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>

Documentation/networking/device_drivers/aquantia/atlantic.txt

@@ -325,6 +325,21 @@ Supported ethtool options
  Example:
  ethtool -N eth0 flow-type udp4 action 0 loc 32
 
+ UDP GSO hardware offload
+ ---------------------------------
+ UDP GSO allows to boost UDP tx rates by offloading UDP headers allocation
+ into hardware. A special userspace socket option is required for this,
+ could be validated with /kernel/tools/testing/selftests/net/
+
+    udpgso_bench_tx -u -4 -D 10.0.1.1 -s 6300 -S 100
+
+ Will cause sending out of 100 byte sized UDP packets formed from single
+ 6300 bytes user buffer.
+
+ UDP GSO is configured by:
+
+    ethtool -K eth0 tx-udp-segmentation on
+
  Private flags (testing)
  ---------------------------------
 

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 @@ unsigned int aq_nic_map_skb(struct aq_nic_s *self, struct sk_buff *skb,
 	     --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,

drivers/net/ethernet/aquantia/atlantic/aq_ring.h

@@ -65,19 +65,20 @@ struct __packed aq_ring_buff_s {
 	};
 	union {
 		struct {
-			u16 len;
+			u32 len:16;
 			u32 is_ip_cso:1;
 			u32 is_udp_cso:1;
 			u32 is_tcp_cso:1;
 			u32 is_cso_err:1;
 			u32 is_sop:1;
 			u32 is_eop:1;
-			u32 is_gso:1;
+			u32 is_gso_tcp:1;
+			u32 is_gso_udp:1;
 			u32 is_mapped:1;
 			u32 is_cleaned:1;
 			u32 is_error:1;
 			u32 is_vlan:1;
-			u32 rsvd3:5;
+			u32 rsvd3:4;
 			u16 eop_index;
 			u16 rsvd4;
 		};

drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_a0.c

@@ -454,7 +454,7 @@ static int hw_atl_a0_hw_ring_tx_xmit(struct aq_hw_s *self,
 
 		buff = &ring->buff_ring[ring->sw_tail];
 
-		if (buff->is_gso) {
+		if (buff->is_gso_tcp) {
 			txd->ctl |= (buff->len_l3 << 31) |
 				(buff->len_l2 << 24) |
 				HW_ATL_A0_TXD_CTL_CMD_TCP |

drivers/net/ethernet/aquantia/atlantic/hw_atl/hw_atl_b0.c

@@ -43,7 +43,9 @@
 			NETIF_F_NTUPLE |  \
 			NETIF_F_HW_VLAN_CTAG_FILTER | \
 			NETIF_F_HW_VLAN_CTAG_RX |     \
-			NETIF_F_HW_VLAN_CTAG_TX,      \
+			NETIF_F_HW_VLAN_CTAG_TX |     \
+			NETIF_F_GSO_UDP_L4      |     \
+			NETIF_F_GSO_PARTIAL,          \
 	.hw_priv_flags = IFF_UNICAST_FLT, \
 	.flow_control = true,		  \
 	.mtu = HW_ATL_B0_MTU_JUMBO,	  \
@@ -533,8 +535,9 @@ static int hw_atl_b0_hw_ring_tx_xmit(struct aq_hw_s *self,
 
 		buff = &ring->buff_ring[ring->sw_tail];
 
-		if (buff->is_gso) {
-			txd->ctl |= HW_ATL_B0_TXD_CTL_CMD_TCP;
+		if (buff->is_gso_tcp || buff->is_gso_udp) {
+			if (buff->is_gso_tcp)
+				txd->ctl |= HW_ATL_B0_TXD_CTL_CMD_TCP;
 			txd->ctl |= HW_ATL_B0_TXD_CTL_DESC_TYPE_TXC;
 			txd->ctl |= (buff->len_l3 << 31) |
 				    (buff->len_l2 << 24);
@@ -554,7 +557,7 @@ static int hw_atl_b0_hw_ring_tx_xmit(struct aq_hw_s *self,
 			txd->ctl |= buff->vlan_tx_tag << 4;
 			is_vlan = true;
 		}
-		if (!buff->is_gso && !buff->is_vlan) {
+		if (!buff->is_gso_tcp && !buff->is_gso_udp && !buff->is_vlan) {
 			buff_pa_len = buff->len;
 
 			txd->buf_addr = buff->pa;