net/mlx5: add more details to flow dump
authorHaifei Luo <haifeil@nvidia.com>
Mon, 31 May 2021 02:22:08 +0000 (05:22 +0300)
committerRaslan Darawsheh <rasland@nvidia.com>
Thu, 8 Jul 2021 20:09:27 +0000 (22:09 +0200)
Currently the flow dump provides few information about actions
- just the pointers. Add implementations to display details for
counter, modify_hdr and encap_decap actions.

For counter, the regular flow operation query is engaged and
the counter content information is provided, including hits
and bytes values.For modify_hdr, encap_and decap actions,
the information stored in the ipool objects is dumped.

There are the formats of information presented in the dump:
Counter: rec_type,id,hits,bytes
Modify_hdr: rec_type,id,actions_number,actions
Encap_decap: rec_type,id,buf

Signed-off-by: Haifei Luo <haifeil@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
drivers/net/mlx5/mlx5.h
drivers/net/mlx5/mlx5_flow.c

index 0f4b239..36d7c9c 100644 (file)
@@ -1398,6 +1398,13 @@ struct rte_hairpin_peer_info {
        uint16_t manual_bind;
 };
 
+#define BUF_SIZE 1024
+enum dr_dump_rec_type {
+       DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT = 4410,
+       DR_DUMP_REC_TYPE_PMD_MODIFY_HDR = 4420,
+       DR_DUMP_REC_TYPE_PMD_COUNTER = 4430,
+};
+
 /* mlx5.c */
 
 int mlx5_getenv_int(const char *);
@@ -1629,6 +1636,14 @@ int mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
                       bool clear, uint64_t *pkts, uint64_t *bytes);
 int mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow,
                        FILE *file, struct rte_flow_error *error);
+int save_dump_file(const unsigned char *data, uint32_t size,
+               uint32_t type, uint32_t id, void *arg, FILE *file);
+int mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
+       struct rte_flow_query_count *count, struct rte_flow_error *error);
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+int mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev, struct rte_flow *flow,
+               FILE *file, struct rte_flow_error *error);
+#endif
 void mlx5_flow_rxq_dynf_metadata_set(struct rte_eth_dev *dev);
 int mlx5_flow_get_aged_flows(struct rte_eth_dev *dev, void **contexts,
                        uint32_t nb_contexts, struct rte_flow_error *error);
index 7e15ca0..cac4e6a 100644 (file)
@@ -7898,6 +7898,157 @@ mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
        return 0;
 }
 
+int
+save_dump_file(const uint8_t *data, uint32_t size,
+       uint32_t type, uint32_t id, void *arg, FILE *file)
+{
+       char line[BUF_SIZE];
+       uint32_t out = 0;
+       uint32_t k;
+       uint32_t actions_num;
+       struct rte_flow_query_count *count;
+
+       memset(line, 0, BUF_SIZE);
+       switch (type) {
+       case DR_DUMP_REC_TYPE_PMD_MODIFY_HDR:
+               actions_num = *(uint32_t *)(arg);
+               out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,%d,",
+                               type, id, actions_num);
+               break;
+       case DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT:
+               out += snprintf(line + out, BUF_SIZE - out, "%d,0x%x,",
+                               type, id);
+               break;
+       case DR_DUMP_REC_TYPE_PMD_COUNTER:
+               count = (struct rte_flow_query_count *)arg;
+               fprintf(file, "%d,0x%x,%" PRIu64 ",%" PRIu64 "\n", type,
+                               id, count->hits, count->bytes);
+               return 0;
+       default:
+               return -1;
+       }
+
+       for (k = 0; k < size; k++) {
+               /* Make sure we do not overrun the line buffer length. */
+               if (out >= BUF_SIZE - 4) {
+                       line[out] = '\0';
+                       break;
+               }
+               out += snprintf(line + out, BUF_SIZE - out, "%02x",
+                               (data[k]) & 0xff);
+       }
+       fprintf(file, "%s\n", line);
+       return 0;
+}
+
+int
+mlx5_flow_query_counter(struct rte_eth_dev *dev, struct rte_flow *flow,
+       struct rte_flow_query_count *count, struct rte_flow_error *error)
+{
+       struct rte_flow_action action[2];
+       enum mlx5_flow_drv_type ftype;
+       const struct mlx5_flow_driver_ops *fops;
+
+       if (!flow) {
+               return rte_flow_error_set(error, ENOENT,
+                               RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                               NULL,
+                               "invalid flow handle");
+       }
+       action[0].type = RTE_FLOW_ACTION_TYPE_COUNT;
+       action[1].type = RTE_FLOW_ACTION_TYPE_END;
+       if (flow->counter) {
+               memset(count, 0, sizeof(struct rte_flow_query_count));
+               ftype = (enum mlx5_flow_drv_type)(flow->drv_type);
+               MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN &&
+                                               ftype < MLX5_FLOW_TYPE_MAX);
+               fops = flow_get_drv_ops(ftype);
+               return fops->query(dev, flow, action, count, error);
+       }
+       return -1;
+}
+
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+/**
+ * Dump flow ipool data to file
+ *
+ * @param[in] dev
+ *   The pointer to Ethernet device.
+ * @param[in] file
+ *   A pointer to a file for output.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL. PMDs initialize this
+ *   structure in case of error only.
+ * @return
+ *   0 on success, a negative value otherwise.
+ */
+int
+mlx5_flow_dev_dump_ipool(struct rte_eth_dev *dev,
+       struct rte_flow *flow, FILE *file,
+       struct rte_flow_error *error)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_flow_dv_modify_hdr_resource  *modify_hdr;
+       struct mlx5_flow_dv_encap_decap_resource *encap_decap;
+       uint32_t handle_idx;
+       struct mlx5_flow_handle *dh;
+       struct rte_flow_query_count count;
+       uint32_t actions_num;
+       const uint8_t *data;
+       size_t size;
+       uint32_t id;
+       uint32_t type;
+
+       if (!flow) {
+               return rte_flow_error_set(error, ENOENT,
+                       RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+                       NULL,
+                       "invalid flow handle");
+       }
+       handle_idx = flow->dev_handles;
+       while (handle_idx) {
+               dh = mlx5_ipool_get(priv->sh->ipool
+                       [MLX5_IPOOL_MLX5_FLOW], handle_idx);
+               if (!dh)
+                       continue;
+               handle_idx = dh->next.next;
+               id = (uint32_t)(uintptr_t)dh->drv_flow;
+
+               /* query counter */
+               type = DR_DUMP_REC_TYPE_PMD_COUNTER;
+               if (!mlx5_flow_query_counter(dev, flow, &count, error))
+                       save_dump_file(NULL, 0, type,
+                                       id, (void *)&count, file);
+
+               /* Get modify_hdr and encap_decap buf from ipools. */
+               encap_decap = NULL;
+               modify_hdr = dh->dvh.modify_hdr;
+
+               if (dh->dvh.rix_encap_decap) {
+                       encap_decap = mlx5_ipool_get(priv->sh->ipool
+                                               [MLX5_IPOOL_DECAP_ENCAP],
+                                               dh->dvh.rix_encap_decap);
+               }
+               if (modify_hdr) {
+                       data = (const uint8_t *)modify_hdr->actions;
+                       size = (size_t)(modify_hdr->actions_num) * 8;
+                       actions_num = modify_hdr->actions_num;
+                       type = DR_DUMP_REC_TYPE_PMD_MODIFY_HDR;
+                       save_dump_file(data, size, type, id,
+                                       (void *)(&actions_num), file);
+               }
+               if (encap_decap) {
+                       data = encap_decap->buf;
+                       size = encap_decap->size;
+                       type = DR_DUMP_REC_TYPE_PMD_PKT_REFORMAT;
+                       save_dump_file(data, size, type,
+                                               id, NULL, file);
+               }
+       }
+       return 0;
+}
+#endif
+
 /**
  * Dump flow raw hw data to file
  *
@@ -7922,6 +8073,9 @@ mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
        int ret;
        struct mlx5_flow_handle *dh;
        struct rte_flow *flow;
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       uint32_t idx;
+#endif
 
        if (!priv->config.dv_flow_en) {
                if (fputs("device dv flow disabled\n", file) <= 0)
@@ -7930,16 +8084,25 @@ mlx5_flow_dev_dump(struct rte_eth_dev *dev, struct rte_flow *flow_idx,
        }
 
        /* dump all */
-       if (!flow_idx)
+       if (!flow_idx) {
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+               ILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_RTE_FLOW],
+                                               priv->flows, idx, flow, next)
+                       mlx5_flow_dev_dump_ipool(dev, flow, file, error);
+#endif
                return mlx5_devx_cmd_flow_dump(sh->fdb_domain,
                                        sh->rx_domain,
                                        sh->tx_domain, file);
+       }
        /* dump one */
        flow = mlx5_ipool_get(priv->sh->ipool
                        [MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
        if (!flow)
                return -ENOENT;
 
+#ifdef HAVE_IBV_FLOW_DV_SUPPORT
+       mlx5_flow_dev_dump_ipool(dev, flow, file, error);
+#endif
        handle_idx = flow->dev_handles;
        while (handle_idx) {
                dh = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW],