Suggestions for properly abstracting multiple screens.

Dave Airlie airlied at gmail.com
Mon Feb 25 23:04:00 PST 2008


On Tue, Feb 26, 2008 at 4:58 PM, JoJo jojo <onetwojojo at gmail.com> wrote:
> that....makes.......sense ;-0
>
>  so instead of device "pci bus id"
>  we use virtual device "vgaArbitrater n"
>
>  X -> vgaArbitrator -> PCI bus  -> GPU
>
>  the two scenarios were:-
>  a) 1 parent X controlling  all the crtc running child X on screens
>  b) independent X controlling individual crtcs
>
>  vgaArbitrater will support b) making a) unnecessary
>
>  The objective of vgaArbitrator, originally was to do this cleanly,
>  for multiple gpu. Now vgaArbitrator will expose individual crtc as a
>  device node for X
>  (I was thinking it would only expose 1 device to X per gpu)
>
>  I suppose the way devices & screens are handled in X will also change.
>

Nope, the vga aribiter is only for arbitration between VGA registers
on multiple GPU, its just part of the puzzle to getting multi-seat
working..

it doesn't get involved at all to do two users on one GPU scenarios.
all this stuff will happen with kernel modesetting and the drm before
we get the arbiter into shape.

the arbiter only matters for the *multiple* GPU case.

Dave.



More information about the xorg mailing list