[Xorg] The big multiconsole nasty

[Egbert Eich]

Keith Packard writes:
 > 
 > Around 16 o'clock on Jun 29, Alan Cox wrote:
 > 
 > > 1. Multiple video cards means making RAC work across *multiple* X
 > > servers running at once on a system each with its own console
 > 
 > A significant amount of the RAC effort involves sharing the standard VGA 
 > I/O port and video memory addresses; if we accept that supporting multiple ISA
 > video cards isn't along the critical path, we can probably delay a large 
 > portion of this work and focus only on getting cards POSTed at boot time.

We have never been able to support multiple ISA video cards.
However RAC is an importand issue for a lot of PCI (AGP) graphics cards
at least during initialization.

 > 
 > > 3. Borrowing things like BIOS mode switch functionality is going to
 > > get really hairy.
 > 
 > But may well be necessary for the forseeable future -- Intel has no 
 > current plan to document the i810 (et al) mode selection code, so we're 
 > stuck with using the BIOS.

It is not only Intel that creates these problems. For many brand new video
cards the VESA driver is the only option. Distributors like to use the
VESA driver as fallback to give customers a grapical login - even if there
isn't a native driver around. 
  no grapical login == very unhappy customer

 > 
 > Seems like we need some level of kernel arbitration to keep the system
 > stable; it could be as little as some PCI remapping ioctls and a little 
 > lock for shared regions.
 > 

We should not need any PCI remapping if the kernel itself does the right
mapping. The remapping in the Xserver originated in the days when kernels
didn't do any PCI validations and BIOSes were were not able to map multiple
graphics cards correctly.
What we may need is an IOCTL to set VGA routing on bridges. 

Egbert.