sched/nohz: Optimize get_nohz_timer_target()
authorWanpeng Li <wanpengli@tencent.com>
Mon, 13 Jan 2020 00:50:27 +0000 (08:50 +0800)
committerIngo Molnar <mingo@kernel.org>
Tue, 28 Jan 2020 20:36:57 +0000 (21:36 +0100)
On a machine, CPU 0 is used for housekeeping, the other 39 CPUs in the
same socket are in nohz_full mode. We can observe huge time burn in the
loop for seaching nearest busy housekeeper cpu by ftrace.

  2)               |                        get_nohz_timer_target() {
  2)   0.240 us    |                          housekeeping_test_cpu();
  2)   0.458 us    |                          housekeeping_test_cpu();

  ...

  2)   0.292 us    |                          housekeeping_test_cpu();
  2)   0.240 us    |                          housekeeping_test_cpu();
  2)   0.227 us    |                          housekeeping_any_cpu();
  2) + 43.460 us   |                        }

This patch optimizes the searching logic by finding a nearest housekeeper
CPU in the housekeeping cpumask, it can minimize the worst searching time
from ~44us to < 10us in my testing. In addition, the last iterated busy
housekeeper can become a random candidate while current CPU is a better
fallback if it is a housekeeper.

Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lkml.kernel.org/r/1578876627-11938-1-git-send-email-wanpengli@tencent.com
kernel/sched/core.c

index 55b9a9c53b915356d919de1503bd6aaeaaee6dd0..a8a5d5b6f5cf7ed2168380c3d4f45dc5881c8de4 100644 (file)
@@ -552,27 +552,32 @@ void resched_cpu(int cpu)
  */
 int get_nohz_timer_target(void)
 {
-       int i, cpu = smp_processor_id();
+       int i, cpu = smp_processor_id(), default_cpu = -1;
        struct sched_domain *sd;
 
-       if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
-               return cpu;
+       if (housekeeping_cpu(cpu, HK_FLAG_TIMER)) {
+               if (!idle_cpu(cpu))
+                       return cpu;
+               default_cpu = cpu;
+       }
 
        rcu_read_lock();
        for_each_domain(cpu, sd) {
-               for_each_cpu(i, sched_domain_span(sd)) {
+               for_each_cpu_and(i, sched_domain_span(sd),
+                       housekeeping_cpumask(HK_FLAG_TIMER)) {
                        if (cpu == i)
                                continue;
 
-                       if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
+                       if (!idle_cpu(i)) {
                                cpu = i;
                                goto unlock;
                        }
                }
        }
 
-       if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
-               cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
+       if (default_cpu == -1)
+               default_cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
+       cpu = default_cpu;
 unlock:
        rcu_read_unlock();
        return cpu;