[PATCH] sched/fair: Optimize select_idle_cpu

[Cheng Jian]

select_idle_cpu will scan the LLC domain for idle CPUs,
it's always expensive. so commit
    1ad3aaf3fcd2 ("sched/core: Implement new approach to scale select_idle_cpu()")
introduces a way to limit how many CPUs we scan.

But this also lead to the following issue:

Our threads are all bind to the front CPUs of the LLC domain,
and now all the threads runs on the last CPU of them. nr is
always less than the cpumask_weight, for_each_cpu_wrap can't
find the CPU which our threads can run on, so the threads stay
at the last CPU all the time.

Fixes: 1ad3aaf3fcd2 ("sched/core: Implement new approach to scale select_idle_cpu()")
Signed-off-by: Cheng Jian <cj.chengjian@xxxxxxxxxx>
---
 kernel/sched/fair.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 08a233e97a01..16a29b570803 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5834,6 +5834,7 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
 	s64 delta;
 	int this = smp_processor_id();
 	int cpu, nr = INT_MAX, si_cpu = -1;
+	struct cpumask cpus;
 
 	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
 	if (!this_sd)
@@ -5859,11 +5860,11 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
 
 	time = cpu_clock(this);
 
-	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
+	cpumask_and(&cpus, sched_domain_span(sd), p->cpus_ptr);
+
+	for_each_cpu_wrap(cpu, &cpus, target) {
 		if (!--nr)
 			return si_cpu;
-		if (!cpumask_test_cpu(cpu, p->cpus_ptr))
-			continue;
 		if (available_idle_cpu(cpu))
 			break;
 		if (si_cpu == -1 && sched_idle_cpu(cpu))
-- 
2.20.1