Re: [patch 3/6] sched: use rq->avg_event to resurrect nohzratelimiting

[Peter Zijlstra]

On Fri, 2011-11-18 at 16:19 +0100, Mike Galbraith wrote:
> Entering nohz at high frequency eats cycles, ratelimit it
> just as we do idle_balance(), and for the same reason.
> 
> Signed-off-by: Mike Galbraith <efault@xxxxxx>

Arjan, can we do something smart with the idle governor which already
guestimates the idle duration for us?

> ---
>  include/linux/sched.h    |    2 ++
>  kernel/sched.c           |    7 +++++++
>  kernel/time/tick-sched.c |    2 +-
>  3 files changed, 10 insertions(+), 1 deletion(-)
> 
> Index: linux-3.2.git/include/linux/sched.h
> ===================================================================
> --- linux-3.2.git.orig/include/linux/sched.h
> +++ linux-3.2.git/include/linux/sched.h
> @@ -1979,8 +1979,10 @@ static inline void idle_task_exit(void)
>  
>  #if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
>  extern void wake_up_idle_cpu(int cpu);
> +extern int sched_needs_cpu(int cpu);
>  #else
>  static inline void wake_up_idle_cpu(int cpu) { }
> +static inline void sched_needs_cpu(int cpu) { }
>  #endif
>  
>  extern unsigned int sysctl_sched_latency;
> Index: linux-3.2.git/kernel/sched.c
> ===================================================================
> --- linux-3.2.git.orig/kernel/sched.c
> +++ linux-3.2.git/kernel/sched.c
> @@ -1419,6 +1419,13 @@ static inline bool got_nohz_idle_kick(vo
>  	return idle_cpu(smp_processor_id()) && this_rq()->nohz_balance_kick;
>  }
>  
> +#ifdef CONFIG_SMP
> +int sched_needs_cpu(int cpu)
> +{
> +	return cpu_rq(cpu)->avg_event < sysctl_sched_migration_cost;
> +}
> +#endif
> +
>  #else /* CONFIG_NO_HZ */
>  
>  static inline bool got_nohz_idle_kick(void)
> Index: linux-3.2.git/kernel/time/tick-sched.c
> ===================================================================
> --- linux-3.2.git.orig/kernel/time/tick-sched.c
> +++ linux-3.2.git/kernel/time/tick-sched.c
> @@ -352,7 +352,7 @@ void tick_nohz_stop_sched_tick(int inidl
>  	} while (read_seqretry(&xtime_lock, seq));
>  
>  	if (rcu_needs_cpu(cpu) || printk_needs_cpu(cpu) ||
> -	    arch_needs_cpu(cpu)) {
> +	    arch_needs_cpu(cpu) || sched_needs_cpu(cpu)) {
>  		next_jiffies = last_jiffies + 1;
>  		delta_jiffies = 1;
>  	} else {
> 
> 

--
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/