[PATCH V5 18/20] ARM: exynos: cpuidle: Pass the AFTR callback to the platform_data

[daniel.lezcano@linaro.org (Daniel Lezcano)]

On 05/21/2014 10:10 AM, Arnd Bergmann wrote:
> On Wednesday 21 May 2014 09:15:34 Daniel Lezcano wrote:
>> On 05/15/2014 10:40 PM, Kukjin Kim wrote:
>>
>> [ ... ]
>>
>>>>>> Exynos cpuidle is not a device on the SoC, so I don't think there is
>>>>>> any
>>>>>> way to represent it in DT. The only thing I could see this is matching
>>>>>> root node with a central SoC driver that instantiates specific
>>>>>> subdevices, such as cpufreq and cpuidle, but I don't see any available
>>>>>> infrastructure for this.
>>>>>
>>>>> There is a RFC for defining generic idle states [1].
>>>>>
>>>>> The idea behind using the platform driver framework is to unify the code
>>>>> across the different drivers and separate the PM / cpuidle code.
>>>>>
>>>>> By this way, we can move the different drivers to drivers/cpuidle and
>>>>> store them in a single place. That make easier the tracking, the review
>>>>> and the maintenance.
>
> Yes, that would be great. I only looked briefly at the series now, doesn't
> that require the use of psci?

No, because PSCI is for some specific platform (eg. calxeda), all the 
other drivers are legacy and manually handling the PM via some low level 
callbacks. This is why all drivers were implemented all over the place 
making so difficult to maintain them. Little by little, we split the PM 
callbacks from the idle algorithm so reducing the platform dependency 
with the generic code.

The PSCI implements such PM callbacks in the firmware directly and are 
accessed through an API. PSCI is, let's say some kindof nextgen cpuidle. 
It is similar than mwait on Intel. But if a new platform does not have 
such firmware, then the cpuidle driver will have the legacy format.

> That's not a bad idea of course, but it
> doesn't solve the problem I raised here.
>
>>>>> I am ok to by using platform_device_register_resndata() but I would
>>>>> prefer to do that a bit later by converting the other drivers too. That
>>>>> will be easier if we have them grouped in a single directory (this is
>>>>> what does this patchset at the end).
>>>>>
>>>>> As there are some more work based on this patchset and the link error
>>>>> could be fixed as an independent patch, I would recommend to
>>>>> re-integrate it in the tree as asked by Bartlomiej.
>>>>
>>>> In general, it would be nice to have everything done properly, but I'd
>>>> consider Daniel's series as a huge improvement already and a nice
>>>> intermediate step towards further clean-up.
>>>>
>>>> So based on the comments quoted above, instead of stalling the
>>>> development, I'd suggest to accept this series and then move forward.
>>>>
>>> I'm fine.
>>>
>>> Arnd, how about you?
>>>
>>> - Kukjin
>>
>> Arnd ?
>
> Sorry for the delay.

No problem.

> Yes, let's do it this way once more, but please
> come up with something better for the future that doesn't tie the
> cpuidle method to the root compatible string as this does.

ok.

Thanks

   -- Daniel

-- 
  <http://www.linaro.org/> Linaro.org ? Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog