Re: [patch 5/8] hrtimer remove state field

[Thomas Gleixner <tglx@linutronix.de>]

On Sun, 2006-03-12 at 17:00 +0100, Roman Zippel wrote:
> Hi,
> 
> On Sun, 12 Mar 2006, Thomas Gleixner wrote:
> 
> > How do you want to prevent that a signal is dequeued on one CPU while
> > the softirq expires the timer on another CPU ? This can not be
> > prevented.
> 
> This should not be possible in first place, otherwise it's a bug.
> The original problem was a broken state machine, is that so hard to 
> believe? If there is another problem, please provide more details.

Roman,

there was a state machine problem caused by something similar.

But the problem I described now happened with the current patch queue -
without the hrtimer_active() check. I have no direct access to the
machine which lets this surface and I just tried to reconstruct the
scenario from the sparse information which was provided by the customer.
All I can tell, that it is related to something similar and a requeue
happens where none should happen.

I agree, that it should not be handled in the hrtimer code. It has to be
fixed in the posix-timer code.

I make the check a BUG_ON(!hrtimer_active(timer) so it might show up in
-mm again. Ok ?

	tglx