xfs: Don't wake the aild once per second
authorDave Chinner <david@fromorbit.com>
Mon, 11 Jan 2010 11:49:58 +0000 (11:49 +0000)
committerAlex Elder <aelder@sgi.com>
Fri, 15 Jan 2010 21:32:46 +0000 (15:32 -0600)
Now that the AIL push algorithm is traversal safe, we don't need a
watchdog function in the xfsaild to catch pushes that fail to make
progress. Remove the watchdog timeout and make pushes purely driven
by demand. This will remove the once-per-second wakeup that is seen
when the filesystem is idle and make laptop power misers happy.

Signed-off-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
fs/xfs/linux-2.6/xfs_super.c
fs/xfs/xfs_trans_ail.c

index 77414db..9f2e398 100644 (file)
@@ -877,12 +877,11 @@ xfsaild(
 {
        struct xfs_ail  *ailp = data;
        xfs_lsn_t       last_pushed_lsn = 0;
-       long            tout = 0;
+       long            tout = 0; /* milliseconds */
 
        while (!kthread_should_stop()) {
-               if (tout)
-                       schedule_timeout_interruptible(msecs_to_jiffies(tout));
-               tout = 1000;
+               schedule_timeout_interruptible(tout ?
+                               msecs_to_jiffies(tout) : MAX_SCHEDULE_TIMEOUT);
 
                /* swsusp */
                try_to_freeze();
index 2ffc570..063dfbd 100644 (file)
@@ -237,14 +237,15 @@ out:
 }
 
 /*
- * Function that does the work of pushing on the AIL
+ * xfsaild_push does the work of pushing on the AIL.  Returning a timeout of
+ * zero indicates that the caller should sleep until woken.
  */
 long
 xfsaild_push(
        struct xfs_ail  *ailp,
        xfs_lsn_t       *last_lsn)
 {
-       long            tout = 1000; /* milliseconds */
+       long            tout = 0;
        xfs_lsn_t       last_pushed_lsn = *last_lsn;
        xfs_lsn_t       target =  ailp->xa_target;
        xfs_lsn_t       lsn;
@@ -262,7 +263,7 @@ xfsaild_push(
                 */
                xfs_trans_ail_cursor_done(ailp, cur);
                spin_unlock(&ailp->xa_lock);
-               last_pushed_lsn = 0;
+               *last_lsn = 0;
                return tout;
        }
 
@@ -279,7 +280,6 @@ xfsaild_push(
         * prevents use from spinning when we can't do anything or there is
         * lots of contention on the AIL lists.
         */
-       tout = 10;
        lsn = lip->li_lsn;
        flush_log = stuck = count = 0;
        while ((XFS_LSN_CMP(lip->li_lsn, target) < 0)) {
@@ -376,14 +376,14 @@ xfsaild_push(
 
        if (!count) {
                /* We're past our target or empty, so idle */
-               tout = 1000;
+               last_pushed_lsn = 0;
        } else if (XFS_LSN_CMP(lsn, target) >= 0) {
                /*
                 * We reached the target so wait a bit longer for I/O to
                 * complete and remove pushed items from the AIL before we
                 * start the next scan from the start of the AIL.
                 */
-               tout += 20;
+               tout = 50;
                last_pushed_lsn = 0;
        } else if ((stuck * 100) / count > 90) {
                /*
@@ -395,11 +395,14 @@ xfsaild_push(
                 * Backoff a bit more to allow some I/O to complete before
                 * continuing from where we were.
                 */
-               tout += 10;
+               tout = 20;
+       } else {
+               /* more to do, but wait a short while before continuing */
+               tout = 10;
        }
        *last_lsn = last_pushed_lsn;
        return tout;
-}      /* xfsaild_push */
+}
 
 
 /*