net/mlx5: add connection tracking context update
authorBing Zhao <bingz@nvidia.com>
Wed, 5 May 2021 12:23:24 +0000 (15:23 +0300)
committerRaslan Darawsheh <rasland@nvidia.com>
Wed, 5 May 2021 12:30:16 +0000 (14:30 +0200)
When updating a connection tracking context, two separate parts
could be updated.
First, the direction. This will only update the traffic direction
recorded in the software for flow creation.
Second, the TCP parameters. The hardware context will be updated
via the WQE. This update will be blocked until the hardware status
is updated and ready for the next flow creation.

Signed-off-by: Bing Zhao <bingz@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
drivers/net/mlx5/mlx5_flow_dv.c

index 618ce6b..eaa97f7 100644 (file)
@@ -14246,6 +14246,60 @@ __flow_dv_action_rss_update(struct rte_eth_dev *dev, uint32_t idx,
        return ret;
 }
 
+/*
+ * Updates in place conntrack context or direction.
+ * Context update should be synchronized.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] idx
+ *   The conntrack object ID to be updated.
+ * @param[in] update
+ *   Pointer to the structure of information to update.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL. Initialized in case of
+ *   error only.
+ *
+ * @return
+ *   0 on success, otherwise negative errno value.
+ */
+static int
+__flow_dv_action_ct_update(struct rte_eth_dev *dev, uint32_t idx,
+                          const struct rte_flow_modify_conntrack *update,
+                          struct rte_flow_error *error)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_aso_ct_action *ct;
+       const struct rte_flow_action_conntrack *new_prf;
+       int ret = 0;
+
+       ct = flow_aso_ct_get_by_idx(dev, idx);
+       if (!ct->refcnt)
+               return rte_flow_error_set(error, ENOMEM,
+                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                         NULL,
+                                         "CT object is inactive");
+       new_prf = &update->new_ct;
+       if (update->direction)
+               ct->is_original = !!new_prf->is_original_dir;
+       if (update->state) {
+               ret = mlx5_aso_ct_update_by_wqe(priv->sh, ct, new_prf);
+               if (ret)
+                       return rte_flow_error_set(error, EIO,
+                                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                       NULL,
+                                       "Failed to send CT context update WQE");
+               /* Block until ready or a failure. */
+               ret = mlx5_aso_ct_available(priv->sh, ct);
+               if (ret)
+                       rte_flow_error_set(error, rte_errno,
+                                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                          NULL,
+                                          "Timeout to get the CT update");
+       }
+       return ret;
+}
+
 /**
  * Updates in place shared action configuration, lock free,
  * (mutex should be acquired by caller).
@@ -14281,6 +14335,8 @@ flow_dv_action_update(struct rte_eth_dev *dev,
        case MLX5_INDIRECT_ACTION_TYPE_RSS:
                action_conf = ((const struct rte_flow_action *)update)->conf;
                return __flow_dv_action_rss_update(dev, idx, action_conf, err);
+       case MLX5_INDIRECT_ACTION_TYPE_CT:
+               return __flow_dv_action_ct_update(dev, idx, update, err);
        default:
                return rte_flow_error_set(err, ENOTSUP,
                                          RTE_FLOW_ERROR_TYPE_ACTION,