mlx5: add VLAN filtering for broadcast and IPv6 multicast
authorAdrien Mazarguil <adrien.mazarguil@6wind.com>
Thu, 3 Mar 2016 14:27:38 +0000 (15:27 +0100)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Wed, 16 Mar 2016 17:52:18 +0000 (18:52 +0100)
Unlike promiscuous and allmulticast flows, those should remain
VLAN-specific.

Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
doc/guides/rel_notes/release_16_04.rst
drivers/net/mlx5/mlx5_rxmode.c
drivers/net/mlx5/mlx5_rxq.c
drivers/net/mlx5/mlx5_rxtx.h
drivers/net/mlx5/mlx5_vlan.c

index 0f6fd4c..5b386f1 100644 (file)
@@ -283,6 +283,10 @@ Drivers
   Done to prevent port initialization on non-Ethernet link layers and
   to report an error.
 
+* **mlx5: Applied VLAN filtering to broadcast and IPv6 multicast flows.**
+
+  Prevented reception of multicast frames outside of configured VLANs.
+
 * **aesni_mb: Fixed wrong return value when creating a device.**
 
   cryptodev_aesni_mb_init() was returning the device id of the device created,
index 730527e..2bc005e 100644 (file)
@@ -74,6 +74,7 @@ static const struct special_flow_init special_flow_init[] = {
 #endif /* HAVE_FLOW_SPEC_IPV6 */
                        1 << HASH_RXQ_ETH |
                        0,
+               .per_vlan = 0,
        },
        [HASH_RXQ_FLOW_TYPE_ALLMULTI] = {
                .dst_mac_val = "\x01\x00\x00\x00\x00\x00",
@@ -87,6 +88,7 @@ static const struct special_flow_init special_flow_init[] = {
 #endif /* HAVE_FLOW_SPEC_IPV6 */
                        1 << HASH_RXQ_ETH |
                        0,
+               .per_vlan = 0,
        },
        [HASH_RXQ_FLOW_TYPE_BROADCAST] = {
                .dst_mac_val = "\xff\xff\xff\xff\xff\xff",
@@ -100,6 +102,7 @@ static const struct special_flow_init special_flow_init[] = {
 #endif /* HAVE_FLOW_SPEC_IPV6 */
                        1 << HASH_RXQ_ETH |
                        0,
+               .per_vlan = 1,
        },
 #ifdef HAVE_FLOW_SPEC_IPV6
        [HASH_RXQ_FLOW_TYPE_IPV6MULTI] = {
@@ -110,24 +113,28 @@ static const struct special_flow_init special_flow_init[] = {
                        1 << HASH_RXQ_IPV6 |
                        1 << HASH_RXQ_ETH |
                        0,
+               .per_vlan = 1,
        },
 #endif /* HAVE_FLOW_SPEC_IPV6 */
 };
 
 /**
- * Enable a special flow in a hash RX queue.
+ * Enable a special flow in a hash RX queue for a given VLAN index.
  *
  * @param hash_rxq
  *   Pointer to hash RX queue structure.
  * @param flow_type
  *   Special flow type.
+ * @param vlan_index
+ *   VLAN index to use.
  *
  * @return
  *   0 on success, errno value on failure.
  */
 static int
-hash_rxq_special_flow_enable(struct hash_rxq *hash_rxq,
-                            enum hash_rxq_flow_type flow_type)
+hash_rxq_special_flow_enable_vlan(struct hash_rxq *hash_rxq,
+                                 enum hash_rxq_flow_type flow_type,
+                                 unsigned int vlan_index)
 {
        struct priv *priv = hash_rxq->priv;
        struct ibv_exp_flow *flow;
@@ -136,12 +143,15 @@ hash_rxq_special_flow_enable(struct hash_rxq *hash_rxq,
        struct ibv_exp_flow_spec_eth *spec = &data->spec;
        const uint8_t *mac;
        const uint8_t *mask;
+       unsigned int vlan_enabled = (priv->vlan_filter_n &&
+                                    special_flow_init[flow_type].per_vlan);
+       unsigned int vlan_id = priv->vlan_filter[vlan_index];
 
        /* Check if flow is relevant for this hash_rxq. */
        if (!(special_flow_init[flow_type].hash_types & (1 << hash_rxq->type)))
                return 0;
        /* Check if flow already exists. */
-       if (hash_rxq->special_flow[flow_type] != NULL)
+       if (hash_rxq->special_flow[flow_type][vlan_index] != NULL)
                return 0;
 
        /*
@@ -164,12 +174,14 @@ hash_rxq_special_flow_enable(struct hash_rxq *hash_rxq,
                                mac[0], mac[1], mac[2],
                                mac[3], mac[4], mac[5],
                        },
+                       .vlan_tag = (vlan_enabled ? htons(vlan_id) : 0),
                },
                .mask = {
                        .dst_mac = {
                                mask[0], mask[1], mask[2],
                                mask[3], mask[4], mask[5],
                        },
+                       .vlan_tag = (vlan_enabled ? htons(0xfff) : 0),
                },
        };
 
@@ -184,9 +196,77 @@ hash_rxq_special_flow_enable(struct hash_rxq *hash_rxq,
                        return errno;
                return EINVAL;
        }
-       hash_rxq->special_flow[flow_type] = flow;
-       DEBUG("%p: enabling special flow %s (%d)",
-             (void *)hash_rxq, hash_rxq_flow_type_str(flow_type), flow_type);
+       hash_rxq->special_flow[flow_type][vlan_index] = flow;
+       DEBUG("%p: special flow %s (index %d) VLAN %u (index %u) enabled",
+             (void *)hash_rxq, hash_rxq_flow_type_str(flow_type), flow_type,
+             vlan_id, vlan_index);
+       return 0;
+}
+
+/**
+ * Disable a special flow in a hash RX queue for a given VLAN index.
+ *
+ * @param hash_rxq
+ *   Pointer to hash RX queue structure.
+ * @param flow_type
+ *   Special flow type.
+ * @param vlan_index
+ *   VLAN index to use.
+ */
+static void
+hash_rxq_special_flow_disable_vlan(struct hash_rxq *hash_rxq,
+                                  enum hash_rxq_flow_type flow_type,
+                                  unsigned int vlan_index)
+{
+       struct ibv_exp_flow *flow =
+               hash_rxq->special_flow[flow_type][vlan_index];
+
+       if (flow == NULL)
+               return;
+       claim_zero(ibv_exp_destroy_flow(flow));
+       hash_rxq->special_flow[flow_type][vlan_index] = NULL;
+       DEBUG("%p: special flow %s (index %d) VLAN %u (index %u) disabled",
+             (void *)hash_rxq, hash_rxq_flow_type_str(flow_type), flow_type,
+             hash_rxq->priv->vlan_filter[vlan_index], vlan_index);
+}
+
+/**
+ * Enable a special flow in a hash RX queue.
+ *
+ * @param hash_rxq
+ *   Pointer to hash RX queue structure.
+ * @param flow_type
+ *   Special flow type.
+ * @param vlan_index
+ *   VLAN index to use.
+ *
+ * @return
+ *   0 on success, errno value on failure.
+ */
+static int
+hash_rxq_special_flow_enable(struct hash_rxq *hash_rxq,
+                            enum hash_rxq_flow_type flow_type)
+{
+       struct priv *priv = hash_rxq->priv;
+       unsigned int i = 0;
+       int ret;
+
+       assert((unsigned int)flow_type < RTE_DIM(hash_rxq->special_flow));
+       assert(RTE_DIM(hash_rxq->special_flow[flow_type]) ==
+              RTE_DIM(priv->vlan_filter));
+       /* Add a special flow for each VLAN filter when relevant. */
+       do {
+               ret = hash_rxq_special_flow_enable_vlan(hash_rxq, flow_type, i);
+               if (ret) {
+                       /* Failure, rollback. */
+                       while (i != 0)
+                               hash_rxq_special_flow_disable_vlan(hash_rxq,
+                                                                  flow_type,
+                                                                  --i);
+                       return ret;
+               }
+       } while (special_flow_init[flow_type].per_vlan &&
+                ++i < priv->vlan_filter_n);
        return 0;
 }
 
@@ -202,12 +282,11 @@ static void
 hash_rxq_special_flow_disable(struct hash_rxq *hash_rxq,
                              enum hash_rxq_flow_type flow_type)
 {
-       if (hash_rxq->special_flow[flow_type] == NULL)
-               return;
-       claim_zero(ibv_exp_destroy_flow(hash_rxq->special_flow[flow_type]));
-       hash_rxq->special_flow[flow_type] = NULL;
-       DEBUG("%p: special flow %s (%d) disabled",
-             (void *)hash_rxq, hash_rxq_flow_type_str(flow_type), flow_type);
+       unsigned int i;
+
+       assert((unsigned int)flow_type < RTE_DIM(hash_rxq->special_flow));
+       for (i = 0; (i != RTE_DIM(hash_rxq->special_flow[flow_type])); ++i)
+               hash_rxq_special_flow_disable_vlan(hash_rxq, flow_type, i);
 }
 
 /**
index 0f5ac65..df6fd92 100644 (file)
@@ -541,7 +541,10 @@ priv_destroy_hash_rxqs(struct priv *priv)
                assert(hash_rxq->qp != NULL);
                /* Also check that there are no remaining flows. */
                for (j = 0; (j != RTE_DIM(hash_rxq->special_flow)); ++j)
-                       assert(hash_rxq->special_flow[j] == NULL);
+                       for (k = 0;
+                            (k != RTE_DIM(hash_rxq->special_flow[j]));
+                            ++k)
+                               assert(hash_rxq->special_flow[j][k] == NULL);
                for (j = 0; (j != RTE_DIM(hash_rxq->mac_flow)); ++j)
                        for (k = 0; (k != RTE_DIM(hash_rxq->mac_flow[j])); ++k)
                                assert(hash_rxq->mac_flow[j][k] == NULL);
index e85cf93..6ac1a5a 100644 (file)
@@ -195,6 +195,7 @@ struct special_flow_init {
        uint8_t dst_mac_val[6];
        uint8_t dst_mac_mask[6];
        unsigned int hash_types;
+       unsigned int per_vlan:1;
 };
 
 enum hash_rxq_flow_type {
@@ -231,7 +232,8 @@ struct hash_rxq {
        enum hash_rxq_type type; /* Hash RX queue type. */
        /* MAC flow steering rules, one per VLAN ID. */
        struct ibv_exp_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
-       struct ibv_exp_flow *special_flow[MLX5_MAX_SPECIAL_FLOWS];
+       struct ibv_exp_flow *special_flow
+               [MLX5_MAX_SPECIAL_FLOWS][MLX5_MAX_VLAN_IDS];
 };
 
 /* TX element. */
index fa9e3b8..ea7af1e 100644 (file)
@@ -98,9 +98,10 @@ vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
                priv->vlan_filter[priv->vlan_filter_n] = vlan_id;
                ++priv->vlan_filter_n;
        }
-       /* Rehash MAC flows in all hash RX queues. */
+       /* Rehash flows in all hash RX queues. */
        priv_mac_addrs_disable(priv);
-       return priv_mac_addrs_enable(priv);
+       priv_special_flow_disable_all(priv);
+       return priv_rehash_flows(priv);
 }
 
 /**