2013-05-28 15:02:19

by Jörg Otte

[permalink] [raw]
Subject: [v3.10-rc] intel_pstate question

Hi,

the sysfs-interface of intel_pstate is different to that of the acpi driver:
- intel_pstate shows current cpu-frequency in "cpuinfo_cur_freq" whereas
the acpi counterpart uses "scaling_cur_freq".
- moreover cpuinfo_cur_freq is readable by root only, whereas
scaling_cur_freq is world readable.

A monitoring tool must take care of theese differences.
Is this intentional?

Thanks, Jörg

Please CC me in answeres, I'm not subscribed.


2013-05-28 15:52:51

by Dirk Brandewie

[permalink] [raw]
Subject: Re: [v3.10-rc] intel_pstate question

Hi Jörg,

On 05/28/2013 08:02 AM, Jörg Otte wrote:
> Hi,
>
> the sysfs-interface of intel_pstate is different to that of the acpi driver:
> - intel_pstate shows current cpu-frequency in "cpuinfo_cur_freq" whereas
> the acpi counterpart uses "scaling_cur_freq".
> - moreover cpuinfo_cur_freq is readable by root only, whereas
> scaling_cur_freq is world readable.
>
> A monitoring tool must take care of theese differences.
> Is this intentional?
>

This is a side effect of the interface that intel_pstate presents to the cpufreq
core.

cpuinfo_cur_freq reports the value returned from the scaling driver.

scaling_cur_freq reports the current frequency in the governors policy structure.

intel_pstate has an internal governor so the core does not try to interpret
the policy structure.

I don't know why cpuinfo_cur_info is read-only root maybe Rafael knows the history.


> Thanks, Jörg
>
> Please CC me in answeres, I'm not subscribed.
>

2013-05-29 09:25:13

by Jörg Otte

[permalink] [raw]
Subject: Re: [v3.10-rc] intel_pstate question

2013/5/28 Dirk Brandewie <[email protected]>:
> Hi Jörg,
>
>
> On 05/28/2013 08:02 AM, Jörg Otte wrote:
>>
>> Hi,
>>
>> the sysfs-interface of intel_pstate is different to that of the acpi
>> driver:
>> - intel_pstate shows current cpu-frequency in "cpuinfo_cur_freq" whereas
>> the acpi counterpart uses "scaling_cur_freq".
>> - moreover cpuinfo_cur_freq is readable by root only, whereas
>> scaling_cur_freq is world readable.
>>
>> A monitoring tool must take care of theese differences.
>> Is this intentional?
>>
>
> This is a side effect of the interface that intel_pstate presents to the
> cpufreq
> core.
>
> cpuinfo_cur_freq reports the value returned from the scaling driver.
>
> scaling_cur_freq reports the current frequency in the governors policy
> structure.
>
> intel_pstate has an internal governor so the core does not try to interpret
> the policy structure.
>
> I don't know why cpuinfo_cur_info is read-only root maybe Rafael knows the
> history.
>
Thank you Dirk, for the information!

Just another question: Now that processor 3a is supported, I am giving
pstate driver a try on my notebook. This is the first time I see frequencies
between above 2.5 and up to 3.1 GHz. With cpufreq driver I never saw that
frequencies.
So I think pstate driver uses turbo mode if appropriate (in contrast to
cpufreq). is this correct?

Thanks, Jörg