James,

I tried the patch, and it does produce a cleaner and smaller driver.
Overall, I like it. Some observations:

1. Without the fb_set_var() hook, switching from X messes up the
console.  I would guess this will be addressed by the console?

fbcon_switch has to be rewritten. I'm going threw the process of cleaning
up the upper fbcon layer. Its such a mess. Yuck!!!

2. Console panning/wrapping does not work.  updatevar includes a check
"if (con == info->currcon)", and my guess is info->currcon is obsoleted
so the check always fails.

It worked before. Strange. Do you mean for you console panning doesn't
work on the visiable VC or a non visible VC?

3. fbdev can be loaded without taking over the console.  After running
an fb-based application, exiting fbdev messes up the vga console
(actually hangs the system).  Should the fbdev driver provide the
capability to restore the VGA state then, ie at info->fb_release?

Yes!!! Of course there is the issue is the framebuffer that actual one
used for vgacon or is it independent, thinking multihead here.

4. The initial font loaded is 8x8. It seems that 8x16 fonts are limited
for the SGI console console only.  Any reason why?

I have no idea why that is. You can select a differnt font.

5.  The cfb_* drawing functions still behave erratically, especially in
emacs.  Geert has made some versions that work correctly for me.  This
was discussed in a thread sometimes ago.

Where are the patchs. I like to incorporate them into BK.

Attached is a diff that will allow the logo to be drawn at 8-bpp
pseudocolor:

Applied.