On Fri, Aug 25, 2000 at 01:16:07PM +0200, Andrei A. Dergatchev wrote:
> [...]

> > I'm not cluefull enough to know why DMA would cause the machine checks -
> > obviously something's getting fratzed on a bus somewhere.  Is this more
> > likely to be a driver or hardware problem?
> 
> I definitely recall reading something like "earlier models of *** had
> problems
> with *** chip". I'm not sure whether it was about DMA chip (quite likely)
> and
> the pc164. I tried to dig in mail archives and failed :-(
> 
> Some opinions from discussion of this problem from redhat axp-list :
> 
> http://www.lib.uaa.alaska.edu/axp-list/archive/2000-01/0032.html
> "I think this is more likely a programming problem. This error
> can be caused by accessing non-existent memory. The pyxis chip
> ECC checking is mainly done when non-processor accesses are
> occurring, such as PCI bus master DMA, and getting a double bit
> error is very unlikely if you have not gotten any single bit
> errors. Other things which can cause it are PCI bus parity
> errors and PCI bus time-outs. "

Early Pyxis revs had problems DMA crossing page boundaries (IIRC). 
That is why e.g. Miata has restrictions on what you can put in the 64 bit
PCI slots. I have not heard about the same problem on PC164 which I think
uses the Alcor chipset and not the Pyxis.

W/

-- 
Wilko Bulte  	 					wilko@freebsd.org
							Arnhem, the Netherlands