+void dsa_tag_generic_flow_dissect(const struct sk_buff *skb, __be16 *proto,
+				  int *offset)
+{
+	const struct dsa_device_ops *ops = skb->dev->dsa_ptr->tag_ops;
+	int tag_len = ops->overhead;
+
+	*offset = tag_len;
+	*proto = ((__be16 *)skb->data)[(tag_len / 2) - 1];
+}
+EXPORT_SYMBOL(dsa_tag_generic_flow_dissect);

If you look where this is used:

#if IS_ENABLED(CONFIG_NET_DSA)
                if (unlikely(skb->dev && netdev_uses_dsa(skb->dev) &&
                             proto == htons(ETH_P_XDSA))) {
                        const struct dsa_device_ops *ops;
                        int offset = 0;

                        ops = skb->dev->dsa_ptr->tag_ops;
                        if (ops->flow_dissect &&
                            !ops->flow_dissect(skb, &proto, &offset)) {
                                hlen -= offset;
                                nhoff += offset;
                        }
                }
#endif

We have already done ops = skb->dev->dsa_ptr->tag_ops once. But since
it is in a different compilation unit, the optimise has no idea about
this. So building on my last comment, it would be nice to make this an
inline function to help out the optimiser.

       Andrew