Re: [BUG] 2.6.29.1 debugobjects warning

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Ingo Molnar (mingo@elte.hu) wrote:
> 
> * Andrew Morton <akpm@linux-foundation.org> wrote:
> 
> > On Thu, 23 Apr 2009 10:00:02 -0400
> > Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> wrote:
> > 
> > > ODEBUG: init active object type: timer_list
> 
> > It seems to be complaining that cpufreq_governor_dbs() is running 
> > init_timer() against a timer which has already been initialised 
> > once.
> 
> Not just already initialized - but also active. There's these states 
> for an object:
> 
>         ODEBUG_STATE_NONE,
>         ODEBUG_STATE_INIT,
>         ODEBUG_STATE_INACTIVE,
>         ODEBUG_STATE_ACTIVE,
>         ODEBUG_STATE_DESTROYED,
> 
> So the 'init active object type' warning above suggests that an 
> init_timer() has been done on an already running timer. If true then 
> that is a bad bug - can corrupt timer state, etc.
> 
> Thomas, do you agree?
> 

This would fit with the behavior I've noticed and submitted a patch for
later in this thread. Basically, the incorrect use of
"cancel_delayed_work()" instead of "cancel_delayed_work_sync()" may
leave the timer active if the race explained in the mail below happen.

http://lkml.org/lkml/2009/4/24/19

Maybe we should audit other users of
schedule_delayed_work/cancel_delayed_work that are re-arming themselves
in the workqueue handler without any proper cancel_delayed_work_sync()
nor flush_workqueue() calls on teardown.

Mathieu

> 	Ingo

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68