Re: [PATCH 6/8] tick/nohz: Only wakeup a single target cpu when kicking a task

[Frederic Weisbecker <frederic@kernel.org>]

On Mon, May 10, 2021 at 12:48:14PM +0200, Peter Zijlstra wrote:
> On Mon, May 10, 2021 at 12:39:01PM +0200, Frederic Weisbecker wrote:
> > On Wed, May 05, 2021 at 03:43:28PM +0200, Peter Zijlstra wrote:
> > > That had me looking at tick_nohz_task_switch(), does we want the below?
> > > 
> > > 
> > > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > > index 9143163fa678..ff45fc513ba7 100644
> > > --- a/kernel/sched/core.c
> > > +++ b/kernel/sched/core.c
> > > @@ -4207,6 +4207,7 @@ static struct rq *finish_task_switch(struct task_struct *prev)
> > >  	vtime_task_switch(prev);
> > >  	perf_event_task_sched_in(prev, current);
> > >  	finish_task(prev);
> > > +	tick_nohz_task_switch();
> > >  	finish_lock_switch(rq);
> > >  	finish_arch_post_lock_switch();
> > >  	kcov_finish_switch(current);
> > > @@ -4252,7 +4253,6 @@ static struct rq *finish_task_switch(struct task_struct *prev)
> > >  		put_task_struct_rcu_user(prev);
> > >  	}
> > >  
> > > -	tick_nohz_task_switch();
> > >  	return rq;
> > >  }
> > >  
> > > diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
> > > index 828b091501ca..ea079be9097f 100644
> > > --- a/kernel/time/tick-sched.c
> > > +++ b/kernel/time/tick-sched.c
> > > @@ -447,13 +447,10 @@ void tick_nohz_dep_clear_signal(struct signal_struct *sig, enum tick_dep_bits bi
> > >   */
> > >  void __tick_nohz_task_switch(void)
> > >  {
> > > -	unsigned long flags;
> > >  	struct tick_sched *ts;
> > >  
> > > -	local_irq_save(flags);
> > > -
> > >  	if (!tick_nohz_full_cpu(smp_processor_id()))
> > > -		goto out;
> > > +		return;
> > >  
> > >  	ts = this_cpu_ptr(&tick_cpu_sched);
> > >  
> > > @@ -462,8 +459,6 @@ void __tick_nohz_task_switch(void)
> > >  		    atomic_read(&current->signal->tick_dep_mask))
> > >  			tick_nohz_full_kick();
> > >  	}
> > > -out:
> > > -	local_irq_restore(flags);
> > >  }
> > >  
> > >  /* Get the boot-time nohz CPU list from the kernel parameters. */
> > 
> > 
> > Sure, I'll take your SoB on that too, ok?
> 
> OK, but please also test it :-) I didn't even ask a compiler it's
> opinion on the thing.

Of course, there will be another version of the patchset + usual testing :-)