net/i40e: move RSS to flow API
authorWei Zhao <wei.zhao1@intel.com>
Wed, 10 Jan 2018 02:10:26 +0000 (10:10 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 16 Jan 2018 17:47:49 +0000 (18:47 +0100)
Rte_flow was defined to include RSS, this patch moves i40e
existing RSS to rte_flow. The old RSS configuration is kept
as it was, and can be deprecated in the future.

Signed-off-by: Wei Zhao <wei.zhao1@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
doc/guides/rel_notes/release_18_02.rst
drivers/net/i40e/i40e_ethdev.c
drivers/net/i40e/i40e_ethdev.h
drivers/net/i40e/i40e_flow.c

index 6c755bf..24a5080 100644 (file)
@@ -61,6 +61,13 @@ New Features
   1, 2, 4, 8 or 16. If no such parameter is configured, the number of queues
   per VF is 4 by default.
 
+* **Added the i40e ethernet driver to support RSS with flow API.**
+
+  Rte_flow actually defined to include RSS, but till now, RSS is out of
+  rte_flow. This patch is to support i40e NIC with existing RSS
+  configuration using rte_flow API.It also enable queue region configuration
+  using flow API for i40e.
+
 * **Added NVGRE and UDP tunnels support in Solarflare network PMD.**
 
   Added support for NVGRE, VXLAN and GENEVE tunnels.
index 83bcec6..f8abd5b 100644 (file)
@@ -1313,6 +1313,10 @@ eth_i40e_dev_init(struct rte_eth_dev *dev)
        /* initialize queue region configuration */
        i40e_init_queue_region_conf(dev);
 
+       /* initialize rss configuration from rte_flow */
+       memset(&pf->rss_info, 0,
+               sizeof(struct i40e_rte_flow_rss_conf));
+
        return 0;
 
 err_init_fdir_filter_list:
@@ -11123,12 +11127,23 @@ i40e_tunnel_filter_restore(struct i40e_pf *pf)
        }
 }
 
+/* Restore rss filter */
+static inline void
+i40e_rss_filter_restore(struct i40e_pf *pf)
+{
+       struct i40e_rte_flow_rss_conf *conf =
+                                       &pf->rss_info;
+       if (conf->num)
+               i40e_config_rss_filter(pf, conf, TRUE);
+}
+
 static void
 i40e_filter_restore(struct i40e_pf *pf)
 {
        i40e_ethertype_filter_restore(pf);
        i40e_tunnel_filter_restore(pf);
        i40e_fdir_filter_restore(pf);
+       i40e_rss_filter_restore(pf);
 }
 
 static bool
@@ -11583,6 +11598,82 @@ i40e_cloud_filter_qinq_create(struct i40e_pf *pf)
        return ret;
 }
 
+int
+i40e_config_rss_filter(struct i40e_pf *pf,
+               struct i40e_rte_flow_rss_conf *conf, bool add)
+{
+       struct i40e_hw *hw = I40E_PF_TO_HW(pf);
+       uint32_t i, lut = 0;
+       uint16_t j, num;
+       struct rte_eth_rss_conf rss_conf = conf->rss_conf;
+       struct i40e_rte_flow_rss_conf *rss_info = &pf->rss_info;
+
+       if (!add) {
+               if (memcmp(conf, rss_info,
+                       sizeof(struct i40e_rte_flow_rss_conf)) == 0) {
+                       i40e_pf_disable_rss(pf);
+                       memset(rss_info, 0,
+                               sizeof(struct i40e_rte_flow_rss_conf));
+                       return 0;
+               }
+               return -EINVAL;
+       }
+
+       if (rss_info->num)
+               return -EINVAL;
+
+       /* If both VMDQ and RSS enabled, not all of PF queues are configured.
+        * It's necessary to calculate the actual PF queues that are configured.
+        */
+       if (pf->dev_data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_VMDQ_FLAG)
+               num = i40e_pf_calc_configured_queues_num(pf);
+       else
+               num = pf->dev_data->nb_rx_queues;
+
+       num = RTE_MIN(num, conf->num);
+       PMD_DRV_LOG(INFO, "Max of contiguous %u PF queues are configured",
+                       num);
+
+       if (num == 0) {
+               PMD_DRV_LOG(ERR, "No PF queues are configured to enable RSS");
+               return -ENOTSUP;
+       }
+
+       /* Fill in redirection table */
+       for (i = 0, j = 0; i < hw->func_caps.rss_table_size; i++, j++) {
+               if (j == num)
+                       j = 0;
+               lut = (lut << 8) | (conf->queue[j] & ((0x1 <<
+                       hw->func_caps.rss_table_entry_width) - 1));
+               if ((i & 3) == 3)
+                       I40E_WRITE_REG(hw, I40E_PFQF_HLUT(i >> 2), lut);
+       }
+
+       if ((rss_conf.rss_hf & pf->adapter->flow_types_mask) == 0) {
+               i40e_pf_disable_rss(pf);
+               return 0;
+       }
+       if (rss_conf.rss_key == NULL || rss_conf.rss_key_len <
+               (I40E_PFQF_HKEY_MAX_INDEX + 1) * sizeof(uint32_t)) {
+               /* Random default keys */
+               static uint32_t rss_key_default[] = {0x6b793944,
+                       0x23504cb5, 0x5bea75b6, 0x309f4f12, 0x3dc0a2b8,
+                       0x024ddcdf, 0x339b8ca0, 0x4c4af64a, 0x34fac605,
+                       0x55d85839, 0x3a58997d, 0x2ec938e1, 0x66031581};
+
+               rss_conf.rss_key = (uint8_t *)rss_key_default;
+               rss_conf.rss_key_len = (I40E_PFQF_HKEY_MAX_INDEX + 1) *
+                                                       sizeof(uint32_t);
+       }
+
+       return i40e_hw_rss_hash_set(pf, &rss_conf);
+
+       rte_memcpy(rss_info,
+               conf, sizeof(struct i40e_rte_flow_rss_conf));
+
+       return 0;
+}
+
 RTE_INIT(i40e_init_log);
 static void
 i40e_init_log(void)
index 84a5133..69ea6c1 100644 (file)
@@ -867,6 +867,13 @@ struct i40e_customized_pctype {
        bool valid;   /* Check if it's valid */
 };
 
+struct i40e_rte_flow_rss_conf {
+       struct rte_eth_rss_conf rss_conf; /**< RSS parameters. */
+       uint16_t queue_region_conf; /**< Queue region config flag */
+       uint16_t num; /**< Number of entries in queue[]. */
+       uint16_t queue[I40E_MAX_Q_PER_TC]; /**< Queues indices to use. */
+};
+
 /*
  * Structure to store private data specific for PF instance.
  */
@@ -921,6 +928,7 @@ struct i40e_pf {
        struct i40e_fdir_info fdir; /* flow director info */
        struct i40e_ethertype_rule ethertype; /* Ethertype filter rule */
        struct i40e_tunnel_rule tunnel; /* Tunnel filter rule */
+       struct i40e_rte_flow_rss_conf rss_info; /* rss info */
        struct i40e_queue_regions queue_region; /* queue region info */
        struct i40e_fc_conf fc_conf; /* Flow control conf */
        struct i40e_mirror_rule_list mirror_list;
@@ -1047,6 +1055,7 @@ union i40e_filter_t {
        struct i40e_fdir_filter_conf fdir_filter;
        struct rte_eth_tunnel_filter_conf tunnel_filter;
        struct i40e_tunnel_filter_conf consistent_tunnel_filter;
+       struct i40e_rte_flow_rss_conf rss_conf;
 };
 
 typedef int (*parse_filter_t)(struct rte_eth_dev *dev,
@@ -1177,6 +1186,8 @@ void i40e_init_queue_region_conf(struct rte_eth_dev *dev);
 void i40e_flex_payload_reg_set_default(struct i40e_hw *hw);
 int i40e_set_rss_key(struct i40e_vsi *vsi, uint8_t *key, uint8_t key_len);
 int i40e_set_rss_lut(struct i40e_vsi *vsi, uint8_t *lut, uint16_t lut_size);
+int i40e_config_rss_filter(struct i40e_pf *pf,
+               struct i40e_rte_flow_rss_conf *conf, bool add);
 
 #define I40E_DEV_TO_PCI(eth_dev) \
        RTE_DEV_TO_PCI((eth_dev)->device)
index 6feb7aa..cd9a9b6 100644 (file)
@@ -110,6 +110,8 @@ static int i40e_flow_flush_fdir_filter(struct i40e_pf *pf);
 static int i40e_flow_flush_ethertype_filter(struct i40e_pf *pf);
 static int i40e_flow_flush_tunnel_filter(struct i40e_pf *pf);
 static int
+i40e_flow_flush_rss_filter(struct rte_eth_dev *dev);
+static int
 i40e_flow_parse_qinq_filter(struct rte_eth_dev *dev,
                              const struct rte_flow_attr *attr,
                              const struct rte_flow_item pattern[],
@@ -4108,6 +4110,316 @@ i40e_flow_parse_qinq_filter(struct rte_eth_dev *dev,
        return ret;
 }
 
+/**
+ * This function is used to do configuration i40e existing RSS with rte_flow.
+ * It also enable queue region configuration using flow API for i40e.
+ * pattern can be used indicate what parameters will be include in flow,
+ * like user_priority or flowtype for queue region or HASH function for RSS.
+ * Action is used to transmit parameter like queue index and HASH
+ * function for RSS, or flowtype for queue region configuration.
+ * For example:
+ * pattern:
+ * Case 1: only ETH, indicate  flowtype for queue region will be parsed.
+ * Case 2: only VLAN, indicate user_priority for queue region will be parsed.
+ * Case 3: none, indicate RSS related will be parsed in action.
+ * Any pattern other the ETH or VLAN will be treated as invalid except END.
+ * So, pattern choice is depened on the purpose of configuration of
+ * that flow.
+ * action:
+ * action RSS will be uaed to transmit valid parameter with
+ * struct rte_flow_action_rss for all the 3 case.
+ */
+static int
+i40e_flow_parse_rss_pattern(__rte_unused struct rte_eth_dev *dev,
+                            const struct rte_flow_item *pattern,
+                            struct rte_flow_error *error,
+                            uint8_t *action_flag,
+                            struct i40e_queue_regions *info)
+{
+       const struct rte_flow_item_vlan *vlan_spec, *vlan_mask;
+       const struct rte_flow_item *item = pattern;
+       enum rte_flow_item_type item_type;
+
+       if (item->type == RTE_FLOW_ITEM_TYPE_END)
+               return 0;
+
+       for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+               if (item->last) {
+                       rte_flow_error_set(error, EINVAL,
+                                          RTE_FLOW_ERROR_TYPE_ITEM,
+                                          item,
+                                          "Not support range");
+                       return -rte_errno;
+               }
+               item_type = item->type;
+               switch (item_type) {
+               case RTE_FLOW_ITEM_TYPE_ETH:
+                       *action_flag = 1;
+                       break;
+               case RTE_FLOW_ITEM_TYPE_VLAN:
+                       vlan_spec =
+                               (const struct rte_flow_item_vlan *)item->spec;
+                       vlan_mask =
+                               (const struct rte_flow_item_vlan *)item->mask;
+                       if (vlan_spec && vlan_mask) {
+                               if (vlan_mask->tci ==
+                                       rte_cpu_to_be_16(I40E_TCI_MASK)) {
+                                       info->region[0].user_priority[0] =
+                                               (vlan_spec->tci >> 13) & 0x7;
+                                       info->region[0].user_priority_num = 1;
+                                       info->queue_region_number = 1;
+                                       *action_flag = 0;
+                               }
+                       }
+                       break;
+               default:
+                       rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ITEM,
+                                       item,
+                                       "Not support range");
+                       return -rte_errno;
+               }
+       }
+
+       return 0;
+}
+
+static int
+i40e_flow_parse_rss_action(struct rte_eth_dev *dev,
+                           const struct rte_flow_action *actions,
+                           struct rte_flow_error *error,
+                           uint8_t *action_flag,
+                           struct i40e_queue_regions *conf_info,
+                           union i40e_filter_t *filter)
+{
+       const struct rte_flow_action *act;
+       const struct rte_flow_action_rss *rss;
+       struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       struct i40e_queue_regions *info = &pf->queue_region;
+       struct i40e_rte_flow_rss_conf *rss_config =
+                       &filter->rss_conf;
+       struct i40e_rte_flow_rss_conf *rss_info = &pf->rss_info;
+       uint16_t i, j, n, tmp;
+       uint32_t index = 0;
+
+       NEXT_ITEM_OF_ACTION(act, actions, index);
+       rss = (const struct rte_flow_action_rss *)act->conf;
+
+       /**
+        * rss only supports forwarding,
+        * check if the first not void action is RSS.
+        */
+       if (act->type != RTE_FLOW_ACTION_TYPE_RSS) {
+               memset(rss_config, 0, sizeof(struct i40e_rte_flow_rss_conf));
+               rte_flow_error_set(error, EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ACTION,
+                       act, "Not supported action.");
+               return -rte_errno;
+       }
+
+       if (action_flag) {
+               for (n = 0; n < 64; n++) {
+                       if (rss->rss_conf->rss_hf & (1 << n)) {
+                               conf_info->region[0].hw_flowtype[0] = n;
+                               conf_info->region[0].flowtype_num = 1;
+                               conf_info->queue_region_number = 1;
+                               break;
+                       }
+               }
+       }
+
+       for (n = 0; n < conf_info->queue_region_number; n++) {
+               if (conf_info->region[n].user_priority_num ||
+                               conf_info->region[n].flowtype_num) {
+                       if (!((rte_is_power_of_2(rss->num)) &&
+                                       rss->num <= 64)) {
+                               PMD_DRV_LOG(ERR, "The region sizes should be any of the following values: 1, 2, 4, 8, 16, 32, 64 as long as the "
+                               "total number of queues do not exceed the VSI allocation");
+                               return -rte_errno;
+                       }
+
+                       if (conf_info->region[n].user_priority[n] >=
+                                       I40E_MAX_USER_PRIORITY) {
+                               PMD_DRV_LOG(ERR, "the user priority max index is 7");
+                               return -rte_errno;
+                       }
+
+                       if (conf_info->region[n].hw_flowtype[n] >=
+                                       I40E_FILTER_PCTYPE_MAX) {
+                               PMD_DRV_LOG(ERR, "the hw_flowtype or PCTYPE max index is 63");
+                               return -rte_errno;
+                       }
+
+                       if (rss_info->num < rss->num ||
+                               rss_info->queue[0] < rss->queue[0] ||
+                               (rss->queue[0] + rss->num >
+                                       rss_info->num + rss_info->queue[0])) {
+                               rte_flow_error_set(error, EINVAL,
+                                       RTE_FLOW_ERROR_TYPE_ACTION,
+                                       act,
+                                       "no valid queues");
+                               return -rte_errno;
+                       }
+
+                       for (i = 0; i < info->queue_region_number; i++) {
+                               if (info->region[i].queue_num == rss->num &&
+                                       info->region[i].queue_start_index ==
+                                               rss->queue[0])
+                                       break;
+                       }
+
+                       if (i == info->queue_region_number) {
+                               if (i > I40E_REGION_MAX_INDEX) {
+                                       PMD_DRV_LOG(ERR, "the queue region max index is 7");
+                                       return -rte_errno;
+                               }
+
+                               info->region[i].queue_num =
+                                       rss->num;
+                               info->region[i].queue_start_index =
+                                       rss->queue[0];
+                               info->region[i].region_id =
+                                       info->queue_region_number;
+
+                               j = info->region[i].user_priority_num;
+                               tmp = conf_info->region[n].user_priority[0];
+                               if (conf_info->region[n].user_priority_num) {
+                                       info->region[i].user_priority[j] = tmp;
+                                       info->region[i].user_priority_num++;
+                               }
+
+                               j = info->region[i].flowtype_num;
+                               tmp = conf_info->region[n].hw_flowtype[0];
+                               if (conf_info->region[n].flowtype_num) {
+                                       info->region[i].hw_flowtype[j] = tmp;
+                                       info->region[i].flowtype_num++;
+                               }
+                               info->queue_region_number++;
+                       } else {
+                               j = info->region[i].user_priority_num;
+                               tmp = conf_info->region[n].user_priority[0];
+                               if (conf_info->region[n].user_priority_num) {
+                                       info->region[i].user_priority[j] = tmp;
+                                       info->region[i].user_priority_num++;
+                               }
+
+                               j = info->region[i].flowtype_num;
+                               tmp = conf_info->region[n].hw_flowtype[0];
+                               if (conf_info->region[n].flowtype_num) {
+                                       info->region[i].hw_flowtype[j] = tmp;
+                                       info->region[i].flowtype_num++;
+                               }
+                       }
+               }
+
+               rss_config->queue_region_conf = TRUE;
+               return 0;
+       }
+
+       if (!rss || !rss->num) {
+               rte_flow_error_set(error, EINVAL,
+                               RTE_FLOW_ERROR_TYPE_ACTION,
+                               act,
+                               "no valid queues");
+               return -rte_errno;
+       }
+
+       for (n = 0; n < rss->num; n++) {
+               if (rss->queue[n] >= dev->data->nb_rx_queues) {
+                       rte_flow_error_set(error, EINVAL,
+                                  RTE_FLOW_ERROR_TYPE_ACTION,
+                                  act,
+                                  "queue id > max number of queues");
+                       return -rte_errno;
+               }
+       }
+       if (rss->rss_conf)
+               rss_config->rss_conf = *rss->rss_conf;
+       else
+               rss_config->rss_conf.rss_hf =
+                       pf->adapter->flow_types_mask;
+
+       for (n = 0; n < rss->num; ++n)
+               rss_config->queue[n] = rss->queue[n];
+       rss_config->num = rss->num;
+       index++;
+
+       /* check if the next not void action is END */
+       NEXT_ITEM_OF_ACTION(act, actions, index);
+       if (act->type != RTE_FLOW_ACTION_TYPE_END) {
+               memset(rss_config, 0, sizeof(struct i40e_rte_flow_rss_conf));
+               rte_flow_error_set(error, EINVAL,
+                       RTE_FLOW_ERROR_TYPE_ACTION,
+                       act, "Not supported action.");
+               return -rte_errno;
+       }
+       rss_config->queue_region_conf = FALSE;
+
+       return 0;
+}
+
+static int
+i40e_parse_rss_filter(struct rte_eth_dev *dev,
+                       const struct rte_flow_attr *attr,
+                       const struct rte_flow_item pattern[],
+                       const struct rte_flow_action actions[],
+                       union i40e_filter_t *filter,
+                       struct rte_flow_error *error)
+{
+       int ret;
+       struct i40e_queue_regions info;
+       uint8_t action_flag = 0;
+
+       memset(&info, 0, sizeof(struct i40e_queue_regions));
+
+       ret = i40e_flow_parse_rss_pattern(dev, pattern,
+                                       error, &action_flag, &info);
+       if (ret)
+               return ret;
+
+       ret = i40e_flow_parse_rss_action(dev, actions, error,
+                                       &action_flag, &info, filter);
+       if (ret)
+               return ret;
+
+       ret = i40e_flow_parse_attr(attr, error);
+       if (ret)
+               return ret;
+
+       cons_filter_type = RTE_ETH_FILTER_HASH;
+
+       return 0;
+}
+
+static int
+i40e_config_rss_filter_set(struct rte_eth_dev *dev,
+               struct i40e_rte_flow_rss_conf *conf)
+{
+       struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       if (conf->queue_region_conf) {
+               i40e_flush_queue_region_all_conf(dev, hw, pf, 1);
+               conf->queue_region_conf = 0;
+       } else {
+               i40e_config_rss_filter(pf, conf, 1);
+       }
+       return 0;
+}
+
+static int
+i40e_config_rss_filter_del(struct rte_eth_dev *dev,
+               struct i40e_rte_flow_rss_conf *conf)
+{
+       struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       i40e_flush_queue_region_all_conf(dev, hw, pf, 0);
+
+       i40e_config_rss_filter(pf, conf, 0);
+       return 0;
+}
+
 static int
 i40e_flow_validate(struct rte_eth_dev *dev,
                   const struct rte_flow_attr *attr,
@@ -4144,6 +4456,17 @@ i40e_flow_validate(struct rte_eth_dev *dev,
 
        memset(&cons_filter, 0, sizeof(cons_filter));
 
+       /* Get the non-void item of action */
+       while ((actions + i)->type == RTE_FLOW_ACTION_TYPE_VOID)
+               i++;
+
+       if ((actions + i)->type == RTE_FLOW_ACTION_TYPE_RSS) {
+               ret = i40e_parse_rss_filter(dev, attr, pattern,
+                                       actions, &cons_filter, error);
+               return ret;
+       }
+
+       i = 0;
        /* Get the non-void item number of pattern */
        while ((pattern + i)->type != RTE_FLOW_ITEM_TYPE_END) {
                if ((pattern + i)->type != RTE_FLOW_ITEM_TYPE_VOID)
@@ -4231,6 +4554,11 @@ i40e_flow_create(struct rte_eth_dev *dev,
                flow->rule = TAILQ_LAST(&pf->tunnel.tunnel_list,
                                        i40e_tunnel_filter_list);
                break;
+       case RTE_ETH_FILTER_HASH:
+               ret = i40e_config_rss_filter_set(dev,
+                           &cons_filter.rss_conf);
+               flow->rule = &pf->rss_info;
+               break;
        default:
                goto free_flow;
        }
@@ -4269,6 +4597,10 @@ i40e_flow_destroy(struct rte_eth_dev *dev,
                ret = i40e_flow_add_del_fdir_filter(dev,
                       &((struct i40e_fdir_filter *)flow->rule)->fdir, 0);
                break;
+       case RTE_ETH_FILTER_HASH:
+               ret = i40e_config_rss_filter_del(dev,
+                          (struct i40e_rte_flow_rss_conf *)flow->rule);
+               break;
        default:
                PMD_DRV_LOG(WARNING, "Filter type (%d) not supported",
                            filter_type);
@@ -4411,6 +4743,14 @@ i40e_flow_flush(struct rte_eth_dev *dev, struct rte_flow_error *error)
                return -rte_errno;
        }
 
+       ret = i40e_flow_flush_rss_filter(dev);
+       if (ret) {
+               rte_flow_error_set(error, -ret,
+                                  RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
+                                  "Failed to flush rss flows.");
+               return -rte_errno;
+       }
+
        return ret;
 }
 
@@ -4506,3 +4846,19 @@ i40e_flow_flush_tunnel_filter(struct i40e_pf *pf)
 
        return ret;
 }
+
+/* remove the rss filter */
+static int
+i40e_flow_flush_rss_filter(struct rte_eth_dev *dev)
+{
+       struct i40e_pf *pf = I40E_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       struct i40e_rte_flow_rss_conf *rss_info = &pf->rss_info;
+       struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       int32_t ret = -EINVAL;
+
+       ret = i40e_flush_queue_region_all_conf(dev, hw, pf, 0);
+
+       if (rss_info->num)
+               ret = i40e_config_rss_filter(pf, rss_info, FALSE);
+       return ret;
+}