Re: Any O2 R10K patches I should try?

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

macallan%NetBSD.org@localhost wrote:

> > My first attempts have been with R10K O2s.  I can boot then with
> > the latest -current.  But networking will cause these machines to
> > reset after only 15-20 seconds.  cvs, ftp and scp have been tried.
> > I've checked the mailing list archives, but don't see this addressed
> > particularly in the last few months.

> I have an r5k O2, mec works just fine - I'd suspect the r10k needs  
> some additional cache flushes / invalidates. An off the shelf PCI  
> ethernet card might do the trick too.

The problem is not in R10000 itself, but
SGI's non-coherent cache hardware design:
http://mail-index.netbsd.org/port-sgimips/2000/06/29/0006.html

I'm afraid hacking on complicated workaround for
such stupid ;-p hardware design would be a bit boring...

> > I've also noticed the R10Ks don't recognize memory over 256M.  Is this
> > still a limitation in the port for O2s?
> 
> That's still a limitation on all O2 variants, although there might be  
> a way to access an additional 512MB without going LP64.

It's current NetBSD/mips pmap limitation. On O2 only 256MB RAM
is mapped at 0x0-0x10000000 for compatibility, and whole RAM is
mapped at higher address which can't be accessed via KSEG0,
and current mips pmap can't handle such memory.

Maybe we should take a look at OpenBSD/mips64..
---
Izumi Tsutsui