FWIW, I didn't put the intentional d3drm, camera leaks here because it seems that the device controls the leaked references via some sort of destroy callback on the device. (See the todo_wine tests in this patch).

+    todo_wine ok(ref4 > frame_ref2, "Expected ref4 > frame_ref2, got frame_ref2 = %u, ref4 = %u.\n", frame_ref2, ref4);
+
+    IDirect3DRMDevice3_Release(device3);
+    IDirect3DRMDevice_Release(device1);

I tried to release d3drm explicitly myself, like how I did with the texture tests. But the tests would then crash here, at IDirect3DRMDevice_Release, where the device gets destroyed. Which probably indicates that the device stores a list of leaked references and releases them on getting destroyed. For now I've decided to avoid the leak, thus avoiding the use of a destroy callback, and implementing this as a separate later on if an application demands it, unless it's safer to get this done on priority.

Cheers,