On Sun, 7 Nov 2004 21:28:06 +0100, Markus W Kilbinger
<kilbi@rad.rwth-aachen.de> wrote:
> >> Kevin Lahey <kml@patheticgeek.net> writes:
> 
> >> Looking at the kernel backtraces that I get when things go wrong,
> >> it seems more like a MIPS-specific VM issue than a Tulip driver
> >> issue. I agree that it is a problem.
> 
> My impression, too! (see
> http://mail-index.netbsd.org/port-cobalt/2004/11/04/0000.html )
> 
> I've played with cobalt/cobalt/bus.c dcache handling during pci i/o
> lst day; I still get corrupted files!
> 
> Decreasing sysctl's vm.max{anon,exec,file}max diminishes the problem
> for me (but does not solve the problem)... and also goes into VM
> direction!
> 
> Now, how to find somebody who _can_ investigate this further?
> 
> Shouldn't other mips people (e. g. sgimips) see the same problems?
> 
> Markus.
> 

I think it was Izumi who said that the PCI implementation on the
cobalt is weird and this is what was causing it. I don't know the
details. They were talking for a while about getting a change into the
machine independent code though, which is going to great lengths to
get it fixed.

Andy