Re: [PATCH 2/3] sched, timer: Use atomics for thread_group_cputimer to improve scalability

[Jason Low <jason.low2@hp.com>]

On Wed, 2015-04-15 at 16:07 +0530, Preeti U Murthy wrote:
> On 04/15/2015 04:39 AM, Jason Low wrote:
> >  /*
> > @@ -885,11 +890,8 @@ static void check_thread_timers(struct task_struct *tsk,
> >  static void stop_process_timers(struct signal_struct *sig)
> >  {
> >  	struct thread_group_cputimer *cputimer = &sig->cputimer;
> > -	unsigned long flags;
> > 
> > -	raw_spin_lock_irqsave(&cputimer->lock, flags);
> > -	cputimer->running = 0;
> > -	raw_spin_unlock_irqrestore(&cputimer->lock, flags);
> > +	WRITE_ONCE(cputimer->running, 0);
> 
> Why do a WRITE_ONCE() here ?

Perhaps Peter can confirm/elaborate, but since we're now updating the
running field without the lock, we use WRITE_ONCE to guarantee that this
doesn't get optimized in any way. This can also serve as "documentation"
that we're writing to a shared variable without a lock.

>  Maybe you should explicitly mention this
> through a comment like Steven pointed out about all
> WRITE/READ/ACCESS_ONCE() usage.

Yeah, we should add a comment here.

Thanks,
Jason