net: dsa: Use prepare/commit phase in dsa_slave_vlan_rx_add_vid()

We were skipping the prepare phase which causes some problems with at
least a couple of drivers:

- mv88e6xxx chooses to skip programming VID = 0 with -EOPNOTSUPP in
  the prepare phase, but we would still try to force this VID since we
  would only call the commit phase and so we would get the driver to
  return -EINVAL instead

- qca8k does not currently have a port_vlan_add() callback implemented,
  yet we would try to call that unconditionally leading to a NPD

Fix both issues by conforming to the current model doing a
prepare/commit phase, this makes us consistent throughout the code and
assumptions.

Reported-by: Heiner Kallweit <hkallweit1@gmail.com>
Reported-by: Michal Vokáč <michal.vokac@ysoft.com>
Fixes: 061f6a5 ("net: dsa: Add ndo_vlan_rx_{add, kill}_vid implementation")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/dsa/slave.c

@@ -993,6 +993,7 @@ static int dsa_slave_vlan_rx_add_vid(struct net_device *dev, __be16 proto,
 		/* This API only allows programming tagged, non-PVID VIDs */
 		.flags = 0,
 	};
+	struct switchdev_trans trans;
 	struct bridge_vlan_info info;
 	int ret;
 
@@ -1009,11 +1010,13 @@ static int dsa_slave_vlan_rx_add_vid(struct net_device *dev, __be16 proto,
 			return -EBUSY;
 	}
 
-	ret = dsa_port_vlan_add(dp, &vlan, NULL);
+	trans.ph_prepare = true;
+	ret = dsa_port_vlan_add(dp, &vlan, &trans);
 	if (ret == -EOPNOTSUPP)
-		ret = 0;
+		return 0;
 
-	return ret;
+	trans.ph_prepare = false;
+	return dsa_port_vlan_add(dp, &vlan, &trans);
 }
 
 static int dsa_slave_vlan_rx_kill_vid(struct net_device *dev, __be16 proto,