[PATCH 05/10] perf: Unroll perf_event_read_value() in perf_read_group()

[Sukadev Bhattiprolu]

Unroll the calls to perf_event_read_value() in perf_read_group()
so we can later optimize out parts we don't need for group events.

Signed-off-by: Sukadev Bhattiprolu <sukadev at linux.vnet.ibm.com>
---
 kernel/events/core.c |   17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index a6bd09d..0ce3012 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -3794,7 +3794,14 @@ static int perf_read_group(struct perf_event *event,
 
 	lockdep_assert_held(&ctx->mutex);
 
-	count = perf_event_read_value(leader, &enabled, &running);
+	enabled = running = 0;
+
+	mutex_lock(&leader->child_mutex);
+
+	perf_event_read(leader);
+	count = perf_event_aggregate(leader, &enabled, &running);
+
+	mutex_unlock(&leader->child_mutex);
 
 	values[n++] = 1 + leader->nr_siblings;
 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
@@ -3815,7 +3822,13 @@ static int perf_read_group(struct perf_event *event,
 	list_for_each_entry(sub, &leader->sibling_list, group_entry) {
 		n = 0;
 
-		values[n++] = perf_event_read_value(sub, &enabled, &running);
+		mutex_lock(&leader->child_mutex);
+
+		perf_event_read(sub);
+		values[n++] = perf_event_aggregate(sub, &enabled, &running);
+
+		mutex_unlock(&leader->child_mutex);
+
 		if (read_format & PERF_FORMAT_ID)
 			values[n++] = primary_event_id(sub);
 
-- 
1.7.9.5