+ cs_flags = txq_ol_cksum_to_cs(buf);
+ /* Copy metadata from mbuf if valid */
+ metadata = buf->ol_flags & PKT_TX_METADATA ? buf->tx_metadata :
+ 0;
+ /* Retrieve packet information. */
+ length = PKT_LEN(buf);
+ /* Start new session if:
+ * - multi-segment packet
+ * - no space left even for a dseg
+ * - next packet can be inlined with a new WQE
+ * - cs_flag differs
+ */
+ if (mpw.state == MLX5_MPW_ENHANCED_STATE_OPENED) {
+ if ((inl_pad + sizeof(struct mlx5_wqe_data_seg) >
+ mpw_room) ||
+ (length <= txq->inline_max_packet_sz &&
+ inl_pad + sizeof(inl_hdr) + length >
+ mpw_room) ||
+ (mpw.wqe->eseg.flow_table_metadata != metadata) ||
+ (mpw.wqe->eseg.cs_flags != cs_flags))
+ max_wqe -= mlx5_empw_close(txq, &mpw);
+ }
+ if (unlikely(mpw.state == MLX5_MPW_STATE_CLOSED)) {
+ /* In Enhanced MPW, inline as much as the budget is
+ * allowed. The remaining space is to be filled with
+ * dsegs. If the title WQEBB isn't padded, it will have
+ * 2 dsegs there.
+ */
+ mpw_room = RTE_MIN(MLX5_WQE_SIZE_MAX,
+ (max_inline ? max_inline :
+ pkts_n * MLX5_WQE_DWORD_SIZE) +
+ MLX5_WQE_SIZE);
+ if (unlikely(max_wqe * MLX5_WQE_SIZE < mpw_room))
+ break;
+ /* Don't pad the title WQEBB to not waste WQ. */
+ mlx5_empw_new(txq, &mpw, 0);
+ mpw_room -= mpw.total_len;
+ inl_pad = 0;
+ do_inline = length <= txq->inline_max_packet_sz &&
+ sizeof(inl_hdr) + length <= mpw_room &&
+ !txq->mpw_hdr_dseg;
+ mpw.wqe->eseg.cs_flags = cs_flags;
+ mpw.wqe->eseg.flow_table_metadata = metadata;
+ } else {
+ /* Evaluate whether the next packet can be inlined.
+ * Inlininig is possible when:
+ * - length is less than configured value
+ * - length fits for remaining space
+ * - not required to fill the title WQEBB with dsegs
+ */
+ do_inline =
+ length <= txq->inline_max_packet_sz &&
+ inl_pad + sizeof(inl_hdr) + length <=
+ mpw_room &&
+ (!txq->mpw_hdr_dseg ||
+ mpw.total_len >= MLX5_WQE_SIZE);
+ }
+ if (max_inline && do_inline) {
+ /* Inline packet into WQE. */
+ unsigned int max;
+
+ assert(mpw.state == MLX5_MPW_ENHANCED_STATE_OPENED);
+ assert(length == DATA_LEN(buf));
+ inl_hdr = rte_cpu_to_be_32(length | MLX5_INLINE_SEG);
+ addr = rte_pktmbuf_mtod(buf, uintptr_t);
+ mpw.data.raw = (volatile void *)
+ ((uintptr_t)mpw.data.raw + inl_pad);
+ max = tx_mlx5_wq_tailroom(txq,
+ (void *)(uintptr_t)mpw.data.raw);
+ /* Copy inline header. */
+ mpw.data.raw = (volatile void *)
+ mlx5_copy_to_wq(
+ (void *)(uintptr_t)mpw.data.raw,
+ &inl_hdr,
+ sizeof(inl_hdr),
+ (void *)(uintptr_t)txq->wqes,
+ max);
+ max = tx_mlx5_wq_tailroom(txq,
+ (void *)(uintptr_t)mpw.data.raw);
+ /* Copy packet data. */
+ mpw.data.raw = (volatile void *)
+ mlx5_copy_to_wq(
+ (void *)(uintptr_t)mpw.data.raw,
+ (void *)addr,
+ length,
+ (void *)(uintptr_t)txq->wqes,
+ max);
+ ++mpw.pkts_n;
+ mpw.total_len += (inl_pad + sizeof(inl_hdr) + length);
+ /* No need to get completion as the entire packet is
+ * copied to WQ. Free the buf right away.
+ */
+ rte_pktmbuf_free_seg(buf);
+ mpw_room -= (inl_pad + sizeof(inl_hdr) + length);
+ /* Add pad in the next packet if any. */
+ inl_pad = (((uintptr_t)mpw.data.raw +
+ (MLX5_WQE_DWORD_SIZE - 1)) &
+ ~(MLX5_WQE_DWORD_SIZE - 1)) -
+ (uintptr_t)mpw.data.raw;
+ } else {
+ /* No inline. Load a dseg of packet pointer. */
+ volatile rte_v128u32_t *dseg;
+
+ assert(mpw.state == MLX5_MPW_ENHANCED_STATE_OPENED);
+ assert((inl_pad + sizeof(*dseg)) <= mpw_room);
+ assert(length == DATA_LEN(buf));
+ if (!tx_mlx5_wq_tailroom(txq,
+ (void *)((uintptr_t)mpw.data.raw
+ + inl_pad)))
+ dseg = (volatile void *)txq->wqes;
+ else
+ dseg = (volatile void *)
+ ((uintptr_t)mpw.data.raw +
+ inl_pad);
+ (*txq->elts)[elts_head++ & elts_m] = buf;
+ addr_64 = rte_cpu_to_be_64(rte_pktmbuf_mtod(buf,
+ uintptr_t));
+ *dseg = (rte_v128u32_t) {
+ rte_cpu_to_be_32(length),
+ mlx5_tx_mb2mr(txq, buf),
+ addr_64,
+ addr_64 >> 32,
+ };
+ mpw.data.raw = (volatile void *)(dseg + 1);
+ mpw.total_len += (inl_pad + sizeof(*dseg));
+ ++j;
+ ++mpw.pkts_n;
+ mpw_room -= (inl_pad + sizeof(*dseg));
+ inl_pad = 0;
+ }
+#ifdef MLX5_PMD_SOFT_COUNTERS
+ /* Increment sent bytes counter. */
+ txq->stats.obytes += length;
+#endif
+ ++i;
+ } while (i < pkts_n);
+ /* Take a shortcut if nothing must be sent. */
+ if (unlikely(i == 0))
+ return 0;
+ /* Check whether completion threshold has been reached. */
+ if (txq->elts_comp + j >= MLX5_TX_COMP_THRESH ||
+ (uint16_t)(txq->wqe_ci - txq->mpw_comp) >=
+ (1 << txq->wqe_n) / MLX5_TX_COMP_THRESH_INLINE_DIV) {
+ volatile struct mlx5_wqe *wqe = mpw.wqe;
+
+ /* A CQE slot must always be available. */
+ assert((1u << txq->cqe_n) - (txq->cq_pi++ - txq->cq_ci));
+ /* Request completion on last WQE. */
+ wqe->ctrl[2] = rte_cpu_to_be_32(8);
+ /* Save elts_head in unused "immediate" field of WQE. */
+ wqe->ctrl[3] = elts_head;
+ txq->elts_comp = 0;
+ txq->mpw_comp = txq->wqe_ci;
+ } else {
+ txq->elts_comp += j;
+ }
+#ifdef MLX5_PMD_SOFT_COUNTERS
+ /* Increment sent packets counter. */
+ txq->stats.opackets += i;
+#endif
+ if (mpw.state == MLX5_MPW_ENHANCED_STATE_OPENED)
+ mlx5_empw_close(txq, &mpw);
+ /* Ring QP doorbell. */
+ mlx5_tx_dbrec(txq, mpw.wqe);
+ txq->elts_head = elts_head;
+ return i;
+}
+
+/**
+ * DPDK callback for TX with Enhanced MPW support.
+ *
+ * @param dpdk_txq
+ * Generic pointer to TX queue structure.
+ * @param[in] pkts
+ * Packets to transmit.
+ * @param pkts_n
+ * Number of packets in array.
+ *
+ * @return
+ * Number of packets successfully transmitted (<= pkts_n).
+ */
+uint16_t
+mlx5_tx_burst_empw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
+{
+ struct mlx5_txq_data *txq = (struct mlx5_txq_data *)dpdk_txq;
+ uint16_t nb_tx = 0;
+
+ while (pkts_n > nb_tx) {
+ uint16_t n;
+ uint16_t ret;
+
+ n = txq_count_contig_multi_seg(&pkts[nb_tx], pkts_n - nb_tx);
+ if (n) {
+ ret = mlx5_tx_burst(dpdk_txq, &pkts[nb_tx], n);
+ if (!ret)
+ break;
+ nb_tx += ret;
+ }
+ n = txq_count_contig_single_seg(&pkts[nb_tx], pkts_n - nb_tx);
+ if (n) {
+ ret = txq_burst_empw(txq, &pkts[nb_tx], n);
+ if (!ret)
+ break;
+ nb_tx += ret;
+ }
+ }
+ return nb_tx;
+}
+
+/**
+ * Translate RX completion flags to packet type.
+ *
+ * @param[in] rxq
+ * Pointer to RX queue structure.
+ * @param[in] cqe
+ * Pointer to CQE.
+ *
+ * @note: fix mlx5_dev_supported_ptypes_get() if any change here.
+ *
+ * @return
+ * Packet type for struct rte_mbuf.
+ */
+static inline uint32_t
+rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe)
+{
+ uint8_t idx;
+ uint8_t pinfo = cqe->pkt_info;
+ uint16_t ptype = cqe->hdr_type_etc;
+
+ /*
+ * The index to the array should have:
+ * bit[1:0] = l3_hdr_type
+ * bit[4:2] = l4_hdr_type
+ * bit[5] = ip_frag
+ * bit[6] = tunneled
+ * bit[7] = outer_l3_type
+ */
+ idx = ((pinfo & 0x3) << 6) | ((ptype & 0xfc00) >> 10);
+ return mlx5_ptype_table[idx] | rxq->tunnel * !!(idx & (1 << 6));
+}
+
+/**
+ * Get size of the next packet for a given CQE. For compressed CQEs, the
+ * consumer index is updated only once all packets of the current one have
+ * been processed.
+ *
+ * @param rxq
+ * Pointer to RX queue.
+ * @param cqe
+ * CQE to process.
+ * @param[out] mcqe
+ * Store pointer to mini-CQE if compressed. Otherwise, the pointer is not
+ * written.
+ *
+ * @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 mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
+ uint16_t cqe_cnt, volatile struct mlx5_mini_cqe8 **mcqe)
+{
+ struct rxq_zip *zip = &rxq->zip;
+ uint16_t cqe_n = cqe_cnt + 1;
+ int len = 0;
+ uint16_t idx, end;
+
+ /* Process compressed data in the CQE and mini arrays. */
+ if (zip->ai) {
+ volatile struct mlx5_mini_cqe8 (*mc)[8] =
+ (volatile struct mlx5_mini_cqe8 (*)[8])
+ (uintptr_t)(&(*rxq->cqes)[zip->ca & cqe_cnt].pkt_info);
+
+ len = rte_be_to_cpu_32((*mc)[zip->ai & 7].byte_cnt);
+ *mcqe = &(*mc)[zip->ai & 7];
+ if ((++zip->ai & 7) == 0) {
+ /* Invalidate consumed CQEs */
+ idx = zip->ca;
+ end = zip->na;
+ while (idx != end) {
+ (*rxq->cqes)[idx & cqe_cnt].op_own =
+ MLX5_CQE_INVALIDATE;
+ ++idx;
+ }