Re: [RFC PATCH] gpio: add GPIO hogging mechanism

[boris brezillon <b.brezillon@overkiz.com>]

Hello,

On 19/12/2013 19:22, Felipe Balbi wrote:
> On Thu, Dec 19, 2013 at 06:18:40PM +0100, boris brezillon wrote:
>> Hello Felipe,
>>
>> On 19/12/2013 17:47, Felipe Balbi wrote:
>>> On Thu, Dec 19, 2013 at 08:41:09AM -0800, Greg Kroah-Hartman wrote:
>>>> On Thu, Dec 19, 2013 at 03:34:31PM +0100, Boris BREZILLON wrote:
>>>>> GPIO hogging is a way to request and configure specific GPIO without
>>>>> explicitly requesting it in the device driver.
>>>>>
>>>>> The request and configuration procedure is handled in the core device
>>>>> driver code before the driver probe function is called.
>>>>>
>>>>> It allows specific GPIOs to be configured without any driver specific code.
>>>>>
>>>>> Particularly usefull when a external device is connected to a bus and the
>>>>> bus connections depends on an external switch controlled by a GPIO pin.
>>> for external switches, you probably need a pinctrl-gpio driver.
>> Do you mean using pinctrl pinconf to configure the PIN as output-high or
>> output-low ?
>>
>> This was my first proposal
>> (see https://www.mail-archive.com/devicetree@vger.kernel.org/msg05829.html).

My mistake: this is not exactly what I proposed in my patch series.

Actually, I was directly requesting the pin connected to the external 
switch as
OUTPUT + (OUTPUT-LEVEL) according the the device needs:
- output high for MMC slot
- output low for SPI device

And, I guess this is why Linus asked me to find a better solution.

IMHO your approach is, by far, much better. You expose the external switch
as a PIN muxing device and the devices connected to through this PIN mux
block just have to request the appropriate PIN states.

In my specific case this would give the following:
- MMC conf for mmc slot 0
- SPI conf for the SPI device

With your approach the HW representation is better: the different signals
controlled by the external switch can be seen using debugfs, and device
tree definition is closer to the real HW design.

> that's quite a weird argument from Linus W, considering you _do_ have a
> discrete mux on the board.

> We have quite a few of such "crazy" scenarios here at TI and we were
> going to send a pinctrl-gpio driver. If that's not acceptable, then I
> suppose there is no way to boot from NAND on a board where NAND signals
> go through a discrete mux where the select signal is a GPIO pin.
>

Please keep going with the submission process: I'm really interested in this
driver (BTW could you add me in the CC list ?).

Linus, tell me if I'm wrong, but I think, the pinctrl-gpio is the right 
way to solve
the at91rm9200ek board use case.

This does not mean, the gpio hogs approach does not solve other issues (see
https://lkml.org/lkml/2013/8/29/333 and
https://www.mail-archive.com/linux-gpio@vger.kernel.org/msg01084.html), 
but in
my specific case, I'd rather use the pinctrl-gpio driver.

Best Regards,

Boris