[PATCH v6 0/4] Runtime Interpreted Power Sequences

[Stephen Warren]

On 10/03/2012 02:24 AM, Alex Courbot wrote:
> On 09/14/2012 12:24 AM, Stephen Warren wrote:
>> On 09/13/2012 01:29 AM, Mark Brown wrote:
>>> On Thu, Sep 13, 2012 at 04:26:34PM +0900, Alex Courbot wrote:
>>>> On Thursday 13 September 2012 15:19:30 Mark Brown wrote:
>>>>>> On Thursday 13 September 2012 14:25:53 Mark Brown wrote:
>>>>>>> It would be sensible to make sure that the framework is done in
>>>>>>> such a
>>>>>>> way that drivers can use it - there will be drivers (perhaps not
>>>>>>> display
>>>>>>> ones) that have a known power sequence and which could benefit
>>>>>>> from the
>>>>>>> ability to use library code to implement it based on the user simply
>>>>>>> supplying named resources.
>>>
>>>>>> Not sure I understand what you mean, but things should be working
>>>>>> this way
>>>>>> already - regulators and PWMs are acquired by name using the standard
>>>>>> regulator_get() and pwm_get() functions. GPIOs do not, AFAIK, have
>>>>>> a way
>>>>>> to be referenced by name so their number is used instead.
>>>
>>>>> Right, but the sequencing for enabling them is currently open coded in
>>>>> each driver.
>>>
>>>> Mmm then I'm afraid I don't see what you wanted to say initially -
>>>> could you
>>>> elaborate?
>>>
>>> The driver knows the power sequence.  Having to type the same sequence
>>> into the DT or platform data for each board using the device wouuld be
>>> retarded so we need the drivers to be able to give the sequence to the
>>> library if they're going to be able to reuse it (which is a lot of what
>>> Tomi is talking about).
>>
>> I believe that's trivial to implement. The relevant function is:
>>
>> struct power_seq_set *devm_power_seq_set_build(struct device *dev,
>>            struct platform_power_seq_set *pseq);
>>
>> It's up to the driver whether pseq comes from platform data or is
>> hard-coded into the driver (or not provided at all, for the DT case).
>> So, the only change needed to convert a "hard-coded" driver to this API
>> is to convert the current custom data structure (or code) that describes
>> the sequence into a struct platform_power_seq_set.
> 
> If we go this way (which looks good IMO!), then maybe we should abandon
> that "platform" denomination and merge platform_power_seq* structures
> with the currently private power_seq*, and also replace the "building"
> step with a resources acquisition one. Calling these structures
> "platform" implies they are for platform data while they can be used to
> perform more flexible things as Mark mentioned.

That all seems reasonable.

> Also making the resolved
> resource visible would allow drivers to "patch" generic sequences with
> the proper GPIO numbers at runtime.

That doesn't sound like a great idea to me, but we can simply avoid
doing this even though it's technically possible.

> We would also avoid a few memory
> copies and both design and usage would be simplified, at the cost of
> having more things exposed. How does that sound?

Sounds fine to me at least.