Current primary plane update flow when blob is enabled (for zero copy display sharing) shows fence synchronization problems when multi planes are referencing a same single large FB (i.e. multi displays in extended mode). This is because there is only one fence bound to the FB and this single fence is re-used asynchronously when flushing all associated planes.
The way to prevent this is to assign the fence for each plane so that flushing one plane won't be affecting or affected by other plane's flush operation. The 1st patch "drm/virtio: .release ops for virtgpu fence release" which adds device specific release ops is for making the virtio_gpu fence freed upon the last dma_fence_put call. The 2nd patch "drm/virtio: fence created per cursor/plane update" contains the main implementation of per-plane fence. Dongwon Kim (2): drm/virtio: .release ops for virtgpu fence release drm/virtio: fence created per cursor/plane update drivers/gpu/drm/virtio/virtgpu_drv.h | 1 - drivers/gpu/drm/virtio/virtgpu_fence.c | 8 ++ drivers/gpu/drm/virtio/virtgpu_plane.c | 103 ++++++++++--------------- 3 files changed, 47 insertions(+), 65 deletions(-) -- 2.20.1
