Re: [PATCH] mm, vmstat: Use cond_resched only when !CONFIG_PREEMPT

From: KAMEZAWA Hiroyuki
Date: Tue May 31 2011 - 00:36:51 EST


On Tue, 31 May 2011 09:58:40 +0600
Rakib Mullick <rakib.mullick@xxxxxxxxx> wrote:

> On Tue, May 31, 2011 at 9:18 AM, KAMEZAWA Hiroyuki
> <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> > On Tue, 31 May 2011 09:13:47 +0600
> > Rakib Mullick <rakib.mullick@xxxxxxxxx> wrote:
> >
> >> On Tue, May 31, 2011 at 5:38 AM, KAMEZAWA Hiroyuki
> >> <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> >> > On Mon, 30 May 2011 22:59:04 +0600
> >> > Rakib Mullick <rakib.mullick@xxxxxxxxx> wrote:
> >> >
> >> >> commit 468fd62ed9 (vmstats: add cond_resched() to refresh_cpu_vm_stats()) added cond_resched() in refresh_cpu_vm_stats. Purpose of that patch was to allow other threads to run in non-preemptive case. This patch, makes sure that cond_resched() gets called when !CONFIG_PREEMPT is set. In a preemptiable kernel we don't need to call cond_resched().
> >> >>
> >> >> Signed-off-by: Rakib Mullick <rakib.mullick@xxxxxxxxx>
> >> >
> >> > Hmm, what benefit do we get by adding this extra #ifdef in the code directly ?
> >> > Other cond_resched() callers are not guilty in !CONFIG_PREEMPT ?
> >> >
> >> Well, in preemptible kernel this context will get preempted if
> >> requires, so we don't need cond_resched(). If you checkout the git log
> >> of the mentioned commit, you'll find the explanation. It says:
> >> Â Â Â Â "Adding a cond_resched() to allow other threads to run in the
> >> non-preemptive
> >> Â Â case."
> >>
> >
> > IOW, my question is "why only this cond_resched() should be fixed ?"
>
> cond_resched() forces this thread to be scheduled. I'm just trying
> pointing out the use of cond_resched(), until unless I'm not missing
> anything.
>
> > What's bad with all cond_resched() in the kernel as no-op in CONFIG_PREEMPT ?
> >
> cond_resched() basically checks whether it needs to be scheduled or
> not. But, we know in advance that we don't need cond_resched in
> CONFIG_PREEMPT.
>
> Thanks,

Then just remove _all_ cond_resched() by defining noop function in header.
Please don't fix in ugly way.

#ifdef CONFIG_PEERMPT
static void cond_resched()
{
}
#endif

Thanks,
-Kame




--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/