FYI, I suspect for now the best might be to let filesystems that define ->update_times work as-is and not tie them into the infrastructure. At least for XFS I suspect the lazy updates might better be handled internally, although I'm not entirely sure yet.