Re: Bug#110273: ITP: lvcool -- "cools" the AMD K7 (Athlon/Duron) CPU
On Tue, Aug 28, 2001 at 05:07:09PM +0200, Eduard Bloch wrote:
> #include <hallo.h>
> Andreas Bombe wrote on Tue Aug 28, 2001 um 04:42:53PM:
>
> > > This tool is not necessary; ACPI support in the kernel will do it
> > > automatically.
> >
> > No, it won't. The disconnect is disabled by most BIOSes for various
> > reasons[1] and ACPI essentially just executes BIOS provided bytecode.
>
> Exactly.
Yep, ACPI will do STPGNT by itself.
All lvcool adds is setting 1 (in words: one) bit in a hardware register
once. Do we really have to add another Debian package to the thousands
just for setting one bit?
> > However the general idea of lvcool is bullshit. Linux doesn't need
>
> Maybe. But it works.
>
> > stupid idle loop programs, it's open source for hell's sake! All we
>
> Okay, then go and make it better. Currently, I think lvcool is most
> efficient then kernel-based solution, kapm-idled. LVcool takes all the
> remaining CPU time, but kapm-idled seems to be interupted very often, so
> the real idle time where HLT-commands or the ACPI-modes can be entered
> is not as large as using an active user-space program.
I'm not sure whether APM handles the STPGNT mode. I guess it depends on
the BIOS again.
Now to your comparison kapm-idled/lvcool timeslices. The APM code uses
insight into the scheduling situation to idle only when no processes are
ready to run. This may cause it not to idle constantly with certain
program run patterns. If you want it to idle more often, just remove
the checks.
In contrast, all lvcool does (and can do) is set its nice value to the
highest possible (20). It will get normally scheduled by the kernel and
run its idle loop even under 100% CPU load, taking away cycles from your
kernel compile/quake/<insert favourite CPU hog>.
And it would only idle one CPU for SMP systems. lvcool is just a gross,
ugly kludge.
--
Andreas E. Bombe <andreas.bombe@munich.netsurf.de> DSA key 0x04880A44
Reply to: