RE: [Xen-users] Re: Status of frequency scaling

["Petersson, Mats" <Mats.Petersson@xxxxxxx>]

> -----Original Message-----
> From: xen-users-bounces@xxxxxxxxxxxxxxxxxxx 
> [mailto:xen-users-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of 
> Goswin von Brederlow
> Sent: 02 May 2007 09:09
> To: xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-users] Re: Status of frequency scaling
> 
> Stephan Seitz <nur-ab-sal@xxxxxx> writes:
> 
> > On Mon, Apr 30, 2007 at 11:47:03AM +0200, Petersson, Mats wrote:
> >> Well, the current official status is "it doesn't work at all".
> >
> > Hm, okay.
> >
> >>There are several problems, some of which are relatively 
> easy to fix and
> >>some that aren't:
> >>2. There is no power-management policy for the entire 
> system, so dom0
> >>usage will determine the entire system's performance. This 
> doesn't work
> >> very well, as Dom0 isn't supposed to be heavily loaded, DomU's are.
> >
> > Understandable. In my case, Dom0 would be by Desktop system 
> while DomU
> > would be the firewall (for now a different host), so I 
> could live with
> > this restriction.
> 
> It shouldn't be too hard to write a userspace governor that check the
> cpu usage of all domains (like xentop does) and then scales the
> frequency to match.

Absolutely right. It's just that no one has deemed it important enough to 
actually sit down and do it. 

> 
> I have xen on my laptop and the batery live goes down drastically
> without frequency scaling (understandably). I would like to have at
> least manual control so I can turn the CPU down when on batery.
> 
> >>3. Guest may need to be informed of the frequency change. 
> Some kernels
> >>may not appreciate that the CPU frequency is chaning "without it's
> >>knowledge". [This is probably not a big problem on modern 
> OS's, but some
> >> older ones may be very un-co-operative on this account].
> >
> > Ah yes, I remember, that a Solaris system within VMware server
> > didnÃâât like the frequency changes of the Linux host.
> > I donÃâât have any problems with XP within VMware server at 
> my notebook.
> 
> Does Linux care?

Depends a little bit on what your doing inside linux, but in general you can 
run Linux with "unstable" clocks. I did that when I first wrote a PowerNow! 
driver for Linux (that driver, after quite a lot of modification, is now the 
AMD official driver, but not much of my code is there now 4 or so years 
later...). 
> 
> >>> I have an AMD64. powerno-k8 works find with non-XEN kernels, but
> >>> with XEN-kernels I get the error:
> >>>
> >>> powernow-k8: Found 1 AMD Athlon(tm) 64 Processor 3400+ processors
> >>> (version 2.00.00)
> >>> powernow-k8: BIOS error - no PSB or ACPI _PSS objects
> >>This is strange - as far as I can remember, I had powernow 
> working on my
> >>machine not so long ago (3.0.4 or so) with only the traps.c 
> patched to
> >> allow writing of the FID/VID controls registers.
> >
> > Could you send me the patch for traps.c? I would like to try it.
> 
> Me too.

Patch attached for anyone to see... It's a bit old, but I believe it still 
applies to traps.c. 

Please be aware that this code is "untested" and without any warranty of 
actually functioning correctly or doing what you want in any other way...

--
Mats
> 
> MfG
>         Goswin
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
> 
> 
>

Attachment: patch.msr_powernow
Description: patch.msr_powernow

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users