Merge raven.il.steeleye.com:/home/jejb/BK/scsi-aic7xxx-2.5

into raven.il.steeleye.com:/home/jejb/BK/scsi-misc-2.5

drivers/scsi/hosts.c

@@ -383,6 +383,7 @@ struct Scsi_Host * scsi_register(Scsi_Host_Template *shost_tp, int xtr_bytes)
 	scsi_assign_lock(shost, &shost->default_lock);
 	INIT_LIST_HEAD(&shost->my_devices);
 	INIT_LIST_HEAD(&shost->eh_cmd_q);
+	INIT_LIST_HEAD(&shost->starved_list);
 
 	init_waitqueue_head(&shost->host_wait);
 	shost->dma_channel = 0xff;
@@ -619,7 +620,6 @@ void scsi_host_busy_dec_and_test(struct Scsi_Host *shost, Scsi_Device *sdev)
 
 	spin_lock_irqsave(shost->host_lock, flags);
 	shost->host_busy--;
-	sdev->device_busy--;
 	if (shost->in_recovery && shost->host_failed &&
 	    (shost->host_busy == shost->host_failed))
 	{

drivers/scsi/hosts.h

@@ -380,6 +380,7 @@ struct Scsi_Host
     struct scsi_host_cmd_pool *cmd_pool;
     spinlock_t            free_list_lock;
     struct list_head      free_list;   /* backup store of cmd structs */
+    struct list_head      starved_list;
 
     spinlock_t		  default_lock;
     spinlock_t		  *host_lock;
@@ -470,12 +471,6 @@ struct Scsi_Host
      */
     unsigned reverse_ordering:1;
 
-    /*
-     * Indicates that one or more devices on this host were starved, and
-     * when the device becomes less busy that we need to feed them.
-     */
-    unsigned some_device_starved:1;
-   
     /*
      * Host has rejected a command because it was busy.
      */

drivers/scsi/scsi.c

@@ -447,8 +447,6 @@ int scsi_dispatch_cmd(Scsi_Cmnd * SCpnt)
 
 	host = SCpnt->device->host;
 
-	ASSERT_LOCK(host->host_lock, 0);
-
 	/* Assign a unique nonzero serial_number. */
 	if (++serial_number == 0)
 		serial_number = 1;
@@ -574,8 +572,6 @@ void scsi_init_cmd_from_req(Scsi_Cmnd * SCpnt, Scsi_Request * SRpnt)
 {
 	struct Scsi_Host *host = SCpnt->device->host;
 
-	ASSERT_LOCK(host->host_lock, 0);
-
 	SCpnt->owner = SCSI_OWNER_MIDLEVEL;
 	SRpnt->sr_command = SCpnt;
 
@@ -819,12 +815,11 @@ void scsi_finish_command(Scsi_Cmnd * SCpnt)
 	struct Scsi_Host *host;
 	Scsi_Device *device;
 	Scsi_Request * SRpnt;
+	unsigned int flags;
 
 	host = SCpnt->device->host;
 	device = SCpnt->device;
 
-	ASSERT_LOCK(host->host_lock, 0);
-
         /*
          * We need to protect the decrement, as otherwise a race condition
          * would exist.  Fiddling with SCpnt isn't a problem as the
@@ -833,6 +828,9 @@ void scsi_finish_command(Scsi_Cmnd * SCpnt)
          * shared.
          */
 	scsi_host_busy_dec_and_test(host, device);
+	spin_lock_irqsave(SCpnt->device->request_queue->queue_lock, flags);
+	SCpnt->device->device_busy--;
+	spin_unlock_irqrestore(SCpnt->device->request_queue->queue_lock, flags);
 
         /*
          * Clear the flags which say that the device/host is no longer

drivers/scsi/scsi.h

@@ -417,7 +417,8 @@ extern void scsi_setup_cmd_retry(Scsi_Cmnd *SCpnt);
 extern void scsi_io_completion(Scsi_Cmnd * SCpnt, int good_sectors,
 			       int block_sectors);
 extern int scsi_queue_insert(struct scsi_cmnd *cmd, int reason);
-extern request_queue_t *scsi_alloc_queue(struct Scsi_Host *shost);
+extern void scsi_queue_next_request(request_queue_t *q, struct scsi_cmnd *cmd);
+extern request_queue_t *scsi_alloc_queue(struct scsi_device *sdev);
 extern void scsi_free_queue(request_queue_t *q);
 extern int scsi_init_queue(void);
 extern void scsi_exit_queue(void);
@@ -530,6 +531,16 @@ struct scsi_dev_info_list {
 extern struct list_head scsi_dev_info_list;
 extern int scsi_dev_info_list_add_str(char *);
 
+/*
+ * scsi_target: representation of a scsi target, for now, this is only
+ * used for single_lun devices. If no one has active IO to the target,
+ * starget_sdev_user is NULL, else it points to the active sdev.
+ */
+struct scsi_target {
+	struct scsi_device *starget_sdev_user;
+	unsigned int starget_refcnt;
+};
+
 /*
  *  The scsi_device struct contains what we know about each given scsi
  *  device.
@@ -554,8 +565,10 @@ struct scsi_device {
 	struct Scsi_Host *host;
 	request_queue_t *request_queue;
 	volatile unsigned short device_busy;	/* commands actually active on low-level */
+	spinlock_t sdev_lock;           /* also the request queue_lock */
 	spinlock_t list_lock;
 	struct list_head cmd_list;	/* queue of in use SCSI Command structures */
+	struct list_head starved_entry;
         Scsi_Cmnd *current_cmnd;	/* currently active command */
 	unsigned short queue_depth;	/* How deep of a queue we want */
 	unsigned short last_queue_full_depth; /* These two are used by */
@@ -586,6 +599,7 @@ struct scsi_device {
 	unsigned char current_tag;	/* current tag */
 //	unsigned char sync_min_period;	/* Not less than this period */
 //	unsigned char sync_max_offset;	/* Not greater than this offset */
+	struct scsi_target      *sdev_target;   /* used only for single_lun */
 
 	unsigned online:1;
 	unsigned writeable:1;
@@ -616,8 +630,6 @@ struct scsi_device {
 					 * because we did a bus reset. */
 	unsigned ten:1;		/* support ten byte read / write */
 	unsigned remap:1;	/* support remapping  */
-	unsigned starved:1;	/* unable to process commands because
-				   host busy */
 //	unsigned sync:1;	/* Sync transfer state, managed by host */
 //	unsigned wide:1;	/* WIDE transfer state, managed by host */
 

drivers/scsi/scsi_error.c

@@ -431,8 +431,6 @@ static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, int timeout)
 	unsigned long flags;
 	int rtn = SUCCESS;
 
-	ASSERT_LOCK(host->host_lock, 0);
-
 	/*
 	 * we will use a queued command if possible, otherwise we will
 	 * emulate the queuing and calling of completion function ourselves.
@@ -1405,8 +1403,6 @@ static void scsi_restart_operations(struct Scsi_Host *shost)
 	struct scsi_device *sdev;
 	unsigned long flags;
 
-	ASSERT_LOCK(shost->host_lock, 0);
-
 	/*
 	 * If the door was locked, we need to insert a door lock request
 	 * onto the head of the SCSI request queue for the device.  There
@@ -1434,18 +1430,11 @@ static void scsi_restart_operations(struct Scsi_Host *shost)
 	 * now that error recovery is done, we will need to ensure that these
 	 * requests are started.
 	 */
-	spin_lock_irqsave(shost->host_lock, flags);
 	list_for_each_entry(sdev, &shost->my_devices, siblings) {
-		if ((shost->can_queue > 0 &&
-		     (shost->host_busy >= shost->can_queue))
-		    || (shost->host_blocked)
-		    || (shost->host_self_blocked)) {
-			break;
-		}
-
+		spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
 		__blk_run_queue(sdev->request_queue);
+		spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
 	}
-	spin_unlock_irqrestore(shost->host_lock, flags);
 }
 
 /**
@@ -1681,6 +1670,7 @@ scsi_reset_provider(struct scsi_device *dev, int flag)
 	struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
 	struct request req;
 	int rtn;
+	struct request_queue *q;
 
 	scmd->request = &req;
 	memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
@@ -1735,6 +1725,8 @@ scsi_reset_provider(struct scsi_device *dev, int flag)
 	}
 
 	scsi_delete_timer(scmd);
+	q = scmd->device->request_queue;
 	scsi_put_command(scmd);
+	scsi_queue_next_request(q, NULL);
 	return rtn;
 }

drivers/scsi/scsi_lib.c

@@ -92,6 +92,7 @@ int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
 {
 	struct Scsi_Host *host = cmd->device->host;
 	struct scsi_device *device = cmd->device;
+	unsigned long flags;
 
 	SCSI_LOG_MLQUEUE(1,
 		 printk("Inserting command %p into mlqueue\n", cmd));
@@ -130,6 +131,9 @@ int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
 	 * Decrement the counters, since these commands are no longer
 	 * active on the host/device.
 	 */
+	spin_lock_irqsave(device->request_queue->queue_lock, flags);
+	device->device_busy--;
+	spin_unlock_irqrestore(device->request_queue->queue_lock, flags);
 	scsi_host_busy_dec_and_test(host, device);
 
 	/*
@@ -174,14 +178,18 @@ void scsi_do_req(struct scsi_request *sreq, const void *cmnd,
 		 void (*done)(struct scsi_cmnd *),
 		 int timeout, int retries)
 {
+	struct request_queue *q;
+
 	/*
 	 * If the upper level driver is reusing these things, then
 	 * we should release the low-level block now.  Another one will
 	 * be allocated later when this request is getting queued.
 	 */
 	if (sreq->sr_command) {
+		q = sreq->sr_command->device->request_queue;
 		scsi_put_command(sreq->sr_command);
 		sreq->sr_command = NULL;
+		scsi_queue_next_request(q, NULL);
 	}
 
 	/*
@@ -228,6 +236,7 @@ static void scsi_wait_done(struct scsi_cmnd *cmd)
 void scsi_wait_req(struct scsi_request *sreq, const void *cmnd, void *buffer,
 		   unsigned bufflen, int timeout, int retries)
 {
+	struct request_queue *q;
 	DECLARE_COMPLETION(wait);
 	
 	sreq->sr_request->waiting = &wait;
@@ -239,7 +248,9 @@ void scsi_wait_req(struct scsi_request *sreq, const void *cmnd, void *buffer,
 	sreq->sr_request->waiting = NULL;
 
 	if (sreq->sr_command) {
+		q = sreq->sr_command->device->request_queue;
 		scsi_put_command(sreq->sr_command);
+		scsi_queue_next_request(q, NULL);
 		sreq->sr_command = NULL;
 	}
 }
@@ -315,6 +326,53 @@ void scsi_setup_cmd_retry(struct scsi_cmnd *cmd)
 	cmd->underflow = cmd->old_underflow;
 }
 
+/*
+ * Called for single_lun devices on IO completion. Clear starget_sdev_user,
+ * and call __blk_run_queue for all the scsi_devices on the target -
+ * including current_sdev first.
+ *
+ * Called with *no* scsi locks held.
+ */
+static void scsi_single_lun_run(struct scsi_device *current_sdev)
+{
+	struct scsi_device *sdev;
+	unsigned int flags, flags2;
+
+	spin_lock_irqsave(current_sdev->host->host_lock, flags);
+	WARN_ON(!current_sdev->sdev_target->starget_sdev_user);
+	current_sdev->sdev_target->starget_sdev_user = NULL;
+	spin_unlock_irqrestore(current_sdev->host->host_lock, flags);
+
+	/*
+	 * Call __blk_run_queue for all LUNs on the target, starting with
+	 * current_sdev. We race with others (to set starget_sdev_user),
+	 * but in most cases, we will be first. Ideally, each LU on the
+	 * target would get some limited time or requests on the target.
+	 */
+	spin_lock_irqsave(current_sdev->request_queue->queue_lock, flags2);
+	__blk_run_queue(current_sdev->request_queue);
+	spin_unlock_irqrestore(current_sdev->request_queue->queue_lock, flags2);
+
+	spin_lock_irqsave(current_sdev->host->host_lock, flags);
+	if (current_sdev->sdev_target->starget_sdev_user) {
+		/*
+		 * After unlock, this races with anyone clearing
+		 * starget_sdev_user, but we (should) always enter this
+		 * function again, avoiding any problems.
+		 */
+		spin_unlock_irqrestore(current_sdev->host->host_lock, flags);
+		return;
+	}
+	spin_unlock_irqrestore(current_sdev->host->host_lock, flags);
+
+	list_for_each_entry(sdev, &current_sdev->same_target_siblings,
+			    same_target_siblings) {
+		spin_lock_irqsave(sdev->request_queue->queue_lock, flags2);
+		__blk_run_queue(sdev->request_queue);
+		spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags2);
+	}
+}
+
 /*
  * Function:    scsi_queue_next_request()
  *
@@ -351,16 +409,12 @@ void scsi_setup_cmd_retry(struct scsi_cmnd *cmd)
  *		permutations grows as 2**N, and if too many more special cases
  *		get added, we start to get screwed.
  */
-static void scsi_queue_next_request(request_queue_t *q, struct scsi_cmnd *cmd)
+void scsi_queue_next_request(request_queue_t *q, struct scsi_cmnd *cmd)
 {
-	struct scsi_device *sdev, *sdev2;
+	struct scsi_device *sdev;
 	struct Scsi_Host *shost;
 	unsigned long flags;
-	int all_clear;
-
-	ASSERT_LOCK(q->queue_lock, 0);
 
-	spin_lock_irqsave(q->queue_lock, flags);
 	if (cmd != NULL) {
 
 		/*
@@ -369,6 +423,7 @@ static void scsi_queue_next_request(request_queue_t *q, struct scsi_cmnd *cmd)
 		 * in which case we need to request the blocks that come after
 		 * the bad sector.
 		 */
+		spin_lock_irqsave(q->queue_lock, flags);
 		cmd->request->special = cmd;
 		if (blk_rq_tagged(cmd->request))
 			blk_queue_end_tag(q, cmd->request);
@@ -381,62 +436,52 @@ static void scsi_queue_next_request(request_queue_t *q, struct scsi_cmnd *cmd)
 		cmd->request->flags |= REQ_SPECIAL;
 		cmd->request->flags &= ~REQ_DONTPREP;
 		__elv_add_request(q, cmd->request, 0, 0);
+		spin_unlock_irqrestore(q->queue_lock, flags);
 	}
 
-	/*
-	 * Just hit the requeue function for the queue.
-	 */
-	__blk_run_queue(q);
-
 	sdev = q->queuedata;
-	shost = sdev->host;
 
-	/*
-	 * If this is a single-lun device, and we are currently finished
-	 * with this device, then see if we need to get another device
-	 * started.  FIXME(eric) - if this function gets too cluttered
-	 * with special case code, then spin off separate versions and
-	 * use function pointers to pick the right one.
-	 */
-	if (sdev->single_lun && sdev->device_busy == 0 &&
-	    !shost->host_blocked && !shost->host_self_blocked &&
-	    !((shost->can_queue > 0) && (shost->host_busy >= shost->can_queue))
-	    && elv_queue_empty(q)) {
-		list_for_each_entry(sdev2, &sdev->same_target_siblings,
-			       same_target_siblings) {
-			if (!sdev2->device_blocked &&
-			    !elv_queue_empty(sdev2->request_queue)) {
-				__blk_run_queue(sdev2->request_queue);
-				break;
-			}
-		}
-	}
+	if (sdev->single_lun)
+		scsi_single_lun_run(sdev);
 
-	/*
-	 * Now see whether there are other devices on the bus which
-	 * might be starved.  If so, hit the request function.  If we
-	 * don't find any, then it is safe to reset the flag.  If we
-	 * find any device that it is starved, it isn't safe to reset the
-	 * flag as the queue function releases the lock and thus some
-	 * other device might have become starved along the way.
-	 */
-	all_clear = 1;
-	if (shost->some_device_starved) {
-		list_for_each_entry(sdev, &shost->my_devices, siblings) {
-			if (shost->can_queue > 0 &&
-			    shost->host_busy >= shost->can_queue)
-				break;
-			if (shost->host_blocked || shost->host_self_blocked)
-				break;
-			if (sdev->device_blocked || !sdev->starved)
-				continue;
-			__blk_run_queue(sdev->request_queue);
-			all_clear = 0;
-		}
+	shost = sdev->host;
+	spin_lock_irqsave(shost->host_lock, flags);
+	while (!list_empty(&shost->starved_list) &&
+	       !shost->host_blocked && !shost->host_self_blocked &&
+		!((shost->can_queue > 0) &&
+		  (shost->host_busy >= shost->can_queue))) {
+		/*
+		 * As long as shost is accepting commands and we have
+		 * starved queues, call __blk_run_queue. scsi_request_fn
+		 * drops the queue_lock and can add us back to the
+		 * starved_list.
+		 *
+		 * host_lock protects the starved_list and starved_entry.
+		 * scsi_request_fn must get the host_lock before checking
+		 * or modifying starved_list or starved_entry.
+		 */
+		sdev = list_entry(shost->starved_list.next,
+					  struct scsi_device, starved_entry);
+		list_del_init(&sdev->starved_entry);
+		spin_unlock_irqrestore(shost->host_lock, flags);
+
+		spin_lock_irqsave(sdev->request_queue->queue_lock, flags);
+		__blk_run_queue(sdev->request_queue);
+		spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
 
-		if (sdev == NULL && all_clear)
-			shost->some_device_starved = 0;
+		spin_lock_irqsave(shost->host_lock, flags);
+		if (unlikely(!list_empty(&sdev->starved_entry)))
+			/*
+			 * sdev lost a race, and was put back on the
+			 * starved list. This is unlikely but without this
+			 * in theory we could loop forever.
+			 */
+			break;
 	}
+	spin_unlock_irqrestore(shost->host_lock, flags);
+
+	spin_lock_irqsave(q->queue_lock, flags);
+	__blk_run_queue(q);
 	spin_unlock_irqrestore(q->queue_lock, flags);
 }
 
@@ -470,8 +515,6 @@ static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int uptodate,
 	struct request *req = cmd->request;
 	unsigned long flags;
 
-	ASSERT_LOCK(q->queue_lock, 0);
-
 	/*
 	 * If there are blocks left over at the end, set up the command
 	 * to queue the remainder of them.
@@ -569,8 +612,6 @@ static void scsi_release_buffers(struct scsi_cmnd *cmd)
 {
 	struct request *req = cmd->request;
 
-	ASSERT_LOCK(cmd->device->host->host_lock, 0);
-
 	/*
 	 * Free up any indirection buffers we allocated for DMA purposes. 
 	 */
@@ -651,8 +692,6 @@ void scsi_io_completion(struct scsi_cmnd *cmd, int good_sectors,
 	 *	would be used if we just wanted to retry, for example.
 	 *
 	 */
-	ASSERT_LOCK(q->queue_lock, 0);
-
 	/*
 	 * Free up any indirection buffers we allocated for DMA purposes. 
 	 * For the case of a READ, we need to copy the data out of the
@@ -923,22 +962,6 @@ static int scsi_init_io(struct scsi_cmnd *cmd)
 	return BLKPREP_KILL;
 }
 
-/*
- * The target associated with myself can only handle one active command at
- * a time. Scan through all of the luns on the same target as myself,
- * return 1 if any are active.
- */
-static int check_all_luns(struct scsi_device *myself)
-{
-	struct scsi_device *sdev;
-
-	list_for_each_entry(sdev, &myself->same_target_siblings,
-			    same_target_siblings)
-		if (sdev->device_busy)
-			return 1;
-	return 0;
-}
-
 static int scsi_prep_fn(struct request_queue *q, struct request *req)
 {
 	struct Scsi_Device_Template *sdt;
@@ -1039,6 +1062,74 @@ static int scsi_prep_fn(struct request_queue *q, struct request *req)
 	return BLKPREP_OK;
 }
 
+/*
+ * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
+ * return 0.
+ *
+ * Called with the queue_lock held.
+ */
+static inline int scsi_dev_queue_ready(struct request_queue *q,
+				  struct scsi_device *sdev)
+{
+	if (sdev->device_busy >= sdev->queue_depth)
+		return 0;
+	if (sdev->device_busy == 0 && sdev->device_blocked) {
+		/*
+		 * unblock after device_blocked iterates to zero
+		 */
+		if (--sdev->device_blocked == 0) {
+			SCSI_LOG_MLQUEUE(3,
+				printk("scsi%d (%d:%d) unblocking device at"
+				       " zero depth\n", sdev->host->host_no,
+				       sdev->id, sdev->lun));
+		} else {
+			blk_plug_device(q);
+			return 0;
+		}
+	}
+	if (sdev->device_blocked)
+		return 0;
+
+	return 1;
+}
+
+/*
+ * scsi_host_queue_ready: if we can send requests to shost, return 1 else
+ * return 0. We must end up running the queue again whenever 0 is
+ * returned, else IO can hang.
+ *
+ * Called with host_lock held.
+ */
+static inline int scsi_host_queue_ready(struct request_queue *q,
+				   struct Scsi_Host *shost,
+				   struct scsi_device *sdev)
+{
+	if (shost->in_recovery)
+		return 0;
+	if (shost->host_busy == 0 && shost->host_blocked) {
+		/*
+		 * unblock after host_blocked iterates to zero
+		 */
+		if (--shost->host_blocked == 0) {
+			SCSI_LOG_MLQUEUE(3,
+				printk("scsi%d unblocking host at zero depth\n",
+					shost->host_no));
+		} else {
+			blk_plug_device(q);
+			return 0;
+		}
+	}
+	if (!list_empty(&sdev->starved_entry))
+		return 0;
+	if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
+	    shost->host_blocked || shost->host_self_blocked) {
+		list_add_tail(&sdev->starved_entry, &shost->starved_list);
+		return 0;
+	}
+
+	return 1;
+}
+
 /*
  * Function:    scsi_request_fn()
  *
@@ -1056,92 +1147,61 @@ static void scsi_request_fn(request_queue_t *q)
 	struct Scsi_Host *shost = sdev->host;
 	struct scsi_cmnd *cmd;
 	struct request *req;
-
-	ASSERT_LOCK(q->queue_lock, 1);
+	unsigned int flags;
 
 	/*
 	 * To start with, we keep looping until the queue is empty, or until
 	 * the host is no longer able to accept any more requests.
 	 */
 	for (;;) {
+		if (blk_queue_plugged(q))
+			goto completed;
+
 		/*
-		 * Check this again - each time we loop through we will have
-		 * released the lock and grabbed it again, so each time
-		 * we need to check to see if the queue is plugged or not.
+		 * get next queueable request.  We do this early to make sure
+		 * that the request is fully prepared even if we cannot 
+		 * accept it.
 		 */
-		if (shost->in_recovery || blk_queue_plugged(q))
-			return;
-
-		if (sdev->device_busy >= sdev->queue_depth)
-			break;
-
-		if (sdev->single_lun && check_all_luns(sdev))
-			break;
+		req = elv_next_request(q);
 
-		if (shost->host_busy == 0 && shost->host_blocked) {
-			/* unblock after host_blocked iterates to zero */
-			if (--shost->host_blocked == 0) {
-				SCSI_LOG_MLQUEUE(3,
-					printk("scsi%d unblocking host at zero depth\n",
-						shost->host_no));
-			} else {
+		if (!req) {
+			/*
+			 * If the device is busy, a returning I/O will
+			 * restart the queue. Otherwise, we have to plug
+			 * the queue
+			 */
+			if (sdev->device_busy == 0)
 				blk_plug_device(q);
-				break;
-			}
+			goto completed;
 		}
 
-		if (sdev->device_busy == 0 && sdev->device_blocked) {
-			/* unblock after device_blocked iterates to zero */
-			if (--sdev->device_blocked == 0) {
-				SCSI_LOG_MLQUEUE(3,
-					printk("scsi%d (%d:%d) unblocking device at zero depth\n",
-						shost->host_no, sdev->id, sdev->lun));
-			} else {
-				blk_plug_device(q);
-				break;
-			}
-		}
+		if (!scsi_dev_queue_ready(q, sdev))
+			goto completed;
 
 		/*
-		 * If the device cannot accept another request, then quit.
+		 * Remove the request from the request list.
 		 */
-		if (sdev->device_blocked)
-			break;
-		if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
-		    shost->host_blocked || shost->host_self_blocked) {
-			/*
-			 * If we are unable to process any commands at all for
-			 * this device, then we consider it to be starved.
-			 * What this means is that there are no outstanding
-			 * commands for this device and hence we need a
-			 * little help getting it started again
-			 * once the host isn't quite so busy.
-			 */
-			if (sdev->device_busy == 0) {
-				sdev->starved = 1;
-				shost->some_device_starved = 1;
-			}
-			break;
-		} else
-			sdev->starved = 0;
+		if (!(blk_queue_tagged(q) && (blk_queue_start_tag(q, req) == 0)))
+			blkdev_dequeue_request(req);
 
-		/*
-		 * get next queueable request.  We do this early to make sure
-		 * that the request is fully prepared even if we cannot 
-		 * accept it.  If there is no request, we'll detect this
-		 * lower down.
-		 */
-		req = elv_next_request(q);
+		sdev->device_busy++;
+		spin_unlock_irq(q->queue_lock);
 
-		if (!req) {
-			/* If the device is busy, a returning I/O
-			 * will restart the queue.  Otherwise, we have
-			 * to plug the queue */
-			if(sdev->device_busy == 0)
-				blk_plug_device(q);
-			break;
+		spin_lock_irqsave(shost->host_lock, flags);
+		if (!scsi_host_queue_ready(q, shost, sdev))
+			goto host_lock_held;
+
+		if (sdev->single_lun) {
+			if (sdev->sdev_target->starget_sdev_user &&
+			    (sdev->sdev_target->starget_sdev_user != sdev))
+				goto host_lock_held;
+			else
+				sdev->sdev_target->starget_sdev_user = sdev;
 		}
 
+		shost->host_busy++;
+		spin_unlock_irqrestore(shost->host_lock, flags);
+
 		cmd = req->special;
 
 		/*
@@ -1150,25 +1210,6 @@ static void scsi_request_fn(request_queue_t *q)
 		 */
 		BUG_ON(!cmd);
 
-		/*
-		 * Finally, before we release the lock, we copy the
-		 * request to the command block, and remove the
-		 * request from the request list.  Note that we always
-		 * operate on the queue head - there is absolutely no
-		 * reason to search the list, because all of the
-		 * commands in this queue are for the same device.
-		 */
-		if (!(blk_queue_tagged(q) && (blk_queue_start_tag(q, req) == 0)))
-			blkdev_dequeue_request(req);
-	
-		/*
-		 * Now bump the usage count for both the host and the
-		 * device.
-		 */
-		shost->host_busy++;
-		sdev->device_busy++;
-		spin_unlock_irq(q->queue_lock);
-
 		/*
 		 * Finally, initialize any error handling parameters, and set up
 		 * the timers for timeouts.
@@ -1186,6 +1227,24 @@ static void scsi_request_fn(request_queue_t *q)
 		 */
 		spin_lock_irq(q->queue_lock);
 	}
+completed:
+	return;
+
+host_lock_held:
+	spin_unlock_irqrestore(shost->host_lock, flags);
+	/*
+	 * lock q, handle tag, requeue req, and decrement device_busy. We
+	 * must return with queue_lock held.
+	 *
+	 * Decrementing device_busy without checking it is OK, as all such
+	 * cases (host limits or settings) should run the queue at some
+	 * later time.
+	 */
+	spin_lock_irq(q->queue_lock);
+	if (blk_rq_tagged(req))
+		blk_queue_end_tag(q, req);
+	__elv_add_request(q, req, 0, 0);
+	sdev->device_busy--;
 }
 
 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
@@ -1207,15 +1266,20 @@ u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
 	return BLK_BOUNCE_HIGH;
 }
 
-request_queue_t *scsi_alloc_queue(struct Scsi_Host *shost)
+request_queue_t *scsi_alloc_queue(struct scsi_device *sdev)
 {
 	request_queue_t *q;
+	struct Scsi_Host *shost;
 
 	q = kmalloc(sizeof(*q), GFP_ATOMIC);
 	if (!q)
 		return NULL;
 	memset(q, 0, sizeof(*q));
 
+	/*
+	 * XXX move host code to scsi_register
+	 */
+	shost = sdev->host;
 	if (!shost->max_sectors) {
 		/*
 		 * Driver imposes no hard sector transfer limit.
@@ -1224,7 +1288,7 @@ request_queue_t *scsi_alloc_queue(struct Scsi_Host *shost)
 		shost->max_sectors = SCSI_DEFAULT_MAX_SECTORS;
 	}
 
-	blk_init_queue(q, scsi_request_fn, shost->host_lock);
+	blk_init_queue(q, scsi_request_fn, &sdev->sdev_lock);
 	blk_queue_prep_rq(q, scsi_prep_fn);
 
 	blk_queue_max_hw_segments(q, shost->sg_tablesize);

drivers/scsi/scsi_scan.c

@@ -387,7 +387,7 @@ static void print_inquiry(unsigned char *inq_result)
  *     Scsi_Device pointer, or NULL on failure.
  **/
 static struct scsi_device *scsi_alloc_sdev(struct Scsi_Host *shost,
-	       	struct request_queue **q, uint channel, uint id, uint lun)
+	       	uint channel, uint id, uint lun)
 {
 	struct scsi_device *sdev, *device;
 
@@ -407,6 +407,7 @@ static struct scsi_device *scsi_alloc_sdev(struct Scsi_Host *shost,
 	INIT_LIST_HEAD(&sdev->siblings);
 	INIT_LIST_HEAD(&sdev->same_target_siblings);
 	INIT_LIST_HEAD(&sdev->cmd_list);
+	INIT_LIST_HEAD(&sdev->starved_entry);
 	spin_lock_init(&sdev->list_lock);
 
 	/*
@@ -421,14 +422,10 @@ static struct scsi_device *scsi_alloc_sdev(struct Scsi_Host *shost,
 	 */
 	sdev->borken = 1;
 
-	if (!q || *q == NULL) {
-		sdev->request_queue = scsi_alloc_queue(shost);
-		if (!sdev->request_queue)
-			goto out_free_dev;
-	} else {
-		sdev->request_queue = *q;
-		*q = NULL;
-	}
+	spin_lock_init(&sdev->sdev_lock);
+	sdev->request_queue = scsi_alloc_queue(sdev);
+	if (!sdev->request_queue)
+		goto out_free_dev;
 
 	sdev->request_queue->queuedata = sdev;
 	scsi_adjust_queue_depth(sdev, 0, sdev->host->cmd_per_lun);
@@ -468,10 +465,7 @@ static struct scsi_device *scsi_alloc_sdev(struct Scsi_Host *shost,
 	return sdev;
 
 out_free_queue:
-	if (q && sdev->request_queue) {
-		*q = sdev->request_queue;
-		sdev->request_queue = NULL;
-	} else if (sdev->request_queue)
+	if (sdev->request_queue)
 		scsi_free_queue(sdev->request_queue);
 
 out_free_dev:
@@ -491,6 +485,8 @@ static struct scsi_device *scsi_alloc_sdev(struct Scsi_Host *shost,
  **/
 static void scsi_free_sdev(struct scsi_device *sdev)
 {
+	unsigned int flags;
+
 	list_del(&sdev->siblings);
 	list_del(&sdev->same_target_siblings);
 
@@ -500,6 +496,15 @@ static void scsi_free_sdev(struct scsi_device *sdev)
 		sdev->host->hostt->slave_destroy(sdev);
 	if (sdev->inquiry)
 		kfree(sdev->inquiry);
+	spin_lock_irqsave(sdev->host->host_lock, flags);
+	list_del(&sdev->starved_entry);
+	if (sdev->single_lun) {
+		sdev->sdev_target->starget_refcnt--;
+		if (sdev->sdev_target->starget_refcnt == 0)
+			kfree(sdev->sdev_target);
+	}
+	spin_unlock_irqrestore(sdev->host->host_lock, flags);
+
 	kfree(sdev);
 }
 
@@ -1135,6 +1140,10 @@ static void scsi_probe_lun(Scsi_Request *sreq, char *inq_result,
 static int scsi_add_lun(Scsi_Device *sdev, Scsi_Request *sreq,
 		char *inq_result, int *bflags)
 {
+	struct scsi_device *sdev_sibling;
+	struct scsi_target *starget;
+	unsigned int flags;
+
 	/*
 	 * XXX do not save the inquiry, since it can change underneath us,
 	 * save just vendor/model/rev.
@@ -1256,10 +1265,38 @@ static int scsi_add_lun(Scsi_Device *sdev, Scsi_Request *sreq,
 
 	/*
 	 * If we need to allow I/O to only one of the luns attached to
-	 * this target id at a time, then we set this flag.
+	 * this target id at a time set single_lun, and allocate or modify
+	 * sdev_target.
 	 */
-	if (*bflags & BLIST_SINGLELUN)
+	if (*bflags & BLIST_SINGLELUN) {
 		sdev->single_lun = 1;
+		spin_lock_irqsave(sdev->host->host_lock, flags);
+		starget = NULL;
+		/*
+		 * Search for an existing target for this sdev.
+		 */
+		list_for_each_entry(sdev_sibling, &sdev->same_target_siblings,
+				    same_target_siblings) {
+			if (sdev_sibling->sdev_target != NULL) {
+				starget = sdev_sibling->sdev_target;
+				break;
+			}
+		}
+		if (!starget) {
+			starget = kmalloc(sizeof(*starget), GFP_ATOMIC);
+			if (!starget) {
+				printk(ALLOC_FAILURE_MSG, __FUNCTION__);
+				spin_unlock_irqrestore(sdev->host->host_lock,
+						       flags);
+				return SCSI_SCAN_NO_RESPONSE;
+			}
+			starget->starget_refcnt = 0;
+			starget->starget_sdev_user = NULL;
+		}
+		starget->starget_refcnt++;
+		sdev->sdev_target = starget;
+		spin_unlock_irqrestore(sdev->host->host_lock, flags);
+	}
 
 	/* if the device needs this changing, it may do so in the detect
 	 * function */
@@ -1288,15 +1325,15 @@ static int scsi_add_lun(Scsi_Device *sdev, Scsi_Request *sreq,
  *     SCSI_SCAN_LUN_PRESENT: a new Scsi_Device was allocated and initialized
  **/
 static int scsi_probe_and_add_lun(struct Scsi_Host *host,
-		struct request_queue **q, uint channel, uint id, uint lun,
-		int *bflagsp, struct scsi_device **sdevp)
+		uint channel, uint id, uint lun, int *bflagsp,
+		struct scsi_device **sdevp)
 {
 	struct scsi_device *sdev;
 	struct scsi_request *sreq;
 	unsigned char *result;
 	int bflags, res = SCSI_SCAN_NO_RESPONSE;
 
-	sdev = scsi_alloc_sdev(host, q, channel, id, lun);
+	sdev = scsi_alloc_sdev(host, channel, id, lun);
 	if (!sdev)
 		goto out;
 	sreq = scsi_allocate_request(sdev);
@@ -1350,13 +1387,8 @@ static int scsi_probe_and_add_lun(struct Scsi_Host *host,
 	if (res == SCSI_SCAN_LUN_PRESENT) {
 		if (sdevp)
 			*sdevp = sdev;
-	} else {
-		if (q) {
-			*q = sdev->request_queue;
-			sdev->request_queue = NULL;
-		}
+	} else
 		scsi_free_sdev(sdev);
-	}
  out:
 	return res;
 }
@@ -1374,9 +1406,8 @@ static int scsi_probe_and_add_lun(struct Scsi_Host *host,
  *
  *     Modifies sdevscan->lun.
  **/
-static void scsi_sequential_lun_scan(struct Scsi_Host *shost,
-	       	struct request_queue **q, uint channel, uint id,
-	       	int bflags, int lun0_res, int scsi_level)
+static void scsi_sequential_lun_scan(struct Scsi_Host *shost, uint channel,
+		uint id, int bflags, int lun0_res, int scsi_level)
 {
 	unsigned int sparse_lun, lun, max_dev_lun;
 
@@ -1444,7 +1475,7 @@ static void scsi_sequential_lun_scan(struct Scsi_Host *shost,
 	 * sparse_lun.
 	 */
 	for (lun = 1; lun < max_dev_lun; ++lun)
-		if ((scsi_probe_and_add_lun(shost, q, channel, id, lun,
+		if ((scsi_probe_and_add_lun(shost, channel, id, lun,
 		      NULL, NULL) != SCSI_SCAN_LUN_PRESENT) && !sparse_lun)
 			return;
 }
@@ -1497,8 +1528,7 @@ static int scsilun_to_int(ScsiLun *scsilun)
  *     0: scan completed (or no memory, so further scanning is futile)
  *     1: no report lun scan, or not configured
  **/
-static int scsi_report_lun_scan(Scsi_Device *sdev, struct request_queue **q,
-		int bflags)
+static int scsi_report_lun_scan(Scsi_Device *sdev, int bflags)
 {
 #ifdef CONFIG_SCSI_REPORT_LUNS
 
@@ -1659,8 +1689,8 @@ static int scsi_report_lun_scan(Scsi_Device *sdev, struct request_queue **q,
 		} else {
 			int res;
 
-			res = scsi_probe_and_add_lun(sdev->host, q,
-				sdev->channel, sdev->id, lun, NULL, NULL);
+			res = scsi_probe_and_add_lun(sdev->host, sdev->channel,
+				sdev->id, lun, NULL, NULL);
 			if (res == SCSI_SCAN_NO_RESPONSE) {
 				/*
 				 * Got some results, but now none, abort.
@@ -1688,8 +1718,7 @@ struct scsi_device *scsi_add_device(struct Scsi_Host *shost,
 	struct scsi_device *sdev;
 	int error = -ENODEV, res;
 
-	res = scsi_probe_and_add_lun(shost, NULL, channel, id, lun,
-			NULL, &sdev);
+	res = scsi_probe_and_add_lun(shost, channel, id, lun, NULL, &sdev);
 	if (res == SCSI_SCAN_LUN_PRESENT)
 		error = scsi_attach_device(sdev);
 
@@ -1730,8 +1759,8 @@ int scsi_remove_device(struct scsi_device *sdev)
  *     First try a REPORT LUN scan, if that does not scan the target, do a
  *     sequential scan of LUNs on the target id.
  **/
-static void scsi_scan_target(struct Scsi_Host *shost, struct request_queue **q,
-			     unsigned int channel, unsigned int id)
+static void scsi_scan_target(struct Scsi_Host *shost, unsigned int channel,
+			     unsigned int id)
 {
 	int bflags = 0;
 	int res;
@@ -1747,14 +1776,14 @@ static void scsi_scan_target(struct Scsi_Host *shost, struct request_queue **q,
 	 * Scan LUN 0, if there is some response, scan further. Ideally, we
 	 * would not configure LUN 0 until all LUNs are scanned.
 	 */
-	res = scsi_probe_and_add_lun(shost, q, channel, id, 0, &bflags, &sdev);
+	res = scsi_probe_and_add_lun(shost, channel, id, 0, &bflags, &sdev);
 	if (res == SCSI_SCAN_LUN_PRESENT) {
-		if (scsi_report_lun_scan(sdev, q, bflags) != 0)
+		if (scsi_report_lun_scan(sdev, bflags) != 0)
 			/*
 			 * The REPORT LUN did not scan the target,
 			 * do a sequential scan.
 			 */
-			scsi_sequential_lun_scan(shost, q, channel, id, bflags,
+			scsi_sequential_lun_scan(shost, channel, id, bflags,
 				       	res, sdev->scsi_level);
 	} else if (res == SCSI_SCAN_TARGET_PRESENT) {
 		/*
@@ -1763,7 +1792,7 @@ static void scsi_scan_target(struct Scsi_Host *shost, struct request_queue **q,
 		 * sequential lun scan with a bflags of SPARSELUN and
 		 * a default scsi level of SCSI_2
 		 */
-		scsi_sequential_lun_scan(shost, q, channel, id, BLIST_SPARSELUN,
+		scsi_sequential_lun_scan(shost, channel, id, BLIST_SPARSELUN,
 				SCSI_SCAN_TARGET_PRESENT, SCSI_2);
 	}
 }
@@ -1778,7 +1807,6 @@ static void scsi_scan_target(struct Scsi_Host *shost, struct request_queue **q,
  **/
 void scsi_scan_host(struct Scsi_Host *shost)
 {
-	struct request_queue *q = NULL;
 	uint channel, id, order_id;
 
 	/*
@@ -1803,12 +1831,9 @@ void scsi_scan_host(struct Scsi_Host *shost)
 				order_id = shost->max_id - id - 1;
 			else
 				order_id = id;
-			scsi_scan_target(shost, &q, channel, order_id);
+			scsi_scan_target(shost, channel, order_id);
 		}
 	}
-
-	if (q)
-		scsi_free_queue(q);
 }
 
 void scsi_forget_host(struct Scsi_Host *shost)
@@ -1847,7 +1872,7 @@ struct scsi_device *scsi_get_host_dev(struct Scsi_Host *shost)
 {
 	struct scsi_device *sdev;
 
-	sdev = scsi_alloc_sdev(shost, NULL, 0, shost->this_id, 0);
+	sdev = scsi_alloc_sdev(shost, 0, shost->this_id, 0);
 	if (sdev) {
 		sdev->borken = 0;
 	}

drivers/scsi/scsi_syms.c

@@ -60,6 +60,8 @@ EXPORT_SYMBOL(scsi_allocate_request);
 EXPORT_SYMBOL(scsi_release_request);
 EXPORT_SYMBOL(scsi_wait_req);
 EXPORT_SYMBOL(scsi_do_req);
+EXPORT_SYMBOL(scsi_get_command);
+EXPORT_SYMBOL(scsi_put_command);
 
 EXPORT_SYMBOL(scsi_report_bus_reset);
 EXPORT_SYMBOL(scsi_block_requests);

drivers/scsi/sd.c

@@ -123,7 +123,7 @@ static int sd_major(int major_idx)
 	case 1 ... 7:
 		return SCSI_DISK1_MAJOR + major_idx - 1;
 	case 8 ... 15:
-		return SCSI_DISK8_MAJOR + major_idx;
+		return SCSI_DISK8_MAJOR + major_idx - 8;
 	default:
 		BUG();
 		return 0;	/* shut up gcc */