[Mesa-dev] [PATCH v4 7/9] st/va: add headless support, i.e. VA_DISPLAY_DRM

[Emil Velikov]

On 30 October 2015 at 14:48, Daniel Stone <daniel at fooishbar.org> wrote:
> Hi Julien,
>
> On 30 October 2015 at 11:32, Julien Isorce <julien.isorce at gmail.com> wrote:
>> Thx for pointing this out. Where are the files related to
>> winsys/presentation :) ?
>
> The one you're changing. ;)
>
> src/gbm already has a great deal of infrastructure for handling direct
> connections to DRM/KMS devices, buffer allocation, et al.
> src/egl/drivers/dri2/platform_x11.c already has the support for
> connecting to the X server, displaying buffers inside a window using
> DRI2/DRI3, etc etc. src/egl/drivers/dri2/platform_wayland.c implements
> everything you need to do the same with Wayland.
>
> It would be really great if it were possible to reuse all that
> infrastructure rather than reinventing it inside the VA-API code.
>
>> Is your remark a blocker for landing the patches I submitted ? Maybe we can
>> still land them and then if you could guide me what I should change to use
>> newer api that would be great.
>
> Right, as I said it's certainly not your fault! I'd be happy to see
> this series land if there were an aspiration towards unifying the
> codepaths.
>
> At the moment, the winsys support in src/egl is, unsurprisingly,
> pretty closely tied to EGL. So I guess it would be a matter of
> modifying the existing internal EGL platform API to be able to use it
> from vl_*, rather than straight reuse. DItto for gbm if there's
> anything useful in there that could use the same codepaths.
>
Bth, I'm not sure if one can share much with the dri backed loaders
(libgl/libegl/libgbm) - on the drm/gbm side from we (vl) don't require
anything special.
In theory the idea is great, in practise I think that things will end
up quite ugly pretty quickly.

-Emil