diff --git a/pfe_ctrl/pfe_eth.c b/pfe_ctrl/pfe_eth.c
index a8891d9..aad7741 100644
--- a/pfe_ctrl/pfe_eth.c
+++ b/pfe_ctrl/pfe_eth.c
@@ -2665,6 +2665,25 @@
 	priv->cpu_id = -1;
 
 	pfe_eth_fast_tx_timeout_init(priv);
+	/* Initialize mdio */
+	if (minfo[id].enabled) {
+
+		if ((err = pfe_eth_mdio_init(priv, &minfo[id]))) {
+			netdev_err(dev, "%s: pfe_eth_mdio_init() failed\n", __func__);
+			goto err2;
+		}
+	}
+	/* For unused interface, skip adding the network device. The code is
+	 * added here instead of configuring einfo because:
+	 * - id and gem_id is used interchangeably in the pfe control logic.
+	 * - Even fix the id with gem_id, there is a gotcha, the mdio bus
+	 *   setting is only associated with the first gem register. The logic
+	 *   can be found in gemac_set_mdc_div.
+	 */
+	if (!strcmp(einfo[id].name, "unused")) {
+		netif_info(priv, probe, dev, "%s: Skip unused interface: %d\n", __func__, id);
+		return 0;
+	}
 
 	/* Copy the station address into the dev structure, */
 	memcpy(dev->dev_addr, einfo[id].mac_addr, ETH_ALEN);
@@ -2674,16 +2693,7 @@
 	if (err < 0) {
 		netdev_err(dev, "%s: dev_alloc_name(%s) failed\n", __func__, einfo[id].name);
 		err = -EINVAL;
-		goto err2;
-	}
-
-	/* Initialize mdio */
-	if (minfo[id].enabled) {
-
-		if ((err = pfe_eth_mdio_init(priv, &minfo[id]))) {
-			netdev_err(dev, "%s: pfe_eth_mdio_init() failed\n", __func__);
-			goto err2;
-		}
+		goto err3;
 	}
 
 	dev->mtu = 1500;