[wayland-devel] RFC: Ideas to run weston on nvidia/ATI blob with hardware acceleration

[microcai]

2012/7/19 microcai <microcai at fedoraproject.org>:
> The only thing wayland needed is a way to share buffers.
>
> Currently the reference implementation uses EGL(with wayland
> extension) to render, switch to another render api is painful and
> useless.
>
> The only reason that NVIDIA can't cooperate with weston is that, they
> don't provide EGL implementation with wayland extension.
>
> NVIDIA blob doesn't provide libEGL.so right? We can implement stubs
> that wraps libEGL to NVIDIA blob's  libGL.so as a way of building  EGL
> wayland platform on top of NVIDIA blob.
>
> What about KMS ?
>
> Let the blob DDX do the job! We fire up an X server with uses NVIDIA
> blob but with no input to be used by weston. weston create an
> FULLSCREEN GLX window( for mutil monitor, create every fullscreen
> window for every monitor)  and uses that as  output surface.
> Note, this X have no support for input device, and can be specify
> builded for use with weston only.  since we create a fullscreen window
> and do direct rendering, the X layer won't hurt us much. And weston
> didn't get input from X, so no input latency.
>
> What about BUFFER SHARE?
>
> Currently we can use shm, in time NVIDIA(hopefully) will add an GL
> extension that allow us to create share-able buffers among different
> processes. Then we can use that extension instead.
>
> Why not just use X?
> Well, in this way, we can have a working wayland on top of NVIDIA
> blob. So developers/distros won't have to deal with X11 and can start
> migration to wayland right away long before the eval NVIDIA make the
> right choice.
>
> In this way, we eliminated the last Barrier that bind us to X, the blob.
>
> The above idea also apply to ATI blob.

For NVIDIA blob to work, we need two things happen:

1) weston side:
         x11-backend will be changed to only use X for draw, get input
directly though /dev/input/event*
         weston-launch detects an NVIDIA card and NVIDIA blob present,
launches a special builded X and run weston inside that X.

2) xorg side:
        able to build a minimal X that has only the ability to load
video drivers and creat GLX window.
        no input drivers
       no input support
       no  core X render api support
       no font support
       no extensions except GLX


3) Mesa side
use glx to implement EGL and wayland platform.