Re: [PATCH v2 1/4] PM / EM: add devices to Energy Model

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

On 2/13/20 10:59 AM, Dietmar Eggemann wrote:
> On 06/02/2020 14:46, lukasz.luba@arm.com wrote:
>> From: Lukasz Luba <lukasz.luba@arm.com>
> 
> [..]
> 
>> @@ -26,7 +28,7 @@ framework, and interested clients reading the data from it::
> 
> s/::/: ?
> 
>>          | Thermal (IPA) |  | Scheduler (EAS) |  |     Other     |
>>          +---------------+  +-----------------+  +---------------+
>>                  |                   | em_pd_energy()    |
>> -               |                   | em_cpu_get()      |
>> +               |  em_get_pd()      | em_cpu_get()      |
>>                  +---------+         |         +---------+
> 
> em_get_pd() and em_cpu_get()? Why not em_pd_get()? em_cpu_get() is a
> specific em_get_pd(). right?

Yes. I will rename 'em_get_pd' to 'em_pd_get'

> 
> [...]
> 
>> @@ -85,13 +89,20 @@ API.
>>   2.3 Accessing performance domains
>>   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>   
>> +There is two API functions which provide the access to the energy model:
>> +em_cpu_get() which takes CPU id as an argument and em_get_pd() with device
>> +pointer as an argument. It depends on the subsystem which interface it is
>> +going to use, but in case of CPU devices both functions return the same
>> +performance domain.
> 
> There is probably a reason why we need this specific function for CPU
> devices? The reason should be described. People might ask why
> em_get_pd() is not sufficient.

True, good point. I will extend the comment in em_cpu_get().

> 
> [...]
> 
>> - * A "performance domain" represents a group of CPUs whose performance is
>> - * scaled together. All CPUs of a performance domain must have the same
>> - * micro-architecture. Performance domains often have a 1-to-1 mapping with
>> - * CPUFreq policies.
>> + * In case of CPU device, a "performance domain" represents a group of CPUs
>> + * whose performance is scaled together. All CPUs of a performance domain
>> + * must have the same micro-architecture. Performance domains often have
>> + * a 1-to-1 mapping with CPUFreq policies.
>> + * In case of other devices the 'priv' field is unused.
>>    */
>>   struct em_perf_domain {
>> -	struct em_cap_state *table;
>> -	int nr_cap_states;
>> -	unsigned long cpus[0];
>> +	struct em_perf_state *table;
>> +	int nr_perf_states;
>> +	void *priv;
> 
> In case you go back to the variable length field plus type field to
> distingush EM devices, keep cpus[0] as the name.

OK, I will.

> 
> [..]
> 
>>   /**
>> - * em_pd_energy() - Estimates the energy consumed by the CPUs of a perf. domain
>> + * em_pd_energy() - Estimates the energy consumed by the CPUs of a perf.
>> +			domain
> 
> Why this change?

hmmm, that's odd, maybe there was 'device' then I changed it back to
'CPUs' but forgot to move the 'domain' to the old place.

> 
> [...]
> 
>> @@ -141,12 +210,12 @@ static struct em_perf_domain *em_create_pd(cpumask_t *span, int nr_states,
>>   		 */
>>   		opp_eff = freq / power;
>>   		if (opp_eff >= prev_opp_eff)
>> -			pr_warn("pd%d: hertz/watts ratio non-monotonically decreasing: em_cap_state %d >= em_cap_state%d\n",
>> -					cpu, i, i - 1);
>> +			dev_warn(dev, "energy_model: hertz/watts ratio non-monotonically decreasing: em_perf_state %d >= em_perf_state%d\n",
> 
> s/energy_model/EM ?

OK, I will rename them in all places.

Thank you for the review.

Regards,
Lukasz