[Xorg] The big multiconsole nasty

[John Dennis]

> You can't make the assumptions about VGA routing. When secondary video
> cards are reset by running their ROM, at the very least they enable
> their VGA support. Sometimes they remap the bridges too.

The first part I'll buy, the second part about VBIOS remapping
(upstream) bridges feels wrong, but nothing would surprise me.

> We need kernel support for:

Precisely, its the kernel that should be managing all this.

> #3 needs a special case. The kernel has to track which VGA device the
> BIOS left active. When an app asks for the VBIOS ROM from this device,
> you return the memory contents at C000:0. Some laptops compress the
> VBIOS image of the primary adapter and this is where the uncompressed
> version is stored. The kernel has to remember the boot device so that
> if someone changes the active VGA device we will still know which
> adapter the image at C000:0 belongs too.

Yes, and this is compounded by the fact that the PCI spec demands that
BIOS images be executed from host memory, not card memory. CPU
architecture and OS designs impose restrictions on executable memory
requirements that the kernel is aware but generic user level code may be
ignorant of, not to mention the low level firmware the OS cooperates
with, best to let the kernel/firmware do its job.

-- 
John Dennis <jdennis at redhat.com>