Vinay Devadas wrote:

>> P0 is not equal to turbo mode. P0 means the system is requesting to
>> the speed of the max performance. When Turbo mode is enabled, P0 give
>> the system a chance to go into turbo mode, but whether the processor
>> enter turbo mode or not is determined by the hardware. So, even if
>> APERF/MPERF may be < 1, if DTrace says we are in P0, we should let
>> PowerTOP and users know the system is requesting max performance
>> speed, not max non turbo speed.
> 
> I agree with what you say above, but I am sorry I need more
> clarification on the display issue.
> 
> Now say max non-turbo frequency is 2400 and P0 is 2401 as
> displayed by powertop. PowerTop does not diplay P0, P1 ...
> etc. It only displays frequencies 2000, 2400, 2401 etc. i.e.
> the display is something like,
> P-state Frequency
> 2000
> 2400
> 2401
> 
> Now considering your case, let us say IDA is enabled and
> Dtrace says we are in P0. CURRENTLY powertop display 100% in
> 2401. However, sampling the APERF/MPERF values, if the ratio
> <=1 we are running at 2400 MHz (*I agree in P0, but it is 2400
> and NOT 2401*). If the ratio is larger than one, then we are
> running at some frequency higher than 2400Hz.
> 
> This is precisely what I do. IDA enabled and APERF/MPERF <=1,
> I display 100% next to 2400Mhz (This means we are in P0 and
> running at 2400 Mhz). If APERF/MPERF > 1, I display AVERAGE
> frequency instead of *2401 Mhz* and the *time spent there*.
> Just because IDA is enabled will it not be incorrect to
> display 100% next to 2401 just because Dtrace probe tells we
> are in P0 as powertop does currently ?
> 

The frequencies is designated as P-state level.
If powertop display 100% at 2400Mhz, how could the users know it's
in P0 or P1? And When APERF/MPERF < 1 in P0, the actual speed
may < 2400 due to T-state or hardware temperature limitation. why put
it in 2400 statistics? The frequencies here are P0,P1, ... Pn.
I understand you want to show the actual speed, but doing it just in P0
may make more sense. 

Another thing comes in my mind is, if the acutal speed 2400< P0 < 2401,
let's display 2401 as well. An increment < 1Mhz is meaningless. Let's
set
the resolution of P-state frequency to 1Mhz.

Thanks,
-Aubrey


Reply via email to