[PATCH v5 00/13] PRIME Synchronization

[Dave Airlie]

On 13 April 2016 at 14:17, Alex Goins <agoins at nvidia.com> wrote:
> Hello all,
>
> These patches change the xserver to support setting up PRIME with double
> buffering, and implement double buffered PRIME sink and source support in
> the modesetting driver. In addition to these changes, I've upstreamed a
> couple of patches to the i915 DRM driver that mesh with these, and have
> implemented double buffered PRIME source support in the NVIDIA proprietary
> driver (pending release.)
>
> Previous cover letters:
>     v4: https://lists.x.org/archives/xorg-devel/2016-March/048944.html
>     v3: http://lists.x.org/archives/xorg-devel/2016-February/048677.html
>     v2: http://lists.x.org/archives/xorg-devel/2016-January/048434.html
>     v1: http://lists.x.org/archives/xorg-devel/2015-November/048039.html
>
> I wasn't able to get any of my questions answered in the last thread, so I
> addressed the issues how I saw fit. Aside from some small tweaks, the biggest
> changes in this patch set involve resolving the ambiguity with
> rrSetScanoutPixmap(). Instead of using multiple calls to rrSetScanoutPixmap() to
> setup scanout pixmaps for flipping, the scanout pixmap setting is done
> implicitly in rrEnableSharedPixmapFlipping().
>
> There are two new patches that fix outstanding bugs with
> drmmode_set_scanout_pixmap(), with details in their respective commit messages:
>     modesetting: Internal storage of scanout pixmaps
>     modesetting: Always tear down scanout pixmap
>
> Getting RRReplaceScanoutPixmap() working with synchronization would require an
> ABI change to pass in two pixmaps instead of one, so I just made it fail
> gracefully in the synchronized case. None of the drivers that I implemented
> PRIME synchronization for seem to use RRReplaceScanoutPixmap(). In fact, I
> believe it is currently broken with the modesetting driver without the above 2
> patches.

Okay I've finally gotten around to playing with these today. I'm using
a i915 + nouveau
system with nouveau running as the master. Both using modesetting DDX.

The basics seem to work okay, but I am seeing some issues I'm not 100% sure
are the fault of these patches.

Scenario 1:
start X, start mutter against X (using DRI3), start gnome-terminal,
drag g-t around
seems smooth.
set provider output, xrandr in a new display, drag g-t around, I get
choppy rendering
on the main display, the offloaded display is smooth!

I don't think this happens with DRI2 mutter, so I'm not 100% sure what
is going on there,
I assume it's something to do with page flipping not being allowed for
present anymore.

Scenario 2:
start X, set provider output, xrandr in a new display, start mutter in
DRI3 mode, things
go horribly wrong. Again it seems fine in DRI2 mode. so I expect this
is some interaction
with the present extension fighting this.

I'm going to try and look at the interfaces a bit more now that I have
it running on a machine.

Dave.