net/mlx5: move rearm and clock queue CQ creation to common
authorMichael Baum <michaelba@nvidia.com>
Wed, 6 Jan 2021 08:19:29 +0000 (08:19 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Thu, 14 Jan 2021 09:12:36 +0000 (10:12 +0100)
Using common function for CQ creation at rearm queue and clock queue.

Signed-off-by: Michael Baum <michaelba@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
drivers/net/mlx5/mlx5.h
drivers/net/mlx5/mlx5_rxtx.c
drivers/net/mlx5/mlx5_txpp.c

index 92a5d04..ba2a8c4 100644 (file)
@@ -26,6 +26,7 @@
 #include <mlx5_prm.h>
 #include <mlx5_common_mp.h>
 #include <mlx5_common_mr.h>
+#include <mlx5_common_devx.h>
 
 #include "mlx5_defs.h"
 #include "mlx5_utils.h"
@@ -616,13 +617,7 @@ struct mlx5_flow_id_pool {
 /* Tx pacing queue structure - for Clock and Rearm queues. */
 struct mlx5_txpp_wq {
        /* Completion Queue related data.*/
-       struct mlx5_devx_obj *cq;
-       void *cq_umem;
-       union {
-               volatile void *cq_buf;
-               volatile struct mlx5_cqe *cqes;
-       };
-       volatile uint32_t *cq_dbrec;
+       struct mlx5_devx_cq cq_obj;
        uint32_t cq_ci:24;
        uint32_t arm_sn:2;
        /* Send Queue related data.*/
index 65a1f99..3497765 100644 (file)
@@ -2325,7 +2325,7 @@ mlx5_tx_wseg_init(struct mlx5_txq_data *restrict txq,
 
        qs = RTE_PTR_ADD(wqe, MLX5_WSEG_SIZE);
        qs->max_index = rte_cpu_to_be_32(wci);
-       qs->qpn_cqn = rte_cpu_to_be_32(txq->sh->txpp.clock_queue.cq->id);
+       qs->qpn_cqn = rte_cpu_to_be_32(txq->sh->txpp.clock_queue.cq_obj.cq->id);
        qs->reserved0 = RTE_BE32(0);
        qs->reserved1 = RTE_BE32(0);
 }
index e998de3..bd679c2 100644 (file)
@@ -13,6 +13,7 @@
 #include <rte_eal_paging.h>
 
 #include <mlx5_malloc.h>
+#include <mlx5_common_devx.h>
 
 #include "mlx5.h"
 #include "mlx5_rxtx.h"
@@ -134,12 +135,7 @@ mlx5_txpp_destroy_send_queue(struct mlx5_txpp_wq *wq)
                claim_zero(mlx5_os_umem_dereg(wq->sq_umem));
        if (wq->sq_buf)
                mlx5_free((void *)(uintptr_t)wq->sq_buf);
-       if (wq->cq)
-               claim_zero(mlx5_devx_cmd_destroy(wq->cq));
-       if (wq->cq_umem)
-               claim_zero(mlx5_os_umem_dereg(wq->cq_umem));
-       if (wq->cq_buf)
-               mlx5_free((void *)(uintptr_t)wq->cq_buf);
+       mlx5_devx_cq_destroy(&wq->cq_obj);
        memset(wq, 0, sizeof(*wq));
 }
 
@@ -188,19 +184,6 @@ mlx5_txpp_doorbell_rearm_queue(struct mlx5_dev_ctx_shared *sh, uint16_t ci)
        rte_wmb();
 }
 
-static void
-mlx5_txpp_fill_cqe_rearm_queue(struct mlx5_dev_ctx_shared *sh)
-{
-       struct mlx5_txpp_wq *wq = &sh->txpp.rearm_queue;
-       struct mlx5_cqe *cqe = (struct mlx5_cqe *)(uintptr_t)wq->cqes;
-       uint32_t i;
-
-       for (i = 0; i < MLX5_TXPP_REARM_CQ_SIZE; i++) {
-               cqe->op_own = (MLX5_CQE_INVALID << 4) | MLX5_CQE_OWNER_MASK;
-               ++cqe;
-       }
-}
-
 static void
 mlx5_txpp_fill_wqe_rearm_queue(struct mlx5_dev_ctx_shared *sh)
 {
@@ -236,7 +219,8 @@ mlx5_txpp_fill_wqe_rearm_queue(struct mlx5_dev_ctx_shared *sh)
                index = (i * MLX5_TXPP_REARM / 2 + MLX5_TXPP_REARM / 2) &
                        ((1 << MLX5_CQ_INDEX_WIDTH) - 1);
                qs->max_index = rte_cpu_to_be_32(index);
-               qs->qpn_cqn = rte_cpu_to_be_32(sh->txpp.clock_queue.cq->id);
+               qs->qpn_cqn =
+                          rte_cpu_to_be_32(sh->txpp.clock_queue.cq_obj.cq->id);
        }
 }
 
@@ -246,7 +230,9 @@ mlx5_txpp_create_rearm_queue(struct mlx5_dev_ctx_shared *sh)
 {
        struct mlx5_devx_create_sq_attr sq_attr = { 0 };
        struct mlx5_devx_modify_sq_attr msq_attr = { 0 };
-       struct mlx5_devx_cq_attr cq_attr = { 0 };
+       struct mlx5_devx_cq_attr cq_attr = {
+               .uar_page_id = mlx5_os_get_devx_uar_page_id(sh->tx_uar),
+       };
        struct mlx5_txpp_wq *wq = &sh->txpp.rearm_queue;
        size_t page_size;
        uint32_t umem_size, umem_dbrec;
@@ -257,48 +243,16 @@ mlx5_txpp_create_rearm_queue(struct mlx5_dev_ctx_shared *sh)
                DRV_LOG(ERR, "Failed to get mem page size");
                return -ENOMEM;
        }
-       /* Allocate memory buffer for CQEs and doorbell record. */
-       umem_size = sizeof(struct mlx5_cqe) * MLX5_TXPP_REARM_CQ_SIZE;
-       umem_dbrec = RTE_ALIGN(umem_size, MLX5_DBR_SIZE);
-       umem_size += MLX5_DBR_SIZE;
-       wq->cq_buf = mlx5_malloc(MLX5_MEM_RTE | MLX5_MEM_ZERO, umem_size,
-                                page_size, sh->numa_node);
-       if (!wq->cq_buf) {
-               DRV_LOG(ERR, "Failed to allocate memory for Rearm Queue.");
-               return -ENOMEM;
-       }
-       /* Register allocated buffer in user space with DevX. */
-       wq->cq_umem = mlx5_os_umem_reg(sh->ctx,
-                                              (void *)(uintptr_t)wq->cq_buf,
-                                              umem_size,
-                                              IBV_ACCESS_LOCAL_WRITE);
-       if (!wq->cq_umem) {
-               rte_errno = errno;
-               DRV_LOG(ERR, "Failed to register umem for Rearm Queue.");
-               goto error;
-       }
        /* Create completion queue object for Rearm Queue. */
-       cq_attr.uar_page_id = mlx5_os_get_devx_uar_page_id(sh->tx_uar);
-       cq_attr.eqn = sh->eqn;
-       cq_attr.q_umem_valid = 1;
-       cq_attr.q_umem_offset = 0;
-       cq_attr.q_umem_id = mlx5_os_get_umem_id(wq->cq_umem);
-       cq_attr.db_umem_valid = 1;
-       cq_attr.db_umem_offset = umem_dbrec;
-       cq_attr.db_umem_id = mlx5_os_get_umem_id(wq->cq_umem);
-       cq_attr.log_cq_size = rte_log2_u32(MLX5_TXPP_REARM_CQ_SIZE);
-       cq_attr.log_page_size = rte_log2_u32(page_size);
-       wq->cq = mlx5_devx_cmd_create_cq(sh->ctx, &cq_attr);
-       if (!wq->cq) {
-               rte_errno = errno;
+       ret = mlx5_devx_cq_create(sh->ctx, &wq->cq_obj,
+                                 log2above(MLX5_TXPP_REARM_CQ_SIZE), &cq_attr,
+                                 sh->numa_node);
+       if (ret) {
                DRV_LOG(ERR, "Failed to create CQ for Rearm Queue.");
-               goto error;
+               return ret;
        }
-       wq->cq_dbrec = RTE_PTR_ADD(wq->cq_buf, umem_dbrec);
        wq->cq_ci = 0;
        wq->arm_sn = 0;
-       /* Mark all CQEs initially as invalid. */
-       mlx5_txpp_fill_cqe_rearm_queue(sh);
        /*
         * Allocate memory buffer for Send Queue WQEs.
         * There should be no WQE leftovers in the cyclic queue.
@@ -329,7 +283,7 @@ mlx5_txpp_create_rearm_queue(struct mlx5_dev_ctx_shared *sh)
        sq_attr.state = MLX5_SQC_STATE_RST;
        sq_attr.tis_lst_sz = 1;
        sq_attr.tis_num = sh->tis->id;
-       sq_attr.cqn = wq->cq->id;
+       sq_attr.cqn = wq->cq_obj.cq->id;
        sq_attr.cd_master = 1;
        sq_attr.wq_attr.uar_page = mlx5_os_get_devx_uar_page_id(sh->tx_uar);
        sq_attr.wq_attr.wq_type = MLX5_WQ_TYPE_CYCLIC;
@@ -472,7 +426,11 @@ mlx5_txpp_create_clock_queue(struct mlx5_dev_ctx_shared *sh)
 {
        struct mlx5_devx_create_sq_attr sq_attr = { 0 };
        struct mlx5_devx_modify_sq_attr msq_attr = { 0 };
-       struct mlx5_devx_cq_attr cq_attr = { 0 };
+       struct mlx5_devx_cq_attr cq_attr = {
+               .use_first_only = 1,
+               .overrun_ignore = 1,
+               .uar_page_id = mlx5_os_get_devx_uar_page_id(sh->tx_uar),
+       };
        struct mlx5_txpp_wq *wq = &sh->txpp.clock_queue;
        size_t page_size;
        uint32_t umem_size, umem_dbrec;
@@ -493,46 +451,14 @@ mlx5_txpp_create_clock_queue(struct mlx5_dev_ctx_shared *sh)
        }
        sh->txpp.ts_p = 0;
        sh->txpp.ts_n = 0;
-       /* Allocate memory buffer for CQEs and doorbell record. */
-       umem_size = sizeof(struct mlx5_cqe) * MLX5_TXPP_CLKQ_SIZE;
-       umem_dbrec = RTE_ALIGN(umem_size, MLX5_DBR_SIZE);
-       umem_size += MLX5_DBR_SIZE;
-       wq->cq_buf = mlx5_malloc(MLX5_MEM_RTE | MLX5_MEM_ZERO, umem_size,
-                                       page_size, sh->numa_node);
-       if (!wq->cq_buf) {
-               DRV_LOG(ERR, "Failed to allocate memory for Clock Queue.");
-               return -ENOMEM;
-       }
-       /* Register allocated buffer in user space with DevX. */
-       wq->cq_umem = mlx5_os_umem_reg(sh->ctx,
-                                              (void *)(uintptr_t)wq->cq_buf,
-                                              umem_size,
-                                              IBV_ACCESS_LOCAL_WRITE);
-       if (!wq->cq_umem) {
-               rte_errno = errno;
-               DRV_LOG(ERR, "Failed to register umem for Clock Queue.");
-               goto error;
-       }
        /* Create completion queue object for Clock Queue. */
-       cq_attr.use_first_only = 1;
-       cq_attr.overrun_ignore = 1;
-       cq_attr.uar_page_id = mlx5_os_get_devx_uar_page_id(sh->tx_uar);
-       cq_attr.eqn = sh->eqn;
-       cq_attr.q_umem_valid = 1;
-       cq_attr.q_umem_offset = 0;
-       cq_attr.q_umem_id = mlx5_os_get_umem_id(wq->cq_umem);
-       cq_attr.db_umem_valid = 1;
-       cq_attr.db_umem_offset = umem_dbrec;
-       cq_attr.db_umem_id = mlx5_os_get_umem_id(wq->cq_umem);
-       cq_attr.log_cq_size = rte_log2_u32(MLX5_TXPP_CLKQ_SIZE);
-       cq_attr.log_page_size = rte_log2_u32(page_size);
-       wq->cq = mlx5_devx_cmd_create_cq(sh->ctx, &cq_attr);
-       if (!wq->cq) {
-               rte_errno = errno;
+       ret = mlx5_devx_cq_create(sh->ctx, &wq->cq_obj,
+                                 log2above(MLX5_TXPP_CLKQ_SIZE), &cq_attr,
+                                 sh->numa_node);
+       if (ret) {
                DRV_LOG(ERR, "Failed to create CQ for Clock Queue.");
                goto error;
        }
-       wq->cq_dbrec = RTE_PTR_ADD(wq->cq_buf, umem_dbrec);
        wq->cq_ci = 0;
        /* Allocate memory buffer for Send Queue WQEs. */
        if (sh->txpp.test) {
@@ -578,7 +504,7 @@ mlx5_txpp_create_clock_queue(struct mlx5_dev_ctx_shared *sh)
                sq_attr.static_sq_wq = 1;
        }
        sq_attr.state = MLX5_SQC_STATE_RST;
-       sq_attr.cqn = wq->cq->id;
+       sq_attr.cqn = wq->cq_obj.cq->id;
        sq_attr.packet_pacing_rate_limit_index = sh->txpp.pp_id;
        sq_attr.wq_attr.cd_slave = 1;
        sq_attr.wq_attr.uar_page = mlx5_os_get_devx_uar_page_id(sh->tx_uar);
@@ -629,12 +555,13 @@ mlx5_txpp_cq_arm(struct mlx5_dev_ctx_shared *sh)
        struct mlx5_txpp_wq *aq = &sh->txpp.rearm_queue;
        uint32_t arm_sn = aq->arm_sn << MLX5_CQ_SQN_OFFSET;
        uint32_t db_hi = arm_sn | MLX5_CQ_DBR_CMD_ALL | aq->cq_ci;
-       uint64_t db_be = rte_cpu_to_be_64(((uint64_t)db_hi << 32) | aq->cq->id);
+       uint64_t db_be =
+               rte_cpu_to_be_64(((uint64_t)db_hi << 32) | aq->cq_obj.cq->id);
        base_addr = mlx5_os_get_devx_uar_base_addr(sh->tx_uar);
        uint32_t *addr = RTE_PTR_ADD(base_addr, MLX5_CQ_DOORBELL);
 
        rte_compiler_barrier();
-       aq->cq_dbrec[MLX5_CQ_ARM_DB] = rte_cpu_to_be_32(db_hi);
+       aq->cq_obj.db_rec[MLX5_CQ_ARM_DB] = rte_cpu_to_be_32(db_hi);
        rte_wmb();
 #ifdef RTE_ARCH_64
        *(uint64_t *)addr = db_be;
@@ -732,7 +659,7 @@ static inline void
 mlx5_txpp_update_timestamp(struct mlx5_dev_ctx_shared *sh)
 {
        struct mlx5_txpp_wq *wq = &sh->txpp.clock_queue;
-       struct mlx5_cqe *cqe = (struct mlx5_cqe *)(uintptr_t)wq->cqes;
+       struct mlx5_cqe *cqe = (struct mlx5_cqe *)(uintptr_t)wq->cq_obj.cqes;
        union {
                rte_int128_t u128;
                struct mlx5_cqe_ts cts;
@@ -807,7 +734,7 @@ mlx5_txpp_handle_rearm_queue(struct mlx5_dev_ctx_shared *sh)
        do {
                volatile struct mlx5_cqe *cqe;
 
-               cqe = &wq->cqes[cq_ci & (MLX5_TXPP_REARM_CQ_SIZE - 1)];
+               cqe = &wq->cq_obj.cqes[cq_ci & (MLX5_TXPP_REARM_CQ_SIZE - 1)];
                ret = check_cqe(cqe, MLX5_TXPP_REARM_CQ_SIZE, cq_ci);
                switch (ret) {
                case MLX5_CQE_STATUS_ERR:
@@ -839,7 +766,7 @@ mlx5_txpp_handle_rearm_queue(struct mlx5_dev_ctx_shared *sh)
                }
                /* Update doorbell record to notify hardware. */
                rte_compiler_barrier();
-               *wq->cq_dbrec = rte_cpu_to_be_32(cq_ci);
+               *wq->cq_obj.db_rec = rte_cpu_to_be_32(cq_ci);
                rte_wmb();
                wq->cq_ci = cq_ci;
                /* Fire new requests to Rearm Queue. */
@@ -934,9 +861,8 @@ mlx5_txpp_start_service(struct mlx5_dev_ctx_shared *sh)
        }
        /* Subscribe CQ event to the event channel controlled by the driver. */
        ret = mlx5_os_devx_subscribe_devx_event(sh->txpp.echan,
-                                                  sh->txpp.rearm_queue.cq->obj,
-                                                  sizeof(event_nums),
-                                                  event_nums, 0);
+                                           sh->txpp.rearm_queue.cq_obj.cq->obj,
+                                            sizeof(event_nums), event_nums, 0);
        if (ret) {
                DRV_LOG(ERR, "Failed to subscribe CQE event.");
                rte_errno = errno;
@@ -1138,7 +1064,8 @@ mlx5_txpp_read_clock(struct rte_eth_dev *dev, uint64_t *timestamp)
 
        if (sh->txpp.refcnt) {
                struct mlx5_txpp_wq *wq = &sh->txpp.clock_queue;
-               struct mlx5_cqe *cqe = (struct mlx5_cqe *)(uintptr_t)wq->cqes;
+               struct mlx5_cqe *cqe =
+                               (struct mlx5_cqe *)(uintptr_t)wq->cq_obj.cqes;
                union {
                        rte_int128_t u128;
                        struct mlx5_cqe_ts cts;