Well, after a week of hacking away I've managed to do it. I've got a
driver with all the primitives running away on my DECstation beside
me. I've a few things left to do, including verifying support for the
3D cards, tidying up the code and fixing all the offsets for Alpha,
and making sure multi-headed machines will work. Oh, and I've still
got to do the rcons glue.

There are still couple of problems:

1) The reset of the STIC and stamp corrupts the screen a little, so
   we have to clear it, nuking everything said by the kernel so far. Also,
   the screen is dead until you give the stamp at least one command to 
   do initally. Is there a hack to 'rememeber' what has been passed
   through rcons (something like a single page of buffer would do)?.

2) I have to put a call to the VM system in init_mach(). This is because
   we need to allocate 96kB + 32kB of slop, wired down, physically
   contigious and aligned on a 32kB boundary; the reason for putting it
   in init_mach() is that it has to be allocated before the buffers, so
   it gets into the first 8MB of memory (buffers take up 6MB on my box,
   along with kernel space that's over 8MB). 28kB of the slop can be freed 
   in px_attach() (or the whole shebang, if it's not used). I'm convinced 
   there's a better way to do this.

3) The px_blitc() routine excects the font data to be layed out a
   particular way. Also, the little PROM font looks ridiculous on the big
   screens. I'm currently using Courier medium, 18pt from X11R6 on my
   box, and it looks delicious ;). I assume this font is under an MIT
   style license... How about adding a kernel option?

I really like these cards, and they're *fast*. I have another PMAG-CA
which I'll use to test the multi-headed support. I haven't seen anything
on this list in a while, have I 'been unsubscribed' or has there been
no trafic?

Regards,
Andy.