Re: [PATCH v3 3/3] [RFC] CPUFreq: Add support for cpu-perf-dependencies

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

Hi Viresh,

On 11/6/20 9:20 AM, Viresh Kumar wrote:
> On 02-11-20, 12:01, Nicola Mazzucato wrote:
>> This is a continuation of the previous v2, where we focused mostly on the
>> dt binding.
>>
>> I am seeking some feedback/comments on the following two approaches.
>>
>> Intro:
>> We have seen that in a system where performance control and hardware
>> description do not match (i.e. per-cpu), we still need the information of
>> how the v/f lines are shared among the cpus. We call this information
>> "performance dependencies".
>> We got this info through the opp-shared (the previous 2 patches aim for
>> that).
>>
>> Problem:
>> How do we share such info (retrieved from a cpufreq driver) to other
>> consumers that rely on it? I have two proposals.
> 
> I haven't really stop thinking about what and how we should solve
> this, but I have few concerns first.
> 
>> 2) drivers/thermal/cpufreq_cooling: Replace related_cpus with dependent_cpus
> 
> I am not sure if I understand completely on how this is going to be
> modified/work.
> 
> The only use of related_cpus in the cooling driver is in the helper
> cdev->get_requested_power(), where we need to find the total power
> being consumed by devices controlled by the cooling device. Right ?
> 
> Now the cooling devices today are very closely related to the cpufreq
> policy, the registration function itself takes a cpufreq policy as an
> argument.
> 
> Consider that you have an octa-core platform and all the CPUs are
> dependent on each other. With your suggested changes and hw control,
> we will have different cpufreq policies for each CPU. And so we will
> have a cooling device, cdev, for each CPU as well. When the IPA
> governor calls cdev->get_requested_power(), why should we ever bother
> to traverse the list of dependent_cpus and not related_cpus only ?
> 
> Otherwise the same CPU will have its load contributed to the power of
> 8 cooling devices.
> 

Good question.

How about a different interface for those cpufreq drivers?
That new registration API would allow to specify the cpumask.
Or rely on EM cpumask: em_span_cpus(em)

Currently we have two ways to register cooling device:
1. when the cpufreq driver set a flag CPUFREQ_IS_COOLING_DEV, the core
will register cooling device
2. cpufreq driver can explicitly call the registration function:
cpufreq_cooling_register() with 'policy' as argument

That would need substantial change to the cpufreq cooling code, from
policy oriented to custom driver's cpumask (like EM registration).

Regards,
Lukasz