[PATCH] pxa: fix gpio_to_irq() for GPIOs on expander chips

[Eric Miao]

On Wed, Dec 23, 2009 at 6:23 PM, Marc Zyngier <maz at misterjones.org> wrote:
> On Wed, 23 Dec 2009 04:07:34 -0500
> Haojian Zhuang <haojian.zhuang at gmail.com> wrote:
>
>> I not prefer the patch in that link
>> (http://www.spinics.net/lists/arm-kernel/msg53803.html). I think
>> gpio_to_irq() macro is enough. When we handle gpio expander, we could
>> implement an independant gpio_to_irq() function in expander driver.
>> It's like Marc's patch
>> ((http://www.spinics.net/lists/arm-kernel/msg79587.html).
>
> I'm not sure I really get it. In my patch, the pca953x_gpio_to_irq()
> function can only be called from __gpio_to_irq(). As such, it is only
> useful with Philipp's patch when used on PXA.
>
>> In platform driver, we can declare the relationship directly in below.
>>
>>  static struct pca953x_platform_data gpio_exp[] = {
>>         [0] = {
>>                .gpio_base      = PXA_GPIO_IRQ_NUM,
>>                .irq_base       = PXA_GPIO_IRQ_BASE + PXA_GPIO_IRQ_NUM,
>>         },
>>         [1] = {
>>                .gpio_base      = PXA_GPIO_IRQ_NUM + 16,
>>                .irq_base       = PXA_GPIO_IRQ_BASE + PXA_GPIO_IRQ_NUM + 16,
>>         },
>>  };
>>
>> But I don't suggest to append ARCH_NR_GPIOS in gpio.h. Could it be
>> appended into platform driver only? Since not each platform have more
>> than 256 GPIOs.
>
> We could mimic what s3c2410 does to provide some extra GPIO space for
> platforms that require it (arch/arm/mach-s3c2410/include/mach/gpio.h):
>
> #define ARCH_NR_GPIOS   (256 + CONFIG_S3C24XX_GPIO_EXTRA)
>

That's going to be making it impossible for mach-* to be built into a
single kernel (call me ambitious :-).

Well, I actually don't mind increasing GPIO number to a fixed large
enough value. It seems to hit the same problem as IRQ which I'm
thinking of cleanup these days, the trade-off of using a large fixed
number is that some spaces are wasted indeed.

>        M.
> --
> Fast. Cheap. Reliable. Pick two.
>