From: Mikulas Patocka <mpatocka@redhat.com>

IO may be submitted to a worker thread with queue_io().  queue_io() sets
DMF_QUEUE_IO_TO_THREAD so that all further IO is queued for the thread. When
the thread finishes its work, it clears DMF_QUEUE_IO_TO_THREAD and from this
point on, requests are submitted from dm_request again.

[FIXME: This patch needs breaking down further with more documentation. - agk]

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

---
 drivers/md/dm.c |   40 +++++++++++++---------------------------
 1 files changed, 13 insertions(+), 27 deletions(-)

Index: linux-2.6.29/drivers/md/dm.c
===================================================================
--- linux-2.6.29.orig/drivers/md/dm.c	2009-04-01 13:51:52.000000000 +0100
+++ linux-2.6.29/drivers/md/dm.c	2009-04-01 13:51:53.000000000 +0100
@@ -436,21 +436,15 @@ static void end_io_acct(struct dm_io *io
 /*
  * Add the bio to the list of deferred io.
  */
-static int queue_io(struct mapped_device *md, struct bio *bio)
+static void queue_io(struct mapped_device *md, struct bio *bio)
 {
 	down_write(&md->io_lock);
-
-	if (!test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) {
-		up_write(&md->io_lock);
-		return 1;
-	}
-
 	spin_lock_irq(&md->deferred_lock);
 	bio_list_add(&md->deferred, bio);
 	spin_unlock_irq(&md->deferred_lock);
-
+	if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
+		queue_work(md->wq, &md->work);
 	up_write(&md->io_lock);
-	return 0;		/* deferred successfully */
 }
 
 /*
@@ -915,7 +909,6 @@ out:
  */
 static int dm_request(struct request_queue *q, struct bio *bio)
 {
-	int r;
 	int rw = bio_data_dir(bio);
 	struct mapped_device *md = q->queuedata;
 	int cpu;
@@ -940,7 +933,7 @@ static int dm_request(struct request_que
 	 * If we're suspended or the thread is processing barriers
 	 * we have to queue this io for later.
 	 */
-	while (test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) {
+	if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))) {
 		up_read(&md->io_lock);
 
 		if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
@@ -949,18 +942,9 @@ static int dm_request(struct request_que
 			return 0;
 		}
 
-		r = queue_io(md, bio);
-		if (r <= 0) {
-			if (r < 0)
-				bio_io_error(bio);
-			return 0;
-		}
+		queue_io(md, bio);
 
-		/*
-		 * We're in a while loop, because someone could suspend
-		 * before we get to the following read lock.
-		 */
-		down_read(&md->io_lock);
+		return 0;
 	}
 
 	__split_and_process_bio(md, bio);
@@ -1424,27 +1408,28 @@ static void dm_wq_work(struct work_struc
 
 	down_write(&md->io_lock);
 
-	while (1) {
+	while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
 		spin_lock_irq(&md->deferred_lock);
 		c = bio_list_pop(&md->deferred);
 		spin_unlock_irq(&md->deferred_lock);
 
+		up_write(&md->io_lock);
+
 		if (!c) {
-			clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
 			clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
 			break;
 		}
 
 		__split_and_process_bio(md, c);
-	}
 
-	up_write(&md->io_lock);
+		down_write(&md->io_lock);
+	}
 }
 
 static void dm_queue_flush(struct mapped_device *md)
 {
+	clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
 	queue_work(md->wq, &md->work);
-	flush_workqueue(md->wq);
 }
 
 /*
@@ -1574,6 +1559,7 @@ int dm_suspend(struct mapped_device *md,
 	/*
 	 * Wait for the already-mapped ios to complete.
 	 */
+	flush_workqueue(md->wq);
 	r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
 
 	down_write(&md->io_lock);