18.12.2020 00:19, Daniel Lezcano пишет:
> On 17/12/2020 21:28, Dmitry Osipenko wrote:
>> 17.12.2020 22:36, Daniel Lezcano пишет:
>>>>>> + type = "critical";
>>>>>> + };
>>>>>> + };
>>>>>> +
>>>>>> + cooling-maps {
>>>>>> + map0 {
>>>>>> + trip = <&trip0>;
>>>>>> + cooling-device = <&cpu0 THERMAL_NO_LIMIT THERMAL_NO_LIMIT>;
>>>>> You should add all CPUs here.
>>>>
>>>> All CPU cores are coupled on Tegra in regards to CPUFreq, hence I think
>>>> it won't make any difference if secondary CPU cores will be added here,
>>>> isn't it?
>>> The explanation is in the description of commit ef4734500407ce4d
>>
>> I think that really only makes sense if CPU cores have independent clock
>> rate management.
>
> ATM I did not see any ARM platform having a clock line per CPU but I may
> be wrong.
>
>> IIRC, I actually made some research about this in the
>> past and intentionally removed the secondary cores from the
>> cooling-device since they didn't make any difference for a coupled CPU
>> cores.
>>
>> That commit also says:
>>
>> "But as soon as this CPU ordering changes and any other CPU is used to
>> bring up the cooling device, we will start seeing failures."
>>
>> I don't quite understand to what "failures" that commit referrers. I
>> tried to change the cpu0 to cpu1 in the cooling-device and don't see any
>> failures. Could you please clarify this?
>>
>> In general it should be fine to add all the cores to the cooling-device
>> and I'll do it in v3, but I want to make it clear why this is needed.
>
> AFAIR, if CPU0 is unplugged the cooling device can not rebind to CPU1.
> And if CPU0 is plugged in again, the cooling device fails to initialize.
>
> And, if the CPUs are mapped with the physical CPU0 to Linux numbering
> CPU1, the cooling device mapping will fail.
Alright, thank you.
