Hi,

Making a custom kernel is quite easy, given the way NetBSD is set up.
And would only making a Sun4/4C/4M (4D? 4U?) kernel improve performance
enough or would one still need to make a really specific kernel that
only includes the drivers that are really needed? Otherwise, all that
you are really offering is a kernel that has been compiled after only
deleting a few machine-types from the top of the configuration-file, and
that is something that is very easy to have the user do once the system
is up and running.

I have had quite a number of different Sparcs around the house by now, and
always made specific kernels for these machines. Even when my fastest one
was an ELC that was not much of a problem.

Just my 2 cents etc :-)

> I had a thought about the next release of NetBSD/sparc: I think we ought to
> provide SUN4, SUN4C, SUN4M kernels in addition to GENERIC in the release.
> It's a terrific achievement that NetBSD has one GENERIC kernel where it
> otherwise might have three (or four if we ever get around to sun4d), but
> that comes at an overhead cost for critical path code, plus code space
> taken up by drivers that do not exist. My concern is that removing this
> overhead means compiling a system-specific kernel, and I'm not sure the
> average NetBSD/sparc user is doing that. That's why I think we should
> provide a canned solution.
> 
> Comments?
> 
> 	Erik <fair@clock.org>