On Thu, Aug 20, 2015 at 09:05:07AM +0200, Sascha Hauer wrote:

On Mon, Aug 10, 2015 at 08:53:34AM +0200, Ludovic Desroches wrote:

quoted

Hi Sascha,

On Wed, Aug 05, 2015 at 09:31:17AM +0200, Sascha Hauer wrote:

quoted

On Fri, Jul 31, 2015 at 05:08:07PM +0200, Ludovic Desroches wrote:

quoted

Hi,

Following our discussion, I send an RFC version of my driver. RFC because it is
not totally achieved, some cleanup and feature addition is needed.

At least, we could discuss about the 'core' part. I have used the pinmux
property as Mediatek driver. Patch 3 is the internal dt files we are using.

As you can imagine I am fine with the binding, so I can add my acked-by
once you send a non-RFC version.

Great, I'm glad to hear that.
 

quoted

The only thing I never understood is what's so special about GPIOs that
they have to bypass the pinctrl framework and instead a gpio_request
magically translates a gpio into a pin.

Not sure to really understand your concern here... Do you mean I could
get rid of gpio_request_enable()?

I would expect a gpio to be a pin like every other pin, hence configured
via the pinctrl framework and not implicitly via gpio_request().

It is not an issue to get rid of gpio_request_*(). I was thinking it was
a bonus to provide it.

quoted

quoted

Wouldn't it make sense to at
least add the pins in their GPIO mode to
arch/arm/boot/dts/sama5d2-pinfunc.h?

It is done, PIN_PA0 could be used for this purpose.

I would expect a define like:

#define PIN_PA3__GPIO                   PINMUX_PIN(PIN_PA3, 0, 2)

PIN_PAx only contains the pin number, but not the function.

Ok I can do this change for v2.


Ludovic