Btrfs: reduce CPU contention while waiting for delayed extent operations

We batch up operations to the extent allocation tree, which allows
us to deal with the recursive nature of using the extent allocation
tree to allocate extents to the extent allocation tree.

It also provides a mechanism to sort and collect extent
operations, which makes it much more efficient to record extents
that are close together.

The delayed extent operations must all be finished before the
running transaction commits, so we have code to make sure and run a few
of the batched operations when closing our transaction handles.

This creates a great deal of contention for the locks in the
delayed extent operation tree, and also contention for the lock on the
extent allocation tree itself.  All the extra contention just slows
down the operations and doesn't get things done any faster.

This commit changes things to use a wait queue instead.  As procs
want to run the delayed operations, one of them races in and gets
permission to hit the tree, and the others step back and wait for
progress to be made.

Signed-off-by: Chris Mason <chris.mason@fusionio.com>

1 files changed, 56 insertions, 4 deletions

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index f13402104c96..87b0e856b6d0 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2438,6 +2438,16 @@ int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
+static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
+		      int count)
+{
+	int val = atomic_read(&delayed_refs->ref_seq);
+
+	if (val < seq || val >= seq + count)
+		return 1;
+	return 0;
+}
+
 /*
  * this starts processing the delayed reference count updates and
  * extent insertions we have queued up so far.  count can be
@@ -2472,6 +2482,44 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 
 	delayed_refs = &trans->transaction->delayed_refs;
 	INIT_LIST_HEAD(&cluster);
+	if (count == 0) {
+		count = delayed_refs->num_entries * 2;
+		run_most = 1;
+	}
+
+	if (!run_all && !run_most) {
+		int old;
+		int seq = atomic_read(&delayed_refs->ref_seq);
+
+progress:
+		old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
+		if (old) {
+			DEFINE_WAIT(__wait);
+			if (delayed_refs->num_entries < 16348)
+				return 0;
+
+			prepare_to_wait(&delayed_refs->wait, &__wait,
+					TASK_UNINTERRUPTIBLE);
+
+			old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
+			if (old) {
+				schedule();
+				finish_wait(&delayed_refs->wait, &__wait);
+
+				if (!refs_newer(delayed_refs, seq, 256))
+					goto progress;
+				else
+					return 0;
+			} else {
+				finish_wait(&delayed_refs->wait, &__wait);
+				goto again;
+			}
+		}
+
+	} else {
+		atomic_inc(&delayed_refs->procs_running_refs);
+	}
+
 again:
 	loops = 0;
 	spin_lock(&delayed_refs->lock);
@@ -2480,10 +2528,6 @@ again:
 	delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
 #endif
 
-	if (count == 0) {
-		count = delayed_refs->num_entries * 2;
-		run_most = 1;
-	}
 	while (1) {
 		if (!(run_all || run_most) &&
 		    delayed_refs->num_heads_ready < 64)
@@ -2505,9 +2549,12 @@ again:
 		if (ret < 0) {
 			spin_unlock(&delayed_refs->lock);
 			btrfs_abort_transaction(trans, root, ret);
+			atomic_dec(&delayed_refs->procs_running_refs);
 			return ret;
 		}
 
+		atomic_add(ret, &delayed_refs->ref_seq);
+
 		count -= min_t(unsigned long, ret, count);
 
 		if (count == 0)
@@ -2576,6 +2623,11 @@ again:
 		goto again;
 	}
 out:
+	atomic_dec(&delayed_refs->procs_running_refs);
+	smp_mb();
+	if (waitqueue_active(&delayed_refs->wait))
+		wake_up(&delayed_refs->wait);
+
 	spin_unlock(&delayed_refs->lock);
 	assert_qgroups_uptodate(trans);
 	return 0;