2010-01-12 10:42:37

by Rakib Mullick

Subject: [PATCH] sched: Move level 'again' after get_cpu() in sched_exec.

sched: move level again after get_cpu().

We can get this_cpu by calling get_cpu() once and it remains
same. We don't need to call it again.

Signed-off-by: Rakib Mullick <[email protected]>
---

--- linus/kernel/sched.c 2010-01-06 20:11:10.000000000 +0600
+++ rakib/kernel/sched.c 2010-01-12 16:01:54.000000000 +0600
@@ -3143,8 +3143,8 @@ void sched_exec(void)
unsigned long flags;
struct rq *rq;

-again:
this_cpu = get_cpu();
+again:
dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0);
if (dest_cpu == this_cpu) {
put_cpu();

2010-01-12 10:50:42

by Peter Zijlstra

Subject: Re: [PATCH] sched: Move level 'again' after get_cpu() in sched_exec.

On Tue, 2010-01-12 at 16:42 +0600, Rakib Mullick wrote:
> sched: move level again after get_cpu().
>
> We can get this_cpu by calling get_cpu() once and it remains
> same. We don't need to call it again.

Yes we do... get_cpu() needs to be paired with put_cpu() and they
disable/enable preemption.

Since there's an unconditional put_cpu() before the goto again, we need
to do the get_cpu().

Plenty of kernel debug options would have informed you of this if you'd
tried running it.

> Signed-off-by: Rakib Mullick <[email protected]>
> ---
>
> --- linus/kernel/sched.c 2010-01-06 20:11:10.000000000 +0600
> +++ rakib/kernel/sched.c 2010-01-12 16:01:54.000000000 +0600
> @@ -3143,8 +3143,8 @@ void sched_exec(void)
> unsigned long flags;
> struct rq *rq;
>
> -again:
> this_cpu = get_cpu();
> +again:
> dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0);
> if (dest_cpu == this_cpu) {
> put_cpu();