case FAILED:
789 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +0200790 case ADD_TO_MLQUEUE:
791 rtn = NEEDS_RETRY;
792 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 default:
794 rtn = FAILED;
795 break;
796 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100797 } else {
Brian King292148f2007-01-30 17:51:17 -0600798 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100799 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 }
801
Boaz Harroshe1c23462007-10-08 16:36:45 +0200802 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return rtn;
804}
805
806/**
807 * scsi_request_sense - Request sense data from a particular target.
808 * @scmd: SCSI cmd for request sense.
809 *
810 * Notes:
811 * Some hosts automatically obtain this information, others require
812 * that we obtain it on our own. This function will *not* return until
813 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800814 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815static int scsi_request_sense(struct scsi_cmnd *scmd)
816{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200817 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818}
819
820/**
821 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
822 * @scmd: Original SCSI cmd that eh has finished.
823 * @done_q: Queue for processed commands.
824 *
825 * Notes:
826 * We don't want to use the normal command completion while we are are
827 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500828 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 * keep a list of pending commands for final completion, and once we
830 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800831 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900832void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200835 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 list_move_tail(&scmd->eh_entry, done_q);
837}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900838EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
840/**
841 * scsi_eh_get_sense - Get device sense data.
842 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500843 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 *
845 * Description:
846 * See if we need to request sense information. if so, then get it
847 * now, so we have a better idea of what to do.
848 *
849 * Notes:
850 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500851 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 * it down before we request it.
853 *
854 * All drivers should request sense information internally these days,
855 * so for now all I have to say is tough noogies if you end up in here.
856 *
857 * XXX: Long term this code should go away, but that needs an audit of
858 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800859 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800860int scsi_eh_get_sense(struct list_head *work_q,
861 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200863 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 int rtn;
865
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200866 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200867 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 SCSI_SENSE_VALID(scmd))
869 continue;
870
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400871 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
872 "%s: requesting sense\n",
873 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 rtn = scsi_request_sense(scmd);
875 if (rtn != SUCCESS)
876 continue;
877
878 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
879 " result %x\n", scmd,
880 scmd->result));
881 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
882
883 rtn = scsi_decide_disposition(scmd);
884
885 /*
886 * if the result was normal, then just pass it along to the
887 * upper level.
888 */
889 if (rtn == SUCCESS)
890 /* we don't want this command reissued, just
891 * finished with the sense data, so set
892 * retries to the max allowed to ensure it
893 * won't get reissued */
894 scmd->retries = scmd->allowed;
895 else if (rtn != NEEDS_RETRY)
896 continue;
897
898 scsi_eh_finish_cmd(scmd, done_q);
899 }
900
901 return list_empty(work_q);
902}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800903EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500907 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 *
909 * Return value:
910 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800911 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912static int scsi_eh_tur(struct scsi_cmnd *scmd)
913{
914 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
915 int retry_cnt = 1, rtn;
916
917retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400918 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700921 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200922
923 switch (rtn) {
924 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 if (retry_cnt--)
926 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200927 /*FALLTHRU*/
928 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500929 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200930 default:
931 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500932 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}
934
935/**
Rob Landleyeb448202007-11-03 13:30:39 -0500936 * scsi_eh_abort_cmds - abort pending commands.
937 * @work_q: &list_head for pending commands.
938 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 *
940 * Decription:
941 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -0500942 * running command. This only works out to be the case if we have one
943 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 * no sense to try and abort the command, since as far as the shost
945 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800946 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947static int scsi_eh_abort_cmds(struct list_head *work_q,
948 struct list_head *done_q)
949{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200950 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 int rtn;
952
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200953 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200954 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 continue;
956 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
957 "0x%p\n", current->comm,
958 scmd));
959 rtn = scsi_try_to_abort_cmd(scmd);
960 if (rtn == SUCCESS) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200961 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 if (!scsi_device_online(scmd->device) ||
963 !scsi_eh_tur(scmd)) {
964 scsi_eh_finish_cmd(scmd, done_q);
965 }
966
967 } else
968 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
969 " cmd failed:"
970 "0x%p\n",
971 current->comm,
972 scmd));
973 }
974
975 return list_empty(work_q);
976}
977
978/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500980 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 *
982 * Return value:
983 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800984 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
986{
987 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Christoph Hellwig631c2282006-07-08 20:42:15 +0200989 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -0500990 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Brian Kinged773e62007-03-29 15:25:52 -0500992 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -0600993 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -0500994
Christoph Hellwig631c2282006-07-08 20:42:15 +0200995 if (rtn == SUCCESS)
996 return 0;
997 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 return 1;
1000}
1001
1002 /**
1003 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001004 * @shost: &scsi host being recovered.
1005 * @work_q: &list_head for pending commands.
1006 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 *
1008 * Notes:
1009 * If commands are failing due to not ready, initializing command required,
1010 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001011 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012static int scsi_eh_stu(struct Scsi_Host *shost,
1013 struct list_head *work_q,
1014 struct list_head *done_q)
1015{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001016 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 struct scsi_device *sdev;
1018
1019 shost_for_each_device(sdev, shost) {
1020 stu_scmd = NULL;
1021 list_for_each_entry(scmd, work_q, eh_entry)
1022 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1023 scsi_check_sense(scmd) == FAILED ) {
1024 stu_scmd = scmd;
1025 break;
1026 }
1027
1028 if (!stu_scmd)
1029 continue;
1030
1031 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1032 " 0x%p\n", current->comm, sdev));
1033
1034 if (!scsi_eh_try_stu(stu_scmd)) {
1035 if (!scsi_device_online(sdev) ||
1036 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001037 list_for_each_entry_safe(scmd, next,
1038 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 if (scmd->device == sdev)
1040 scsi_eh_finish_cmd(scmd, done_q);
1041 }
1042 }
1043 } else {
1044 SCSI_LOG_ERROR_RECOVERY(3,
1045 printk("%s: START_UNIT failed to sdev:"
1046 " 0x%p\n", current->comm, sdev));
1047 }
1048 }
1049
1050 return list_empty(work_q);
1051}
1052
1053
1054/**
1055 * scsi_eh_bus_device_reset - send bdr if needed
1056 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001057 * @work_q: &list_head for pending commands.
1058 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 *
1060 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001061 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 * devices that are jammed or not - if we have multiple devices, it
1063 * makes no sense to try bus_device_reset - we really would need to try
1064 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001065 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1067 struct list_head *work_q,
1068 struct list_head *done_q)
1069{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001070 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 struct scsi_device *sdev;
1072 int rtn;
1073
1074 shost_for_each_device(sdev, shost) {
1075 bdr_scmd = NULL;
1076 list_for_each_entry(scmd, work_q, eh_entry)
1077 if (scmd->device == sdev) {
1078 bdr_scmd = scmd;
1079 break;
1080 }
1081
1082 if (!bdr_scmd)
1083 continue;
1084
1085 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1086 " 0x%p\n", current->comm,
1087 sdev));
1088 rtn = scsi_try_bus_device_reset(bdr_scmd);
1089 if (rtn == SUCCESS) {
1090 if (!scsi_device_online(sdev) ||
1091 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001092 list_for_each_entry_safe(scmd, next,
1093 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (scmd->device == sdev)
1095 scsi_eh_finish_cmd(scmd,
1096 done_q);
1097 }
1098 }
1099 } else {
1100 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1101 " failed sdev:"
1102 "0x%p\n",
1103 current->comm,
1104 sdev));
1105 }
1106 }
1107
1108 return list_empty(work_q);
1109}
1110
1111/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001112 * scsi_eh_target_reset - send target reset if needed
1113 * @shost: scsi host being recovered.
1114 * @work_q: &list_head for pending commands.
1115 * @done_q: &list_head for processed commands.
1116 *
1117 * Notes:
1118 * Try a target reset.
1119 */
1120static int scsi_eh_target_reset(struct Scsi_Host *shost,
1121 struct list_head *work_q,
1122 struct list_head *done_q)
1123{
1124 struct scsi_cmnd *scmd, *tgtr_scmd, *next;
James Bottomleyc82dc882008-09-12 16:46:51 -05001125 unsigned int id = 0;
Mike Christie30bd7df2008-02-29 18:25:19 -06001126 int rtn;
1127
James Bottomleyc82dc882008-09-12 16:46:51 -05001128 do {
Mike Christie30bd7df2008-02-29 18:25:19 -06001129 tgtr_scmd = NULL;
1130 list_for_each_entry(scmd, work_q, eh_entry) {
1131 if (id == scmd_id(scmd)) {
1132 tgtr_scmd = scmd;
1133 break;
1134 }
1135 }
James Bottomleyc82dc882008-09-12 16:46:51 -05001136 if (!tgtr_scmd) {
1137 /* not one exactly equal; find the next highest */
1138 list_for_each_entry(scmd, work_q, eh_entry) {
1139 if (scmd_id(scmd) > id &&
1140 (!tgtr_scmd ||
1141 scmd_id(tgtr_scmd) > scmd_id(scmd)))
1142 tgtr_scmd = scmd;
1143 }
1144 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001145 if (!tgtr_scmd)
James Bottomleyc82dc882008-09-12 16:46:51 -05001146 /* no more commands, that's it */
1147 break;
Mike Christie30bd7df2008-02-29 18:25:19 -06001148
1149 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1150 "to target %d\n",
1151 current->comm, id));
1152 rtn = scsi_try_target_reset(tgtr_scmd);
1153 if (rtn == SUCCESS) {
1154 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1155 if (id == scmd_id(scmd))
1156 if (!scsi_device_online(scmd->device) ||
1157 !scsi_eh_tur(tgtr_scmd))
1158 scsi_eh_finish_cmd(scmd,
1159 done_q);
1160 }
1161 } else
1162 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1163 " failed target: "
1164 "%d\n",
1165 current->comm, id));
James Bottomleyc82dc882008-09-12 16:46:51 -05001166 id++;
1167 } while(id != 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001168
1169 return list_empty(work_q);
1170}
1171
1172/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001174 * @shost: &scsi host being recovered.
1175 * @work_q: &list_head for pending commands.
1176 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001177 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1179 struct list_head *work_q,
1180 struct list_head *done_q)
1181{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001182 struct scsi_cmnd *scmd, *chan_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 unsigned int channel;
1184 int rtn;
1185
1186 /*
1187 * we really want to loop over the various channels, and do this on
1188 * a channel by channel basis. we should also check to see if any
1189 * of the failed commands are on soft_reset devices, and if so, skip
1190 * the reset.
1191 */
1192
1193 for (channel = 0; channel <= shost->max_channel; channel++) {
1194 chan_scmd = NULL;
1195 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001196 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 chan_scmd = scmd;
1198 break;
1199 /*
1200 * FIXME add back in some support for
1201 * soft_reset devices.
1202 */
1203 }
1204 }
1205
1206 if (!chan_scmd)
1207 continue;
1208 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1209 " %d\n", current->comm,
1210 channel));
1211 rtn = scsi_try_bus_reset(chan_scmd);
1212 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001213 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001214 if (channel == scmd_channel(scmd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (!scsi_device_online(scmd->device) ||
1216 !scsi_eh_tur(scmd))
1217 scsi_eh_finish_cmd(scmd,
1218 done_q);
1219 }
1220 } else {
1221 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1222 " failed chan: %d\n",
1223 current->comm,
1224 channel));
1225 }
1226 }
1227 return list_empty(work_q);
1228}
1229
1230/**
1231 * scsi_eh_host_reset - send a host reset
1232 * @work_q: list_head for processed commands.
1233 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001234 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235static int scsi_eh_host_reset(struct list_head *work_q,
1236 struct list_head *done_q)
1237{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001238 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
1241 if (!list_empty(work_q)) {
1242 scmd = list_entry(work_q->next,
1243 struct scsi_cmnd, eh_entry);
1244
1245 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1246 , current->comm));
1247
1248 rtn = scsi_try_host_reset(scmd);
1249 if (rtn == SUCCESS) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001250 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 if (!scsi_device_online(scmd->device) ||
1252 (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1253 !scsi_eh_tur(scmd))
1254 scsi_eh_finish_cmd(scmd, done_q);
1255 }
1256 } else {
1257 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1258 " failed\n",
1259 current->comm));
1260 }
1261 }
1262 return list_empty(work_q);
1263}
1264
1265/**
1266 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1267 * @work_q: list_head for processed commands.
1268 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001269 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270static void scsi_eh_offline_sdevs(struct list_head *work_q,
1271 struct list_head *done_q)
1272{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001273 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001275 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001276 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1277 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001279 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 /*
1281 * FIXME: Handle lost cmds.
1282 */
1283 }
1284 scsi_eh_finish_cmd(scmd, done_q);
1285 }
1286 return;
1287}
1288
1289/**
Mike Christie4a274462008-08-19 18:45:31 -05001290 * scsi_noretry_cmd - determinte if command should be failed fast
1291 * @scmd: SCSI cmd to examine.
1292 */
1293int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1294{
1295 switch (host_byte(scmd->result)) {
1296 case DID_OK:
1297 break;
1298 case DID_BUS_BUSY:
1299 return blk_failfast_transport(scmd->request);
1300 case DID_PARITY:
1301 return blk_failfast_dev(scmd->request);
1302 case DID_ERROR:
1303 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1304 status_byte(scmd->result) == RESERVATION_CONFLICT)
1305 return 0;
1306 /* fall through */
1307 case DID_SOFT_ERROR:
1308 return blk_failfast_driver(scmd->request);
1309 }
1310
1311 switch (status_byte(scmd->result)) {
1312 case CHECK_CONDITION:
1313 /*
1314 * assume caller has checked sense and determinted
1315 * the check condition was retryable.
1316 */
1317 return blk_failfast_dev(scmd->request);
1318 }
1319
1320 return 0;
1321}
1322
1323/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1325 * @scmd: SCSI cmd to examine.
1326 *
1327 * Notes:
1328 * This is *only* called when we are examining the status after sending
1329 * out the actual data command. any commands that are queued for error
1330 * recovery (e.g. test_unit_ready) do *not* come through here.
1331 *
1332 * When this routine returns failed, it means the error handler thread
1333 * is woken. In cases where the error code indicates an error that
1334 * doesn't require the error handler read (i.e. we don't need to
1335 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001336 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337int scsi_decide_disposition(struct scsi_cmnd *scmd)
1338{
1339 int rtn;
1340
1341 /*
1342 * if the device is offline, then we clearly just pass the result back
1343 * up to the top level.
1344 */
1345 if (!scsi_device_online(scmd->device)) {
1346 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1347 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001348 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 return SUCCESS;
1350 }
1351
1352 /*
1353 * first check the host byte, to see if there is anything in there
1354 * that would indicate what we need to do.
1355 */
1356 switch (host_byte(scmd->result)) {
1357 case DID_PASSTHROUGH:
1358 /*
1359 * no matter what, pass this through to the upper layer.
1360 * nuke this special code so that it looks like we are saying
1361 * did_ok.
1362 */
1363 scmd->result &= 0xff00ffff;
1364 return SUCCESS;
1365 case DID_OK:
1366 /*
1367 * looks good. drop through, and check the next byte.
1368 */
1369 break;
1370 case DID_NO_CONNECT:
1371 case DID_BAD_TARGET:
1372 case DID_ABORT:
1373 /*
1374 * note - this means that we just report the status back
1375 * to the top level driver, not that we actually think
1376 * that it indicates SUCCESS.
1377 */
1378 return SUCCESS;
1379 /*
1380 * when the low level driver returns did_soft_error,
1381 * it is responsible for keeping an internal retry counter
1382 * in order to avoid endless loops (db)
1383 *
1384 * actually this is a bug in this function here. we should
1385 * be mindful of the maximum number of retries specified
1386 * and not get stuck in a loop.
1387 */
1388 case DID_SOFT_ERROR:
1389 goto maybe_retry;
1390 case DID_IMM_RETRY:
1391 return NEEDS_RETRY;
1392
bf341912005-04-12 17:49:09 -05001393 case DID_REQUEUE:
1394 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001395 case DID_TRANSPORT_DISRUPTED:
1396 /*
1397 * LLD/transport was disrupted during processing of the IO.
1398 * The transport class is now blocked/blocking,
1399 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001400 * based on its timers and recovery capablilities if
1401 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001402 */
Mike Christie939c2282008-11-04 19:47:19 -06001403 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001404 case DID_TRANSPORT_FAILFAST:
1405 /*
1406 * The transport decided to failfast the IO (most likely
1407 * the fast io fail tmo fired), so send IO directly upwards.
1408 */
1409 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 case DID_ERROR:
1411 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1412 status_byte(scmd->result) == RESERVATION_CONFLICT)
1413 /*
1414 * execute reservation conflict processing code
1415 * lower down
1416 */
1417 break;
1418 /* fallthrough */
1419
1420 case DID_BUS_BUSY:
1421 case DID_PARITY:
1422 goto maybe_retry;
1423 case DID_TIME_OUT:
1424 /*
1425 * when we scan the bus, we get timeout messages for
1426 * these commands if there is no device available.
1427 * other hosts report did_no_connect for the same thing.
1428 */
1429 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1430 scmd->cmnd[0] == INQUIRY)) {
1431 return SUCCESS;
1432 } else {
1433 return FAILED;
1434 }
1435 case DID_RESET:
1436 return SUCCESS;
1437 default:
1438 return FAILED;
1439 }
1440
1441 /*
1442 * next, check the message byte.
1443 */
1444 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1445 return FAILED;
1446
1447 /*
1448 * check the status byte to see if this indicates anything special.
1449 */
1450 switch (status_byte(scmd->result)) {
1451 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001452 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 /*
1454 * the case of trying to send too many commands to a
1455 * tagged queueing device.
1456 */
1457 case BUSY:
1458 /*
1459 * device can't talk to us at the moment. Should only
1460 * occur (SAM-3) when the task queue is empty, so will cause
1461 * the empty queue handling to trigger a stall in the
1462 * device.
1463 */
1464 return ADD_TO_MLQUEUE;
1465 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -07001466 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001469 case TASK_ABORTED:
1470 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 case CHECK_CONDITION:
1472 rtn = scsi_check_sense(scmd);
1473 if (rtn == NEEDS_RETRY)
1474 goto maybe_retry;
1475 /* if rtn == FAILED, we have no sense information;
1476 * returning FAILED will wake the error handler thread
1477 * to collect the sense and redo the decide
1478 * disposition */
1479 return rtn;
1480 case CONDITION_GOOD:
1481 case INTERMEDIATE_GOOD:
1482 case INTERMEDIATE_C_GOOD:
1483 case ACA_ACTIVE:
1484 /*
1485 * who knows? FIXME(eric)
1486 */
1487 return SUCCESS;
1488
1489 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001490 sdev_printk(KERN_INFO, scmd->device,
1491 "reservation conflict\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 return SUCCESS; /* causes immediate i/o error */
1493 default:
1494 return FAILED;
1495 }
1496 return FAILED;
1497
1498 maybe_retry:
1499
1500 /* we requeue for retry because the error was retryable, and
1501 * the request was not marked fast fail. Note that above,
1502 * even if the request is marked fast fail, we still requeue
1503 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001504 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001505 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 return NEEDS_RETRY;
1507 } else {
1508 /*
1509 * no more retries - report this one back to upper level.
1510 */
1511 return SUCCESS;
1512 }
1513}
1514
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001515static void eh_lock_door_done(struct request *req, int uptodate)
1516{
1517 __blk_put_request(req->q, req);
1518}
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 * scsi_eh_lock_door - Prevent medium removal for the specified device
1522 * @sdev: SCSI device to prevent medium removal
1523 *
1524 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001525 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 *
1527 * Notes:
1528 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1529 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001530 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531static void scsi_eh_lock_door(struct scsi_device *sdev)
1532{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001533 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
James Bottomley91bc31f2009-05-17 09:30:48 -05001535 /*
1536 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1537 * request becomes available
1538 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001539 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001541 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1542 req->cmd[1] = 0;
1543 req->cmd[2] = 0;
1544 req->cmd[3] = 0;
1545 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1546 req->cmd[5] = 0;
1547
1548 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1549
1550 req->cmd_type = REQ_TYPE_BLOCK_PC;
1551 req->cmd_flags |= REQ_QUIET;
1552 req->timeout = 10 * HZ;
1553 req->retries = 5;
1554
1555 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556}
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558/**
1559 * scsi_restart_operations - restart io operations to the specified host.
1560 * @shost: Host we are restarting.
1561 *
1562 * Notes:
1563 * When we entered the error handler, we blocked all further i/o to
1564 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001565 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566static void scsi_restart_operations(struct Scsi_Host *shost)
1567{
1568 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001569 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
1571 /*
1572 * If the door was locked, we need to insert a door lock request
1573 * onto the head of the SCSI request queue for the device. There
1574 * is no point trying to lock the door of an off-line device.
1575 */
1576 shost_for_each_device(sdev, shost) {
1577 if (scsi_device_online(sdev) && sdev->locked)
1578 scsi_eh_lock_door(sdev);
1579 }
1580
1581 /*
1582 * next free up anything directly waiting upon the host. this
1583 * will be requests for character device operations, and also for
1584 * ioctls to queued block devices.
1585 */
1586 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001587 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
James Bottomley939647e2005-09-18 15:05:20 -05001589 spin_lock_irqsave(shost->host_lock, flags);
1590 if (scsi_host_set_state(shost, SHOST_RUNNING))
1591 if (scsi_host_set_state(shost, SHOST_CANCEL))
1592 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1593 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595 wake_up(&shost->host_wait);
1596
1597 /*
1598 * finally we need to re-initiate requests that may be pending. we will
1599 * have had everything blocked while error handling is taking place, and
1600 * now that error recovery is done, we will need to ensure that these
1601 * requests are started.
1602 */
1603 scsi_run_host_queues(shost);
1604}
1605
1606/**
1607 * scsi_eh_ready_devs - check device ready state and recover if not.
1608 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001609 * @work_q: &list_head for pending commands.
1610 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001611 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001612void scsi_eh_ready_devs(struct Scsi_Host *shost,
1613 struct list_head *work_q,
1614 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615{
1616 if (!scsi_eh_stu(shost, work_q, done_q))
1617 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001618 if (!scsi_eh_target_reset(shost, work_q, done_q))
1619 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1620 if (!scsi_eh_host_reset(work_q, done_q))
1621 scsi_eh_offline_sdevs(work_q,
1622 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001624EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625
1626/**
1627 * scsi_eh_flush_done_q - finish processed commands or retry them.
1628 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001629 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001630void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001632 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001634 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1635 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001637 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001638 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1640 " retry cmd: %p\n",
1641 current->comm,
1642 scmd));
1643 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1644 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001645 /*
1646 * If just we got sense for the device (called
1647 * scsi_eh_get_sense), scmd->result is already
1648 * set, do not set DRIVER_TIMEOUT.
1649 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 if (!scmd->result)
1651 scmd->result |= (DRIVER_TIMEOUT << 24);
1652 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1653 " cmd: %p\n",
1654 current->comm, scmd));
1655 scsi_finish_command(scmd);
1656 }
1657 }
1658}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001659EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661/**
1662 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1663 * @shost: Host to unjam.
1664 *
1665 * Notes:
1666 * When we come in here, we *know* that all commands on the bus have
1667 * either completed, failed or timed out. we also know that no further
1668 * commands are being sent to the host, so things are relatively quiet
1669 * and we have freedom to fiddle with things as we wish.
1670 *
1671 * This is only the *default* implementation. it is possible for
1672 * individual drivers to supply their own version of this function, and
1673 * if the maintainer wishes to do this, it is strongly suggested that
1674 * this function be taken as a template and modified. this function
1675 * was designed to correctly handle problems for about 95% of the
1676 * different cases out there, and it should always provide at least a
1677 * reasonable amount of error recovery.
1678 *
1679 * Any command marked 'failed' or 'timeout' must eventually have
1680 * scsi_finish_cmd() called for it. we do all of the retry stuff
1681 * here, so when we restart the host after we return it should have an
1682 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001683 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684static void scsi_unjam_host(struct Scsi_Host *shost)
1685{
1686 unsigned long flags;
1687 LIST_HEAD(eh_work_q);
1688 LIST_HEAD(eh_done_q);
1689
1690 spin_lock_irqsave(shost->host_lock, flags);
1691 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1692 spin_unlock_irqrestore(shost->host_lock, flags);
1693
1694 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1695
1696 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1697 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1698 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1699
1700 scsi_eh_flush_done_q(&eh_done_q);
1701}
1702
1703/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001704 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 * @data: Host for which we are running.
1706 *
1707 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001708 * This is the main error handling loop. This is run as a kernel thread
1709 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001710 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711int scsi_error_handler(void *data)
1712{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001713 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001716 * We use TASK_INTERRUPTIBLE so that the thread is not
1717 * counted against the load average as a running process.
1718 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001719 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001721 set_current_state(TASK_INTERRUPTIBLE);
1722 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001723 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001724 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001725 SCSI_LOG_ERROR_RECOVERY(1,
1726 printk("Error handler scsi_eh_%d sleeping\n",
1727 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001728 schedule();
1729 set_current_state(TASK_INTERRUPTIBLE);
1730 continue;
1731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
James Bottomley3ed7a472005-09-19 09:50:04 -05001733 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001734 SCSI_LOG_ERROR_RECOVERY(1,
1735 printk("Error handler scsi_eh_%d waking up\n",
1736 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 /*
1739 * We have a host that is failing for some reason. Figure out
1740 * what we need to do to get it up and online again (if we can).
1741 * If we fail, we end up taking the thing offline.
1742 */
Christoph Hellwig9227c332006-04-01 19:21:04 +02001743 if (shost->transportt->eh_strategy_handler)
1744 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 else
1746 scsi_unjam_host(shost);
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 /*
1749 * Note - if the above fails completely, the action is to take
1750 * individual devices offline and flush the queue of any
1751 * outstanding requests that may have been pending. When we
1752 * restart, we restart any I/O to any other devices on the bus
1753 * which are still online.
1754 */
1755 scsi_restart_operations(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001756 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001758 __set_current_state(TASK_RUNNING);
1759
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001760 SCSI_LOG_ERROR_RECOVERY(1,
1761 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001762 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 return 0;
1764}
1765
1766/*
1767 * Function: scsi_report_bus_reset()
1768 *
1769 * Purpose: Utility function used by low-level drivers to report that
1770 * they have observed a bus reset on the bus being handled.
1771 *
1772 * Arguments: shost - Host in question
1773 * channel - channel on which reset was observed.
1774 *
1775 * Returns: Nothing
1776 *
1777 * Lock status: Host lock must be held.
1778 *
1779 * Notes: This only needs to be called if the reset is one which
1780 * originates from an unknown location. Resets originated
1781 * by the mid-level itself don't need to call this, but there
1782 * should be no harm.
1783 *
1784 * The main purpose of this is to make sure that a CHECK_CONDITION
1785 * is properly treated.
1786 */
1787void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1788{
1789 struct scsi_device *sdev;
1790
1791 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001792 if (channel == sdev_channel(sdev))
1793 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 }
1795}
1796EXPORT_SYMBOL(scsi_report_bus_reset);
1797
1798/*
1799 * Function: scsi_report_device_reset()
1800 *
1801 * Purpose: Utility function used by low-level drivers to report that
1802 * they have observed a device reset on the device being handled.
1803 *
1804 * Arguments: shost - Host in question
1805 * channel - channel on which reset was observed
1806 * target - target on which reset was observed
1807 *
1808 * Returns: Nothing
1809 *
1810 * Lock status: Host lock must be held
1811 *
1812 * Notes: This only needs to be called if the reset is one which
1813 * originates from an unknown location. Resets originated
1814 * by the mid-level itself don't need to call this, but there
1815 * should be no harm.
1816 *
1817 * The main purpose of this is to make sure that a CHECK_CONDITION
1818 * is properly treated.
1819 */
1820void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1821{
1822 struct scsi_device *sdev;
1823
1824 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001825 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001826 target == sdev_id(sdev))
1827 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 }
1829}
1830EXPORT_SYMBOL(scsi_report_device_reset);
1831
1832static void
1833scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1834{
1835}
1836
1837/*
1838 * Function: scsi_reset_provider
1839 *
1840 * Purpose: Send requested reset to a bus or device at any phase.
1841 *
1842 * Arguments: device - device to send reset to
1843 * flag - reset type (see scsi.h)
1844 *
1845 * Returns: SUCCESS/FAILURE.
1846 *
1847 * Notes: This is used by the SCSI Generic driver to provide
1848 * Bus/Device reset capability.
1849 */
1850int
1851scsi_reset_provider(struct scsi_device *dev, int flag)
1852{
1853 struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
James Smartd7a1bb02006-03-08 14:50:12 -05001854 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001856 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 int rtn;
1858
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001859 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001861
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001862 scmd->cmnd = req.cmd;
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001865 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
1867 scmd->cmd_len = 0;
1868
1869 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
James Smartd7a1bb02006-03-08 14:50:12 -05001871 spin_lock_irqsave(shost->host_lock, flags);
1872 shost->tmf_in_progress = 1;
1873 spin_unlock_irqrestore(shost->host_lock, flags);
1874
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 switch (flag) {
1876 case SCSI_TRY_RESET_DEVICE:
1877 rtn = scsi_try_bus_device_reset(scmd);
1878 if (rtn == SUCCESS)
1879 break;
1880 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001881 case SCSI_TRY_RESET_TARGET:
1882 rtn = scsi_try_target_reset(scmd);
1883 if (rtn == SUCCESS)
1884 break;
1885 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 case SCSI_TRY_RESET_BUS:
1887 rtn = scsi_try_bus_reset(scmd);
1888 if (rtn == SUCCESS)
1889 break;
1890 /* FALLTHROUGH */
1891 case SCSI_TRY_RESET_HOST:
1892 rtn = scsi_try_host_reset(scmd);
1893 break;
1894 default:
1895 rtn = FAILED;
1896 }
1897
James Smartd7a1bb02006-03-08 14:50:12 -05001898 spin_lock_irqsave(shost->host_lock, flags);
1899 shost->tmf_in_progress = 0;
1900 spin_unlock_irqrestore(shost->host_lock, flags);
1901
1902 /*
1903 * be sure to wake up anyone who was sleeping or had their queue
1904 * suspended while we performed the TMF.
1905 */
1906 SCSI_LOG_ERROR_RECOVERY(3,
1907 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001908 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001909
1910 wake_up(&shost->host_wait);
1911
1912 scsi_run_host_queues(shost);
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 scsi_next_command(scmd);
1915 return rtn;
1916}
1917EXPORT_SYMBOL(scsi_reset_provider);
1918
1919/**
1920 * scsi_normalize_sense - normalize main elements from either fixed or
1921 * descriptor sense data format into a common format.
1922 *
1923 * @sense_buffer: byte array containing sense data returned by device
1924 * @sb_len: number of valid bytes in sense_buffer
1925 * @sshdr: pointer to instance of structure that common
1926 * elements are written to.
1927 *
1928 * Notes:
1929 * The "main elements" from sense data are: response_code, sense_key,
1930 * asc, ascq and additional_length (only for descriptor format).
1931 *
1932 * Typically this function can be called after a device has
1933 * responded to a SCSI command with the CHECK_CONDITION status.
1934 *
1935 * Return value:
1936 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001937 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1939 struct scsi_sense_hdr *sshdr)
1940{
James Bottomley33aa6872005-08-28 11:31:14 -05001941 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 return 0;
1943
1944 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1945
1946 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05001947
1948 if (!scsi_sense_valid(sshdr))
1949 return 0;
1950
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 if (sshdr->response_code >= 0x72) {
1952 /*
1953 * descriptor format
1954 */
1955 if (sb_len > 1)
1956 sshdr->sense_key = (sense_buffer[1] & 0xf);
1957 if (sb_len > 2)
1958 sshdr->asc = sense_buffer[2];
1959 if (sb_len > 3)
1960 sshdr->ascq = sense_buffer[3];
1961 if (sb_len > 7)
1962 sshdr->additional_length = sense_buffer[7];
1963 } else {
1964 /*
1965 * fixed format
1966 */
1967 if (sb_len > 2)
1968 sshdr->sense_key = (sense_buffer[2] & 0xf);
1969 if (sb_len > 7) {
1970 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1971 sb_len : (sense_buffer[7] + 8);
1972 if (sb_len > 12)
1973 sshdr->asc = sense_buffer[12];
1974 if (sb_len > 13)
1975 sshdr->ascq = sense_buffer[13];
1976 }
1977 }
1978
1979 return 1;
1980}
1981EXPORT_SYMBOL(scsi_normalize_sense);
1982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1984 struct scsi_sense_hdr *sshdr)
1985{
1986 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09001987 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988}
1989EXPORT_SYMBOL(scsi_command_normalize_sense);
1990
1991/**
Rob Landleyeb448202007-11-03 13:30:39 -05001992 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 * @sense_buffer: byte array of descriptor format sense data
1994 * @sb_len: number of valid bytes in sense_buffer
1995 * @desc_type: value of descriptor type to find
1996 * (e.g. 0 -> information)
1997 *
1998 * Notes:
1999 * only valid when sense data is in descriptor format
2000 *
2001 * Return value:
2002 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002003 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2005 int desc_type)
2006{
2007 int add_sen_len, add_len, desc_len, k;
2008 const u8 * descp;
2009
2010 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2011 return NULL;
2012 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2013 return NULL;
2014 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2015 add_sen_len : (sb_len - 8);
2016 descp = &sense_buffer[8];
2017 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2018 descp += desc_len;
2019 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2020 desc_len = add_len + 2;
2021 if (descp[0] == desc_type)
2022 return descp;
2023 if (add_len < 0) // short descriptor ??
2024 break;
2025 }
2026 return NULL;
2027}
2028EXPORT_SYMBOL(scsi_sense_desc_find);
2029
2030/**
Rob Landleyeb448202007-11-03 13:30:39 -05002031 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 * @sense_buffer: byte array of sense data
2033 * @sb_len: number of valid bytes in sense_buffer
2034 * @info_out: pointer to 64 integer where 8 or 4 byte information
2035 * field will be placed if found.
2036 *
2037 * Return value:
2038 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002039 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2041 u64 * info_out)
2042{
2043 int j;
2044 const u8 * ucp;
2045 u64 ull;
2046
2047 if (sb_len < 7)
2048 return 0;
2049 switch (sense_buffer[0] & 0x7f) {
2050 case 0x70:
2051 case 0x71:
2052 if (sense_buffer[0] & 0x80) {
2053 *info_out = (sense_buffer[3] << 24) +
2054 (sense_buffer[4] << 16) +
2055 (sense_buffer[5] << 8) + sense_buffer[6];
2056 return 1;
2057 } else
2058 return 0;
2059 case 0x72:
2060 case 0x73:
2061 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2062 0 /* info desc */);
2063 if (ucp && (0xa == ucp[1])) {
2064 ull = 0;
2065 for (j = 0; j < 8; ++j) {
2066 if (j > 0)
2067 ull <<= 8;
2068 ull |= ucp[4 + j];
2069 }
2070 *info_out = ull;
2071 return 1;
2072 } else
2073 return 0;
2074 default:
2075 return 0;
2076 }
2077}
2078EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002079
2080/**
2081 * scsi_build_sense_buffer - build sense data in a buffer
2082 * @desc: Sense format (non zero == descriptor format,
2083 * 0 == fixed format)
2084 * @buf: Where to build sense data
2085 * @key: Sense key
2086 * @asc: Additional sense code
2087 * @ascq: Additional sense code qualifier
2088 *
2089 **/
2090void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2091{
2092 if (desc) {
2093 buf[0] = 0x72; /* descriptor, current */
2094 buf[1] = key;
2095 buf[2] = asc;
2096 buf[3] = ascq;
2097 buf[7] = 0;
2098 } else {
2099 buf[0] = 0x70; /* fixed, current */
2100 buf[2] = key;
2101 buf[7] = 0xa;
2102 buf[12] = asc;
2103 buf[13] = ascq;
2104 }
2105}
2106EXPORT_SYMBOL(scsi_build_sense_buffer);