Subject: Re: MMU faults while trying to install 1.5
To: John Vrolijk <email@example.com>
From: Ignatios Souvatzis <firstname.lastname@example.org>
Date: 12/20/2000 20:32:07
On Wed, Dec 20, 2000 at 07:54:45AM +0100, John Vrolijk wrote:
> A4K in tower. 16MB onboard RAM
> Apollo 4060: 50Mhz '060, 64MB RAM (2*32MB)
> I tried netbsd.INSTALL and netbsd.generic.
> Could it be that I have some faulty hardware ?
Apollo hardware have a bad reputation. I know of two people who had one,
one with success, one without (who was later led to believe, by independent
evidence, that NetBSD was not at fault).
> Is there someway I can test the MMU on the 060 ?
Oh, the MMU will be fine. They have some wacky timing, I think, so that
depending on weather or phase of moon, and bus load (number and specific type
of memory chips etc.) it works or not.
NetBSD will notice this a bit earlier than AmigaOS, as NetBSD _depends_
on cache flushes (== fast line transfers) back-to-back with DMA transfers
and MMU operations.
When the data is misread/miswritten during this process, you'll see, e.g,
wrong program pointers or wrong instructions, leading to later misbehaviour,
e.g. random "segmentation faults".
All I can advise you is to maybe set jumpers (if there are any) to make memory
access slower, get faster RAM, play with the -m, -n and -p switches to
slow down memory access, etc... Some of this might work.
The other victim got a semi-working NetBSD kernel by changing the code to
never use the data cache... but this assumes you have a machine to compile
the code on, and I don't think you'd like the performance.