drivers/scsi/libata-eh.c

   1 /*
   2  *  libata-eh.c - libata error handling
   3  *
   4  *  Maintained by:  Jeff Garzik <jgarzik@pobox.com>
   5  *                  Please ALWAYS copy linux-ide@vger.kernel.org
   6  *                  on emails.
   7  *
   8  *  Copyright 2006 Tejun Heo <htejun@gmail.com>
   9  *
  10  *
  11  *  This program is free software; you can redistribute it and/or
  12  *  modify it under the terms of the GNU General Public License as
  13  *  published by the Free Software Foundation; either version 2, or
  14  *  (at your option) any later version.
  15  *
  16  *  This program is distributed in the hope that it will be useful,
  17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  *  General Public License for more details.
  20  *
  21  *  You should have received a copy of the GNU General Public License
  22  *  along with this program; see the file COPYING.  If not, write to
  23  *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
  24  *  USA.
  25  *
  26  *
  27  *  libata documentation is available via 'make {ps|pdf}docs',
  28  *  as Documentation/DocBook/libata.*
  29  *
  30  *  Hardware documentation available from http://www.t13.org/ and
  31  *  http://www.sata-io.org/
  32  *
  33  */
  34
  35 #include <linux/config.h>
  36 #include <linux/kernel.h>
  37 #include <scsi/scsi.h>
  38 #include <scsi/scsi_host.h>
  39 #include <scsi/scsi_eh.h>
  40 #include <scsi/scsi_device.h>
  41 #include <scsi/scsi_cmnd.h>
  42
  43 #include <linux/libata.h>
  44
  45 #include "libata.h"
  46
  47 /**
  48  *      ata_scsi_timed_out - SCSI layer time out callback
  49  *      @cmd: timed out SCSI command
  50  *
  51  *      Handles SCSI layer timeout.  We race with normal completion of
  52  *      the qc for @cmd.  If the qc is already gone, we lose and let
  53  *      the scsi command finish (EH_HANDLED).  Otherwise, the qc has
  54  *      timed out and EH should be invoked.  Prevent ata_qc_complete()
  55  *      from finishing it by setting EH_SCHEDULED and return
  56  *      EH_NOT_HANDLED.
  57  *
  58  *      LOCKING:
  59  *      Called from timer context
  60  *
  61  *      RETURNS:
  62  *      EH_HANDLED or EH_NOT_HANDLED
  63  */
  64 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
  65 {
  66         struct Scsi_Host *host = cmd->device->host;
  67         struct ata_port *ap = ata_shost_to_port(host);
  68         unsigned long flags;
  69         struct ata_queued_cmd *qc;
  70         enum scsi_eh_timer_return ret = EH_HANDLED;
  71
  72         DPRINTK("ENTER\n");
  73
  74         spin_lock_irqsave(&ap->host_set->lock, flags);
  75         qc = ata_qc_from_tag(ap, ap->active_tag);
  76         if (qc) {
  77                 WARN_ON(qc->scsicmd != cmd);
  78                 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
  79                 qc->err_mask |= AC_ERR_TIMEOUT;
  80                 ret = EH_NOT_HANDLED;
  81         }
  82         spin_unlock_irqrestore(&ap->host_set->lock, flags);
  83
  84         DPRINTK("EXIT, ret=%d\n", ret);
  85         return ret;
  86 }
  87
  88 /**
  89  *      ata_scsi_error - SCSI layer error handler callback
  90  *      @host: SCSI host on which error occurred
  91  *
  92  *      Handles SCSI-layer-thrown error events.
  93  *
  94  *      LOCKING:
  95  *      Inherited from SCSI layer (none, can sleep)
  96  *
  97  *      RETURNS:
  98  *      Zero.
  99  */
 100 void ata_scsi_error(struct Scsi_Host *host)
 101 {
 102         struct ata_port *ap = ata_shost_to_port(host);
 103
 104         DPRINTK("ENTER\n");
 105
 106         /* synchronize with IRQ handler and port task */
 107         spin_unlock_wait(&ap->host_set->lock);
 108         ata_port_flush_task(ap);
 109
 110         WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
 111
 112         ap->ops->eng_timeout(ap);
 113
 114         WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
 115
 116         scsi_eh_flush_done_q(&ap->eh_done_q);
 117
 118         DPRINTK("EXIT\n");
 119 }
 120
 121 /**
 122  *      ata_qc_timeout - Handle timeout of queued command
 123  *      @qc: Command that timed out
 124  *
 125  *      Some part of the kernel (currently, only the SCSI layer)
 126  *      has noticed that the active command on port @ap has not
 127  *      completed after a specified length of time.  Handle this
 128  *      condition by disabling DMA (if necessary) and completing
 129  *      transactions, with error if necessary.
 130  *
 131  *      This also handles the case of the "lost interrupt", where
 132  *      for some reason (possibly hardware bug, possibly driver bug)
 133  *      an interrupt was not delivered to the driver, even though the
 134  *      transaction completed successfully.
 135  *
 136  *      LOCKING:
 137  *      Inherited from SCSI layer (none, can sleep)
 138  */
 139 static void ata_qc_timeout(struct ata_queued_cmd *qc)
 140 {
 141         struct ata_port *ap = qc->ap;
 142         struct ata_host_set *host_set = ap->host_set;
 143         u8 host_stat = 0, drv_stat;
 144         unsigned long flags;
 145
 146         DPRINTK("ENTER\n");
 147
 148         ap->hsm_task_state = HSM_ST_IDLE;
 149
 150         spin_lock_irqsave(&host_set->lock, flags);
 151
 152         switch (qc->tf.protocol) {
 153
 154         case ATA_PROT_DMA:
 155         case ATA_PROT_ATAPI_DMA:
 156                 host_stat = ap->ops->bmdma_status(ap);
 157
 158                 /* before we do anything else, clear DMA-Start bit */
 159                 ap->ops->bmdma_stop(qc);
 160
 161                 /* fall through */
 162
 163         default:
 164                 ata_altstatus(ap);
 165                 drv_stat = ata_chk_status(ap);
 166
 167                 /* ack bmdma irq events */
 168                 ap->ops->irq_clear(ap);
 169
 170                 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
 171                                "stat 0x%x host_stat 0x%x\n",
 172                                qc->tf.command, drv_stat, host_stat);
 173
 174                 /* complete taskfile transaction */
 175                 qc->err_mask |= ac_err_mask(drv_stat);
 176                 break;
 177         }
 178
 179         spin_unlock_irqrestore(&host_set->lock, flags);
 180
 181         ata_eh_qc_complete(qc);
 182
 183         DPRINTK("EXIT\n");
 184 }
 185
 186 /**
 187  *      ata_eng_timeout - Handle timeout of queued command
 188  *      @ap: Port on which timed-out command is active
 189  *
 190  *      Some part of the kernel (currently, only the SCSI layer)
 191  *      has noticed that the active command on port @ap has not
 192  *      completed after a specified length of time.  Handle this
 193  *      condition by disabling DMA (if necessary) and completing
 194  *      transactions, with error if necessary.
 195  *
 196  *      This also handles the case of the "lost interrupt", where
 197  *      for some reason (possibly hardware bug, possibly driver bug)
 198  *      an interrupt was not delivered to the driver, even though the
 199  *      transaction completed successfully.
 200  *
 201  *      LOCKING:
 202  *      Inherited from SCSI layer (none, can sleep)
 203  */
 204 void ata_eng_timeout(struct ata_port *ap)
 205 {
 206         DPRINTK("ENTER\n");
 207
 208         ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
 209
 210         DPRINTK("EXIT\n");
 211 }
 212
 213 /**
 214  *      ata_qc_schedule_eh - schedule qc for error handling
 215  *      @qc: command to schedule error handling for
 216  *
 217  *      Schedule error handling for @qc.  EH will kick in as soon as
 218  *      other commands are drained.
 219  *
 220  *      LOCKING:
 221  *      spin_lock_irqsave(host_set lock)
 222  */
 223 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
 224 {
 225         struct ata_port *ap = qc->ap;
 226
 227         WARN_ON(!ap->ops->error_handler);
 228
 229         qc->flags |= ATA_QCFLAG_FAILED;
 230         qc->ap->flags |= ATA_FLAG_EH_PENDING;
 231
 232         /* The following will fail if timeout has already expired.
 233          * ata_scsi_error() takes care of such scmds on EH entry.
 234          * Note that ATA_QCFLAG_FAILED is unconditionally set after
 235          * this function completes.
 236          */
 237         scsi_req_abort_cmd(qc->scsicmd);
 238 }
 239
 240 /**
 241  *      ata_port_schedule_eh - schedule error handling without a qc
 242  *      @ap: ATA port to schedule EH for
 243  *
 244  *      Schedule error handling for @ap.  EH will kick in as soon as
 245  *      all commands are drained.
 246  *
 247  *      LOCKING:
 248  *      spin_lock_irqsave(host_set lock)
 249  */
 250 void ata_port_schedule_eh(struct ata_port *ap)
 251 {
 252         WARN_ON(!ap->ops->error_handler);
 253
 254         ap->flags |= ATA_FLAG_EH_PENDING;
 255         ata_schedule_scsi_eh(ap->host);
 256
 257         DPRINTK("port EH scheduled\n");
 258 }
 259
 260 /**
 261  *      ata_port_abort - abort all qc's on the port
 262  *      @ap: ATA port to abort qc's for
 263  *
 264  *      Abort all active qc's of @ap and schedule EH.
 265  *
 266  *      LOCKING:
 267  *      spin_lock_irqsave(host_set lock)
 268  *
 269  *      RETURNS:
 270  *      Number of aborted qc's.
 271  */
 272 int ata_port_abort(struct ata_port *ap)
 273 {
 274         int tag, nr_aborted = 0;
 275
 276         WARN_ON(!ap->ops->error_handler);
 277
 278         for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
 279                 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
 280
 281                 if (qc) {
 282                         qc->flags |= ATA_QCFLAG_FAILED;
 283                         ata_qc_complete(qc);
 284                         nr_aborted++;
 285                 }
 286         }
 287
 288         if (!nr_aborted)
 289                 ata_port_schedule_eh(ap);
 290
 291         return nr_aborted;
 292 }
 293
 294 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
 295 {
 296         /* nada */
 297 }
 298
 299 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
 300 {
 301         struct ata_port *ap = qc->ap;
 302         struct scsi_cmnd *scmd = qc->scsicmd;
 303         unsigned long flags;
 304
 305         spin_lock_irqsave(&ap->host_set->lock, flags);
 306         qc->scsidone = ata_eh_scsidone;
 307         __ata_qc_complete(qc);
 308         WARN_ON(ata_tag_valid(qc->tag));
 309         spin_unlock_irqrestore(&ap->host_set->lock, flags);
 310
 311         scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
 312 }
 313
 314 /**
 315  *      ata_eh_qc_complete - Complete an active ATA command from EH
 316  *      @qc: Command to complete
 317  *
 318  *      Indicate to the mid and upper layers that an ATA command has
 319  *      completed.  To be used from EH.
 320  */
 321 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
 322 {
 323         struct scsi_cmnd *scmd = qc->scsicmd;
 324         scmd->retries = scmd->allowed;
 325         __ata_eh_qc_complete(qc);
 326 }
 327
 328 /**
 329  *      ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
 330  *      @qc: Command to retry
 331  *
 332  *      Indicate to the mid and upper layers that an ATA command
 333  *      should be retried.  To be used from EH.
 334  *
 335  *      SCSI midlayer limits the number of retries to scmd->allowed.
 336  *      scmd->retries is decremented for commands which get retried
 337  *      due to unrelated failures (qc->err_mask is zero).
 338  */
 339 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
 340 {
 341         struct scsi_cmnd *scmd = qc->scsicmd;
 342         if (!qc->err_mask && scmd->retries)
 343                 scmd->retries--;
 344         __ata_eh_qc_complete(qc);
 345 }