Re: [patch 2.6.25-rc3] lockdep: add spin_lock_irq_nested()

[David Brownell <david-b@pacbell.net>]

On Tuesday 26 February 2008, Peter Zijlstra wrote:
> 
> On Mon, 2008-02-25 at 14:33 -0800, David Brownell wrote:
> 
> > > +#ifdef CONFIG_LOCKDEP
> > > +
> > > +/* tell lockdep that this IRQ's locks and its parent's locks are in
> > > + * different categories, so that it won't detect false recursion.
> > > + */
> > > +static struct lock_class_key gpio_lock_class;
> > > +
> > > +static inline void mark_gpio_locking(unsigned gpio_irq)
> > > +{
> > > +	lockdep_set_class(&irq_desc[gpio_irq].lock, &gpio_lock_class);
> > > +}
> > > +
> > > +#else
> > > +
> > > +	...
> 
> Glad to hear this works out for you.

Yeah, glad to have a localized fix rather than mucking with genirq.


> Just one note, you don't need the #ifdef mess here. struct
> lock_class_key is 0 bytes on !LOCKDEP and lockdep_set_class*() is
> defined away as well.

Hmm, then kernel/irq/handle.c shouldn't need #ifdefs either ... ;)