sched: Reassign prev and switch_count when reacquire_kernel_lock() fail
authorYong Zhang <yong.zhang0@gmail.com>
Mon, 11 Jan 2010 06:21:25 +0000 (14:21 +0800)
committerIngo Molnar <mingo@elte.hu>
Thu, 21 Jan 2010 12:39:04 +0000 (13:39 +0100)
Assume A->B schedule is processing, if B have acquired BKL before and it
need reschedule this time. Then on B's context, it will go to
need_resched_nonpreemptible for reschedule. But at this time, prev and
switch_count are related to A. It's wrong and will lead to incorrect
scheduler statistics.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <2674af741001102238w7b0ddcadref00d345e2181d11@mail.gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched.c

index c535cc4f6428bcd09405c2f4a3322f9b41de7eaa..4508fe7048be5d91ee5a01daef3fe9de8d37b909 100644 (file)
@@ -5530,8 +5530,11 @@ need_resched_nonpreemptible:
 
        post_schedule(rq);
 
-       if (unlikely(reacquire_kernel_lock(current) < 0))
+       if (unlikely(reacquire_kernel_lock(current) < 0)) {
+               prev = rq->curr;
+               switch_count = &prev->nivcsw;
                goto need_resched_nonpreemptible;
+       }
 
        preempt_enable_no_resched();
        if (need_resched())