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 = (struct ata_port *) &host->hostdata[0];
  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 int ata_scsi_error(struct Scsi_Host *host)
 101 {
 102         struct ata_port *ap = (struct ata_port *)&host->hostdata[0];
 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         return 0;
 120 }
 121
 122 /**
 123  *      ata_qc_timeout - Handle timeout of queued command
 124  *      @qc: Command that timed out
 125  *
 126  *      Some part of the kernel (currently, only the SCSI layer)
 127  *      has noticed that the active command on port @ap has not
 128  *      completed after a specified length of time.  Handle this
 129  *      condition by disabling DMA (if necessary) and completing
 130  *      transactions, with error if necessary.
 131  *
 132  *      This also handles the case of the "lost interrupt", where
 133  *      for some reason (possibly hardware bug, possibly driver bug)
 134  *      an interrupt was not delivered to the driver, even though the
 135  *      transaction completed successfully.
 136  *
 137  *      LOCKING:
 138  *      Inherited from SCSI layer (none, can sleep)
 139  */
 140
 141 static void ata_qc_timeout(struct ata_queued_cmd *qc)
 142 {
 143         struct ata_port *ap = qc->ap;
 144         struct ata_host_set *host_set = ap->host_set;
 145         u8 host_stat = 0, drv_stat;
 146         unsigned long flags;
 147
 148         DPRINTK("ENTER\n");
 149
 150         ap->hsm_task_state = HSM_ST_IDLE;
 151
 152         spin_lock_irqsave(&host_set->lock, flags);
 153
 154         switch (qc->tf.protocol) {
 155
 156         case ATA_PROT_DMA:
 157         case ATA_PROT_ATAPI_DMA:
 158                 host_stat = ap->ops->bmdma_status(ap);
 159
 160                 /* before we do anything else, clear DMA-Start bit */
 161                 ap->ops->bmdma_stop(qc);
 162
 163                 /* fall through */
 164
 165         default:
 166                 ata_altstatus(ap);
 167                 drv_stat = ata_chk_status(ap);
 168
 169                 /* ack bmdma irq events */
 170                 ap->ops->irq_clear(ap);
 171
 172                 printk(KERN_ERR "ata%u: command 0x%x timeout, stat 0x%x host_stat 0x%x\n",
 173                        ap->id, qc->tf.command, drv_stat, host_stat);
 174
 175                 ap->hsm_task_state = HSM_ST_IDLE;
 176
 177                 /* complete taskfile transaction */
 178                 qc->err_mask |= AC_ERR_TIMEOUT;
 179                 break;
 180         }
 181
 182         spin_unlock_irqrestore(&host_set->lock, flags);
 183
 184         ata_eh_qc_complete(qc);
 185
 186         DPRINTK("EXIT\n");
 187 }
 188
 189 /**
 190  *      ata_eng_timeout - Handle timeout of queued command
 191  *      @ap: Port on which timed-out command is active
 192  *
 193  *      Some part of the kernel (currently, only the SCSI layer)
 194  *      has noticed that the active command on port @ap has not
 195  *      completed after a specified length of time.  Handle this
 196  *      condition by disabling DMA (if necessary) and completing
 197  *      transactions, with error if necessary.
 198  *
 199  *      This also handles the case of the "lost interrupt", where
 200  *      for some reason (possibly hardware bug, possibly driver bug)
 201  *      an interrupt was not delivered to the driver, even though the
 202  *      transaction completed successfully.
 203  *
 204  *      LOCKING:
 205  *      Inherited from SCSI layer (none, can sleep)
 206  */
 207 void ata_eng_timeout(struct ata_port *ap)
 208 {
 209         DPRINTK("ENTER\n");
 210
 211         ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
 212
 213         DPRINTK("EXIT\n");
 214 }
 215
 216 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
 217 {
 218         /* nada */
 219 }
 220
 221 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
 222 {
 223         struct ata_port *ap = qc->ap;
 224         struct scsi_cmnd *scmd = qc->scsicmd;
 225         unsigned long flags;
 226
 227         spin_lock_irqsave(&ap->host_set->lock, flags);
 228         qc->scsidone = ata_eh_scsidone;
 229         __ata_qc_complete(qc);
 230         WARN_ON(ata_tag_valid(qc->tag));
 231         spin_unlock_irqrestore(&ap->host_set->lock, flags);
 232
 233         scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
 234 }
 235
 236 /**
 237  *      ata_eh_qc_complete - Complete an active ATA command from EH
 238  *      @qc: Command to complete
 239  *
 240  *      Indicate to the mid and upper layers that an ATA command has
 241  *      completed.  To be used from EH.
 242  */
 243 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
 244 {
 245         struct scsi_cmnd *scmd = qc->scsicmd;
 246         scmd->retries = scmd->allowed;
 247         __ata_eh_qc_complete(qc);
 248 }
 249
 250 /**
 251  *      ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
 252  *      @qc: Command to retry
 253  *
 254  *      Indicate to the mid and upper layers that an ATA command
 255  *      should be retried.  To be used from EH.
 256  *
 257  *      SCSI midlayer limits the number of retries to scmd->allowed.
 258  *      scmd->retries is decremented for commands which get retried
 259  *      due to unrelated failures (qc->err_mask is zero).
 260  */
 261 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
 262 {
 263         struct scsi_cmnd *scmd = qc->scsicmd;
 264         if (!qc->err_mask && scmd->retries)
 265                 scmd->retries--;
 266         __ata_eh_qc_complete(qc);
 267 }