Re: [mmotm and linux-next][PATCH] irq: enclose irq_desc_lock_class in CONFIG_LOCKDEP

[Ingo Molnar <mingo@elte.hu>]

* KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> wrote:

> >> > actually, this breaks the build on !SPARSEIRQ because we will use that
> >> > class in the non-sparseirq case. So we've converted a build warning to
> >> > a build failure ;-)
> >>
> >> Please give me your .config and tell me your arch. my ia64 box (ia64 is
> >> !SPARSEIRQ) can build the akpm patch.
> >
> > The expected build failure is obvious from reading the code:
> >
> >  #ifdef CONFIG_TRACE_IRQFLAGS
> >  void early_init_irq_lock_class(void)
> >  {
> >  #ifndef CONFIG_SPARSE_IRQ
> >         struct irq_desc *desc;
> >         int i;
> >
> >         for_each_irq_desc(i, desc) {
> >                 if (!desc)
> >                         continue;
> >
> >                 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
> >
> > Note that it's an #ifndef sparseirq, not an #ifdef sparseirq condition.
> 
> I see. thanks.
> It seems my first proposal is better.
> 
> or, following #ifdef ?
> 
> #if defined(CONFIG_SPARSE_IRQ) || defined(CONFIG_TRACE_IRQFLAGS)
> 
>  /*
>  * lockdep: we want to handle all irq_desc locks as a single lock-class:
>  */
>  static struct lock_class_key irq_desc_lock_class;

instead of increasing the #ifdef jungle, how about removing some? For 
example is this distinction:

> >  #ifndef CONFIG_SPARSE_IRQ

really needed? We should use symmetric lock class annotations, regardless 
of how irq_desc[] is laid out.

	Ingo