Re: [PATCH 8/9] psi: pressure stall information for CPU, memory, and IO

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Aug 21, 2018 at 03:44:13PM -0400, Johannes Weiner wrote:

> > > +		for (s = PSI_NONIDLE; s >= 0; s--) {
> > > +			u32 time, delta;
> > > +
> > > +			time = READ_ONCE(groupc->times[s]);
> > > +			/*
> > > +			 * In addition to already concluded states, we
> > > +			 * also incorporate currently active states on
> > > +			 * the CPU, since states may last for many
> > > +			 * sampling periods.
> > > +			 *
> > > +			 * This way we keep our delta sampling buckets
> > > +			 * small (u32) and our reported pressure close
> > > +			 * to what's actually happening.
> > > +			 */
> > > +			if (test_state(groupc->tasks, cpu, s)) {
> > > +				/*
> > > +				 * We can race with a state change and
> > > +				 * need to make sure the state_start
> > > +				 * update is ordered against the
> > > +				 * updates to the live state and the
> > > +				 * time buckets (groupc->times).
> > > +				 *
> > > +				 * 1. If we observe task state that
> > > +				 * needs to be recorded, make sure we
> > > +				 * see state_start from when that
> > > +				 * state went into effect or we'll
> > > +				 * count time from the previous state.
> > > +				 *
> > > +				 * 2. If the time delta has already
> > > +				 * been added to the bucket, make sure
> > > +				 * we don't see it in state_start or
> > > +				 * we'll count it twice.
> > > +				 *
> > > +				 * If the time delta is out of
> > > +				 * state_start but not in the time
> > > +				 * bucket yet, we'll miss it entirely
> > > +				 * and handle it in the next period.
> > > +				 */
> > > +				smp_rmb();
> > > +				time += cpu_clock(cpu) - groupc->state_start;
> > > +			}
> > 
> > The alternative is adding an update to scheduler_tick(), that would
> > ensure you're never more than nr_cpu_ids * TICK_NSEC behind.
> 
> I wasn't able to convert *all* states to tick updates like this.
> 
> The reason is that, while testing rq->curr for PF_MEMSTALL is cheap,
> other tasks associated with the rq could be from any cgroup in the
> system. That means we'd have to do for_each_cgroup() on every tick to
> keep the groupc->times that closely uptodate, and that wouldn't scale.
> We tend to have hundreds of them, some setups have thousands.
> 
> Since we don't need to be *that* current, I left the on-demand update
> inside the aggregator for now. It's a bit trickier, but much cheaper.

ARGH indeed; I was thinking we only need to update current. But because
we're tracking blocked state that doesn't work.

Sorry for that :/