Re: [PATCH V3] thermal/core/power_allocator: avoid thermal cdev can not be reset

[Lukasz Luba <lukasz.luba@arm.com>]

On 4/14/23 16:06, Daniel Lezcano wrote:
> On 14/04/2023 16:18, Lukasz Luba wrote:
>>
>>
>> On 4/14/23 12:12, Daniel Lezcano wrote:
>>> On 13/04/2023 10:40, Di Shen wrote:
>>>> We have discussed this question in patch-v1:
>>>> https://lore.kernel.org/all/f6aaa5f1-495d-a158-14d8-ddb2bffbd9c2@arm.com/
>>>>
>>>> Simply put, we use the trip_temp in the Android System; set different
>>>> trip_temp for thermal control of different scenarios.
>>>
>>> The changes are dealing with the trip points and trying to detect the 
>>> threshold. That part should be handled in the thermal core or thermal 
>>> trip side, not in the governor.
>>>
>>> AFAICT, if a trip point is changed, then the power allocator should 
>>> be reset, including the cdev state.
>>>
>>> It would be more convenient to add an ops to the governor ops 
>>> structure to reset the governor and then call it when a trip point is 
>>> changed in thermal_zone_set_trip()
>>>
>>>
>>
>> Sounds reasonable to have a proper API and fwk handling this corner
>> case scenario.
>> Although, if there is a need for a 'easy-to-backport' fix for IPA only,
>> I agree with this patch, since it's straight forward to put in some
>> Android kernel. We can later fix the framework to handle this properly.
>> Anyway, both ways are OK to me.
> 
> Unfortunately, we can not do the maintenance of the Linux kernel based 
> on an 'easy-to-backport' policy to Android.
> 
> This patch could be applied from-list to Android as a hotfix. But for 
> Linux the fix should be more elaborated. One solution is to add a 
> 'reset' ops and call it from the trip point update function.

Fair enough.

> 
> Did you double check the issue is not impacting the other governors too ?

No, unfortunately, I haven't checked other governors.