Re: Clean BIOS booting Mandrake 9 via LILO

On Wed, 2002-11-27 at 23:12, Oliver Schwartz wrote:

> You mean does X program all the registers? No, currently it doesn't. 
> Originally the driver used only the framebuffer mode, i.e. the kernel 
> framebuffer driver would be responsible for mode switching (which 
> fails because of the missing VESA bios).
> 
> For the XVideo extensions I added most of the register setup, although 
> the driver still uses the framebuffer mode. At the moment it will 
> save the current register state, do all the calculations for the new 
> mode and then setup the registers - but not all of them. To make 
> things work I simply commented out those registers that produced 
> garbage on my screen when programmed. Whether this is due to a wrong 
> modeline or some other driver issue I haven't investigated yet.
> 
> In particular, none of the timing parameter is currently set in the X 
> driver. If the graphics chipset is not initialized by the BIOS then X 
> won't output anything either.

I have managed to get the LILO boot thing working stably, but
unfortunately its using the MS Xcode list right now, so unfortunately
its not suitable for CVS yet :-(.  There's clearly something bad
happening during the boot process that is related to the cache, I can't
figure out what; its not fixed but merely not being triggered by this
build.  But for the moment this particular build works stably and I once
again am drawn back to my bete noir, getting video up in cromwell.  A
little bit more detail on how I am hoping to attack it this time:

I have hooked up two v1.0 xboxes, one has the Filtror with cromwell in
it booting a 'blank screen' Linux, set up for 192.168.0.64, and the
other is booting from XtenderV1.1 -> Mandrake 9 boot CD, and is set up
for 192.168.0.65.

My plan is to give both an NFS mount on to a directory on my main
machine, ssh into both from my main machine, and write a small app using
mmap() (compiled by one box into the NFS shared directory on the main
machine, so its visible to both Xboxes) which will allow me to fairly
easily try all kinds of register copying between the machines, all in
Linux.  Essentially I hope to copy the register guts out of a live,
working machine and deliver them into the cold, unliving flesh of the
video chip on the second, cromwell-booted machine to see if We Can
Create The Mystical Spark Of Video Life In It, using these creepy
techniques.

I had a look through the kernel code you kindly pointed at Oliver, and
there's clearly a lot of useful stuff in there, especially the bunch of
helper functions which specify which register pairs hide their own
personal address/data maps.

There's one thing that's really worrying me, in
./drivers/video/riva/nv4ref.h at the beginning are defined "Magic values
to lock/unlock extended regs"

Do you know what this is about?  It looks like by default not all regs
are accessible for read :-(

-Andy



-------------------------------------------------------
This SF.net email is sponsored by: Get the new Palm Tungsten T 
handheld. Power & Color in a compact size! 
http://ads.sourceforge.net/cgi-bin/redirect.pl?palm0002en