Re: [PATCH] tracing/sched: Check preempt_count() for current when reading task->state
From: Ingo Molnar
Date: Thu Dec 11 2014 - 11:50:40 EST
* Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> On Thu, 11 Dec 2014 13:31:21 +0100
> Ingo Molnar <mingo@xxxxxxxxxx> wrote:
> > > What overhead are you worried about? Note, this is in the
> > > schedule tracepoint and does not affect the scheduler itself
> > > (as long as the tracepoint is not enabled).
> > Scheduler tracepoints are pretty popular, so I'm worried about
> > their complexity when they are activated.
> > > I'm also thinking that as long as "prev" is always guaranteed
> > > to be "current" we can remove the check and just use
> > > preempt_count() always. But I'm worried that we can't
> > > guaranteed that.
> > You could add a WARN_ON_ONCE() or so to double check that
> > assumption?
> I actually thought about that, but that gives us the same overhead as
> the branch we want to remove.
> But if you are going for simpler, then that would make sense.
> > > What other ideas do you have? Because wrong data is worse than
> > > the overhead of the above code. If Thomas taught me anything,
> > > it's that!
> > My idea is to have simpler, yet correct code. And ponies!
> So something like this instead?
> -- Steve
> diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
> index 0a68d5ae584e..782018b135ff 100644
> --- a/include/trace/events/sched.h
> +++ b/include/trace/events/sched.h
> @@ -97,10 +97,12 @@ static inline long __trace_sched_switch_state(struct task_struct *p)
> long state = p->state;
> #ifdef CONFIG_PREEMPT
> + WARN_ON_ONCE(p != current);
> * For all intents and purposes a preempted task is a running task.
> - if (task_preempt_count(p) & PREEMPT_ACTIVE)
> + if (preempt_count() & PREEMPT_ACTIVE)
> state = TASK_RUNNING | TASK_STATE_MAX;
Yeah, that looks a lot better IMHO, 'p' is supposed to be the
current task, at least on a booted up system with a working
scheduler. Not sure about transient initialization states such as
very early boot and idle thread initialization - but it might
work out for them as well.
If the WARN_ON_ONCE() remains silent on your testbox then I'd
suggest removing the WARN_ON_ONCE(), the change looks good to me:
Acked-by: Ingo Molnar <mingo@xxxxxxxxxx>
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/