Re: [PATCH] sched/core: fix formatting issues in sched_can_stop_tick()

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Apr 14, 2014 at 09:47:41PM +0530, Viresh Kumar wrote:
> sched_can_stop_tick() was using 7 spaces instead of 8 spaces or a 'tab' at the
> beginning of each line. Which doesn't align with the Coding Guidelines.
> 
> Also it removes the *rq variable as it was used at only one place and hence we
> can directly use this_rq() instead.
> 
> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
> ---
> I don't think rq = tihs_rq() has to be done before smp_mb(), in case yes sorry
> for this patch :(
> 
>  kernel/sched/core.c | 16 ++++++----------
>  1 file changed, 6 insertions(+), 10 deletions(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 268a45e..13299c5 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -666,18 +666,14 @@ static inline bool got_nohz_idle_kick(void)
>  #ifdef CONFIG_NO_HZ_FULL
>  bool sched_can_stop_tick(void)
>  {
> +	/* Make sure rq->nr_running update is visible after the IPI */
> +	smp_rmb();
>  
> +	/* More than one running task need preemption */
> +	if (this_rq()->nr_running > 1)
> +		return false;
>  
> +	return true;
>  }
>  #endif /* CONFIG_NO_HZ_FULL */

AFAICT the smp_rmb() is entirely spurious, arch interrupts should ensure
consistency on their own. That is:

  CPU 0		  CPU 1

[w] X = 1
    IPI 1	    <int>
		[r] r = X

Should act as if there was a full memory barrier, making it so that the
read on CPU1 observes the write on CPU0.