Re: [RFC/PATCH] debug workqueue deadlocks with lockdep

[Ingo Molnar <mingo@elte.hu>]

* Johannes Berg <johannes@sipsolutions.net> wrote:

> > > +#define create_workqueue(name) \
> > > +({								\
> > > +	static struct lock_class_key __key;			\
> > > +	struct workqueue_struct *__wq;				\
> > > +								\
> > > +	__wq = __create_workqueue((name), 0, 0, &__key);	\
> > > +	__wq;							\
> > > +})
> > 
> > Why do we need __wq ?
> 
> No particular reason I think, I copied some other code doing it that 
> way.

yep, should be fine doing this:

 #define create_workqueue(name)				\
 ({							\
	static struct lock_class_key __key;		\
							\
	__create_workqueue((name), 0, 0, &__key);	\
 })

(and the return value of __create_workqueue() will be the 'return value' 
of the macro as well.)

> > 	+ extern struct workqueue_struct *__create_workqueue_key(..., key);
> > 	+ #define __create_workqueue(...)	\
> > 	+	static struct lock_class_key __key;	\
> > 	+	__create_workqueue_key(..., key);	\
> > 
> > but this is a matter of taste.

the above macro should at minimum be encapsulated with
do { ... } while (0) so that __create_workqueue() is a single C 
statement.

	Ingo