Re: [PATCH v2] sched/fair: update scale invariance of PELT

[Peter Zijlstra]

On Sat, Apr 29, 2017 at 12:09:24AM +0200, Peter Zijlstra wrote:
> On Mon, Apr 10, 2017 at 11:18:29AM +0200, Vincent Guittot wrote:
> > +++ b/include/linux/sched.h
> > @@ -313,6 +313,7 @@ struct load_weight {
> >   */
> >  struct sched_avg {
> >  	u64				last_update_time;
> > +	u64				stolen_idle_time;
> >  	u64				load_sum;
> >  	u32				util_sum;
> >  	u32				period_contrib;
> 
> > +		if (sa->util_sum < (LOAD_AVG_MAX * 1000)) {
> > +			/*
> > +			 * Add the idle time stolen by running at lower compute
> > +			 * capacity
> > +			 */
> > +			delta += sa->stolen_idle_time;
> > +		}
> > +		sa->stolen_idle_time = 0;
> 
> 
> So I was wondering if stolen_idle_time really needs to be a u64. Afaict
> we'll be at LOAD_AVG_MAX after LOAD_AVG_MAX_N periods, or LOAD_AVG_MAX_N
> * LOAD_AVG_PERIOD time, which ends up being 11040.

* 1024 or course, but still easily fits in u32.