If the guest state is paused before it gets a response for the current scanout frame submission (resource-flush), it won't flush new frames after being restored as it still waits for the old response, which is accepted as a scanout render done signal. So it's needed to unblock the current scanout render pipeline before the run state is changed to make sure the guest receives the response for the current frame submission.
v2: Giving some time for the fence to be signaled before flushing the pipeline Cc: Marc-André Lureau <marcandre.lur...@redhat.com> Cc: Vivek Kasireddy <vivek.kasire...@intel.com> Signed-off-by: Dongwon Kim <dongwon....@intel.com> --- ui/gtk.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) diff --git a/ui/gtk.c b/ui/gtk.c index 810d7fc796..ea8d07833e 100644 --- a/ui/gtk.c +++ b/ui/gtk.c @@ -678,6 +678,25 @@ static const DisplayGLCtxOps egl_ctx_ops = { static void gd_change_runstate(void *opaque, bool running, RunState state) { GtkDisplayState *s = opaque; + int i; + + if (state == RUN_STATE_SAVE_VM) { + for (i = 0; i < s->nb_vcs; i++) { + VirtualConsole *vc = &s->vc[i]; + + if (vc->gfx.guest_fb.dmabuf && + vc->gfx.guest_fb.dmabuf->fence_fd >= 0) { + eglClientWaitSync(qemu_egl_display, + vc->gfx.guest_fb.dmabuf->sync, + EGL_SYNC_FLUSH_COMMANDS_BIT_KHR, + 100000000); + + /* force flushing current scanout blob rendering process + * just in case the fence is still not signaled */ + gd_hw_gl_flushed(vc); + } + } + } gd_update_caption(s); } -- 2.34.1