Re: [PATCH] sched/nohz: Optimize get nohz timer target

[Frederic Weisbecker <frederic@kernel.org>]

On Wed, Jun 19, 2019 at 08:42:19AM +0800, Wanpeng Li wrote:
> Cc Frederic,
> On Tue, 18 Jun 2019 at 11:13, Wanpeng Li <kernellwp@gmail.com> wrote:
> >
> > From: Wanpeng Li <wanpengli@tencent.com>
> >
> > On a machine, cpu 0 is used for housekeeping, other 39 cpus 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.
> >
> > Cc: Ingo Molnar <mingo@redhat.com>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
> > ---
> >  kernel/sched/core.c | 5 +++--
> >  1 file changed, 3 insertions(+), 2 deletions(-)
> >
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 83bd6bb..db550cf 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -548,11 +548,12 @@ int get_nohz_timer_target(void)
> >
> >         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;
> >                         }

Nice, but you also need to handle the default case that doesn't make much sense anymore.
It hasn't ever been clear anyway. The last iterated buzy housekeeper can become
a random candidate while current CPU is a better fallback if it is a housekeeper. Also
you're enhancing housekeeping_any_cpu() in another patch so give it a better chance:

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 4778c48a7fda..c5229d71540a 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -539,27 +539,32 @@ 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;