more on VIA 686B (trials)

more on VIA 686B (trials)

[David Monniaux]

Due to a catastrophic fan short-circuit, I was forced to exchange my
686A-based motherboard for a 686B. Bad idea!

The 686A MB (MSI-6330 aka K7T-Pro) worked perfectly well: no crashes,
UDMA 66. It accepted Athlon-optimized kernels.

The 686B MB (K7T-Lite) crashed if used with DMA (any kind - mdma0 to UDMA100),
whatever version of the "VIA fixes" was in place.
Furthermore, upgrading to BIOS 2.7 (instead of 2.5), including a so-called
"SB Live fix", made the system permanently unstable.
On top of that, running an Athlon-optimized kernel (whether or not it
was compile with =march=i686 (egcs-1.1.2) or -march=athlon (gcc 2.96)
immediately oops and crash during /etc/rc!

I replaced this mobo+Duron with an ASUS A7V133+Athlon, which
work perfectly well.
Athlon-optimized kernel, UDMA100, no problem whatsoever.

So we have two kinds of problems:
- *certain* 686B motherboards crash if used with an Athlon kernel
  (and it does not depend on the compiler options, rather on hand-made
  Athlon optimizations)
- *certain* 686B motherboards will crash if used with any kind of DMA
  under heavy disk access
- some 686B motherboards have absolutely no problem.

Crash test:
for i in `seq 1 30` ; do echo $i; tar xfz X410src-1.tgz ; rm -rf xc; done

All the above is valid for kernel version 2.4.2-RedHat to 2.4.5ac9...

Another lesson: the MSI K7T-Lite is absolute crap. The manual sucks,
and the BIOS upgrades supposed to make the machine stabler actually
make it randomly hiccup (EVEN under Windows, which is what I suppose
those mobos are supposed to run).

-- 
David Monniaux            http://www.di.ens.fr/~monniaux
Laboratoire d'informatique de l'École Normale Supérieure,
Paris, France

Re: more on VIA 686B (trials)

[Alan Cox]

> Abit KT7A, kernel oops right after boot... :(  Can be solved to turning off 
> 'Enhance Chip Performance' in the BIOS, but then our chip performance is 
> un'Enhance'd, and we can't have that!  So back to the K6 kernel.

And praying it doesnt go wrong on you - has it not occurred to you that the
extremely high throughput copies that the mmx copy we use causes will
occasionally happen by chance and get you anyway ?

Alan

Re: more on VIA 686B (trials)

[Cory Watson]

On Thursday 14 June 2001 12:44 pm, David Monniaux wrote:
> So we have two kinds of problems:
> - *certain* 686B motherboards crash if used with an Athlon kernel
>   (and it does not depend on the compiler options, rather on hand-made
>   Athlon optimizations)

Abit KT7A, kernel oops right after boot... :(  Can be solved to turning off 
'Enhance Chip Performance' in the BIOS, but then our chip performance is 
un'Enhance'd, and we can't have that!  So back to the K6 kernel.

-- 
Cory 'G' Watson
   Dad are you vicariously living through me in the hope that my
accomplishments will validate your mediocre life and in some way compensate 
for
all the opportunities you botched ?	  -- Calvin

Re: more on VIA 686B (trials)

[Rachel Greenham]

David Monniaux wrote:

>I replaced this mobo+Duron with an ASUS A7V133+Athlon, which
>work perfectly well.
>Athlon-optimized kernel, UDMA100, no problem whatsoever.
>
Which is odd, because that's exactly my combination (ASUS A7V133 + 
Athlon), and I get crashes with DMA on anything from 2.4.3-ac7 onwards, 
but up to 2.4.3-ac6 is rock-steady. (my crash test is "bonnie -s 1024" 
:-)) I wonder what's different between our machines (apart from distro, 
which I wouldn't expect to be relevant)? Clock speed? We tried 
downclocking my Athlon to 1.0 GHz but it made no difference.

I've been tinkering (have no kernel programming experience) with 
selectively forward-porting the 2.4.3-ac6 code to newer kernels for my 
own use at least, but haven't got it right yet.

-- 
Rachel

Re: more on VIA 686B (trials)

[Cory Watson]

On Thursday 14 June 2001 01:10 pm, Alan Cox wrote:
> And praying it doesnt go wrong on you - has it not occurred to you that the
> extremely high throughput copies that the mmx copy we use causes will
> occasionally happen by chance and get you anyway ?

Yeah, it's occurred to me, but it's yet to happen.  I fought with the thing 
for 48 hours, and after settling on a K6 kernel w/'Enhanced' enabled, I've 
not had a single problem.  I suppose it's possible that I've not seen the 
problem, but I've put this box through it's paces many times over since I 
settled.  I gave it another go with 2.4.6-pre3 this morning, hoping something 
had snuck in ;)

-- 
Cory 'G' Watson
   You know what we need, Hobbes? We need an attitude. Yeah, you can't be cool
if you don't have an attitude.	  -- Calvin