Re: [PATCH v3] sched/fair: update scale invariance of PELT

[Peter Zijlstra]

On Fri, May 18, 2018 at 02:08:51PM +0200, Peter Zijlstra wrote:
> However, I think we can shrink util_avg unconditionally, which would
> give us exactly the 4 byte hole we need.

There is of course the problem that we track rq util_avg as a straight
sum of entity util_avg, such that if you migrate/wake a bunch of tasks
into the rq the sum can exceed the natural bounds, although it will
settle down back to it.

If we're really worried we could try and detect overflows on the
summing and clip.