On Sat, 17 Jun 2000, David O'Brien wrote:

> On Sat, Jun 17, 2000 at 07:33:54AM -0700, Eduardo Horvath wrote:
> > On Sat, 17 Jun 2000, David O'Brien wrote:
> > sparc64 describes 64-bit SPARC binaries.  sparc describes 32-bit SPARC
> > binaries.
> 
> I view "sparc64" as the platform.  I guess that was the wrong terminology
> for NetBSD.

`sparc64' describes two different things.

It describes a machine that runs the 64-bit SPARC v9 instruction set,
which is a superset of the SPARC v8 instruction set.

It also describes an ABI based upon the 64-bit SPARC v9 instruction set.

A SPARC v9 machine can be configured to run the SPARC v8 ABI.

> > The recommended way is to use a NetBSD/sparc userland with a
> > NetBSD/sparc64 32-bit kernel.  That's the most stable combination.
> 
> Yes.  But doing even this is harder than it needs to be.  I spent 3 hours
> trying to do this, including hitting the "Memory Address not Aligned"
> problem that was reported in April.  You replied you were working on that
> problem, but I could not find updated bits anywhere.  Also you've got two
> 32-bit sun4u kernel's -- 1.4U and 1.4X.  Both are dated mid-April.  There
> is a sparc snapshot from mid April, but is that 1.4U or 1.4X, or even
> something else?  The sparc miniroot seems useless on a sun4u(U1).  I
> tried two sets of installation approaches people posted in April, neither
> has been successful for me.  Installing NetBSD on sparc(sun4m), pmax, and
> VAX hasn't been this hard.

Welcome to life on the bleeding edge B^>.

The problem is that hardware support is limited.  We are running on a
rather small sample of machines, and it seems that whenever we fix some
problem on one type of machine it breaks another type.  I'm currently
working on support for PCI machines, which means, unfortunately,
everything's broken at the moment.

Although from the description I don't think your problem has anyghing to
do with kernel/userland mismatch.  Sounds like it's entirely in the
kernel.

> Would it be possible to put a sun4u-32bit snapshot together that is
> consistent with itself?

I expect to get around to this once 1.5 had been branched and is less of a
moving target.

Eduardo Horvath