Re: [PATCH v3 2/3] gpio: pca953x: add support for pca9505

[Gregory CLEMENT <gregory.clement@free-electrons.com>]

On 01/25/2013 09:16 AM, Linus Walleij wrote:
> On Tue, Jan 22, 2013 at 10:10 PM, Gregory CLEMENT
> <gregory.clement@free-electrons.com> wrote:
> 
>> Now that pca953x driver can handle GPIO expanders with more than 32
>> bits this patch adds the support for the pca9505 which cam with 40
>> GPIOs.
>>
>> Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com>
> 
> Patch applied, thanks.
> 
> But guys, this driver contains some horrific stuff, look at this:
> 
> static int pxa_gpio_nums(void)
> {
>         int count = 0;
> 
> #ifdef CONFIG_ARCH_PXA
>         if (cpu_is_pxa25x()) {
> #ifdef CONFIG_CPU_PXA26x
>                 count = 89;
>                 gpio_type = PXA26X_GPIO;
> #elif defined(CONFIG_PXA25x)
>                 count = 84;
>                 gpio_type = PXA26X_GPIO;
> #endif /* CONFIG_CPU_PXA26x */
>         } else if (cpu_is_pxa27x()) {
>                 count = 120;
>                 gpio_type = PXA27X_GPIO;
>         } else if (cpu_is_pxa93x()) {
>                 count = 191;
>                 gpio_type = PXA93X_GPIO;
>         } else if (cpu_is_pxa3xx()) {
>                 count = 127;
>                 gpio_type = PXA3XX_GPIO;
>         }
> #endif /* CONFIG_ARCH_PXA */
> 
> #ifdef CONFIG_ARCH_MMP
>         if (cpu_is_pxa168() || cpu_is_pxa910()) {
>                 count = 127;
>                 gpio_type = MMP_GPIO;
>         } else if (cpu_is_mmp2()) {
>                 count = 191;
>                 gpio_type = MMP_GPIO;
>         }
> #endif /* CONFIG_ARCH_MMP */
>         return count;
> }
> 
> This is totally killing all attempts at a single kernel for multiple
> machines in this family. The above should not be #ifdef's but instead
> use either platform data or the compatible property to figure out which
> one to use.
> 
> It's fine to introduce new compatible= strings or device names to
> distinguish between these.
> 
> As an example, in pinctrl-nomadik.c we have this:
> 
> static const struct of_device_id nmk_pinctrl_match[] = {
>         {
>                 .compatible = "stericsson,nmk_pinctrl",
>                 .data = (void *)PINCTRL_NMK_DB8500,
>         },
>         {},
> };
> 
> static const struct platform_device_id nmk_pinctrl_id[] = {
>         { "pinctrl-stn8815", PINCTRL_NMK_STN8815 },
>         { "pinctrl-db8500", PINCTRL_NMK_DB8500 },
>         { "pinctrl-db8540", PINCTRL_NMK_DB8540 },
>         { }
> };
> 
> static struct platform_driver nmk_pinctrl_driver = {
>         .driver = {
>                 .owner = THIS_MODULE,
>                 .name = "pinctrl-nomadik",
>                 .of_match_table = nmk_pinctrl_match,
>         },
>         .probe = nmk_pinctrl_probe,
>         .id_table = nmk_pinctrl_id,
> };
> 
> 
> The first match is for DT boot the latter for using the platform
> device name directly.
> 
> Then in the probefunction we do:
> 
> static int nmk_pinctrl_probe(struct platform_device *pdev)
> {
>         const struct platform_device_id *platid = platform_get_device_id(pdev);
> (...)
>         if (platid)
>                 version = platid->driver_data;
>         else if (np) {
>                 const struct of_device_id *match;
> 
>                 match = of_match_device(nmk_pinctrl_match, &pdev->dev);
>                 if (!match)
>                         return -ENODEV;
>                 version = (unsigned int) match->data;
>         }
> (...)
>         if (version == PINCTRL_NMK_STN8815)
>                 nmk_pinctrl_stn8815_init(&npct->soc);
>         if (version == PINCTRL_NMK_DB8500)
>                 nmk_pinctrl_db8500_init(&npct->soc);
>         if (version == PINCTRL_NMK_DB8540)
>                 nmk_pinctrl_db8540_init(&npct->soc);
> }
> 
> Surely a scheme like this must be possible to use to distinguish between
> the different versions at runtime rather than using these #ifdefs?
> 

Well, at the beginning I thought adding support for pca9505 was just a matter
of a couple of lines to add. Then I realized that I need to handle the 40 bits
case, and I ended up refactoring all access to the registers. So now I am on it,
it seems I am volunteer to continue to improve this driver.

However I won't be able to test it, the only PXA based platform I have is a
Zaurus SL-C3100 which embeds a PXA270 if I remember well, but I doubt it come
with gpio expander on i2c.

Gregory