[PATCH v2 2/2] cpufreq: ti: Add cpufreq driver to determine available OPPs at runtime

[Dave Gerlach]

On 09/23/2016 12:19 AM, Viresh Kumar wrote:
> On 21-09-16, 14:34, Dave Gerlach wrote:
>> Viresh,
>> On 09/07/2016 10:39 PM, Viresh Kumar wrote:
>>> On 07-09-16, 10:04, Dave Gerlach wrote:
>>>>>> +static const struct of_device_id ti_cpufreq_of_match[] = {
>>>>>> +	{ .compatible = "operating-points-v2-ti-am3352-cpu",
>>>>>> +	  .data = &am3x_soc_data, },
>>>>>> +	{ .compatible = "operating-points-v2-ti-am4372-cpu",
>>>>>> +	  .data = &am4x_soc_data, },
>>>>>> +	{ .compatible = "operating-points-v2-ti-dra7-cpu",
>>>>>> +	  .data = &dra7_soc_data },
>>>>>
>>>>> You should be using your SoC compatible strings here. OPP compatible
>>>>> property isn't supposed to be (mis)used for this purpose.
>>>>>
>>>>
>>>> Referring to my comments in patch 1, what if we end up changing the bindings
>>>> based on DT maintainer comments? We will have these compatible strings, and
>>>> at that point is it acceptable to match against them? Or is it still better
>>>> to match to SoC compatibles? I think it makes sense to just probe against
>>>> these.
>>>
>>> But even then I think these are not correct. You should have added a
>>> single compatible string: operating-points-v2-ti-cpu.
>>>
>>> As the properties will stay the same across machines. And then you
>>> need to use SoC strings here.
>>>
>>
>> Are you opposed to moving _of_get_opp_desc_node from
>> drivers/base/power/opp/opp.h to include/linux/pm_opp.h and renaming it
>> appropriately?
>
> I am not opposed to that, but ...
>
>> If I move the ti properties out of the cpu node, as discussed in patch 1 of
>> this series, and into the operating-points-v2 table, I need a way to get the
>> operating-points-v2 device node and I think it makes sense to reuse this as
>> it is what the opp framework uses internally to parse the phandle to the opp
>> table.
>
> I am not sure if those registers belong to the OPP bindings. What are those
> registers really? What all can be read from them? Why shouldn't they be present
> as a separate node in DT on the respective bus? Look at how it is done for
> sti-cpufreq driver.
>

The sti-cpufreq driver in v4.8-rc7 appears to do what I am already doing 
in this revision of the patch, reading from a syscon phandle that is 
part of the cpu node in the DT which is what I was told not to do.

The register I am referencing in the syscon is a bit-field describing 
which OPPs are valid for the system, so it is very relevant to the OPP 
binding. They really are already present in a separate node, I'm just 
indexing into a syscon, same as the sti-cpufreq driver appears to be doing.

Regards,
Dave