From 53863232ef961778aa414b700ed88a48e8e871e6 Mon Sep 17 00:00:00 2001
From: Chris Mason <chris.mason@oracle.com>
Date: Fri, 15 Aug 2008 15:34:18 -0400
Subject: Btrfs: Lower contention on the csum mutex

This takes the csum mutex deeper in the call chain and releases it
more often.

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

diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 958cd8b..2ee3017 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -48,6 +48,7 @@ struct btrfs_worker_thread {
 
 	/* number of things on the pending list */
 	atomic_t num_pending;
+
 	unsigned long sequence;
 
 	/* protects the pending list. */
@@ -242,7 +243,7 @@ static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
 	worker = list_entry(next, struct btrfs_worker_thread, worker_list);
 	atomic_inc(&worker->num_pending);
 	worker->sequence++;
-	if (worker->sequence % 4 == 0)
+	if (worker->sequence % workers->idle_thresh == 0)
 		list_move_tail(next, &workers->worker_list);
 	return worker;
 }
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 1aed1f4..92e14dd 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1386,7 +1386,14 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	 * devices
 	 */
 	fs_info->submit_workers.idle_thresh = 64;
-	fs_info->workers.idle_thresh = 32;
+
+	/* fs_info->workers is responsible for checksumming file data
+	 * blocks and metadata.  Using a larger idle thresh allows each
+	 * worker thread to operate on things in roughly the order they
+	 * were sent by the writeback daemons, improving overall locality
+	 * of the IO going down the pipe.
+	 */
+	fs_info->workers.idle_thresh = 128;
 
 	btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
 	btrfs_init_workers(&fs_info->endio_workers, "endio",
diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c
index 51aba8c..9454e0a 100644
--- a/fs/btrfs/file-item.c
+++ b/fs/btrfs/file-item.c
@@ -321,6 +321,7 @@ again:
 	file_key.offset = offset;
 	btrfs_set_key_type(&file_key, BTRFS_CSUM_ITEM_KEY);
 
+	mutex_lock(&BTRFS_I(inode)->csum_mutex);
 	item = btrfs_lookup_csum(trans, root, path, objectid, offset, 1);
 	if (!IS_ERR(item)) {
 		leaf = path->nodes[0];
@@ -367,7 +368,7 @@ again:
 	ret = btrfs_search_slot(trans, root, &file_key, path,
 				BTRFS_CRC32_SIZE, 1);
 	if (ret < 0)
-		goto fail;
+		goto fail_unlock;
 	if (ret == 0) {
 		BUG();
 	}
@@ -411,10 +412,10 @@ insert:
 	ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 				      ins_size);
 	if (ret < 0)
-		goto fail;
+		goto fail_unlock;
 	if (ret != 0) {
 		WARN_ON(1);
-		goto fail;
+		goto fail_unlock;
 	}
 csum:
 	leaf = path->nodes[0];
@@ -427,6 +428,8 @@ found:
 	item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
 				      btrfs_item_size_nr(leaf, path->slots[0]));
 	eb_token = NULL;
+	mutex_unlock(&BTRFS_I(inode)->csum_mutex);
+	cond_resched();
 next_sector:
 
 	if (!eb_token ||
@@ -467,13 +470,18 @@ next_sector:
 		eb_token = NULL;
 	}
 	btrfs_mark_buffer_dirty(path->nodes[0]);
+	cond_resched();
 	if (total_bytes < sums->len) {
 		btrfs_release_path(root, path);
 		goto again;
 	}
-fail:
+out:
 	btrfs_free_path(path);
 	return ret;
+
+fail_unlock:
+	mutex_unlock(&BTRFS_I(inode)->csum_mutex);
+	goto out;
 }
 
 int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 33b9908..6510789 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -415,10 +415,8 @@ static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 	btrfs_set_trans_block_group(trans, inode);
 	list_for_each(cur, list) {
 		sum = list_entry(cur, struct btrfs_ordered_sum, list);
-		mutex_lock(&BTRFS_I(inode)->csum_mutex);
 		btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 				       inode, sum);
-		mutex_unlock(&BTRFS_I(inode)->csum_mutex);
 	}
 	return 0;
 }
-- 
cgit v0.10.2