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> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include <linux/kernel.h> |
Rafael J. Wysocki | 8314418 | 2007-07-17 04:03:35 -0700 | [diff] [blame] | 22 | #include <linux/freezer.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" |
Adrian Bunk | 79ee830 | 2007-08-10 14:50:42 -0700 | [diff] [blame] | 39 | #include "scsi_transport_api.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | |
| 41 | #define SENSE_TIMEOUT (10*HZ) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 65 | */ |
Tejun Heo | f8bbfc2 | 2006-05-19 21:07:05 +0900 | [diff] [blame] | 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. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 89 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | * scsi_times_out - Timeout function for normal scsi commands. |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 116 | * @req: request that is timing out. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | * |
| 118 | * Notes: |
| 119 | * We do not need to lock this. There is the potential for a race |
| 120 | * only in that the normal completion handling might run, but if the |
| 121 | * normal completion function determines that the timer has already |
| 122 | * fired, then it mustn't do anything. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 123 | */ |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 124 | enum blk_eh_timer_return scsi_times_out(struct request *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | { |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 126 | struct scsi_cmnd *scmd = req->special; |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 127 | enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; |
James Bottomley | 6c5f8ce | 2007-03-16 17:44:41 -0500 | [diff] [blame] | 128 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | scsi_log_completion(scmd, TIMEOUT_ERROR); |
| 130 | |
James Smart | c829c39 | 2006-03-13 08:28:57 -0500 | [diff] [blame] | 131 | if (scmd->device->host->transportt->eh_timed_out) |
Christoph Hellwig | fa99078 | 2008-11-05 15:03:15 +0100 | [diff] [blame] | 132 | rtn = scmd->device->host->transportt->eh_timed_out(scmd); |
James Bottomley | 6c5f8ce | 2007-03-16 17:44:41 -0500 | [diff] [blame] | 133 | else if (scmd->device->host->hostt->eh_timed_out) |
Christoph Hellwig | fa99078 | 2008-11-05 15:03:15 +0100 | [diff] [blame] | 134 | rtn = scmd->device->host->hostt->eh_timed_out(scmd); |
James Bottomley | 6c5f8ce | 2007-03-16 17:44:41 -0500 | [diff] [blame] | 135 | |
Christoph Hellwig | fa99078 | 2008-11-05 15:03:15 +0100 | [diff] [blame] | 136 | if (unlikely(rtn == BLK_EH_NOT_HANDLED && |
| 137 | !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) { |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 138 | scmd->result |= DID_TIME_OUT << 16; |
Christoph Hellwig | fa99078 | 2008-11-05 15:03:15 +0100 | [diff] [blame] | 139 | rtn = BLK_EH_HANDLED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | } |
Jens Axboe | 242f9dc | 2008-09-14 05:55:09 -0700 | [diff] [blame] | 141 | |
Christoph Hellwig | fa99078 | 2008-11-05 15:03:15 +0100 | [diff] [blame] | 142 | return rtn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 143 | } |
| 144 | |
| 145 | /** |
| 146 | * scsi_block_when_processing_errors - Prevent cmds from being queued. |
| 147 | * @sdev: Device on which we are performing recovery. |
| 148 | * |
| 149 | * Description: |
| 150 | * We block until the host is out of error recovery, and then check to |
| 151 | * see whether the host or the device is offline. |
| 152 | * |
| 153 | * Return value: |
| 154 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 155 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | int scsi_block_when_processing_errors(struct scsi_device *sdev) |
| 157 | { |
| 158 | int online; |
| 159 | |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 160 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
| 162 | online = scsi_device_online(sdev); |
| 163 | |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 164 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | online)); |
| 166 | |
| 167 | return online; |
| 168 | } |
| 169 | EXPORT_SYMBOL(scsi_block_when_processing_errors); |
| 170 | |
| 171 | #ifdef CONFIG_SCSI_LOGGING |
| 172 | /** |
| 173 | * scsi_eh_prt_fail_stats - Log info on failures. |
| 174 | * @shost: scsi host being recovered. |
| 175 | * @work_q: Queue of scsi cmds to process. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 176 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, |
| 178 | struct list_head *work_q) |
| 179 | { |
| 180 | struct scsi_cmnd *scmd; |
| 181 | struct scsi_device *sdev; |
| 182 | int total_failures = 0; |
| 183 | int cmd_failed = 0; |
| 184 | int cmd_cancel = 0; |
| 185 | int devices_failed = 0; |
| 186 | |
| 187 | shost_for_each_device(sdev, shost) { |
| 188 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 189 | if (scmd->device == sdev) { |
| 190 | ++total_failures; |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 191 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | ++cmd_cancel; |
| 193 | else |
| 194 | ++cmd_failed; |
| 195 | } |
| 196 | } |
| 197 | |
| 198 | if (cmd_cancel || cmd_failed) { |
| 199 | SCSI_LOG_ERROR_RECOVERY(3, |
James Bottomley | 9ccfc75 | 2005-10-02 11:45:08 -0500 | [diff] [blame] | 200 | sdev_printk(KERN_INFO, sdev, |
| 201 | "%s: cmds failed: %d, cancel: %d\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 202 | __func__, cmd_failed, |
James Bottomley | 9ccfc75 | 2005-10-02 11:45:08 -0500 | [diff] [blame] | 203 | cmd_cancel)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | cmd_cancel = 0; |
| 205 | cmd_failed = 0; |
| 206 | ++devices_failed; |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d" |
| 211 | " devices require eh work\n", |
| 212 | total_failures, devices_failed)); |
| 213 | } |
| 214 | #endif |
| 215 | |
| 216 | /** |
| 217 | * scsi_check_sense - Examine scsi cmd sense |
| 218 | * @scmd: Cmd to have sense checked. |
| 219 | * |
| 220 | * Return value: |
| 221 | * SUCCESS or FAILED or NEEDS_RETRY |
| 222 | * |
| 223 | * Notes: |
| 224 | * When a deferred error is detected the current command has |
| 225 | * not been executed and needs retrying. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 226 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | static int scsi_check_sense(struct scsi_cmnd *scmd) |
| 228 | { |
Chandra Seetharaman | a6a8d9f | 2008-05-01 14:49:46 -0700 | [diff] [blame] | 229 | struct scsi_device *sdev = scmd->device; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | struct scsi_sense_hdr sshdr; |
| 231 | |
| 232 | if (! scsi_command_normalize_sense(scmd, &sshdr)) |
| 233 | return FAILED; /* no valid sense data */ |
| 234 | |
| 235 | if (scsi_sense_is_deferred(&sshdr)) |
| 236 | return NEEDS_RETRY; |
| 237 | |
Chandra Seetharaman | a6a8d9f | 2008-05-01 14:49:46 -0700 | [diff] [blame] | 238 | if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh && |
| 239 | sdev->scsi_dh_data->scsi_dh->check_sense) { |
| 240 | int rc; |
| 241 | |
| 242 | rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr); |
| 243 | if (rc != SCSI_RETURN_NOT_HANDLED) |
| 244 | return rc; |
| 245 | /* handler does not care. Drop down to default handling */ |
| 246 | } |
| 247 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | /* |
| 249 | * Previous logic looked for FILEMARK, EOM or ILI which are |
| 250 | * mainly associated with tapes and returned SUCCESS. |
| 251 | */ |
| 252 | if (sshdr.response_code == 0x70) { |
| 253 | /* fixed format */ |
| 254 | if (scmd->sense_buffer[2] & 0xe0) |
| 255 | return SUCCESS; |
| 256 | } else { |
| 257 | /* |
| 258 | * descriptor format: look for "stream commands sense data |
| 259 | * descriptor" (see SSC-3). Assume single sense data |
| 260 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. |
| 261 | */ |
| 262 | if ((sshdr.additional_length > 3) && |
| 263 | (scmd->sense_buffer[8] == 0x4) && |
| 264 | (scmd->sense_buffer[11] & 0xe0)) |
| 265 | return SUCCESS; |
| 266 | } |
| 267 | |
| 268 | switch (sshdr.sense_key) { |
| 269 | case NO_SENSE: |
| 270 | return SUCCESS; |
| 271 | case RECOVERED_ERROR: |
| 272 | return /* soft_error */ SUCCESS; |
| 273 | |
| 274 | case ABORTED_COMMAND: |
Martin K. Petersen | 511e44f | 2008-07-17 04:28:33 -0400 | [diff] [blame] | 275 | if (sshdr.asc == 0x10) /* DIF */ |
| 276 | return SUCCESS; |
| 277 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | return NEEDS_RETRY; |
| 279 | case NOT_READY: |
| 280 | case UNIT_ATTENTION: |
| 281 | /* |
| 282 | * if we are expecting a cc/ua because of a bus reset that we |
| 283 | * performed, treat this just as a retry. otherwise this is |
| 284 | * information that we should pass up to the upper-level driver |
| 285 | * so that we can deal with it there. |
| 286 | */ |
| 287 | if (scmd->device->expecting_cc_ua) { |
| 288 | scmd->device->expecting_cc_ua = 0; |
| 289 | return NEEDS_RETRY; |
| 290 | } |
| 291 | /* |
| 292 | * if the device is in the process of becoming ready, we |
| 293 | * should retry. |
| 294 | */ |
| 295 | if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) |
| 296 | return NEEDS_RETRY; |
| 297 | /* |
| 298 | * if the device is not started, we need to wake |
| 299 | * the error handler to start the motor |
| 300 | */ |
| 301 | if (scmd->device->allow_restart && |
| 302 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) |
| 303 | return FAILED; |
| 304 | return SUCCESS; |
| 305 | |
| 306 | /* these three are not supported */ |
| 307 | case COPY_ABORTED: |
| 308 | case VOLUME_OVERFLOW: |
| 309 | case MISCOMPARE: |
| 310 | return SUCCESS; |
| 311 | |
| 312 | case MEDIUM_ERROR: |
Luben Tuikov | fd1b494 | 2006-11-29 19:45:23 -0800 | [diff] [blame] | 313 | if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ |
| 314 | sshdr.asc == 0x13 || /* AMNF DATA FIELD */ |
| 315 | sshdr.asc == 0x14) { /* RECORD NOT FOUND */ |
| 316 | return SUCCESS; |
| 317 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | return NEEDS_RETRY; |
| 319 | |
| 320 | case HARDWARE_ERROR: |
| 321 | if (scmd->device->retry_hwerror) |
Mike Anderson | bb0003c | 2008-08-12 12:11:58 -0700 | [diff] [blame] | 322 | return ADD_TO_MLQUEUE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | else |
| 324 | return SUCCESS; |
| 325 | |
| 326 | case ILLEGAL_REQUEST: |
| 327 | case BLANK_CHECK: |
| 328 | case DATA_PROTECT: |
| 329 | default: |
| 330 | return SUCCESS; |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | /** |
| 335 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. |
| 336 | * @scmd: SCSI cmd to examine. |
| 337 | * |
| 338 | * Notes: |
| 339 | * This is *only* called when we are examining the status of commands |
| 340 | * queued during error recovery. the main difference here is that we |
| 341 | * don't allow for the possibility of retries here, and we are a lot |
| 342 | * more restrictive about what we consider acceptable. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 343 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) |
| 345 | { |
| 346 | /* |
| 347 | * first check the host byte, to see if there is anything in there |
| 348 | * that would indicate what we need to do. |
| 349 | */ |
| 350 | if (host_byte(scmd->result) == DID_RESET) { |
| 351 | /* |
| 352 | * rats. we are already in the error handler, so we now |
| 353 | * get to try and figure out what to do next. if the sense |
| 354 | * is valid, we have a pretty good idea of what to do. |
| 355 | * if not, we mark it as FAILED. |
| 356 | */ |
| 357 | return scsi_check_sense(scmd); |
| 358 | } |
| 359 | if (host_byte(scmd->result) != DID_OK) |
| 360 | return FAILED; |
| 361 | |
| 362 | /* |
| 363 | * next, check the message byte. |
| 364 | */ |
| 365 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) |
| 366 | return FAILED; |
| 367 | |
| 368 | /* |
| 369 | * now, check the status byte to see if this indicates |
| 370 | * anything special. |
| 371 | */ |
| 372 | switch (status_byte(scmd->result)) { |
| 373 | case GOOD: |
| 374 | case COMMAND_TERMINATED: |
| 375 | return SUCCESS; |
| 376 | case CHECK_CONDITION: |
| 377 | return scsi_check_sense(scmd); |
| 378 | case CONDITION_GOOD: |
| 379 | case INTERMEDIATE_GOOD: |
| 380 | case INTERMEDIATE_C_GOOD: |
| 381 | /* |
| 382 | * who knows? FIXME(eric) |
| 383 | */ |
| 384 | return SUCCESS; |
Michael Reed | 5f91bb0 | 2009-08-10 11:59:28 -0500 | [diff] [blame] | 385 | case RESERVATION_CONFLICT: |
| 386 | /* |
| 387 | * let issuer deal with this, it could be just fine |
| 388 | */ |
| 389 | return SUCCESS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | case BUSY: |
| 391 | case QUEUE_FULL: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | default: |
| 393 | return FAILED; |
| 394 | } |
| 395 | return FAILED; |
| 396 | } |
| 397 | |
| 398 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 399 | * scsi_eh_done - Completion function for error handling. |
| 400 | * @scmd: Cmd that is done. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 401 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | static void scsi_eh_done(struct scsi_cmnd *scmd) |
| 403 | { |
Michael Reed | 8563167 | 2005-12-07 21:46:27 -0600 | [diff] [blame] | 404 | struct completion *eh_action; |
| 405 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 406 | SCSI_LOG_ERROR_RECOVERY(3, |
| 407 | printk("%s scmd: %p result: %x\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 408 | __func__, scmd, scmd->result)); |
Michael Reed | 8563167 | 2005-12-07 21:46:27 -0600 | [diff] [blame] | 409 | |
| 410 | eh_action = scmd->device->host->eh_action; |
| 411 | if (eh_action) |
| 412 | complete(eh_action); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | } |
| 414 | |
| 415 | /** |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 416 | * scsi_try_host_reset - ask host adapter to reset itself |
| 417 | * @scmd: SCSI cmd to send hsot reset. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 418 | */ |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 419 | static int scsi_try_host_reset(struct scsi_cmnd *scmd) |
| 420 | { |
| 421 | unsigned long flags; |
| 422 | int rtn; |
| 423 | |
| 424 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 425 | __func__)); |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 426 | |
| 427 | if (!scmd->device->host->hostt->eh_host_reset_handler) |
| 428 | return FAILED; |
| 429 | |
| 430 | rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd); |
| 431 | |
| 432 | if (rtn == SUCCESS) { |
| 433 | if (!scmd->device->host->hostt->skip_settle_delay) |
| 434 | ssleep(HOST_RESET_SETTLE_TIME); |
| 435 | spin_lock_irqsave(scmd->device->host->host_lock, flags); |
| 436 | scsi_report_bus_reset(scmd->device->host, |
| 437 | scmd_channel(scmd)); |
| 438 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
| 439 | } |
| 440 | |
| 441 | return rtn; |
| 442 | } |
| 443 | |
| 444 | /** |
| 445 | * scsi_try_bus_reset - ask host to perform a bus reset |
| 446 | * @scmd: SCSI cmd to send bus reset. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 447 | */ |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 448 | static int scsi_try_bus_reset(struct scsi_cmnd *scmd) |
| 449 | { |
| 450 | unsigned long flags; |
| 451 | int rtn; |
| 452 | |
| 453 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 454 | __func__)); |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 455 | |
| 456 | if (!scmd->device->host->hostt->eh_bus_reset_handler) |
| 457 | return FAILED; |
| 458 | |
| 459 | rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd); |
| 460 | |
| 461 | if (rtn == SUCCESS) { |
| 462 | if (!scmd->device->host->hostt->skip_settle_delay) |
| 463 | ssleep(BUS_RESET_SETTLE_TIME); |
| 464 | spin_lock_irqsave(scmd->device->host->host_lock, flags); |
| 465 | scsi_report_bus_reset(scmd->device->host, |
| 466 | scmd_channel(scmd)); |
| 467 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
| 468 | } |
| 469 | |
| 470 | return rtn; |
| 471 | } |
| 472 | |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 473 | static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) |
| 474 | { |
| 475 | sdev->was_reset = 1; |
| 476 | sdev->expecting_cc_ua = 1; |
| 477 | } |
| 478 | |
| 479 | /** |
| 480 | * scsi_try_target_reset - Ask host to perform a target reset |
| 481 | * @scmd: SCSI cmd used to send a target reset |
| 482 | * |
| 483 | * Notes: |
| 484 | * There is no timeout for this operation. if this operation is |
| 485 | * unreliable for a given host, then the host itself needs to put a |
| 486 | * timer on it, and set the host back to a consistent state prior to |
| 487 | * returning. |
| 488 | */ |
| 489 | static int scsi_try_target_reset(struct scsi_cmnd *scmd) |
| 490 | { |
| 491 | unsigned long flags; |
| 492 | int rtn; |
| 493 | |
| 494 | if (!scmd->device->host->hostt->eh_target_reset_handler) |
| 495 | return FAILED; |
| 496 | |
| 497 | rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd); |
| 498 | if (rtn == SUCCESS) { |
| 499 | spin_lock_irqsave(scmd->device->host->host_lock, flags); |
| 500 | __starget_for_each_device(scsi_target(scmd->device), NULL, |
| 501 | __scsi_report_device_reset); |
| 502 | spin_unlock_irqrestore(scmd->device->host->host_lock, flags); |
| 503 | } |
| 504 | |
| 505 | return rtn; |
| 506 | } |
| 507 | |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 508 | /** |
| 509 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev |
| 510 | * @scmd: SCSI cmd used to send BDR |
| 511 | * |
| 512 | * Notes: |
| 513 | * There is no timeout for this operation. if this operation is |
| 514 | * unreliable for a given host, then the host itself needs to put a |
| 515 | * timer on it, and set the host back to a consistent state prior to |
| 516 | * returning. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 517 | */ |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 518 | static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) |
| 519 | { |
| 520 | int rtn; |
| 521 | |
| 522 | if (!scmd->device->host->hostt->eh_device_reset_handler) |
| 523 | return FAILED; |
| 524 | |
| 525 | rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd); |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 526 | if (rtn == SUCCESS) |
| 527 | __scsi_report_device_reset(scmd->device, NULL); |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 528 | return rtn; |
| 529 | } |
| 530 | |
| 531 | static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) |
| 532 | { |
| 533 | if (!scmd->device->host->hostt->eh_abort_handler) |
| 534 | return FAILED; |
| 535 | |
| 536 | return scmd->device->host->hostt->eh_abort_handler(scmd); |
| 537 | } |
| 538 | |
| 539 | /** |
| 540 | * scsi_try_to_abort_cmd - Ask host to abort a running command. |
| 541 | * @scmd: SCSI cmd to abort from Lower Level. |
| 542 | * |
| 543 | * Notes: |
| 544 | * This function will not return until the user's completion function |
| 545 | * has been called. there is no timeout on this operation. if the |
| 546 | * author of the low-level driver wishes this operation to be timed, |
| 547 | * they can provide this facility themselves. helper functions in |
| 548 | * scsi_error.c can be supplied to make this easier to do. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 549 | */ |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 550 | static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd) |
| 551 | { |
| 552 | /* |
| 553 | * scsi_done was called just after the command timed out and before |
| 554 | * we had a chance to process it. (db) |
| 555 | */ |
| 556 | if (scmd->serial_number == 0) |
| 557 | return SUCCESS; |
| 558 | return __scsi_try_to_abort_cmd(scmd); |
| 559 | } |
| 560 | |
| 561 | static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) |
| 562 | { |
| 563 | if (__scsi_try_to_abort_cmd(scmd) != SUCCESS) |
| 564 | if (scsi_try_bus_device_reset(scmd) != SUCCESS) |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 565 | if (scsi_try_target_reset(scmd) != SUCCESS) |
| 566 | if (scsi_try_bus_reset(scmd) != SUCCESS) |
| 567 | scsi_try_host_reset(scmd); |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 568 | } |
| 569 | |
| 570 | /** |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 571 | * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory |
Christoph Hellwig | 2dc611d | 2006-11-04 20:04:21 +0100 | [diff] [blame] | 572 | * @scmd: SCSI command structure to hijack |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 573 | * @ses: structure to save restore information |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 574 | * @cmnd: CDB to send. Can be NULL if no new cmnd is needed |
Boaz Harrosh | 64a87b2 | 2008-04-30 11:19:47 +0300 | [diff] [blame] | 575 | * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 576 | * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) |
Christoph Hellwig | 2dc611d | 2006-11-04 20:04:21 +0100 | [diff] [blame] | 577 | * |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 578 | * This function is used to save a scsi command information before re-execution |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 579 | * as part of the error recovery process. If @sense_bytes is 0 the command |
| 580 | * sent must be one that does not transfer any data. If @sense_bytes != 0 |
| 581 | * @cmnd is ignored and this functions sets up a REQUEST_SENSE command |
| 582 | * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 583 | */ |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 584 | void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, |
| 585 | unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | { |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 587 | struct scsi_device *sdev = scmd->device; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 588 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 589 | /* |
| 590 | * We need saved copies of a number of fields - this is because |
| 591 | * error handling may need to overwrite these with different values |
| 592 | * to run different commands, and once error handling is complete, |
| 593 | * we will need to restore these values prior to running the actual |
| 594 | * command. |
| 595 | */ |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 596 | ses->cmd_len = scmd->cmd_len; |
Boaz Harrosh | 64a87b2 | 2008-04-30 11:19:47 +0300 | [diff] [blame] | 597 | ses->cmnd = scmd->cmnd; |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 598 | ses->data_direction = scmd->sc_data_direction; |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 599 | ses->sdb = scmd->sdb; |
Boaz Harrosh | 6f9a35e | 2007-12-13 13:50:53 +0200 | [diff] [blame] | 600 | ses->next_rq = scmd->request->next_rq; |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 601 | ses->result = scmd->result; |
Alan Stern | 1226570 | 2008-07-21 10:25:52 -0400 | [diff] [blame] | 602 | ses->underflow = scmd->underflow; |
Martin K. Petersen | db007fc | 2008-07-17 04:28:31 -0400 | [diff] [blame] | 603 | ses->prot_op = scmd->prot_op; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 604 | |
Martin K. Petersen | db007fc | 2008-07-17 04:28:31 -0400 | [diff] [blame] | 605 | scmd->prot_op = SCSI_PROT_NORMAL; |
Boaz Harrosh | 64a87b2 | 2008-04-30 11:19:47 +0300 | [diff] [blame] | 606 | scmd->cmnd = ses->eh_cmnd; |
| 607 | memset(scmd->cmnd, 0, BLK_MAX_CDB); |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 608 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
Boaz Harrosh | 6f9a35e | 2007-12-13 13:50:53 +0200 | [diff] [blame] | 609 | scmd->request->next_rq = NULL; |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 610 | |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 611 | if (sense_bytes) { |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 612 | scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, |
| 613 | sense_bytes); |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 614 | sg_init_one(&ses->sense_sgl, scmd->sense_buffer, |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 615 | scmd->sdb.length); |
| 616 | scmd->sdb.table.sgl = &ses->sense_sgl; |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 617 | scmd->sc_data_direction = DMA_FROM_DEVICE; |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 618 | scmd->sdb.table.nents = 1; |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 619 | scmd->cmnd[0] = REQUEST_SENSE; |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 620 | scmd->cmnd[4] = scmd->sdb.length; |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 621 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 622 | } else { |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 623 | scmd->sc_data_direction = DMA_NONE; |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 624 | if (cmnd) { |
Boaz Harrosh | 64a87b2 | 2008-04-30 11:19:47 +0300 | [diff] [blame] | 625 | BUG_ON(cmnd_size > BLK_MAX_CDB); |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 626 | memcpy(scmd->cmnd, cmnd, cmnd_size); |
| 627 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 628 | } |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 629 | } |
| 630 | |
| 631 | scmd->underflow = 0; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 632 | |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 633 | if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 635 | (sdev->lun << 5 & 0xe0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 637 | /* |
| 638 | * Zero the sense buffer. The scsi spec mandates that any |
| 639 | * untransferred sense data should be interpreted as being zero. |
| 640 | */ |
FUJITA Tomonori | b80ca4f | 2008-01-13 15:46:13 +0900 | [diff] [blame] | 641 | memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 642 | } |
| 643 | EXPORT_SYMBOL(scsi_eh_prep_cmnd); |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 644 | |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 645 | /** |
| 646 | * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory |
| 647 | * @scmd: SCSI command structure to restore |
Bartlomiej Zolnierkiewicz | 477e608 | 2009-04-27 20:54:22 +0200 | [diff] [blame] | 648 | * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 649 | * |
Bartlomiej Zolnierkiewicz | 477e608 | 2009-04-27 20:54:22 +0200 | [diff] [blame] | 650 | * Undo any damage done by above scsi_eh_prep_cmnd(). |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 651 | */ |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 652 | void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) |
| 653 | { |
| 654 | /* |
| 655 | * Restore original data |
| 656 | */ |
| 657 | scmd->cmd_len = ses->cmd_len; |
Boaz Harrosh | 64a87b2 | 2008-04-30 11:19:47 +0300 | [diff] [blame] | 658 | scmd->cmnd = ses->cmnd; |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 659 | scmd->sc_data_direction = ses->data_direction; |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 660 | scmd->sdb = ses->sdb; |
Boaz Harrosh | 6f9a35e | 2007-12-13 13:50:53 +0200 | [diff] [blame] | 661 | scmd->request->next_rq = ses->next_rq; |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 662 | scmd->result = ses->result; |
Alan Stern | 1226570 | 2008-07-21 10:25:52 -0400 | [diff] [blame] | 663 | scmd->underflow = ses->underflow; |
Martin K. Petersen | db007fc | 2008-07-17 04:28:31 -0400 | [diff] [blame] | 664 | scmd->prot_op = ses->prot_op; |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 665 | } |
| 666 | EXPORT_SYMBOL(scsi_eh_restore_cmnd); |
| 667 | |
| 668 | /** |
| 669 | * scsi_send_eh_cmnd - submit a scsi command as part of error recory |
| 670 | * @scmd: SCSI command structure to hijack |
| 671 | * @cmnd: CDB to send |
| 672 | * @cmnd_size: size in bytes of @cmnd |
| 673 | * @timeout: timeout for this request |
| 674 | * @sense_bytes: size of sense data to copy or 0 |
| 675 | * |
| 676 | * This function is used to send a scsi command down to a target device |
| 677 | * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. |
| 678 | * |
| 679 | * Return value: |
| 680 | * SUCCESS or FAILED or NEEDS_RETRY |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 681 | */ |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 682 | static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, |
| 683 | int cmnd_size, int timeout, unsigned sense_bytes) |
| 684 | { |
| 685 | struct scsi_device *sdev = scmd->device; |
| 686 | struct Scsi_Host *shost = sdev->host; |
| 687 | DECLARE_COMPLETION_ONSTACK(done); |
| 688 | unsigned long timeleft; |
| 689 | unsigned long flags; |
| 690 | struct scsi_eh_save ses; |
| 691 | int rtn; |
| 692 | |
| 693 | scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 694 | shost->eh_action = &done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 696 | spin_lock_irqsave(shost->host_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | scsi_log_send(scmd); |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 698 | shost->hostt->queuecommand(scmd, scsi_eh_done); |
| 699 | spin_unlock_irqrestore(shost->host_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 700 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 701 | timeleft = wait_for_completion_timeout(&done, timeout); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 702 | |
| f59114b | 2005-04-17 15:00:23 -0500 | [diff] [blame] | 703 | shost->eh_action = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 704 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 705 | scsi_log_completion(scmd, SUCCESS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 706 | |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 707 | SCSI_LOG_ERROR_RECOVERY(3, |
| 708 | printk("%s: scmd: %p, timeleft: %ld\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 709 | __func__, scmd, timeleft)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | |
| 711 | /* |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 712 | * If there is time left scsi_eh_done got called, and we will |
| 713 | * examine the actual status codes to see whether the command |
| 714 | * actually did complete normally, else tell the host to forget |
| 715 | * about this command. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | */ |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 717 | if (timeleft) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | rtn = scsi_eh_completed_normally(scmd); |
| 719 | SCSI_LOG_ERROR_RECOVERY(3, |
| 720 | printk("%s: scsi_eh_completed_normally %x\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 721 | __func__, rtn)); |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 722 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 723 | switch (rtn) { |
| 724 | case SUCCESS: |
| 725 | case NEEDS_RETRY: |
| 726 | case FAILED: |
| 727 | break; |
Hannes Reinecke | 6e883b0 | 2009-09-17 17:00:26 +0200 | [diff] [blame] | 728 | case ADD_TO_MLQUEUE: |
| 729 | rtn = NEEDS_RETRY; |
| 730 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | default: |
| 732 | rtn = FAILED; |
| 733 | break; |
| 734 | } |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 735 | } else { |
Brian King | 292148f | 2007-01-30 17:51:17 -0600 | [diff] [blame] | 736 | scsi_abort_eh_cmnd(scmd); |
Christoph Hellwig | 7dfdc9a | 2005-10-31 18:49:52 +0100 | [diff] [blame] | 737 | rtn = FAILED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 738 | } |
| 739 | |
Boaz Harrosh | e1c2346 | 2007-10-08 16:36:45 +0200 | [diff] [blame] | 740 | scsi_eh_restore_cmnd(scmd, &ses); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | return rtn; |
| 742 | } |
| 743 | |
| 744 | /** |
| 745 | * scsi_request_sense - Request sense data from a particular target. |
| 746 | * @scmd: SCSI cmd for request sense. |
| 747 | * |
| 748 | * Notes: |
| 749 | * Some hosts automatically obtain this information, others require |
| 750 | * that we obtain it on our own. This function will *not* return until |
| 751 | * the command either times out, or it completes. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 752 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | static int scsi_request_sense(struct scsi_cmnd *scmd) |
| 754 | { |
Boaz Harrosh | 55db6c1 | 2007-10-08 16:35:19 +0200 | [diff] [blame] | 755 | return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | } |
| 757 | |
| 758 | /** |
| 759 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. |
| 760 | * @scmd: Original SCSI cmd that eh has finished. |
| 761 | * @done_q: Queue for processed commands. |
| 762 | * |
| 763 | * Notes: |
| 764 | * We don't want to use the normal command completion while we are are |
| 765 | * still handling errors - it may cause other commands to be queued, |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 766 | * and that would disturb what we are doing. Thus we really want to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | * keep a list of pending commands for final completion, and once we |
| 768 | * are ready to leave error handling we handle completion for real. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 769 | */ |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 770 | 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] | 771 | { |
| 772 | scmd->device->host->host_failed--; |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 773 | scmd->eh_eflags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 774 | list_move_tail(&scmd->eh_entry, done_q); |
| 775 | } |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 776 | EXPORT_SYMBOL(scsi_eh_finish_cmd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 777 | |
| 778 | /** |
| 779 | * scsi_eh_get_sense - Get device sense data. |
| 780 | * @work_q: Queue of commands to process. |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 781 | * @done_q: Queue of processed commands. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 782 | * |
| 783 | * Description: |
| 784 | * See if we need to request sense information. if so, then get it |
| 785 | * now, so we have a better idea of what to do. |
| 786 | * |
| 787 | * Notes: |
| 788 | * This has the unfortunate side effect that if a shost adapter does |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 789 | * not automatically request sense information, we end up shutting |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | * it down before we request it. |
| 791 | * |
| 792 | * All drivers should request sense information internally these days, |
| 793 | * so for now all I have to say is tough noogies if you end up in here. |
| 794 | * |
| 795 | * XXX: Long term this code should go away, but that needs an audit of |
| 796 | * all LLDDs first. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 797 | */ |
Darrick J. Wong | dca84e4 | 2007-01-26 14:08:49 -0800 | [diff] [blame] | 798 | int scsi_eh_get_sense(struct list_head *work_q, |
| 799 | struct list_head *done_q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 800 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 801 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | int rtn; |
| 803 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 804 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 805 | if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 806 | SCSI_SENSE_VALID(scmd)) |
| 807 | continue; |
| 808 | |
Jeff Garzik | 3bf743e | 2005-10-24 18:04:06 -0400 | [diff] [blame] | 809 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
| 810 | "%s: requesting sense\n", |
| 811 | current->comm)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | rtn = scsi_request_sense(scmd); |
| 813 | if (rtn != SUCCESS) |
| 814 | continue; |
| 815 | |
| 816 | SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p" |
| 817 | " result %x\n", scmd, |
| 818 | scmd->result)); |
| 819 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd)); |
| 820 | |
| 821 | rtn = scsi_decide_disposition(scmd); |
| 822 | |
| 823 | /* |
| 824 | * if the result was normal, then just pass it along to the |
| 825 | * upper level. |
| 826 | */ |
| 827 | if (rtn == SUCCESS) |
| 828 | /* we don't want this command reissued, just |
| 829 | * finished with the sense data, so set |
| 830 | * retries to the max allowed to ensure it |
| 831 | * won't get reissued */ |
| 832 | scmd->retries = scmd->allowed; |
| 833 | else if (rtn != NEEDS_RETRY) |
| 834 | continue; |
| 835 | |
| 836 | scsi_eh_finish_cmd(scmd, done_q); |
| 837 | } |
| 838 | |
| 839 | return list_empty(work_q); |
| 840 | } |
Darrick J. Wong | dca84e4 | 2007-01-26 14:08:49 -0800 | [diff] [blame] | 841 | EXPORT_SYMBOL_GPL(scsi_eh_get_sense); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 842 | |
| 843 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | * scsi_eh_tur - Send TUR to device. |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 845 | * @scmd: &scsi_cmnd to send TUR |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | * |
| 847 | * Return value: |
| 848 | * 0 - Device is ready. 1 - Device NOT ready. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 849 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | static int scsi_eh_tur(struct scsi_cmnd *scmd) |
| 851 | { |
| 852 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; |
| 853 | int retry_cnt = 1, rtn; |
| 854 | |
| 855 | retry_tur: |
Mike Christie | 0db99e3 | 2006-08-26 03:00:22 -0400 | [diff] [blame] | 856 | rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 857 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 859 | __func__, scmd, rtn)); |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 860 | |
| 861 | switch (rtn) { |
| 862 | case NEEDS_RETRY: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 863 | if (retry_cnt--) |
| 864 | goto retry_tur; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 865 | /*FALLTHRU*/ |
| 866 | case SUCCESS: |
Alan Stern | e47373e | 2005-03-30 15:05:45 -0500 | [diff] [blame] | 867 | return 0; |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 868 | default: |
| 869 | return 1; |
Alan Stern | e47373e | 2005-03-30 15:05:45 -0500 | [diff] [blame] | 870 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | } |
| 872 | |
| 873 | /** |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 874 | * scsi_eh_abort_cmds - abort pending commands. |
| 875 | * @work_q: &list_head for pending commands. |
| 876 | * @done_q: &list_head for processed commands. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | * |
| 878 | * Decription: |
| 879 | * Try and see whether or not it makes sense to try and abort the |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 880 | * running command. This only works out to be the case if we have one |
| 881 | * command that has timed out. If the command simply failed, it makes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | * no sense to try and abort the command, since as far as the shost |
| 883 | * adapter is concerned, it isn't running. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 884 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 885 | static int scsi_eh_abort_cmds(struct list_head *work_q, |
| 886 | struct list_head *done_q) |
| 887 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 888 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 889 | int rtn; |
| 890 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 891 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 892 | if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | continue; |
| 894 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:" |
| 895 | "0x%p\n", current->comm, |
| 896 | scmd)); |
| 897 | rtn = scsi_try_to_abort_cmd(scmd); |
| 898 | if (rtn == SUCCESS) { |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 899 | scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 900 | if (!scsi_device_online(scmd->device) || |
| 901 | !scsi_eh_tur(scmd)) { |
| 902 | scsi_eh_finish_cmd(scmd, done_q); |
| 903 | } |
| 904 | |
| 905 | } else |
| 906 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting" |
| 907 | " cmd failed:" |
| 908 | "0x%p\n", |
| 909 | current->comm, |
| 910 | scmd)); |
| 911 | } |
| 912 | |
| 913 | return list_empty(work_q); |
| 914 | } |
| 915 | |
| 916 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 917 | * scsi_eh_try_stu - Send START_UNIT to device. |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 918 | * @scmd: &scsi_cmnd to send START_UNIT |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | * |
| 920 | * Return value: |
| 921 | * 0 - Device is ready. 1 - Device NOT ready. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 922 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) |
| 924 | { |
| 925 | 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] | 926 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 927 | if (scmd->device->allow_restart) { |
Brian King | ed773e6 | 2007-03-29 15:25:52 -0500 | [diff] [blame] | 928 | int i, rtn = NEEDS_RETRY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | |
Brian King | ed773e6 | 2007-03-29 15:25:52 -0500 | [diff] [blame] | 930 | for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) |
James Bottomley | 9728c08 | 2008-11-30 10:32:26 -0600 | [diff] [blame] | 931 | rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); |
Brian King | ed773e6 | 2007-03-29 15:25:52 -0500 | [diff] [blame] | 932 | |
Christoph Hellwig | 631c228 | 2006-07-08 20:42:15 +0200 | [diff] [blame] | 933 | if (rtn == SUCCESS) |
| 934 | return 0; |
| 935 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 936 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 937 | return 1; |
| 938 | } |
| 939 | |
| 940 | /** |
| 941 | * scsi_eh_stu - send START_UNIT if needed |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 942 | * @shost: &scsi host being recovered. |
| 943 | * @work_q: &list_head for pending commands. |
| 944 | * @done_q: &list_head for processed commands. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 945 | * |
| 946 | * Notes: |
| 947 | * If commands are failing due to not ready, initializing command required, |
| 948 | * try revalidating the device, which will end up sending a start unit. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 949 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 950 | static int scsi_eh_stu(struct Scsi_Host *shost, |
| 951 | struct list_head *work_q, |
| 952 | struct list_head *done_q) |
| 953 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 954 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | struct scsi_device *sdev; |
| 956 | |
| 957 | shost_for_each_device(sdev, shost) { |
| 958 | stu_scmd = NULL; |
| 959 | list_for_each_entry(scmd, work_q, eh_entry) |
| 960 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && |
| 961 | scsi_check_sense(scmd) == FAILED ) { |
| 962 | stu_scmd = scmd; |
| 963 | break; |
| 964 | } |
| 965 | |
| 966 | if (!stu_scmd) |
| 967 | continue; |
| 968 | |
| 969 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:" |
| 970 | " 0x%p\n", current->comm, sdev)); |
| 971 | |
| 972 | if (!scsi_eh_try_stu(stu_scmd)) { |
| 973 | if (!scsi_device_online(sdev) || |
| 974 | !scsi_eh_tur(stu_scmd)) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 975 | list_for_each_entry_safe(scmd, next, |
| 976 | work_q, eh_entry) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 977 | if (scmd->device == sdev) |
| 978 | scsi_eh_finish_cmd(scmd, done_q); |
| 979 | } |
| 980 | } |
| 981 | } else { |
| 982 | SCSI_LOG_ERROR_RECOVERY(3, |
| 983 | printk("%s: START_UNIT failed to sdev:" |
| 984 | " 0x%p\n", current->comm, sdev)); |
| 985 | } |
| 986 | } |
| 987 | |
| 988 | return list_empty(work_q); |
| 989 | } |
| 990 | |
| 991 | |
| 992 | /** |
| 993 | * scsi_eh_bus_device_reset - send bdr if needed |
| 994 | * @shost: scsi host being recovered. |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 995 | * @work_q: &list_head for pending commands. |
| 996 | * @done_q: &list_head for processed commands. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 997 | * |
| 998 | * Notes: |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 999 | * Try a bus device reset. Still, look to see whether we have multiple |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | * devices that are jammed or not - if we have multiple devices, it |
| 1001 | * makes no sense to try bus_device_reset - we really would need to try |
| 1002 | * a bus_reset instead. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1003 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1004 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, |
| 1005 | struct list_head *work_q, |
| 1006 | struct list_head *done_q) |
| 1007 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1008 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | struct scsi_device *sdev; |
| 1010 | int rtn; |
| 1011 | |
| 1012 | shost_for_each_device(sdev, shost) { |
| 1013 | bdr_scmd = NULL; |
| 1014 | list_for_each_entry(scmd, work_q, eh_entry) |
| 1015 | if (scmd->device == sdev) { |
| 1016 | bdr_scmd = scmd; |
| 1017 | break; |
| 1018 | } |
| 1019 | |
| 1020 | if (!bdr_scmd) |
| 1021 | continue; |
| 1022 | |
| 1023 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:" |
| 1024 | " 0x%p\n", current->comm, |
| 1025 | sdev)); |
| 1026 | rtn = scsi_try_bus_device_reset(bdr_scmd); |
| 1027 | if (rtn == SUCCESS) { |
| 1028 | if (!scsi_device_online(sdev) || |
| 1029 | !scsi_eh_tur(bdr_scmd)) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1030 | list_for_each_entry_safe(scmd, next, |
| 1031 | work_q, eh_entry) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | if (scmd->device == sdev) |
| 1033 | scsi_eh_finish_cmd(scmd, |
| 1034 | done_q); |
| 1035 | } |
| 1036 | } |
| 1037 | } else { |
| 1038 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR" |
| 1039 | " failed sdev:" |
| 1040 | "0x%p\n", |
| 1041 | current->comm, |
| 1042 | sdev)); |
| 1043 | } |
| 1044 | } |
| 1045 | |
| 1046 | return list_empty(work_q); |
| 1047 | } |
| 1048 | |
| 1049 | /** |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1050 | * scsi_eh_target_reset - send target reset if needed |
| 1051 | * @shost: scsi host being recovered. |
| 1052 | * @work_q: &list_head for pending commands. |
| 1053 | * @done_q: &list_head for processed commands. |
| 1054 | * |
| 1055 | * Notes: |
| 1056 | * Try a target reset. |
| 1057 | */ |
| 1058 | static int scsi_eh_target_reset(struct Scsi_Host *shost, |
| 1059 | struct list_head *work_q, |
| 1060 | struct list_head *done_q) |
| 1061 | { |
| 1062 | struct scsi_cmnd *scmd, *tgtr_scmd, *next; |
James Bottomley | c82dc88 | 2008-09-12 16:46:51 -0500 | [diff] [blame] | 1063 | unsigned int id = 0; |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1064 | int rtn; |
| 1065 | |
James Bottomley | c82dc88 | 2008-09-12 16:46:51 -0500 | [diff] [blame] | 1066 | do { |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1067 | tgtr_scmd = NULL; |
| 1068 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 1069 | if (id == scmd_id(scmd)) { |
| 1070 | tgtr_scmd = scmd; |
| 1071 | break; |
| 1072 | } |
| 1073 | } |
James Bottomley | c82dc88 | 2008-09-12 16:46:51 -0500 | [diff] [blame] | 1074 | if (!tgtr_scmd) { |
| 1075 | /* not one exactly equal; find the next highest */ |
| 1076 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 1077 | if (scmd_id(scmd) > id && |
| 1078 | (!tgtr_scmd || |
| 1079 | scmd_id(tgtr_scmd) > scmd_id(scmd))) |
| 1080 | tgtr_scmd = scmd; |
| 1081 | } |
| 1082 | } |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1083 | if (!tgtr_scmd) |
James Bottomley | c82dc88 | 2008-09-12 16:46:51 -0500 | [diff] [blame] | 1084 | /* no more commands, that's it */ |
| 1085 | break; |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1086 | |
| 1087 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset " |
| 1088 | "to target %d\n", |
| 1089 | current->comm, id)); |
| 1090 | rtn = scsi_try_target_reset(tgtr_scmd); |
| 1091 | if (rtn == SUCCESS) { |
| 1092 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1093 | if (id == scmd_id(scmd)) |
| 1094 | if (!scsi_device_online(scmd->device) || |
| 1095 | !scsi_eh_tur(tgtr_scmd)) |
| 1096 | scsi_eh_finish_cmd(scmd, |
| 1097 | done_q); |
| 1098 | } |
| 1099 | } else |
| 1100 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset" |
| 1101 | " failed target: " |
| 1102 | "%d\n", |
| 1103 | current->comm, id)); |
James Bottomley | c82dc88 | 2008-09-12 16:46:51 -0500 | [diff] [blame] | 1104 | id++; |
| 1105 | } while(id != 0); |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1106 | |
| 1107 | return list_empty(work_q); |
| 1108 | } |
| 1109 | |
| 1110 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1111 | * scsi_eh_bus_reset - send a bus reset |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 1112 | * @shost: &scsi host being recovered. |
| 1113 | * @work_q: &list_head for pending commands. |
| 1114 | * @done_q: &list_head for processed commands. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1115 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, |
| 1117 | struct list_head *work_q, |
| 1118 | struct list_head *done_q) |
| 1119 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1120 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | unsigned int channel; |
| 1122 | int rtn; |
| 1123 | |
| 1124 | /* |
| 1125 | * we really want to loop over the various channels, and do this on |
| 1126 | * a channel by channel basis. we should also check to see if any |
| 1127 | * of the failed commands are on soft_reset devices, and if so, skip |
| 1128 | * the reset. |
| 1129 | */ |
| 1130 | |
| 1131 | for (channel = 0; channel <= shost->max_channel; channel++) { |
| 1132 | chan_scmd = NULL; |
| 1133 | list_for_each_entry(scmd, work_q, eh_entry) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1134 | if (channel == scmd_channel(scmd)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1135 | chan_scmd = scmd; |
| 1136 | break; |
| 1137 | /* |
| 1138 | * FIXME add back in some support for |
| 1139 | * soft_reset devices. |
| 1140 | */ |
| 1141 | } |
| 1142 | } |
| 1143 | |
| 1144 | if (!chan_scmd) |
| 1145 | continue; |
| 1146 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:" |
| 1147 | " %d\n", current->comm, |
| 1148 | channel)); |
| 1149 | rtn = scsi_try_bus_reset(chan_scmd); |
| 1150 | if (rtn == SUCCESS) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1151 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1152 | if (channel == scmd_channel(scmd)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1153 | if (!scsi_device_online(scmd->device) || |
| 1154 | !scsi_eh_tur(scmd)) |
| 1155 | scsi_eh_finish_cmd(scmd, |
| 1156 | done_q); |
| 1157 | } |
| 1158 | } else { |
| 1159 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST" |
| 1160 | " failed chan: %d\n", |
| 1161 | current->comm, |
| 1162 | channel)); |
| 1163 | } |
| 1164 | } |
| 1165 | return list_empty(work_q); |
| 1166 | } |
| 1167 | |
| 1168 | /** |
| 1169 | * scsi_eh_host_reset - send a host reset |
| 1170 | * @work_q: list_head for processed commands. |
| 1171 | * @done_q: list_head for processed commands. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1172 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1173 | static int scsi_eh_host_reset(struct list_head *work_q, |
| 1174 | struct list_head *done_q) |
| 1175 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1176 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | int rtn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | |
| 1179 | if (!list_empty(work_q)) { |
| 1180 | scmd = list_entry(work_q->next, |
| 1181 | struct scsi_cmnd, eh_entry); |
| 1182 | |
| 1183 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n" |
| 1184 | , current->comm)); |
| 1185 | |
| 1186 | rtn = scsi_try_host_reset(scmd); |
| 1187 | if (rtn == SUCCESS) { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1188 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1189 | if (!scsi_device_online(scmd->device) || |
| 1190 | (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) || |
| 1191 | !scsi_eh_tur(scmd)) |
| 1192 | scsi_eh_finish_cmd(scmd, done_q); |
| 1193 | } |
| 1194 | } else { |
| 1195 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST" |
| 1196 | " failed\n", |
| 1197 | current->comm)); |
| 1198 | } |
| 1199 | } |
| 1200 | return list_empty(work_q); |
| 1201 | } |
| 1202 | |
| 1203 | /** |
| 1204 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover |
| 1205 | * @work_q: list_head for processed commands. |
| 1206 | * @done_q: list_head for processed commands. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1207 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | static void scsi_eh_offline_sdevs(struct list_head *work_q, |
| 1209 | struct list_head *done_q) |
| 1210 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1211 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1212 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1213 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
Matthew Wilcox | 31765d7 | 2007-08-17 11:02:10 -0600 | [diff] [blame] | 1214 | sdev_printk(KERN_INFO, scmd->device, "Device offlined - " |
| 1215 | "not ready after error recovery\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1216 | scsi_device_set_state(scmd->device, SDEV_OFFLINE); |
Christoph Hellwig | 3111b0d | 2005-06-19 13:43:26 +0200 | [diff] [blame] | 1217 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | /* |
| 1219 | * FIXME: Handle lost cmds. |
| 1220 | */ |
| 1221 | } |
| 1222 | scsi_eh_finish_cmd(scmd, done_q); |
| 1223 | } |
| 1224 | return; |
| 1225 | } |
| 1226 | |
| 1227 | /** |
Mike Christie | 4a27446 | 2008-08-19 18:45:31 -0500 | [diff] [blame] | 1228 | * scsi_noretry_cmd - determinte if command should be failed fast |
| 1229 | * @scmd: SCSI cmd to examine. |
| 1230 | */ |
| 1231 | int scsi_noretry_cmd(struct scsi_cmnd *scmd) |
| 1232 | { |
| 1233 | switch (host_byte(scmd->result)) { |
| 1234 | case DID_OK: |
| 1235 | break; |
| 1236 | case DID_BUS_BUSY: |
| 1237 | return blk_failfast_transport(scmd->request); |
| 1238 | case DID_PARITY: |
| 1239 | return blk_failfast_dev(scmd->request); |
| 1240 | case DID_ERROR: |
| 1241 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && |
| 1242 | status_byte(scmd->result) == RESERVATION_CONFLICT) |
| 1243 | return 0; |
| 1244 | /* fall through */ |
| 1245 | case DID_SOFT_ERROR: |
| 1246 | return blk_failfast_driver(scmd->request); |
| 1247 | } |
| 1248 | |
| 1249 | switch (status_byte(scmd->result)) { |
| 1250 | case CHECK_CONDITION: |
| 1251 | /* |
| 1252 | * assume caller has checked sense and determinted |
| 1253 | * the check condition was retryable. |
| 1254 | */ |
| 1255 | return blk_failfast_dev(scmd->request); |
| 1256 | } |
| 1257 | |
| 1258 | return 0; |
| 1259 | } |
| 1260 | |
| 1261 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1262 | * scsi_decide_disposition - Disposition a cmd on return from LLD. |
| 1263 | * @scmd: SCSI cmd to examine. |
| 1264 | * |
| 1265 | * Notes: |
| 1266 | * This is *only* called when we are examining the status after sending |
| 1267 | * out the actual data command. any commands that are queued for error |
| 1268 | * recovery (e.g. test_unit_ready) do *not* come through here. |
| 1269 | * |
| 1270 | * When this routine returns failed, it means the error handler thread |
| 1271 | * is woken. In cases where the error code indicates an error that |
| 1272 | * doesn't require the error handler read (i.e. we don't need to |
| 1273 | * abort/reset), this function should return SUCCESS. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1274 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1275 | int scsi_decide_disposition(struct scsi_cmnd *scmd) |
| 1276 | { |
| 1277 | int rtn; |
| 1278 | |
| 1279 | /* |
| 1280 | * if the device is offline, then we clearly just pass the result back |
| 1281 | * up to the top level. |
| 1282 | */ |
| 1283 | if (!scsi_device_online(scmd->device)) { |
| 1284 | SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report" |
| 1285 | " as SUCCESS\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 1286 | __func__)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | return SUCCESS; |
| 1288 | } |
| 1289 | |
| 1290 | /* |
| 1291 | * first check the host byte, to see if there is anything in there |
| 1292 | * that would indicate what we need to do. |
| 1293 | */ |
| 1294 | switch (host_byte(scmd->result)) { |
| 1295 | case DID_PASSTHROUGH: |
| 1296 | /* |
| 1297 | * no matter what, pass this through to the upper layer. |
| 1298 | * nuke this special code so that it looks like we are saying |
| 1299 | * did_ok. |
| 1300 | */ |
| 1301 | scmd->result &= 0xff00ffff; |
| 1302 | return SUCCESS; |
| 1303 | case DID_OK: |
| 1304 | /* |
| 1305 | * looks good. drop through, and check the next byte. |
| 1306 | */ |
| 1307 | break; |
| 1308 | case DID_NO_CONNECT: |
| 1309 | case DID_BAD_TARGET: |
| 1310 | case DID_ABORT: |
| 1311 | /* |
| 1312 | * note - this means that we just report the status back |
| 1313 | * to the top level driver, not that we actually think |
| 1314 | * that it indicates SUCCESS. |
| 1315 | */ |
| 1316 | return SUCCESS; |
| 1317 | /* |
| 1318 | * when the low level driver returns did_soft_error, |
| 1319 | * it is responsible for keeping an internal retry counter |
| 1320 | * in order to avoid endless loops (db) |
| 1321 | * |
| 1322 | * actually this is a bug in this function here. we should |
| 1323 | * be mindful of the maximum number of retries specified |
| 1324 | * and not get stuck in a loop. |
| 1325 | */ |
| 1326 | case DID_SOFT_ERROR: |
| 1327 | goto maybe_retry; |
| 1328 | case DID_IMM_RETRY: |
| 1329 | return NEEDS_RETRY; |
| 1330 | |
| bf34191 | 2005-04-12 17:49:09 -0500 | [diff] [blame] | 1331 | case DID_REQUEUE: |
| 1332 | return ADD_TO_MLQUEUE; |
Mike Christie | a4dfaa6 | 2008-08-19 18:45:25 -0500 | [diff] [blame] | 1333 | case DID_TRANSPORT_DISRUPTED: |
| 1334 | /* |
| 1335 | * LLD/transport was disrupted during processing of the IO. |
| 1336 | * The transport class is now blocked/blocking, |
| 1337 | * and the transport will decide what to do with the IO |
Mike Christie | 939c228 | 2008-11-04 19:47:19 -0600 | [diff] [blame] | 1338 | * based on its timers and recovery capablilities if |
| 1339 | * there are enough retries. |
Mike Christie | a4dfaa6 | 2008-08-19 18:45:25 -0500 | [diff] [blame] | 1340 | */ |
Mike Christie | 939c228 | 2008-11-04 19:47:19 -0600 | [diff] [blame] | 1341 | goto maybe_retry; |
Mike Christie | a4dfaa6 | 2008-08-19 18:45:25 -0500 | [diff] [blame] | 1342 | case DID_TRANSPORT_FAILFAST: |
| 1343 | /* |
| 1344 | * The transport decided to failfast the IO (most likely |
| 1345 | * the fast io fail tmo fired), so send IO directly upwards. |
| 1346 | */ |
| 1347 | return SUCCESS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | case DID_ERROR: |
| 1349 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && |
| 1350 | status_byte(scmd->result) == RESERVATION_CONFLICT) |
| 1351 | /* |
| 1352 | * execute reservation conflict processing code |
| 1353 | * lower down |
| 1354 | */ |
| 1355 | break; |
| 1356 | /* fallthrough */ |
| 1357 | |
| 1358 | case DID_BUS_BUSY: |
| 1359 | case DID_PARITY: |
| 1360 | goto maybe_retry; |
| 1361 | case DID_TIME_OUT: |
| 1362 | /* |
| 1363 | * when we scan the bus, we get timeout messages for |
| 1364 | * these commands if there is no device available. |
| 1365 | * other hosts report did_no_connect for the same thing. |
| 1366 | */ |
| 1367 | if ((scmd->cmnd[0] == TEST_UNIT_READY || |
| 1368 | scmd->cmnd[0] == INQUIRY)) { |
| 1369 | return SUCCESS; |
| 1370 | } else { |
| 1371 | return FAILED; |
| 1372 | } |
| 1373 | case DID_RESET: |
| 1374 | return SUCCESS; |
| 1375 | default: |
| 1376 | return FAILED; |
| 1377 | } |
| 1378 | |
| 1379 | /* |
| 1380 | * next, check the message byte. |
| 1381 | */ |
| 1382 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) |
| 1383 | return FAILED; |
| 1384 | |
| 1385 | /* |
| 1386 | * check the status byte to see if this indicates anything special. |
| 1387 | */ |
| 1388 | switch (status_byte(scmd->result)) { |
| 1389 | case QUEUE_FULL: |
| 1390 | /* |
| 1391 | * the case of trying to send too many commands to a |
| 1392 | * tagged queueing device. |
| 1393 | */ |
| 1394 | case BUSY: |
| 1395 | /* |
| 1396 | * device can't talk to us at the moment. Should only |
| 1397 | * occur (SAM-3) when the task queue is empty, so will cause |
| 1398 | * the empty queue handling to trigger a stall in the |
| 1399 | * device. |
| 1400 | */ |
| 1401 | return ADD_TO_MLQUEUE; |
| 1402 | case GOOD: |
| 1403 | case COMMAND_TERMINATED: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1404 | return SUCCESS; |
Vladislav Bolkhovitin | a9b589d | 2008-11-06 13:57:52 +0300 | [diff] [blame] | 1405 | case TASK_ABORTED: |
| 1406 | goto maybe_retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | case CHECK_CONDITION: |
| 1408 | rtn = scsi_check_sense(scmd); |
| 1409 | if (rtn == NEEDS_RETRY) |
| 1410 | goto maybe_retry; |
| 1411 | /* if rtn == FAILED, we have no sense information; |
| 1412 | * returning FAILED will wake the error handler thread |
| 1413 | * to collect the sense and redo the decide |
| 1414 | * disposition */ |
| 1415 | return rtn; |
| 1416 | case CONDITION_GOOD: |
| 1417 | case INTERMEDIATE_GOOD: |
| 1418 | case INTERMEDIATE_C_GOOD: |
| 1419 | case ACA_ACTIVE: |
| 1420 | /* |
| 1421 | * who knows? FIXME(eric) |
| 1422 | */ |
| 1423 | return SUCCESS; |
| 1424 | |
| 1425 | case RESERVATION_CONFLICT: |
James Bottomley | 9ccfc75 | 2005-10-02 11:45:08 -0500 | [diff] [blame] | 1426 | sdev_printk(KERN_INFO, scmd->device, |
| 1427 | "reservation conflict\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1428 | return SUCCESS; /* causes immediate i/o error */ |
| 1429 | default: |
| 1430 | return FAILED; |
| 1431 | } |
| 1432 | return FAILED; |
| 1433 | |
| 1434 | maybe_retry: |
| 1435 | |
| 1436 | /* we requeue for retry because the error was retryable, and |
| 1437 | * the request was not marked fast fail. Note that above, |
| 1438 | * even if the request is marked fast fail, we still requeue |
| 1439 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ |
Brian King | 8884efa | 2006-02-24 17:10:04 -0600 | [diff] [blame] | 1440 | if ((++scmd->retries) <= scmd->allowed |
Mike Christie | 4a27446 | 2008-08-19 18:45:31 -0500 | [diff] [blame] | 1441 | && !scsi_noretry_cmd(scmd)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1442 | return NEEDS_RETRY; |
| 1443 | } else { |
| 1444 | /* |
| 1445 | * no more retries - report this one back to upper level. |
| 1446 | */ |
| 1447 | return SUCCESS; |
| 1448 | } |
| 1449 | } |
| 1450 | |
FUJITA Tomonori | f078727 | 2008-12-14 01:23:45 +0900 | [diff] [blame] | 1451 | static void eh_lock_door_done(struct request *req, int uptodate) |
| 1452 | { |
| 1453 | __blk_put_request(req->q, req); |
| 1454 | } |
| 1455 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1456 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | * scsi_eh_lock_door - Prevent medium removal for the specified device |
| 1458 | * @sdev: SCSI device to prevent medium removal |
| 1459 | * |
| 1460 | * Locking: |
James Bottomley | 91bc31f | 2009-05-17 09:30:48 -0500 | [diff] [blame] | 1461 | * We must be called from process context. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | * |
| 1463 | * Notes: |
| 1464 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the |
| 1465 | * head of the devices request queue, and continue. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1466 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | static void scsi_eh_lock_door(struct scsi_device *sdev) |
| 1468 | { |
FUJITA Tomonori | f078727 | 2008-12-14 01:23:45 +0900 | [diff] [blame] | 1469 | struct request *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1470 | |
James Bottomley | 91bc31f | 2009-05-17 09:30:48 -0500 | [diff] [blame] | 1471 | /* |
| 1472 | * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a |
| 1473 | * request becomes available |
| 1474 | */ |
FUJITA Tomonori | f078727 | 2008-12-14 01:23:45 +0900 | [diff] [blame] | 1475 | req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | |
FUJITA Tomonori | f078727 | 2008-12-14 01:23:45 +0900 | [diff] [blame] | 1477 | req->cmd[0] = ALLOW_MEDIUM_REMOVAL; |
| 1478 | req->cmd[1] = 0; |
| 1479 | req->cmd[2] = 0; |
| 1480 | req->cmd[3] = 0; |
| 1481 | req->cmd[4] = SCSI_REMOVAL_PREVENT; |
| 1482 | req->cmd[5] = 0; |
| 1483 | |
| 1484 | req->cmd_len = COMMAND_SIZE(req->cmd[0]); |
| 1485 | |
| 1486 | req->cmd_type = REQ_TYPE_BLOCK_PC; |
| 1487 | req->cmd_flags |= REQ_QUIET; |
| 1488 | req->timeout = 10 * HZ; |
| 1489 | req->retries = 5; |
| 1490 | |
| 1491 | blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | } |
| 1493 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1494 | /** |
| 1495 | * scsi_restart_operations - restart io operations to the specified host. |
| 1496 | * @shost: Host we are restarting. |
| 1497 | * |
| 1498 | * Notes: |
| 1499 | * When we entered the error handler, we blocked all further i/o to |
| 1500 | * this device. we need to 'reverse' this process. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1501 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1502 | static void scsi_restart_operations(struct Scsi_Host *shost) |
| 1503 | { |
| 1504 | struct scsi_device *sdev; |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 1505 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | |
| 1507 | /* |
| 1508 | * If the door was locked, we need to insert a door lock request |
| 1509 | * onto the head of the SCSI request queue for the device. There |
| 1510 | * is no point trying to lock the door of an off-line device. |
| 1511 | */ |
| 1512 | shost_for_each_device(sdev, shost) { |
| 1513 | if (scsi_device_online(sdev) && sdev->locked) |
| 1514 | scsi_eh_lock_door(sdev); |
| 1515 | } |
| 1516 | |
| 1517 | /* |
| 1518 | * next free up anything directly waiting upon the host. this |
| 1519 | * will be requests for character device operations, and also for |
| 1520 | * ioctls to queued block devices. |
| 1521 | */ |
| 1522 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 1523 | __func__)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | |
James Bottomley | 939647e | 2005-09-18 15:05:20 -0500 | [diff] [blame] | 1525 | spin_lock_irqsave(shost->host_lock, flags); |
| 1526 | if (scsi_host_set_state(shost, SHOST_RUNNING)) |
| 1527 | if (scsi_host_set_state(shost, SHOST_CANCEL)) |
| 1528 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); |
| 1529 | spin_unlock_irqrestore(shost->host_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1530 | |
| 1531 | wake_up(&shost->host_wait); |
| 1532 | |
| 1533 | /* |
| 1534 | * finally we need to re-initiate requests that may be pending. we will |
| 1535 | * have had everything blocked while error handling is taking place, and |
| 1536 | * now that error recovery is done, we will need to ensure that these |
| 1537 | * requests are started. |
| 1538 | */ |
| 1539 | scsi_run_host_queues(shost); |
| 1540 | } |
| 1541 | |
| 1542 | /** |
| 1543 | * scsi_eh_ready_devs - check device ready state and recover if not. |
| 1544 | * @shost: host to be recovered. |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 1545 | * @work_q: &list_head for pending commands. |
| 1546 | * @done_q: &list_head for processed commands. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1547 | */ |
Darrick J. Wong | dca84e4 | 2007-01-26 14:08:49 -0800 | [diff] [blame] | 1548 | void scsi_eh_ready_devs(struct Scsi_Host *shost, |
| 1549 | struct list_head *work_q, |
| 1550 | struct list_head *done_q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1551 | { |
| 1552 | if (!scsi_eh_stu(shost, work_q, done_q)) |
| 1553 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1554 | if (!scsi_eh_target_reset(shost, work_q, done_q)) |
| 1555 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) |
| 1556 | if (!scsi_eh_host_reset(work_q, done_q)) |
| 1557 | scsi_eh_offline_sdevs(work_q, |
| 1558 | done_q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1559 | } |
Darrick J. Wong | dca84e4 | 2007-01-26 14:08:49 -0800 | [diff] [blame] | 1560 | EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1561 | |
| 1562 | /** |
| 1563 | * scsi_eh_flush_done_q - finish processed commands or retry them. |
| 1564 | * @done_q: list_head of processed commands. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1565 | */ |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 1566 | void scsi_eh_flush_done_q(struct list_head *done_q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | { |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1568 | struct scsi_cmnd *scmd, *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1569 | |
Christoph Hellwig | 937abeaa | 2005-06-19 13:43:56 +0200 | [diff] [blame] | 1570 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
| 1571 | list_del_init(&scmd->eh_entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1572 | if (scsi_device_online(scmd->device) && |
Mike Christie | 4a27446 | 2008-08-19 18:45:31 -0500 | [diff] [blame] | 1573 | !scsi_noretry_cmd(scmd) && |
Brian King | 8884efa | 2006-02-24 17:10:04 -0600 | [diff] [blame] | 1574 | (++scmd->retries <= scmd->allowed)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1575 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush" |
| 1576 | " retry cmd: %p\n", |
| 1577 | current->comm, |
| 1578 | scmd)); |
| 1579 | scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); |
| 1580 | } else { |
Patrick Mansfield | 793698c | 2005-05-16 17:42:15 -0700 | [diff] [blame] | 1581 | /* |
| 1582 | * If just we got sense for the device (called |
| 1583 | * scsi_eh_get_sense), scmd->result is already |
| 1584 | * set, do not set DRIVER_TIMEOUT. |
| 1585 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1586 | if (!scmd->result) |
| 1587 | scmd->result |= (DRIVER_TIMEOUT << 24); |
| 1588 | SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish" |
| 1589 | " cmd: %p\n", |
| 1590 | current->comm, scmd)); |
| 1591 | scsi_finish_command(scmd); |
| 1592 | } |
| 1593 | } |
| 1594 | } |
Tejun Heo | 041c5fc | 2006-01-23 13:09:36 +0900 | [diff] [blame] | 1595 | EXPORT_SYMBOL(scsi_eh_flush_done_q); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1596 | |
| 1597 | /** |
| 1598 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. |
| 1599 | * @shost: Host to unjam. |
| 1600 | * |
| 1601 | * Notes: |
| 1602 | * When we come in here, we *know* that all commands on the bus have |
| 1603 | * either completed, failed or timed out. we also know that no further |
| 1604 | * commands are being sent to the host, so things are relatively quiet |
| 1605 | * and we have freedom to fiddle with things as we wish. |
| 1606 | * |
| 1607 | * This is only the *default* implementation. it is possible for |
| 1608 | * individual drivers to supply their own version of this function, and |
| 1609 | * if the maintainer wishes to do this, it is strongly suggested that |
| 1610 | * this function be taken as a template and modified. this function |
| 1611 | * was designed to correctly handle problems for about 95% of the |
| 1612 | * different cases out there, and it should always provide at least a |
| 1613 | * reasonable amount of error recovery. |
| 1614 | * |
| 1615 | * Any command marked 'failed' or 'timeout' must eventually have |
| 1616 | * scsi_finish_cmd() called for it. we do all of the retry stuff |
| 1617 | * here, so when we restart the host after we return it should have an |
| 1618 | * empty queue. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1619 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1620 | static void scsi_unjam_host(struct Scsi_Host *shost) |
| 1621 | { |
| 1622 | unsigned long flags; |
| 1623 | LIST_HEAD(eh_work_q); |
| 1624 | LIST_HEAD(eh_done_q); |
| 1625 | |
| 1626 | spin_lock_irqsave(shost->host_lock, flags); |
| 1627 | list_splice_init(&shost->eh_cmd_q, &eh_work_q); |
| 1628 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 1629 | |
| 1630 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); |
| 1631 | |
| 1632 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) |
| 1633 | if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) |
| 1634 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); |
| 1635 | |
| 1636 | scsi_eh_flush_done_q(&eh_done_q); |
| 1637 | } |
| 1638 | |
| 1639 | /** |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1640 | * scsi_error_handler - SCSI error handler thread |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1641 | * @data: Host for which we are running. |
| 1642 | * |
| 1643 | * Notes: |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1644 | * This is the main error handling loop. This is run as a kernel thread |
| 1645 | * for every SCSI host and handles all error handling activity. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1646 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1647 | int scsi_error_handler(void *data) |
| 1648 | { |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1649 | struct Scsi_Host *shost = data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1650 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1651 | /* |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1652 | * We use TASK_INTERRUPTIBLE so that the thread is not |
| 1653 | * counted against the load average as a running process. |
| 1654 | * We never actually get interrupted because kthread_run |
Frederik Schwarzer | c03264a | 2008-12-23 00:54:00 +0100 | [diff] [blame] | 1655 | * disables signal delivery for the created thread. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1656 | */ |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1657 | set_current_state(TASK_INTERRUPTIBLE); |
| 1658 | while (!kthread_should_stop()) { |
Tejun Heo | ee7863b | 2006-05-15 20:57:20 +0900 | [diff] [blame] | 1659 | if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1660 | shost->host_failed != shost->host_busy) { |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1661 | SCSI_LOG_ERROR_RECOVERY(1, |
| 1662 | printk("Error handler scsi_eh_%d sleeping\n", |
| 1663 | shost->host_no)); |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1664 | schedule(); |
| 1665 | set_current_state(TASK_INTERRUPTIBLE); |
| 1666 | continue; |
| 1667 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1668 | |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1669 | __set_current_state(TASK_RUNNING); |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1670 | SCSI_LOG_ERROR_RECOVERY(1, |
| 1671 | printk("Error handler scsi_eh_%d waking up\n", |
| 1672 | shost->host_no)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1673 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | /* |
| 1675 | * We have a host that is failing for some reason. Figure out |
| 1676 | * what we need to do to get it up and online again (if we can). |
| 1677 | * If we fail, we end up taking the thing offline. |
| 1678 | */ |
Christoph Hellwig | 9227c33 | 2006-04-01 19:21:04 +0200 | [diff] [blame] | 1679 | if (shost->transportt->eh_strategy_handler) |
| 1680 | shost->transportt->eh_strategy_handler(shost); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | else |
| 1682 | scsi_unjam_host(shost); |
| 1683 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | /* |
| 1685 | * Note - if the above fails completely, the action is to take |
| 1686 | * individual devices offline and flush the queue of any |
| 1687 | * outstanding requests that may have been pending. When we |
| 1688 | * restart, we restart any I/O to any other devices on the bus |
| 1689 | * which are still online. |
| 1690 | */ |
| 1691 | scsi_restart_operations(shost); |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1692 | set_current_state(TASK_INTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1693 | } |
Steven Rostedt | 461a0ff | 2005-10-19 08:22:13 -0400 | [diff] [blame] | 1694 | __set_current_state(TASK_RUNNING); |
| 1695 | |
Christoph Hellwig | ad42eb1 | 2005-10-29 01:01:55 +0200 | [diff] [blame] | 1696 | SCSI_LOG_ERROR_RECOVERY(1, |
| 1697 | printk("Error handler scsi_eh_%d exiting\n", shost->host_no)); |
James Bottomley | 3ed7a47 | 2005-09-19 09:50:04 -0500 | [diff] [blame] | 1698 | shost->ehandler = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | return 0; |
| 1700 | } |
| 1701 | |
| 1702 | /* |
| 1703 | * Function: scsi_report_bus_reset() |
| 1704 | * |
| 1705 | * Purpose: Utility function used by low-level drivers to report that |
| 1706 | * they have observed a bus reset on the bus being handled. |
| 1707 | * |
| 1708 | * Arguments: shost - Host in question |
| 1709 | * channel - channel on which reset was observed. |
| 1710 | * |
| 1711 | * Returns: Nothing |
| 1712 | * |
| 1713 | * Lock status: Host lock must be held. |
| 1714 | * |
| 1715 | * Notes: This only needs to be called if the reset is one which |
| 1716 | * originates from an unknown location. Resets originated |
| 1717 | * by the mid-level itself don't need to call this, but there |
| 1718 | * should be no harm. |
| 1719 | * |
| 1720 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 1721 | * is properly treated. |
| 1722 | */ |
| 1723 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) |
| 1724 | { |
| 1725 | struct scsi_device *sdev; |
| 1726 | |
| 1727 | __shost_for_each_device(sdev, shost) { |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1728 | if (channel == sdev_channel(sdev)) |
| 1729 | __scsi_report_device_reset(sdev, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1730 | } |
| 1731 | } |
| 1732 | EXPORT_SYMBOL(scsi_report_bus_reset); |
| 1733 | |
| 1734 | /* |
| 1735 | * Function: scsi_report_device_reset() |
| 1736 | * |
| 1737 | * Purpose: Utility function used by low-level drivers to report that |
| 1738 | * they have observed a device reset on the device being handled. |
| 1739 | * |
| 1740 | * Arguments: shost - Host in question |
| 1741 | * channel - channel on which reset was observed |
| 1742 | * target - target on which reset was observed |
| 1743 | * |
| 1744 | * Returns: Nothing |
| 1745 | * |
| 1746 | * Lock status: Host lock must be held |
| 1747 | * |
| 1748 | * Notes: This only needs to be called if the reset is one which |
| 1749 | * originates from an unknown location. Resets originated |
| 1750 | * by the mid-level itself don't need to call this, but there |
| 1751 | * should be no harm. |
| 1752 | * |
| 1753 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 1754 | * is properly treated. |
| 1755 | */ |
| 1756 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) |
| 1757 | { |
| 1758 | struct scsi_device *sdev; |
| 1759 | |
| 1760 | __shost_for_each_device(sdev, shost) { |
Jeff Garzik | 422c0d6 | 2005-10-24 18:05:09 -0400 | [diff] [blame] | 1761 | if (channel == sdev_channel(sdev) && |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1762 | target == sdev_id(sdev)) |
| 1763 | __scsi_report_device_reset(sdev, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1764 | } |
| 1765 | } |
| 1766 | EXPORT_SYMBOL(scsi_report_device_reset); |
| 1767 | |
| 1768 | static void |
| 1769 | scsi_reset_provider_done_command(struct scsi_cmnd *scmd) |
| 1770 | { |
| 1771 | } |
| 1772 | |
| 1773 | /* |
| 1774 | * Function: scsi_reset_provider |
| 1775 | * |
| 1776 | * Purpose: Send requested reset to a bus or device at any phase. |
| 1777 | * |
| 1778 | * Arguments: device - device to send reset to |
| 1779 | * flag - reset type (see scsi.h) |
| 1780 | * |
| 1781 | * Returns: SUCCESS/FAILURE. |
| 1782 | * |
| 1783 | * Notes: This is used by the SCSI Generic driver to provide |
| 1784 | * Bus/Device reset capability. |
| 1785 | */ |
| 1786 | int |
| 1787 | scsi_reset_provider(struct scsi_device *dev, int flag) |
| 1788 | { |
| 1789 | struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL); |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1790 | struct Scsi_Host *shost = dev->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1791 | struct request req; |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1792 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1793 | int rtn; |
| 1794 | |
FUJITA Tomonori | 4f54eec | 2008-04-29 09:54:37 +0200 | [diff] [blame] | 1795 | blk_rq_init(NULL, &req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | scmd->request = &req; |
Christoph Hellwig | b4edcbc | 2005-06-19 13:40:52 +0200 | [diff] [blame] | 1797 | |
Boaz Harrosh | 64a87b2 | 2008-04-30 11:19:47 +0300 | [diff] [blame] | 1798 | scmd->cmnd = req.cmd; |
| 1799 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1800 | scmd->scsi_done = scsi_reset_provider_done_command; |
Boaz Harrosh | 30b0c37 | 2007-12-13 13:47:40 +0200 | [diff] [blame] | 1801 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1802 | |
| 1803 | scmd->cmd_len = 0; |
| 1804 | |
| 1805 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1806 | |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1807 | spin_lock_irqsave(shost->host_lock, flags); |
| 1808 | shost->tmf_in_progress = 1; |
| 1809 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 1810 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1811 | switch (flag) { |
| 1812 | case SCSI_TRY_RESET_DEVICE: |
| 1813 | rtn = scsi_try_bus_device_reset(scmd); |
| 1814 | if (rtn == SUCCESS) |
| 1815 | break; |
| 1816 | /* FALLTHROUGH */ |
Mike Christie | 30bd7df | 2008-02-29 18:25:19 -0600 | [diff] [blame] | 1817 | case SCSI_TRY_RESET_TARGET: |
| 1818 | rtn = scsi_try_target_reset(scmd); |
| 1819 | if (rtn == SUCCESS) |
| 1820 | break; |
| 1821 | /* FALLTHROUGH */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1822 | case SCSI_TRY_RESET_BUS: |
| 1823 | rtn = scsi_try_bus_reset(scmd); |
| 1824 | if (rtn == SUCCESS) |
| 1825 | break; |
| 1826 | /* FALLTHROUGH */ |
| 1827 | case SCSI_TRY_RESET_HOST: |
| 1828 | rtn = scsi_try_host_reset(scmd); |
| 1829 | break; |
| 1830 | default: |
| 1831 | rtn = FAILED; |
| 1832 | } |
| 1833 | |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1834 | spin_lock_irqsave(shost->host_lock, flags); |
| 1835 | shost->tmf_in_progress = 0; |
| 1836 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 1837 | |
| 1838 | /* |
| 1839 | * be sure to wake up anyone who was sleeping or had their queue |
| 1840 | * suspended while we performed the TMF. |
| 1841 | */ |
| 1842 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1843 | printk("%s: waking up host to restart after TMF\n", |
Harvey Harrison | cadbd4a | 2008-07-03 23:47:27 -0700 | [diff] [blame] | 1844 | __func__)); |
James Smart | d7a1bb0 | 2006-03-08 14:50:12 -0500 | [diff] [blame] | 1845 | |
| 1846 | wake_up(&shost->host_wait); |
| 1847 | |
| 1848 | scsi_run_host_queues(shost); |
| 1849 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1850 | scsi_next_command(scmd); |
| 1851 | return rtn; |
| 1852 | } |
| 1853 | EXPORT_SYMBOL(scsi_reset_provider); |
| 1854 | |
| 1855 | /** |
| 1856 | * scsi_normalize_sense - normalize main elements from either fixed or |
| 1857 | * descriptor sense data format into a common format. |
| 1858 | * |
| 1859 | * @sense_buffer: byte array containing sense data returned by device |
| 1860 | * @sb_len: number of valid bytes in sense_buffer |
| 1861 | * @sshdr: pointer to instance of structure that common |
| 1862 | * elements are written to. |
| 1863 | * |
| 1864 | * Notes: |
| 1865 | * The "main elements" from sense data are: response_code, sense_key, |
| 1866 | * asc, ascq and additional_length (only for descriptor format). |
| 1867 | * |
| 1868 | * Typically this function can be called after a device has |
| 1869 | * responded to a SCSI command with the CHECK_CONDITION status. |
| 1870 | * |
| 1871 | * Return value: |
| 1872 | * 1 if valid sense data information found, else 0; |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1873 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1874 | int scsi_normalize_sense(const u8 *sense_buffer, int sb_len, |
| 1875 | struct scsi_sense_hdr *sshdr) |
| 1876 | { |
James Bottomley | 33aa687 | 2005-08-28 11:31:14 -0500 | [diff] [blame] | 1877 | if (!sense_buffer || !sb_len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1878 | return 0; |
| 1879 | |
| 1880 | memset(sshdr, 0, sizeof(struct scsi_sense_hdr)); |
| 1881 | |
| 1882 | sshdr->response_code = (sense_buffer[0] & 0x7f); |
James Bottomley | 33aa687 | 2005-08-28 11:31:14 -0500 | [diff] [blame] | 1883 | |
| 1884 | if (!scsi_sense_valid(sshdr)) |
| 1885 | return 0; |
| 1886 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1887 | if (sshdr->response_code >= 0x72) { |
| 1888 | /* |
| 1889 | * descriptor format |
| 1890 | */ |
| 1891 | if (sb_len > 1) |
| 1892 | sshdr->sense_key = (sense_buffer[1] & 0xf); |
| 1893 | if (sb_len > 2) |
| 1894 | sshdr->asc = sense_buffer[2]; |
| 1895 | if (sb_len > 3) |
| 1896 | sshdr->ascq = sense_buffer[3]; |
| 1897 | if (sb_len > 7) |
| 1898 | sshdr->additional_length = sense_buffer[7]; |
| 1899 | } else { |
| 1900 | /* |
| 1901 | * fixed format |
| 1902 | */ |
| 1903 | if (sb_len > 2) |
| 1904 | sshdr->sense_key = (sense_buffer[2] & 0xf); |
| 1905 | if (sb_len > 7) { |
| 1906 | sb_len = (sb_len < (sense_buffer[7] + 8)) ? |
| 1907 | sb_len : (sense_buffer[7] + 8); |
| 1908 | if (sb_len > 12) |
| 1909 | sshdr->asc = sense_buffer[12]; |
| 1910 | if (sb_len > 13) |
| 1911 | sshdr->ascq = sense_buffer[13]; |
| 1912 | } |
| 1913 | } |
| 1914 | |
| 1915 | return 1; |
| 1916 | } |
| 1917 | EXPORT_SYMBOL(scsi_normalize_sense); |
| 1918 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | int scsi_command_normalize_sense(struct scsi_cmnd *cmd, |
| 1920 | struct scsi_sense_hdr *sshdr) |
| 1921 | { |
| 1922 | return scsi_normalize_sense(cmd->sense_buffer, |
FUJITA Tomonori | b80ca4f | 2008-01-13 15:46:13 +0900 | [diff] [blame] | 1923 | SCSI_SENSE_BUFFERSIZE, sshdr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1924 | } |
| 1925 | EXPORT_SYMBOL(scsi_command_normalize_sense); |
| 1926 | |
| 1927 | /** |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 1928 | * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1929 | * @sense_buffer: byte array of descriptor format sense data |
| 1930 | * @sb_len: number of valid bytes in sense_buffer |
| 1931 | * @desc_type: value of descriptor type to find |
| 1932 | * (e.g. 0 -> information) |
| 1933 | * |
| 1934 | * Notes: |
| 1935 | * only valid when sense data is in descriptor format |
| 1936 | * |
| 1937 | * Return value: |
| 1938 | * pointer to start of (first) descriptor if found else NULL |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1939 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1940 | const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len, |
| 1941 | int desc_type) |
| 1942 | { |
| 1943 | int add_sen_len, add_len, desc_len, k; |
| 1944 | const u8 * descp; |
| 1945 | |
| 1946 | if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7]))) |
| 1947 | return NULL; |
| 1948 | if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73)) |
| 1949 | return NULL; |
| 1950 | add_sen_len = (add_sen_len < (sb_len - 8)) ? |
| 1951 | add_sen_len : (sb_len - 8); |
| 1952 | descp = &sense_buffer[8]; |
| 1953 | for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) { |
| 1954 | descp += desc_len; |
| 1955 | add_len = (k < (add_sen_len - 1)) ? descp[1]: -1; |
| 1956 | desc_len = add_len + 2; |
| 1957 | if (descp[0] == desc_type) |
| 1958 | return descp; |
| 1959 | if (add_len < 0) // short descriptor ?? |
| 1960 | break; |
| 1961 | } |
| 1962 | return NULL; |
| 1963 | } |
| 1964 | EXPORT_SYMBOL(scsi_sense_desc_find); |
| 1965 | |
| 1966 | /** |
Rob Landley | eb44820 | 2007-11-03 13:30:39 -0500 | [diff] [blame] | 1967 | * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1968 | * @sense_buffer: byte array of sense data |
| 1969 | * @sb_len: number of valid bytes in sense_buffer |
| 1970 | * @info_out: pointer to 64 integer where 8 or 4 byte information |
| 1971 | * field will be placed if found. |
| 1972 | * |
| 1973 | * Return value: |
| 1974 | * 1 if information field found, 0 if not found. |
Randy Dunlap | dc8875e | 2007-11-15 15:42:30 -0800 | [diff] [blame] | 1975 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1976 | int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, |
| 1977 | u64 * info_out) |
| 1978 | { |
| 1979 | int j; |
| 1980 | const u8 * ucp; |
| 1981 | u64 ull; |
| 1982 | |
| 1983 | if (sb_len < 7) |
| 1984 | return 0; |
| 1985 | switch (sense_buffer[0] & 0x7f) { |
| 1986 | case 0x70: |
| 1987 | case 0x71: |
| 1988 | if (sense_buffer[0] & 0x80) { |
| 1989 | *info_out = (sense_buffer[3] << 24) + |
| 1990 | (sense_buffer[4] << 16) + |
| 1991 | (sense_buffer[5] << 8) + sense_buffer[6]; |
| 1992 | return 1; |
| 1993 | } else |
| 1994 | return 0; |
| 1995 | case 0x72: |
| 1996 | case 0x73: |
| 1997 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, |
| 1998 | 0 /* info desc */); |
| 1999 | if (ucp && (0xa == ucp[1])) { |
| 2000 | ull = 0; |
| 2001 | for (j = 0; j < 8; ++j) { |
| 2002 | if (j > 0) |
| 2003 | ull <<= 8; |
| 2004 | ull |= ucp[4 + j]; |
| 2005 | } |
| 2006 | *info_out = ull; |
| 2007 | return 1; |
| 2008 | } else |
| 2009 | return 0; |
| 2010 | default: |
| 2011 | return 0; |
| 2012 | } |
| 2013 | } |
| 2014 | EXPORT_SYMBOL(scsi_get_sense_info_fld); |
FUJITA Tomonori | 3bc6a26 | 2008-03-25 09:26:49 +0900 | [diff] [blame] | 2015 | |
| 2016 | /** |
| 2017 | * scsi_build_sense_buffer - build sense data in a buffer |
| 2018 | * @desc: Sense format (non zero == descriptor format, |
| 2019 | * 0 == fixed format) |
| 2020 | * @buf: Where to build sense data |
| 2021 | * @key: Sense key |
| 2022 | * @asc: Additional sense code |
| 2023 | * @ascq: Additional sense code qualifier |
| 2024 | * |
| 2025 | **/ |
| 2026 | void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq) |
| 2027 | { |
| 2028 | if (desc) { |
| 2029 | buf[0] = 0x72; /* descriptor, current */ |
| 2030 | buf[1] = key; |
| 2031 | buf[2] = asc; |
| 2032 | buf[3] = ascq; |
| 2033 | buf[7] = 0; |
| 2034 | } else { |
| 2035 | buf[0] = 0x70; /* fixed, current */ |
| 2036 | buf[2] = key; |
| 2037 | buf[7] = 0xa; |
| 2038 | buf[12] = asc; |
| 2039 | buf[13] = ascq; |
| 2040 | } |
| 2041 | } |
| 2042 | EXPORT_SYMBOL(scsi_build_sense_buffer); |