From patchwork Tue Apr 13 00:19:53 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Li Zhang X-Patchwork-Id: 91144 X-Patchwork-Delegate: rasland@nvidia.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id CF61AA0524; Tue, 13 Apr 2021 02:20:05 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id B3A7E16091A; Tue, 13 Apr 2021 02:20:05 +0200 (CEST) Received: from mellanox.co.il (mail-il-dmz.mellanox.com [193.47.165.129]) by mails.dpdk.org (Postfix) with ESMTP id 599B416091F for ; Tue, 13 Apr 2021 02:20:04 +0200 (CEST) Received: from Internal Mail-Server by MTLPINE1 (envelope-from lizh@nvidia.com) with SMTP; 13 Apr 2021 03:19:59 +0300 Received: from nvidia.com (c-235-17-1-009.mtl.labs.mlnx [10.235.17.9]) by labmailer.mlnx (8.13.8/8.13.8) with ESMTP id 13D0JwNJ007332; Tue, 13 Apr 2021 03:19:59 +0300 From: Li Zhang To: dekelp@nvidia.com, orika@nvidia.com, viacheslavo@nvidia.com, matan@nvidia.com, shahafs@nvidia.com Cc: dev@dpdk.org, thomas@monjalon.net, rasland@nvidia.com, roniba@nvidia.com Date: Tue, 13 Apr 2021 03:19:53 +0300 Message-Id: <20210413001954.2000292-4-lizh@nvidia.com> X-Mailer: git-send-email 2.21.0 In-Reply-To: <20210413001954.2000292-1-lizh@nvidia.com> References: <20210401081624.1482490-1-lizh@nvidia.com> <20210413001954.2000292-1-lizh@nvidia.com> MIME-Version: 1.0 Subject: [dpdk-dev] [PATCH v3 3/4] net/mlx5: prepare sub-policy for a flow with meter X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 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" When a flow has a RSS action, the driver splits each sub flow finally is configured with a different HW TIR action. Any RSS action configured in meter policy may cause a split in the flow configuration. To save performance, any TIR action will be configured in different flow table, so policy can be split to sub-policies per TIR in the flow creation time. Create a function to prepare the policy and its sub-policies for a configured flow with meter. Signed-off-by: Li Zhang Acked-by: Matan Azrad --- drivers/net/mlx5/mlx5_flow.h | 10 +++ drivers/net/mlx5/mlx5_flow_dv.c | 144 ++++++++++++++++++++++++++++++++ 2 files changed, 154 insertions(+) diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h index 89e43f2de6..cc9b37b9eb 100644 --- a/drivers/net/mlx5/mlx5_flow.h +++ b/drivers/net/mlx5/mlx5_flow.h @@ -1095,6 +1095,11 @@ typedef int (*mlx5_flow_create_mtr_tbls_t)(struct rte_eth_dev *dev, typedef void (*mlx5_flow_destroy_mtr_tbls_t)(struct rte_eth_dev *dev, struct mlx5_flow_meter_info *fm); typedef void (*mlx5_flow_destroy_mtr_drop_tbls_t)(struct rte_eth_dev *dev); +typedef struct mlx5_flow_meter_sub_policy * + (*mlx5_flow_meter_sub_policy_rss_prepare_t) + (struct rte_eth_dev *dev, + struct mlx5_flow_meter_policy *mtr_policy, + struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS]); typedef uint32_t (*mlx5_flow_mtr_alloc_t) (struct rte_eth_dev *dev); typedef void (*mlx5_flow_mtr_free_t)(struct rte_eth_dev *dev, @@ -1187,6 +1192,7 @@ struct mlx5_flow_driver_ops { mlx5_flow_destroy_policy_rules_t destroy_policy_rules; mlx5_flow_create_def_policy_t create_def_policy; mlx5_flow_destroy_def_policy_t destroy_def_policy; + mlx5_flow_meter_sub_policy_rss_prepare_t meter_sub_policy_rss_prepare; mlx5_flow_counter_alloc_t counter_alloc; mlx5_flow_counter_free_t counter_free; mlx5_flow_counter_query_t counter_query; @@ -1418,6 +1424,10 @@ int mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev, void mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev, struct mlx5_flow_meter_info *fm); void mlx5_flow_destroy_mtr_drop_tbls(struct rte_eth_dev *dev); +struct mlx5_flow_meter_sub_policy *mlx5_flow_meter_sub_policy_rss_prepare + (struct rte_eth_dev *dev, + struct mlx5_flow_meter_policy *mtr_policy, + struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS]); int mlx5_flow_dv_discover_counter_offset_support(struct rte_eth_dev *dev); int mlx5_shared_action_flush(struct rte_eth_dev *dev); void mlx5_release_tunnel_hub(struct mlx5_dev_ctx_shared *sh, uint16_t port_id); diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c index f789f2454e..ed17bd903f 100644 --- a/drivers/net/mlx5/mlx5_flow_dv.c +++ b/drivers/net/mlx5/mlx5_flow_dv.c @@ -14857,6 +14857,149 @@ flow_dv_create_mtr_tbls(struct rte_eth_dev *dev, return -1; } +/** + * Find the policy table for prefix table with RSS. + * + * @param[in] dev + * Pointer to Ethernet device. + * @param[in] policy_id + * Policy index. + * @param[in] rss_desc + * Pointer to rss_desc + * @return + * Pointer to table set on success, NULL otherwise and rte_errno is set. + */ +static struct mlx5_flow_meter_sub_policy * +flow_dv_meter_sub_policy_rss_prepare(struct rte_eth_dev *dev, + struct mlx5_flow_meter_policy *mtr_policy, + struct mlx5_flow_rss_desc *rss_desc[MLX5_MTR_RTE_COLORS]) +{ + struct mlx5_priv *priv = dev->data->dev_private; + struct mlx5_flow_meter_sub_policy *sub_policy = NULL; + uint32_t sub_policy_idx = 0; + uint32_t hrxq_idx[MLX5_MTR_RTE_COLORS] = {0}; + uint32_t i, j; + struct mlx5_hrxq *hrxq; + struct mlx5_flow_handle dh; + struct mlx5_meter_policy_action_container *act_cnt; + uint32_t domain = MLX5_MTR_DOMAIN_INGRESS; + uint16_t sub_policy_num; + + rte_spinlock_lock(&mtr_policy->sl); + for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) { + if (!rss_desc[i]) + continue; + hrxq_idx[i] = mlx5_hrxq_get(dev, rss_desc[i]); + if (!hrxq_idx[i]) { + rte_spinlock_unlock(&mtr_policy->sl); + return NULL; + } + } + sub_policy_num = (mtr_policy->sub_policy_num >> + (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) & + MLX5_MTR_SUB_POLICY_NUM_MASK; + for (i = 0; i < sub_policy_num; + i++) { + for (j = 0; j < MLX5_MTR_RTE_COLORS; j++) { + if (rss_desc[j] && + hrxq_idx[j] != + mtr_policy->sub_policys[domain][i]->rix_hrxq[j]) + break; + } + if (j >= MLX5_MTR_RTE_COLORS) { + /* + * Found the sub policy table with + * the same queue per color + */ + rte_spinlock_unlock(&mtr_policy->sl); + for (j = 0; j < MLX5_MTR_RTE_COLORS; j++) + mlx5_hrxq_release(dev, hrxq_idx[j]); + return mtr_policy->sub_policys[domain][i]; + } + } + /* Create sub policy. */ + if (!mtr_policy->sub_policys[domain][0]->rix_hrxq[0]) { + /* Reuse the first dummy sub_policy*/ + sub_policy = mtr_policy->sub_policys[domain][0]; + sub_policy_idx = sub_policy->idx; + } else { + sub_policy = mlx5_ipool_zmalloc + (priv->sh->ipool[MLX5_IPOOL_MTR_POLICY], + &sub_policy_idx); + if (!sub_policy || + sub_policy_idx > MLX5_MAX_SUB_POLICY_TBL_NUM) + goto rss_sub_policy_error; + sub_policy->idx = sub_policy_idx; + sub_policy->main_policy = mtr_policy; + } + for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) { + if (!rss_desc[i]) + continue; + sub_policy->rix_hrxq[i] = hrxq_idx[i]; + /* + * Overwrite the last action from + * RSS action to Queue action. + */ + hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ], + hrxq_idx[i]); + if (!hrxq) { + DRV_LOG(ERR, "Failed to create policy hrxq"); + goto rss_sub_policy_error; + } + act_cnt = &mtr_policy->act_cnt[i]; + if (act_cnt->rix_mark || act_cnt->modify_hdr) { + memset(&dh, 0, sizeof(struct mlx5_flow_handle)); + if (act_cnt->rix_mark) + dh.mark = 1; + dh.fate_action = MLX5_FLOW_FATE_QUEUE; + dh.rix_hrxq = hrxq_idx[i]; + flow_drv_rxq_flags_set(dev, &dh); + } + } + if (__flow_dv_create_policy_acts_rules(dev, mtr_policy, + sub_policy, domain)) { + DRV_LOG(ERR, "Failed to create policy " + "rules per domain."); + goto rss_sub_policy_error; + } + if (sub_policy != mtr_policy->sub_policys[domain][0]) { + i = (mtr_policy->sub_policy_num >> + (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) & + MLX5_MTR_SUB_POLICY_NUM_MASK; + mtr_policy->sub_policys[domain][i] = sub_policy; + i++; + if (i > MLX5_MTR_RSS_MAX_SUB_POLICY) + goto rss_sub_policy_error; + mtr_policy->sub_policy_num &= ~(MLX5_MTR_SUB_POLICY_NUM_MASK << + (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)); + mtr_policy->sub_policy_num |= + (i & MLX5_MTR_SUB_POLICY_NUM_MASK) << + (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain); + } + rte_spinlock_unlock(&mtr_policy->sl); + return sub_policy; +rss_sub_policy_error: + if (sub_policy) { + __flow_dv_destroy_sub_policy_rules(dev, sub_policy); + if (sub_policy != mtr_policy->sub_policys[domain][0]) { + i = (mtr_policy->sub_policy_num >> + (MLX5_MTR_SUB_POLICY_NUM_SHIFT * domain)) & + MLX5_MTR_SUB_POLICY_NUM_MASK; + mtr_policy->sub_policys[domain][i] = NULL; + mlx5_ipool_free + (priv->sh->ipool[MLX5_IPOOL_MTR_POLICY], + sub_policy->idx); + } + } + for (i = 0; i < MLX5_MTR_RTE_COLORS; i++) + mlx5_hrxq_release(dev, hrxq_idx[i]); + if (sub_policy_idx) + mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MTR_POLICY], + sub_policy_idx); + rte_spinlock_unlock(&mtr_policy->sl); + return NULL; +} + /** * Validate the batch counter support in root table. * @@ -15447,6 +15590,7 @@ const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops = { .destroy_policy_rules = flow_dv_destroy_policy_rules, .create_def_policy = flow_dv_create_def_policy, .destroy_def_policy = flow_dv_destroy_def_policy, + .meter_sub_policy_rss_prepare = flow_dv_meter_sub_policy_rss_prepare, .counter_alloc = flow_dv_counter_allocate, .counter_free = flow_dv_counter_free, .counter_query = flow_dv_counter_query,