Re: [RFC PATCH 2/5] sched: Add NOHZ_STATS_KICK

[Morten Rasmussen]

On Fri, Dec 22, 2017 at 09:42:47PM +0100, Peter Zijlstra wrote:
> On Fri, Dec 22, 2017 at 07:56:29PM +0100, Peter Zijlstra wrote:
> > Right; but I figured we'd try and do it 'right' and see how horrible it
> > is before we try and do funny things.
> 
> So now it should have a 32ms tick for up to .5s when the system goes
> completely idle.
> 
> No idea how bad that is..

Does it mean that the 32ms tick will keep going forever if the system
doesn't go completely idle? Some tiny background task or a slightly
bigger one with a longer period?

Do we actually care about stale values if the system is completely idle?

Instead of hacking select_task_rq_fair() to kick off a stats update as
Vincent already proposed, why can't we just modify Brendan's
CPU_NEWLY_IDLE proposal to do a stats update from idle_balance() every
32ms regardless of whether we need to load-balance?

This way we should get updates if there is anything running, we don't
touch the wake-up path, we don't cause any additional wake-ups, and we
don't need a timer. What am I missing?