Re: Compal EFL30 ACPI

[Bruno Ducrot <ducrot-kk6yZipjEM5g9hUCZPvPmw@public.gmane.org>]

On Wed, Sep 07, 2005 at 11:23:58AM +0200, Erik Slagter wrote:
> On Wed, 2005-09-07 at 11:18 +0200, Bruno Ducrot wrote:
> > On Wed, Sep 07, 2005 at 10:33:07AM +0200, Erik Slagter wrote:
> > > > Right. That's probably a sign that you do not have the 2.6.xx cpufreq
> > > > CPU speed adjustment drivers present in that SuSE kernel of yours. If
> > > > you enable AMD PowerNow / Intel SpeedStep cpufreq drivers, this problem
> > > > will go away. Then, if you want to control the speed, download
> > > > cpufreqd, which will let you set up everything just exactly how you
> > > > want it using cpufreqd.conf.
> > > 
> > > I was under the impression that the "acpi" cpufreq driver was
> > > preferable.
> > 
> > You were wrong.
> 
> Can please someone elaborate?
> 
> I remember to have read that it's best to use the acpi method for
> frequency scaling, unless that's not available?! Also I suppose acpi
> knows best how to drive it's own hardware?

The cpufreq ACPI IO driver use some ioports (as its name suggest) in
order to perform p-state transition.  Those ioports likely trigger
an SMM trap which means that some propritary BIOS code will be run.
It's fine to do so, but the problem is that there will be more latencies
just to put the processor under SMM.  It's also fine to do so on older
platforms, since anyway the latency introduced by putting the processor
in SMM will be low compared to the time taken for a transition.  But
this will become too high on, say, centrino platforms.

Cheers,

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf