Hi Mark,

I am Vinay, working on the Turbo Mode RFE on PowerTop.

So this means that, when a *Turbo Mode supported* processor runs at P0, then it 
may actually be running at some frequency *greater than or equal to its P1 
frequency (2400 in the example)*. So using APERF/MPERF will help us determine 
the average frequency in P0 for such conditions. Is that correct ?

Thanks,
Vinay  

----- Original Message -----
From: Mark Haywood <[email protected]>
Date: Wednesday, July 23, 2008 12:21 pm
Subject: Re: [tesla-dev] Question on PowerTop V 1.1
To: Eric Saxe <Eric.Saxe at Sun.COM>
Cc: Vinay Devadas <Vinay.Devadas at Sun.COM>, tesla-dev at opensolaris.org


> Eric Saxe wrote:
> > Mark Haywood wrote:
> >>
> >> No. When it runs at P0, then the minimum frequency it will operate 
> at 
> >> is 2400MHz. But it is very likely to run faster than that and yes, 
> I 
> >> believe it may even run faster than 2401MHz. So, the "average 
> >> frequency" as far as I know is the average speed while in P0. I 
> >> believe that the APERF/MPERF registers will be used to determine 
> the 
> >> average frequency. But I'm not working on that, so maybe I should 
> let 
> >> someone else answer that. ;-)
> >>   
> > Is it the case here that there's essentially two "P0" states being 
> > enumerated by ACPI (One for turbo mode enabled, and one for disabled)?
> 
> It depends what you mean by P0 states. By definition, P0 is the first 
> 
> P-state returned when evaluating the _PSS. The P0 frequency for Turbo 
> 
> Mode enabled systems is always supposed to be defined as the P1 
> frequency + 1MHz.
> 
> If you were to disable Turbo Mode (assuming the BIOS gives you the 
> option), then I believe you would find your supported_frequencies_Hz 
> to be:
> 
> 800000000:1200000000:1600000000:2000000000:2400000000
> 
> Why did Intel define the Turbo Mode P-state P0 this way you ask? I 
> suspect that they did it this way so that an OS could basically 
> support 
> Turbo Mode without any modifications. An OS that is totally ignorant 
> of 
> Turbo Mode will still request Turbo Mode (by requesting P0) when the 
> CPUs are heavily utilized at P1 (not yet over clocked). That means 
> existing Windows, Linux and other OSes that have been in the can long 
> 
> before Turbo Mode existed can still take advantage of it.
> 
> > This is what the cpu_info kstat is showing on Kuriakose's laptop 
> > (thanks Vinay):
> >
> > supported_frequencies_Hz
> > 800000000:1200000000:1600000000:2000000000:2400000000:2401000000
> >
> > I'm wondering what the essential difference is between the 2400 
> state 
> > and the 2401 state (besides 1 MHz, and i'm guessing that's not even 
> 
> > the difference)...
> 
> The 2401 state says feel free to run 2400 or faster if conditions allow.
> 
> Mark
> 
> >
> > Thanks,
> > -Eric
> 
> _______________________________________________
> tesla-dev mailing list
> tesla-dev at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/tesla-dev

Reply via email to