Re: [PATCH 2/3] Add hard_irq_disable()

["Satyam Sharma" <satyam.sharma@gmail.com>]

Hi Andrew,

On 5/10/07, Andrew Morton <akpm@linux-foundation.org> wrote:
> On Thu, 10 May 2007 15:25:58 +1000 Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
>
> > --- linux-cell.orig/include/linux/interrupt.h 2007-05-10 14:51:22.000000000 +1000
> > +++ linux-cell/include/linux/interrupt.h      2007-05-10 15:18:04.000000000 +1000
> > @@ -241,6 +241,16 @@ static inline void __deprecated save_and
> >  #define save_and_cli(x)      save_and_cli(&x)
> >  #endif /* CONFIG_SMP */
> >
> > +/* Some architectures might implement lazy enabling/disabling of
> > + * interrupts. In some cases, such as stop_machine, we might want
> > + * to ensure that after a local_irq_disable(), interrupts have
> > + * really been disabled in hardware. Such architectures need to
> > + * implement the following hook.
> > + */
> > +#ifndef hard_irq_disable
> > +#define hard_irq_disable()   do { } while(0)
> > +#endif
>
> We absolutely require that the architecture's hard_irq_disable() be defined
> when we do this.  If it happens that the definition of hard_irq_disable()
> is implemented three levels deep in nested includes then we risk getting
> into a situation where different .c files see different implementations of
> hard_irq_disable(), which could lead to confusing results, to say the least.

So you're saying that this mechanism forces the arch (that really
wants hard_irq_disable) to _#define_ hard_irq_disable (as a macro),
and if it implements it as an inline function, for example, then we're
screwed?

> Your implementation comes via the inclusion of system.h which then includes
> hw_irq.h.  That's perhaps a little fragile and it would be better to
>
> a) include in the comment the name of the arch file which must implement
>    hard_irq_disable() and
>
> b) include that file directly from this one.

Hmmm. How about:

1. Introduce some CONFIG_WANTS_HARD_IRQ_DISABLE that is #defined (or
left undefined) by the arch/.../defconfig (depending upon whether or
not that arch implements a hard_irq_disable() for itself or not)

2. Then pull-in that code into include/linux/interrupt.h somehow
(through some known / fixed header file, or through asm/system.h, or
anyhow -- it doesn't really matter)

3. And:

#ifndef CONFIG_WANTS_HARD_IRQ_DISABLE
#define hard_irq_disable() do { } while(0)
#endif

We don't need to standardize on some particular arch-specific header
filename in this case.

Comments?

Satyam