net: atlantic: implement UDP GSO offload

atlantic hardware does support UDP hardware segmentation offload.
This allows user to specify one large contiguous buffer with data
which then will be split automagically into multiple UDP packets
of specified size.

Bulk sending of large UDP streams lowers CPU usage and increases
bandwidth.

We did estimations both with udpgso_bench_tx test tool and with modified
iperf3 measurement tool (4 streams, multithread, 200b packet size)
over AQC<->AQC 10G link. Flow control is disabled to prevent RX side
impact on measurements.

No UDP GSO:
	iperf3 -c 10.0.1.2 -u -b0 -l 200 -P4 --multithread
UDP GSO:
	iperf3 -c 10.0.1.2 -u -b0 -l 12600 --udp-lso 200 -P4 --multithread

Mode          CPU   iperf speed    Line speed   Packets per second
-------------------------------------------------------------
NO UDP GSO    350%   3.07 Gbps      3.8 Gbps     1,919,419
SW UDP GSO    200%   5.55 Gbps      6.4 Gbps     3,286,144
HW UDP GSO    90%    6.80 Gbps      8.4 Gbps     4,273,117

Signed-off-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 28 insertions, 24 deletions

diff --git a/drivers/net/ethernet/aquantia/atlantic/aq_nic.c b/drivers/net/ethernet/aquantia/atlantic/aq_nic.c
index 7ad8eb535d28..a17a4da7bc15 100644
--- a/drivers/net/ethernet/aquantia/atlantic/aq_nic.c
+++ b/drivers/net/ethernet/aquantia/atlantic/aq_nic.c
@@ -309,6 +309,7 @@ void aq_nic_ndev_init(struct aq_nic_s *self)
 	self->ndev->vlan_features |= NETIF_F_HW_CSUM | NETIF_F_RXCSUM |
 				     NETIF_F_RXHASH | NETIF_F_SG |
 				     NETIF_F_LRO | NETIF_F_TSO;
+	self->ndev->gso_partial_features = NETIF_F_GSO_UDP_L4;
 	self->ndev->priv_flags = aq_hw_caps->hw_priv_flags;
 	self->ndev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
 
@@ -472,11 +473,18 @@ unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
 {
 	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
 	struct aq_ring_buff_s *first = NULL;
+	u8 ipver = ip_hdr(skb)->version;
 	struct aq_ring_buff_s *dx_buff;
 	bool need_context_tag = false;
 	unsigned int frag_count = 0U;
 	unsigned int ret = 0U;
 	unsigned int dx;
+	u8 l4proto = 0;
+
+	if (ipver == 4)
+		l4proto = ip_hdr(skb)->protocol;
+	else if (ipver == 6)
+		l4proto = ipv6_hdr(skb)->nexthdr;
 
 	dx = ring->sw_tail;
 	dx_buff = &ring->buff_ring[dx];
@@ -484,14 +492,24 @@ unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
 
 	if (unlikely(skb_is_gso(skb))) {
 		dx_buff->mss = skb_shinfo(skb)->gso_size;
-		dx_buff->is_gso = 1U;
+		if (l4proto == IPPROTO_TCP) {
+			dx_buff->is_gso_tcp = 1U;
+			dx_buff->len_l4 = tcp_hdrlen(skb);
+		} else if (l4proto == IPPROTO_UDP) {
+			dx_buff->is_gso_udp = 1U;
+			dx_buff->len_l4 = sizeof(struct udphdr);
+			/* UDP GSO Hardware does not replace packet length. */
+			udp_hdr(skb)->len = htons(dx_buff->mss +
+						  dx_buff->len_l4);
+		} else {
+			WARN_ONCE(true, "Bad GSO mode");
+			goto exit;
+		}
 		dx_buff->len_pkt = skb->len;
 		dx_buff->len_l2 = ETH_HLEN;
-		dx_buff->len_l3 = ip_hdrlen(skb);
-		dx_buff->len_l4 = tcp_hdrlen(skb);
+		dx_buff->len_l3 = skb_network_header_len(skb);
 		dx_buff->eop_index = 0xffffU;
-		dx_buff->is_ipv6 =
-			(ip_hdr(skb)->version == 6) ? 1U : 0U;
+		dx_buff->is_ipv6 = (ipver == 6);
 		need_context_tag = true;
 	}
 
@@ -525,24 +543,9 @@ unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
 	++ret;
 
 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
-		dx_buff->is_ip_cso = (htons(ETH_P_IP) == skb->protocol) ?
-			1U : 0U;
-
-		if (ip_hdr(skb)->version == 4) {
-			dx_buff->is_tcp_cso =
-				(ip_hdr(skb)->protocol == IPPROTO_TCP) ?
-					1U : 0U;
-			dx_buff->is_udp_cso =
-				(ip_hdr(skb)->protocol == IPPROTO_UDP) ?
-					1U : 0U;
-		} else if (ip_hdr(skb)->version == 6) {
-			dx_buff->is_tcp_cso =
-				(ipv6_hdr(skb)->nexthdr == NEXTHDR_TCP) ?
-					1U : 0U;
-			dx_buff->is_udp_cso =
-				(ipv6_hdr(skb)->nexthdr == NEXTHDR_UDP) ?
-					1U : 0U;
-		}
+		dx_buff->is_ip_cso = (htons(ETH_P_IP) == skb->protocol);
+		dx_buff->is_tcp_cso = (l4proto == IPPROTO_TCP);
+		dx_buff->is_udp_cso = (l4proto == IPPROTO_UDP);
 	}
 
 	for (; nr_frags--; ++frag_count) {
@@ -597,7 +600,8 @@ mapping_error:
 	     --ret, dx = aq_ring_next_dx(ring, dx)) {
 		dx_buff = &ring->buff_ring[dx];
 
-		if (!dx_buff->is_gso && !dx_buff->is_vlan && dx_buff->pa) {
+		if (!(dx_buff->is_gso_tcp || dx_buff->is_gso_udp) &&
+		    !dx_buff->is_vlan && dx_buff->pa) {
 			if (unlikely(dx_buff->is_sop)) {
 				dma_unmap_single(aq_nic_get_dev(self),
 						 dx_buff->pa,