Steven Rostedt
2013-Nov-15 03:57 UTC
[PATCH] btrfs: Use trace condition for get_extent tracepoint
Doing an if statement to test some condition to know if we should
trigger a tracepoint is pointless when tracing is disabled. This just
adds overhead and wastes a branch prediction. This is why the
TRACE_EVENT_CONDITION() was created. It places the check inside the jump
label so that the branch does not happen unless tracing is enabled.
That is, instead of doing:
if (em)
trace_btrfs_get_extent(root, em);
Which is basically this:
if (em)
if (static_key(trace_btrfs_get_extent)) {
Using a TRACE_EVENT_CONDITION() we can just do:
trace_btrfs_get_extent(root, em);
And the condition trace event will do:
if (static_key(trace_btrfs_get_extent)) {
if (em) {
...
The static key is a non conditional jump (or nop) that is faster than
having to check if em is NULL or not.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 51e3afa..a9ad918 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -6173,8 +6173,7 @@ insert:
write_unlock(&em_tree->lock);
out:
- if (em)
- trace_btrfs_get_extent(root, em);
+ trace_btrfs_get_extent(root, em);
if (path)
btrfs_free_path(path);
diff --git a/include/trace/events/btrfs.h b/include/trace/events/btrfs.h
index f18b3b7..4832d75 100644
--- a/include/trace/events/btrfs.h
+++ b/include/trace/events/btrfs.h
@@ -162,12 +162,14 @@ DEFINE_EVENT(btrfs__inode, btrfs_inode_evict,
{ EXTENT_FLAG_LOGGING, "LOGGING" }, \
{ EXTENT_FLAG_FILLING, "FILLING" })
-TRACE_EVENT(btrfs_get_extent,
+TRACE_EVENT_CONDITION(btrfs_get_extent,
TP_PROTO(struct btrfs_root *root, struct extent_map *map),
TP_ARGS(root, map),
+ TP_CONDITION(map),
+
TP_STRUCT__entry(
__field( u64, root_objectid )
__field( u64, start )