Xref: utzoo comp.unix.i386:7829 comp.unix.questions:24705
Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!silma!cichlid!aab
From: aab@cichlid.com (Andrew A. Burgess)
Newsgroups: comp.unix.i386,comp.unix.questions
Subject: Re: need help with FATAL error in unix 3.2u
Keywords: parity error, NMI, floppy access, gateway2000
Message-ID: <1990Aug15.193416.8426@cichlid.com>
Date: 15 Aug 90 19:34:16 GMT
References: <601@bigbroth.UUCP>
Organization: Consulting Software Engineer, Felton CA
Lines: 45

In article <601@bigbroth.UUCP> rk@bigbroth.UUCP (rohan kelley) writes:
>Problems with unix 3.2u BellTech (Interactive) installation in
>Gateway2000-25 cache system.
>
>Error message:
>FATAL:Parity error on the motherboard
>PANIC:Parity error address unknown
>     Trying to dump 1024 pages (etc)
...
>Crash occurs when accessing the floppy drive (either 0 or 1) but only
>at intermittent times. Commands current have been cpio and format.  If
...

I had a similar problem with an old AMI motherboard once. I first noticed that
it would drop about one byte per million when reading from the tape drive. 
Repeating the read was successful. Note that there were no error messages
from the tape drive -- it thought the tape had read correctly. I only became
aware of a problem when doing a compare of disk to tape after a backup.

Then I noticed that reading floppys had a similar problem. The only common
denominator I could think of was that both subsystems used the motherboard
DMA controller to transfer data. So I created a little test under DOS.
I made a 500Kb file of random data and put two copies on a 1.2MB floppy.
I then ran the DOS comp (file compare) program endlessly. The files would
miscompare in a few minutes. Eventually the program would crash. My guess was
that this one bye in a million was not just vanishing but instead being
written to a 'random' location.

This could be your problem. If so then a replacement motherboard would
solve it (assuming you have a marginal component somewhere rather than
a bad motherboard design).

If the dealer is willing to swap and maybe give you another week or so to
test, you might get lucky. Then again this is a WILD GUESS!

You might also try writing a test program like mine.

Good luck
Andy

-- 
Andy Burgess
Independent Consultant
uunet!silma!cichlid!aab
aab@cichlid.com