scsi_error.c

基于组件方式开发操作系统的OSKIT源代码

   */
  switch (status_byte(SCpnt->result))
    {
    case QUEUE_FULL:
      /*
       * The case of trying to send too many commands to a tagged queueing
       * device.
       */
      return ADD_TO_MLQUEUE;
    case GOOD:
    case COMMAND_TERMINATED:
      return SUCCESS;
    case CHECK_CONDITION:
      rtn = scsi_check_sense(SCpnt);
      if( rtn == NEEDS_RETRY )
	{
	  goto maybe_retry;
	}
      return rtn;
    case CONDITION_GOOD:
    case INTERMEDIATE_GOOD:
    case INTERMEDIATE_C_GOOD:
      /*
       * Who knows?  FIXME(eric)
       */
      return SUCCESS;
    case BUSY:
    case RESERVATION_CONFLICT:
      goto maybe_retry;
    default:
      return FAILED;
    }
  return FAILED;

maybe_retry:

  if ((++SCpnt->retries) < SCpnt->allowed)
    {
      return NEEDS_RETRY;
    }
  else
    {
      return FAILED;
    }
}

/*
 * Function:	scsi_eh_completed_normally
 *
 * Purpose:	Examine a command block that has come back from the low-level
 *		and figure out what to do next.
 *
 * Returns:	SUCCESS		- pass on to upper level.
 *		FAILED		- pass on to error handler thread.
 *		RETRY		- command should be retried.
 *		SOFTERR		- command succeeded, but we need to log
 *				  a soft error.
 *
 * Notes:	This is *ONLY* called when we are examining the status
 *		of commands queued during error recovery.  The main
 *		difference here is that we don't allow for the possibility
 *		of retries here, and we are a lot more restrictive about what
 *              we consider acceptable.
 */
STATIC int scsi_eh_completed_normally (Scsi_Cmnd * SCpnt)
{
  int	rtn;
  /*
   * First check the host byte, to see if there is anything in there
   * that would indicate what we need to do.
   */
  if( host_byte(SCpnt->result) == DID_RESET )
    {
     if (SCpnt->flags & IS_RESETTING )
       {
	 /*
	  * OK, this is normal.  We don't know whether in fact the
	  * command in question really needs to be rerun or not - 
	  * if this was the original data command then the answer is yes,
	  * otherwise we just flag it as success.
	  */
	 SCpnt->flags &= ~IS_RESETTING;
	 return NEEDS_RETRY;
       }

     /*
      * Rats.  We are already in the error handler, so we now get to try
      * and figure out what to do next.  If the sense is valid, we have
      * a pretty good idea of what to do.  If not, we mark it as failed.
      */
     return scsi_check_sense (SCpnt);
    }

  if(host_byte(SCpnt->result) != DID_OK )
  {
      return FAILED;
  }

  /*
   * Next, check the message byte.
   */
  if( msg_byte(SCpnt->result) != COMMAND_COMPLETE )
    {
      return FAILED;
    }

  /*
   * Now, check the status byte to see if this indicates anything special.
   */
  switch (status_byte(SCpnt->result))
    {
    case GOOD:
    case COMMAND_TERMINATED:
      return SUCCESS;
    case CHECK_CONDITION:
      rtn = scsi_check_sense(SCpnt);
      if( rtn == NEEDS_RETRY )
	{
	  return FAILED;
	}
      return rtn;
    case CONDITION_GOOD:
    case INTERMEDIATE_GOOD:
    case INTERMEDIATE_C_GOOD:
      /*
       * Who knows?  FIXME(eric)
       */
      return SUCCESS;
    case BUSY:
    case QUEUE_FULL:
    case RESERVATION_CONFLICT:
    default:
      return FAILED;
    }
  return FAILED;
}

/*
 * Function:	scsi_check_sense
 *
 * Purpose:	Examine sense information - give suggestion as to what
 *		we should do with it.
 */
STATIC  int scsi_check_sense (Scsi_Cmnd * SCpnt)
{
    if ( !scsi_sense_valid(SCpnt) ) 
      {
	return FAILED;
      }

    if (SCpnt->sense_buffer[2] & 0xe0)
	return SUCCESS;

    switch (SCpnt->sense_buffer[2] & 0xf)
    {
    case NO_SENSE:
	return SUCCESS;
    case RECOVERED_ERROR:
	return /* SOFT_ERROR */ SUCCESS;

    case ABORTED_COMMAND:
	return NEEDS_RETRY;
    case NOT_READY:
    case UNIT_ATTENTION:
        /*
         * If we are expecting a CC/UA because of a bus reset that we
         * performed, treat this just as a retry.  Otherwise this is
         * information that we should pass up to the upper-level driver
         * so that we can deal with it there.
         */
        if( SCpnt->device->expecting_cc_ua )
        {
            SCpnt->device->expecting_cc_ua = 0;
            return NEEDS_RETRY;
        }
	return SUCCESS;

    /* these three are not supported */
    case COPY_ABORTED:
    case VOLUME_OVERFLOW:
    case MISCOMPARE:
        return SUCCESS;

    case MEDIUM_ERROR:
	return NEEDS_RETRY;

    case ILLEGAL_REQUEST:
    case BLANK_CHECK:
    case DATA_PROTECT:
    case HARDWARE_ERROR:
    default:
	return SUCCESS;
    }
}


/*
 * Function:	scsi_restart_operations
 *
 * Purpose:	Restart IO operations to the specified host.
 *
 * Arguments:	host  - host that we are restarting
 *
 * Returns:	Nothing
 *
 * Notes:	When we entered the error handler, we blocked all further
 *		I/O to this device.  We need to 'reverse' this process.
 */
STATIC void
scsi_restart_operations(struct Scsi_Host * host)
{
  Scsi_Device * SDpnt;

  /*
   * Next free up anything directly waiting upon the host.  This will be
   * requests for character device operations, and also for ioctls to queued
   * block devices.
   */
  SCSI_LOG_ERROR_RECOVERY(5,printk("scsi_error.c: Waking up host to restart\n"));

   wake_up(&host->host_wait);

   /*
    * Finally, block devices need an extra kick in the pants.  This is because
    * the request queueing mechanism may have queued lots of pending requests
    * and there won't be a process waiting in a place where we can simply wake
    * it up.  Thus we simply go through and call the request function to goose
    * the various top level drivers and get things moving again.
    */
   for( SDpnt = host->host_queue; SDpnt; SDpnt = SDpnt->next )
     {
       SCSI_LOG_ERROR_RECOVERY(5,printk("Calling request function to restart things...\n"));

       if( SDpnt->scsi_request_fn != NULL )
	 (*SDpnt->scsi_request_fn)();
     }
}

/*
 * Function:	scsi_unjam_host
 *
 * Purpose:	Attempt to fix a host which has a command that failed for
 *		some reason.
 *
 * Arguments:	host	- host that needs unjamming.
 * 
 * Returns:	Nothing
 *
 * Notes:	When we come in here, we *know* that all commands on the
 *		bus have either completed, failed or timed out.  We also
 *		know that no further commands are being sent to the host,
 *		so things are relatively quiet and we have freedom to
 *		fiddle with things as we wish.
 *
 * Additional note:  This is only the *default* implementation.  It is possible
 *		for individual drivers to supply their own version of this
 *		function, and if the maintainer wishes to do this, it is
 *		strongly suggested that this function be taken as a template
 *		and modified.  This function was designed to correctly handle
 *		problems for about 95% of the different cases out there, and
 *		it should always provide at least a reasonable amount of error
 *		recovery.
 *
 * Note3:       Any command marked 'FAILED' or 'TIMEOUT' must eventually
 *              have scsi_finish_command() called for it.  We do all of
 *              the retry stuff here, so when we restart the host after we
 *              return it should have an empty queue.
 */
STATIC int
scsi_unjam_host(struct Scsi_Host * host)
{
  int           devices_failed;
  int           numfailed;
  int           ourrtn;
  int		rtn = FALSE;
  int		result;
  Scsi_Cmnd   * SCloop;
  Scsi_Cmnd   * SCpnt;
  Scsi_Device * SDpnt;
  Scsi_Device * SDloop;
  Scsi_Cmnd   * SCdone;
  int           timed_out;

  SCdone = NULL;

  /*
   * First, protect against any sort of race condition.  If any of the outstanding
   * commands are in states that indicate that we are not yet blocked (i.e. we are
   * not in a quiet state) then we got woken up in error.  If we ever end up here,
   * we need to re-examine some of the assumptions.
   */
  for(SDpnt=host->host_queue; SDpnt; SDpnt = SDpnt->next)
  {
      for(SCpnt=SDpnt->device_queue; SCpnt; SCpnt = SCpnt->next)
      {
          if( SCpnt->state == SCSI_STATE_FAILED 
              || SCpnt->state == SCSI_STATE_TIMEOUT 
              || SCpnt->state == SCSI_STATE_INITIALIZING
              || SCpnt->state == SCSI_STATE_UNUSED)
          {
              continue;
          }

          /*
           * Rats.  Something is still floating around out there.  This could
           * be the result of the fact that the upper level drivers are still frobbing
           * commands that might have succeeded.  There are two outcomes.  One is that
           * the command block will eventually be freed, and the other one is that
           * the command will be queued and will be finished along the way.
           */
          SCSI_LOG_ERROR_RECOVERY(1,printk("Error handler prematurely woken - commands still active (%p %x %d)\n", SCpnt, SCpnt->state, SCpnt->target));

/*
 *        panic("SCSI Error handler woken too early\n");
 *
 * This is no longer a problem, since now the code cares only about
 * SCSI_STATE_TIMEOUT and SCSI_STATE_FAILED.
 * Other states are useful only to release active commands when devices are
 * set offline. If (host->host_active == host->host_busy) we can safely assume
 * that there are no commands in state other then TIMEOUT od FAILED. (DB)
 *
 * FIXME:
 * It is not easy to release correctly commands according to their state when 
 * devices are set offline, when the state is neither TIMEOUT nor FAILED.
 * When a device is set offline, we can have some command with
 * rq_status=RQ_SCSY_BUSY, owner=SCSI_STATE_HIGHLEVEL, 
 * state=SCSI_STATE_INITIALIZING and the driver module cannot be released.
 * (DB, 17 May 1998)
 */
      }
  }

  /*
   * Next, see if we need to request sense information.  if so,
   * then get it now, so we have a better idea of what to do.
   * FIXME(eric) this has the unfortunate side effect that if a host
   * adapter does not automatically request sense information, that we end
   * up shutting it down before we request it.  All hosts should be doing this
   * anyways, so for now all I have to say is tough noogies if you end up in here.
   * On second thought, this is probably a good idea.  We *really* want to give
   * authors an incentive to automatically request this.
   */
  SCSI_LOG_ERROR_RECOVERY(3,printk("scsi_unjam_host: Checking to see if we need to request sense\n"));

  for(SDpnt=host->host_queue; SDpnt; SDpnt = SDpnt->next)
  {
      for(SCpnt=SDpnt->device_queue; SCpnt; SCpnt = SCpnt->next)
      {
          if( SCpnt->state != SCSI_STATE_FAILED || scsi_sense_valid(SCpnt) )
          {
              continue;
          }

          SCSI_LOG_ERROR_RECOVERY(2,printk("scsi_unjam_host: Requesting sense for %d\n",
                                           SCpnt->target));
          rtn = scsi_request_sense(SCpnt);
          if( rtn != SUCCESS )
          {
              continue;
          }

          SCSI_LOG_ERROR_RECOVERY(3,printk("Sense requested for %p - result %x\n",
                                           SCpnt, SCpnt->result));
          SCSI_LOG_ERROR_RECOVERY(3,print_sense("bh",SCpnt));
                  
          result = scsi_decide_disposition(SCpnt);

          /*
           * If the result was normal, then just pass it along to the
           * upper level.
           */
          if( result == SUCCESS )
          {
              SCpnt->host->host_failed--;
              scsi_eh_finish_command(&SCdone, SCpnt);
          }

          if( result != NEEDS_RETRY )
          {
              continue;
          }

          /* 
           * We only come in here if we want to retry a
           * command.  The test to see whether the command
           * should be retried should be keeping track of the
           * number of tries, so we don't end up looping, of
           * course.  
           */
          SCpnt->state = NEEDS_RETRY;
          rtn = scsi_eh_retry_command(SCpnt);
          if( rtn != SUCCESS )
          {
              continue;
          }

          /*
           * We eventually hand this one back to the top level.
           */
          SCpnt->host->host_failed--;
          scsi_eh_finish_command(&SCdone, SCpnt);
      }
  }

  /*
   * Go through the list of commands and figure out where we stand and how bad things
   * really are.
   */
  numfailed = 0;
  timed_out = 0;
  devices_failed = 0;
  for(SDpnt=host->host_queue; SDpnt; SDpnt = SDpnt->next)
  {
      unsigned int device_error = 0;

      for(SCpnt=SDpnt->device_queue; SCpnt; SCpnt = SCpnt->next)
      {
          if( SCpnt->state == SCSI_STATE_FAILED )
          {
              SCSI_LOG_ERROR_RECOVERY(5,printk("Command to ID %d failed\n", 
                                               SCpnt->target));
              numfailed++;
              device_error++;
          }
          if( SCpnt->state == SCSI_STATE_TIMEOUT )
          {
              SCSI_LOG_ERROR_RECOVERY(5,printk("Command to ID %d timedout\n", 
                                               SCpnt->target));
              timed_out++;
              device_error++;
          }
      }
      if( device_error > 0 )
      {
          devices_failed++;
      }
  }

  SCSI_LOG_ERROR_RECOVERY(2,printk("Total of %d+%d commands on %d devices require eh work\n", 
                                   numfailed, timed_out, devices_failed));

  if( host->host_failed == 0 )
  {
      ourrtn = TRUE;
      goto leave;
  }


  /*
   * Next, try and see whether or not it makes sense to try and abort
   * the running command.  This only works out to be the case if we have
   * one command that has timed out.  If the command simply failed, it
   * makes no sense to try and abort the command, since as far as the
   * host adapter is concerned, it isn't running.
   */

  SCSI_LOG_ERROR_RECOVERY(3,printk("scsi_unjam_host: Checking to see if we want to try abort\n"));

  for(SDpnt=host->host_queue; SDpnt; SDpnt = SDpnt->next)
  {
      for(SCloop=SDpnt->device_queue; SCloop; SCloop = SCloop->next)
      {
          if( SCloop->state != SCSI_STATE_TIMEOUT )
          {
              continue;
          }

	  rtn = scsi_try_to_abort_command(SCloop, ABORT_TIMEOUT);
	  if( rtn == SUCCESS )
          {
	      rtn = scsi_test_unit_ready(SCloop);
              
	      if( rtn == SUCCESS && scsi_unit_is_ready(SCloop) )
              {
		  rtn = scsi_eh_retry_command(SCloop);
                  
		  if( rtn == SUCCESS )
                  {
                      SCloop->host->host_failed--;
		      scsi_eh_finish_command(&SCdone,SCloop);
                  }
              }
          }
      }
  }
  
  /*
   * If we have corrected all of the problems, then we are done.
   */
  if( host->host_failed == 0 )
  {
      ourrtn = TRUE;
      goto leave;
  }

  /*
   * Either the abort wasn't appropriate, or it didn't succeed.
   * Now try a bus device reset.  Still, look to see whether we have
   * multiple devices that are jammed or not - if we have multiple devices,
   * it makes no sense to try BUS_DEVICE_RESET - we really would need
   * to try a BUS_RESET instead.
   *
   * Does this make sense - should we try BDR on each device individually?
   * Yes, definitely.
   */
  SCSI_LOG_ERROR_RECOVERY(3,printk("scsi_unjam_host: Checking to see if we want to try BDR\n"));

  for(SDpnt=host->host_queue; SDpnt; SDpnt = SDpnt->next)
  {
      for(SCloop=SDpnt->device_queue; SCloop; SCloop = SCloop->next)
      {
          if( SCloop->state == SCSI_STATE_FAILED 
              || SCloop->state == SCSI_STATE_TIMEOUT )
          {
              break;
          }
      }