Merge branch 'enic-vxlan-offload'

Govindarajulu Varadarajan says:

====================
enic: add vxlan offload support

This series adds vxlan offload support for enic driver. The first
patch adds vxlan devcmd for configuring vxland offload parameters.
Second patch adds ndo_udp_tunnel_add/del and offload on rx path.
There are to modes in which fw supports vxlan offload.

mode 0: fcoe bit is set for encapsulated packet. fcoe_fc_crc_ok is set
if checksum of csum is ok. This bit is or of ip_csum_ok and
tcp_udp_csum_ok

mode 2: BIT(0) in rss_hash is set if it is encapsulated packet.
        BIT(1) is set if outer_ip_csum_ok/
        BIT(2) is set if outer_tcp_csum_ok

Some hw supports only mode 0, some support mode 0 and 2. Driver gets
the supported modes bitmap using get_supported_feature_ver devcmd
and selects the highest mode both driver and fw supports.

Third patch adds offload support on tx path by adding
enic_features_check().

v2: Order local variable declarations from longest to shortest line,
    on all three patches.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

drivers/net/ethernet/cisco/enic/enic.h

@@ -135,6 +135,11 @@ struct enic_rfs_flw_tbl {
 	struct timer_list rfs_may_expire;
 };
 
+struct vxlan_offload {
+	u16 vxlan_udp_port_number;
+	u8 patch_level;
+};
+
 /* Per-instance private data structure */
 struct enic {
 	struct net_device *netdev;
@@ -175,6 +180,7 @@ struct enic {
 	/* receive queue cache line section */
 	____cacheline_aligned struct vnic_rq rq[ENIC_RQ_MAX];
 	unsigned int rq_count;
+	struct vxlan_offload vxlan;
 	u64 rq_truncated_pkts;
 	u64 rq_bad_fcs;
 	struct napi_struct napi[ENIC_RQ_MAX + ENIC_WQ_MAX];

drivers/net/ethernet/cisco/enic/enic_main.c

@@ -45,6 +45,7 @@
 #endif
 #include <linux/crash_dump.h>
 #include <net/busy_poll.h>
+#include <net/vxlan.h>
 
 #include "cq_enet_desc.h"
 #include "vnic_dev.h"
@@ -176,6 +177,134 @@ static void enic_unset_affinity_hint(struct enic *enic)
 		irq_set_affinity_hint(enic->msix_entry[i].vector, NULL);
 }
 
+static void enic_udp_tunnel_add(struct net_device *netdev,
+				struct udp_tunnel_info *ti)
+{
+	struct enic *enic = netdev_priv(netdev);
+	__be16 port = ti->port;
+	int err;
+
+	spin_lock_bh(&enic->devcmd_lock);
+
+	if (ti->type != UDP_TUNNEL_TYPE_VXLAN) {
+		netdev_info(netdev, "udp_tnl: only vxlan tunnel offload supported");
+		goto error;
+	}
+
+	if (ti->sa_family != AF_INET) {
+		netdev_info(netdev, "vxlan: only IPv4 offload supported");
+		goto error;
+	}
+
+	if (enic->vxlan.vxlan_udp_port_number) {
+		if (ntohs(port) == enic->vxlan.vxlan_udp_port_number)
+			netdev_warn(netdev, "vxlan: udp port already offloaded");
+		else
+			netdev_info(netdev, "vxlan: offload supported for only one UDP port");
+
+		goto error;
+	}
+
+	err = vnic_dev_overlay_offload_cfg(enic->vdev,
+					   OVERLAY_CFG_VXLAN_PORT_UPDATE,
+					   ntohs(port));
+	if (err)
+		goto error;
+
+	err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
+					    enic->vxlan.patch_level);
+	if (err)
+		goto error;
+
+	enic->vxlan.vxlan_udp_port_number = ntohs(port);
+
+	netdev_info(netdev, "vxlan fw-vers-%d: offload enabled for udp port: %d, sa_family: %d ",
+		    (int)enic->vxlan.patch_level, ntohs(port), ti->sa_family);
+
+	goto unlock;
+
+error:
+	netdev_info(netdev, "failed to offload udp port: %d, sa_family: %d, type: %d",
+		    ntohs(port), ti->sa_family, ti->type);
+unlock:
+	spin_unlock_bh(&enic->devcmd_lock);
+}
+
+static void enic_udp_tunnel_del(struct net_device *netdev,
+				struct udp_tunnel_info *ti)
+{
+	struct enic *enic = netdev_priv(netdev);
+	int err;
+
+	spin_lock_bh(&enic->devcmd_lock);
+
+	if ((ti->sa_family != AF_INET) ||
+	    ((ntohs(ti->port) != enic->vxlan.vxlan_udp_port_number)) ||
+	    (ti->type != UDP_TUNNEL_TYPE_VXLAN)) {
+		netdev_info(netdev, "udp_tnl: port:%d, sa_family: %d, type: %d not offloaded",
+			    ntohs(ti->port), ti->sa_family, ti->type);
+		goto unlock;
+	}
+
+	err = vnic_dev_overlay_offload_ctrl(enic->vdev, OVERLAY_FEATURE_VXLAN,
+					    OVERLAY_OFFLOAD_DISABLE);
+	if (err) {
+		netdev_err(netdev, "vxlan: del offload udp port: %d failed",
+			   ntohs(ti->port));
+		goto unlock;
+	}
+
+	enic->vxlan.vxlan_udp_port_number = 0;
+
+	netdev_info(netdev, "vxlan: del offload udp port %d, family %d\n",
+		    ntohs(ti->port), ti->sa_family);
+
+unlock:
+	spin_unlock_bh(&enic->devcmd_lock);
+}
+
+static netdev_features_t enic_features_check(struct sk_buff *skb,
+					     struct net_device *dev,
+					     netdev_features_t features)
+{
+	const struct ethhdr *eth = (struct ethhdr *)skb_inner_mac_header(skb);
+	struct enic *enic = netdev_priv(dev);
+	struct udphdr *udph;
+	u16 port = 0;
+	u16 proto;
+
+	if (!skb->encapsulation)
+		return features;
+
+	features = vxlan_features_check(skb, features);
+
+	/* hardware only supports IPv4 vxlan tunnel */
+	if (vlan_get_protocol(skb) != htons(ETH_P_IP))
+		goto out;
+
+	/* hardware does not support offload of ipv6 inner pkt */
+	if (eth->h_proto != ntohs(ETH_P_IP))
+		goto out;
+
+	proto = ip_hdr(skb)->protocol;
+
+	if (proto == IPPROTO_UDP) {
+		udph = udp_hdr(skb);
+		port = be16_to_cpu(udph->dest);
+	}
+
+	/* HW supports offload of only one UDP port. Remove CSUM and GSO MASK
+	 * for other UDP port tunnels
+	 */
+	if (port  != enic->vxlan.vxlan_udp_port_number)
+		goto out;
+
+	return features;
+
+out:
+	return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
+}
+
 int enic_is_dynamic(struct enic *enic)
 {
 	return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
@@ -504,20 +633,19 @@ static int enic_queue_wq_skb_csum_l4(struct enic *enic, struct vnic_wq *wq,
 	return err;
 }
 
-static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
-				 struct sk_buff *skb, unsigned int mss,
-				 int vlan_tag_insert, unsigned int vlan_tag,
-				 int loopback)
+static void enic_preload_tcp_csum_encap(struct sk_buff *skb)
 {
-	unsigned int frag_len_left = skb_headlen(skb);
-	unsigned int len_left = skb->len - frag_len_left;
-	unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
-	int eop = (len_left == 0);
-	unsigned int len;
-	dma_addr_t dma_addr;
-	unsigned int offset = 0;
-	skb_frag_t *frag;
+	if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
+		inner_ip_hdr(skb)->check = 0;
+		inner_tcp_hdr(skb)->check =
+			~csum_tcpudp_magic(inner_ip_hdr(skb)->saddr,
+					   inner_ip_hdr(skb)->daddr, 0,
+					   IPPROTO_TCP, 0);
+	}
+}
 
+static void enic_preload_tcp_csum(struct sk_buff *skb)
+{
 	/* Preload TCP csum field with IP pseudo hdr calculated
 	 * with IP length set to zero.  HW will later add in length
 	 * to each TCP segment resulting from the TSO.
@@ -531,6 +659,30 @@ static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
 		tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 			&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 	}
+}
+
+static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
+				 struct sk_buff *skb, unsigned int mss,
+				 int vlan_tag_insert, unsigned int vlan_tag,
+				 int loopback)
+{
+	unsigned int frag_len_left = skb_headlen(skb);
+	unsigned int len_left = skb->len - frag_len_left;
+	int eop = (len_left == 0);
+	unsigned int offset = 0;
+	unsigned int hdr_len;
+	dma_addr_t dma_addr;
+	unsigned int len;
+	skb_frag_t *frag;
+
+	if (skb->encapsulation) {
+		hdr_len = skb_inner_transport_header(skb) - skb->data;
+		hdr_len += inner_tcp_hdrlen(skb);
+		enic_preload_tcp_csum_encap(skb);
+	} else {
+		hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
+		enic_preload_tcp_csum(skb);
+	}
 
 	/* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 	 * for the main skb fragment
@@ -579,6 +731,38 @@ static int enic_queue_wq_skb_tso(struct enic *enic, struct vnic_wq *wq,
 	return 0;
 }
 
+static inline int enic_queue_wq_skb_encap(struct enic *enic, struct vnic_wq *wq,
+					  struct sk_buff *skb,
+					  int vlan_tag_insert,
+					  unsigned int vlan_tag, int loopback)
+{
+	unsigned int head_len = skb_headlen(skb);
+	unsigned int len_left = skb->len - head_len;
+	/* Hardware will overwrite the checksum fields, calculating from
+	 * scratch and ignoring the value placed by software.
+	 * Offload mode = 00
+	 * mss[2], mss[1], mss[0] bits are set
+	 */
+	unsigned int mss_or_csum = 7;
+	int eop = (len_left == 0);
+	dma_addr_t dma_addr;
+	int err = 0;
+
+	dma_addr = pci_map_single(enic->pdev, skb->data, head_len,
+				  PCI_DMA_TODEVICE);
+	if (unlikely(enic_dma_map_check(enic, dma_addr)))
+		return -ENOMEM;
+
+	enic_queue_wq_desc_ex(wq, skb, dma_addr, head_len, mss_or_csum, 0,
+			      vlan_tag_insert, vlan_tag,
+			      WQ_ENET_OFFLOAD_MODE_CSUM, eop, 1 /* SOP */, eop,
+			      loopback);
+	if (!eop)
+		err = enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
+
+	return err;
+}
+
 static inline void enic_queue_wq_skb(struct enic *enic,
 	struct vnic_wq *wq, struct sk_buff *skb)
 {
@@ -601,6 +785,9 @@ static inline void enic_queue_wq_skb(struct enic *enic,
 		err = enic_queue_wq_skb_tso(enic, wq, skb, mss,
 					    vlan_tag_insert, vlan_tag,
 					    loopback);
+	else if (skb->encapsulation)
+		err = enic_queue_wq_skb_encap(enic, wq, skb, vlan_tag_insert,
+					      vlan_tag, loopback);
 	else if	(skb->ip_summed == CHECKSUM_PARTIAL)
 		err = enic_queue_wq_skb_csum_l4(enic, wq, skb, vlan_tag_insert,
 						vlan_tag, loopback);
@@ -1113,6 +1300,7 @@ static void enic_rq_indicate_buf(struct vnic_rq *rq,
 	u8 packet_error;
 	u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
 	u32 rss_hash;
+	bool outer_csum_ok = true, encap = false;
 
 	if (skipped)
 		return;
@@ -1161,7 +1349,8 @@ static void enic_rq_indicate_buf(struct vnic_rq *rq,
 		skb_put(skb, bytes_written);
 		skb->protocol = eth_type_trans(skb, netdev);
 		skb_record_rx_queue(skb, q_number);
-		if (netdev->features & NETIF_F_RXHASH) {
+		if ((netdev->features & NETIF_F_RXHASH) && rss_hash &&
+		    (type == 3)) {
 			switch (rss_type) {
 			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv4:
 			case CQ_ENET_RQ_DESC_RSS_TYPE_TCP_IPv6:
@@ -1175,15 +1364,39 @@ static void enic_rq_indicate_buf(struct vnic_rq *rq,
 				break;
 			}
 		}
+		if (enic->vxlan.vxlan_udp_port_number) {
+			switch (enic->vxlan.patch_level) {
+			case 0:
+				if (fcoe) {
+					encap = true;
+					outer_csum_ok = fcoe_fc_crc_ok;
+				}
+				break;
+			case 2:
+				if ((type == 7) &&
+				    (rss_hash & BIT(0))) {
+					encap = true;
+					outer_csum_ok = (rss_hash & BIT(1)) &&
+							(rss_hash & BIT(2));
+				}
+				break;
+			}
+		}
 
 		/* Hardware does not provide whole packet checksum. It only
 		 * provides pseudo checksum. Since hw validates the packet
 		 * checksum but not provide us the checksum value. use
 		 * CHECSUM_UNNECESSARY.
+		 *
+		 * In case of encap pkt tcp_udp_csum_ok/tcp_udp_csum_ok is
+		 * inner csum_ok. outer_csum_ok is set by hw when outer udp
+		 * csum is correct or is zero.
 		 */
-		if ((netdev->features & NETIF_F_RXCSUM) && tcp_udp_csum_ok &&
-		    ipv4_csum_ok)
+		if ((netdev->features & NETIF_F_RXCSUM) && !csum_not_calc &&
+		    tcp_udp_csum_ok && ipv4_csum_ok && outer_csum_ok) {
 			skb->ip_summed = CHECKSUM_UNNECESSARY;
+			skb->csum_level = encap;
+		}
 
 		if (vlan_stripped)
 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
@@ -2285,6 +2498,9 @@ static const struct net_device_ops enic_netdev_dynamic_ops = {
 #ifdef CONFIG_RFS_ACCEL
 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
 #endif
+	.ndo_udp_tunnel_add	= enic_udp_tunnel_add,
+	.ndo_udp_tunnel_del	= enic_udp_tunnel_del,
+	.ndo_features_check	= enic_features_check,
 };
 
 static const struct net_device_ops enic_netdev_ops = {
@@ -2308,6 +2524,9 @@ static const struct net_device_ops enic_netdev_ops = {
 #ifdef CONFIG_RFS_ACCEL
 	.ndo_rx_flow_steer	= enic_rx_flow_steer,
 #endif
+	.ndo_udp_tunnel_add	= enic_udp_tunnel_add,
+	.ndo_udp_tunnel_del	= enic_udp_tunnel_del,
+	.ndo_features_check	= enic_features_check,
 };
 
 static void enic_dev_deinit(struct enic *enic)
@@ -2683,6 +2902,39 @@ static int enic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 		netdev->hw_features |= NETIF_F_RXHASH;
 	if (ENIC_SETTING(enic, RXCSUM))
 		netdev->hw_features |= NETIF_F_RXCSUM;
+	if (ENIC_SETTING(enic, VXLAN)) {
+		u64 patch_level;
+
+		netdev->hw_enc_features |= NETIF_F_RXCSUM		|
+					   NETIF_F_TSO			|
+					   NETIF_F_TSO_ECN		|
+					   NETIF_F_GSO_UDP_TUNNEL	|
+					   NETIF_F_HW_CSUM		|
+					   NETIF_F_GSO_UDP_TUNNEL_CSUM;
+		netdev->hw_features |= netdev->hw_enc_features;
+		/* get bit mask from hw about supported offload bit level
+		 * BIT(0) = fw supports patch_level 0
+		 *	    fcoe bit = encap
+		 *	    fcoe_fc_crc_ok = outer csum ok
+		 * BIT(1) = always set by fw
+		 * BIT(2) = fw supports patch_level 2
+		 *	    BIT(0) in rss_hash = encap
+		 *	    BIT(1,2) in rss_hash = outer_ip_csum_ok/
+		 *				   outer_tcp_csum_ok
+		 * used in enic_rq_indicate_buf
+		 */
+		err = vnic_dev_get_supported_feature_ver(enic->vdev,
+							 VIC_FEATURE_VXLAN,
+							 &patch_level);
+		if (err)
+			patch_level = 0;
+		/* mask bits that are supported by driver
+		 */
+		patch_level &= BIT_ULL(0) | BIT_ULL(2);
+		patch_level = fls(patch_level);
+		patch_level = patch_level ? patch_level - 1 : 0;
+		enic->vxlan.patch_level = patch_level;
+	}
 
 	netdev->features |= netdev->hw_features;
 	netdev->vlan_features |= netdev->features;

drivers/net/ethernet/cisco/enic/vnic_dev.c

@@ -1247,3 +1247,37 @@ int vnic_dev_classifier(struct vnic_dev *vdev, u8 cmd, u16 *entry,
 
 	return ret;
 }
+
+int vnic_dev_overlay_offload_ctrl(struct vnic_dev *vdev, u8 overlay, u8 config)
+{
+	u64 a0 = overlay;
+	u64 a1 = config;
+	int wait = 1000;
+
+	return vnic_dev_cmd(vdev, CMD_OVERLAY_OFFLOAD_CTRL, &a0, &a1, wait);
+}
+
+int vnic_dev_overlay_offload_cfg(struct vnic_dev *vdev, u8 overlay,
+				 u16 vxlan_udp_port_number)
+{
+	u64 a1 = vxlan_udp_port_number;
+	u64 a0 = overlay;
+	int wait = 1000;
+
+	return vnic_dev_cmd(vdev, CMD_OVERLAY_OFFLOAD_CFG, &a0, &a1, wait);
+}
+
+int vnic_dev_get_supported_feature_ver(struct vnic_dev *vdev, u8 feature,
+				       u64 *supported_versions)
+{
+	u64 a0 = feature;
+	int wait = 1000;
+	u64 a1 = 0;
+	int ret;
+
+	ret = vnic_dev_cmd(vdev, CMD_GET_SUPP_FEATURE_VER, &a0, &a1, wait);
+	if (!ret)
+		*supported_versions = a0;
+
+	return ret;
+}

drivers/net/ethernet/cisco/enic/vnic_dev.h

@@ -179,5 +179,10 @@ int vnic_dev_set_mac_addr(struct vnic_dev *vdev, u8 *mac_addr);
 int vnic_dev_classifier(struct vnic_dev *vdev, u8 cmd, u16 *entry,
 			struct filter *data);
 int vnic_devcmd_init(struct vnic_dev *vdev);
+int vnic_dev_overlay_offload_ctrl(struct vnic_dev *vdev, u8 overlay, u8 config);
+int vnic_dev_overlay_offload_cfg(struct vnic_dev *vdev, u8 overlay,
+				 u16 vxlan_udp_port_number);
+int vnic_dev_get_supported_feature_ver(struct vnic_dev *vdev, u8 feature,
+				       u64 *supported_versions);
 
 #endif /* _VNIC_DEV_H_ */

drivers/net/ethernet/cisco/enic/vnic_devcmd.h

@@ -406,6 +406,31 @@ enum vnic_devcmd_cmd {
 	 * in: (u32) a0=Queue Pair number
 	 */
 	CMD_QP_STATS_CLEAR = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 63),
+
+	/* Use this devcmd for agreeing on the highest common version supported
+	 * by both driver and fw for features who need such a facility.
+	 * in:	(u64) a0 = feature (driver requests for the supported versions
+	 *	on this feature)
+	 * out: (u64) a0 = bitmap of all supported versions for that feature
+	 */
+	CMD_GET_SUPP_FEATURE_VER = _CMDC(_CMD_DIR_RW, _CMD_VTYPE_ENET, 69),
+
+	/* Control (Enable/Disable) overlay offloads on the given vnic
+	 * in: (u8) a0 = OVERLAY_FEATURE_NVGRE : NVGRE
+	 *	    a0 = OVERLAY_FEATURE_VXLAN : VxLAN
+	 * in: (u8) a1 = OVERLAY_OFFLOAD_ENABLE : Enable or
+	 *	    a1 = OVERLAY_OFFLOAD_DISABLE : Disable or
+	 *	    a1 = OVERLAY_OFFLOAD_ENABLE_V2 : Enable with version 2
+	 */
+	CMD_OVERLAY_OFFLOAD_CTRL = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 72),
+
+	/* Configuration of overlay offloads feature on a given vNIC
+	 * in: (u8) a0 = DEVCMD_OVERLAY_NVGRE : NVGRE
+	 *	    a0 = DEVCMD_OVERLAY_VXLAN : VxLAN
+	 * in: (u8) a1 = VXLAN_PORT_UPDATE : VxLAN
+	 * in: (u16) a2 = unsigned short int port information
+	 */
+	CMD_OVERLAY_OFFLOAD_CFG = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 73),
 };
 
 /* CMD_ENABLE2 flags */
@@ -657,4 +682,30 @@ struct devcmd2_result {
 #define DEVCMD2_RING_SIZE	32
 #define DEVCMD2_DESC_SIZE	128
 
+enum overlay_feature_t {
+	OVERLAY_FEATURE_NVGRE = 1,
+	OVERLAY_FEATURE_VXLAN,
+	OVERLAY_FEATURE_MAX,
+};
+
+enum overlay_ofld_cmd {
+	OVERLAY_OFFLOAD_ENABLE,
+	OVERLAY_OFFLOAD_DISABLE,
+	OVERLAY_OFFLOAD_ENABLE_P2,
+	OVERLAY_OFFLOAD_MAX,
+};
+
+#define OVERLAY_CFG_VXLAN_PORT_UPDATE	0
+
+/* Use this enum to get the supported versions for each of these features
+ * If you need to use the devcmd_get_supported_feature_version(), add
+ * the new feature into this enum and install function handler in devcmd.c
+ */
+enum vic_feature_t {
+	VIC_FEATURE_VXLAN,
+	VIC_FEATURE_RDMA,
+	VIC_FEATURE_VXLAN_PATCH,
+	VIC_FEATURE_MAX,
+};
+
 #endif /* _VNIC_DEVCMD_H_ */

drivers/net/ethernet/cisco/enic/vnic_enet.h

@@ -48,6 +48,7 @@ struct vnic_enet_config {
 #define VENETF_RSSHASH_IPV6_EX	0x200	/* Hash on IPv6 extended fields */
 #define VENETF_RSSHASH_TCPIPV6_EX 0x400	/* Hash on TCP + IPv6 ext. fields */
 #define VENETF_LOOP		0x800	/* Loopback enabled */
+#define VENETF_VXLAN		0x10000	/* VxLAN offload */
 
 #define VENET_INTR_TYPE_MIN	0	/* Timer specs min interrupt spacing */
 #define VENET_INTR_TYPE_IDLE	1	/* Timer specs idle time before irq */