[PATCH 1/1] gpio: drop CONFIG_OF_GPIO in the definition of the struct gpio_chip

[nicolas.ferre@atmel.com (Nicolas Ferre)]

On 02/13/2012 03:33 PM, Rob Herring :
> On 02/13/2012 08:12 AM, Jean-Christophe PLAGNIOL-VILLARD wrote:
>> On 08:00 Mon 13 Feb     , Rob Herring wrote:
>>> On 02/13/2012 03:23 AM, Jean-Christophe PLAGNIOL-VILLARD wrote:
>>>> this will allow to avoid the ifdef CONFIG_OF_GPIO in gpio drivers
>>>>
>>>
>>> So would always enabling CONFIG_OF on at91 and then only 1 sub-arch is
>>> paying the price for additional space. Then we would only have ifdefs in
>>> the truly cross-platform gpio drivers.
>> we talk about 12bytes and here it's force to put a ifdef in the gpio drivers
>> in platform_device we don't do so. So why should be do it here??
> 
> But you are affecting everyone. I'm not saying we shouldn't do this. I'm
> just highlighting the reason this is a problem and gpio is just one
> instance. I think sub-arches should either always select OF or not.
> Trying to do both is just going to be a pain to maintain.

(nothing related to the particular case discussed here, but...)

Rob,

It may be a pain to maintain (and I can tell it is), but we do not have
the choice: the AT91 family is simply too widely spread in terms of
number of SoC and release date of those SoC. We must maintain all of
them working while we are doing the effort to switch most of the
material to DT. As this work takes time I fear that the transition
period will last for a long time.

Older SoCs will surely never be converted to device-tree as both the
lack of workforce, customer habits and amount of boards existing and
working correctly in mainline will go against this.

Best regards,
-- 
Nicolas Ferre