sched: Optimize task_sched_runtime()

Large multi-threaded apps like to hit this using do_sys_times() and
then queue up on the rq->lock.

Avoid when possible.

Larry reported ~20% performance increase his test case.

Reported-by: Larry Woodman <lwoodman@redhat.com>
Suggested-by: Paul Turner <pjt@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Link: http://lkml.kernel.org/r/20131111172925.GG26898@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 1deccd7..c180860 100644 (file)
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2253,6 +2253,20 @@ unsigned long long task_sched_runtime(struct task_struct *p)
        struct rq *rq;
        u64 ns = 0;
 
+#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
+       /*
+        * 64-bit doesn't need locks to atomically read a 64bit value.
+        * So we have a optimization chance when the task's delta_exec is 0.
+        * Reading ->on_cpu is racy, but this is ok.
+        *
+        * If we race with it leaving cpu, we'll take a lock. So we're correct.
+        * If we race with it entering cpu, unaccounted time is 0. This is
+        * indistinguishable from the read occurring a few cycles earlier.
+        */
+       if (!p->on_cpu)
+               return p->se.sum_exec_runtime;
+#endif
+
        rq = task_rq_lock(p, &flags);
        ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
        task_rq_unlock(rq, p, &flags);