On Mon, Jun 03, 2019 at 06:09:53PM +0200, Oleg Nesterov wrote:
> On 06/03, Peter Zijlstra wrote:
> >
> > It now also has concurrency on wakeup; but afaict that's harmless, we'll
> > get racing stores of p->state = TASK_RUNNING, much the same as if there
> > was a remote wakeup vs a wait-loop terminating early.
> >
> > I suppose the tracepoint consumers might have to deal with some
> > artifacts there, but that's their problem.
>
> I guess you mean that trace_sched_waking/wakeup can be reported twice if
> try_to_wake_up(current) races with ttwu_remote(). And ttwu_stat().
Right, one local one remote, and you get them things twice.
> > > --- a/kernel/sched/core.c
> > > +++ b/kernel/sched/core.c
> > > @@ -1990,6 +1990,28 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
> > > unsigned long flags;
> > > int cpu, success = 0;
> > >
> > > + if (p == current) {
> > > + /*
> > > + * We're waking current, this means 'p->on_rq' and 'task_cpu(p)
> > > + * == smp_processor_id()'. Together this means we can special
> > > + * case the whole 'p->on_rq && ttwu_remote()' case below
> > > + * without taking any locks.
> > > + *
> > > + * In particular:
> > > + * - we rely on Program-Order guarantees for all the ordering,
> > > + * - we're serialized against set_special_state() by virtue of
> > > + * it disabling IRQs (this allows not taking ->pi_lock).
> > > + */
> > > + if (!(p->state & state))
> > > + goto out;
> > > +
> > > + success = 1;
> > > + trace_sched_waking(p);
> > > + p->state = TASK_RUNNING;
> > > + trace_sched_woken(p);
> ^^^^^^^^^^^^^^^^^
> trace_sched_wakeup(p) ?
Uhm,, yah.
> I see nothing wrong... but probably this is because I don't fully understand
> this change. In particular, I don't really understand who else can benefit from
> this optimization...
Pretty much every wait-loop, where the wakeup happens from IRQ context
on the same CPU, before we've hit schedule().
Now, I've no idea if that's many, but I much prefer to keep this magic
inside try_to_wake_up() than spreading it around.
