[PATCH v3] gpio: UniPhier: add driver for UniPhier GPIO controller

[Linus Walleij]

On Thu, Jul 16, 2015 at 9:35 AM, Masahiro Yamada
<yamada.masahiro at socionext.com> wrote:
> 2015-07-16 16:07 GMT+09:00 Linus Walleij <linus.walleij at linaro.org>:

>>> ngpio == 248 for some SoCs,
>>> and ngpio == 136 for some, etc.
>>
>> That is the wrong way to handle different SoC. That should be handled
>> by different compatible strings, and then you select the number of GPIOs
>> for the version corresponding to that compatibe string.
>>
>>>>> +static const struct of_device_id uniphier_gpio_match[] = {
>>>>> +       { .compatible = "socionext,uniphier-gpio" },
>>>>> +       { /* sentinel */ }
>>>>> +};
>>
>> i.e. you should use the .data field of of_device_id to carry variant-specific
>> information.
>
>
> Currently, I want to use this driver on 7 SoCs
>
> PH1-sLD3:    ngpio == 136
> PH1-LD4 :    ngpio == 136
> PH1-Pro4:    ngpio == 248
> PH1-sLD8:    ngpio == 136
> PH1-Pro5:    ngpio == 248
> ProXstream2: ngpio == 232
> PH1-LD6b:    ngpio == 232
>
> So, should I describe the OF match table like this?
>
> static const struct of_device_id uniphier_gpio_match[] = {
>        { .compatible = "socionext,ph1-sld3-gpio"       .data = (void *)136 },
>        { .compatible = "socionext,ph1-ld4-gpio"        .data = (void *)136 },
>        { .compatible = "socionext,ph1-pro4-gpio"       .data = (void *)248 },
>        { .compatible = "socionext,ph1-sld8-gpio"       .data = (void *)136 },
>        { .compatible = "socionext,ph1-pro5-gpio"       .data = (void *)248 },
>        { .compatible = "socionext,proxstream2-gpio",   .data = (void *)232 },
>        { .compatible = "socionext,ph1-ld6b-gpio",      .data = (void *)232 },
>        { /* sentinel */ }
> };

Yes.

> One disadvantage for this way is that
> I need to touch the driver file every time I add a new SoC support.

That is appropriate since it is a new hardware. This is the same as
the fact that we touch the kernel to add new USB IDs and PCI IDs
every time a new hardware comes out for x86, we should know
what hardware we are toying with.

Since you seem to have a pin controller in parallel anyways I see
it as natural to do this at the same time as you do that anyways.

Yours,
Linus Walleij