]> git.baikalelectronics.ru Git - kernel.git/commitdiff
net/mlx5e: Slave representors sharing unique metadata for match
authorVu Pham <vuhuong@mellanox.com>
Mon, 2 Mar 2020 18:33:49 +0000 (10:33 -0800)
committerSaeed Mahameed <saeedm@mellanox.com>
Thu, 28 May 2020 01:13:49 +0000 (18:13 -0700)
Bonded slave representors' vports must share a unique metadata
for match.

On enslaving event of slave representor to lag device, allocate
new unique "bond_metadata" for match if this is the first slave.
The subsequent enslaved representors will share the same unique
"bond_metadata".

On unslaving event of slave representor, reset the slave
representor's vport to use its own default metadata.

Replace ingress acl and rx rules of the slave representors' vports
using new vport->bond_metadata.

Signed-off-by: Vu Pham <vuhuong@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/en/rep/bond.c
drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
drivers/net/ethernet/mellanox/mlx5/core/en_rep.h

index 932e94362ceb5f6188643aa4a2ec82a2ec55d712..13500f60bef612b420802e98599c8ea7e71ad2fa 100644 (file)
@@ -71,6 +71,7 @@ static void mlx5e_rep_bond_metadata_release(struct mlx5e_rep_bond_metadata *mdat
        netdev_dbg(mdata->lag_dev, "destroy rep_bond_metadata(%d)\n",
                   mdata->metadata_reg_c_0);
        list_del(&mdata->list);
+       mlx5_esw_match_metadata_free(mdata->esw, mdata->metadata_reg_c_0);
        WARN_ON(!list_empty(&mdata->slaves_list));
        kfree(mdata);
 }
@@ -82,6 +83,8 @@ int mlx5e_rep_bond_enslave(struct mlx5_eswitch *esw, struct net_device *netdev,
        struct mlx5e_rep_bond_slave_entry *s_entry;
        struct mlx5e_rep_bond_metadata *mdata;
        struct mlx5e_rep_priv *rpriv;
+       struct mlx5e_priv *priv;
+       int err;
 
        ASSERT_RTNL();
 
@@ -96,6 +99,11 @@ int mlx5e_rep_bond_enslave(struct mlx5_eswitch *esw, struct net_device *netdev,
                mdata->lag_dev = lag_dev;
                mdata->esw = esw;
                INIT_LIST_HEAD(&mdata->slaves_list);
+               mdata->metadata_reg_c_0 = mlx5_esw_match_metadata_alloc(esw);
+               if (!mdata->metadata_reg_c_0) {
+                       kfree(mdata);
+                       return -ENOSPC;
+               }
                list_add(&mdata->list, &rpriv->uplink_priv.bond->metadata_list);
 
                netdev_dbg(lag_dev, "create rep_bond_metadata(%d)\n",
@@ -103,14 +111,33 @@ int mlx5e_rep_bond_enslave(struct mlx5_eswitch *esw, struct net_device *netdev,
        }
 
        s_entry = kzalloc(sizeof(*s_entry), GFP_KERNEL);
-       if (!s_entry)
-               return -ENOMEM;
+       if (!s_entry) {
+               err = -ENOMEM;
+               goto entry_alloc_err;
+       }
 
        s_entry->netdev = netdev;
+       priv = netdev_priv(netdev);
+       rpriv = priv->ppriv;
+
+       err = mlx5_esw_acl_ingress_vport_bond_update(esw, rpriv->rep->vport,
+                                                    mdata->metadata_reg_c_0);
+       if (err)
+               goto ingress_err;
+
        mdata->slaves++;
        list_add_tail(&s_entry->list, &mdata->slaves_list);
+       netdev_dbg(netdev, "enslave rep vport(%d) lag_dev(%s) metadata(0x%x)\n",
+                  rpriv->rep->vport, lag_dev->name, mdata->metadata_reg_c_0);
 
        return 0;
+
+ingress_err:
+       kfree(s_entry);
+entry_alloc_err:
+       if (!mdata->slaves)
+               mlx5e_rep_bond_metadata_release(mdata);
+       return err;
 }
 
 /* This must be called under rtnl_lock */
@@ -121,6 +148,7 @@ void mlx5e_rep_bond_unslave(struct mlx5_eswitch *esw,
        struct mlx5e_rep_bond_slave_entry *s_entry;
        struct mlx5e_rep_bond_metadata *mdata;
        struct mlx5e_rep_priv *rpriv;
+       struct mlx5e_priv *priv;
 
        ASSERT_RTNL();
 
@@ -133,7 +161,16 @@ void mlx5e_rep_bond_unslave(struct mlx5_eswitch *esw,
        if (!s_entry)
                return;
 
+       priv = netdev_priv(netdev);
+       rpriv = priv->ppriv;
+
+       mlx5_esw_acl_ingress_vport_bond_update(esw, rpriv->rep->vport, 0);
+       mlx5e_rep_bond_update(priv, false);
        list_del(&s_entry->list);
+
+       netdev_dbg(netdev, "unslave rep vport(%d) lag_dev(%s) metadata(0x%x)\n",
+                  rpriv->rep->vport, lag_dev->name, mdata->metadata_reg_c_0);
+
        if (--mdata->slaves == 0)
                mlx5e_rep_bond_metadata_release(mdata);
        kfree(s_entry);
@@ -163,6 +200,7 @@ static void mlx5e_rep_changelowerstate_event(struct net_device *netdev, void *pt
        struct net_device *dev;
        u16 acl_vport_num;
        u16 fwd_vport_num;
+       int err;
 
        if (!mlx5e_rep_is_lag_netdev(netdev))
                return;
@@ -187,11 +225,28 @@ static void mlx5e_rep_changelowerstate_event(struct net_device *netdev, void *pt
                rpriv = priv->ppriv;
                acl_vport_num = rpriv->rep->vport;
                if (acl_vport_num != fwd_vport_num) {
-                       mlx5_esw_acl_egress_vport_bond(priv->mdev->priv.eswitch,
-                                                      fwd_vport_num,
-                                                      acl_vport_num);
+                       /* Only single rx_rule for unique bond_metadata should be
+                        * present, delete it if it's saved as passive vport's
+                        * rx_rule with destination as passive vport's root_ft
+                        */
+                       mlx5e_rep_bond_update(priv, true);
+                       err = mlx5_esw_acl_egress_vport_bond(priv->mdev->priv.eswitch,
+                                                            fwd_vport_num,
+                                                            acl_vport_num);
+                       if (err)
+                               netdev_warn(dev,
+                                           "configure slave vport(%d) egress fwd, err(%d)",
+                                           acl_vport_num, err);
                }
        }
+
+       /* Insert new rx_rule for unique bond_metadata, save it as active vport's
+        * rx_rule with new destination as active vport's root_ft
+        */
+       err = mlx5e_rep_bond_update(netdev_priv(netdev), false);
+       if (err)
+               netdev_warn(netdev, "configure active slave vport(%d) rx_rule, err(%d)",
+                           fwd_vport_num, err);
 }
 
 static void mlx5e_rep_changeupper_event(struct net_device *netdev, void *ptr)
index 12593d75e8850df50b8d8142a2d43b9d06c22f9b..af89a4803c7d251d58f033e45b469383c474a7d7 100644 (file)
@@ -854,6 +854,24 @@ static int mlx5e_create_rep_vport_rx_rule(struct mlx5e_priv *priv)
        return 0;
 }
 
+static void rep_vport_rx_rule_destroy(struct mlx5e_priv *priv)
+{
+       struct mlx5e_rep_priv *rpriv = priv->ppriv;
+
+       if (!rpriv->vport_rx_rule)
+               return;
+
+       mlx5_del_flow_rules(rpriv->vport_rx_rule);
+       rpriv->vport_rx_rule = NULL;
+}
+
+int mlx5e_rep_bond_update(struct mlx5e_priv *priv, bool cleanup)
+{
+       rep_vport_rx_rule_destroy(priv);
+
+       return cleanup ? 0 : mlx5e_create_rep_vport_rx_rule(priv);
+}
+
 static int mlx5e_init_rep_rx(struct mlx5e_priv *priv)
 {
        struct mlx5_core_dev *mdev = priv->mdev;
@@ -918,9 +936,7 @@ err_close_drop_rq:
 
 static void mlx5e_cleanup_rep_rx(struct mlx5e_priv *priv)
 {
-       struct mlx5e_rep_priv *rpriv = priv->ppriv;
-
-       mlx5_del_flow_rules(rpriv->vport_rx_rule);
+       rep_vport_rx_rule_destroy(priv);
        mlx5e_destroy_rep_root_ft(priv);
        mlx5e_destroy_ttc_table(priv, &priv->fs.ttc);
        mlx5e_destroy_direct_tirs(priv, priv->direct_tir);
index ed741b6e6af2dd3dfe8a3737ed9b844ed761f91b..da9f1686d5254f6ad3d9741ad08bfc8038a10726 100644 (file)
@@ -222,6 +222,7 @@ int mlx5e_rep_bond_enslave(struct mlx5_eswitch *esw, struct net_device *netdev,
 void mlx5e_rep_bond_unslave(struct mlx5_eswitch *esw,
                            const struct net_device *netdev,
                            const struct net_device *lag_dev);
+int mlx5e_rep_bond_update(struct mlx5e_priv *priv, bool cleanup);
 
 bool mlx5e_is_uplink_rep(struct mlx5e_priv *priv);
 int mlx5e_add_sqs_fwd_rules(struct mlx5e_priv *priv);