Re: WARNING: at kernel/lockdep.c:690 __lock_acquire+0x168/0x164b()

[David Rientjes]

On Thu, 20 Oct 2011, Tejun Heo wrote:

> > Tejun, would you like to revert f59de8992aa6 ("lockdep: Clear whole 
> > lockdep_map on initialization") since it fixes this lockdep warning?
> 
> Hmmm... the issue was that kmemcheck noticed that memory regions in
> lockdep_map are accessed before being set to any value.  I'm feeling
> dim as usual and don't understand what's going on here.  The function
> looks like the following.
> 
> 
>  void lockdep_init_map(struct lockdep_map *lock, const char *name,
> 		       struct lock_class_key *key, int subclass)
>  {
> 	 memset(lock, 0, sizeof(*lock));
> 
>  #ifdef CONFIG_LOCK_STAT
> 	 lock->cpu = raw_smp_processor_id();
>  #endif
> 	 if (DEBUG_LOCKS_WARN_ON(!name)) {
> 		 lock->name = "NULL";
> 		 return;
> 	 }
> 
> 	 lock->name = name;
> 
> 
> So, according to this thread, the problem is that the memset() clears
> lock->name field, right?

Right, and reverting f59de8992aa6 ("lockdep: Clear whole lockdep_map on 
initialization") seems to fix the lockdep warning.

> But how can that be a problem?  lock->name
> is always set to either "NULL" or @name.  Why would clearing it before
> setting make any difference?  What am I missing?
> 

The scheduler (in sched_fair and sched_rt) calls lock_set_subclass() which 
sets the name in double_unlock_balance() to set the name but there's a 
race between when that is cleared with the memset() and setting of 
lock->name where lockdep can find them to match.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/