X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;ds=sidebyside;f=drivers%2Fnet%2Fmlx5%2Fmlx5_rxtx.c;h=56c7f780fbd7bf3b1f14844d432882d6cce8340a;hb=a4193ae3bc4fcaf6a81b472563466946771a5f8e;hp=2a66d52ecb16a3221d629a675cd93846dda94506;hpb=6579c27c11a5fb07b03cea837d51ed64ba73a763;p=dpdk.git

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 2a66d52ecb..56c7f780fb 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -69,6 +69,35 @@
 #include "mlx5_defs.h"
 #include "mlx5_prm.h"
 
+static inline int
+check_cqe(volatile struct mlx5_cqe *cqe,
+	  unsigned int cqes_n, const uint16_t ci)
+	  __attribute__((always_inline));
+
+static inline void
+txq_complete(struct txq *txq) __attribute__((always_inline));
+
+static inline uint32_t
+txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
+	__attribute__((always_inline));
+
+static inline void
+mlx5_tx_dbrec(struct txq *txq, volatile struct mlx5_wqe *wqe)
+	__attribute__((always_inline));
+
+static inline uint32_t
+rxq_cq_to_pkt_type(volatile struct mlx5_cqe *cqe)
+	__attribute__((always_inline));
+
+static inline int
+mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe *cqe,
+		 uint16_t cqe_cnt, uint32_t *rss_hash)
+		 __attribute__((always_inline));
+
+static inline uint32_t
+rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe *cqe)
+		   __attribute__((always_inline));
+
 #ifndef NDEBUG
 
 /**
@@ -81,10 +110,10 @@
  *   0 the first time.
  */
 static inline int
-check_cqe64_seen(volatile struct mlx5_cqe64 *cqe)
+check_cqe_seen(volatile struct mlx5_cqe *cqe)
 {
 	static const uint8_t magic[] = "seen";
-	volatile uint8_t (*buf)[sizeof(cqe->rsvd40)] = &cqe->rsvd40;
+	volatile uint8_t (*buf)[sizeof(cqe->rsvd0)] = &cqe->rsvd0;
 	int ret = 1;
 	unsigned int i;
 
@@ -98,11 +127,6 @@ check_cqe64_seen(volatile struct mlx5_cqe64 *cqe)
 
 #endif /* NDEBUG */
 
-static inline int
-check_cqe64(volatile struct mlx5_cqe64 *cqe,
-	    unsigned int cqes_n, const uint16_t ci)
-	    __attribute__((always_inline));
-
 /**
  * Check whether CQE is valid.
  *
@@ -117,8 +141,8 @@ check_cqe64(volatile struct mlx5_cqe64 *cqe,
  *   0 on success, 1 on failure.
  */
 static inline int
-check_cqe64(volatile struct mlx5_cqe64 *cqe,
-		unsigned int cqes_n, const uint16_t ci)
+check_cqe(volatile struct mlx5_cqe *cqe,
+	  unsigned int cqes_n, const uint16_t ci)
 {
 	uint16_t idx = ci & cqes_n;
 	uint8_t op_own = cqe->op_own;
@@ -136,14 +160,14 @@ check_cqe64(volatile struct mlx5_cqe64 *cqe,
 		if ((syndrome == MLX5_CQE_SYNDROME_LOCAL_LENGTH_ERR) ||
 		    (syndrome == MLX5_CQE_SYNDROME_REMOTE_ABORTED_ERR))
 			return 0;
-		if (!check_cqe64_seen(cqe))
+		if (!check_cqe_seen(cqe))
 			ERROR("unexpected CQE error %u (0x%02x)"
 			      " syndrome 0x%02x",
 			      op_code, op_code, syndrome);
 		return 1;
 	} else if ((op_code != MLX5_CQE_RESP_SEND) &&
 		   (op_code != MLX5_CQE_REQ)) {
-		if (!check_cqe64_seen(cqe))
+		if (!check_cqe_seen(cqe))
 			ERROR("unexpected CQE opcode %u (0x%02x)",
 			      op_code, op_code);
 		return 1;
@@ -152,8 +176,23 @@ check_cqe64(volatile struct mlx5_cqe64 *cqe,
 	return 0;
 }
 
-static inline void
-txq_complete(struct txq *txq) __attribute__((always_inline));
+/**
+ * Return the address of the WQE.
+ *
+ * @param txq
+ *   Pointer to TX queue structure.
+ * @param  wqe_ci
+ *   WQE consumer index.
+ *
+ * @return
+ *   WQE address.
+ */
+static inline uintptr_t *
+tx_mlx5_wqe(struct txq *txq, uint16_t ci)
+{
+	ci &= ((1 << txq->wqe_n) - 1);
+	return (uintptr_t *)((uintptr_t)txq->wqes + ci * MLX5_WQE_SIZE);
+}
 
 /**
  * Manage TX completions.
@@ -172,25 +211,25 @@ txq_complete(struct txq *txq)
 	uint16_t elts_free = txq->elts_tail;
 	uint16_t elts_tail;
 	uint16_t cq_ci = txq->cq_ci;
-	volatile struct mlx5_cqe64 *cqe = NULL;
-	volatile struct mlx5_wqe *wqe;
+	volatile struct mlx5_cqe *cqe = NULL;
+	volatile struct mlx5_wqe_ctrl *ctrl;
 
 	do {
-		volatile struct mlx5_cqe64 *tmp;
+		volatile struct mlx5_cqe *tmp;
 
-		tmp = &(*txq->cqes)[cq_ci & cqe_cnt].cqe64;
-		if (check_cqe64(tmp, cqe_n, cq_ci))
+		tmp = &(*txq->cqes)[cq_ci & cqe_cnt];
+		if (check_cqe(tmp, cqe_n, cq_ci))
 			break;
 		cqe = tmp;
 #ifndef NDEBUG
 		if (MLX5_CQE_FORMAT(cqe->op_own) == MLX5_COMPRESSED) {
-			if (!check_cqe64_seen(cqe))
+			if (!check_cqe_seen(cqe))
 				ERROR("unexpected compressed CQE, TX stopped");
 			return;
 		}
 		if ((MLX5_CQE_OPCODE(cqe->op_own) == MLX5_CQE_RESP_ERR) ||
 		    (MLX5_CQE_OPCODE(cqe->op_own) == MLX5_CQE_REQ_ERR)) {
-			if (!check_cqe64_seen(cqe))
+			if (!check_cqe_seen(cqe))
 				ERROR("unexpected error CQE, TX stopped");
 			return;
 		}
@@ -199,9 +238,9 @@ txq_complete(struct txq *txq)
 	} while (1);
 	if (unlikely(cqe == NULL))
 		return;
-	wqe = &(*txq->wqes)[htons(cqe->wqe_counter) &
-			    ((1 << txq->wqe_n) - 1)].hdr;
-	elts_tail = wqe->ctrl[3];
+	ctrl = (volatile struct mlx5_wqe_ctrl *)
+		tx_mlx5_wqe(txq, ntohs(cqe->wqe_counter));
+	elts_tail = ctrl->ctrl3;
 	assert(elts_tail < (1 << txq->wqe_n));
 	/* Free buffers. */
 	while (elts_free != elts_tail) {
@@ -246,10 +285,6 @@ txq_mb2mp(struct rte_mbuf *buf)
 	return buf->pool;
 }
 
-static inline uint32_t
-txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
-	__attribute__((always_inline));
-
 /**
  * Get Memory Region (MR) <-> Memory Pool (MP) association from txq->mp2mr[].
  * Add MP to txq->mp2mr[] if it's not registered yet. If mp2mr[] is full,
@@ -292,57 +327,20 @@ txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
  *
  * @param txq
  *   Pointer to TX queue structure.
+ * @param wqe
+ *   Pointer to the last WQE posted in the NIC.
  */
 static inline void
-mlx5_tx_dbrec(struct txq *txq)
+mlx5_tx_dbrec(struct txq *txq, volatile struct mlx5_wqe *wqe)
 {
-	uint8_t *dst = (uint8_t *)((uintptr_t)txq->bf_reg + txq->bf_offset);
-	uint32_t data[4] = {
-		htonl((txq->wqe_ci << 8) | MLX5_OPCODE_SEND),
-		htonl(txq->qp_num_8s),
-		0,
-		0,
-	};
+	uint64_t *dst = (uint64_t *)((uintptr_t)txq->bf_reg);
+	volatile uint64_t *src = ((volatile uint64_t *)wqe);
+
 	rte_wmb();
 	*txq->qp_db = htonl(txq->wqe_ci);
 	/* Ensure ordering between DB record and BF copy. */
 	rte_wmb();
-	rte_mov16(dst, (uint8_t *)data);
-	txq->bf_offset ^= (1 << txq->bf_buf_size);
-}
-
-/**
- * Prefetch a CQE.
- *
- * @param txq
- *   Pointer to TX queue structure.
- * @param cqe_ci
- *   CQE consumer index.
- */
-static inline void
-tx_prefetch_cqe(struct txq *txq, uint16_t ci)
-{
-	volatile struct mlx5_cqe *cqe;
-
-	cqe = &(*txq->cqes)[ci & ((1 << txq->cqe_n) - 1)];
-	rte_prefetch0(cqe);
-}
-
-/**
- * Prefetch a WQE.
- *
- * @param txq
- *   Pointer to TX queue structure.
- * @param  wqe_ci
- *   WQE consumer index.
- */
-static inline void
-tx_prefetch_wqe(struct txq *txq, uint16_t ci)
-{
-	volatile struct mlx5_wqe64 *wqe;
-
-	wqe = &(*txq->wqes)[ci & ((1 << txq->wqe_n) - 1)];
-	rte_prefetch0(wqe);
+	*dst = *src;
 }
 
 /**
@@ -368,7 +366,7 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	unsigned int j = 0;
 	unsigned int max;
 	unsigned int comp;
-	volatile struct mlx5_wqe *wqe = NULL;
+	volatile struct mlx5_wqe_v *wqe = NULL;
 	unsigned int segs_n = 0;
 	struct rte_mbuf *buf = NULL;
 	uint8_t *raw;
@@ -376,8 +374,6 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (unlikely(!pkts_n))
 		return 0;
 	/* Prefetch first packet cacheline. */
-	tx_prefetch_cqe(txq, txq->cq_ci);
-	tx_prefetch_cqe(txq, txq->cq_ci + 1);
 	rte_prefetch0(*pkts);
 	/* Start processing. */
 	txq_complete(txq);
@@ -385,10 +381,14 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (max > elts_n)
 		max -= elts_n;
 	do {
-		volatile struct mlx5_wqe_data_seg *dseg = NULL;
+		volatile rte_v128u32_t *dseg = NULL;
 		uint32_t length;
 		unsigned int ds = 0;
 		uintptr_t addr;
+		uint64_t naddr;
+		uint16_t pkt_inline_sz = MLX5_WQE_DWORD_SIZE;
+		uint16_t ehdr;
+		uint8_t cs_flags = 0;
 #ifdef MLX5_PMD_SOFT_COUNTERS
 		uint32_t total_length = 0;
 #endif
@@ -407,13 +407,15 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		--segs_n;
 		if (!segs_n)
 			--pkts_n;
-		wqe = &(*txq->wqes)[txq->wqe_ci &
-				    ((1 << txq->wqe_n) - 1)].hdr;
-		tx_prefetch_wqe(txq, txq->wqe_ci + 1);
+		wqe = (volatile struct mlx5_wqe_v *)
+			tx_mlx5_wqe(txq, txq->wqe_ci);
+		rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
 		if (pkts_n > 1)
 			rte_prefetch0(*pkts);
 		addr = rte_pktmbuf_mtod(buf, uintptr_t);
 		length = DATA_LEN(buf);
+		ehdr = (((uint8_t *)addr)[1] << 8) |
+		       ((uint8_t *)addr)[0];
 #ifdef MLX5_PMD_SOFT_COUNTERS
 		total_length = length;
 #endif
@@ -431,30 +433,22 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		/* Should we enable HW CKSUM offload */
 		if (buf->ol_flags &
 		    (PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) {
-			wqe->eseg.cs_flags =
-				MLX5_ETH_WQE_L3_CSUM |
-				MLX5_ETH_WQE_L4_CSUM;
-		} else {
-			wqe->eseg.cs_flags = 0;
+			cs_flags = MLX5_ETH_WQE_L3_CSUM | MLX5_ETH_WQE_L4_CSUM;
 		}
-		raw  = (uint8_t *)(uintptr_t)&wqe->eseg.inline_hdr[0];
-		/* Start the know and common part of the WQE structure. */
-		wqe->ctrl[0] = htonl((txq->wqe_ci << 8) | MLX5_OPCODE_SEND);
-		wqe->ctrl[2] = 0;
-		wqe->ctrl[3] = 0;
-		wqe->eseg.rsvd0 = 0;
-		wqe->eseg.rsvd1 = 0;
-		wqe->eseg.mss = 0;
-		wqe->eseg.rsvd2 = 0;
-		/* Start by copying the Ethernet Header. */
-		rte_mov16((uint8_t *)raw, (uint8_t *)addr);
+		raw = ((uint8_t *)(uintptr_t)wqe) + 2 * MLX5_WQE_DWORD_SIZE;
+		/*
+		 * Start by copying the Ethernet header minus the first two
+		 * bytes which will be appended at the end of the Ethernet
+		 * segment.
+		 */
+		memcpy((uint8_t *)raw, ((uint8_t *)addr) + 2, 16);
 		length -= MLX5_WQE_DWORD_SIZE;
 		addr += MLX5_WQE_DWORD_SIZE;
 		/* Replace the Ethernet type by the VLAN if necessary. */
 		if (buf->ol_flags & PKT_TX_VLAN_PKT) {
 			uint32_t vlan = htonl(0x81000000 | buf->vlan_tci);
 
-			memcpy((uint8_t *)(raw + MLX5_WQE_DWORD_SIZE -
+			memcpy((uint8_t *)(raw + MLX5_WQE_DWORD_SIZE - 2 -
 					   sizeof(vlan)),
 			       &vlan, sizeof(vlan));
 			addr -= sizeof(vlan);
@@ -462,14 +456,18 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		}
 		/* Inline if enough room. */
 		if (txq->max_inline != 0) {
-			uintptr_t end =
-				(uintptr_t)&(*txq->wqes)[1 << txq->wqe_n];
+			uintptr_t end = (uintptr_t)
+				(((uintptr_t)txq->wqes) +
+				 (1 << txq->wqe_n) * MLX5_WQE_SIZE);
 			uint16_t max_inline =
 				txq->max_inline * RTE_CACHE_LINE_SIZE;
-			uint16_t pkt_inline_sz = MLX5_WQE_DWORD_SIZE;
 			uint16_t room;
 
-			raw += MLX5_WQE_DWORD_SIZE;
+			/*
+			 * raw starts two bytes before the boundary to
+			 * continue the above copy of packet data.
+			 */
+			raw += MLX5_WQE_DWORD_SIZE - 2;
 			room = end - (uintptr_t)raw;
 			if (room > max_inline) {
 				uintptr_t addr_end = (addr + max_inline) &
@@ -485,24 +483,26 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				/* Sanity check. */
 				assert(addr <= addr_end);
 			}
-			/* Store the inlined packet size in the WQE. */
-			wqe->eseg.inline_hdr_sz = htons(pkt_inline_sz);
 			/*
-			 * 2 DWORDs consumed by the WQE header + 1 DSEG +
+			 * 2 DWORDs consumed by the WQE header + ETH segment +
 			 * the size of the inline part of the packet.
 			 */
 			ds = 2 + MLX5_WQE_DS(pkt_inline_sz - 2);
 			if (length > 0) {
-				dseg = (struct mlx5_wqe_data_seg *)
+				dseg = (volatile rte_v128u32_t *)
 					((uintptr_t)wqe +
 					 (ds * MLX5_WQE_DWORD_SIZE));
 				if ((uintptr_t)dseg >= end)
-					dseg = (struct mlx5_wqe_data_seg *)
-						((uintptr_t)&(*txq->wqes)[0]);
+					dseg = (volatile rte_v128u32_t *)
+					       txq->wqes;
 				goto use_dseg;
 			} else if (!segs_n) {
 				goto next_pkt;
 			} else {
+				/* dseg will be advance as part of next_seg */
+				dseg = (volatile rte_v128u32_t *)
+					((uintptr_t)wqe +
+					 ((ds - 1) * MLX5_WQE_DWORD_SIZE));
 				goto next_seg;
 			}
 		} else {
@@ -510,16 +510,17 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			 * No inline has been done in the packet, only the
 			 * Ethernet Header as been stored.
 			 */
-			wqe->eseg.inline_hdr_sz = htons(MLX5_WQE_DWORD_SIZE);
-			dseg = (struct mlx5_wqe_data_seg *)
+			dseg = (volatile rte_v128u32_t *)
 				((uintptr_t)wqe + (3 * MLX5_WQE_DWORD_SIZE));
 			ds = 3;
 use_dseg:
 			/* Add the remaining packet as a simple ds. */
-			*dseg = (struct mlx5_wqe_data_seg) {
-				.addr = htonll(addr),
-				.byte_count = htonl(length),
-				.lkey = txq_mp2mr(txq, txq_mb2mp(buf)),
+			naddr = htonll(addr);
+			*dseg = (rte_v128u32_t){
+				htonl(length),
+				txq_mp2mr(txq, txq_mb2mp(buf)),
+				naddr,
+				naddr >> 32,
 			};
 			++ds;
 			if (!segs_n)
@@ -539,9 +540,9 @@ next_seg:
 			unsigned int n = (txq->wqe_ci + ((ds + 3) / 4)) &
 				((1 << txq->wqe_n) - 1);
 
-			dseg = (struct mlx5_wqe_data_seg *)
-				((uintptr_t)&(*txq->wqes)[n]);
-			tx_prefetch_wqe(txq, n + 1);
+			dseg = (volatile rte_v128u32_t *)
+			       tx_mlx5_wqe(txq, n);
+			rte_prefetch0(tx_mlx5_wqe(txq, n + 1));
 		} else {
 			++dseg;
 		}
@@ -553,10 +554,12 @@ next_seg:
 		total_length += length;
 #endif
 		/* Store segment information. */
-		*dseg = (struct mlx5_wqe_data_seg) {
-			.addr = htonll(rte_pktmbuf_mtod(buf, uintptr_t)),
-			.byte_count = htonl(length),
-			.lkey = txq_mp2mr(txq, txq_mb2mp(buf)),
+		naddr = htonll(rte_pktmbuf_mtod(buf, uintptr_t));
+		*dseg = (rte_v128u32_t){
+			htonl(length),
+			txq_mp2mr(txq, txq_mb2mp(buf)),
+			naddr,
+			naddr >> 32,
 		};
 		(*txq->elts)[elts_head] = buf;
 		elts_head = (elts_head + 1) & (elts_n - 1);
@@ -568,7 +571,19 @@ next_seg:
 			--pkts_n;
 next_pkt:
 		++i;
-		wqe->ctrl[1] = htonl(txq->qp_num_8s | ds);
+		/* Initialize known and common part of the WQE structure. */
+		wqe->ctrl = (rte_v128u32_t){
+			htonl((txq->wqe_ci << 8) | MLX5_OPCODE_SEND),
+			htonl(txq->qp_num_8s | ds),
+			0,
+			0,
+		};
+		wqe->eseg = (rte_v128u32_t){
+			0,
+			cs_flags,
+			0,
+			(ehdr << 16) | htons(pkt_inline_sz),
+		};
 		txq->wqe_ci += (ds + 3) / 4;
 #ifdef MLX5_PMD_SOFT_COUNTERS
 		/* Increment sent bytes counter. */
@@ -581,10 +596,13 @@ next_pkt:
 	/* Check whether completion threshold has been reached. */
 	comp = txq->elts_comp + i + j;
 	if (comp >= MLX5_TX_COMP_THRESH) {
+		volatile struct mlx5_wqe_ctrl *w =
+			(volatile struct mlx5_wqe_ctrl *)wqe;
+
 		/* Request completion on last WQE. */
-		wqe->ctrl[2] = htonl(8);
+		w->ctrl2 = htonl(8);
 		/* Save elts_head in unused "immediate" field of WQE. */
-		wqe->ctrl[3] = elts_head;
+		w->ctrl3 = elts_head;
 		txq->elts_comp = 0;
 	} else {
 		txq->elts_comp = comp;
@@ -594,7 +612,7 @@ next_pkt:
 	txq->stats.opackets += i;
 #endif
 	/* Ring QP doorbell. */
-	mlx5_tx_dbrec(txq);
+	mlx5_tx_dbrec(txq, (volatile struct mlx5_wqe *)wqe);
 	txq->elts_head = elts_head;
 	return i;
 }
@@ -615,20 +633,20 @@ mlx5_mpw_new(struct txq *txq, struct mlx5_mpw *mpw, uint32_t length)
 	uint16_t idx = txq->wqe_ci & ((1 << txq->wqe_n) - 1);
 	volatile struct mlx5_wqe_data_seg (*dseg)[MLX5_MPW_DSEG_MAX] =
 		(volatile struct mlx5_wqe_data_seg (*)[])
-		(uintptr_t)&(*txq->wqes)[(idx + 1) & ((1 << txq->wqe_n) - 1)];
+		tx_mlx5_wqe(txq, idx + 1);
 
 	mpw->state = MLX5_MPW_STATE_OPENED;
 	mpw->pkts_n = 0;
 	mpw->len = length;
 	mpw->total_len = 0;
-	mpw->wqe = (volatile struct mlx5_wqe *)&(*txq->wqes)[idx].hdr;
+	mpw->wqe = (volatile struct mlx5_wqe *)tx_mlx5_wqe(txq, idx);
 	mpw->wqe->eseg.mss = htons(length);
 	mpw->wqe->eseg.inline_hdr_sz = 0;
 	mpw->wqe->eseg.rsvd0 = 0;
 	mpw->wqe->eseg.rsvd1 = 0;
 	mpw->wqe->eseg.rsvd2 = 0;
 	mpw->wqe->ctrl[0] = htonl((MLX5_OPC_MOD_MPW << 24) |
-				  (txq->wqe_ci << 8) | MLX5_OPCODE_LSO_MPW);
+				  (txq->wqe_ci << 8) | MLX5_OPCODE_TSO);
 	mpw->wqe->ctrl[2] = 0;
 	mpw->wqe->ctrl[3] = 0;
 	mpw->data.dseg[0] = (volatile struct mlx5_wqe_data_seg *)
@@ -663,8 +681,8 @@ mlx5_mpw_close(struct txq *txq, struct mlx5_mpw *mpw)
 		++txq->wqe_ci;
 	else
 		txq->wqe_ci += 2;
-	tx_prefetch_wqe(txq, txq->wqe_ci);
-	tx_prefetch_wqe(txq, txq->wqe_ci + 1);
+	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
+	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
 }
 
 /**
@@ -697,9 +715,8 @@ mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (unlikely(!pkts_n))
 		return 0;
 	/* Prefetch first packet cacheline. */
-	tx_prefetch_cqe(txq, txq->cq_ci);
-	tx_prefetch_wqe(txq, txq->wqe_ci);
-	tx_prefetch_wqe(txq, txq->wqe_ci + 1);
+	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
+	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
 	/* Start processing. */
 	txq_complete(txq);
 	max = (elts_n - (elts_head - txq->elts_tail));
@@ -802,7 +819,7 @@ mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	/* Ring QP doorbell. */
 	if (mpw.state == MLX5_MPW_STATE_OPENED)
 		mlx5_mpw_close(txq, &mpw);
-	mlx5_tx_dbrec(txq);
+	mlx5_tx_dbrec(txq, mpw.wqe);
 	txq->elts_head = elts_head;
 	return i;
 }
@@ -827,10 +844,10 @@ mlx5_mpw_inline_new(struct txq *txq, struct mlx5_mpw *mpw, uint32_t length)
 	mpw->pkts_n = 0;
 	mpw->len = length;
 	mpw->total_len = 0;
-	mpw->wqe = (volatile struct mlx5_wqe *)&(*txq->wqes)[idx].hdr;
+	mpw->wqe = (volatile struct mlx5_wqe *)tx_mlx5_wqe(txq, idx);
 	mpw->wqe->ctrl[0] = htonl((MLX5_OPC_MOD_MPW << 24) |
 				  (txq->wqe_ci << 8) |
-				  MLX5_OPCODE_LSO_MPW);
+				  MLX5_OPCODE_TSO);
 	mpw->wqe->ctrl[2] = 0;
 	mpw->wqe->ctrl[3] = 0;
 	mpw->wqe->eseg.mss = htons(length);
@@ -902,9 +919,8 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 	if (unlikely(!pkts_n))
 		return 0;
 	/* Prefetch first packet cacheline. */
-	tx_prefetch_cqe(txq, txq->cq_ci);
-	tx_prefetch_wqe(txq, txq->wqe_ci);
-	tx_prefetch_wqe(txq, txq->wqe_ci + 1);
+	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
+	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
 	/* Start processing. */
 	txq_complete(txq);
 	max = (elts_n - (elts_head - txq->elts_tail));
@@ -1005,14 +1021,15 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 			addr = rte_pktmbuf_mtod(buf, uintptr_t);
 			(*txq->elts)[elts_head] = buf;
 			/* Maximum number of bytes before wrapping. */
-			max = ((uintptr_t)&(*txq->wqes)[1 << txq->wqe_n] -
+			max = ((((uintptr_t)(txq->wqes)) +
+				(1 << txq->wqe_n) *
+				MLX5_WQE_SIZE) -
 			       (uintptr_t)mpw.data.raw);
 			if (length > max) {
 				rte_memcpy((void *)(uintptr_t)mpw.data.raw,
 					   (void *)addr,
 					   max);
-				mpw.data.raw =
-					(volatile void *)&(*txq->wqes)[0];
+				mpw.data.raw = (volatile void *)txq->wqes;
 				rte_memcpy((void *)(uintptr_t)mpw.data.raw,
 					   (void *)(addr + max),
 					   length - max);
@@ -1024,9 +1041,8 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 				mpw.data.raw += length;
 			}
 			if ((uintptr_t)mpw.data.raw ==
-			    (uintptr_t)&(*txq->wqes)[1 << txq->wqe_n])
-				mpw.data.raw =
-					(volatile void *)&(*txq->wqes)[0];
+			    (uintptr_t)tx_mlx5_wqe(txq, 1 << txq->wqe_n))
+				mpw.data.raw = (volatile void *)txq->wqes;
 			++mpw.pkts_n;
 			++j;
 			if (mpw.pkts_n == MLX5_MPW_DSEG_MAX) {
@@ -1071,7 +1087,7 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 		mlx5_mpw_inline_close(txq, &mpw);
 	else if (mpw.state == MLX5_MPW_STATE_OPENED)
 		mlx5_mpw_close(txq, &mpw);
-	mlx5_tx_dbrec(txq);
+	mlx5_tx_dbrec(txq, mpw.wqe);
 	txq->elts_head = elts_head;
 	return i;
 }
@@ -1088,34 +1104,33 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
  *   Packet type for struct rte_mbuf.
  */
 static inline uint32_t
-rxq_cq_to_pkt_type(volatile struct mlx5_cqe64 *cqe)
+rxq_cq_to_pkt_type(volatile struct mlx5_cqe *cqe)
 {
 	uint32_t pkt_type;
 	uint8_t flags = cqe->l4_hdr_type_etc;
-	uint8_t info = cqe->rsvd0[0];
 
-	if (info & IBV_EXP_CQ_RX_TUNNEL_PACKET)
+	if (cqe->pkt_info & MLX5_CQE_RX_TUNNEL_PACKET)
 		pkt_type =
 			TRANSPOSE(flags,
-				  IBV_EXP_CQ_RX_OUTER_IPV4_PACKET,
-				  RTE_PTYPE_L3_IPV4) |
+				  MLX5_CQE_RX_OUTER_IPV4_PACKET,
+				  RTE_PTYPE_L3_IPV4_EXT_UNKNOWN) |
 			TRANSPOSE(flags,
-				  IBV_EXP_CQ_RX_OUTER_IPV6_PACKET,
-				  RTE_PTYPE_L3_IPV6) |
+				  MLX5_CQE_RX_OUTER_IPV6_PACKET,
+				  RTE_PTYPE_L3_IPV6_EXT_UNKNOWN) |
 			TRANSPOSE(flags,
-				  IBV_EXP_CQ_RX_IPV4_PACKET,
-				  RTE_PTYPE_INNER_L3_IPV4) |
+				  MLX5_CQE_RX_IPV4_PACKET,
+				  RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN) |
 			TRANSPOSE(flags,
-				  IBV_EXP_CQ_RX_IPV6_PACKET,
-				  RTE_PTYPE_INNER_L3_IPV6);
+				  MLX5_CQE_RX_IPV6_PACKET,
+				  RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN);
 	else
 		pkt_type =
 			TRANSPOSE(flags,
 				  MLX5_CQE_L3_HDR_TYPE_IPV6,
-				  RTE_PTYPE_L3_IPV6) |
+				  RTE_PTYPE_L3_IPV6_EXT_UNKNOWN) |
 			TRANSPOSE(flags,
 				  MLX5_CQE_L3_HDR_TYPE_IPV4,
-				  RTE_PTYPE_L3_IPV4);
+				  RTE_PTYPE_L3_IPV4_EXT_UNKNOWN);
 	return pkt_type;
 }
 
@@ -1128,14 +1143,16 @@ rxq_cq_to_pkt_type(volatile struct mlx5_cqe64 *cqe)
  *   Pointer to RX queue.
  * @param cqe
  *   CQE to process.
+ * @param[out] rss_hash
+ *   Packet RSS Hash result.
  *
  * @return
  *   Packet size in bytes (0 if there is none), -1 in case of completion
  *   with error.
  */
 static inline int
-mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
-		 uint16_t cqe_cnt)
+mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe *cqe,
+		 uint16_t cqe_cnt, uint32_t *rss_hash)
 {
 	struct rxq_zip *zip = &rxq->zip;
 	uint16_t cqe_n = cqe_cnt + 1;
@@ -1145,9 +1162,10 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
 	if (zip->ai) {
 		volatile struct mlx5_mini_cqe8 (*mc)[8] =
 			(volatile struct mlx5_mini_cqe8 (*)[8])
-			(uintptr_t)(&(*rxq->cqes)[zip->ca & cqe_cnt].cqe64);
+			(uintptr_t)(&(*rxq->cqes)[zip->ca & cqe_cnt]);
 
 		len = ntohl((*mc)[zip->ai & 7].byte_cnt);
+		*rss_hash = ntohl((*mc)[zip->ai & 7].rx_hash_result);
 		if ((++zip->ai & 7) == 0) {
 			/*
 			 * Increment consumer index to skip the number of
@@ -1158,11 +1176,11 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
 			zip->na += 8;
 		}
 		if (unlikely(rxq->zip.ai == rxq->zip.cqe_cnt)) {
-			uint16_t idx = rxq->cq_ci;
+			uint16_t idx = rxq->cq_ci + 1;
 			uint16_t end = zip->cq_ci;
 
 			while (idx != end) {
-				(*rxq->cqes)[idx & cqe_cnt].cqe64.op_own =
+				(*rxq->cqes)[idx & cqe_cnt].op_own =
 					MLX5_CQE_INVALIDATE;
 				++idx;
 			}
@@ -1174,7 +1192,7 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
 		int ret;
 		int8_t op_own;
 
-		ret = check_cqe64(cqe, cqe_n, rxq->cq_ci);
+		ret = check_cqe(cqe, cqe_n, rxq->cq_ci);
 		if (unlikely(ret == 1))
 			return 0;
 		++rxq->cq_ci;
@@ -1183,7 +1201,7 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
 			volatile struct mlx5_mini_cqe8 (*mc)[8] =
 				(volatile struct mlx5_mini_cqe8 (*)[8])
 				(uintptr_t)(&(*rxq->cqes)[rxq->cq_ci &
-							  cqe_cnt].cqe64);
+							  cqe_cnt]);
 
 			/* Fix endianness. */
 			zip->cqe_cnt = ntohl(cqe->byte_cnt);
@@ -1202,9 +1220,11 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
 			zip->cq_ci = rxq->cq_ci + zip->cqe_cnt;
 			/* Get packet size to return. */
 			len = ntohl((*mc)[0].byte_cnt);
+			*rss_hash = ntohl((*mc)[0].rx_hash_result);
 			zip->ai = 1;
 		} else {
 			len = ntohl(cqe->byte_cnt);
+			*rss_hash = ntohl(cqe->rx_hash_res);
 		}
 		/* Error while receiving packet. */
 		if (unlikely(MLX5_CQE_OPCODE(op_own) == MLX5_CQE_RESP_ERR))
@@ -1225,38 +1245,32 @@ mlx5_rx_poll_len(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe,
  *   Offload flags (ol_flags) for struct rte_mbuf.
  */
 static inline uint32_t
-rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe64 *cqe)
+rxq_cq_to_ol_flags(struct rxq *rxq, volatile struct mlx5_cqe *cqe)
 {
 	uint32_t ol_flags = 0;
 	uint8_t l3_hdr = (cqe->l4_hdr_type_etc) & MLX5_CQE_L3_HDR_TYPE_MASK;
 	uint8_t l4_hdr = (cqe->l4_hdr_type_etc) & MLX5_CQE_L4_HDR_TYPE_MASK;
-	uint8_t info = cqe->rsvd0[0];
 
 	if ((l3_hdr == MLX5_CQE_L3_HDR_TYPE_IPV4) ||
 	    (l3_hdr == MLX5_CQE_L3_HDR_TYPE_IPV6))
-		ol_flags |=
-			(!(cqe->hds_ip_ext & MLX5_CQE_L3_OK) *
-			 PKT_RX_IP_CKSUM_BAD);
+		ol_flags |= TRANSPOSE(cqe->hds_ip_ext,
+				      MLX5_CQE_L3_OK,
+				      PKT_RX_IP_CKSUM_GOOD);
 	if ((l4_hdr == MLX5_CQE_L4_HDR_TYPE_TCP) ||
 	    (l4_hdr == MLX5_CQE_L4_HDR_TYPE_TCP_EMP_ACK) ||
 	    (l4_hdr == MLX5_CQE_L4_HDR_TYPE_TCP_ACK) ||
 	    (l4_hdr == MLX5_CQE_L4_HDR_TYPE_UDP))
+		ol_flags |= TRANSPOSE(cqe->hds_ip_ext,
+				      MLX5_CQE_L4_OK,
+				      PKT_RX_L4_CKSUM_GOOD);
+	if ((cqe->pkt_info & MLX5_CQE_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
 		ol_flags |=
-			(!(cqe->hds_ip_ext & MLX5_CQE_L4_OK) *
-			 PKT_RX_L4_CKSUM_BAD);
-	/*
-	 * PKT_RX_IP_CKSUM_BAD and PKT_RX_L4_CKSUM_BAD are used in place
-	 * of PKT_RX_EIP_CKSUM_BAD because the latter is not functional
-	 * (its value is 0).
-	 */
-	if ((info & IBV_EXP_CQ_RX_TUNNEL_PACKET) && (rxq->csum_l2tun))
-		ol_flags |=
-			TRANSPOSE(~cqe->l4_hdr_type_etc,
-				  IBV_EXP_CQ_RX_OUTER_IP_CSUM_OK,
-				  PKT_RX_IP_CKSUM_BAD) |
-			TRANSPOSE(~cqe->l4_hdr_type_etc,
-				  IBV_EXP_CQ_RX_OUTER_TCP_UDP_CSUM_OK,
-				  PKT_RX_L4_CKSUM_BAD);
+			TRANSPOSE(cqe->l4_hdr_type_etc,
+				  MLX5_CQE_RX_OUTER_IP_CSUM_OK,
+				  PKT_RX_IP_CKSUM_GOOD) |
+			TRANSPOSE(cqe->l4_hdr_type_etc,
+				  MLX5_CQE_RX_OUTER_TCP_UDP_CSUM_OK,
+				  PKT_RX_L4_CKSUM_GOOD);
 	return ol_flags;
 }
 
@@ -1282,16 +1296,17 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	const unsigned int sges_n = rxq->sges_n;
 	struct rte_mbuf *pkt = NULL;
 	struct rte_mbuf *seg = NULL;
-	volatile struct mlx5_cqe64 *cqe =
-		&(*rxq->cqes)[rxq->cq_ci & cqe_cnt].cqe64;
+	volatile struct mlx5_cqe *cqe =
+		&(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
 	unsigned int i = 0;
 	unsigned int rq_ci = rxq->rq_ci << sges_n;
-	int len;
+	int len; /* keep its value across iterations. */
 
 	while (pkts_n) {
 		unsigned int idx = rq_ci & wqe_cnt;
 		volatile struct mlx5_wqe_data_seg *wqe = &(*rxq->wqes)[idx];
 		struct rte_mbuf *rep = (*rxq->elts)[idx];
+		uint32_t rss_hash_res = 0;
 
 		if (pkt)
 			NEXT(seg) = rep;
@@ -1311,17 +1326,18 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			}
 			while (pkt != seg) {
 				assert(pkt != (*rxq->elts)[idx]);
-				seg = NEXT(pkt);
+				rep = NEXT(pkt);
 				rte_mbuf_refcnt_set(pkt, 0);
 				__rte_mbuf_raw_free(pkt);
-				pkt = seg;
+				pkt = rep;
 			}
 			break;
 		}
 		if (!pkt) {
-			cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt].cqe64;
-			len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt);
-			if (len == 0) {
+			cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
+			len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt,
+					       &rss_hash_res);
+			if (!len) {
 				rte_mbuf_refcnt_set(rep, 0);
 				__rte_mbuf_raw_free(rep);
 				break;
@@ -1338,12 +1354,26 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			/* Update packet information. */
 			pkt->packet_type = 0;
 			pkt->ol_flags = 0;
+			if (rss_hash_res && rxq->rss_hash) {
+				pkt->hash.rss = rss_hash_res;
+				pkt->ol_flags = PKT_RX_RSS_HASH;
+			}
+			if (rxq->mark &&
+			    ((cqe->sop_drop_qpn !=
+			      htonl(MLX5_FLOW_MARK_INVALID)) ||
+			     (cqe->sop_drop_qpn !=
+			      htonl(MLX5_FLOW_MARK_DEFAULT)))) {
+				pkt->hash.fdir.hi =
+					mlx5_flow_mark_get(cqe->sop_drop_qpn);
+				pkt->ol_flags &= ~PKT_RX_RSS_HASH;
+				pkt->ol_flags |= PKT_RX_FDIR | PKT_RX_FDIR_ID;
+			}
 			if (rxq->csum | rxq->csum_l2tun | rxq->vlan_strip |
 			    rxq->crc_present) {
 				if (rxq->csum) {
 					pkt->packet_type =
 						rxq_cq_to_pkt_type(cqe);
-					pkt->ol_flags =
+					pkt->ol_flags |=
 						rxq_cq_to_ol_flags(rxq, cqe);
 				}
 				if (cqe->l4_hdr_type_etc &