Still smells like a libdrm bug to me, surely drmGetDevice2 can't rely on its caller initializing the device pointer.