> Hi Thiemo.
> > The solution is to use a special kernel with compiled in cache barriers
> > (needs a not-yet written compiler patch) and careful managing of page
> > table accesses from userspace (needs rmap, and thus 2.5 Kernel).
> I noticed this guy talking about O2 and having some patches and precompiled
> kernels. Are these in anyway related to what you are talking about?
Partially yes, the O2 as a ELF32 machine allows a different (and more
performant) solution for the speculative store problem.
> > I'm working sowly on it, but I get constantly distracted by toolchain
> > issues. :-)
> I looked arround your website and saw "mips64-linux-ip28-2002-06-28.tar.gz"
> whoo.. Sounds exiting =]
I know it's somewhat outdated. :-)
> > > There was some talk in Febuary on a guy got Linux up and running on an
> > > Origin 200 - and some bootable cd's. That sounded prety interesting =]
> > The Origin is easier to suppot, it has coherent I/O.
> And you are probably going to tell me that this is the same reson that the
> O2 is working, right?
Only the R5000 O2 is working. The R10000 O2 is the problematic one.
> > > Btw: I also got my hands on an Octane (IP30) with an R10000 (195 Mhz) -
> > > I haven't tried to install on this one, but that might be interesting
> > > too..
> > This one is also I/O coherent, so chances are much better than for IP28.
> Are you saying I might get this one working with the Debian boot-images
> that I already tried, or?
It might work, but I know way to little about the Octane to say something