Re: [ANNOUNCE] 2.6.15-rc5-hrt2 - hrtimers based high resolution patches

[Thomas Gleixner <tglx@linutronix.de>]

On Tue, 2006-01-10 at 18:59 -0500, Steven Rostedt wrote:
> And this _is_ protected, but I just discovered that this does _not_
> protect against hrtimer_start!
> 
> In hrtimer_start we have:
> 
> 	base = lock_hrtimer_base(timer, &flags);
> 
> 	/* Remove an active timer from the queue: */
> 	remove_hrtimer(timer, base);
> 
> Which can be called after that spin_unlock_irq is done by the
> run_hrtimer_queue, and we will hit a bug (as I did).  This is not an
> easy race to hit. 

Right, but there should be actually no use case where this happens.

For now I prefer to add a 

BUG_ON(base->curr_timer == timer); 

into hrtimer_start to find the real reason.

> Here's an example of a race for this problem:
> In posix-timers.c: commen_timer_set:
> 
>    If we get preempted between hrtimer_try_to_cancel and
>    hrtimer_restart.
> 
>    Then a new thread adds the timer back (by a threaded program).

I don't see how this should happen.

sys_timer_settime()
	timr = lock_timer(timer_id, flag);

Now the k_itimer, which is the container of the hrtimer is locked.

common_timer_set()
	if (hrtimer_try_to_cancel() < 0)
		return TIMER_RETRY;
	....
	hrtimer_start();
	return;

back in sys_timer_settime()

	unlock_timer(timr);

So how gets this timer added back, when the container lock is held
across the complete operation ?

> Now the question is what's the right solution?  Can hrtimer_start
> schedule?  Probably not, maybe we should add something to check this and
> have hrtimer_start return -1 if it is running, and let who ever called
> it figure out what to do?  Maybe have a hrtimer_cancel_start atomic
> operation? As well as a hrtimer_try_to_cancel_start?

The posix timer code does already the Right Thing, as it uses
hrtimer_try_to_cancel() and reacts on the return value. This is
necessary to prevent a dead lock with the k_itimer lock. 

I try to reproduce this with your test program on my test boxes.

	tglx