[PATCH v2 0/5] virgl: fence fd support

[Rob Herring]

On Fri, Feb 16, 2018 at 9:12 AM, Gustavo Padovan <gustavo at padovan.org> wrote:
> From: Gustavo Padovan <gustavo.padovan at collabora.com>
>
> Hi,
>
> So I finally got around to finish this work! :)
> Here are the updated patchset with fixes for Rob Herring incorporated.
> This follow pretty much the same semantics of other drivers that
> implemented explicit fence support. It extends the execbuf ioctl to have
> an fence_fd argument that when entering the kernel carries the in-fence
> and when returning carries back the out-fence.
>
> The mesa patch can be found at
>
> https://gitlab.collabora.com/padovan/mesa
>
> Please review!
>
> Regards,
>
> Gustavo
>
> Gustavo Padovan (5):
>   drm/virtio: add virtio_gpu_alloc_fence()
>   drm/virtio: add uapi for in and out explicit fences
>   drm/virtio: add in-fences support for explicit synchronization
>   drm/virtio: add out-fences support for explicit synchronization
>   drm/virtio: bump driver version after explicit synchronization
>     addition
>
>  drivers/gpu/drm/virtio/virtgpu_drv.h   |  21 +++---
>  drivers/gpu/drm/virtio/virtgpu_fence.c |  41 ++++++++----
>  drivers/gpu/drm/virtio/virtgpu_ioctl.c | 115 +++++++++++++++++++++++++++------
>  drivers/gpu/drm/virtio/virtgpu_plane.c |  46 +++++++++++--
>  drivers/gpu/drm/virtio/virtgpu_vq.c    |  16 ++---
>  include/uapi/drm/virtgpu_drm.h         |  13 +++-
>  6 files changed, 197 insertions(+), 55 deletions(-)

I tested this on Android. Without the 16ms sleep in the vsync callback
to fake the timing, I get some errors about invalid fences after the
splash screen and when the UI comes up.

BTW, I'm testing on a version of drm_hwcomposer with sw_sync and GL
compositing removed.

At least with that one line hack/fix:

Tested-by: Rob Herring <robh at kernel.org>

Rob