sched: Fix rebalance interval calculation
Sisir Koppaka [Sat, 26 Mar 2011 12:52:55 +0000 (17:52 +0530)]
The interval for checking scheduling domains if they are due to be
balanced currently depends on boot state NR_CPUS, which may not
accurately reflect the number of online CPUs at the time of check.

Thus replace NR_CPUS with num_online_cpus().

 (ed: Should only affect those who set NR_CPUS really high, such as 4096
      or so :-)

Signed-off-by: Sisir Koppaka <sisir.koppaka@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <AANLkTikqHWid2Q93F5U5Qw5snJH8C5PXoa7J6=6hYO94@mail.gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/sched_fair.c

index 3f7ec9e..c7ec5c8 100644 (file)
@@ -22,6 +22,7 @@
 
 #include <linux/latencytop.h>
 #include <linux/sched.h>
+#include <linux/cpumask.h>
 
 /*
  * Targeted preemption latency for CPU-bound tasks:
@@ -3850,8 +3851,8 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
                interval = msecs_to_jiffies(interval);
                if (unlikely(!interval))
                        interval = 1;
-               if (interval > HZ*NR_CPUS/10)
-                       interval = HZ*NR_CPUS/10;
+               if (interval > HZ*num_online_cpus()/10)
+                       interval = HZ*num_online_cpus()/10;
 
                need_serialize = sd->flags & SD_SERIALIZE;