Subject: Re: Pyxis chip on a 500a.
To: Andrew Gallatin <gallatin@cs.duke.edu>
From: Wilko Bulte <wilko@yedi.iaf.nl>
List: port-alpha
Date: 12/30/1999 13:49:00
On Wed, Dec 29, 1999 at 01:34:39PM -0500, Andrew Gallatin wrote:
> 
> Jeff Roberson writes:
>  > I booted a freebsd, and a linux kernel on the machine last night.  Both,
>  > to my surprise, worked.  Although freebsd doesn't support ide on the
> 
> FreeBSD 4.0-current supports IDE on alphas.  Make sure you have the
> ata device in your kernel config file:
> 
> As to why FreeBSD works and NetBSD doesn't, I'm not sure.  I used to
> have some machines with buggy pyxis revs, and I remember that bursty
> writes to dense space pci memory may also trigger a lockup.  I don't
> think there is anything inherently broken in NetBSD, it may just be
> that FreeBSD & linux draw the screen differently & don't trigger the
> hardware bug.

The bug description in the Pyxis hardware manual reads:

"DMA PAGE BOUNDARY SOLUTION"
...
PCI DMA reads that attempt to cross 8K page boundaries cause data
corruption problems. A fix has been implementes <bla... describes
an extensive hardware fix using some programmable logic etc>
...

The latest revs of the MX5-based Miatas have a hardware fix included.
The later GL-based Miatas use a newer rev Pyxis that does not have the
bug.

To be honest I don't see any reason for this bug to trigger during
kernel startup/device probing. No PCI DMA going on (right?)

Wilko
-- 
Wilko Bulte 		Arnhem, The Netherlands	  - The FreeBSD Project 
    			WWW : http://www.tcja.nl  http://www.freebsd.org