There is Mach64 drivers just don know how to make them. Here is the site 

http://www.retinalburn.net/linux/dri_status.html

On the status page it says AGP drivers for nebsd are working. I just
dontknow how to do it. The drivers are in the mach64 branch of dri.

On Thu, 9 Oct 2003, Frederick Bruckman wrote:
> Date: Thu, 9 Oct 2003 15:25:20 -0500 (CDT)
> To: Nicholas Ricketts <nightshade@phreaker.net>
> From: Frederick Bruckman <fredb@immanent.net>
> Subject: Re: help with video dri/drm problems
> 
> On Thu, 9 Oct 2003, Nicholas Ricketts wrote:
> 
> > How do you compile DRM/DRI. I have downloaded the macch64 branch of DRi
> from
> > thier CVS and did a make world. But I dont see any drm modules anywhere
> > after I do the build. Where do you compile DRM??
> 
> There doesn't seem to be a BSD DRM module for mach64. For say, mga,
> though, in the built tree, in
> "programs/Xserver/hw/xfree86/os-support/bsd/drm/kernel/mga", type
> 
>     make opt_drm.h
>     touch locators.h
>     make <options, as below>.
> 
> The other modules are in adjacent directories. gcc-3.3.1 broke some of
> them, but I think at least mga, tdfx and r128 built. Fixes to the
> others would probably not be difficult, but it's still "hacker's only"
> at this point.
> 
> That's all I know. Erik is the DRI guy who has been doing the work on
> this. You should probably direct any more questions to Erik.
> 
> > On Thu, 9 Oct 2003, Frederick Bruckman wrote:
> > > Date: Thu, 9 Oct 2003 08:03:35 -0500 (CDT)
> > > To: Jaromir Dolecek <jdolecek@NetBSD.org>
> > > From: Frederick Bruckman <fredb@immanent.net>
> > > Subject: Re: help with video dri/drm problems
> > >
> > > On Thu, 9 Oct 2003, Jaromir Dolecek wrote:
> > >
> > > > Frederick Bruckman wrote:
> > > > > by hand. It had to be force loaded, with "-f", because there
> seems to
> > > > > be something wrong with the LKM versioning scheme: it complained
> that
> > > > > module is <some random word>, kernel is <some other random
> words>,
> > > >
> > > > This means the LKM was compiled with different options than
> > > > kernel, and so the LKM is not compatible with running kernel.
> > > > Specifically, if the LKM isn't compiled with MULTIPROCESSOR,
> > > > it will crash kernel compiled with MULTIPROCESSOR.
> > > >
> > > > You NEED to compile LKMs with exactly same options as kernel, such
> > > > as:
> > > >
> > > > cd src/sys/lkm
> > > > make clean
> > > > DBG="-O2 -DMULTIPROCESSOR" make dependall
> > > > make install
> > >
> > > Ahh. It was indeed a multiprocessor kernel. -DMULTIPROCESSOR isn't
> > > even defined anywhere in the kernel "Makefile", however; it's set in
> > > opt_multiprocessor.h. So how do you know, of all the options in the
> > > kernel config file, which options count for this purpose?
> > >
> > > > Do NOT use forced LKM load. It WILL make the kernel crash once the
> > > > LKM would do anything. I thought the message emitted by modload
> > > > with -f is discouraging enough. I guess I'd need to improve it
> > > > further :)
> > >
> > > Well, I had to see what would happen for myself, warning or no
> > > warning. The initial error on mismatch could stand to be clearer.
> > >
> > > Thanks for heads ups. I will give it another try.
> 
> Frederick
>