Re: IBM 6x86MX233 blacklisted ?

Kurt Garloff (garloff@kg1.ping.de)
Mon, 2 Feb 1998 21:02:41 +0100 (CET)


On Mon, 2 Feb 1998, Steffen Grunewald wrote:

> Hi list,
>
> recently I got an offer for an IBM6x86MX233 based computer. Looks
> pretty good (even a bit faster than iPII-233 due to the larger cache)...
> Are there any arguments against buying such a beast ?
> Is there a FAQ on Linux' CPU support ? Perhaps a comparison of
> performance and throughput ?
> Are there (besides the ones that appeared in this mailing list past
> weekend) any blacklisted single-processor main boards ?

Linux supports every i386 model and up. However, in Linux-2.0.xx, there is
no special code to recognize the 6x86 processors. So, the 6x86 is reported
as a 486, whereas the 6x86MX is reported as a Pentium or PentiumPro (the
difference is, that the 6x86MX has CPUID enabled by default.) The report
as 486 or 686 has no problems, as far as I know.

The 6x86 (and 6x86MX) have some special configuration registers which can
be used to optimize performance such as the Wt_Alloc (Allocate Cache lines
on Write). If you want to use these, you need a kernel patch or a program
to set up the config registers.

There are several programs and patches floating around. You can find my
kernel patch against 2.0.33 (a pretty fat one) on
ftp://student.physik.uni-dortmund.de/pub/linux/kernel/patches/cyrix
and some programs on
ftp://student.physik.uni-dortmund.de/pub/linux/kernel/6x86/
(cpudev, set-6x86)
I call the patch fat, because it does a lot of the setup during the kernel
initialization. Most of it could be done in userspace (root) by programs.
This alternative was followed by Andrew Balsa (and may be included into
2.1.xx). He got a small kernel patch (mainly 6x86 recognition) and lets
the program do the setup. Have a look at it's site:
http://www.tux.org/~balsa/linux/cyrix
You can find the set6x86-1.5 there. It includes the small kernel patch and
the set6x86 program.

If you got a system, where stability is important and you fear bad users,
you have to apply one of the patches or use the set6x86 program to work
around a processor bug, which is know as Hidden-CLI or Coma bug. Just like
the Pentium F00FC7C8 bug, it let's any (non-privileged!) user be able to
crash the system. (The Pentium bugfix made it into the 2.0.32 kernel
however.)

--

I got good experience with boards from FIC. I got a PA-2010+, which uses the VIA VPX chipset. It supports 75 and 83 MHz with async PCI bus, the 6x86(MX) linear burst feature, up to 512MB cacheable area and EDO and SDRam. If you'd like to try 2.5x83.3MHz (208) instead of 3x66MHz (200), this would be the right joyce. (Or did you get a 2.5x75MHz (188) model?

--

I doubt, that the 6x86MX-233 is faster than a iPII-233. It is possibly faster than a iP233MMX as long as you don't start doing floating point calculations or using the MMX features, but the iPII-233 is superior.

Kurt Garloff, Dortmund <K.Garloff@ping.de> PGP key on http://www.student.physik.uni-dortmund.de/homepages/garloff