Re: An extremely simplified pinctrl bindings proposal

[Stephen Warren <swarren@nvidia.com>]

On 02/05/2012 07:07 PM, Thomas Abraham wrote:
> Hi Stephen,
> 
> On 4 February 2012 21:31, Stephen Warren <swarren@nvidia.com> wrote:
>> Sorry, I haven't had a chance to read any of the pincrl emails from
>> Friday onwards. However, I thought a bit more about this, and decided
>> to propose someting much simpler:
>>
>> Thoughts:
>>
>> * Defining all the pins, groups, functions, ... takes a lot of space,
>>  whether it's in static data in pinctrl drivers or in the device tree.
>>  The lists must also be stored in RAM at runtime.
>>
>> * It's been very difficult to come up with a generic description of all
>>  pin controller's capabilities. This is true even irrespective of device
>>  tree; think pin config where we've agonized over whether we can create
>>  a standardized list of pin config properties, or need to allow each
>>  pinctrl driver to define its own set of properties, etc.
>>
>> * The only real use of the lists is for debugfs. Drivers shouldn't expect
>>  to directly request specific pinctrl settings, since that would encode
>>  knowledge of an individual SoC's pin controller. This should be
>>  abstracted from drivers.
>>
>> * The data in debugfs could easily be replaced by a raw register dump
>>  coupled with a SoC-specific script to print out what each register
>>  means.
>>
>> My proposal below is to radically simplify the pinctrl subsystem, and
>> make it little more than a system to execute a list of arbitrary register
>> writes.
> 
> Thanks for your work on pinctrl bindings proposal.
> 
> With this new approach, how much of the pinctrl susbystem is used in
> device tree mode.

Probably not a lot of the actual implementation. I'd assume that the
APIs called by device drivers would remain constant, or roughly so, in
order to still provide a simple interface for drivers.

> Last time I had proposed something similar to this
> (http://marc.info/?l=linux-kernel&m=132697880016289&w=2), you were not
> happy about that approach since the pinctrl subsystem is largely
> under-utilized (http://marc.info/?l=linux-kernel&m=132735884622374&w=2),
> expect for providing a interface which individual
> pinctrl/pinmux/pinconfig drivers will use to implement a function that
> programs the hardware.

That's true.

AS LinusW says in his later reply, losing the semantic representation in
the pinctrl system is a pity, which is the main reasoning behind my
previous response. However, I'm beginning to lean towards the simplicity
of something like a list of register writes trumping the lack of semantics.

> There need not be pinmux/pinctrl/pinconfig bindings that are designed
> for the linux pinctrl subsystem. DT allows specifying the
> pinconfig/pinmux properties in a simple way, which was not possible in
> non-dt case and hence the pinctrl subsystem. Other OS'es which might
> not have a pinctrl subsystem, similar to what linux has, should also
> find the pinctrl bindings useful.

True.

I think that pinctrl in a non-DT system probably could do something like
this DT binding proposal though, so we need not have DT/non-DT work
differently.

-- 
nvpublic