amd-xgbe: Add support for VXLAN offload capabilities

The hardware has the capability to perform checksum offload support
(both Tx and Rx) and TSO support for VXLAN packets. Add the support
required to enable this.

The hardware can only support a single VXLAN port for offload. If more
than one VXLAN port is added then the offload capabilities have to be
disabled and can no longer be advertised.

Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 363 insertions, 2 deletions

diff --git a/drivers/net/ethernet/amd/xgbe/xgbe-drv.c b/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
index 7498bb81f918..608693d11bd7 100644
--- a/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
+++ b/drivers/net/ethernet/amd/xgbe/xgbe-drv.c
@@ -124,6 +124,7 @@
 #include <linux/if_ether.h>
 #include <linux/net_tstamp.h>
 #include <linux/phy.h>
+#include <net/vxlan.h>
 
 #include "xgbe.h"
 #include "xgbe-common.h"
@@ -756,6 +757,7 @@ void xgbe_get_all_hw_features(struct xgbe_prv_data *pdata)
 					      ADDMACADRSEL);
 	hw_feat->ts_src      = XGMAC_GET_BITS(mac_hfr0, MAC_HWF0R, TSSTSSEL);
 	hw_feat->sa_vlan_ins = XGMAC_GET_BITS(mac_hfr0, MAC_HWF0R, SAVLANINS);
+	hw_feat->vxn         = XGMAC_GET_BITS(mac_hfr0, MAC_HWF0R, VXN);
 
 	/* Hardware feature register 1 */
 	hw_feat->rx_fifo_size  = XGMAC_GET_BITS(mac_hfr1, MAC_HWF1R,
@@ -860,6 +862,8 @@ void xgbe_get_all_hw_features(struct xgbe_prv_data *pdata)
 			(hw_feat->ts_src == 3) ? "internal/external" : "n/a");
 		dev_dbg(pdata->dev, "  SA/VLAN insertion         : %s\n",
 			hw_feat->sa_vlan_ins ? "yes" : "no");
+		dev_dbg(pdata->dev, "  VXLAN/NVGRE support       : %s\n",
+			hw_feat->vxn ? "yes" : "no");
 
 		/* Hardware feature register 1 */
 		dev_dbg(pdata->dev, "  RX fifo size              : %u\n",
@@ -903,6 +907,116 @@ void xgbe_get_all_hw_features(struct xgbe_prv_data *pdata)
 	}
 }
 
+static void xgbe_disable_vxlan_offloads(struct xgbe_prv_data *pdata)
+{
+	struct net_device *netdev = pdata->netdev;
+
+	if (!pdata->vxlan_offloads_set)
+		return;
+
+	netdev_info(netdev, "disabling VXLAN offloads\n");
+
+	netdev->hw_enc_features &= ~(NETIF_F_SG |
+				     NETIF_F_IP_CSUM |
+				     NETIF_F_IPV6_CSUM |
+				     NETIF_F_RXCSUM |
+				     NETIF_F_TSO |
+				     NETIF_F_TSO6 |
+				     NETIF_F_GRO |
+				     NETIF_F_GSO_UDP_TUNNEL |
+				     NETIF_F_GSO_UDP_TUNNEL_CSUM);
+
+	netdev->features &= ~(NETIF_F_GSO_UDP_TUNNEL |
+			      NETIF_F_GSO_UDP_TUNNEL_CSUM);
+
+	pdata->vxlan_offloads_set = 0;
+}
+
+static void xgbe_disable_vxlan_hw(struct xgbe_prv_data *pdata)
+{
+	if (!pdata->vxlan_port_set)
+		return;
+
+	pdata->hw_if.disable_vxlan(pdata);
+
+	pdata->vxlan_port_set = 0;
+	pdata->vxlan_port = 0;
+}
+
+static void xgbe_disable_vxlan_accel(struct xgbe_prv_data *pdata)
+{
+	xgbe_disable_vxlan_offloads(pdata);
+
+	xgbe_disable_vxlan_hw(pdata);
+}
+
+static void xgbe_enable_vxlan_offloads(struct xgbe_prv_data *pdata)
+{
+	struct net_device *netdev = pdata->netdev;
+
+	if (pdata->vxlan_offloads_set)
+		return;
+
+	netdev_info(netdev, "enabling VXLAN offloads\n");
+
+	netdev->hw_enc_features |= NETIF_F_SG |
+				   NETIF_F_IP_CSUM |
+				   NETIF_F_IPV6_CSUM |
+				   NETIF_F_RXCSUM |
+				   NETIF_F_TSO |
+				   NETIF_F_TSO6 |
+				   NETIF_F_GRO |
+				   pdata->vxlan_features;
+
+	netdev->features |= pdata->vxlan_features;
+
+	pdata->vxlan_offloads_set = 1;
+}
+
+static void xgbe_enable_vxlan_hw(struct xgbe_prv_data *pdata)
+{
+	struct xgbe_vxlan_data *vdata;
+
+	if (pdata->vxlan_port_set)
+		return;
+
+	if (list_empty(&pdata->vxlan_ports))
+		return;
+
+	vdata = list_first_entry(&pdata->vxlan_ports,
+				 struct xgbe_vxlan_data, list);
+
+	pdata->vxlan_port_set = 1;
+	pdata->vxlan_port = be16_to_cpu(vdata->port);
+
+	pdata->hw_if.enable_vxlan(pdata);
+}
+
+static void xgbe_enable_vxlan_accel(struct xgbe_prv_data *pdata)
+{
+	/* VXLAN acceleration desired? */
+	if (!pdata->vxlan_features)
+		return;
+
+	/* VXLAN acceleration possible? */
+	if (pdata->vxlan_force_disable)
+		return;
+
+	xgbe_enable_vxlan_hw(pdata);
+
+	xgbe_enable_vxlan_offloads(pdata);
+}
+
+static void xgbe_reset_vxlan_accel(struct xgbe_prv_data *pdata)
+{
+	xgbe_disable_vxlan_hw(pdata);
+
+	if (pdata->vxlan_features)
+		xgbe_enable_vxlan_offloads(pdata);
+
+	pdata->vxlan_force_disable = 0;
+}
+
 static void xgbe_napi_enable(struct xgbe_prv_data *pdata, unsigned int add)
 {
 	struct xgbe_channel *channel;
@@ -1226,6 +1340,8 @@ static int xgbe_start(struct xgbe_prv_data *pdata)
 	hw_if->enable_tx(pdata);
 	hw_if->enable_rx(pdata);
 
+	udp_tunnel_get_rx_info(netdev);
+
 	netif_tx_start_all_queues(netdev);
 
 	xgbe_start_timers(pdata);
@@ -1267,6 +1383,8 @@ static void xgbe_stop(struct xgbe_prv_data *pdata)
 	xgbe_stop_timers(pdata);
 	flush_workqueue(pdata->dev_workqueue);
 
+	xgbe_reset_vxlan_accel(pdata);
+
 	hw_if->disable_tx(pdata);
 	hw_if->disable_rx(pdata);
 
@@ -1555,10 +1673,18 @@ static int xgbe_prep_tso(struct sk_buff *skb, struct xgbe_packet_data *packet)
 	if (ret)
 		return ret;
 
-	packet->header_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
-	packet->tcp_header_len = tcp_hdrlen(skb);
+	if (XGMAC_GET_BITS(packet->attributes, TX_PACKET_ATTRIBUTES, VXLAN)) {
+		packet->header_len = skb_inner_transport_offset(skb) +
+				     inner_tcp_hdrlen(skb);
+		packet->tcp_header_len = inner_tcp_hdrlen(skb);
+	} else {
+		packet->header_len = skb_transport_offset(skb) +
+				     tcp_hdrlen(skb);
+		packet->tcp_header_len = tcp_hdrlen(skb);
+	}
 	packet->tcp_payload_len = skb->len - packet->header_len;
 	packet->mss = skb_shinfo(skb)->gso_size;
+
 	DBGPR("  packet->header_len=%u\n", packet->header_len);
 	DBGPR("  packet->tcp_header_len=%u, packet->tcp_payload_len=%u\n",
 	      packet->tcp_header_len, packet->tcp_payload_len);
@@ -1573,6 +1699,49 @@ static int xgbe_prep_tso(struct sk_buff *skb, struct xgbe_packet_data *packet)
 	return 0;
 }
 
+static bool xgbe_is_vxlan(struct xgbe_prv_data *pdata, struct sk_buff *skb)
+{
+	struct xgbe_vxlan_data *vdata;
+
+	if (pdata->vxlan_force_disable)
+		return false;
+
+	if (!skb->encapsulation)
+		return false;
+
+	if (skb->ip_summed != CHECKSUM_PARTIAL)
+		return false;
+
+	switch (skb->protocol) {
+	case htons(ETH_P_IP):
+		if (ip_hdr(skb)->protocol != IPPROTO_UDP)
+			return false;
+		break;
+
+	case htons(ETH_P_IPV6):
+		if (ipv6_hdr(skb)->nexthdr != IPPROTO_UDP)
+			return false;
+		break;
+
+	default:
+		return false;
+	}
+
+	/* See if we have the UDP port in our list */
+	list_for_each_entry(vdata, &pdata->vxlan_ports, list) {
+		if ((skb->protocol == htons(ETH_P_IP)) &&
+		    (vdata->sa_family == AF_INET) &&
+		    (vdata->port == udp_hdr(skb)->dest))
+			return true;
+		else if ((skb->protocol == htons(ETH_P_IPV6)) &&
+			 (vdata->sa_family == AF_INET6) &&
+			 (vdata->port == udp_hdr(skb)->dest))
+			return true;
+	}
+
+	return false;
+}
+
 static int xgbe_is_tso(struct sk_buff *skb)
 {
 	if (skb->ip_summed != CHECKSUM_PARTIAL)
@@ -1621,6 +1790,10 @@ static void xgbe_packet_info(struct xgbe_prv_data *pdata,
 		XGMAC_SET_BITS(packet->attributes, TX_PACKET_ATTRIBUTES,
 			       CSUM_ENABLE, 1);
 
+	if (xgbe_is_vxlan(pdata, skb))
+		XGMAC_SET_BITS(packet->attributes, TX_PACKET_ATTRIBUTES,
+			       VXLAN, 1);
+
 	if (skb_vlan_tag_present(skb)) {
 		/* VLAN requires an extra descriptor if tag is different */
 		if (skb_vlan_tag_get(skb) != ring->tx.cur_vlan_ctag)
@@ -2050,18 +2223,83 @@ static int xgbe_setup_tc(struct net_device *netdev, enum tc_setup_type type,
 	return 0;
 }
 
+static netdev_features_t xgbe_fix_features(struct net_device *netdev,
+					   netdev_features_t features)
+{
+	struct xgbe_prv_data *pdata = netdev_priv(netdev);
+	netdev_features_t vxlan_base, vxlan_mask;
+
+	vxlan_base = NETIF_F_GSO_UDP_TUNNEL | NETIF_F_RX_UDP_TUNNEL_PORT;
+	vxlan_mask = vxlan_base | NETIF_F_GSO_UDP_TUNNEL_CSUM;
+
+	pdata->vxlan_features = features & vxlan_mask;
+
+	/* Only fix VXLAN-related features */
+	if (!pdata->vxlan_features)
+		return features;
+
+	/* If VXLAN isn't supported then clear any features:
+	 *   This is needed because NETIF_F_RX_UDP_TUNNEL_PORT gets
+	 *   automatically set if ndo_udp_tunnel_add is set.
+	 */
+	if (!pdata->hw_feat.vxn)
+		return features & ~vxlan_mask;
+
+	/* VXLAN CSUM requires VXLAN base */
+	if ((features & NETIF_F_GSO_UDP_TUNNEL_CSUM) &&
+	    !(features & NETIF_F_GSO_UDP_TUNNEL)) {
+		netdev_notice(netdev,
+			      "forcing tx udp tunnel support\n");
+		features |= NETIF_F_GSO_UDP_TUNNEL;
+	}
+
+	/* Can't do one without doing the other */
+	if ((features & vxlan_base) != vxlan_base) {
+		netdev_notice(netdev,
+			      "forcing both tx and rx udp tunnel support\n");
+		features |= vxlan_base;
+	}
+
+	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
+		if (!(features & NETIF_F_GSO_UDP_TUNNEL_CSUM)) {
+			netdev_notice(netdev,
+				      "forcing tx udp tunnel checksumming on\n");
+			features |= NETIF_F_GSO_UDP_TUNNEL_CSUM;
+		}
+	} else {
+		if (features & NETIF_F_GSO_UDP_TUNNEL_CSUM) {
+			netdev_notice(netdev,
+				      "forcing tx udp tunnel checksumming off\n");
+			features &= ~NETIF_F_GSO_UDP_TUNNEL_CSUM;
+		}
+	}
+
+	pdata->vxlan_features = features & vxlan_mask;
+
+	/* Adjust UDP Tunnel based on current state */
+	if (pdata->vxlan_force_disable) {
+		netdev_notice(netdev,
+			      "VXLAN acceleration disabled, turning off udp tunnel features\n");
+		features &= ~vxlan_mask;
+	}
+
+	return features;
+}
+
 static int xgbe_set_features(struct net_device *netdev,
 			     netdev_features_t features)
 {
 	struct xgbe_prv_data *pdata = netdev_priv(netdev);
 	struct xgbe_hw_if *hw_if = &pdata->hw_if;
 	netdev_features_t rxhash, rxcsum, rxvlan, rxvlan_filter;
+	netdev_features_t udp_tunnel;
 	int ret = 0;
 
 	rxhash = pdata->netdev_features & NETIF_F_RXHASH;
 	rxcsum = pdata->netdev_features & NETIF_F_RXCSUM;
 	rxvlan = pdata->netdev_features & NETIF_F_HW_VLAN_CTAG_RX;
 	rxvlan_filter = pdata->netdev_features & NETIF_F_HW_VLAN_CTAG_FILTER;
+	udp_tunnel = pdata->netdev_features & NETIF_F_GSO_UDP_TUNNEL;
 
 	if ((features & NETIF_F_RXHASH) && !rxhash)
 		ret = hw_if->enable_rss(pdata);
@@ -2085,6 +2323,11 @@ static int xgbe_set_features(struct net_device *netdev,
 	else if (!(features & NETIF_F_HW_VLAN_CTAG_FILTER) && rxvlan_filter)
 		hw_if->disable_rx_vlan_filtering(pdata);
 
+	if ((features & NETIF_F_GSO_UDP_TUNNEL) && !udp_tunnel)
+		xgbe_enable_vxlan_accel(pdata);
+	else if (!(features & NETIF_F_GSO_UDP_TUNNEL) && udp_tunnel)
+		xgbe_disable_vxlan_accel(pdata);
+
 	pdata->netdev_features = features;
 
 	DBGPR("<--xgbe_set_features\n");
@@ -2092,6 +2335,111 @@ static int xgbe_set_features(struct net_device *netdev,
 	return 0;
 }
 
+static void xgbe_udp_tunnel_add(struct net_device *netdev,
+				struct udp_tunnel_info *ti)
+{
+	struct xgbe_prv_data *pdata = netdev_priv(netdev);
+	struct xgbe_vxlan_data *vdata;
+
+	if (!pdata->hw_feat.vxn)
+		return;
+
+	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
+		return;
+
+	pdata->vxlan_port_count++;
+
+	netif_dbg(pdata, drv, netdev,
+		  "adding VXLAN tunnel, family=%hx/port=%hx\n",
+		  ti->sa_family, be16_to_cpu(ti->port));
+
+	if (pdata->vxlan_force_disable)
+		return;
+
+	vdata = kzalloc(sizeof(*vdata), GFP_ATOMIC);
+	if (!vdata) {
+		/* Can no longer properly track VXLAN ports */
+		pdata->vxlan_force_disable = 1;
+		netif_dbg(pdata, drv, netdev,
+			  "internal error, disabling VXLAN accelerations\n");
+
+		xgbe_disable_vxlan_accel(pdata);
+
+		return;
+	}
+	vdata->sa_family = ti->sa_family;
+	vdata->port = ti->port;
+
+	list_add_tail(&vdata->list, &pdata->vxlan_ports);
+
+	/* First port added? */
+	if (pdata->vxlan_port_count == 1) {
+		xgbe_enable_vxlan_accel(pdata);
+
+		return;
+	}
+}
+
+static void xgbe_udp_tunnel_del(struct net_device *netdev,
+				struct udp_tunnel_info *ti)
+{
+	struct xgbe_prv_data *pdata = netdev_priv(netdev);
+	struct xgbe_vxlan_data *vdata;
+
+	if (!pdata->hw_feat.vxn)
+		return;
+
+	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
+		return;
+
+	netif_dbg(pdata, drv, netdev,
+		  "deleting VXLAN tunnel, family=%hx/port=%hx\n",
+		  ti->sa_family, be16_to_cpu(ti->port));
+
+	/* Don't need safe version since loop terminates with deletion */
+	list_for_each_entry(vdata, &pdata->vxlan_ports, list) {
+		if (vdata->sa_family != ti->sa_family)
+			continue;
+
+		if (vdata->port != ti->port)
+			continue;
+
+		list_del(&vdata->list);
+		kfree(vdata);
+
+		break;
+	}
+
+	pdata->vxlan_port_count--;
+	if (!pdata->vxlan_port_count) {
+		xgbe_reset_vxlan_accel(pdata);
+
+		return;
+	}
+
+	if (pdata->vxlan_force_disable)
+		return;
+
+	/* See if VXLAN tunnel id needs to be changed */
+	vdata = list_first_entry(&pdata->vxlan_ports,
+				 struct xgbe_vxlan_data, list);
+	if (pdata->vxlan_port == be16_to_cpu(vdata->port))
+		return;
+
+	pdata->vxlan_port = be16_to_cpu(vdata->port);
+	pdata->hw_if.set_vxlan_id(pdata);
+}
+
+static netdev_features_t xgbe_features_check(struct sk_buff *skb,
+					     struct net_device *netdev,
+					     netdev_features_t features)
+{
+	features = vlan_features_check(skb, features);
+	features = vxlan_features_check(skb, features);
+
+	return features;
+}
+
 static const struct net_device_ops xgbe_netdev_ops = {
 	.ndo_open		= xgbe_open,
 	.ndo_stop		= xgbe_close,
@@ -2109,7 +2457,11 @@ static const struct net_device_ops xgbe_netdev_ops = {
 	.ndo_poll_controller	= xgbe_poll_controller,
 #endif
 	.ndo_setup_tc		= xgbe_setup_tc,
+	.ndo_fix_features	= xgbe_fix_features,
 	.ndo_set_features	= xgbe_set_features,
+	.ndo_udp_tunnel_add	= xgbe_udp_tunnel_add,
+	.ndo_udp_tunnel_del	= xgbe_udp_tunnel_del,
+	.ndo_features_check	= xgbe_features_check,
 };
 
 const struct net_device_ops *xgbe_get_netdev_ops(void)
@@ -2422,6 +2774,15 @@ skip_data:
 			skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 		if (XGMAC_GET_BITS(packet->attributes,
+				   RX_PACKET_ATTRIBUTES, TNP)) {
+			skb->encapsulation = 1;
+
+			if (XGMAC_GET_BITS(packet->attributes,
+					   RX_PACKET_ATTRIBUTES, TNPCSUM_DONE))
+				skb->csum_level = 1;
+		}
+
+		if (XGMAC_GET_BITS(packet->attributes,
 				   RX_PACKET_ATTRIBUTES, VLAN_CTAG))
 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 					       packet->vlan_ctag);