+ rxq_ctrl = container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+ if (!rxq_ctrl->irq)
+ return 0;
+ rxq_obj = mlx5_rxq_obj_get(dev, rx_queue_id);
+ if (!rxq_obj) {
+ rte_errno = EINVAL;
+ return -rte_errno;
+ }
+ ret = mlx5_glue->get_cq_event(rxq_obj->channel, &ev_cq, &ev_ctx);
+ if (ret || ev_cq != rxq_obj->cq) {
+ rte_errno = EINVAL;
+ goto exit;
+ }
+ rxq_data->cq_arm_sn++;
+ mlx5_glue->ack_cq_events(rxq_obj->cq, 1);
+ mlx5_rxq_obj_release(rxq_obj);
+ return 0;
+exit:
+ ret = rte_errno; /* Save rte_errno before cleanup. */
+ if (rxq_obj)
+ mlx5_rxq_obj_release(rxq_obj);
+ DRV_LOG(WARNING, "port %u unable to disable interrupt on Rx queue %d",
+ dev->data->port_id, rx_queue_id);
+ rte_errno = ret; /* Restore rte_errno. */
+ return -rte_errno;
+}
+
+/**
+ * Create a CQ Verbs object.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param priv
+ * Pointer to device private data.
+ * @param rxq_data
+ * Pointer to Rx queue data.
+ * @param cqe_n
+ * Number of CQEs in CQ.
+ * @param rxq_obj
+ * Pointer to Rx queue object data.
+ *
+ * @return
+ * The Verbs object initialised, NULL otherwise and rte_errno is set.
+ */
+static struct ibv_cq *
+mlx5_ibv_cq_new(struct rte_eth_dev *dev, struct mlx5_priv *priv,
+ struct mlx5_rxq_data *rxq_data,
+ unsigned int cqe_n, struct mlx5_rxq_obj *rxq_obj)
+{
+ struct {
+ struct ibv_cq_init_attr_ex ibv;
+ struct mlx5dv_cq_init_attr mlx5;
+ } cq_attr;
+
+ cq_attr.ibv = (struct ibv_cq_init_attr_ex){
+ .cqe = cqe_n,
+ .channel = rxq_obj->channel,
+ .comp_mask = 0,
+ };
+ cq_attr.mlx5 = (struct mlx5dv_cq_init_attr){
+ .comp_mask = 0,
+ };
+ if (priv->config.cqe_comp && !rxq_data->hw_timestamp &&
+ !rxq_data->lro) {
+ cq_attr.mlx5.comp_mask |=
+ MLX5DV_CQ_INIT_ATTR_MASK_COMPRESSED_CQE;
+#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
+ cq_attr.mlx5.cqe_comp_res_format =
+ mlx5_rxq_mprq_enabled(rxq_data) ?
+ MLX5DV_CQE_RES_FORMAT_CSUM_STRIDX :
+ MLX5DV_CQE_RES_FORMAT_HASH;
+#else
+ cq_attr.mlx5.cqe_comp_res_format = MLX5DV_CQE_RES_FORMAT_HASH;
+#endif
+ /*
+ * For vectorized Rx, it must not be doubled in order to
+ * make cq_ci and rq_ci aligned.
+ */
+ if (mlx5_rxq_check_vec_support(rxq_data) < 0)
+ cq_attr.ibv.cqe *= 2;
+ } else if (priv->config.cqe_comp && rxq_data->hw_timestamp) {
+ DRV_LOG(DEBUG,
+ "port %u Rx CQE compression is disabled for HW"
+ " timestamp",
+ dev->data->port_id);
+ } else if (priv->config.cqe_comp && rxq_data->lro) {
+ DRV_LOG(DEBUG,
+ "port %u Rx CQE compression is disabled for LRO",
+ dev->data->port_id);
+ }
+#ifdef HAVE_IBV_MLX5_MOD_CQE_128B_PAD
+ if (priv->config.cqe_pad) {
+ cq_attr.mlx5.comp_mask |= MLX5DV_CQ_INIT_ATTR_MASK_FLAGS;
+ cq_attr.mlx5.flags |= MLX5DV_CQ_INIT_ATTR_FLAGS_CQE_PAD;
+ }
+#endif
+ return mlx5_glue->cq_ex_to_cq(mlx5_glue->dv_create_cq(priv->sh->ctx,
+ &cq_attr.ibv,
+ &cq_attr.mlx5));
+}
+
+/**
+ * Create a WQ Verbs object.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param priv
+ * Pointer to device private data.
+ * @param rxq_data
+ * Pointer to Rx queue data.
+ * @param idx
+ * Queue index in DPDK Rx queue array
+ * @param wqe_n
+ * Number of WQEs in WQ.
+ * @param rxq_obj
+ * Pointer to Rx queue object data.
+ *
+ * @return
+ * The Verbs object initialised, NULL otherwise and rte_errno is set.
+ */
+static struct ibv_wq *
+mlx5_ibv_wq_new(struct rte_eth_dev *dev, struct mlx5_priv *priv,
+ struct mlx5_rxq_data *rxq_data, uint16_t idx,
+ unsigned int wqe_n, struct mlx5_rxq_obj *rxq_obj)
+{
+ struct {
+ struct ibv_wq_init_attr ibv;
+#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
+ struct mlx5dv_wq_init_attr mlx5;
+#endif
+ } wq_attr;
+
+ wq_attr.ibv = (struct ibv_wq_init_attr){
+ .wq_context = NULL, /* Could be useful in the future. */
+ .wq_type = IBV_WQT_RQ,
+ /* Max number of outstanding WRs. */
+ .max_wr = wqe_n >> rxq_data->sges_n,
+ /* Max number of scatter/gather elements in a WR. */
+ .max_sge = 1 << rxq_data->sges_n,
+ .pd = priv->sh->pd,
+ .cq = rxq_obj->cq,
+ .comp_mask = IBV_WQ_FLAGS_CVLAN_STRIPPING | 0,
+ .create_flags = (rxq_data->vlan_strip ?
+ IBV_WQ_FLAGS_CVLAN_STRIPPING : 0),
+ };
+ /* By default, FCS (CRC) is stripped by hardware. */
+ if (rxq_data->crc_present) {
+ wq_attr.ibv.create_flags |= IBV_WQ_FLAGS_SCATTER_FCS;
+ wq_attr.ibv.comp_mask |= IBV_WQ_INIT_ATTR_FLAGS;
+ }
+ if (priv->config.hw_padding) {
+#if defined(HAVE_IBV_WQ_FLAG_RX_END_PADDING)
+ wq_attr.ibv.create_flags |= IBV_WQ_FLAG_RX_END_PADDING;
+ wq_attr.ibv.comp_mask |= IBV_WQ_INIT_ATTR_FLAGS;
+#elif defined(HAVE_IBV_WQ_FLAGS_PCI_WRITE_END_PADDING)
+ wq_attr.ibv.create_flags |= IBV_WQ_FLAGS_PCI_WRITE_END_PADDING;
+ wq_attr.ibv.comp_mask |= IBV_WQ_INIT_ATTR_FLAGS;
+#endif
+ }
+#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
+ wq_attr.mlx5 = (struct mlx5dv_wq_init_attr){
+ .comp_mask = 0,
+ };
+ if (mlx5_rxq_mprq_enabled(rxq_data)) {
+ struct mlx5dv_striding_rq_init_attr *mprq_attr =
+ &wq_attr.mlx5.striding_rq_attrs;
+
+ wq_attr.mlx5.comp_mask |= MLX5DV_WQ_INIT_ATTR_MASK_STRIDING_RQ;
+ *mprq_attr = (struct mlx5dv_striding_rq_init_attr){
+ .single_stride_log_num_of_bytes = rxq_data->strd_sz_n,
+ .single_wqe_log_num_of_strides = rxq_data->strd_num_n,
+ .two_byte_shift_en = MLX5_MPRQ_TWO_BYTE_SHIFT,
+ };
+ }
+ rxq_obj->wq = mlx5_glue->dv_create_wq(priv->sh->ctx, &wq_attr.ibv,
+ &wq_attr.mlx5);
+#else
+ rxq_obj->wq = mlx5_glue->create_wq(priv->sh->ctx, &wq_attr.ibv);
+#endif
+ if (rxq_obj->wq) {
+ /*
+ * Make sure number of WRs*SGEs match expectations since a queue
+ * cannot allocate more than "desc" buffers.
+ */
+ if (wq_attr.ibv.max_wr != (wqe_n >> rxq_data->sges_n) ||
+ wq_attr.ibv.max_sge != (1u << rxq_data->sges_n)) {
+ DRV_LOG(ERR,
+ "port %u Rx queue %u requested %u*%u but got"
+ " %u*%u WRs*SGEs",
+ dev->data->port_id, idx,
+ wqe_n >> rxq_data->sges_n,
+ (1 << rxq_data->sges_n),
+ wq_attr.ibv.max_wr, wq_attr.ibv.max_sge);
+ claim_zero(mlx5_glue->destroy_wq(rxq_obj->wq));
+ rxq_obj->wq = NULL;
+ rte_errno = EINVAL;
+ }
+ }
+ return rxq_obj->wq;
+}
+
+/**
+ * Fill common fields of create RQ attributes structure.
+ *
+ * @param rxq_data
+ * Pointer to Rx queue data.
+ * @param cqn
+ * CQ number to use with this RQ.
+ * @param rq_attr
+ * RQ attributes structure to fill..
+ */
+static void
+mlx5_devx_create_rq_attr_fill(struct mlx5_rxq_data *rxq_data, uint32_t cqn,
+ struct mlx5_devx_create_rq_attr *rq_attr)
+{
+ rq_attr->state = MLX5_RQC_STATE_RST;
+ rq_attr->vsd = (rxq_data->vlan_strip) ? 0 : 1;
+ rq_attr->cqn = cqn;
+ rq_attr->scatter_fcs = (rxq_data->crc_present) ? 1 : 0;
+}
+
+/**
+ * Fill common fields of DevX WQ attributes structure.
+ *
+ * @param priv
+ * Pointer to device private data.
+ * @param rxq_ctrl
+ * Pointer to Rx queue control structure.
+ * @param wq_attr
+ * WQ attributes structure to fill..
+ */
+static void
+mlx5_devx_wq_attr_fill(struct mlx5_priv *priv, struct mlx5_rxq_ctrl *rxq_ctrl,
+ struct mlx5_devx_wq_attr *wq_attr)
+{
+ wq_attr->end_padding_mode = priv->config.cqe_pad ?
+ MLX5_WQ_END_PAD_MODE_ALIGN :
+ MLX5_WQ_END_PAD_MODE_NONE;
+ wq_attr->pd = priv->sh->pdn;
+ wq_attr->dbr_addr = rxq_ctrl->dbr_offset;
+ wq_attr->dbr_umem_id = rxq_ctrl->dbr_umem_id;
+ wq_attr->dbr_umem_valid = 1;
+ wq_attr->wq_umem_id = rxq_ctrl->wq_umem->umem_id;
+ wq_attr->wq_umem_valid = 1;
+}
+
+/**
+ * Create a RQ object using DevX.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param idx
+ * Queue index in DPDK Rx queue array
+ * @param cqn
+ * CQ number to use with this RQ.
+ *
+ * @return
+ * The DevX object initialised, NULL otherwise and rte_errno is set.
+ */
+static struct mlx5_devx_obj *
+mlx5_devx_rq_new(struct rte_eth_dev *dev, uint16_t idx, uint32_t cqn)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
+ struct mlx5_rxq_ctrl *rxq_ctrl =
+ container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+ struct mlx5_devx_create_rq_attr rq_attr;
+ uint32_t wqe_n = 1 << (rxq_data->elts_n - rxq_data->sges_n);
+ uint32_t wq_size = 0;
+ uint32_t wqe_size = 0;
+ uint32_t log_wqe_size = 0;
+ void *buf = NULL;
+ struct mlx5_devx_obj *rq;
+
+ memset(&rq_attr, 0, sizeof(rq_attr));
+ /* Fill RQ attributes. */
+ rq_attr.mem_rq_type = MLX5_RQC_MEM_RQ_TYPE_MEMORY_RQ_INLINE;
+ rq_attr.flush_in_error_en = 1;
+ mlx5_devx_create_rq_attr_fill(rxq_data, cqn, &rq_attr);
+ /* Fill WQ attributes for this RQ. */
+ if (mlx5_rxq_mprq_enabled(rxq_data)) {
+ rq_attr.wq_attr.wq_type = MLX5_WQ_TYPE_CYCLIC_STRIDING_RQ;
+ /*
+ * Number of strides in each WQE:
+ * 512*2^single_wqe_log_num_of_strides.
+ */
+ rq_attr.wq_attr.single_wqe_log_num_of_strides =
+ rxq_data->strd_num_n -
+ MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES;
+ /* Stride size = (2^single_stride_log_num_of_bytes)*64B. */
+ rq_attr.wq_attr.single_stride_log_num_of_bytes =
+ rxq_data->strd_sz_n -
+ MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES;
+ wqe_size = sizeof(struct mlx5_wqe_mprq);
+ } else {
+ rq_attr.wq_attr.wq_type = MLX5_WQ_TYPE_CYCLIC;
+ wqe_size = sizeof(struct mlx5_wqe_data_seg);
+ }
+ log_wqe_size = log2above(wqe_size) + rxq_data->sges_n;
+ rq_attr.wq_attr.log_wq_stride = log_wqe_size;
+ rq_attr.wq_attr.log_wq_sz = rxq_data->elts_n - rxq_data->sges_n;
+ /* Calculate and allocate WQ memory space. */
+ wqe_size = 1 << log_wqe_size; /* round up power of two.*/
+ wq_size = wqe_n * wqe_size;
+ buf = rte_calloc_socket(__func__, 1, wq_size, MLX5_WQE_BUF_ALIGNMENT,
+ rxq_ctrl->socket);
+ if (!buf)
+ return NULL;
+ rxq_data->wqes = buf;
+ rxq_ctrl->wq_umem = mlx5_glue->devx_umem_reg(priv->sh->ctx,
+ buf, wq_size, 0);
+ if (!rxq_ctrl->wq_umem) {
+ rte_free(buf);
+ return NULL;
+ }
+ mlx5_devx_wq_attr_fill(priv, rxq_ctrl, &rq_attr.wq_attr);
+ rq = mlx5_devx_cmd_create_rq(priv->sh->ctx, &rq_attr, rxq_ctrl->socket);
+ if (!rq)
+ rxq_release_rq_resources(rxq_ctrl);
+ return rq;
+}
+
+/**
+ * Create the Rx hairpin queue object.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param idx
+ * Queue index in DPDK Rx queue array
+ *
+ * @return
+ * The hairpin DevX object initialised, NULL otherwise and rte_errno is set.
+ */
+static struct mlx5_rxq_obj *
+mlx5_rxq_obj_hairpin_new(struct rte_eth_dev *dev, uint16_t idx)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct mlx5_rxq_data *rxq_data = (*priv->rxqs)[idx];
+ struct mlx5_rxq_ctrl *rxq_ctrl =
+ container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
+ struct mlx5_devx_create_rq_attr attr = { 0 };
+ struct mlx5_rxq_obj *tmpl = NULL;
+ int ret = 0;
+ uint32_t max_wq_data;
+
+ MLX5_ASSERT(rxq_data);
+ MLX5_ASSERT(!rxq_ctrl->obj);
+ tmpl = rte_calloc_socket(__func__, 1, sizeof(*tmpl), 0,
+ rxq_ctrl->socket);
+ if (!tmpl) {
+ DRV_LOG(ERR,
+ "port %u Rx queue %u cannot allocate verbs resources",
+ dev->data->port_id, rxq_data->idx);
+ rte_errno = ENOMEM;
+ goto error;
+ }
+ tmpl->type = MLX5_RXQ_OBJ_TYPE_DEVX_HAIRPIN;
+ tmpl->rxq_ctrl = rxq_ctrl;
+ attr.hairpin = 1;
+ max_wq_data = priv->config.hca_attr.log_max_hairpin_wq_data_sz;
+ /* Jumbo frames > 9KB should be supported, and more packets. */
+ if (priv->config.log_hp_size != (uint32_t)MLX5_ARG_UNSET) {
+ if (priv->config.log_hp_size > max_wq_data) {
+ DRV_LOG(ERR, "total data size %u power of 2 is "
+ "too large for hairpin",
+ priv->config.log_hp_size);
+ rte_errno = ERANGE;
+ return NULL;
+ }
+ attr.wq_attr.log_hairpin_data_sz = priv->config.log_hp_size;
+ } else {
+ attr.wq_attr.log_hairpin_data_sz =
+ (max_wq_data < MLX5_HAIRPIN_JUMBO_LOG_SIZE) ?
+ max_wq_data : MLX5_HAIRPIN_JUMBO_LOG_SIZE;