Re: [RFC PATCH] sched: Fix performance regression introduced by mm_cid

[Mathieu Desnoyers]

On 2023-04-04 05:05, Peter Zijlstra wrote:
> On Mon, Apr 03, 2023 at 02:13:42PM -0400, Mathieu Desnoyers wrote:
> >   void sched_mm_cid_exit_signals(struct task_struct *t)
> >   {
> >   	struct mm_struct *mm = t->mm;
> > -	unsigned long flags;
> > +	struct rq_flags rf;
> > +	struct rq *rq;
> >   
> >   	if (!mm)
> >   		return;
> > -	local_irq_save(flags);
> > +
> > +	preempt_disable();
> > +	rq = this_rq();
> > +	rq_lock_irqsave(rq, &rf);
> >   	mm_cid_put(mm, t->mm_cid);
> > -	t->mm_cid = -1;
> > +	t->last_mm_cid = t->mm_cid = -1;
> >   	t->mm_cid_active = 0;
> > -	local_irq_restore(flags);
> > +	rq_unlock_irqrestore(rq, &rf);
> > +	preempt_enable();
> >   }
>
> FWIW a *slightly* cheaper form is:
>
> 	preempt_disable();
> 	rq = this_rq();
> 	rq_lock_irqsave(rq, &rf);
> 	preempt_enable_noresched(); /* holding spinlock */
> 	...
> 	rq_unlock_irqrestore(rq, &rf);

Good point, updated for next round.

Thanks,

Mathieu



--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com