cpuidle readouts

[Ken O'Brien]

Hi,

I'm trying to find out how much power is consumed for each cpu state.
I'm getting the following output for cpuidle's "power" attribute. I
thought this was supposed to list how much power in milliwatts is
consumed in each state.

=====
***@mjolnir:/sys/devices/system/cpu/cpu0/cpuidle$ cat state*/power
4294967295
4294967294
4294967293
4294967292
=====

My guess is that these are mock values until proper values are
implemented. Can anyone shed some light on this? i.e., the current
status of this implementation or perhaps improper configuration on my part?

Regards,

Ken

[Arjan van de Ven]

Post by Ken O'Brien
Hi,
I'm trying to find out how much power is consumed for each cpu state.
I'm getting the following output for cpuidle's "power" attribute. I
thought this was supposed to list how much power in milliwatts is
consumed in each state.
=====
4294967295
4294967294
4294967293
4294967292
=====
My guess is that these are mock values until proper values are
implemented. Can anyone shed some light on this? i.e., the current
status of this implementation or perhaps improper configuration on my part?

frankly, I don't think this really make sense...
I wouldn't know what it means if a logical cpu core (half of a
hyperthreading pair or half of an AMD pair) is in a state and the other
half is not. The core is also a very small %age of power of the whole
CPU chip (never mind SOCs)

[k***@public.gmane.org]

Ok, well is there a way to measure total CPU power consumption? Maybe by
reading sysfs?

Post by Arjan van de Ven

Post by Ken O'Brien
Hi,
I'm trying to find out how much power is consumed for each cpu state.
I'm getting the following output for cpuidle's "power" attribute. I
thought this was supposed to list how much power in milliwatts is
consumed in each state.
=====
4294967295
4294967294
4294967293
4294967292
=====
My guess is that these are mock values until proper values are
implemented. Can anyone shed some light on this? i.e., the current
status of this implementation or perhaps improper configuration on my

part?
frankly, I don't think this really make sense...
I wouldn't know what it means if a logical cpu core (half of a
hyperthreading pair or half of an AMD pair) is in a state and the other
half is not. The core is also a very small %age of power of the whole
CPU chip (never mind SOCs)

[Sundar]

Post by k***@public.gmane.org
Ok, well is there a way to measure total CPU power consumption? Maybe by
reading sysfs?

Hi,

You will need some extensive characterization (and measurement) of the
core power w.r.t the CPU frequency and the C-state residencies.

You could create a table of the power consumption based on the core
frequency; and then individual residencies at that frequency; you
should discount all the state beyond the C-state for the CPU when
clock gating comes into effect, if your core supports clock gating at
all. Say for example, running at the lowest frequency constrained to a
100% CPU utilization might be the number for starters.

Cheers!

[Brown, Len]

We put these numbers in sysfs because ACPI had them.
Unfortunately, the ACPI numbers are not useful.
Subsequently, cpuidle changes what these numbers means.
They are an instantaneous relative ranking, not a real power number.
We should remove them from sysfs.

thanks,
-Len

-----Original Message-----
Sent: Sunday, November 20, 2011 9:11 PM
Subject: Re: [Discuss] cpuidle readouts

Post by k***@public.gmane.org
Ok, well is there a way to measure total CPU power consumption? Maybe by
reading sysfs?

Hi,
You will need some extensive characterization (and measurement) of the
core power w.r.t the CPU frequency and the C-state residencies.
You could create a table of the power consumption based on the core
frequency; and then individual residencies at that frequency; you
should discount all the state beyond the C-state for the CPU when
clock gating comes into effect, if your core supports clock gating at
all. Say for example, running at the lowest frequency constrained to a
100% CPU utilization might be the number for starters.
Cheers!
--
---------
The views expressed in this email are personal and do not necessarily
echo my employers.
_______________________________________________
Discuss mailing list
http://lists.lesswatts.org/listinfo/discuss