Re: [GIT PULL] sched: Fix missing preemption opportunity

[Peter Zijlstra <peterz@infradead.org>]

I picked up the patch; will drop it if Ingo also does ;-)

On Thu, Jan 22, 2015 at 06:08:04PM +0100, Frederic Weisbecker wrote:
> +++ b/kernel/sched/core.c
> @@ -2877,6 +2877,21 @@ void __sched schedule_preempt_disabled(void)
>  	preempt_disable();
>  }
>  
> +static void preempt_schedule_common(void)
> +{
> +	do {
> +		__preempt_count_add(PREEMPT_ACTIVE);
> +		__schedule();
> +		__preempt_count_sub(PREEMPT_ACTIVE);
> +
> +		/*
> +		 * Check again in case we missed a preemption opportunity
> +		 * between schedule and now.
> +		 */
> +		barrier();

I do however wonder about this barrier() here; why do we think we need
it?

Is that because test_bit() it 'broken'? The bitops are typically atomic
ops and atomic reads should be through a volatile cast (x86
constant_test_bit doesn't seem to do this).

Should we go audit and fix that?

> +	} while (need_resched());
> +}