net/mlx5: add count flow action
authorNelio Laranjeiro <nelio.laranjeiro@6wind.com>
Thu, 12 Jul 2018 09:31:07 +0000 (11:31 +0200)
committerShahaf Shuler <shahafs@mellanox.com>
Thu, 12 Jul 2018 10:12:27 +0000 (12:12 +0200)
This is only supported by Mellanox OFED.

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>
Acked-by: Yongseok Koh <yskoh@mellanox.com>
drivers/net/mlx5/mlx5.h
drivers/net/mlx5/mlx5_flow.c

index 9949cd3..131be33 100644 (file)
@@ -188,6 +188,8 @@ struct priv {
        struct mlx5_drop drop_queue; /* Flow drop queues. */
        struct mlx5_flows flows; /* RTE Flow rules. */
        struct mlx5_flows ctrl_flows; /* Control flow rules. */
+       LIST_HEAD(counters, mlx5_flow_counter) flow_counters;
+       /* Flow counters. */
        struct {
                uint32_t dev_gen; /* Generation number to flush local caches. */
                rte_rwlock_t rwlock; /* MR Lock. */
index 146f1b8..6fa4e30 100644 (file)
@@ -88,6 +88,7 @@ extern const struct eth_dev_ops mlx5_dev_ops_isolate;
 /* Modify a packet. */
 #define MLX5_FLOW_MOD_FLAG (1u << 0)
 #define MLX5_FLOW_MOD_MARK (1u << 1)
+#define MLX5_FLOW_MOD_COUNT (1u << 2)
 
 /* possible L3 layers protocols filtering. */
 #define MLX5_IP_PROTOCOL_TCP 6
@@ -249,6 +250,17 @@ struct mlx5_flow_verbs {
        uint64_t hash_fields; /**< Verbs hash Rx queue hash fields. */
 };
 
+/* Counters information. */
+struct mlx5_flow_counter {
+       LIST_ENTRY(mlx5_flow_counter) next; /**< Pointer to the next counter. */
+       uint32_t shared:1; /**< Share counter ID with other flow rules. */
+       uint32_t ref_cnt:31; /**< Reference counter. */
+       uint32_t id; /**< Counter ID. */
+       struct ibv_counter_set *cs; /**< Holds the counters for the rule. */
+       uint64_t hits; /**< Number of packets matched by the rule. */
+       uint64_t bytes; /**< Number of bytes matched by the rule. */
+};
+
 /* Flow structure. */
 struct rte_flow {
        TAILQ_ENTRY(rte_flow) next; /**< Pointer to the next flow structure. */
@@ -264,6 +276,7 @@ struct rte_flow {
        LIST_HEAD(verbs, mlx5_flow_verbs) verbs; /**< Verbs flows list. */
        struct mlx5_flow_verbs *cur_verbs;
        /**< Current Verbs flow structure being filled. */
+       struct mlx5_flow_counter *counter; /**< Holds Verbs flow counter. */
        struct rte_flow_action_rss rss;/**< RSS context. */
        uint8_t key[MLX5_RSS_HASH_KEY_LEN]; /**< RSS hash key. */
        uint16_t (*queue)[]; /**< Destination queues to redirect traffic to. */
@@ -275,6 +288,7 @@ static const struct rte_flow_ops mlx5_flow_ops = {
        .destroy = mlx5_flow_destroy,
        .flush = mlx5_flow_flush,
        .isolate = mlx5_flow_isolate,
+       .query = mlx5_flow_query,
 };
 
 /* Convert FDIR request to Generic flow. */
@@ -454,6 +468,80 @@ mlx5_flow_adjust_priority(struct rte_eth_dev *dev, struct rte_flow *flow)
        flow->cur_verbs->attr->priority = priority;
 }
 
+/**
+ * Get a flow counter.
+ *
+ * @param[in] dev
+ *   Pointer to Ethernet device.
+ * @param[in] shared
+ *   Indicate if this counter is shared with other flows.
+ * @param[in] id
+ *   Counter identifier.
+ *
+ * @return
+ *   A pointer to the counter, NULL otherwise and rte_errno is set.
+ */
+static struct mlx5_flow_counter *
+mlx5_flow_counter_new(struct rte_eth_dev *dev, uint32_t shared, uint32_t id)
+{
+       struct priv *priv = dev->data->dev_private;
+       struct mlx5_flow_counter *cnt;
+
+       LIST_FOREACH(cnt, &priv->flow_counters, next) {
+               if (cnt->shared != shared)
+                       continue;
+               if (cnt->id != id)
+                       continue;
+               cnt->ref_cnt++;
+               return cnt;
+       }
+#ifdef HAVE_IBV_DEVICE_COUNTERS_SET_SUPPORT
+
+       struct mlx5_flow_counter tmpl = {
+               .shared = shared,
+               .id = id,
+               .cs = mlx5_glue->create_counter_set
+                       (priv->ctx,
+                        &(struct ibv_counter_set_init_attr){
+                                .counter_set_id = id,
+                        }),
+               .hits = 0,
+               .bytes = 0,
+       };
+
+       if (!tmpl.cs) {
+               rte_errno = errno;
+               return NULL;
+       }
+       cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
+       if (!cnt) {
+               rte_errno = ENOMEM;
+               return NULL;
+       }
+       *cnt = tmpl;
+       LIST_INSERT_HEAD(&priv->flow_counters, cnt, next);
+       return cnt;
+#endif
+       rte_errno = ENOTSUP;
+       return NULL;
+}
+
+/**
+ * Release a flow counter.
+ *
+ * @param[in] counter
+ *   Pointer to the counter handler.
+ */
+static void
+mlx5_flow_counter_release(struct mlx5_flow_counter *counter)
+{
+       if (--counter->ref_cnt == 0) {
+               claim_zero(mlx5_glue->destroy_counter_set(counter->cs));
+               LIST_REMOVE(counter, next);
+               rte_free(counter);
+       }
+}
+
 /**
  * Verify the @p attributes will be correctly understood by the NIC and store
  * them in the @p flow if everything is correct.
@@ -2128,6 +2216,70 @@ mlx5_flow_action_mark(const struct rte_flow_action *action,
        return size;
 }
 
+/**
+ * Convert the @p action into a Verbs specification after ensuring the NIC
+ * will understand and process it correctly.
+ * If the necessary size for the conversion is greater than the @p flow_size,
+ * nothing is written in @p flow, the validation is still performed.
+ *
+ * @param action[in]
+ *   Action configuration.
+ * @param flow[in, out]
+ *   Pointer to flow structure.
+ * @param flow_size[in]
+ *   Size in bytes of the available space in @p flow, if too small, nothing is
+ *   written.
+ * @param error[int, out]
+ *   Pointer to error structure.
+ *
+ * @return
+ *   On success the number of bytes consumed/necessary, if the returned value
+ *   is lesser or equal to @p flow_size, the @p action has fully been
+ *   converted, otherwise another call with this returned memory size should
+ *   be done.
+ *   On error, a negative errno value is returned and rte_errno is set.
+ */
+static int
+mlx5_flow_action_count(struct rte_eth_dev *dev,
+                      const struct rte_flow_action *action,
+                      struct rte_flow *flow,
+                      const size_t flow_size __rte_unused,
+                      struct rte_flow_error *error)
+{
+       const struct rte_flow_action_count *count = action->conf;
+#ifdef HAVE_IBV_DEVICE_COUNTERS_SET_SUPPORT
+       unsigned int size = sizeof(struct ibv_flow_spec_counter_action);
+       struct ibv_flow_spec_counter_action counter = {
+               .type = IBV_FLOW_SPEC_ACTION_COUNT,
+               .size = size,
+       };
+#endif
+
+       if (!flow->counter) {
+               flow->counter = mlx5_flow_counter_new(dev, count->shared,
+                                                     count->id);
+               if (!flow->counter)
+                       return rte_flow_error_set(error, ENOTSUP,
+                                                 RTE_FLOW_ERROR_TYPE_ACTION,
+                                                 action,
+                                                 "cannot get counter"
+                                                 " context.");
+       }
+       if (!((struct priv *)dev->data->dev_private)->config.flow_counter_en)
+               return rte_flow_error_set(error, ENOTSUP,
+                                         RTE_FLOW_ERROR_TYPE_ACTION,
+                                         action,
+                                         "flow counters are not supported.");
+       flow->modifier |= MLX5_FLOW_MOD_COUNT;
+#ifdef HAVE_IBV_DEVICE_COUNTERS_SET_SUPPORT
+       counter.counter_set_handle = flow->counter->cs->handle;
+       if (size <= flow_size)
+               mlx5_flow_spec_verbs_add(flow, &counter, size);
+       return size;
+#endif
+       return 0;
+}
+
 /**
  * Convert the @p action into @p flow after ensuring the NIC will understand
  * and process it correctly.
@@ -2187,6 +2339,10 @@ mlx5_flow_actions(struct rte_eth_dev *dev,
                case RTE_FLOW_ACTION_TYPE_RSS:
                        ret = mlx5_flow_action_rss(dev, actions, flow, error);
                        break;
+               case RTE_FLOW_ACTION_TYPE_COUNT:
+                       ret = mlx5_flow_action_count(dev, actions, flow, remain,
+                                                    error);
+                       break;
                default:
                        return rte_flow_error_set(error, ENOTSUP,
                                                  RTE_FLOW_ERROR_TYPE_ACTION,
@@ -2568,6 +2724,10 @@ mlx5_flow_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
                        verbs->hrxq = NULL;
                }
        }
+       if (flow->counter) {
+               mlx5_flow_counter_release(flow->counter);
+               flow->counter = NULL;
+       }
 }
 
 /**
@@ -3017,6 +3177,88 @@ mlx5_flow_isolate(struct rte_eth_dev *dev,
        return 0;
 }
 
+/**
+ * Query flow counter.
+ *
+ * @param flow
+ *   Pointer to the flow.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+mlx5_flow_query_count(struct rte_flow *flow __rte_unused,
+                     void *data __rte_unused,
+                     struct rte_flow_error *error)
+{
+#ifdef HAVE_IBV_DEVICE_COUNTERS_SET_SUPPORT
+       struct rte_flow_query_count *qc = data;
+       uint64_t counters[2] = {0, 0};
+       struct ibv_query_counter_set_attr query_cs_attr = {
+               .cs = flow->counter->cs,
+               .query_flags = IBV_COUNTER_SET_FORCE_UPDATE,
+       };
+       struct ibv_counter_set_data query_out = {
+               .out = counters,
+               .outlen = 2 * sizeof(uint64_t),
+       };
+       int err = mlx5_glue->query_counter_set(&query_cs_attr, &query_out);
+
+       if (err)
+               return rte_flow_error_set(error, err,
+                                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                         NULL,
+                                         "cannot read counter");
+       qc->hits_set = 1;
+       qc->bytes_set = 1;
+       qc->hits = counters[0] - flow->counter->hits;
+       qc->bytes = counters[1] - flow->counter->bytes;
+       if (qc->reset) {
+               flow->counter->hits = counters[0];
+               flow->counter->bytes = counters[1];
+       }
+       return 0;
+#endif
+       return rte_flow_error_set(error, ENOTSUP,
+                                 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                                 NULL,
+                                 "counters are not available");
+}
+
+/**
+ * Query a flows.
+ *
+ * @see rte_flow_query()
+ * @see rte_flow_ops
+ */
+int
+mlx5_flow_query(struct rte_eth_dev *dev __rte_unused,
+               struct rte_flow *flow,
+               const struct rte_flow_action *actions,
+               void *data,
+               struct rte_flow_error *error)
+{
+       int ret = 0;
+
+       for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+               switch (actions->type) {
+               case RTE_FLOW_ACTION_TYPE_VOID:
+                       break;
+               case RTE_FLOW_ACTION_TYPE_COUNT:
+                       ret = mlx5_flow_query_count(flow, data, error);
+                       break;
+               default:
+                       return rte_flow_error_set(error, ENOTSUP,
+                                                 RTE_FLOW_ERROR_TYPE_ACTION,
+                                                 actions,
+                                                 "action not supported");
+               }
+               if (ret < 0)
+                       return ret;
+       }
+       return 0;
+}
+
 /**
  * Convert a flow director filter to a generic flow.
  *