On Thu, Feb 08, 2018 at 03:00:05PM +0100, Peter Zijlstra wrote:
> On Tue, Feb 06, 2018 at 08:23:05PM +0100, Vincent Guittot wrote:
>
> > @@ -9207,13 +9231,15 @@ void nohz_balance_enter_idle(int cpu)
> > if (!housekeeping_cpu(cpu, HK_FLAG_SCHED))
> > return;
> >
> > + rq->has_blocked_load = 1;
> > +
> > if (rq->nohz_tick_stopped)
> > - return;
>
> this case is difficult... needs thinking
>
> > + goto out;
> >
> > /*
> > * If we're a completely isolated CPU, we don't play.
> > */
> > - if (on_null_domain(cpu_rq(cpu)))
> > + if (on_null_domain(rq))
> > return;
> >
> > rq->nohz_tick_stopped = 1;
> > @@ -9222,6 +9248,13 @@ void nohz_balance_enter_idle(int cpu)
> > atomic_inc(&nohz.nr_cpus);
> >
> > set_cpu_sd_state_idle(cpu);
>
> /*
> * Ensures that if nohz_idle_balance() fails to observe our
> * @idle_cpus_mask store, it must observe the @has_blocked
> * store.
> */
> smp_mb__after_atomic();
>
> > +
> > +out:
> > + /*
> > + * Each time a cpu enter idle, we assume that it has blocked load and
> > + * enable the periodic update of the load of idle cpus
> > + */
> > + WRITE_ONCE(nohz.has_blocked, 1);
> > }
>
>
>
> > @@ -9374,6 +9407,16 @@ static bool nohz_idle_balance(struct rq *this_rq,
> > enum cpu_idle_type idle)
> >
> > SCHED_WARN_ON((flags & NOHZ_KICK_MASK) == NOHZ_BALANCE_KICK);
> >
> > + /*
> > + * We assume there will be no idle load after this update and clear
> > + * the has_blocked flag. If a cpu enters idle in the mean time, it will
> > + * set the has_blocked flag and trig another update of idle load.
> > + * Because a cpu that becomes idle, is added to idle_cpus_mask before
> > + * setting the flag, we are sure to not clear the state and not
> > + * check the load of an idle cpu.
> > + */
> > + WRITE_ONCE(nohz.has_blocked, 0);
>
> /*
> * Ensures that if we miss the CPU, we must see the has_blocked
> * store from nohz_balance_enter_idle().
> */
> smp_mb();
>
> > for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
> > if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
> > continue;
>
>
> I _think_, but my brain isn't quite willing to turn on today.
>
> Without this ordering I think it would be possible to loose has_blocked
> and not observe the CPU either.
>
>
might be the exact case from litmus-tests/R+mbonceonces.litmus but I
can't seem to get herd7 working today.
