Re: Cyrix cooling (Re: RH5.2 won't install on Cyrix)

[Philipp Rumpf <prumpf@jcsbs.lanobis.de>]

> Wouldn´t this be a nice feature of Linux?
> If a server-CPU gets "bored" for more than a time why not setting the
> CPU to slower cycle via the APM?

Careful. APM is an old standard, it relies on the OS calling the BIOS in
privileged mode and there are a lot of broken implementations around (fur-
theremore the current way to do it has races). There are two better ways:

1) ACPI. I do not know much about ACPI, but the spec's there and there are
(were ?) some people writing a driver for it.

2) direct device drivers for the chipset. Configuring the most common chipsets
isn't hard at all (you can do it out of userspace using setpci) and you can
do all the stuff (doing the power management in the load level calculation so
it never gets beyond 1.0 (as long as you don't _need_ your CPU, that is)).


The current concept for PMing (x86) CPU's is to assert STPCLK for some time, then
releasing it again. This time is (1-256) x 32 usec, so it might hurt interrupt
latency if you set it too high. Furthermore, some CPUs have a low-power mode that
can be enabled in an MSR.

> Is there a way to do this today?
> This may increase the livetime of a CPU and maybe the CPU-powersupply
> on the mainboard.

Well, one really nice point is that (at least for the Pentium MMX 200 I have
here) you can switch off the CPU fan when you slow it down to 2-3%. Still
enough for some basic services and quiet).
-
Linux-future: thinking about the future of the Linux kernel
Archive:      http://humbolt.nl.linux.org/lists/
Wish list:    http://users.ox.ac.uk/~mert0236/linux-future.html