]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
locks: avoid taking global lock if possible when waking up blocked waiters
authorJeff Layton <jlayton@redhat.com>
Fri, 21 Jun 2013 12:58:16 +0000 (08:58 -0400)
committerAl Viro <viro@zeniv.linux.org.uk>
Sat, 29 Jun 2013 08:57:43 +0000 (12:57 +0400)
Since we always hold the i_lock when inserting a new waiter onto the
fl_block list, we can avoid taking the global lock at all if we find
that it's empty when we go to wake up blocked waiters.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/locks.c

index ce302d43822ba41a7d3c543ef4f2cc1572a9e26d..84e269fc4c6936bc8c4a5ae66d6e8df5dbd9c141 100644 (file)
@@ -548,7 +548,10 @@ static void locks_delete_block(struct file_lock *waiter)
  * the order they blocked. The documentation doesn't require this but
  * it seems like the reasonable thing to do.
  *
- * Must be called with file_lock_lock held!
+ * Must be called with both the i_lock and file_lock_lock held. The fl_block
+ * list itself is protected by the file_lock_list, but by ensuring that the
+ * i_lock is also held on insertions we can avoid taking the file_lock_lock
+ * in some cases when we see that the fl_block list is empty.
  */
 static void __locks_insert_block(struct file_lock *blocker,
                                        struct file_lock *waiter)
@@ -576,6 +579,16 @@ static void locks_insert_block(struct file_lock *blocker,
  */
 static void locks_wake_up_blocks(struct file_lock *blocker)
 {
+       /*
+        * Avoid taking global lock if list is empty. This is safe since new
+        * blocked requests are only added to the list under the i_lock, and
+        * the i_lock is always held here. Note that removal from the fl_block
+        * list does not require the i_lock, so we must recheck list_empty()
+        * after acquiring the file_lock_lock.
+        */
+       if (list_empty(&blocker->fl_block))
+               return;
+
        spin_lock(&file_lock_lock);
        while (!list_empty(&blocker->fl_block)) {
                struct file_lock *waiter;