diff --git a/contrib/ofed/libibverbs/man/ibv_create_qp_ex.3 b/contrib/ofed/libibverbs/man/ibv_create_qp_ex.3
--- a/contrib/ofed/libibverbs/man/ibv_create_qp_ex.3
+++ b/contrib/ofed/libibverbs/man/ibv_create_qp_ex.3
@@ -69,6 +69,26 @@
 .in -8
 };
 .fi
+
+.nf
+enum ibv_rx_hash_fields {
+.in +8
+IBV_RX_HASH_SRC_IPV4            = 1 << 0,
+IBV_RX_HASH_DST_IPV4            = 1 << 1,
+IBV_RX_HASH_SRC_IPV6            = 1 << 2,
+IBV_RX_HASH_DST_IPV6            = 1 << 3,
+IBV_RX_HASH_SRC_PORT_TCP        = 1 << 4,
+IBV_RX_HASH_DST_PORT_TCP        = 1 << 5,
+IBV_RX_HASH_SRC_PORT_UDP        = 1 << 6,
+IBV_RX_HASH_DST_PORT_UDP        = 1 << 7,
+/* When using tunneling protocol, e.g. VXLAN, then we have an inner (encapsulated packet) and outer.
+ * For applying RSS on the inner packet, then the following field should be set with one of the L3/L4 fields.
+*/
+IBV_RX_HASH_INNER		= (1UL << 31),
+.in -8
+};
+.fi
+
 .PP
 The function
 .B ibv_create_qp_ex()
diff --git a/contrib/ofed/libibverbs/verbs.h b/contrib/ofed/libibverbs/verbs.h
--- a/contrib/ofed/libibverbs/verbs.h
+++ b/contrib/ofed/libibverbs/verbs.h
@@ -236,7 +236,8 @@
 	IBV_RX_HASH_SRC_PORT_TCP	= 1 << 4,
 	IBV_RX_HASH_DST_PORT_TCP	= 1 << 5,
 	IBV_RX_HASH_SRC_PORT_UDP	= 1 << 6,
-	IBV_RX_HASH_DST_PORT_UDP	= 1 << 7
+	IBV_RX_HASH_DST_PORT_UDP	= 1 << 7,
+	IBV_RX_HASH_INNER		= (1UL << 31),
 };
 
 struct ibv_rss_caps {
diff --git a/contrib/ofed/libmlx5/libmlx5.map b/contrib/ofed/libmlx5/libmlx5.map
--- a/contrib/ofed/libmlx5/libmlx5.map
+++ b/contrib/ofed/libmlx5/libmlx5.map
@@ -19,5 +19,6 @@
 
 MLX5_1.3 {
 	global:
+		mlx5dv_create_qp;
 		mlx5dv_create_wq;
 } MLX5_1.2;
diff --git a/contrib/ofed/libmlx5/mlx5-abi.h b/contrib/ofed/libmlx5/mlx5-abi.h
--- a/contrib/ofed/libmlx5/mlx5-abi.h
+++ b/contrib/ofed/libmlx5/mlx5-abi.h
@@ -43,6 +43,7 @@
 enum {
 	MLX5_QP_FLAG_SIGNATURE		= 1 << 0,
 	MLX5_QP_FLAG_SCATTER_CQE	= 1 << 1,
+	MLX5_QP_FLAG_TUNNEL_OFFLOADS	= 1 << 2,
 };
 
 enum {
@@ -178,7 +179,7 @@
 	__u8 reserved[6];
 	__u8 rx_hash_key[128];
 	__u32   comp_mask;
-	__u32   reserved1;
+	__u32   create_flags;
 };
 
 struct mlx5_create_qp_resp_ex {
@@ -302,6 +303,8 @@
 	__u32				reserved;
 	struct mlx5dv_sw_parsing_caps	sw_parsing_caps;
 	struct mlx5_striding_rq_caps	striding_rq_caps;
+	__u32				tunnel_offloads_caps;
+	__u32				reserved4;
 };
 
 #endif /* MLX5_ABI_H */
diff --git a/contrib/ofed/libmlx5/mlx5.h b/contrib/ofed/libmlx5/mlx5.h
--- a/contrib/ofed/libmlx5/mlx5.h
+++ b/contrib/ofed/libmlx5/mlx5.h
@@ -274,6 +274,7 @@
 	struct mlx5dv_ctx_allocators	extern_alloc;
 	struct mlx5dv_sw_parsing_caps	sw_parsing_caps;
 	struct mlx5dv_striding_rq_caps	striding_rq_caps;
+	uint32_t			tunnel_offloads_caps;
 };
 
 struct mlx5_bitmap {
diff --git a/contrib/ofed/libmlx5/mlx5.c b/contrib/ofed/libmlx5/mlx5.c
--- a/contrib/ofed/libmlx5/mlx5.c
+++ b/contrib/ofed/libmlx5/mlx5.c
@@ -641,6 +641,10 @@
 		comp_mask_out |= MLX5DV_CONTEXT_MASK_STRIDING_RQ;
 	}
 
+	if (attrs_out->comp_mask & MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS) {
+		attrs_out->tunnel_offloads_caps = mctx->tunnel_offloads_caps;
+		comp_mask_out |= MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS;
+	}
 
 	attrs_out->comp_mask = comp_mask_out;
 
diff --git a/contrib/ofed/libmlx5/mlx5dv.h b/contrib/ofed/libmlx5/mlx5dv.h
--- a/contrib/ofed/libmlx5/mlx5dv.h
+++ b/contrib/ofed/libmlx5/mlx5dv.h
@@ -59,7 +59,8 @@
 	MLX5DV_CONTEXT_MASK_CQE_COMPRESION	= 1 << 0,
 	MLX5DV_CONTEXT_MASK_SWP			= 1 << 1,
 	MLX5DV_CONTEXT_MASK_STRIDING_RQ		= 1 << 2,
-	MLX5DV_CONTEXT_MASK_RESERVED		= 1 << 3,
+	MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS	= 1 << 3,
+	MLX5DV_CONTEXT_MASK_RESERVED		= 1 << 4,
 };
 
 struct mlx5dv_cqe_comp_caps {
@@ -80,6 +81,13 @@
 	uint32_t supported_qpts;
 };
 
+enum mlx5dv_tunnel_offloads {
+	MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_VXLAN		= 1 << 0,
+	MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GRE		= 1 << 1,
+	MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GENEVE		= 1 << 2,
+};
+
+
 /*
  * Direct verbs device-specific attributes
  */
@@ -90,6 +98,7 @@
 	struct mlx5dv_cqe_comp_caps	cqe_comp_caps;
 	struct mlx5dv_sw_parsing_caps sw_parsing_caps;
 	struct mlx5dv_striding_rq_caps striding_rq_caps;
+	uint32_t	tunnel_offloads_caps;
 };
 
 enum mlx5dv_context_flags {
@@ -115,6 +124,23 @@
 struct ibv_cq_ex *mlx5dv_create_cq(struct ibv_context *context,
 				   struct ibv_cq_init_attr_ex *cq_attr,
 				   struct mlx5dv_cq_init_attr *mlx5_cq_attr);
+
+enum mlx5dv_qp_create_flags {
+	MLX5DV_QP_CREATE_TUNNEL_OFFLOADS = 1 << 0,
+};
+
+enum mlx5dv_qp_init_attr_mask {
+	MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS	= 1 << 0,
+};
+
+struct mlx5dv_qp_init_attr {
+	uint64_t comp_mask;	/* Use enum mlx5dv_qp_init_attr_mask */
+	uint32_t create_flags;	/* Use enum mlx5dv_qp_create_flags */
+};
+
+struct ibv_qp *mlx5dv_create_qp(struct ibv_context *context,
+			struct ibv_qp_init_attr_ex *qp_attr,
+			struct mlx5dv_qp_init_attr *mlx5_qp_attr);
 /*
  * Most device capabilities are exported by ibv_query_device(...),
  * but there is HW device-specific information which is important
diff --git a/contrib/ofed/libmlx5/mlx5dv_query_device.3 b/contrib/ofed/libmlx5/mlx5dv_query_device.3
--- a/contrib/ofed/libmlx5/mlx5dv_query_device.3
+++ b/contrib/ofed/libmlx5/mlx5dv_query_device.3
@@ -45,6 +45,7 @@
 uint8_t         version;
 uint64_t        flags;
 uint64_t        comp_mask;
+uint32_t 	tunnel_offloads_caps;
 .in -8
 };
 
@@ -64,10 +65,21 @@
 .nf
 enum mlx5dv_context_comp_mask {
 .in +8
- MLX5DV_CONTEXT_MASK_CQE_COMPRESION = 1 << 0,
- MLX5DV_CONTEXT_MASK_SWP = 1 << 1,
- MLX5DV_CONTEXT_MASK_STRIDING_RQ = 1 << 2,
- MLX5DV_CONTEXT_MASK_RESERVED = 1 << 3,
+ MLX5DV_CONTEXT_MASK_CQE_COMPRESION	= 1 << 0,
+ MLX5DV_CONTEXT_MASK_SWP		= 1 << 1,
+ MLX5DV_CONTEXT_MASK_STRIDING_RQ	= 1 << 2,
+ MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS	= 1 << 3,
+ MLX5DV_CONTEXT_MASK_RESERVED		= 1 << 4,
+.in -8
+};
+
+.PP
+.nf
+enum mlx5dv_tunnel_offloads {
+.in +8
+ MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_VXLAN   = 1 << 0,
+ MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GRE     = 1 << 1,
+ MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_GENEVE	= 1 << 2,
 .in -8
 };
 
diff --git a/contrib/ofed/libmlx5/verbs.c b/contrib/ofed/libmlx5/verbs.c
--- a/contrib/ofed/libmlx5/verbs.c
+++ b/contrib/ofed/libmlx5/verbs.c
@@ -1177,7 +1177,8 @@
 
 static int mlx5_cmd_create_rss_qp(struct ibv_context *context,
 				 struct ibv_qp_init_attr_ex *attr,
-				 struct mlx5_qp *qp)
+				 struct mlx5_qp *qp,
+				 uint32_t mlx5_create_flags)
 {
 	struct mlx5_create_qp_ex_rss cmd_ex_rss = {};
 	struct mlx5_create_qp_resp_ex resp = {};
@@ -1191,6 +1192,7 @@
 	cmd_ex_rss.rx_hash_fields_mask = attr->rx_hash_conf.rx_hash_fields_mask;
 	cmd_ex_rss.rx_hash_function = attr->rx_hash_conf.rx_hash_function;
 	cmd_ex_rss.rx_key_len = attr->rx_hash_conf.rx_hash_key_len;
+	cmd_ex_rss.create_flags = mlx5_create_flags;
 	memcpy(cmd_ex_rss.rx_hash_key, attr->rx_hash_conf.rx_hash_key,
 			attr->rx_hash_conf.rx_hash_key_len);
 
@@ -1243,6 +1245,10 @@
 					IBV_QP_INIT_ATTR_RX_HASH),
 };
 
+enum {
+	MLX5_DV_CREATE_QP_SUP_COMP_MASK = MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS
+};
+
 enum {
 	MLX5_CREATE_QP_EX2_COMP_MASK = (IBV_QP_INIT_ATTR_CREATE_FLAGS |
 					IBV_QP_INIT_ATTR_MAX_TSO_HEADER |
@@ -1251,7 +1257,8 @@
 };
 
 static struct ibv_qp *create_qp(struct ibv_context *context,
-			 struct ibv_qp_init_attr_ex *attr)
+				struct ibv_qp_init_attr_ex *attr,
+				struct mlx5dv_qp_init_attr *mlx5_qp_attr)
 {
 	struct mlx5_create_qp		cmd;
 	struct mlx5_create_qp_resp	resp;
@@ -1262,6 +1269,7 @@
 	struct ibv_qp		       *ibqp;
 	int32_t				usr_idx = 0;
 	uint32_t			uuar_index;
+	uint32_t			mlx5_create_flags = 0;
 	FILE *fp = ctx->dbg_fp;
 
 	if (attr->comp_mask & ~MLX5_CREATE_QP_SUP_COMP_MASK)
@@ -1283,14 +1291,39 @@
 	memset(&resp, 0, sizeof(resp));
 	memset(&resp_ex, 0, sizeof(resp_ex));
 
+	if (mlx5_qp_attr) {
+		if (!check_comp_mask(mlx5_qp_attr->comp_mask,
+				     MLX5_DV_CREATE_QP_SUP_COMP_MASK)) {
+			mlx5_dbg(fp, MLX5_DBG_QP,
+					"Unsupported vendor comp_mask for create_qp\n");
+			errno = EINVAL;
+			goto err;
+		}
+
+		if (mlx5_qp_attr->comp_mask &
+		    MLX5DV_QP_INIT_ATTR_MASK_QP_CREATE_FLAGS) {
+			if (mlx5_qp_attr->create_flags &
+			    MLX5DV_QP_CREATE_TUNNEL_OFFLOADS) {
+				mlx5_create_flags = MLX5_QP_FLAG_TUNNEL_OFFLOADS;
+			} else {
+				mlx5_dbg(fp, MLX5_DBG_QP,
+					 "Unsupported creation flags requested for create_qp\n");
+				errno = EINVAL;
+				goto err;
+			}
+		}
+	}
+
 	if (attr->comp_mask & IBV_QP_INIT_ATTR_RX_HASH) {
-		ret = mlx5_cmd_create_rss_qp(context, attr, qp);
+		ret = mlx5_cmd_create_rss_qp(context, attr, qp,
+					     mlx5_create_flags);
 		if (ret)
 			goto err;
 
 		return ibqp;
 	}
 
+	cmd.flags = mlx5_create_flags;
 	qp->wq_sig = qp_sig_enabled();
 	if (qp->wq_sig)
 		cmd.flags |= MLX5_QP_FLAG_SIGNATURE;
@@ -1441,7 +1474,7 @@
 	memcpy(&attrx, attr, sizeof(*attr));
 	attrx.comp_mask = IBV_QP_INIT_ATTR_PD;
 	attrx.pd = pd;
-	qp = create_qp(pd->context, &attrx);
+	qp = create_qp(pd->context, &attrx, NULL);
 	if (qp)
 		memcpy(attr, &attrx, sizeof(*attr));
 
@@ -1767,7 +1800,14 @@
 struct ibv_qp *mlx5_create_qp_ex(struct ibv_context *context,
 				 struct ibv_qp_init_attr_ex *attr)
 {
-	return create_qp(context, attr);
+	return create_qp(context, attr, NULL);
+}
+
+struct ibv_qp *mlx5dv_create_qp(struct ibv_context *context,
+				struct ibv_qp_init_attr_ex *qp_attr,
+				struct mlx5dv_qp_init_attr *mlx5_qp_attr)
+{
+	return create_qp(context, qp_attr, mlx5_qp_attr);
 }
 
 int mlx5_get_srq_num(struct ibv_srq *srq, uint32_t *srq_num)
@@ -1852,7 +1892,7 @@
 	init_attr.send_cq = srq_attr->cq;
 	init_attr.recv_cq = srq_attr->cq;
 
-	qp = create_qp(context, &init_attr);
+	qp = create_qp(context, &init_attr, NULL);
 	if (!qp)
 		return NULL;
 
@@ -2095,6 +2135,7 @@
 	mctx->cqe_comp_caps = resp.cqe_comp_caps;
 	mctx->sw_parsing_caps = resp.sw_parsing_caps;
 	mctx->striding_rq_caps = resp.striding_rq_caps.caps;
+	mctx->tunnel_offloads_caps = resp.tunnel_offloads_caps;
 
 	major     = (raw_fw_ver >> 32) & 0xffff;
 	minor     = (raw_fw_ver >> 16) & 0xffff;