Skip to contentSkip to conversation

dm: use cpu_relax when busy waiting for condition

PATCH In review Mike Snitzer 2 participants 2 messages Sep 9, 2010 – Jun 27, 2011 Device mapper

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dm: use cpu_relax when busy waiting for condition

Switch busy loops from using msleep(1) to cpu_relax().

Any improvement is negligible given the rare nature of these busy wait conditions -- but cpu_relax() is considered the more polite method to use for busy loops.

---
 drivers/md/dm-snap.c  |    6 +++---
 drivers/md/dm-table.c |    2 +-
 drivers/md/dm.c       |    2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)
drivers/md/dm-snap.c +3 -3
index eed2101..d1f44c1 100644
@@ -264,12 +264,12 @@ static int __chunk_is_tracked(struct dm_snapshot *s, chunk_t chunk)
 
 /*
  * This conflicting I/O is extremely improbable in the caller,
- * so msleep(1) is sufficient and there is no need for a wait queue.
+ * so cpu_relax() is sufficient and there is no need for a wait queue.
  */
 static void __check_for_conflicting_io(struct dm_snapshot *s, chunk_t chunk)
 {
 	while (__chunk_is_tracked(s, chunk))
-		msleep(1);
+		cpu_relax();
 }
 
 /*
@@ -1302,7 +1302,7 @@ static void snapshot_dtr(struct dm_target *ti)
 	unregister_snapshot(s);
 
 	while (atomic_read(&s->pending_exceptions_count))
-		msleep(1);
+		cpu_relax();
 	/*
 	 * Ensure instructions in mempool_destroy aren't reordered
 	 * before atomic_read.
drivers/md/dm-table.c +1 -1
index f9fc07d..2955d46 100644
@@ -245,7 +245,7 @@ void dm_table_destroy(struct dm_table *t)
 		return;
 
 	while (atomic_read(&t->holders))
-		msleep(1);
+		cpu_relax();
 	smp_mb();
 
 	/* free the indexes */
drivers/md/dm.c +1 -1
index f934e98..4e699e4 100644
@@ -2228,7 +2228,7 @@ static void __dm_destroy(struct mapped_device *md, bool wait)
 	 */
 	if (wait)
 		while (atomic_read(&md->holders))
-			msleep(1);
+			cpu_relax();
 	else if (atomic_read(&md->holders))
 		DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
 		       dm_device_name(md), atomic_read(&md->holders));
#1 lore raw
Quoted 1 line · Mike Snitzer

Switch busy loops from using msleep(1) to cpu_relax().

I'd rather we didn't have any of these, but I think these should stay as msleep. (Would cpu_relax() even work in every case?)

Alasdair