Re: [PATCH 4/8] sched/core: Skip wakeup when task is already running.

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Apr 22, 2014 at 08:56:11PM +0900, Dongsheng Yang wrote:

> >On 04/15/2014 10:53 PM, Peter Zijlstra wrote:
> >>On Tue, Apr 15, 2014 at 09:32:53PM +0900, Dongsheng Yang wrote:
> >>
> >>How can you get there with ->state == RUNNING? try_to_wake_up*() bail
> >>when !(->state & state).

> Hi Peter, after some more investigation, I think I got the problem, which is
> that
> some other task set p->state to TASK_RUNNING without holding p->pi_lock.

That would be a plain bug, you're only ever allowed setting your own
->state, never someone else's. That is:

  for (;;) {
    set_current_state(TASK_INTERRUPTIBLE);
    if (cond)
      break;
    schedule();
  }
  __set_current_state(TASK_RUNNING);

Is the 'normal' way of blocking, and that is very much only touching its
own state. And in the above you cannot trigger the case you're seeing,
because the ->state = TASK_RUNNING, at the end, is _after_ its already
running, so we've completed the wakeup.

So I think you've spotted something rotten and we should catch/fix that.

Your backtrace only shows where we find ->state == TASK_RUNNING, not
where it became such.