[Openchrome-devel] the future of VIA drm/3D,

[Thomas Hellström]

Hi, Dave!

Dave Airlie wrote:
> Hi guys,
>
> I'm just wondering what your views and if there is any consenus on the
> VIA drm going forward.
>
> My hope is that Thomas's codebase is used to reimplement support for
> the latest Chrome9 chipsets on top of, then that is submitted
> upstream.
>
> At the moment I have a long list of outstanding patches for Chrome9
> support to go upstream, however I realise these patches are primarily
> there to
> support the closed source binary userspace that exists. I'm not happy
> to upstream these unless some forward motion on an open source
> userspace exists.
>   
I'm fine with the existing "via" drm driver coexisting with the 
"openchrome" drm driver, and even VIA adding chrome9 support to the old 
one, but I think any _open-source_ Chrome9 efforts should be based on 
the new drm module architecture and gallium3D. The new Chrome9 DRM stuff 
should not be too hard to port over. However, we're still missing shader 
docs for the Chrome9 3D engine for any gallium efforts to commence.

There are no open-source gallium3D Chrome9 efforts yet, but the DRM / 
TTM functionality can be set up and tested with 2D, I guess.

> However if Thomas's code is to be the basis for an open source
> userspace, I can't see the point in merging the old code and
> interfaces. The upstream first policy sucks for people who don't
> upstream code in time, but it does lead to a cleaner Linux kernel over
> time.
>
> Thomas: I think for TTM we could probably look at making the TTM code
> a submodule that is linked to the drivers, if shouldn't require too
> many hooks into the core DRM if at all.
>   
I agree completely. TTM uses drm hashtabs and the linear range manager. 
The drm_global stuff I added can be moved over to ttm_global if we add a 
ttm submodule. DRM doesn't call into TTM at all. We should probably 
discuss, however, whether to add a separate character device for TTM.

Thanks,
Thomas


> Harald: I also have had Red Hat people asking me to try and setup some
> sort of HW stream that I can at least do some regression tests on,
> I've got a lab in the office here in Brisbane where I can setup
> systems for remote access and testing by the team here. I'm not sure
> how much time RH can allocate towards VIA stuff but I'd really like
> some hw to have for testing stuff on.
>
> Also any ideas on future kernel modesetting drivers? gallium etc?
>
> Dave.
>