RE: Processor C states and USB problem

[Nate Lawson <nate-Y6VGUYTwhu0@public.gmane.org>]

On Sun, 18 Jan 2004, Yu, Luming wrote:
> > Yes, but compared to the usage count of C2 that is extremely low, and
> > the CPU is idling all the time (CPU usage is below 1%). I don't know
> > much about C states but I guess it should go into C3 much more often,
> > and the bus mastering activity should show something else
> > than ffffffff
> > when booting with a bare bones kernel and with no extra hw initialized
> > (USB, FireWire, etc.)
>
> I agree that it's unreasonable that C3 usage count is never above 5.
> This is the result of the C state power policy.
>
> Threshold is:
> pr->power.states[ACPI_STATE_C2].promotion.threshold.bm = 0x0F;
> pr->power.states[ACPI_STATE_C3].demotion.threshold.bm = 0x0F;
>
> The condition of promotion or demotion is:
> if (pr->power.bm_activity & cx->demotion.threshold.bm) {
> if (!(pr->power.bm_activity & cx->promotion.threshold.bm)) {
>
> Using the following statement to adjust bm_activity of the last Nth
> schedule quanta.
>
> pr->power.bm_activity <<= 1;
> if(bm_status)
> 	pr->power.bm_activity++;
>
> So it's is not strange that C3 usage count is 4  !
>
> Maybe we need a more sophisticated policy .

In FreeBSD, I clear BM_STS before exiting idle.  If I come back and it's
set, then avoid C3 and demote to C2.  If not and last sleep was > C3
latency, then use C3.  It's unnecessary to count the number of BM_STS
events.

-Nate


-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn