net/funeth: Support UDP segmentation offload

Handle skbs with SKB_GSO_UDP_L4, advertise the offload in features, and
add an ethtool counter for it. Small change to existing TSO code due to
UDP's different header length.
Signed-off-by: Dimitris Michailidis <dmichail@fungible.com>
Link: https://lore.kernel.org/r/20220622223703.59886-1-dmichail@fungible.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/fungible/funeth/funeth_ethtool.c

@@ -78,6 +78,7 @@ static const char * const txq_stat_names[] = {
 	"tx_cso",
 	"tx_tso",
 	"tx_encapsulated_tso",
+	"tx_uso",
 	"tx_more",
 	"tx_queue_stops",
 	"tx_queue_restarts",
@@ -778,6 +779,7 @@ static void fun_get_ethtool_stats(struct net_device *netdev,
 		ADD_STAT(txs.tx_cso);
 		ADD_STAT(txs.tx_tso);
 		ADD_STAT(txs.tx_encap_tso);
+		ADD_STAT(txs.tx_uso);
 		ADD_STAT(txs.tx_more);
 		ADD_STAT(txs.tx_nstops);
 		ADD_STAT(txs.tx_nrestarts);

drivers/net/ethernet/fungible/funeth/funeth_main.c

@@ -1357,7 +1357,8 @@ static const struct net_device_ops fun_netdev_ops = {
 #define GSO_ENCAP_FLAGS (NETIF_F_GSO_GRE | NETIF_F_GSO_IPXIP4 | \
 			 NETIF_F_GSO_IPXIP6 | NETIF_F_GSO_UDP_TUNNEL | \
 			 NETIF_F_GSO_UDP_TUNNEL_CSUM)
-#define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
+#define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN | \
+		   NETIF_F_GSO_UDP_L4)
 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_HW_CSUM | TSO_FLAGS | \
 		   GSO_ENCAP_FLAGS | NETIF_F_HIGHDMA)
 

drivers/net/ethernet/fungible/funeth/funeth_tx.c

@@ -130,6 +130,7 @@ static unsigned int write_pkt_desc(struct sk_buff *skb, struct funeth_txq *q,
 	struct fun_dataop_gl *gle;
 	const struct tcphdr *th;
 	unsigned int ngle, i;
+	unsigned int l4_hlen;
 	u16 flags;
 
 	if (unlikely(map_skb(skb, q->dma_dev, addrs, lens))) {
@@ -178,6 +179,7 @@ static unsigned int write_pkt_desc(struct sk_buff *skb, struct funeth_txq *q,
 						 FUN_ETH_UPDATE_INNER_L3_LEN;
 			}
 			th = inner_tcp_hdr(skb);
+			l4_hlen = __tcp_hdrlen(th);
 			fun_eth_offload_init(&req->offload, flags,
 					     shinfo->gso_size,
 					     tcp_hdr_doff_flags(th), 0,
@@ -185,6 +187,24 @@ static unsigned int write_pkt_desc(struct sk_buff *skb, struct funeth_txq *q,
 					     skb_inner_transport_offset(skb),
 					     skb_network_offset(skb), ol4_ofst);
 			FUN_QSTAT_INC(q, tx_encap_tso);
+		} else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
+			flags = FUN_ETH_INNER_LSO | FUN_ETH_INNER_UDP |
+				FUN_ETH_UPDATE_INNER_L4_CKSUM |
+				FUN_ETH_UPDATE_INNER_L4_LEN |
+				FUN_ETH_UPDATE_INNER_L3_LEN;
+
+			if (ip_hdr(skb)->version == 4)
+				flags |= FUN_ETH_UPDATE_INNER_L3_CKSUM;
+			else
+				flags |= FUN_ETH_INNER_IPV6;
+
+			l4_hlen = sizeof(struct udphdr);
+			fun_eth_offload_init(&req->offload, flags,
+					     shinfo->gso_size,
+					     cpu_to_be16(l4_hlen << 10), 0,
+					     skb_network_offset(skb),
+					     skb_transport_offset(skb), 0, 0);
+			FUN_QSTAT_INC(q, tx_uso);
 		} else {
 			/* HW considers one set of headers as inner */
 			flags = FUN_ETH_INNER_LSO |
@@ -195,6 +215,7 @@ static unsigned int write_pkt_desc(struct sk_buff *skb, struct funeth_txq *q,
 			else
 				flags |= FUN_ETH_UPDATE_INNER_L3_CKSUM;
 			th = tcp_hdr(skb);
+			l4_hlen = __tcp_hdrlen(th);
 			fun_eth_offload_init(&req->offload, flags,
 					     shinfo->gso_size,
 					     tcp_hdr_doff_flags(th), 0,
@@ -209,7 +230,7 @@ static unsigned int write_pkt_desc(struct sk_buff *skb, struct funeth_txq *q,
 
 		extra_pkts = shinfo->gso_segs - 1;
 		extra_bytes = (be16_to_cpu(req->offload.inner_l4_off) +
-			       __tcp_hdrlen(th)) * extra_pkts;
+			       l4_hlen) * extra_pkts;
 	} else if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
 		flags = FUN_ETH_UPDATE_INNER_L4_CKSUM;
 		if (skb->csum_offset == offsetof(struct udphdr, check))

drivers/net/ethernet/fungible/funeth/funeth_txrx.h

@@ -82,6 +82,7 @@ struct funeth_txq_stats {  /* per Tx queue SW counters */
 	u64 tx_cso;        /* # of packets with checksum offload */
 	u64 tx_tso;        /* # of non-encapsulated TSO super-packets */
 	u64 tx_encap_tso;  /* # of encapsulated TSO super-packets */
+	u64 tx_uso;        /* # of non-encapsulated UDP LSO super-packets */
 	u64 tx_more;       /* # of DBs elided due to xmit_more */
 	u64 tx_nstops;     /* # of times the queue has stopped */
 	u64 tx_nrestarts;  /* # of times the queue has restarted */