Re: [PATCH v1 4/4] leds: no longer use unnamed gpios

[Alexandre Courbot <gnurou@gmail.com>]

On Mon, Jan 19, 2015 at 12:43 PM, Alexandre Courbot <gnurou@gmail.com> wrote:
> On Wed, Jan 14, 2015 at 10:20 PM, Olliver Schinagl <oliver@schinagl.nl> wrote:
>>
>> On 14-01-15 13:45, Linus Walleij wrote:
>>>
>>> On Thu, Jan 8, 2015 at 11:12 PM, Dmitry Torokhov
>>> <dmitry.torokhov@gmail.com> wrote:
>>>>
>>>> On Thu, Jan 08, 2015 at 08:40:20AM -0600, Rob Herring wrote:
>>>>>
>>>>> On Thu, Jan 8, 2015 at 2:45 AM, Olliver Schinagl <oliver@schinagl.nl>
>>>>> wrote:
>>>>>>>>
>>>>>>>> --- a/drivers/leds/leds-gpio.c
>>>>>>>> +++ b/drivers/leds/leds-gpio.c
>>>>>>>> @@ -184,7 +184,7 @@ static struct gpio_leds_priv
>>>>>>>> *gpio_leds_create(struct
>>>>>>>> platform_device *pdev)
>>>>>>>>                  struct gpio_led led = {};
>>>>>>>>                  const char *state = NULL;
>>>>>>>>    -             led.gpiod = devm_get_gpiod_from_child(dev, NULL,
>>>>>>>> child);
>>>>>>>> +               led.gpiod = devm_get_gpiod_from_child(dev, "led",
>>>>>>>> child);
>>>>>>>
>>>>>>> Would not this break existing boards using old bindings? You need to
>>>>>>> handle both cases: if you can't located "led-gpios" then you will have
>>>>>>> to
>>>>>>> try just "gpios".
>>>>>>
>>>>>> Very true. I was rather even hoping we could update all bindings, I
>>>>>> don't
>>>>>> mind going through the available dts files to fix them ... But need to
>>>>>> know
>>>>>> that that's the proper way to go before doing the work ;)
>>>>>
>>>>> That will not work. You cannot make changes that require a new dtb
>>>>> with a new kernel. This would also break for the other way around
>>>>> (i.e. a new dtb and old kernel).
>>>>>
>>>>> You would have to search for both led-gpios and gpios. I'm not sure if
>>>>> we can do that generically for all GPIOs. If you had a node with both
>>>>> "blah-gpios" and "gpios", it would break. I would hope there are no
>>>>> such cases like that. We also now have to consider how ACPI identifies
>>>>> GPIOs and whether this makes sense.
>>>>
>>>> I think only the driver itself can know about such "legacy" mappings and
>>>> make a decision.
>>>
>>> Yeah leds-gpio.c will need to be patched to check for "led-gpios" first
>>> and then fall back to "gpios" if not found.
>>
>> yeah I've done the work, just need to double check it and resend a v2.
>>
>> Linus, I assume you want the already applied patches omitted from v2?
>
> Yes, please base new revisions on Linus' devel branch, omitting any
> patches that he has already accepted.

Actually on top of breaking backward compatibility, I think the case
of LED GPIOs is one of thoses where it makes sense to not have a name
(the GPIO usage is obvious from the DT hierarchy, and there cannot be
another one). So I don't feel like this change is really needed (other
patches in this series are still useful though).