From patchwork Thu Jun 18 19:11:15 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Matan Azrad X-Patchwork-Id: 71760 X-Patchwork-Delegate: maxime.coquelin@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 1A9D2A04F1; Thu, 18 Jun 2020 21:11:36 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 98AAB1C068; Thu, 18 Jun 2020 21:11:30 +0200 (CEST) Received: from mellanox.co.il (mail-il-dmz.mellanox.com [193.47.165.129]) by dpdk.org (Postfix) with ESMTP id 13D091BFF5 for ; Thu, 18 Jun 2020 21:11:29 +0200 (CEST) Received: from Internal Mail-Server by MTLPINE1 (envelope-from matan@mellanox.com) with SMTP; 18 Jun 2020 22:11:23 +0300 Received: from pegasus25.mtr.labs.mlnx. (pegasus25.mtr.labs.mlnx [10.210.16.10]) by labmailer.mlnx (8.13.8/8.13.8) with ESMTP id 05IJBIZ7005965; Thu, 18 Jun 2020 22:11:23 +0300 From: Matan Azrad To: Maxime Coquelin Cc: dev@dpdk.org Date: Thu, 18 Jun 2020 19:11:15 +0000 Message-Id: <1592507476-442112-3-git-send-email-matan@mellanox.com> X-Mailer: git-send-email 1.8.3.1 In-Reply-To: <1592507476-442112-1-git-send-email-matan@mellanox.com> References: <1592507476-442112-1-git-send-email-matan@mellanox.com> Subject: [dpdk-dev] [PATCH 2/3] vdpa/mlx5: optimize completion queue poll X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" The vDPA driver uses a CQ in order to know when traffic works were completed by the HW. Each traffic completion adds a CQE to the CQ. When the vDPA driver detects CQEs in the CQ, it triggers the guest notification for the corresponding queue and consumes all of them. There is collapse feature in the HW that configures the HW to write all the CQEs in the first entry of the CQ. Using this feature, the vDPA driver can read only the first CQE, validate that the completion counter inside the CQE was changed and if so, to notify the guest. Use CQ collapse feature in order to improve the poll utilization. Signed-off-by: Matan Azrad --- drivers/vdpa/mlx5/mlx5_vdpa_event.c | 73 ++++++++++++++++++++----------------- 1 file changed, 40 insertions(+), 33 deletions(-) diff --git a/drivers/vdpa/mlx5/mlx5_vdpa_event.c b/drivers/vdpa/mlx5/mlx5_vdpa_event.c index 69c8bf6..25f11fd 100644 --- a/drivers/vdpa/mlx5/mlx5_vdpa_event.c +++ b/drivers/vdpa/mlx5/mlx5_vdpa_event.c @@ -127,12 +127,12 @@ struct mlx5_devx_cq_attr attr; size_t pgsize = sysconf(_SC_PAGESIZE); uint32_t umem_size; - int ret; uint16_t event_nums[1] = {0}; + uint16_t cq_size = 1 << log_desc_n; + int ret; cq->log_desc_n = log_desc_n; - umem_size = sizeof(struct mlx5_cqe) * (1 << log_desc_n) + - sizeof(*cq->db_rec) * 2; + umem_size = sizeof(struct mlx5_cqe) * cq_size + sizeof(*cq->db_rec) * 2; cq->umem_buf = rte_zmalloc(__func__, umem_size, 4096); if (!cq->umem_buf) { DRV_LOG(ERR, "Failed to allocate memory for CQ."); @@ -149,13 +149,13 @@ } attr.q_umem_valid = 1; attr.db_umem_valid = 1; - attr.use_first_only = 0; + attr.use_first_only = 1; attr.overrun_ignore = 0; attr.uar_page_id = priv->uar->page_id; attr.q_umem_id = cq->umem_obj->umem_id; attr.q_umem_offset = 0; attr.db_umem_id = cq->umem_obj->umem_id; - attr.db_umem_offset = sizeof(struct mlx5_cqe) * (1 << log_desc_n); + attr.db_umem_offset = sizeof(struct mlx5_cqe) * cq_size; attr.eqn = priv->eqn; attr.log_cq_size = log_desc_n; attr.log_page_size = rte_log2_u32(pgsize); @@ -187,7 +187,8 @@ } cq->callfd = callfd; /* Init CQ to ones to be in HW owner in the start. */ - memset((void *)(uintptr_t)cq->umem_buf, 0xFF, attr.db_umem_offset); + cq->cqes[0].op_own = MLX5_CQE_OWNER_MASK; + cq->cqes[0].wqe_counter = rte_cpu_to_be_16(cq_size - 1); /* First arming. */ mlx5_vdpa_cq_arm(priv, cq); return 0; @@ -203,34 +204,40 @@ container_of(cq, struct mlx5_vdpa_event_qp, cq); const unsigned int cq_size = 1 << cq->log_desc_n; const unsigned int cq_mask = cq_size - 1; - uint32_t total = 0; - int ret; - - do { - volatile struct mlx5_cqe *cqe = cq->cqes + ((cq->cq_ci + total) - & cq_mask); - - ret = check_cqe(cqe, cq_size, cq->cq_ci + total); - switch (ret) { - case MLX5_CQE_STATUS_ERR: + union { + struct { + uint16_t wqe_counter; + uint8_t rsvd5; + uint8_t op_own; + }; + uint32_t word; + } last_word; + uint16_t next_wqe_counter = cq->cq_ci & cq_mask; + uint16_t cur_wqe_counter; + uint16_t comp; + + last_word.word = rte_read32(&cq->cqes[0].wqe_counter); + cur_wqe_counter = rte_be_to_cpu_16(last_word.wqe_counter); + comp = (cur_wqe_counter + 1u - next_wqe_counter) & cq_mask; + if (comp) { + cq->cq_ci += comp; + MLX5_ASSERT(!!(cq->cq_ci & cq_size) == + MLX5_CQE_OWNER(last_word.op_own)); + MLX5_ASSERT(MLX5_CQE_OPCODE(last_word.op_own) != + MLX5_CQE_INVALID); + if (unlikely(!(MLX5_CQE_OPCODE(last_word.op_own) == + MLX5_CQE_RESP_ERR || + MLX5_CQE_OPCODE(last_word.op_own) == + MLX5_CQE_REQ_ERR))) cq->errors++; - /*fall-through*/ - case MLX5_CQE_STATUS_SW_OWN: - total++; - break; - case MLX5_CQE_STATUS_HW_OWN: - default: - break; - } - } while (ret != MLX5_CQE_STATUS_HW_OWN); - rte_io_wmb(); - cq->cq_ci += total; - /* Ring CQ doorbell record. */ - cq->db_rec[0] = rte_cpu_to_be_32(cq->cq_ci); - rte_io_wmb(); - /* Ring SW QP doorbell record. */ - eqp->db_rec[0] = rte_cpu_to_be_32(cq->cq_ci + cq_size); - return total; + rte_io_wmb(); + /* Ring CQ doorbell record. */ + cq->db_rec[0] = rte_cpu_to_be_32(cq->cq_ci); + rte_io_wmb(); + /* Ring SW QP doorbell record. */ + eqp->db_rec[0] = rte_cpu_to_be_32(cq->cq_ci + cq_size); + } + return comp; } static void