On Wed, Jun 02, 1999 at 02:32:31PM -0700, Robert Keller wrote:
> At 10:56 AM 6/2/99 -0700, William J. Earl wrote:
> >Robert Keller writes:
> > > At 12:46 AM 5/28/99 +0200, Ralf Baechle wrote:
> > > >The R5000 is supported and known to work. You happen to have a
> > > >second level cache on that board?
> > >
> > > how convinced are people that the VR5000 is supported? I'm running
> > > the latest code off the linux.sgi.com cvs server and I'm getting *very*
> > > weird page faulting problems when doing the kernel/user space
> > > hazard shuffle. Its really hard to describe the problems as they are
> > > not deterministic: sometimes the right thing happens, other times
> > > I get restricted instruction exceptions... Both of these can happen on
> > > the very same kernel binary...
> >
> > What sort of system are you using, and what is the hardware
> >configuration (including caches)?
>
> Its an NEC 5074 development board with an VR5000 --
> 32K I and 32K D Primary cache
> no secondary cache.
>
> which vec0 code should I be using?
Like R4k.
> NEC seems to have a bunch of errata and hazards in this part when
> dealing with TLBWR and friends... Is any of the existing code supposed
> to respect these?
I don't have the errata at hand but as I remember we deal correctly with
all the cache and TLB errata documented by IDT/QED. Otherwise for
shure Linux wouldn't run on my R5k Indy.
Is NEC's VR5000 different from the IDT/QED version?
> Also, arch/mips/Makefile makes the compiler use -r8000 -mips2, is
> that right?
Yes.
Ralf
|