Re: [PATCH] RFC: gpio: add API to be strict about GPIO IRQ usage

[Linus Walleij <linus.walleij@linaro.org>]

On Fri, Oct 11, 2013 at 9:10 PM, Stephen Warren <swarren@wwwdotorg.org> wrote:
> On 10/11/2013 02:39 AM, Linus Walleij wrote:

>> However that is not the whole story, is it? We have a gazillion
>> drivers calling irq_create_mapping() in this function, so I would
>> say that things are already a mess here.
>
> I expect things are a mess indeed:-)
>
> I believe that if a driver is only calling irq_create_mapping() inside
> gpio_to_irq(), it's a bug. I think things can operate correctly in one
> of two cases, at least with DT:
>
> 1) irq_create_mapping() is called from both gpio_to_irq() and the
> of_xlate callback for IRQs.
>
> (I don't think this method would work in a board-file-based system where
> of_xlate isn't called for IRQs...)
>
> or:
>
> 2) irq_create_mapping() is called for all IRQs when registering the IRQ
> controller/domain.
>
> To me, (2) is much simpler, and avoids the issue (1) probably has with
> only supporting direct IRQ usage (without something calling gpio_to_irq()).

Hm. Jean-Christophe says (1) is the way to go ... This is
arch/arm/mach-at91/gpio.c I guess.

But this seems to call irq_create_mapping() for all IRQs on
all gpiochips in at91_gpio_irq_setup() which is called by the board
init code?

Maybe I'm not following this :-/

I'll take a look over "my" GPIO drivers and see how I can make
those a bit more elegant so as to set a good example.

Yours,
Linus Walleij