Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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> |
Christoph Hellwig | c5478de | 2005-09-06 14:04:26 +0200 | [diff] [blame] | 23 | #include <linux/kthread.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/interrupt.h> |
| 25 | #include <linux/blkdev.h> |
| 26 | #include <linux/delay.h> |
| 27 | |
| 28 | #include <scsi/scsi.h> |
Christoph Hellwig | beb4048 | 2006-06-10 18:01:03 +0200 | [diff] [blame] | 29 | #include <scsi/scsi_cmnd.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <scsi/scsi_dbg.h> |
| 31 | #include <scsi/scsi_device.h> |
| 32 | #include <scsi/scsi_eh.h> |
James Smart | c829c39 | 2006-03-13 08:28:57 -0500 | [diff] [blame] | 33 | #include <scsi/scsi_transport.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include <scsi/scsi_host.h> |
| 35 | #include <scsi/scsi_ioctl.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | |
| 37 | #include "scsi_priv.h" |
| 38 | #include "scsi_logging.h" |
| 39 | |
| 40 | #define SENSE_TIMEOUT (10*HZ) |
| 41 | #define START_UNIT_TIMEOUT (30*HZ) |
| 42 | |
| 43 | /* |
| 44 | * These should *probably* be handled by the host itself. |
| 45 | * Since it is allowed to sleep, it probably should. |
| 46 | */ |
| 47 | #define BUS_RESET_SETTLE_TIME (10) |
| 48 | #define HOST_RESET_SETTLE_TIME (10) |
| 49 | |
| 50 | /* called with shost->host_lock held */ |
| 51 | void scsi_eh_wakeup(struct Scsi_Host *shost) |
| 52 | { |
| 53 | if (shost->host_busy == shost->host_failed) { |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 54 | wake_up_process(shost->ehandler); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | SCSI_LOG_ERROR_RECOVERY(5, |
| 56 | printk("Waking error handler thread\n")); |
| 57 | } |
| 58 | } |
Tejun Heo | f8bbfc2 | 2006-05-19 21:07:05 +0900 | [diff] [blame] | 59 | |
| 60 | /** |
| 61 | * scsi_schedule_eh - schedule EH for SCSI host |
| 62 | * @shost: SCSI host to invoke error handling on. |
| 63 | * |
| 64 | * Schedule SCSI EH without scmd. |
| 65 | **/ |
| 66 | void scsi_schedule_eh(struct Scsi_Host *shost) |
| 67 | { |
| 68 | unsigned long flags; |
| 69 | |
| 70 | spin_lock_irqsave(shost->host_lock, flags); |
| 71 | |
| 72 | if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || |
| 73 | scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { |
| 74 | shost->host_eh_scheduled++; |
| 75 | scsi_eh_wakeup(shost); |
| 76 | } |
| 77 | |
| 78 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 79 | } |
| 80 | EXPORT_SYMBOL_GPL(scsi_schedule_eh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | |
| 82 | /** |
| 83 | * scsi_eh_scmd_add - add scsi cmd to error handling. |
| 84 | * @scmd: scmd to run eh on. |
| 85 | * @eh_flag: optional SCSI_EH flag. |
| 86 | * |
| 87 | * Return value: |
| 88 | * 0 on failure. |
| 89 | **/ |
| 90 | int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) |
| 91 | { |
| 92 | struct Scsi_Host *shost = scmd->device->host; |
| 93 | unsigned long flags; |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 94 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 96 | if (!shost->ehandler) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | return 0; |
| 98 | |
| 99 | spin_lock_irqsave(shost->host_lock, flags); |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 100 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
| 101 | if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) |
| 102 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 104 | ret = 1; |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 105 | scmd->eh_eflags |= eh_flag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | shost->host_failed++; |
| 108 | scsi_eh_wakeup(shost); |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 109 | out_unlock: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | spin_unlock_irqrestore(shost->host_lock, flags); |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 111 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | } |
| 113 | |
| 114 | /** |
| 115 | * scsi_add_timer - Start timeout timer for a single scsi command. |
| 116 | * @scmd: scsi command that is about to start running. |
| 117 | * @timeout: amount of time to allow this command to run. |
| 118 | * @complete: timeout function to call if timer isn't canceled. |
| 119 | * |
| 120 | * Notes: |
| 121 | * This should be turned into an inline function. Each scsi command |
| 122 | * has its own timer, and as it is added to the queue, we set up the |
| 123 | * timer. When the command completes, we cancel the timer. |
| 124 | **/ |
| 125 | void scsi_add_timer(struct scsi_cmnd *scmd, int timeout, |
| 126 | void (*complete)(struct scsi_cmnd *)) |
| 127 | { |
| 128 | |
| 129 | /* |
| 130 | * If the clock was already running for this command, then |
| 131 | * first delete the timer. The timer handling code gets rather |
| 132 | * confused if we don't do this. |
| 133 | */ |
| 134 | if (scmd->eh_timeout.function) |
| 135 | del_timer(&scmd->eh_timeout); |
| 136 | |
| 137 | scmd->eh_timeout.data = (unsigned long)scmd; |
| 138 | scmd->eh_timeout.expires = jiffies + timeout; |
| 139 | scmd->eh_timeout.function = (void (*)(unsigned long)) complete; |
| 140 | |
| 141 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:" |
| 142 | " %d, (%p)\n", __FUNCTION__, |
| 143 | scmd, timeout, complete)); |
| 144 | |
| 145 | add_timer(&scmd->eh_timeout); |
| 146 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | |
| 148 | /** |
| 149 | * scsi_delete_timer - Delete/cancel timer for a given function. |
| 150 | * @scmd: Cmd that we are canceling timer for |
| 151 | * |
| 152 | * Notes: |
| 153 | * This should be turned into an inline function. |
| 154 | * |
| 155 | * Return value: |
| 156 | * 1 if we were able to detach the timer. 0 if we blew it, and the |
| 157 | * timer function has already started to run. |
| 158 | **/ |
| 159 | int scsi_delete_timer(struct scsi_cmnd *scmd) |
| 160 | { |
| 161 | int rtn; |
| 162 | |
| 163 | rtn = del_timer(&scmd->eh_timeout); |
| 164 | |
| 165 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p," |
| 166 | " rtn: %d\n", __FUNCTION__, |
| 167 | scmd, rtn)); |
| 168 | |
| 169 | scmd->eh_timeout.data = (unsigned long)NULL; |
| 170 | scmd->eh_timeout.function = NULL; |
| 171 | |
| 172 | return rtn; |
| 173 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | |
| 175 | /** |
| 176 | * scsi_times_out - Timeout function for normal scsi commands. |
| 177 | * @scmd: Cmd that is timing out. |
| 178 | * |
| 179 | * Notes: |
| 180 | * We do not need to lock this. There is the potential for a race |
| 181 | * only in that the normal completion handling might run, but if the |
| 182 | * normal completion function determines that the timer has already |
| 183 | * fired, then it mustn't do anything. |
| 184 | **/ |
| 185 | void scsi_times_out(struct scsi_cmnd *scmd) |
| 186 | { |
| 187 | scsi_log_completion(scmd, TIMEOUT_ERROR); |
| 188 | |
James Smart | c829c39 | 2006-03-13 08:28:57 -0500 | [diff] [blame] | 189 | if (scmd->device->host->transportt->eh_timed_out) |
| 190 | switch (scmd->device->host->transportt->eh_timed_out(scmd)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | case EH_HANDLED: |
| 192 | __scsi_done(scmd); |
| 193 | return; |
| 194 | case EH_RESET_TIMER: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | scsi_add_timer(scmd, scmd->timeout_per_command, |
| 196 | scsi_times_out); |
| 197 | return; |
| 198 | case EH_NOT_HANDLED: |
| 199 | break; |
| 200 | } |
| 201 | |
| 202 | if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 203 | scmd->result |= DID_TIME_OUT << 16; |
| 204 | __scsi_done(scmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | } |
| 206 | } |
| 207 | |
| 208 | /** |
| 209 | * scsi_block_when_processing_errors - Prevent cmds from being queued. |
| 210 | * @sdev: Device on which we are performing recovery. |
| 211 | * |
| 212 | * Description: |
| 213 | * We block until the host is out of error recovery, and then check to |
| 214 | * see whether the host or the device is offline. |
| 215 | * |
| 216 | * Return value: |
| 217 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. |
| 218 | **/ |
| 219 | int scsi_block_when_processing_errors(struct scsi_device *sdev) |
| 220 | { |
| 221 | int online; |
| 222 | |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 223 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 224 | |
| 225 | online = scsi_device_online(sdev); |
| 226 | |
| 227 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__, |
| 228 | online)); |
| 229 | |
| 230 | return online; |
| 231 | } |
| 232 | EXPORT_SYMBOL(scsi_block_when_processing_errors); |
| 233 | |
| 234 | #ifdef CONFIG_SCSI_LOGGING |
| 235 | /** |
| 236 | * scsi_eh_prt_fail_stats - Log info on failures. |
| 237 | * @shost: scsi host being recovered. |
| 238 | * @work_q: Queue of scsi cmds to process. |
| 239 | **/ |
| 240 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, |
| 241 | struct list_head *work_q) |
| 242 | { |
| 243 | struct scsi_cmnd *scmd; |
| 244 | struct scsi_device *sdev; |
| 245 | int total_failures = 0; |
| 246 | int cmd_failed = 0; |
| 247 | int cmd_cancel = 0; |
| 248 | int devices_failed = 0; |
| 249 | |
| 250 | shost_for_each_device(sdev, shost) { |
| 251 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 252 | if (scmd->device == sdev) { |
| 253 | ++total_failures; |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 254 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | ++cmd_cancel; |
| 256 | else |
| 257 | ++cmd_failed; |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | if (cmd_cancel || cmd_failed) { |
| 262 | SCSI_LOG_ERROR_RECOVERY(3, |
James Bottomley | 9ccfc75 | 2005-10-02 11:45:08 -0500 | [diff] [blame] | 263 | sdev_printk(KERN_INFO, sdev, |
| 264 | "%s: cmds failed: %d, cancel: %d\n", |
| 265 | __FUNCTION__, cmd_failed, |
| 266 | cmd_cancel)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | cmd_cancel = 0; |
| 268 | cmd_failed = 0; |
| 269 | ++devices_failed; |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" |
| 274 | " devices require eh work\n", |
| 275 | total_failures, devices_failed)); |
| 276 | } |
| 277 | #endif |
| 278 | |
| 279 | /** |
| 280 | * scsi_check_sense - Examine scsi cmd sense |
| 281 | * @scmd: Cmd to have sense checked. |
| 282 | * |
| 283 | * Return value: |
| 284 | * SUCCESS or FAILED or NEEDS_RETRY |
| 285 | * |
| 286 | * Notes: |
| 287 | * When a deferred error is detected the current command has |
| 288 | * not been executed and needs retrying. |
| 289 | **/ |
| 290 | static int scsi_check_sense(struct scsi_cmnd *scmd) |
| 291 | { |
| 292 | struct scsi_sense_hdr sshdr; |
| 293 | |
| 294 | if (! scsi_command_normalize_sense(scmd, &sshdr)) |
| 295 | return FAILED; /* no valid sense data */ |
| 296 | |
| 297 | if (scsi_sense_is_deferred(&sshdr)) |
| 298 | return NEEDS_RETRY; |
| 299 | |
| 300 | /* |
| 301 | * Previous logic looked for FILEMARK, EOM or ILI which are |
| 302 | * mainly associated with tapes and returned SUCCESS. |
| 303 | */ |
| 304 | if (sshdr.response_code == 0x70) { |
| 305 | /* fixed format */ |
| 306 | if (scmd->sense_buffer[2] & 0xe0) |
| 307 | return SUCCESS; |
| 308 | } else { |
| 309 | /* |
| 310 | * descriptor format: look for "stream commands sense data |
| 311 | * descriptor" (see SSC-3). Assume single sense data |
| 312 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. |
| 313 | */ |
| 314 | if ((sshdr.additional_length > 3) && |
| 315 | (scmd->sense_buffer[8] == 0x4) && |
| 316 | (scmd->sense_buffer[11] & 0xe0)) |
| 317 | return SUCCESS; |
| 318 | } |
| 319 | |
| 320 | switch (sshdr.sense_key) { |
| 321 | case NO_SENSE: |
| 322 | return SUCCESS; |
| 323 | case RECOVERED_ERROR: |
| 324 | return /* soft_error */ SUCCESS; |
| 325 | |
| 326 | case ABORTED_COMMAND: |
| 327 | return NEEDS_RETRY; |
| 328 | case NOT_READY: |
| 329 | case UNIT_ATTENTION: |
| 330 | /* |
| 331 | * if we are expecting a cc/ua because of a bus reset that we |
| 332 | * performed, treat this just as a retry. otherwise this is |
| 333 | * information that we should pass up to the upper-level driver |
| 334 | * so that we can deal with it there. |
| 335 | */ |
| 336 | if (scmd->device->expecting_cc_ua) { |
| 337 | scmd->device->expecting_cc_ua = 0; |
| 338 | return NEEDS_RETRY; |
| 339 | } |
| 340 | /* |
| 341 | * if the device is in the process of becoming ready, we |
| 342 | * should retry. |
| 343 | */ |
| 344 | if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) |
| 345 | return NEEDS_RETRY; |
| 346 | /* |
| 347 | * if the device is not started, we need to wake |
| 348 | * the error handler to start the motor |
| 349 | */ |
| 350 | if (scmd->device->allow_restart && |
| 351 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) |
| 352 | return FAILED; |
| 353 | return SUCCESS; |
| 354 | |
| 355 | /* these three are not supported */ |
| 356 | case COPY_ABORTED: |
| 357 | case VOLUME_OVERFLOW: |
| 358 | case MISCOMPARE: |
| 359 | return SUCCESS; |
| 360 | |
| 361 | case MEDIUM_ERROR: |
| 362 | return NEEDS_RETRY; |
| 363 | |
| 364 | case HARDWARE_ERROR: |
| 365 | if (scmd->device->retry_hwerror) |
| 366 | return NEEDS_RETRY; |
| 367 | else |
| 368 | return SUCCESS; |
| 369 | |
| 370 | case ILLEGAL_REQUEST: |
| 371 | case BLANK_CHECK: |
| 372 | case DATA_PROTECT: |
| 373 | default: |
| 374 | return SUCCESS; |
| 375 | } |
| 376 | } |
| 377 | |
| 378 | /** |
| 379 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. |
| 380 | * @scmd: SCSI cmd to examine. |
| 381 | * |
| 382 | * Notes: |
| 383 | * This is *only* called when we are examining the status of commands |
| 384 | * queued during error recovery. the main difference here is that we |
| 385 | * don't allow for the possibility of retries here, and we are a lot |
| 386 | * more restrictive about what we consider acceptable. |
| 387 | **/ |
| 388 | static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) |
| 389 | { |
| 390 | /* |
| 391 | * first check the host byte, to see if there is anything in there |
| 392 | * that would indicate what we need to do. |
| 393 | */ |
| 394 | if (host_byte(scmd->result) == DID_RESET) { |
| 395 | /* |
| 396 | * rats. we are already in the error handler, so we now |
| 397 | * get to try and figure out what to do next. if the sense |
| 398 | * is valid, we have a pretty good idea of what to do. |
| 399 | * if not, we mark it as FAILED. |
| 400 | */ |
| 401 | return scsi_check_sense(scmd); |
| 402 | } |
| 403 | if (host_byte(scmd->result) != DID_OK) |
| 404 | return FAILED; |
| 405 | |
| 406 | /* |
| 407 | * next, check the message byte. |
| 408 | */ |
| 409 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) |
| 410 | return FAILED; |
| 411 | |
| 412 | /* |
| 413 | * now, check the status byte to see if this indicates |
| 414 | * anything special. |
| 415 | */ |
| 416 | switch (status_byte(scmd->result)) { |
| 417 | case GOOD: |
| 418 | case COMMAND_TERMINATED: |
| 419 | return SUCCESS; |
| 420 | case CHECK_CONDITION: |
| 421 | return scsi_check_sense(scmd); |
| 422 | case CONDITION_GOOD: |
| 423 | case INTERMEDIATE_GOOD: |
| 424 | case INTERMEDIATE_C_GOOD: |
| 425 | /* |
| 426 | * who knows? FIXME(eric) |
| 427 | */ |
| 428 | return SUCCESS; |
| 429 | case BUSY: |
| 430 | case QUEUE_FULL: |
| 431 | case RESERVATION_CONFLICT: |
| 432 | default: |
| 433 | return FAILED; |
| 434 | } |
| 435 | return FAILED; |
| 436 | } |
| 437 | |
| 438 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | * scsi_eh_done - Completion function for error handling. |
| 440 | * @scmd: Cmd that is done. |
| 441 | **/ |
| 442 | static void scsi_eh_done(struct scsi_cmnd *scmd) |
| 443 | { |
Michael Reed | 8563167 | 2005-12-07 21:46:27 -0600 | [diff] [blame] | 444 | struct completion *eh_action; |
| 445 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 446 | SCSI_LOG_ERROR_RECOVERY(3, |
| 447 | printk("%s scmd: %p result: %x\n", |
| 448 | __FUNCTION__, scmd, scmd->result)); |
Michael Reed | 8563167 | 2005-12-07 21:46:27 -0600 | [diff] [blame] | 449 | |
| 450 | eh_action = scmd->device->host->eh_action; |
| 451 | if (eh_action) |
| 452 | complete(eh_action); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | } |
| 454 | |
| 455 | /** |
| 456 | * scsi_send_eh_cmnd - send a cmd to a device as part of error recovery. |
| 457 | * @scmd: SCSI Cmd to send. |
| 458 | * @timeout: Timeout for cmd. |
| 459 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | * Return value: |
| 461 | * SUCCESS or FAILED or NEEDS_RETRY |
| 462 | **/ |
Mike Christie | 0db99e3 | 2006-08-26 03:00:22 -0400 | [diff] [blame] | 463 | static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, |
| 464 | int cmnd_size, int timeout, int copy_sense) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | { |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 466 | struct scsi_device *sdev = scmd->device; |
| 467 | struct Scsi_Host *shost = sdev->host; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 468 | int old_result = scmd->result; |
Stefan Richter | 7fbb364 | 2006-09-12 20:35:54 -0700 | [diff] [blame] | 469 | DECLARE_COMPLETION_ONSTACK(done); |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 470 | unsigned long timeleft; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | unsigned long flags; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 472 | unsigned char old_cmnd[MAX_COMMAND_SIZE]; |
| 473 | enum dma_data_direction old_data_direction; |
| 474 | unsigned short old_use_sg; |
| 475 | unsigned char old_cmd_len; |
| 476 | unsigned old_bufflen; |
| 477 | void *old_buffer; |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 478 | int rtn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 480 | /* |
| 481 | * We need saved copies of a number of fields - this is because |
| 482 | * error handling may need to overwrite these with different values |
| 483 | * to run different commands, and once error handling is complete, |
| 484 | * we will need to restore these values prior to running the actual |
| 485 | * command. |
| 486 | */ |
| 487 | old_buffer = scmd->request_buffer; |
| 488 | old_bufflen = scmd->request_bufflen; |
| 489 | memcpy(old_cmnd, scmd->cmnd, sizeof(scmd->cmnd)); |
| 490 | old_data_direction = scmd->sc_data_direction; |
| 491 | old_cmd_len = scmd->cmd_len; |
| 492 | old_use_sg = scmd->use_sg; |
| 493 | |
Mike Christie | 0db99e3 | 2006-08-26 03:00:22 -0400 | [diff] [blame] | 494 | memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); |
| 495 | memcpy(scmd->cmnd, cmnd, cmnd_size); |
| 496 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 497 | if (copy_sense) { |
| 498 | int gfp_mask = GFP_ATOMIC; |
| 499 | |
| 500 | if (shost->hostt->unchecked_isa_dma) |
| 501 | gfp_mask |= __GFP_DMA; |
| 502 | |
| 503 | scmd->sc_data_direction = DMA_FROM_DEVICE; |
| 504 | scmd->request_bufflen = 252; |
| 505 | scmd->request_buffer = kzalloc(scmd->request_bufflen, gfp_mask); |
| 506 | if (!scmd->request_buffer) |
| 507 | return FAILED; |
| 508 | } else { |
| 509 | scmd->request_buffer = NULL; |
| 510 | scmd->request_bufflen = 0; |
| 511 | scmd->sc_data_direction = DMA_NONE; |
| 512 | } |
| 513 | |
| 514 | scmd->underflow = 0; |
| 515 | scmd->use_sg = 0; |
| 516 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 517 | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 518 | if (sdev->scsi_level <= SCSI_2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 520 | (sdev->lun << 5 & 0xe0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 522 | /* |
| 523 | * Zero the sense buffer. The scsi spec mandates that any |
| 524 | * untransferred sense data should be interpreted as being zero. |
| 525 | */ |
| 526 | memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer)); |
| 527 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 528 | shost->eh_action = &done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 530 | spin_lock_irqsave(shost->host_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | scsi_log_send(scmd); |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 532 | shost->hostt->queuecommand(scmd, scsi_eh_done); |
| 533 | spin_unlock_irqrestore(shost->host_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 535 | timeleft = wait_for_completion_timeout(&done, timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 537 | shost->eh_action = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 538 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 539 | scsi_log_completion(scmd, SUCCESS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 541 | SCSI_LOG_ERROR_RECOVERY(3, |
| 542 | printk("%s: scmd: %p, timeleft: %ld\n", |
| 543 | __FUNCTION__, scmd, timeleft)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | |
| 545 | /* |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 546 | * If there is time left scsi_eh_done got called, and we will |
| 547 | * examine the actual status codes to see whether the command |
| 548 | * actually did complete normally, else tell the host to forget |
| 549 | * about this command. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | */ |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 551 | if (timeleft) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 552 | rtn = scsi_eh_completed_normally(scmd); |
| 553 | SCSI_LOG_ERROR_RECOVERY(3, |
| 554 | printk("%s: scsi_eh_completed_normally %x\n", |
| 555 | __FUNCTION__, rtn)); |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 556 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | switch (rtn) { |
| 558 | case SUCCESS: |
| 559 | case NEEDS_RETRY: |
| 560 | case FAILED: |
| 561 | break; |
| 562 | default: |
| 563 | rtn = FAILED; |
| 564 | break; |
| 565 | } |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 566 | } else { |
| 567 | /* |
| 568 | * FIXME(eric) - we are not tracking whether we could |
| 569 | * abort a timed out command or not. not sure how |
| 570 | * we should treat them differently anyways. |
| 571 | */ |
| 572 | if (shost->hostt->eh_abort_handler) |
| 573 | shost->hostt->eh_abort_handler(scmd); |
| 574 | rtn = FAILED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | } |
| 576 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 577 | |
| 578 | /* |
| 579 | * Last chance to have valid sense data. |
| 580 | */ |
| 581 | if (copy_sense) { |
| 582 | if (!SCSI_SENSE_VALID(scmd)) { |
| 583 | memcpy(scmd->sense_buffer, scmd->request_buffer, |
| 584 | sizeof(scmd->sense_buffer)); |
| 585 | } |
| 586 | kfree(scmd->request_buffer); |
| 587 | } |
| 588 | |
| 589 | |
| 590 | /* |
| 591 | * Restore original data |
| 592 | */ |
| 593 | scmd->request_buffer = old_buffer; |
| 594 | scmd->request_bufflen = old_bufflen; |
| 595 | memcpy(scmd->cmnd, old_cmnd, sizeof(scmd->cmnd)); |
| 596 | scmd->sc_data_direction = old_data_direction; |
| 597 | scmd->cmd_len = old_cmd_len; |
| 598 | scmd->use_sg = old_use_sg; |
| 599 | scmd->result = old_result; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | return rtn; |
| 601 | } |
| 602 | |
| 603 | /** |
| 604 | * scsi_request_sense - Request sense data from a particular target. |
| 605 | * @scmd: SCSI cmd for request sense. |
| 606 | * |
| 607 | * Notes: |
| 608 | * Some hosts automatically obtain this information, others require |
| 609 | * that we obtain it on our own. This function will *not* return until |
| 610 | * the command either times out, or it completes. |
| 611 | **/ |
| 612 | static int scsi_request_sense(struct scsi_cmnd *scmd) |
| 613 | { |
| 614 | static unsigned char generic_sense[6] = |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 615 | {REQUEST_SENSE, 0, 0, 0, 252, 0}; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | |
Mike Christie | 0db99e3 | 2006-08-26 03:00:22 -0400 | [diff] [blame] | 617 | return scsi_send_eh_cmnd(scmd, generic_sense, 6, SENSE_TIMEOUT, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | } |
| 619 | |
| 620 | /** |
| 621 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. |
| 622 | * @scmd: Original SCSI cmd that eh has finished. |
| 623 | * @done_q: Queue for processed commands. |
| 624 | * |
| 625 | * Notes: |
| 626 | * We don't want to use the normal command completion while we are are |
| 627 | * still handling errors - it may cause other commands to be queued, |
| 628 | * and that would disturb what we are doing. thus we really want to |
| 629 | * keep a list of pending commands for final completion, and once we |
| 630 | * are ready to leave error handling we handle completion for real. |
| 631 | **/ |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 632 | void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 633 | { |
| 634 | scmd->device->host->host_failed--; |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 635 | scmd->eh_eflags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | list_move_tail(&scmd->eh_entry, done_q); |
| 637 | } |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 638 | EXPORT_SYMBOL(scsi_eh_finish_cmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | |
| 640 | /** |
| 641 | * scsi_eh_get_sense - Get device sense data. |
| 642 | * @work_q: Queue of commands to process. |
| 643 | * @done_q: Queue of proccessed commands.. |
| 644 | * |
| 645 | * Description: |
| 646 | * See if we need to request sense information. if so, then get it |
| 647 | * now, so we have a better idea of what to do. |
| 648 | * |
| 649 | * Notes: |
| 650 | * This has the unfortunate side effect that if a shost adapter does |
| 651 | * not automatically request sense information, that we end up shutting |
| 652 | * it down before we request it. |
| 653 | * |
| 654 | * All drivers should request sense information internally these days, |
| 655 | * so for now all I have to say is tough noogies if you end up in here. |
| 656 | * |
| 657 | * XXX: Long term this code should go away, but that needs an audit of |
| 658 | * all LLDDs first. |
| 659 | **/ |
| 660 | static int scsi_eh_get_sense(struct list_head *work_q, |
| 661 | struct list_head *done_q) |
| 662 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 663 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 664 | int rtn; |
| 665 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 666 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 667 | if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | SCSI_SENSE_VALID(scmd)) |
| 669 | continue; |
| 670 | |
Jeff Garzik | 3bf743e | 2005-10-24 18:04:06 -0400 | [diff] [blame] | 671 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
| 672 | "%s: requesting sense\n", |
| 673 | current->comm)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | rtn = scsi_request_sense(scmd); |
| 675 | if (rtn != SUCCESS) |
| 676 | continue; |
| 677 | |
| 678 | SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" |
| 679 | " result %x\n", scmd, |
| 680 | scmd->result)); |
| 681 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); |
| 682 | |
| 683 | rtn = scsi_decide_disposition(scmd); |
| 684 | |
| 685 | /* |
| 686 | * if the result was normal, then just pass it along to the |
| 687 | * upper level. |
| 688 | */ |
| 689 | if (rtn == SUCCESS) |
| 690 | /* we don't want this command reissued, just |
| 691 | * finished with the sense data, so set |
| 692 | * retries to the max allowed to ensure it |
| 693 | * won't get reissued */ |
| 694 | scmd->retries = scmd->allowed; |
| 695 | else if (rtn != NEEDS_RETRY) |
| 696 | continue; |
| 697 | |
| 698 | scsi_eh_finish_cmd(scmd, done_q); |
| 699 | } |
| 700 | |
| 701 | return list_empty(work_q); |
| 702 | } |
| 703 | |
| 704 | /** |
| 705 | * scsi_try_to_abort_cmd - Ask host to abort a running command. |
| 706 | * @scmd: SCSI cmd to abort from Lower Level. |
| 707 | * |
| 708 | * Notes: |
| 709 | * This function will not return until the user's completion function |
| 710 | * has been called. there is no timeout on this operation. if the |
| 711 | * author of the low-level driver wishes this operation to be timed, |
| 712 | * they can provide this facility themselves. helper functions in |
| 713 | * scsi_error.c can be supplied to make this easier to do. |
| 714 | **/ |
| 715 | static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) |
| 716 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | if (!scmd->device->host->hostt->eh_abort_handler) |
Jeff Garzik | 8fa728a | 2005-05-28 07:54:40 -0400 | [diff] [blame] | 718 | return FAILED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 719 | |
| 720 | /* |
| 721 | * scsi_done was called just after the command timed out and before |
| 722 | * we had a chance to process it. (db) |
| 723 | */ |
| 724 | if (scmd->serial_number == 0) |
| 725 | return SUCCESS; |
Jeff Garzik | 8fa728a | 2005-05-28 07:54:40 -0400 | [diff] [blame] | 726 | return scmd->device->host->hostt->eh_abort_handler(scmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 727 | } |
| 728 | |
| 729 | /** |
| 730 | * scsi_eh_tur - Send TUR to device. |
| 731 | * @scmd: Scsi cmd to send TUR |
| 732 | * |
| 733 | * Return value: |
| 734 | * 0 - Device is ready. 1 - Device NOT ready. |
| 735 | **/ |
| 736 | static int scsi_eh_tur(struct scsi_cmnd *scmd) |
| 737 | { |
| 738 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; |
| 739 | int retry_cnt = 1, rtn; |
| 740 | |
| 741 | retry_tur: |
Mike Christie | 0db99e3 | 2006-08-26 03:00:22 -0400 | [diff] [blame] | 742 | rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", |
| 745 | __FUNCTION__, scmd, rtn)); |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 746 | |
| 747 | switch (rtn) { |
| 748 | case NEEDS_RETRY: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | if (retry_cnt--) |
| 750 | goto retry_tur; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 751 | /*FALLTHRU*/ |
| 752 | case SUCCESS: |
Alan Stern | e47373e | 2005-03-30 15:05:45 -0500 | [diff] [blame] | 753 | return 0; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 754 | default: |
| 755 | return 1; |
Alan Stern | e47373e | 2005-03-30 15:05:45 -0500 | [diff] [blame] | 756 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | } |
| 758 | |
| 759 | /** |
| 760 | * scsi_eh_abort_cmds - abort canceled commands. |
| 761 | * @shost: scsi host being recovered. |
| 762 | * @eh_done_q: list_head for processed commands. |
| 763 | * |
| 764 | * Decription: |
| 765 | * Try and see whether or not it makes sense to try and abort the |
| 766 | * running command. this only works out to be the case if we have one |
| 767 | * command that has timed out. if the command simply failed, it makes |
| 768 | * no sense to try and abort the command, since as far as the shost |
| 769 | * adapter is concerned, it isn't running. |
| 770 | **/ |
| 771 | static int scsi_eh_abort_cmds(struct list_head *work_q, |
| 772 | struct list_head *done_q) |
| 773 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 774 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 775 | int rtn; |
| 776 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 777 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 778 | if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 779 | continue; |
| 780 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" |
| 781 | "0x%p\n", current->comm, |
| 782 | scmd)); |
| 783 | rtn = scsi_try_to_abort_cmd(scmd); |
| 784 | if (rtn == SUCCESS) { |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 785 | scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | if (!scsi_device_online(scmd->device) || |
| 787 | !scsi_eh_tur(scmd)) { |
| 788 | scsi_eh_finish_cmd(scmd, done_q); |
| 789 | } |
| 790 | |
| 791 | } else |
| 792 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" |
| 793 | " cmd failed:" |
| 794 | "0x%p\n", |
| 795 | current->comm, |
| 796 | scmd)); |
| 797 | } |
| 798 | |
| 799 | return list_empty(work_q); |
| 800 | } |
| 801 | |
| 802 | /** |
| 803 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev |
| 804 | * @scmd: SCSI cmd used to send BDR |
| 805 | * |
| 806 | * Notes: |
| 807 | * There is no timeout for this operation. if this operation is |
| 808 | * unreliable for a given host, then the host itself needs to put a |
| 809 | * timer on it, and set the host back to a consistent state prior to |
| 810 | * returning. |
| 811 | **/ |
| 812 | static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) |
| 813 | { |
Jeff Garzik | 94d0e7b8 | 2005-05-28 07:55:48 -0400 | [diff] [blame] | 814 | int rtn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | |
| 816 | if (!scmd->device->host->hostt->eh_device_reset_handler) |
Jeff Garzik | 94d0e7b8 | 2005-05-28 07:55:48 -0400 | [diff] [blame] | 817 | return FAILED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 818 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 819 | rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 820 | if (rtn == SUCCESS) { |
| 821 | scmd->device->was_reset = 1; |
| 822 | scmd->device->expecting_cc_ua = 1; |
| 823 | } |
| 824 | |
| 825 | return rtn; |
| 826 | } |
| 827 | |
| 828 | /** |
| 829 | * scsi_eh_try_stu - Send START_UNIT to device. |
| 830 | * @scmd: Scsi cmd to send START_UNIT |
| 831 | * |
| 832 | * Return value: |
| 833 | * 0 - Device is ready. 1 - Device NOT ready. |
| 834 | **/ |
| 835 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) |
| 836 | { |
| 837 | static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 839 | if (scmd->device->allow_restart) { |
| 840 | int rtn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 841 | |
Mike Christie | 0db99e3 | 2006-08-26 03:00:22 -0400 | [diff] [blame] | 842 | rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, |
| 843 | START_UNIT_TIMEOUT, 0); |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 844 | if (rtn == SUCCESS) |
| 845 | return 0; |
| 846 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 847 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 848 | return 1; |
| 849 | } |
| 850 | |
| 851 | /** |
| 852 | * scsi_eh_stu - send START_UNIT if needed |
| 853 | * @shost: scsi host being recovered. |
| 854 | * @eh_done_q: list_head for processed commands. |
| 855 | * |
| 856 | * Notes: |
| 857 | * If commands are failing due to not ready, initializing command required, |
| 858 | * try revalidating the device, which will end up sending a start unit. |
| 859 | **/ |
| 860 | static int scsi_eh_stu(struct Scsi_Host *shost, |
| 861 | struct list_head *work_q, |
| 862 | struct list_head *done_q) |
| 863 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 864 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | struct scsi_device *sdev; |
| 866 | |
| 867 | shost_for_each_device(sdev, shost) { |
| 868 | stu_scmd = NULL; |
| 869 | list_for_each_entry(scmd, work_q, eh_entry) |
| 870 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && |
| 871 | scsi_check_sense(scmd) == FAILED ) { |
| 872 | stu_scmd = scmd; |
| 873 | break; |
| 874 | } |
| 875 | |
| 876 | if (!stu_scmd) |
| 877 | continue; |
| 878 | |
| 879 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" |
| 880 | " 0x%p\n", current->comm, sdev)); |
| 881 | |
| 882 | if (!scsi_eh_try_stu(stu_scmd)) { |
| 883 | if (!scsi_device_online(sdev) || |
| 884 | !scsi_eh_tur(stu_scmd)) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 885 | list_for_each_entry_safe(scmd, next, |
| 886 | work_q, eh_entry) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 887 | if (scmd->device == sdev) |
| 888 | scsi_eh_finish_cmd(scmd, done_q); |
| 889 | } |
| 890 | } |
| 891 | } else { |
| 892 | SCSI_LOG_ERROR_RECOVERY(3, |
| 893 | printk("%s: START_UNIT failed to sdev:" |
| 894 | " 0x%p\n", current->comm, sdev)); |
| 895 | } |
| 896 | } |
| 897 | |
| 898 | return list_empty(work_q); |
| 899 | } |
| 900 | |
| 901 | |
| 902 | /** |
| 903 | * scsi_eh_bus_device_reset - send bdr if needed |
| 904 | * @shost: scsi host being recovered. |
| 905 | * @eh_done_q: list_head for processed commands. |
| 906 | * |
| 907 | * Notes: |
| 908 | * Try a bus device reset. still, look to see whether we have multiple |
| 909 | * devices that are jammed or not - if we have multiple devices, it |
| 910 | * makes no sense to try bus_device_reset - we really would need to try |
| 911 | * a bus_reset instead. |
| 912 | **/ |
| 913 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, |
| 914 | struct list_head *work_q, |
| 915 | struct list_head *done_q) |
| 916 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 917 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 918 | struct scsi_device *sdev; |
| 919 | int rtn; |
| 920 | |
| 921 | shost_for_each_device(sdev, shost) { |
| 922 | bdr_scmd = NULL; |
| 923 | list_for_each_entry(scmd, work_q, eh_entry) |
| 924 | if (scmd->device == sdev) { |
| 925 | bdr_scmd = scmd; |
| 926 | break; |
| 927 | } |
| 928 | |
| 929 | if (!bdr_scmd) |
| 930 | continue; |
| 931 | |
| 932 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" |
| 933 | " 0x%p\n", current->comm, |
| 934 | sdev)); |
| 935 | rtn = scsi_try_bus_device_reset(bdr_scmd); |
| 936 | if (rtn == SUCCESS) { |
| 937 | if (!scsi_device_online(sdev) || |
| 938 | !scsi_eh_tur(bdr_scmd)) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 939 | list_for_each_entry_safe(scmd, next, |
| 940 | work_q, eh_entry) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 941 | if (scmd->device == sdev) |
| 942 | scsi_eh_finish_cmd(scmd, |
| 943 | done_q); |
| 944 | } |
| 945 | } |
| 946 | } else { |
| 947 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" |
| 948 | " failed sdev:" |
| 949 | "0x%p\n", |
| 950 | current->comm, |
| 951 | sdev)); |
| 952 | } |
| 953 | } |
| 954 | |
| 955 | return list_empty(work_q); |
| 956 | } |
| 957 | |
| 958 | /** |
| 959 | * scsi_try_bus_reset - ask host to perform a bus reset |
| 960 | * @scmd: SCSI cmd to send bus reset. |
| 961 | **/ |
| 962 | static int scsi_try_bus_reset(struct scsi_cmnd *scmd) |
| 963 | { |
| 964 | unsigned long flags; |
| 965 | int rtn; |
| 966 | |
| 967 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", |
| 968 | __FUNCTION__)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | |
| 970 | if (!scmd->device->host->hostt->eh_bus_reset_handler) |
| 971 | return FAILED; |
| 972 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 973 | rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 974 | |
| 975 | if (rtn == SUCCESS) { |
| 976 | if (!scmd->device->host->hostt->skip_settle_delay) |
| 977 | ssleep(BUS_RESET_SETTLE_TIME); |
| 978 | spin_lock_irqsave(scmd->device->host->host_lock, flags); |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 979 | scsi_report_bus_reset(scmd->device->host, |
| 980 | scmd_channel(scmd)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 981 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
| 982 | } |
| 983 | |
| 984 | return rtn; |
| 985 | } |
| 986 | |
| 987 | /** |
| 988 | * scsi_try_host_reset - ask host adapter to reset itself |
| 989 | * @scmd: SCSI cmd to send hsot reset. |
| 990 | **/ |
| 991 | static int scsi_try_host_reset(struct scsi_cmnd *scmd) |
| 992 | { |
| 993 | unsigned long flags; |
| 994 | int rtn; |
| 995 | |
| 996 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", |
| 997 | __FUNCTION__)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | |
| 999 | if (!scmd->device->host->hostt->eh_host_reset_handler) |
| 1000 | return FAILED; |
| 1001 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1003 | |
| 1004 | if (rtn == SUCCESS) { |
| 1005 | if (!scmd->device->host->hostt->skip_settle_delay) |
| 1006 | ssleep(HOST_RESET_SETTLE_TIME); |
| 1007 | spin_lock_irqsave(scmd->device->host->host_lock, flags); |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1008 | scsi_report_bus_reset(scmd->device->host, |
| 1009 | scmd_channel(scmd)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1010 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
| 1011 | } |
| 1012 | |
| 1013 | return rtn; |
| 1014 | } |
| 1015 | |
| 1016 | /** |
| 1017 | * scsi_eh_bus_reset - send a bus reset |
| 1018 | * @shost: scsi host being recovered. |
| 1019 | * @eh_done_q: list_head for processed commands. |
| 1020 | **/ |
| 1021 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, |
| 1022 | struct list_head *work_q, |
| 1023 | struct list_head *done_q) |
| 1024 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1025 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | unsigned int channel; |
| 1027 | int rtn; |
| 1028 | |
| 1029 | /* |
| 1030 | * we really want to loop over the various channels, and do this on |
| 1031 | * a channel by channel basis. we should also check to see if any |
| 1032 | * of the failed commands are on soft_reset devices, and if so, skip |
| 1033 | * the reset. |
| 1034 | */ |
| 1035 | |
| 1036 | for (channel = 0; channel <= shost->max_channel; channel++) { |
| 1037 | chan_scmd = NULL; |
| 1038 | list_for_each_entry(scmd, work_q, eh_entry) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1039 | if (channel == scmd_channel(scmd)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | chan_scmd = scmd; |
| 1041 | break; |
| 1042 | /* |
| 1043 | * FIXME add back in some support for |
| 1044 | * soft_reset devices. |
| 1045 | */ |
| 1046 | } |
| 1047 | } |
| 1048 | |
| 1049 | if (!chan_scmd) |
| 1050 | continue; |
| 1051 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" |
| 1052 | " %d\n", current->comm, |
| 1053 | channel)); |
| 1054 | rtn = scsi_try_bus_reset(chan_scmd); |
| 1055 | if (rtn == SUCCESS) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1056 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1057 | if (channel == scmd_channel(scmd)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | if (!scsi_device_online(scmd->device) || |
| 1059 | !scsi_eh_tur(scmd)) |
| 1060 | scsi_eh_finish_cmd(scmd, |
| 1061 | done_q); |
| 1062 | } |
| 1063 | } else { |
| 1064 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" |
| 1065 | " failed chan: %d\n", |
| 1066 | current->comm, |
| 1067 | channel)); |
| 1068 | } |
| 1069 | } |
| 1070 | return list_empty(work_q); |
| 1071 | } |
| 1072 | |
| 1073 | /** |
| 1074 | * scsi_eh_host_reset - send a host reset |
| 1075 | * @work_q: list_head for processed commands. |
| 1076 | * @done_q: list_head for processed commands. |
| 1077 | **/ |
| 1078 | static int scsi_eh_host_reset(struct list_head *work_q, |
| 1079 | struct list_head *done_q) |
| 1080 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1081 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | int rtn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1083 | |
| 1084 | if (!list_empty(work_q)) { |
| 1085 | scmd = list_entry(work_q->next, |
| 1086 | struct scsi_cmnd, eh_entry); |
| 1087 | |
| 1088 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" |
| 1089 | , current->comm)); |
| 1090 | |
| 1091 | rtn = scsi_try_host_reset(scmd); |
| 1092 | if (rtn == SUCCESS) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1093 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1094 | if (!scsi_device_online(scmd->device) || |
| 1095 | (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || |
| 1096 | !scsi_eh_tur(scmd)) |
| 1097 | scsi_eh_finish_cmd(scmd, done_q); |
| 1098 | } |
| 1099 | } else { |
| 1100 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" |
| 1101 | " failed\n", |
| 1102 | current->comm)); |
| 1103 | } |
| 1104 | } |
| 1105 | return list_empty(work_q); |
| 1106 | } |
| 1107 | |
| 1108 | /** |
| 1109 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover |
| 1110 | * @work_q: list_head for processed commands. |
| 1111 | * @done_q: list_head for processed commands. |
| 1112 | * |
| 1113 | **/ |
| 1114 | static void scsi_eh_offline_sdevs(struct list_head *work_q, |
| 1115 | struct list_head *done_q) |
| 1116 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1117 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1118 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1119 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
James Bottomley | 9ccfc75 | 2005-10-02 11:45:08 -0500 | [diff] [blame] | 1120 | sdev_printk(KERN_INFO, scmd->device, |
| 1121 | "scsi: Device offlined - not" |
| 1122 | " ready after error recovery\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | scsi_device_set_state(scmd->device, SDEV_OFFLINE); |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 1124 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1125 | /* |
| 1126 | * FIXME: Handle lost cmds. |
| 1127 | */ |
| 1128 | } |
| 1129 | scsi_eh_finish_cmd(scmd, done_q); |
| 1130 | } |
| 1131 | return; |
| 1132 | } |
| 1133 | |
| 1134 | /** |
| 1135 | * scsi_decide_disposition - Disposition a cmd on return from LLD. |
| 1136 | * @scmd: SCSI cmd to examine. |
| 1137 | * |
| 1138 | * Notes: |
| 1139 | * This is *only* called when we are examining the status after sending |
| 1140 | * out the actual data command. any commands that are queued for error |
| 1141 | * recovery (e.g. test_unit_ready) do *not* come through here. |
| 1142 | * |
| 1143 | * When this routine returns failed, it means the error handler thread |
| 1144 | * is woken. In cases where the error code indicates an error that |
| 1145 | * doesn't require the error handler read (i.e. we don't need to |
| 1146 | * abort/reset), this function should return SUCCESS. |
| 1147 | **/ |
| 1148 | int scsi_decide_disposition(struct scsi_cmnd *scmd) |
| 1149 | { |
| 1150 | int rtn; |
| 1151 | |
| 1152 | /* |
| 1153 | * if the device is offline, then we clearly just pass the result back |
| 1154 | * up to the top level. |
| 1155 | */ |
| 1156 | if (!scsi_device_online(scmd->device)) { |
| 1157 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" |
| 1158 | " as SUCCESS\n", |
| 1159 | __FUNCTION__)); |
| 1160 | return SUCCESS; |
| 1161 | } |
| 1162 | |
| 1163 | /* |
| 1164 | * first check the host byte, to see if there is anything in there |
| 1165 | * that would indicate what we need to do. |
| 1166 | */ |
| 1167 | switch (host_byte(scmd->result)) { |
| 1168 | case DID_PASSTHROUGH: |
| 1169 | /* |
| 1170 | * no matter what, pass this through to the upper layer. |
| 1171 | * nuke this special code so that it looks like we are saying |
| 1172 | * did_ok. |
| 1173 | */ |
| 1174 | scmd->result &= 0xff00ffff; |
| 1175 | return SUCCESS; |
| 1176 | case DID_OK: |
| 1177 | /* |
| 1178 | * looks good. drop through, and check the next byte. |
| 1179 | */ |
| 1180 | break; |
| 1181 | case DID_NO_CONNECT: |
| 1182 | case DID_BAD_TARGET: |
| 1183 | case DID_ABORT: |
| 1184 | /* |
| 1185 | * note - this means that we just report the status back |
| 1186 | * to the top level driver, not that we actually think |
| 1187 | * that it indicates SUCCESS. |
| 1188 | */ |
| 1189 | return SUCCESS; |
| 1190 | /* |
| 1191 | * when the low level driver returns did_soft_error, |
| 1192 | * it is responsible for keeping an internal retry counter |
| 1193 | * in order to avoid endless loops (db) |
| 1194 | * |
| 1195 | * actually this is a bug in this function here. we should |
| 1196 | * be mindful of the maximum number of retries specified |
| 1197 | * and not get stuck in a loop. |
| 1198 | */ |
| 1199 | case DID_SOFT_ERROR: |
| 1200 | goto maybe_retry; |
| 1201 | case DID_IMM_RETRY: |
| 1202 | return NEEDS_RETRY; |
| 1203 | |
| bf34191 | 2005-04-12 17:49:09 -0500 | [diff] [blame] | 1204 | case DID_REQUEUE: |
| 1205 | return ADD_TO_MLQUEUE; |
| 1206 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1207 | case DID_ERROR: |
| 1208 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && |
| 1209 | status_byte(scmd->result) == RESERVATION_CONFLICT) |
| 1210 | /* |
| 1211 | * execute reservation conflict processing code |
| 1212 | * lower down |
| 1213 | */ |
| 1214 | break; |
| 1215 | /* fallthrough */ |
| 1216 | |
| 1217 | case DID_BUS_BUSY: |
| 1218 | case DID_PARITY: |
| 1219 | goto maybe_retry; |
| 1220 | case DID_TIME_OUT: |
| 1221 | /* |
| 1222 | * when we scan the bus, we get timeout messages for |
| 1223 | * these commands if there is no device available. |
| 1224 | * other hosts report did_no_connect for the same thing. |
| 1225 | */ |
| 1226 | if ((scmd->cmnd[0] == TEST_UNIT_READY || |
| 1227 | scmd->cmnd[0] == INQUIRY)) { |
| 1228 | return SUCCESS; |
| 1229 | } else { |
| 1230 | return FAILED; |
| 1231 | } |
| 1232 | case DID_RESET: |
| 1233 | return SUCCESS; |
| 1234 | default: |
| 1235 | return FAILED; |
| 1236 | } |
| 1237 | |
| 1238 | /* |
| 1239 | * next, check the message byte. |
| 1240 | */ |
| 1241 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) |
| 1242 | return FAILED; |
| 1243 | |
| 1244 | /* |
| 1245 | * check the status byte to see if this indicates anything special. |
| 1246 | */ |
| 1247 | switch (status_byte(scmd->result)) { |
| 1248 | case QUEUE_FULL: |
| 1249 | /* |
| 1250 | * the case of trying to send too many commands to a |
| 1251 | * tagged queueing device. |
| 1252 | */ |
| 1253 | case BUSY: |
| 1254 | /* |
| 1255 | * device can't talk to us at the moment. Should only |
| 1256 | * occur (SAM-3) when the task queue is empty, so will cause |
| 1257 | * the empty queue handling to trigger a stall in the |
| 1258 | * device. |
| 1259 | */ |
| 1260 | return ADD_TO_MLQUEUE; |
| 1261 | case GOOD: |
| 1262 | case COMMAND_TERMINATED: |
| 1263 | case TASK_ABORTED: |
| 1264 | return SUCCESS; |
| 1265 | case CHECK_CONDITION: |
| 1266 | rtn = scsi_check_sense(scmd); |
| 1267 | if (rtn == NEEDS_RETRY) |
| 1268 | goto maybe_retry; |
| 1269 | /* if rtn == FAILED, we have no sense information; |
| 1270 | * returning FAILED will wake the error handler thread |
| 1271 | * to collect the sense and redo the decide |
| 1272 | * disposition */ |
| 1273 | return rtn; |
| 1274 | case CONDITION_GOOD: |
| 1275 | case INTERMEDIATE_GOOD: |
| 1276 | case INTERMEDIATE_C_GOOD: |
| 1277 | case ACA_ACTIVE: |
| 1278 | /* |
| 1279 | * who knows? FIXME(eric) |
| 1280 | */ |
| 1281 | return SUCCESS; |
| 1282 | |
| 1283 | case RESERVATION_CONFLICT: |
James Bottomley | 9ccfc75 | 2005-10-02 11:45:08 -0500 | [diff] [blame] | 1284 | sdev_printk(KERN_INFO, scmd->device, |
| 1285 | "reservation conflict\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | return SUCCESS; /* causes immediate i/o error */ |
| 1287 | default: |
| 1288 | return FAILED; |
| 1289 | } |
| 1290 | return FAILED; |
| 1291 | |
| 1292 | maybe_retry: |
| 1293 | |
| 1294 | /* we requeue for retry because the error was retryable, and |
| 1295 | * the request was not marked fast fail. Note that above, |
| 1296 | * even if the request is marked fast fail, we still requeue |
| 1297 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ |
Brian King | 8884efa | 2006-02-24 17:10:04 -0600 | [diff] [blame] | 1298 | if ((++scmd->retries) <= scmd->allowed |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1299 | && !blk_noretry_request(scmd->request)) { |
| 1300 | return NEEDS_RETRY; |
| 1301 | } else { |
| 1302 | /* |
| 1303 | * no more retries - report this one back to upper level. |
| 1304 | */ |
| 1305 | return SUCCESS; |
| 1306 | } |
| 1307 | } |
| 1308 | |
| 1309 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | * scsi_eh_lock_door - Prevent medium removal for the specified device |
| 1311 | * @sdev: SCSI device to prevent medium removal |
| 1312 | * |
| 1313 | * Locking: |
| 1314 | * We must be called from process context; scsi_allocate_request() |
| 1315 | * may sleep. |
| 1316 | * |
| 1317 | * Notes: |
| 1318 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the |
| 1319 | * head of the devices request queue, and continue. |
| 1320 | * |
| 1321 | * Bugs: |
| 1322 | * scsi_allocate_request() may sleep waiting for existing requests to |
| 1323 | * be processed. However, since we haven't kicked off any request |
| 1324 | * processing for this host, this may deadlock. |
| 1325 | * |
| 1326 | * If scsi_allocate_request() fails for what ever reason, we |
| 1327 | * completely forget to lock the door. |
| 1328 | **/ |
| 1329 | static void scsi_eh_lock_door(struct scsi_device *sdev) |
| 1330 | { |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 1331 | unsigned char cmnd[MAX_COMMAND_SIZE]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1332 | |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 1333 | cmnd[0] = ALLOW_MEDIUM_REMOVAL; |
| 1334 | cmnd[1] = 0; |
| 1335 | cmnd[2] = 0; |
| 1336 | cmnd[3] = 0; |
| 1337 | cmnd[4] = SCSI_REMOVAL_PREVENT; |
| 1338 | cmnd[5] = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1339 | |
brking@us.ibm.com | bb1d107 | 2006-01-23 15:03:22 -0600 | [diff] [blame] | 1340 | scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ, |
Mike Christie | 6e68af6 | 2005-11-11 05:30:27 -0600 | [diff] [blame] | 1341 | 5, NULL, NULL, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1342 | } |
| 1343 | |
| 1344 | |
| 1345 | /** |
| 1346 | * scsi_restart_operations - restart io operations to the specified host. |
| 1347 | * @shost: Host we are restarting. |
| 1348 | * |
| 1349 | * Notes: |
| 1350 | * When we entered the error handler, we blocked all further i/o to |
| 1351 | * this device. we need to 'reverse' this process. |
| 1352 | **/ |
| 1353 | static void scsi_restart_operations(struct Scsi_Host *shost) |
| 1354 | { |
| 1355 | struct scsi_device *sdev; |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 1356 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | |
| 1358 | /* |
| 1359 | * If the door was locked, we need to insert a door lock request |
| 1360 | * onto the head of the SCSI request queue for the device. There |
| 1361 | * is no point trying to lock the door of an off-line device. |
| 1362 | */ |
| 1363 | shost_for_each_device(sdev, shost) { |
| 1364 | if (scsi_device_online(sdev) && sdev->locked) |
| 1365 | scsi_eh_lock_door(sdev); |
| 1366 | } |
| 1367 | |
| 1368 | /* |
| 1369 | * next free up anything directly waiting upon the host. this |
| 1370 | * will be requests for character device operations, and also for |
| 1371 | * ioctls to queued block devices. |
| 1372 | */ |
| 1373 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", |
| 1374 | __FUNCTION__)); |
| 1375 | |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 1376 | spin_lock_irqsave(shost->host_lock, flags); |
| 1377 | if (scsi_host_set_state(shost, SHOST_RUNNING)) |
| 1378 | if (scsi_host_set_state(shost, SHOST_CANCEL)) |
| 1379 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); |
| 1380 | spin_unlock_irqrestore(shost->host_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1381 | |
| 1382 | wake_up(&shost->host_wait); |
| 1383 | |
| 1384 | /* |
| 1385 | * finally we need to re-initiate requests that may be pending. we will |
| 1386 | * have had everything blocked while error handling is taking place, and |
| 1387 | * now that error recovery is done, we will need to ensure that these |
| 1388 | * requests are started. |
| 1389 | */ |
| 1390 | scsi_run_host_queues(shost); |
| 1391 | } |
| 1392 | |
| 1393 | /** |
| 1394 | * scsi_eh_ready_devs - check device ready state and recover if not. |
| 1395 | * @shost: host to be recovered. |
| 1396 | * @eh_done_q: list_head for processed commands. |
| 1397 | * |
| 1398 | **/ |
| 1399 | static void scsi_eh_ready_devs(struct Scsi_Host *shost, |
| 1400 | struct list_head *work_q, |
| 1401 | struct list_head *done_q) |
| 1402 | { |
| 1403 | if (!scsi_eh_stu(shost, work_q, done_q)) |
| 1404 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) |
| 1405 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) |
| 1406 | if (!scsi_eh_host_reset(work_q, done_q)) |
| 1407 | scsi_eh_offline_sdevs(work_q, done_q); |
| 1408 | } |
| 1409 | |
| 1410 | /** |
| 1411 | * scsi_eh_flush_done_q - finish processed commands or retry them. |
| 1412 | * @done_q: list_head of processed commands. |
| 1413 | * |
| 1414 | **/ |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 1415 | void scsi_eh_flush_done_q(struct list_head *done_q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1416 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1417 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1418 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1419 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
| 1420 | list_del_init(&scmd->eh_entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | if (scsi_device_online(scmd->device) && |
| 1422 | !blk_noretry_request(scmd->request) && |
Brian King | 8884efa | 2006-02-24 17:10:04 -0600 | [diff] [blame] | 1423 | (++scmd->retries <= scmd->allowed)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1424 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" |
| 1425 | " retry cmd: %p\n", |
| 1426 | current->comm, |
| 1427 | scmd)); |
| 1428 | scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); |
| 1429 | } else { |
Patrick Mansfield | 793698c | 2005-05-16 17:42:15 -0700 | [diff] [blame] | 1430 | /* |
| 1431 | * If just we got sense for the device (called |
| 1432 | * scsi_eh_get_sense), scmd->result is already |
| 1433 | * set, do not set DRIVER_TIMEOUT. |
| 1434 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1435 | if (!scmd->result) |
| 1436 | scmd->result |= (DRIVER_TIMEOUT << 24); |
| 1437 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" |
| 1438 | " cmd: %p\n", |
| 1439 | current->comm, scmd)); |
| 1440 | scsi_finish_command(scmd); |
| 1441 | } |
| 1442 | } |
| 1443 | } |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 1444 | EXPORT_SYMBOL(scsi_eh_flush_done_q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | |
| 1446 | /** |
| 1447 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. |
| 1448 | * @shost: Host to unjam. |
| 1449 | * |
| 1450 | * Notes: |
| 1451 | * When we come in here, we *know* that all commands on the bus have |
| 1452 | * either completed, failed or timed out. we also know that no further |
| 1453 | * commands are being sent to the host, so things are relatively quiet |
| 1454 | * and we have freedom to fiddle with things as we wish. |
| 1455 | * |
| 1456 | * This is only the *default* implementation. it is possible for |
| 1457 | * individual drivers to supply their own version of this function, and |
| 1458 | * if the maintainer wishes to do this, it is strongly suggested that |
| 1459 | * this function be taken as a template and modified. this function |
| 1460 | * was designed to correctly handle problems for about 95% of the |
| 1461 | * different cases out there, and it should always provide at least a |
| 1462 | * reasonable amount of error recovery. |
| 1463 | * |
| 1464 | * Any command marked 'failed' or 'timeout' must eventually have |
| 1465 | * scsi_finish_cmd() called for it. we do all of the retry stuff |
| 1466 | * here, so when we restart the host after we return it should have an |
| 1467 | * empty queue. |
| 1468 | **/ |
| 1469 | static void scsi_unjam_host(struct Scsi_Host *shost) |
| 1470 | { |
| 1471 | unsigned long flags; |
| 1472 | LIST_HEAD(eh_work_q); |
| 1473 | LIST_HEAD(eh_done_q); |
| 1474 | |
| 1475 | spin_lock_irqsave(shost->host_lock, flags); |
| 1476 | list_splice_init(&shost->eh_cmd_q, &eh_work_q); |
| 1477 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 1478 | |
| 1479 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); |
| 1480 | |
| 1481 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) |
| 1482 | if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) |
| 1483 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); |
| 1484 | |
| 1485 | scsi_eh_flush_done_q(&eh_done_q); |
| 1486 | } |
| 1487 | |
| 1488 | /** |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1489 | * scsi_error_handler - SCSI error handler thread |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | * @data: Host for which we are running. |
| 1491 | * |
| 1492 | * Notes: |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1493 | * This is the main error handling loop. This is run as a kernel thread |
| 1494 | * for every SCSI host and handles all error handling activity. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | **/ |
| 1496 | int scsi_error_handler(void *data) |
| 1497 | { |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1498 | struct Scsi_Host *shost = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1499 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | current->flags |= PF_NOFREEZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | |
| 1502 | /* |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1503 | * We use TASK_INTERRUPTIBLE so that the thread is not |
| 1504 | * counted against the load average as a running process. |
| 1505 | * We never actually get interrupted because kthread_run |
| 1506 | * disables singal delivery for the created thread. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1507 | */ |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1508 | set_current_state(TASK_INTERRUPTIBLE); |
| 1509 | while (!kthread_should_stop()) { |
Tejun Heo | ee7863b | 2006-05-15 20:57:20 +0900 | [diff] [blame] | 1510 | if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1511 | shost->host_failed != shost->host_busy) { |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1512 | SCSI_LOG_ERROR_RECOVERY(1, |
| 1513 | printk("Error handler scsi_eh_%d sleeping\n", |
| 1514 | shost->host_no)); |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1515 | schedule(); |
| 1516 | set_current_state(TASK_INTERRUPTIBLE); |
| 1517 | continue; |
| 1518 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1520 | __set_current_state(TASK_RUNNING); |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1521 | SCSI_LOG_ERROR_RECOVERY(1, |
| 1522 | printk("Error handler scsi_eh_%d waking up\n", |
| 1523 | shost->host_no)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | /* |
| 1526 | * We have a host that is failing for some reason. Figure out |
| 1527 | * what we need to do to get it up and online again (if we can). |
| 1528 | * If we fail, we end up taking the thing offline. |
| 1529 | */ |
Christoph Hellwig | 9227c33 | 2006-04-01 19:21:04 +0200 | [diff] [blame] | 1530 | if (shost->transportt->eh_strategy_handler) |
| 1531 | shost->transportt->eh_strategy_handler(shost); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1532 | else |
| 1533 | scsi_unjam_host(shost); |
| 1534 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1535 | /* |
| 1536 | * Note - if the above fails completely, the action is to take |
| 1537 | * individual devices offline and flush the queue of any |
| 1538 | * outstanding requests that may have been pending. When we |
| 1539 | * restart, we restart any I/O to any other devices on the bus |
| 1540 | * which are still online. |
| 1541 | */ |
| 1542 | scsi_restart_operations(shost); |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1543 | set_current_state(TASK_INTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1544 | } |
Steven Rostedt | 461a0ff | 2005-10-19 08:22:13 -0400 | [diff] [blame] | 1545 | __set_current_state(TASK_RUNNING); |
| 1546 | |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1547 | SCSI_LOG_ERROR_RECOVERY(1, |
| 1548 | printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1549 | shost->ehandler = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1550 | return 0; |
| 1551 | } |
| 1552 | |
| 1553 | /* |
| 1554 | * Function: scsi_report_bus_reset() |
| 1555 | * |
| 1556 | * Purpose: Utility function used by low-level drivers to report that |
| 1557 | * they have observed a bus reset on the bus being handled. |
| 1558 | * |
| 1559 | * Arguments: shost - Host in question |
| 1560 | * channel - channel on which reset was observed. |
| 1561 | * |
| 1562 | * Returns: Nothing |
| 1563 | * |
| 1564 | * Lock status: Host lock must be held. |
| 1565 | * |
| 1566 | * Notes: This only needs to be called if the reset is one which |
| 1567 | * originates from an unknown location. Resets originated |
| 1568 | * by the mid-level itself don't need to call this, but there |
| 1569 | * should be no harm. |
| 1570 | * |
| 1571 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 1572 | * is properly treated. |
| 1573 | */ |
| 1574 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) |
| 1575 | { |
| 1576 | struct scsi_device *sdev; |
| 1577 | |
| 1578 | __shost_for_each_device(sdev, shost) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1579 | if (channel == sdev_channel(sdev)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1580 | sdev->was_reset = 1; |
| 1581 | sdev->expecting_cc_ua = 1; |
| 1582 | } |
| 1583 | } |
| 1584 | } |
| 1585 | EXPORT_SYMBOL(scsi_report_bus_reset); |
| 1586 | |
| 1587 | /* |
| 1588 | * Function: scsi_report_device_reset() |
| 1589 | * |
| 1590 | * Purpose: Utility function used by low-level drivers to report that |
| 1591 | * they have observed a device reset on the device being handled. |
| 1592 | * |
| 1593 | * Arguments: shost - Host in question |
| 1594 | * channel - channel on which reset was observed |
| 1595 | * target - target on which reset was observed |
| 1596 | * |
| 1597 | * Returns: Nothing |
| 1598 | * |
| 1599 | * Lock status: Host lock must be held |
| 1600 | * |
| 1601 | * Notes: This only needs to be called if the reset is one which |
| 1602 | * originates from an unknown location. Resets originated |
| 1603 | * by the mid-level itself don't need to call this, but there |
| 1604 | * should be no harm. |
| 1605 | * |
| 1606 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 1607 | * is properly treated. |
| 1608 | */ |
| 1609 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) |
| 1610 | { |
| 1611 | struct scsi_device *sdev; |
| 1612 | |
| 1613 | __shost_for_each_device(sdev, shost) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1614 | if (channel == sdev_channel(sdev) && |
| 1615 | target == sdev_id(sdev)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1616 | sdev->was_reset = 1; |
| 1617 | sdev->expecting_cc_ua = 1; |
| 1618 | } |
| 1619 | } |
| 1620 | } |
| 1621 | EXPORT_SYMBOL(scsi_report_device_reset); |
| 1622 | |
| 1623 | static void |
| 1624 | scsi_reset_provider_done_command(struct scsi_cmnd *scmd) |
| 1625 | { |
| 1626 | } |
| 1627 | |
| 1628 | /* |
| 1629 | * Function: scsi_reset_provider |
| 1630 | * |
| 1631 | * Purpose: Send requested reset to a bus or device at any phase. |
| 1632 | * |
| 1633 | * Arguments: device - device to send reset to |
| 1634 | * flag - reset type (see scsi.h) |
| 1635 | * |
| 1636 | * Returns: SUCCESS/FAILURE. |
| 1637 | * |
| 1638 | * Notes: This is used by the SCSI Generic driver to provide |
| 1639 | * Bus/Device reset capability. |
| 1640 | */ |
| 1641 | int |
| 1642 | scsi_reset_provider(struct scsi_device *dev, int flag) |
| 1643 | { |
| 1644 | struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1645 | struct Scsi_Host *shost = dev->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1646 | struct request req; |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1647 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | int rtn; |
| 1649 | |
| 1650 | scmd->request = &req; |
| 1651 | memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout)); |
Christoph Hellwig | b4edcbc | 2005-06-19 13:40:52 +0200 | [diff] [blame] | 1652 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd)); |
| 1654 | |
| 1655 | scmd->scsi_done = scsi_reset_provider_done_command; |
| 1656 | scmd->done = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | scmd->request_buffer = NULL; |
| 1658 | scmd->request_bufflen = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1659 | |
| 1660 | scmd->cmd_len = 0; |
| 1661 | |
| 1662 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1663 | |
| 1664 | init_timer(&scmd->eh_timeout); |
| 1665 | |
| 1666 | /* |
| 1667 | * Sometimes the command can get back into the timer chain, |
| 1668 | * so use the pid as an identifier. |
| 1669 | */ |
| 1670 | scmd->pid = 0; |
| 1671 | |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1672 | spin_lock_irqsave(shost->host_lock, flags); |
| 1673 | shost->tmf_in_progress = 1; |
| 1674 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 1675 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1676 | switch (flag) { |
| 1677 | case SCSI_TRY_RESET_DEVICE: |
| 1678 | rtn = scsi_try_bus_device_reset(scmd); |
| 1679 | if (rtn == SUCCESS) |
| 1680 | break; |
| 1681 | /* FALLTHROUGH */ |
| 1682 | case SCSI_TRY_RESET_BUS: |
| 1683 | rtn = scsi_try_bus_reset(scmd); |
| 1684 | if (rtn == SUCCESS) |
| 1685 | break; |
| 1686 | /* FALLTHROUGH */ |
| 1687 | case SCSI_TRY_RESET_HOST: |
| 1688 | rtn = scsi_try_host_reset(scmd); |
| 1689 | break; |
| 1690 | default: |
| 1691 | rtn = FAILED; |
| 1692 | } |
| 1693 | |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1694 | spin_lock_irqsave(shost->host_lock, flags); |
| 1695 | shost->tmf_in_progress = 0; |
| 1696 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 1697 | |
| 1698 | /* |
| 1699 | * be sure to wake up anyone who was sleeping or had their queue |
| 1700 | * suspended while we performed the TMF. |
| 1701 | */ |
| 1702 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1703 | printk("%s: waking up host to restart after TMF\n", |
| 1704 | __FUNCTION__)); |
| 1705 | |
| 1706 | wake_up(&shost->host_wait); |
| 1707 | |
| 1708 | scsi_run_host_queues(shost); |
| 1709 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1710 | scsi_next_command(scmd); |
| 1711 | return rtn; |
| 1712 | } |
| 1713 | EXPORT_SYMBOL(scsi_reset_provider); |
| 1714 | |
| 1715 | /** |
| 1716 | * scsi_normalize_sense - normalize main elements from either fixed or |
| 1717 | * descriptor sense data format into a common format. |
| 1718 | * |
| 1719 | * @sense_buffer: byte array containing sense data returned by device |
| 1720 | * @sb_len: number of valid bytes in sense_buffer |
| 1721 | * @sshdr: pointer to instance of structure that common |
| 1722 | * elements are written to. |
| 1723 | * |
| 1724 | * Notes: |
| 1725 | * The "main elements" from sense data are: response_code, sense_key, |
| 1726 | * asc, ascq and additional_length (only for descriptor format). |
| 1727 | * |
| 1728 | * Typically this function can be called after a device has |
| 1729 | * responded to a SCSI command with the CHECK_CONDITION status. |
| 1730 | * |
| 1731 | * Return value: |
| 1732 | * 1 if valid sense data information found, else 0; |
| 1733 | **/ |
| 1734 | int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, |
| 1735 | struct scsi_sense_hdr *sshdr) |
| 1736 | { |
James Bottomley | 33aa687 | 2005-08-28 11:31:14 -0500 | [diff] [blame] | 1737 | if (!sense_buffer || !sb_len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | return 0; |
| 1739 | |
| 1740 | memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); |
| 1741 | |
| 1742 | sshdr->response_code = (sense_buffer[0] & 0x7f); |
James Bottomley | 33aa687 | 2005-08-28 11:31:14 -0500 | [diff] [blame] | 1743 | |
| 1744 | if (!scsi_sense_valid(sshdr)) |
| 1745 | return 0; |
| 1746 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1747 | if (sshdr->response_code >= 0x72) { |
| 1748 | /* |
| 1749 | * descriptor format |
| 1750 | */ |
| 1751 | if (sb_len > 1) |
| 1752 | sshdr->sense_key = (sense_buffer[1] & 0xf); |
| 1753 | if (sb_len > 2) |
| 1754 | sshdr->asc = sense_buffer[2]; |
| 1755 | if (sb_len > 3) |
| 1756 | sshdr->ascq = sense_buffer[3]; |
| 1757 | if (sb_len > 7) |
| 1758 | sshdr->additional_length = sense_buffer[7]; |
| 1759 | } else { |
| 1760 | /* |
| 1761 | * fixed format |
| 1762 | */ |
| 1763 | if (sb_len > 2) |
| 1764 | sshdr->sense_key = (sense_buffer[2] & 0xf); |
| 1765 | if (sb_len > 7) { |
| 1766 | sb_len = (sb_len < (sense_buffer[7] + 8)) ? |
| 1767 | sb_len : (sense_buffer[7] + 8); |
| 1768 | if (sb_len > 12) |
| 1769 | sshdr->asc = sense_buffer[12]; |
| 1770 | if (sb_len > 13) |
| 1771 | sshdr->ascq = sense_buffer[13]; |
| 1772 | } |
| 1773 | } |
| 1774 | |
| 1775 | return 1; |
| 1776 | } |
| 1777 | EXPORT_SYMBOL(scsi_normalize_sense); |
| 1778 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1779 | int scsi_command_normalize_sense(struct scsi_cmnd *cmd, |
| 1780 | struct scsi_sense_hdr *sshdr) |
| 1781 | { |
| 1782 | return scsi_normalize_sense(cmd->sense_buffer, |
| 1783 | sizeof(cmd->sense_buffer), sshdr); |
| 1784 | } |
| 1785 | EXPORT_SYMBOL(scsi_command_normalize_sense); |
| 1786 | |
| 1787 | /** |
| 1788 | * scsi_sense_desc_find - search for a given descriptor type in |
| 1789 | * descriptor sense data format. |
| 1790 | * |
| 1791 | * @sense_buffer: byte array of descriptor format sense data |
| 1792 | * @sb_len: number of valid bytes in sense_buffer |
| 1793 | * @desc_type: value of descriptor type to find |
| 1794 | * (e.g. 0 -> information) |
| 1795 | * |
| 1796 | * Notes: |
| 1797 | * only valid when sense data is in descriptor format |
| 1798 | * |
| 1799 | * Return value: |
| 1800 | * pointer to start of (first) descriptor if found else NULL |
| 1801 | **/ |
| 1802 | const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, |
| 1803 | int desc_type) |
| 1804 | { |
| 1805 | int add_sen_len, add_len, desc_len, k; |
| 1806 | const u8 * descp; |
| 1807 | |
| 1808 | if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) |
| 1809 | return NULL; |
| 1810 | if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) |
| 1811 | return NULL; |
| 1812 | add_sen_len = (add_sen_len < (sb_len - 8)) ? |
| 1813 | add_sen_len : (sb_len - 8); |
| 1814 | descp = &sense_buffer[8]; |
| 1815 | for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { |
| 1816 | descp += desc_len; |
| 1817 | add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; |
| 1818 | desc_len = add_len + 2; |
| 1819 | if (descp[0] == desc_type) |
| 1820 | return descp; |
| 1821 | if (add_len < 0) // short descriptor ?? |
| 1822 | break; |
| 1823 | } |
| 1824 | return NULL; |
| 1825 | } |
| 1826 | EXPORT_SYMBOL(scsi_sense_desc_find); |
| 1827 | |
| 1828 | /** |
| 1829 | * scsi_get_sense_info_fld - attempts to get information field from |
| 1830 | * sense data (either fixed or descriptor format) |
| 1831 | * |
| 1832 | * @sense_buffer: byte array of sense data |
| 1833 | * @sb_len: number of valid bytes in sense_buffer |
| 1834 | * @info_out: pointer to 64 integer where 8 or 4 byte information |
| 1835 | * field will be placed if found. |
| 1836 | * |
| 1837 | * Return value: |
| 1838 | * 1 if information field found, 0 if not found. |
| 1839 | **/ |
| 1840 | int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, |
| 1841 | u64 * info_out) |
| 1842 | { |
| 1843 | int j; |
| 1844 | const u8 * ucp; |
| 1845 | u64 ull; |
| 1846 | |
| 1847 | if (sb_len < 7) |
| 1848 | return 0; |
| 1849 | switch (sense_buffer[0] & 0x7f) { |
| 1850 | case 0x70: |
| 1851 | case 0x71: |
| 1852 | if (sense_buffer[0] & 0x80) { |
| 1853 | *info_out = (sense_buffer[3] << 24) + |
| 1854 | (sense_buffer[4] << 16) + |
| 1855 | (sense_buffer[5] << 8) + sense_buffer[6]; |
| 1856 | return 1; |
| 1857 | } else |
| 1858 | return 0; |
| 1859 | case 0x72: |
| 1860 | case 0x73: |
| 1861 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, |
| 1862 | 0 /* info desc */); |
| 1863 | if (ucp && (0xa == ucp[1])) { |
| 1864 | ull = 0; |
| 1865 | for (j = 0; j < 8; ++j) { |
| 1866 | if (j > 0) |
| 1867 | ull <<= 8; |
| 1868 | ull |= ucp[4 + j]; |
| 1869 | } |
| 1870 | *info_out = ull; |
| 1871 | return 1; |
| 1872 | } else |
| 1873 | return 0; |
| 1874 | default: |
| 1875 | return 0; |
| 1876 | } |
| 1877 | } |
| 1878 | EXPORT_SYMBOL(scsi_get_sense_info_fld); |