From: Xueming Li <xuemingl@mellanox.com>
Date: Sun, 8 Apr 2018 12:41:20 +0000 (+0800)
Subject: net/mlx5: support generic tunnel offloading
X-Git-Url: http://git.droids-corp.org/?a=commitdiff_plain;h=5f8ba81c4228;p=dpdk.git

net/mlx5: support generic tunnel offloading

This commit adds support for generic tunnel TSO and checksum offload.
PMD will compute the inner/outer headers offset according to the
mbuf fields. Hardware will do calculation based on offsets and types.

Signed-off-by: Xueming Li <xuemingl@mellanox.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
---

diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index b710a10f53..ead66a40c3 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -111,6 +111,11 @@ mlx5_autoconf.h.new: $(RTE_SDK)/buildtools/auto-config-h.sh
 		infiniband/verbs.h \
 		enum IBV_WQ_FLAG_RX_END_PADDING \
 		$(AUTOCONF_OUTPUT)
+	$Q sh -- '$<' '$@' \
+		HAVE_IBV_MLX5_MOD_SWP \
+		infiniband/mlx5dv.h \
+		enum MLX5DV_CONTEXT_MASK_SWP \
+		$(AUTOCONF_OUTPUT)
 	$Q sh -- '$<' '$@' \
 		HAVE_IBV_MLX5_MOD_MPW \
 		infiniband/mlx5dv.h \
diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 887924d076..5f7ec12b86 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -621,6 +621,7 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 	unsigned int mps;
 	unsigned int cqe_comp;
 	unsigned int tunnel_en = 0;
+	unsigned int swp = 0;
 	int idx;
 	int i;
 	struct mlx5dv_context attrs_out = {0};
@@ -696,6 +697,9 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 	}
 	ibv_dev = list[i];
 	DRV_LOG(DEBUG, "device opened");
+#ifdef HAVE_IBV_MLX5_MOD_SWP
+	attrs_out.comp_mask |= MLX5DV_CONTEXT_MASK_SWP;
+#endif
 	/*
 	 * Multi-packet send is supported by ConnectX-4 Lx PF as well
 	 * as all ConnectX-5 devices.
@@ -716,6 +720,11 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 		DRV_LOG(DEBUG, "MPW isn't supported");
 		mps = MLX5_MPW_DISABLED;
 	}
+#ifdef HAVE_IBV_MLX5_MOD_SWP
+	if (attrs_out.comp_mask | MLX5DV_CONTEXT_MASK_SWP)
+		swp = attrs_out.sw_parsing_caps.sw_parsing_offloads;
+	DRV_LOG(DEBUG, "SWP support: %u", swp);
+#endif
 	if (RTE_CACHE_LINE_SIZE == 128 &&
 	    !(attrs_out.flags & MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP))
 		cqe_comp = 0;
@@ -763,6 +772,7 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 			.txqs_inline = MLX5_ARG_UNSET,
 			.inline_max_packet_sz = MLX5_ARG_UNSET,
 			.vf_nl_en = 1,
+			.swp = !!swp,
 		};
 
 		len = snprintf(name, sizeof(name), PCI_PRI_FMT,
@@ -1227,8 +1237,10 @@ RTE_INIT(rte_mlx5_pmd_init);
 static void
 rte_mlx5_pmd_init(void)
 {
-	/* Build the static table for ptype conversion. */
+	/* Build the static tables for Verbs conversion. */
 	mlx5_set_ptype_table();
+	mlx5_set_cksum_table();
+	mlx5_set_swp_types_table();
 	/*
 	 * RDMAV_HUGEPAGES_SAFE tells ibv_fork_init() we intend to use
 	 * huge pages. Calling ibv_fork_init() during init allows
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 64f025d22d..236e832064 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -89,6 +89,7 @@ struct mlx5_dev_config {
 	unsigned int rx_vec_en:1; /* Rx vector is enabled. */
 	unsigned int mpw_hdr_dseg:1; /* Enable DSEGs in the title WQEBB. */
 	unsigned int vf_nl_en:1; /* Enable Netlink requests in VF mode. */
+	unsigned int swp:1; /* Tx generic tunnel checksum and TSO offload. */
 	unsigned int tso_max_payload_sz; /* Maximum TCP payload for TSO. */
 	unsigned int ind_table_max_size; /* Maximum indirection table size. */
 	int txq_inline; /* Maximum packet size for inlining. */
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 8741df1b34..746b94f734 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -1090,11 +1090,14 @@ mlx5_select_tx_function(struct rte_eth_dev *dev)
 	int tso = !!(tx_offloads & (DEV_TX_OFFLOAD_TCP_TSO |
 				    DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
 				    DEV_TX_OFFLOAD_GRE_TNL_TSO));
+	int swp = !!(tx_offloads & (DEV_TX_OFFLOAD_IP_TNL_TSO |
+				    DEV_TX_OFFLOAD_UDP_TNL_TSO |
+				    DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM));
 	int vlan_insert = !!(tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT);
 
 	assert(priv != NULL);
 	/* Select appropriate TX function. */
-	if (vlan_insert || tso)
+	if (vlan_insert || tso || swp)
 		return tx_pkt_burst;
 	if (config->mps == MLX5_MPW_ENHANCED) {
 		if (mlx5_check_vec_tx_support(dev) > 0) {
diff --git a/drivers/net/mlx5/mlx5_prm.h b/drivers/net/mlx5/mlx5_prm.h
index 3e0edb4365..05a6828981 100644
--- a/drivers/net/mlx5/mlx5_prm.h
+++ b/drivers/net/mlx5/mlx5_prm.h
@@ -107,6 +107,30 @@
 /* Inner L4 checksum offload (Tunneled packets only). */
 #define MLX5_ETH_WQE_L4_INNER_CSUM (1u << 5)
 
+/* Outer L4 type is TCP. */
+#define MLX5_ETH_WQE_L4_OUTER_TCP  (0u << 5)
+
+/* Outer L4 type is UDP. */
+#define MLX5_ETH_WQE_L4_OUTER_UDP  (1u << 5)
+
+/* Outer L3 type is IPV4. */
+#define MLX5_ETH_WQE_L3_OUTER_IPV4 (0u << 4)
+
+/* Outer L3 type is IPV6. */
+#define MLX5_ETH_WQE_L3_OUTER_IPV6 (1u << 4)
+
+/* Inner L4 type is TCP. */
+#define MLX5_ETH_WQE_L4_INNER_TCP (0u << 1)
+
+/* Inner L4 type is UDP. */
+#define MLX5_ETH_WQE_L4_INNER_UDP (1u << 1)
+
+/* Inner L3 type is IPV4. */
+#define MLX5_ETH_WQE_L3_INNER_IPV4 (0u << 0)
+
+/* Inner L3 type is IPV6. */
+#define MLX5_ETH_WQE_L3_INNER_IPV6 (1u << 0)
+
 /* Is flow mark valid. */
 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
 #define MLX5_FLOW_MARK_IS_VALID(val) ((val) & 0xffffff00)
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index e9537ce4bb..cd019fb28e 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -47,6 +47,9 @@ uint32_t mlx5_ptype_table[] __rte_cache_aligned = {
 	[0xff] = RTE_PTYPE_ALL_MASK, /* Last entry for errored packet. */
 };
 
+uint8_t mlx5_cksum_table[1 << 10] __rte_cache_aligned;
+uint8_t mlx5_swp_types_table[1 << 10] __rte_cache_aligned;
+
 /**
  * Build a table to translate Rx completion flags to packet type.
  *
@@ -202,6 +205,74 @@ mlx5_set_ptype_table(void)
 		     RTE_PTYPE_INNER_L4_UDP;
 }
 
+/**
+ * Build a table to translate packet to checksum type of Verbs.
+ */
+void
+mlx5_set_cksum_table(void)
+{
+	unsigned int i;
+	uint8_t v;
+
+	/*
+	 * The index should have:
+	 * bit[0] = PKT_TX_TCP_SEG
+	 * bit[2:3] = PKT_TX_UDP_CKSUM, PKT_TX_TCP_CKSUM
+	 * bit[4] = PKT_TX_IP_CKSUM
+	 * bit[8] = PKT_TX_OUTER_IP_CKSUM
+	 * bit[9] = tunnel
+	 */
+	for (i = 0; i < RTE_DIM(mlx5_cksum_table); ++i) {
+		v = 0;
+		if (i & (1 << 9)) {
+			/* Tunneled packet. */
+			if (i & (1 << 8)) /* Outer IP. */
+				v |= MLX5_ETH_WQE_L3_CSUM;
+			if (i & (1 << 4)) /* Inner IP. */
+				v |= MLX5_ETH_WQE_L3_INNER_CSUM;
+			if (i & (3 << 2 | 1 << 0)) /* L4 or TSO. */
+				v |= MLX5_ETH_WQE_L4_INNER_CSUM;
+		} else {
+			/* No tunnel. */
+			if (i & (1 << 4)) /* IP. */
+				v |= MLX5_ETH_WQE_L3_CSUM;
+			if (i & (3 << 2 | 1 << 0)) /* L4 or TSO. */
+				v |= MLX5_ETH_WQE_L4_CSUM;
+		}
+		mlx5_cksum_table[i] = v;
+	}
+}
+
+/**
+ * Build a table to translate packet type of mbuf to SWP type of Verbs.
+ */
+void
+mlx5_set_swp_types_table(void)
+{
+	unsigned int i;
+	uint8_t v;
+
+	/*
+	 * The index should have:
+	 * bit[0:1] = PKT_TX_L4_MASK
+	 * bit[4] = PKT_TX_IPV6
+	 * bit[8] = PKT_TX_OUTER_IPV6
+	 * bit[9] = PKT_TX_OUTER_UDP
+	 */
+	for (i = 0; i < RTE_DIM(mlx5_swp_types_table); ++i) {
+		v = 0;
+		if (i & (1 << 8))
+			v |= MLX5_ETH_WQE_L3_OUTER_IPV6;
+		if (i & (1 << 9))
+			v |= MLX5_ETH_WQE_L4_OUTER_UDP;
+		if (i & (1 << 4))
+			v |= MLX5_ETH_WQE_L3_INNER_IPV6;
+		if ((i & 3) == (PKT_TX_UDP_CKSUM >> 52))
+			v |= MLX5_ETH_WQE_L4_INNER_UDP;
+		mlx5_swp_types_table[i] = v;
+	}
+}
+
 /**
  * Return the size of tailroom of WQ.
  *
@@ -267,7 +338,6 @@ mlx5_copy_to_wq(void *dst, const void *src, size_t n,
 static int
 inline_tso(struct mlx5_txq_data *txq, struct rte_mbuf *buf,
 	   uint32_t *length,
-	   uint8_t *cs_flags,
 	   uintptr_t *addr,
 	   uint16_t *pkt_inline_sz,
 	   uint8_t **raw,
@@ -279,9 +349,8 @@ inline_tso(struct mlx5_txq_data *txq, struct rte_mbuf *buf,
 				    (1 << txq->wqe_n) * MLX5_WQE_SIZE);
 	unsigned int copy_b;
 	uint8_t vlan_sz = (buf->ol_flags & PKT_TX_VLAN_PKT) ? 4 : 0;
-	const uint8_t tunneled = txq->tunnel_en &&
-				 (buf->ol_flags & (PKT_TX_TUNNEL_GRE |
-						   PKT_TX_TUNNEL_VXLAN));
+	const uint8_t tunneled = txq->tunnel_en && (buf->ol_flags &
+				 PKT_TX_TUNNEL_MASK);
 	uint16_t n_wqe;
 
 	*tso_segsz = buf->tso_segsz;
@@ -290,19 +359,15 @@ inline_tso(struct mlx5_txq_data *txq, struct rte_mbuf *buf,
 		txq->stats.oerrors++;
 		return -EINVAL;
 	}
-	if (tunneled) {
+	if (tunneled)
 		*tso_header_sz += buf->outer_l2_len + buf->outer_l3_len;
-		*cs_flags |= MLX5_ETH_WQE_L4_INNER_CSUM;
-	} else {
-		*cs_flags |= MLX5_ETH_WQE_L4_CSUM;
-	}
-	if (unlikely(*tso_header_sz > MLX5_MAX_TSO_HEADER)) {
+	/* First seg must contain all TSO headers. */
+	if (unlikely(*tso_header_sz > MLX5_MAX_TSO_HEADER) ||
+		     *tso_header_sz > DATA_LEN(buf)) {
 		txq->stats.oerrors++;
 		return -EINVAL;
 	}
 	copy_b = *tso_header_sz - *pkt_inline_sz;
-	/* First seg must contain all TSO headers. */
-	assert(copy_b <= *length);
 	if (!copy_b || ((end - (uintptr_t)*raw) < copy_b))
 		return -EAGAIN;
 	n_wqe = (MLX5_WQE_DS(copy_b) - 1 + 3) / 4;
@@ -435,7 +500,7 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (unlikely(!max_wqe))
 		return 0;
 	do {
-		struct rte_mbuf *buf = NULL;
+		struct rte_mbuf *buf = *pkts; /* First_seg. */
 		uint8_t *raw;
 		volatile struct mlx5_wqe_v *wqe = NULL;
 		volatile rte_v128u32_t *dseg = NULL;
@@ -447,15 +512,16 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		uint16_t tso_header_sz = 0;
 		uint16_t ehdr;
 		uint8_t cs_flags;
-		uint64_t tso = 0;
+		uint8_t tso = txq->tso_en && (buf->ol_flags & PKT_TX_TCP_SEG);
+		uint8_t is_vlan = !!(buf->ol_flags & PKT_TX_VLAN_PKT);
+		uint32_t swp_offsets = 0;
+		uint8_t swp_types = 0;
 		uint16_t tso_segsz = 0;
 #ifdef MLX5_PMD_SOFT_COUNTERS
 		uint32_t total_length = 0;
 #endif
 		int ret;
 
-		/* first_seg */
-		buf = *pkts;
 		segs_n = buf->nb_segs;
 		/*
 		 * Make sure there is enough room to store this packet and
@@ -490,10 +556,12 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		if (pkts_n - i > 1)
 			rte_prefetch0(
 			    rte_pktmbuf_mtod(*(pkts + 1), volatile void *));
-		cs_flags = txq_ol_cksum_to_cs(txq, buf);
+		cs_flags = txq_ol_cksum_to_cs(buf);
+		txq_mbuf_to_swp(txq, buf, tso, is_vlan,
+				(uint8_t *)&swp_offsets, &swp_types);
 		raw = ((uint8_t *)(uintptr_t)wqe) + 2 * MLX5_WQE_DWORD_SIZE;
 		/* Replace the Ethernet type by the VLAN if necessary. */
-		if (buf->ol_flags & PKT_TX_VLAN_PKT) {
+		if (is_vlan) {
 			uint32_t vlan = rte_cpu_to_be_32(0x81000000 |
 							 buf->vlan_tci);
 			unsigned int len = 2 * ETHER_ADDR_LEN - 2;
@@ -516,9 +584,8 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			addr += pkt_inline_sz;
 		}
 		raw += MLX5_WQE_DWORD_SIZE;
-		tso = txq->tso_en && (buf->ol_flags & PKT_TX_TCP_SEG);
 		if (tso) {
-			ret = inline_tso(txq, buf, &length, &cs_flags,
+			ret = inline_tso(txq, buf, &length,
 					 &addr, &pkt_inline_sz,
 					 &raw, &max_wqe,
 					 &tso_segsz, &tso_header_sz);
@@ -695,8 +762,9 @@ next_pkt:
 				0,
 			};
 			wqe->eseg = (rte_v128u32_t){
-				0,
-				cs_flags | (rte_cpu_to_be_16(tso_segsz) << 16),
+				swp_offsets,
+				cs_flags | (swp_types << 8) |
+				(rte_cpu_to_be_16(tso_segsz) << 16),
 				0,
 				(ehdr << 16) | rte_cpu_to_be_16(tso_header_sz),
 			};
@@ -709,8 +777,8 @@ next_pkt:
 				0,
 			};
 			wqe->eseg = (rte_v128u32_t){
-				0,
-				cs_flags,
+				swp_offsets,
+				cs_flags | (swp_types << 8),
 				0,
 				(ehdr << 16) | rte_cpu_to_be_16(pkt_inline_sz),
 			};
@@ -882,7 +950,7 @@ mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		}
 		max_elts -= segs_n;
 		--pkts_n;
-		cs_flags = txq_ol_cksum_to_cs(txq, buf);
+		cs_flags = txq_ol_cksum_to_cs(buf);
 		/* Retrieve packet information. */
 		length = PKT_LEN(buf);
 		assert(length);
@@ -1114,7 +1182,7 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 		 * iteration.
 		 */
 		max_wqe = (1u << txq->wqe_n) - (txq->wqe_ci - txq->wqe_pi);
-		cs_flags = txq_ol_cksum_to_cs(txq, buf);
+		cs_flags = txq_ol_cksum_to_cs(buf);
 		/* Retrieve packet information. */
 		length = PKT_LEN(buf);
 		/* Start new session if packet differs. */
@@ -1391,7 +1459,7 @@ txq_burst_empw(struct mlx5_txq_data *txq, struct rte_mbuf **pkts,
 		/* Make sure there is enough room to store this packet. */
 		if (max_elts - j == 0)
 			break;
-		cs_flags = txq_ol_cksum_to_cs(txq, buf);
+		cs_flags = txq_ol_cksum_to_cs(buf);
 		/* Retrieve packet information. */
 		length = PKT_LEN(buf);
 		/* Start new session if:
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 2309aa4f39..14d4418d99 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -168,6 +168,7 @@ struct mlx5_txq_data {
 	uint16_t tso_en:1; /* When set hardware TSO is enabled. */
 	uint16_t tunnel_en:1;
 	/* When set TX offload for tunneled packets are supported. */
+	uint16_t swp_en:1; /* Whether SW parser is enabled. */
 	uint16_t mpw_hdr_dseg:1; /* Enable DSEGs in the title WQEBB. */
 	uint16_t max_inline; /* Multiple of RTE_CACHE_LINE_SIZE to inline. */
 	uint16_t inline_max_packet_sz; /* Max packet size for inlining. */
@@ -280,8 +281,12 @@ uint64_t mlx5_get_tx_port_offloads(struct rte_eth_dev *dev);
 /* mlx5_rxtx.c */
 
 extern uint32_t mlx5_ptype_table[];
+extern uint8_t mlx5_cksum_table[];
+extern uint8_t mlx5_swp_types_table[];
 
 void mlx5_set_ptype_table(void);
+void mlx5_set_cksum_table(void);
+void mlx5_set_swp_types_table(void);
 uint16_t mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts,
 		       uint16_t pkts_n);
 uint16_t mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts,
@@ -614,38 +619,89 @@ mlx5_tx_dbrec(struct mlx5_txq_data *txq, volatile struct mlx5_wqe *wqe)
 }
 
 /**
- * Convert the Checksum offloads to Verbs.
+ * Convert mbuf to Verb SWP.
  *
  * @param txq_data
  *   Pointer to the Tx queue.
  * @param buf
  *   Pointer to the mbuf.
+ * @param tso
+ *   TSO offloads enabled.
+ * @param vlan
+ *   VLAN offloads enabled
+ * @param offsets
+ *   Pointer to the SWP header offsets.
+ * @param swp_types
+ *   Pointer to the SWP header types.
+ */
+static __rte_always_inline void
+txq_mbuf_to_swp(struct mlx5_txq_data *txq, struct rte_mbuf *buf,
+		 uint8_t tso, uint64_t vlan,
+		 uint8_t *offsets, uint8_t *swp_types)
+{
+	uint64_t tunnel = buf->ol_flags & PKT_TX_TUNNEL_MASK;
+	uint16_t idx;
+	uint16_t off;
+	const uint64_t ol_flags_mask = PKT_TX_L4_MASK | PKT_TX_IPV6 |
+				       PKT_TX_OUTER_IPV6;
+
+	if (likely(!tunnel || !txq->swp_en ||
+		   (tunnel != PKT_TX_TUNNEL_UDP && tunnel != PKT_TX_TUNNEL_IP)))
+		return;
+	/*
+	 * The index should have:
+	 * bit[0:1] = PKT_TX_L4_MASK
+	 * bit[4] = PKT_TX_IPV6
+	 * bit[8] = PKT_TX_OUTER_IPV6
+	 * bit[9] = PKT_TX_OUTER_UDP
+	 */
+	idx = (buf->ol_flags & ol_flags_mask) >> 52;
+	if (tunnel == PKT_TX_TUNNEL_UDP)
+		idx |= 1 << 9;
+	*swp_types = mlx5_swp_types_table[idx];
+	/* swp offsets. */
+	off = buf->outer_l2_len + (vlan ? 4 : 0); /* Outer L3 offset. */
+	if (tso || (buf->ol_flags & PKT_TX_OUTER_IP_CKSUM))
+		offsets[1] = off >> 1;
+	off += buf->outer_l3_len; /* Outer L4 offset. */
+	if (tunnel == PKT_TX_TUNNEL_UDP)
+		offsets[0] = off >> 1;
+	off += buf->l2_len; /* Inner L3 offset. */
+	if (tso || (buf->ol_flags & PKT_TX_IP_CKSUM))
+		offsets[3] = off >> 1;
+	off += buf->l3_len; /* Inner L4 offset. */
+	if (tso || ((buf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) ||
+	    ((buf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM))
+		offsets[2] = off >> 1;
+}
+
+/**
+ * Convert the Checksum offloads to Verbs.
+ *
+ * @param buf
+ *   Pointer to the mbuf.
  *
  * @return
- *   the converted cs_flags.
+ *   Converted checksum flags.
  */
 static __rte_always_inline uint8_t
-txq_ol_cksum_to_cs(struct mlx5_txq_data *txq_data, struct rte_mbuf *buf)
+txq_ol_cksum_to_cs(struct rte_mbuf *buf)
 {
-	uint8_t cs_flags = 0;
-
-	/* Should we enable HW CKSUM offload */
-	if (buf->ol_flags &
-	    (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM |
-	     PKT_TX_OUTER_IP_CKSUM)) {
-		if (txq_data->tunnel_en &&
-		    (buf->ol_flags &
-		     (PKT_TX_TUNNEL_GRE | PKT_TX_TUNNEL_VXLAN))) {
-			cs_flags = MLX5_ETH_WQE_L3_INNER_CSUM |
-				   MLX5_ETH_WQE_L4_INNER_CSUM;
-			if (buf->ol_flags & PKT_TX_OUTER_IP_CKSUM)
-				cs_flags |= MLX5_ETH_WQE_L3_CSUM;
-		} else {
-			cs_flags = MLX5_ETH_WQE_L3_CSUM |
-				   MLX5_ETH_WQE_L4_CSUM;
-		}
-	}
-	return cs_flags;
+	uint32_t idx;
+	uint8_t is_tunnel = !!(buf->ol_flags & PKT_TX_TUNNEL_MASK);
+	const uint64_t ol_flags_mask = PKT_TX_TCP_SEG | PKT_TX_L4_MASK |
+				       PKT_TX_IP_CKSUM | PKT_TX_OUTER_IP_CKSUM;
+
+	/*
+	 * The index should have:
+	 * bit[0] = PKT_TX_TCP_SEG
+	 * bit[2:3] = PKT_TX_UDP_CKSUM, PKT_TX_TCP_CKSUM
+	 * bit[4] = PKT_TX_IP_CKSUM
+	 * bit[8] = PKT_TX_OUTER_IP_CKSUM
+	 * bit[9] = tunnel
+	 */
+	idx = ((buf->ol_flags & ol_flags_mask) >> 50) | (!!is_tunnel << 9);
+	return mlx5_cksum_table[idx];
 }
 
 /**
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec.c b/drivers/net/mlx5/mlx5_rxtx_vec.c
index 8b132396de..215a9e5926 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec.c
+++ b/drivers/net/mlx5/mlx5_rxtx_vec.c
@@ -42,8 +42,6 @@
 /**
  * Count the number of packets having same ol_flags and calculate cs_flags.
  *
- * @param txq
- *   Pointer to TX queue structure.
  * @param pkts
  *   Pointer to array of packets.
  * @param pkts_n
@@ -55,8 +53,7 @@
  *   Number of packets having same ol_flags.
  */
 static inline unsigned int
-txq_calc_offload(struct mlx5_txq_data *txq, struct rte_mbuf **pkts,
-		 uint16_t pkts_n, uint8_t *cs_flags)
+txq_calc_offload(struct rte_mbuf **pkts, uint16_t pkts_n, uint8_t *cs_flags)
 {
 	unsigned int pos;
 	const uint64_t ol_mask =
@@ -70,7 +67,7 @@ txq_calc_offload(struct mlx5_txq_data *txq, struct rte_mbuf **pkts,
 	for (pos = 1; pos < pkts_n; ++pos)
 		if ((pkts[pos]->ol_flags ^ pkts[0]->ol_flags) & ol_mask)
 			break;
-	*cs_flags = txq_ol_cksum_to_cs(txq, pkts[0]);
+	*cs_flags = txq_ol_cksum_to_cs(pkts[0]);
 	return pos;
 }
 
@@ -141,7 +138,7 @@ mlx5_tx_burst_vec(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		if (txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS)
 			n = txq_count_contig_single_seg(&pkts[nb_tx], n);
 		if (txq->offloads & MLX5_VEC_TX_CKSUM_OFFLOAD_CAP)
-			n = txq_calc_offload(txq, &pkts[nb_tx], n, &cs_flags);
+			n = txq_calc_offload(&pkts[nb_tx], n, &cs_flags);
 		ret = txq_burst_v(txq, &pkts[nb_tx], n, cs_flags);
 		nb_tx += ret;
 		if (!ret)
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_neon.h b/drivers/net/mlx5/mlx5_rxtx_vec_neon.h
index 84817e7ad2..240d396e35 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec_neon.h
+++ b/drivers/net/mlx5/mlx5_rxtx_vec_neon.h
@@ -142,7 +142,7 @@ txq_scatter_v(struct mlx5_txq_data *txq, struct rte_mbuf **pkts,
 			break;
 		wqe = &((volatile struct mlx5_wqe64 *)
 			 txq->wqes)[wqe_ci & wq_mask].hdr;
-		cs_flags = txq_ol_cksum_to_cs(txq, buf);
+		cs_flags = txq_ol_cksum_to_cs(buf);
 		/* Title WQEBB pointer. */
 		t_wqe = (uint8x16_t *)wqe;
 		dseg = (uint8_t *)(wqe + 1);
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_sse.h b/drivers/net/mlx5/mlx5_rxtx_vec_sse.h
index 83d6e431fa..d10da29f67 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec_sse.h
+++ b/drivers/net/mlx5/mlx5_rxtx_vec_sse.h
@@ -144,7 +144,7 @@ txq_scatter_v(struct mlx5_txq_data *txq, struct rte_mbuf **pkts,
 		}
 		wqe = &((volatile struct mlx5_wqe64 *)
 			 txq->wqes)[wqe_ci & wq_mask].hdr;
-		cs_flags = txq_ol_cksum_to_cs(txq, buf);
+		cs_flags = txq_ol_cksum_to_cs(buf);
 		/* Title WQEBB pointer. */
 		t_wqe = (__m128i *)wqe;
 		dseg = (__m128i *)(wqe + 1);
diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
index 1c5e4c547e..44358744d6 100644
--- a/drivers/net/mlx5/mlx5_txq.c
+++ b/drivers/net/mlx5/mlx5_txq.c
@@ -119,6 +119,9 @@ mlx5_get_tx_port_offloads(struct rte_eth_dev *dev)
 		if (config->tso)
 			offloads |= (DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
 				     DEV_TX_OFFLOAD_GRE_TNL_TSO);
+		if (config->swp)
+			offloads |= (DEV_TX_OFFLOAD_IP_TNL_TSO |
+				     DEV_TX_OFFLOAD_UDP_TNL_TSO);
 	}
 	return offloads;
 }
@@ -686,7 +689,9 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl)
 	int is_empw_func = is_empw_burst_func(tx_pkt_burst);
 	int tso = !!(txq_ctrl->txq.offloads & (DEV_TX_OFFLOAD_TCP_TSO |
 					       DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
-					       DEV_TX_OFFLOAD_GRE_TNL_TSO));
+					       DEV_TX_OFFLOAD_GRE_TNL_TSO |
+					       DEV_TX_OFFLOAD_IP_TNL_TSO |
+					       DEV_TX_OFFLOAD_UDP_TNL_TSO));
 
 	txq_inline = (config->txq_inline == MLX5_ARG_UNSET) ?
 		0 : config->txq_inline;
@@ -767,6 +772,9 @@ txq_set_params(struct mlx5_txq_ctrl *txq_ctrl)
 		txq_ctrl->txq.tso_en = 1;
 	}
 	txq_ctrl->txq.tunnel_en = config->tunnel_en;
+	txq_ctrl->txq.swp_en = ((DEV_TX_OFFLOAD_IP_TNL_TSO |
+				 DEV_TX_OFFLOAD_UDP_TNL_TSO) &
+				txq_ctrl->txq.offloads) && config->swp;
 }
 
 /**