Re: [PATCHv2 3/5] net: rfkill: gpio: remove gpio names

[Stephen Warren <swarren-3lzwWm7+Weoh9ZMKESR00Q@public.gmane.org>]

On 03/06/2014 08:43 PM, Chen-Yu Tsai wrote:
> On Fri, Mar 7, 2014 at 11:41 AM, Linus Walleij <linus.walleij-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org> wrote:
>> On Wed, Mar 5, 2014 at 10:59 AM, Stephen Warren <swarren-3lzwWm7+Weoh9ZMKESR00Q@public.gmane.org> wrote:
>>> On 03/04/2014 07:37 PM, Linus Walleij wrote:
>>>> On Wed, Mar 5, 2014 at 10:18 AM, Stephen Warren <swarren-3lzwWm7+Weoh9ZMKESR00Q@public.gmane.org> wrote:
>>>>> On 03/04/2014 06:43 PM, Linus Walleij wrote:
>>>>
>>>>>> If I understand the situation correctly it's like ACPI does not have named
>>>>>> GPIOs so keeping specifying this in DT GPIO bindings is counter-productive
>>>>>> to the work of abstracting the access to GPIO handlers so that drivers
>>>>>> need not know whether ACPI or DT is used for describing the hardware.
>>>>>
>>>>> For devices that already have both ACPI and DT bindings, we can't
>>>>> pretend they can be the same; they are already potentially different. We
>>>>> simply need to parse DT and ACPI differently, since that's the sway
>>>>> their bindings are defined.
>>>>>
>>>>> For any devices that don't have both ACPI and DT bindings, I agree we
>>>>> should certainly strive to make any new bindings aligned so we don't
>>>>> have to deal with this for them.
>>>>>
>>>>> However, we can't change the past.
>>>>
>>>> Yeah, right, so for this very driver there are no bindings defined (yet)
>>>> and the only device tree I can find referencing it is the Tegra20-paz00
>>>> and it just use gpios = <>;
>>>>
>>>> So in this case I think this patch is the right way forward, but I admit
>>>> I'm really uncertain in the general case.
>>>
>>> If there are no bindings defined at all yet, then we can define both DT
>>> and ACPI bindings to use name-based GPIOs. Index-based lookups aren't a
>>> good way forward.
>>
>> After Mark clarifying that ACPI is going to have named GPIOs I'm
>> totally aligned on this, so OK!
> 
> Glad to hear this, but is it possible to get rid of the index in current
> drivers? Or change the behavior to name-based OR index-based lookups.
> This might break any DTs that have multiple GPIOs defined under one
> property though.

For any bindings that are already defined to use index-based lookups, I
think we have to continue using them, for backwards-compatibility with
old DTs (and I assume old ACPI databases need the same thing).