Re: [PATCH v2] gpio: Add Avionic Design N-bit GPIO expander support

[Thierry Reding <thierry.reding@avionic-design.de>]

[-- Attachment #1: Type: text/plain, Size: 1491 bytes --]

On Fri, Aug 10, 2012 at 10:19:02AM +0200, Linus Walleij wrote:
> On Thu, Aug 9, 2012 at 10:20 PM, Thierry Reding
> <thierry.reding@avionic-design.de> wrote:
> > On Sun, Jul 29, 2012 at 07:13:57PM +0200, Linus Walleij wrote:
> >> > +       client->irq = irq_of_parse_and_map(client->dev.of_node, 0);
> >> > +       if (client->irq == NO_IRQ)
> >>
> >> Just if (!client->irq) since NO_IRQ is 0 nowadays.
> >
> > At the risk of seeming pedantic, NO_IRQ is in fact quite often not 0.
> 
> No. A year back, yes, but not anymore. We went to great lengths in the
> ARM architecture to ensure NO_IRQ is *always 0. Russell spent
> a lot of time on this.
> 
> Consult the following article on LWN:
> http://lwn.net/Articles/470820/
> 
> Then grep your gitlog and you'll see we got rid of it from ARM.

Then why is there still the following in arch/arm/include/asm/irq.h?

	/*
	 * Use this value to indicate lack of interrupt
	 * capability
	 */
	#ifndef NO_IRQ
	#define NO_IRQ	((unsigned int)(-1))
	#endif

> If this driver is for some other arch like openrisc I might accept
> it but please reconsider.

There's nothing in the driver that makes it ARM specific, so it could be
used on other platforms just as well. But as I also said in my previous
mail, in this particular case the value for the interrupt comes from the
call to irq_of_parse_and_map(), which will return 0 on failure,
regardless of the architecture, so there is actually no problem.

Thierry

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]