Re: [PATCH] timer: implement lockdep deadlock detection

[Johannes Berg]

On Tue, 2009-01-27 at 00:07 +0100, Ingo Molnar wrote:
> * Johannes Berg <johannes@xxxxxxxxxxxxxxxx> wrote:
> 
> > This modifies the timer code in a way to allow lockdep to detect 
> > deadlocks resulting from a lock being taken in the timer function as 
> > well as around the del_timer_sync() call.
> > 
> > Validated with this module, otherwise bootup was clean.
> 
> That's a really neat trick ...

Did it with workqueues earlier :)

> Curious: have you hit such a bug recently that motivated you to implement 
> it?

No, I haven't hit a bug, but code inspection made me wonder whether or
not it could trigger. Didn't have a chance to try yet, had no hardware
available for this particular case.

> > +#ifdef CONFIG_LOCKDEP
> > +#define init_timer(timer)						\
> > +	do {								\
> > +		static struct lock_class_key __key;			\
> > +		init_timer_key((timer), #timer, &__key);		\
> > +	} while (0)
> > +#define init_timer_deferrable(timer)					\
> > +	do {								\
> 
> (Style detail: please put a newline after each macro block to make them 
> stand apart a bit more.)

Sure.

> >  int del_timer_sync(struct timer_list *timer)
> >  {
> > +#ifdef CONFIG_LOCKDEP
> > +	unsigned long flags;
> > +
> > +	local_irq_save(flags);
> > +	lock_map_acquire(&timer->lockdep_map);
> > +	lock_map_release(&timer->lockdep_map);
> > +	local_irq_restore(flags);
> > +#endif
> 
> yummie. We have repeat bugs in this area that are rather tricky to find. 
> This will trigger them in a debuggable way.
> 
> > @@ -861,10 +881,21 @@ static inline void __run_timers(struct t
> >  
> >  			set_running_timer(base, timer);
> >  			detach_timer(timer, 1);
> > +
> >  			spin_unlock_irq(&base->lock);
> 
> Yes, the newline is needed there :-) [seriously]

Hah. I was editing code there in that area and removed it later again,
not realising I left the newline.

> >  				int preempt_count = preempt_count();
> > +
> > +				/* Couple the lock chain with the lock chain at
> > +				 * del_timer_sync by acquiring the lock_map around
> > +				 * the fn() call here and in del_timer_sync.
> > +				 */
> 
> Please use the standard multi-line comment style:
> 
>   /*
>    * Comment .....
>    * ...... goes here:
>    */

Oops, yes.

johannes

Attachment: signature.asc
Description: This is a digitally signed message part