It would be nice ivy-mode supported `describe-face`. It does call `completing-read`, but via `complete-face-name`.