> 1. Can't we set ->reporting _after_ checking engine->flags ?
> (unlikely)
I'm not sure...it is early the morning again. ;-)
> 2. Is utrace_barrier() correct???? Note the example above,
> it assumes that utrace_barrier() itself is the barrier
> wrt reporting. But, is it?
I don't think I understand the question.
> > list_for_each_entry_safe(engine, next, &utrace->attached, entry) {
> > - if (engine->flags & UTRACE_EVENT(REAP))
> > + if (engine->flags & UTRACE_EVENT(REAP)) {
> > + /*
> > + * These barriers order anything ->report_reap()
> > + * does to be while ->reporting is set. That way
> > + * any utrace_barrier() call that returns 0 is
> > + * guaranteeing that the callback is complete.
> > + */
> > + utrace->reporting = engine;
> > + smp_mb();
> > engine->ops->report_reap(engine, target);
> > + smp_mb();
> > + utrace->reporting = NULL;
>
> Agreed.
Let's verify/explain the logic of each barrier before we put it in.
> And yes, now it is important to clear ->ops _after_ calling
> ->report_reap(), otherwise utrace_barrier() can take the lock.
Right.
Thanks,
Roland
