Re: [PATCH 2/5] rcu/nocb: Perform deferred wake up before last idle's need_resched() check

[Frederic Weisbecker <frederic@kernel.org>]

On Mon, Feb 08, 2021 at 03:45:50PM +0100, Peter Zijlstra wrote:
> On Mon, Feb 01, 2021 at 12:05:45AM +0100, Frederic Weisbecker wrote:
> 
> > diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> > index 305727ea0677..b601a3aa2152 100644
> > --- a/kernel/sched/idle.c
> > +++ b/kernel/sched/idle.c
> > @@ -55,6 +55,7 @@ __setup("hlt", cpu_idle_nopoll_setup);
> >  static noinline int __cpuidle cpu_idle_poll(void)
> >  {
> >  	trace_cpu_idle(0, smp_processor_id());
> > +	rcu_nocb_flush_deferred_wakeup();
> >  	stop_critical_timings();
> >  	rcu_idle_enter();
> >  	local_irq_enable();
> > @@ -173,6 +174,8 @@ static void cpuidle_idle_call(void)
> >  	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
> >  	int next_state, entered_state;
> >  
> > +	rcu_nocb_flush_deferred_wakeup();
> > +
> >  	/*
> >  	 * Check if the idle task must be rescheduled. If it is the
> >  	 * case, exit the function after re-enabling the local irq.
> 
> Ok if I do this instead?
> 
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -55,7 +55,6 @@ __setup("hlt", cpu_idle_nopoll_setup);
>  static noinline int __cpuidle cpu_idle_poll(void)
>  {
>  	trace_cpu_idle(0, smp_processor_id());
> -	rcu_nocb_flush_deferred_wakeup();
>  	stop_critical_timings();
>  	rcu_idle_enter();
>  	local_irq_enable();
> @@ -174,8 +173,6 @@ static void cpuidle_idle_call(void)
>  	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
>  	int next_state, entered_state;
>  
> -	rcu_nocb_flush_deferred_wakeup();
> -
>  	/*
>  	 * Check if the idle task must be rescheduled. If it is the
>  	 * case, exit the function after re-enabling the local irq.
> @@ -288,6 +285,7 @@ static void do_idle(void)
>  		}
>  
>  		arch_cpu_idle_enter();
> +		rcu_nocb_flush_deferred_wakeup();
>  
>  		/*
>  		 * In poll mode we reenable interrupts and spin. Also if we

Right, I think that should work. Nothing should call_rcu() before the
need_resched() call. And if it does, we still have the nocb_timer to do
the deferred wakeup in the worst case.

Thanks.