net/mlx5e: Refactor ct to use post action infrastructure
authorChris Mi <cmi@nvidia.com>
Tue, 17 Aug 2021 03:23:09 +0000 (11:23 +0800)
committerSaeed Mahameed <saeedm@nvidia.com>
Fri, 20 Aug 2021 04:50:37 +0000 (21:50 -0700)
Move post action table management to common library providing
add/del/get API. Refactor the ct action offload to use the common
API.

Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
drivers/net/ethernet/mellanox/mlx5/core/en/fs.h
drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_act.c
drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_act.h
drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c
drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.h
drivers/net/ethernet/mellanox/mlx5/core/en_rep.h
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c

index e348c27..41684a6 100644 (file)
@@ -7,6 +7,8 @@
 #include "mod_hdr.h"
 #include "lib/fs_ttc.h"
 
+struct mlx5e_post_act;
+
 enum {
        MLX5E_TC_FT_LEVEL = 0,
        MLX5E_TC_TTC_FT_LEVEL,
@@ -19,6 +21,7 @@ struct mlx5e_tc_table {
        struct mutex                    t_lock;
        struct mlx5_flow_table          *t;
        struct mlx5_fs_chains           *chains;
+       struct mlx5e_post_act           *post_act;
 
        struct rhashtable               ht;
 
index cd72955..a3e43e8 100644 (file)
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
 // Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 
+#include "en_tc.h"
 #include "post_act.h"
 #include "mlx5_core.h"
 
@@ -9,8 +10,20 @@ struct mlx5e_post_act {
        struct mlx5_fs_chains *chains;
        struct mlx5_flow_table *ft;
        struct mlx5e_priv *priv;
+       struct xarray ids;
 };
 
+struct mlx5e_post_act_handle {
+       enum mlx5_flow_namespace_type ns_type;
+       struct mlx5_flow_attr *attr;
+       struct mlx5_flow_handle *rule;
+       u32 id;
+};
+
+#define MLX5_POST_ACTION_BITS (mlx5e_tc_attr_to_reg_mappings[FTEID_TO_REG].mlen)
+#define MLX5_POST_ACTION_MAX GENMASK(MLX5_POST_ACTION_BITS - 1, 0)
+#define MLX5_POST_ACTION_MASK MLX5_POST_ACTION_MAX
+
 struct mlx5e_post_act *
 mlx5e_tc_post_act_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
                       enum mlx5_flow_namespace_type ns_type)
@@ -43,6 +56,7 @@ mlx5e_tc_post_act_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
        post_act->chains = chains;
        post_act->ns_type = ns_type;
        post_act->priv = priv;
+       xa_init_flags(&post_act->ids, XA_FLAGS_ALLOC1);
        return post_act;
 
 err_ft:
@@ -57,6 +71,94 @@ mlx5e_tc_post_act_destroy(struct mlx5e_post_act *post_act)
        if (IS_ERR_OR_NULL(post_act))
                return;
 
+       xa_destroy(&post_act->ids);
        mlx5_chains_destroy_global_table(post_act->chains, post_act->ft);
        kfree(post_act);
 }
+
+struct mlx5e_post_act_handle *
+mlx5e_tc_post_act_add(struct mlx5e_post_act *post_act, struct mlx5_flow_attr *attr)
+{
+       u32 attr_sz = ns_to_attr_sz(post_act->ns_type);
+       struct mlx5e_post_act_handle *handle = NULL;
+       struct mlx5_flow_attr *post_attr = NULL;
+       struct mlx5_flow_spec *spec = NULL;
+       int err;
+
+       handle = kzalloc(sizeof(*handle), GFP_KERNEL);
+       spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
+       post_attr = mlx5_alloc_flow_attr(post_act->ns_type);
+       if (!handle || !spec || !post_attr) {
+               kfree(post_attr);
+               kvfree(spec);
+               kfree(handle);
+               return ERR_PTR(-ENOMEM);
+       }
+
+       memcpy(post_attr, attr, attr_sz);
+       post_attr->chain = 0;
+       post_attr->prio = 0;
+       post_attr->ft = post_act->ft;
+       post_attr->inner_match_level = MLX5_MATCH_NONE;
+       post_attr->outer_match_level = MLX5_MATCH_NONE;
+       post_attr->action &= ~(MLX5_FLOW_CONTEXT_ACTION_DECAP);
+
+       handle->ns_type = post_act->ns_type;
+       /* Splits were handled before post action */
+       if (handle->ns_type == MLX5_FLOW_NAMESPACE_FDB)
+               post_attr->esw_attr->split_count = 0;
+
+       err = xa_alloc(&post_act->ids, &handle->id, post_attr,
+                      XA_LIMIT(1, MLX5_POST_ACTION_MAX), GFP_KERNEL);
+       if (err)
+               goto err_xarray;
+
+       /* Post action rule matches on fte_id and executes original rule's
+        * tc rule action
+        */
+       mlx5e_tc_match_to_reg_match(spec, FTEID_TO_REG,
+                                   handle->id, MLX5_POST_ACTION_MASK);
+
+       handle->rule = mlx5_tc_rule_insert(post_act->priv, spec, post_attr);
+       if (IS_ERR(handle->rule)) {
+               err = PTR_ERR(handle->rule);
+               netdev_warn(post_act->priv->netdev, "Failed to add post action rule");
+               goto err_rule;
+       }
+       handle->attr = post_attr;
+
+       kvfree(spec);
+       return handle;
+
+err_rule:
+       xa_erase(&post_act->ids, handle->id);
+err_xarray:
+       kfree(post_attr);
+       kvfree(spec);
+       kfree(handle);
+       return ERR_PTR(err);
+}
+
+void
+mlx5e_tc_post_act_del(struct mlx5e_post_act *post_act, struct mlx5e_post_act_handle *handle)
+{
+       mlx5_tc_rule_delete(post_act->priv, handle->rule, handle->attr);
+       xa_erase(&post_act->ids, handle->id);
+       kfree(handle->attr);
+       kfree(handle);
+}
+
+struct mlx5_flow_table *
+mlx5e_tc_post_act_get_ft(struct mlx5e_post_act *post_act)
+{
+       return post_act->ft;
+}
+
+/* Allocate a header modify action to write the post action handle fte id to a register. */
+int
+mlx5e_tc_post_act_set_handle(struct mlx5_core_dev *dev,
+                            struct mlx5e_post_act_handle *handle,
+                            struct mlx5e_tc_mod_hdr_acts *acts)
+{
+       return mlx5e_tc_match_to_reg_set(dev, acts, handle->ns_type, FTEID_TO_REG, handle->id);
+}
index a7ac69e..b530ec1 100644 (file)
@@ -7,6 +7,10 @@
 #include "en.h"
 #include "lib/fs_chains.h"
 
+struct mlx5_flow_attr;
+struct mlx5e_priv;
+struct mlx5e_tc_mod_hdr_acts;
+
 struct mlx5e_post_act *
 mlx5e_tc_post_act_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
                       enum mlx5_flow_namespace_type ns_type);
@@ -14,4 +18,18 @@ mlx5e_tc_post_act_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
 void
 mlx5e_tc_post_act_destroy(struct mlx5e_post_act *post_act);
 
+struct mlx5e_post_act_handle *
+mlx5e_tc_post_act_add(struct mlx5e_post_act *post_act, struct mlx5_flow_attr *attr);
+
+void
+mlx5e_tc_post_act_del(struct mlx5e_post_act *post_act, struct mlx5e_post_act_handle *handle);
+
+struct mlx5_flow_table *
+mlx5e_tc_post_act_get_ft(struct mlx5e_post_act *post_act);
+
+int
+mlx5e_tc_post_act_set_handle(struct mlx5_core_dev *dev,
+                            struct mlx5e_post_act_handle *handle,
+                            struct mlx5e_tc_mod_hdr_acts *acts);
+
 #endif /* __MLX5_POST_ACTION_H__ */
index 9609692..6c949ab 100644 (file)
@@ -19,6 +19,7 @@
 #include "en/tc_ct.h"
 #include "en/mod_hdr.h"
 #include "en/mapping.h"
+#include "en/tc/post_act.h"
 #include "en.h"
 #include "en_tc.h"
 #include "en_rep.h"
 #define MLX5_CT_STATE_RELATED_BIT BIT(5)
 #define MLX5_CT_STATE_INVALID_BIT BIT(6)
 
-#define MLX5_FTE_ID_BITS (mlx5e_tc_attr_to_reg_mappings[FTEID_TO_REG].mlen)
-#define MLX5_FTE_ID_MAX GENMASK(MLX5_FTE_ID_BITS - 1, 0)
-#define MLX5_FTE_ID_MASK MLX5_FTE_ID_MAX
-
 #define MLX5_CT_LABELS_BITS (mlx5e_tc_attr_to_reg_mappings[LABELS_TO_REG].mlen)
 #define MLX5_CT_LABELS_MASK GENMASK(MLX5_CT_LABELS_BITS - 1, 0)
 
@@ -46,14 +43,13 @@ struct mlx5_tc_ct_priv {
        struct mlx5_core_dev *dev;
        const struct net_device *netdev;
        struct mod_hdr_tbl *mod_hdr_tbl;
-       struct xarray fte_ids;
        struct xarray tuple_ids;
        struct rhashtable zone_ht;
        struct rhashtable ct_tuples_ht;
        struct rhashtable ct_tuples_nat_ht;
        struct mlx5_flow_table *ct;
        struct mlx5_flow_table *ct_nat;
-       struct mlx5_flow_table *post_ct;
+       struct mlx5e_post_act *post_act;
        struct mutex control_lock; /* guards parallel adds/dels */
        struct mapping_ctx *zone_mapping;
        struct mapping_ctx *labels_mapping;
@@ -64,11 +60,9 @@ struct mlx5_tc_ct_priv {
 
 struct mlx5_ct_flow {
        struct mlx5_flow_attr *pre_ct_attr;
-       struct mlx5_flow_attr *post_ct_attr;
        struct mlx5_flow_handle *pre_ct_rule;
-       struct mlx5_flow_handle *post_ct_rule;
+       struct mlx5e_post_act_handle *post_act_handle;
        struct mlx5_ct_ft *ft;
-       u32 fte_id;
        u32 chain_mapping;
 };
 
@@ -768,7 +762,7 @@ mlx5_tc_ct_entry_add_rule(struct mlx5_tc_ct_priv *ct_priv,
                       MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
                       MLX5_FLOW_CONTEXT_ACTION_COUNT;
        attr->dest_chain = 0;
-       attr->dest_ft = ct_priv->post_ct;
+       attr->dest_ft = mlx5e_tc_post_act_get_ft(ct_priv->post_act);
        attr->ft = nat ? ct_priv->ct_nat : ct_priv->ct;
        attr->outer_match_level = MLX5_MATCH_L4;
        attr->counter = entry->counter->counter;
@@ -1432,7 +1426,7 @@ static int tc_ct_pre_ct_add_rules(struct mlx5_ct_ft *ct_ft,
                ctstate |= MLX5_CT_STATE_NAT_BIT;
        mlx5e_tc_match_to_reg_match(spec, CTSTATE_TO_REG, ctstate, ctstate);
 
-       dest.ft = ct_priv->post_ct;
+       dest.ft = mlx5e_tc_post_act_get_ft(ct_priv->post_act);
        rule = mlx5_add_flow_rules(ft, spec, &flow_act, &dest, 1);
        if (IS_ERR(rule)) {
                err = PTR_ERR(rule);
@@ -1716,9 +1710,9 @@ mlx5_tc_ct_del_ft_cb(struct mlx5_tc_ct_priv *ct_priv, struct mlx5_ct_ft *ft)
  *      | do decap
  *      v
  * +---------------------+
- * + pre_ct/pre_ct_nat   +  if matches     +---------------------+
- * + zone+nat match      +---------------->+ post_ct (see below) +
- * +---------------------+  set zone       +---------------------+
+ * + pre_ct/pre_ct_nat   +  if matches     +-------------------------+
+ * + zone+nat match      +---------------->+ post_act (see below) +
+ * +---------------------+  set zone       +-------------------------+
  *      | set zone
  *      v
  * +--------------------+
@@ -1732,7 +1726,7 @@ mlx5_tc_ct_del_ft_cb(struct mlx5_tc_ct_priv *ct_priv, struct mlx5_ct_ft *ft)
  *      | do nat (if needed)
  *      v
  * +--------------+
- * + post_ct      + original filter actions
+ * + post_act  + original filter actions
  * + fte_id match +------------------------>
  * +--------------+
  */
@@ -1746,19 +1740,15 @@ __mlx5_tc_ct_flow_offload(struct mlx5_tc_ct_priv *ct_priv,
        struct mlx5e_priv *priv = netdev_priv(ct_priv->netdev);
        struct mlx5e_tc_mod_hdr_acts pre_mod_acts = {};
        u32 attr_sz = ns_to_attr_sz(ct_priv->ns_type);
-       struct mlx5_flow_spec *post_ct_spec = NULL;
+       struct mlx5e_post_act_handle *handle;
        struct mlx5_flow_attr *pre_ct_attr;
        struct mlx5_modify_hdr *mod_hdr;
-       struct mlx5_flow_handle *rule;
        struct mlx5_ct_flow *ct_flow;
        int chain_mapping = 0, err;
        struct mlx5_ct_ft *ft;
-       u32 fte_id = 1;
 
-       post_ct_spec = kvzalloc(sizeof(*post_ct_spec), GFP_KERNEL);
        ct_flow = kzalloc(sizeof(*ct_flow), GFP_KERNEL);
-       if (!post_ct_spec || !ct_flow) {
-               kvfree(post_ct_spec);
+       if (!ct_flow) {
                kfree(ct_flow);
                return ERR_PTR(-ENOMEM);
        }
@@ -1773,14 +1763,13 @@ __mlx5_tc_ct_flow_offload(struct mlx5_tc_ct_priv *ct_priv,
        }
        ct_flow->ft = ft;
 
-       err = xa_alloc(&ct_priv->fte_ids, &fte_id, ct_flow,
-                      XA_LIMIT(1, MLX5_FTE_ID_MAX), GFP_KERNEL);
-       if (err) {
-               netdev_warn(priv->netdev,
-                           "Failed to allocate fte id, err: %d\n", err);
-               goto err_xarray;
+       handle = mlx5e_tc_post_act_add(ct_priv->post_act, attr);
+       if (IS_ERR(handle)) {
+               err = PTR_ERR(handle);
+               ct_dbg("Failed to allocate post action handle");
+               goto err_post_act_handle;
        }
-       ct_flow->fte_id = fte_id;
+       ct_flow->post_act_handle = handle;
 
        /* Base flow attributes of both rules on original rule attribute */
        ct_flow->pre_ct_attr = mlx5_alloc_flow_attr(ct_priv->ns_type);
@@ -1789,15 +1778,8 @@ __mlx5_tc_ct_flow_offload(struct mlx5_tc_ct_priv *ct_priv,
                goto err_alloc_pre;
        }
 
-       ct_flow->post_ct_attr = mlx5_alloc_flow_attr(ct_priv->ns_type);
-       if (!ct_flow->post_ct_attr) {
-               err = -ENOMEM;
-               goto err_alloc_post;
-       }
-
        pre_ct_attr = ct_flow->pre_ct_attr;
        memcpy(pre_ct_attr, attr, attr_sz);
-       memcpy(ct_flow->post_ct_attr, attr, attr_sz);
 
        /* Modify the original rule's action to fwd and modify, leave decap */
        pre_ct_attr->action = attr->action & MLX5_FLOW_CONTEXT_ACTION_DECAP;
@@ -1823,10 +1805,9 @@ __mlx5_tc_ct_flow_offload(struct mlx5_tc_ct_priv *ct_priv,
                goto err_mapping;
        }
 
-       err = mlx5e_tc_match_to_reg_set(priv->mdev, &pre_mod_acts, ct_priv->ns_type,
-                                       FTEID_TO_REG, fte_id);
+       err = mlx5e_tc_post_act_set_handle(priv->mdev, handle, &pre_mod_acts);
        if (err) {
-               ct_dbg("Failed to set fte_id register mapping");
+               ct_dbg("Failed to set post action handle");
                goto err_mapping;
        }
 
@@ -1857,33 +1838,6 @@ __mlx5_tc_ct_flow_offload(struct mlx5_tc_ct_priv *ct_priv,
        }
        pre_ct_attr->modify_hdr = mod_hdr;
 
-       /* Post ct rule matches on fte_id and executes original rule's
-        * tc rule action
-        */
-       mlx5e_tc_match_to_reg_match(post_ct_spec, FTEID_TO_REG,
-                                   fte_id, MLX5_FTE_ID_MASK);
-
-       /* Put post_ct rule on post_ct flow table */
-       ct_flow->post_ct_attr->chain = 0;
-       ct_flow->post_ct_attr->prio = 0;
-       ct_flow->post_ct_attr->ft = ct_priv->post_ct;
-
-       /* Splits were handled before CT */
-       if (ct_priv->ns_type == MLX5_FLOW_NAMESPACE_FDB)
-               ct_flow->post_ct_attr->esw_attr->split_count = 0;
-
-       ct_flow->post_ct_attr->inner_match_level = MLX5_MATCH_NONE;
-       ct_flow->post_ct_attr->outer_match_level = MLX5_MATCH_NONE;
-       ct_flow->post_ct_attr->action &= ~(MLX5_FLOW_CONTEXT_ACTION_DECAP);
-       rule = mlx5_tc_rule_insert(priv, post_ct_spec,
-                                  ct_flow->post_ct_attr);
-       ct_flow->post_ct_rule = rule;
-       if (IS_ERR(ct_flow->post_ct_rule)) {
-               err = PTR_ERR(ct_flow->post_ct_rule);
-               ct_dbg("Failed to add post ct rule");
-               goto err_insert_post_ct;
-       }
-
        /* Change original rule point to ct table */
        pre_ct_attr->dest_chain = 0;
        pre_ct_attr->dest_ft = nat ? ft->pre_ct_nat.ft : ft->pre_ct.ft;
@@ -1897,28 +1851,21 @@ __mlx5_tc_ct_flow_offload(struct mlx5_tc_ct_priv *ct_priv,
 
        attr->ct_attr.ct_flow = ct_flow;
        dealloc_mod_hdr_actions(&pre_mod_acts);
-       kvfree(post_ct_spec);
 
-       return rule;
+       return ct_flow->pre_ct_rule;
 
 err_insert_orig:
-       mlx5_tc_rule_delete(priv, ct_flow->post_ct_rule,
-                           ct_flow->post_ct_attr);
-err_insert_post_ct:
        mlx5_modify_header_dealloc(priv->mdev, pre_ct_attr->modify_hdr);
 err_mapping:
        dealloc_mod_hdr_actions(&pre_mod_acts);
        mlx5_chains_put_chain_mapping(ct_priv->chains, ct_flow->chain_mapping);
 err_get_chain:
-       kfree(ct_flow->post_ct_attr);
-err_alloc_post:
        kfree(ct_flow->pre_ct_attr);
 err_alloc_pre:
-       xa_erase(&ct_priv->fte_ids, fte_id);
-err_xarray:
+       mlx5e_tc_post_act_del(ct_priv->post_act, handle);
+err_post_act_handle:
        mlx5_tc_ct_del_ft_cb(ct_priv, ft);
 err_ft:
-       kvfree(post_ct_spec);
        kfree(ct_flow);
        netdev_warn(priv->netdev, "Failed to offload ct flow, err %d\n", err);
        return ERR_PTR(err);
@@ -2029,16 +1976,13 @@ __mlx5_tc_ct_delete_flow(struct mlx5_tc_ct_priv *ct_priv,
                            pre_ct_attr);
        mlx5_modify_header_dealloc(priv->mdev, pre_ct_attr->modify_hdr);
 
-       if (ct_flow->post_ct_rule) {
-               mlx5_tc_rule_delete(priv, ct_flow->post_ct_rule,
-                                   ct_flow->post_ct_attr);
+       if (ct_flow->post_act_handle) {
                mlx5_chains_put_chain_mapping(ct_priv->chains, ct_flow->chain_mapping);
-               xa_erase(&ct_priv->fte_ids, ct_flow->fte_id);
+               mlx5e_tc_post_act_del(ct_priv->post_act, ct_flow->post_act_handle);
                mlx5_tc_ct_del_ft_cb(ct_priv, ct_flow->ft);
        }
 
        kfree(ct_flow->pre_ct_attr);
-       kfree(ct_flow->post_ct_attr);
        kfree(ct_flow);
 }
 
@@ -2064,11 +2008,6 @@ static int
 mlx5_tc_ct_init_check_esw_support(struct mlx5_eswitch *esw,
                                  const char **err_msg)
 {
-       if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ignore_flow_level)) {
-               *err_msg = "firmware level support is missing";
-               return -EOPNOTSUPP;
-       }
-
        if (!mlx5_eswitch_vlan_actions_supported(esw->dev, 1)) {
                /* vlan workaround should be avoided for multi chain rules.
                 * This is just a sanity check as pop vlan action should
@@ -2098,20 +2037,9 @@ mlx5_tc_ct_init_check_esw_support(struct mlx5_eswitch *esw,
 }
 
 static int
-mlx5_tc_ct_init_check_nic_support(struct mlx5e_priv *priv,
-                                 const char **err_msg)
-{
-       if (!MLX5_CAP_FLOWTABLE_NIC_RX(priv->mdev, ignore_flow_level)) {
-               *err_msg = "firmware level support is missing";
-               return -EOPNOTSUPP;
-       }
-
-       return 0;
-}
-
-static int
 mlx5_tc_ct_init_check_support(struct mlx5e_priv *priv,
                              enum mlx5_flow_namespace_type ns_type,
+                             struct mlx5e_post_act *post_act,
                              const char **err_msg)
 {
        struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
@@ -2122,10 +2050,14 @@ mlx5_tc_ct_init_check_support(struct mlx5e_priv *priv,
        *err_msg = "tc skb extension missing";
        return -EOPNOTSUPP;
 #endif
+       if (IS_ERR_OR_NULL(post_act)) {
+               *err_msg = "tc ct offload not supported, post action is missing";
+               return -EOPNOTSUPP;
+       }
+
        if (ns_type == MLX5_FLOW_NAMESPACE_FDB)
                return mlx5_tc_ct_init_check_esw_support(esw, err_msg);
-       else
-               return mlx5_tc_ct_init_check_nic_support(priv, err_msg);
+       return 0;
 }
 
 #define INIT_ERR_PREFIX "tc ct offload init failed"
@@ -2133,7 +2065,8 @@ mlx5_tc_ct_init_check_support(struct mlx5e_priv *priv,
 struct mlx5_tc_ct_priv *
 mlx5_tc_ct_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
                struct mod_hdr_tbl *mod_hdr,
-               enum mlx5_flow_namespace_type ns_type)
+               enum mlx5_flow_namespace_type ns_type,
+               struct mlx5e_post_act *post_act)
 {
        struct mlx5_tc_ct_priv *ct_priv;
        struct mlx5_core_dev *dev;
@@ -2142,11 +2075,9 @@ mlx5_tc_ct_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
        int err;
 
        dev = priv->mdev;
-       err = mlx5_tc_ct_init_check_support(priv, ns_type, &msg);
+       err = mlx5_tc_ct_init_check_support(priv, ns_type, post_act, &msg);
        if (err) {
-               mlx5_core_warn(dev,
-                              "tc ct offload not supported, %s\n",
-                              msg);
+               mlx5_core_warn(dev, "tc ct offload not supported, %s\n", msg);
                goto err_support;
        }
 
@@ -2194,16 +2125,7 @@ mlx5_tc_ct_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
                goto err_ct_nat_tbl;
        }
 
-       ct_priv->post_ct = mlx5_chains_create_global_table(chains);
-       if (IS_ERR(ct_priv->post_ct)) {
-               err = PTR_ERR(ct_priv->post_ct);
-               mlx5_core_warn(dev,
-                              "%s, failed to create post ct table err: %d\n",
-                              INIT_ERR_PREFIX, err);
-               goto err_post_ct_tbl;
-       }
-
-       xa_init_flags(&ct_priv->fte_ids, XA_FLAGS_ALLOC1);
+       ct_priv->post_act = post_act;
        mutex_init(&ct_priv->control_lock);
        rhashtable_init(&ct_priv->zone_ht, &zone_params);
        rhashtable_init(&ct_priv->ct_tuples_ht, &tuples_ht_params);
@@ -2211,8 +2133,6 @@ mlx5_tc_ct_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
 
        return ct_priv;
 
-err_post_ct_tbl:
-       mlx5_chains_destroy_global_table(chains, ct_priv->ct_nat);
 err_ct_nat_tbl:
        mlx5_chains_destroy_global_table(chains, ct_priv->ct);
 err_ct_tbl:
@@ -2237,7 +2157,6 @@ mlx5_tc_ct_clean(struct mlx5_tc_ct_priv *ct_priv)
 
        chains = ct_priv->chains;
 
-       mlx5_chains_destroy_global_table(chains, ct_priv->post_ct);
        mlx5_chains_destroy_global_table(chains, ct_priv->ct_nat);
        mlx5_chains_destroy_global_table(chains, ct_priv->ct);
        mapping_destroy(ct_priv->zone_mapping);
@@ -2247,7 +2166,6 @@ mlx5_tc_ct_clean(struct mlx5_tc_ct_priv *ct_priv)
        rhashtable_destroy(&ct_priv->ct_tuples_nat_ht);
        rhashtable_destroy(&ct_priv->zone_ht);
        mutex_destroy(&ct_priv->control_lock);
-       xa_destroy(&ct_priv->fte_ids);
        kfree(ct_priv);
 }
 
index 644cf16..363329f 100644 (file)
@@ -92,7 +92,8 @@ struct mlx5_ct_attr {
 struct mlx5_tc_ct_priv *
 mlx5_tc_ct_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
                struct mod_hdr_tbl *mod_hdr,
-               enum mlx5_flow_namespace_type ns_type);
+               enum mlx5_flow_namespace_type ns_type,
+               struct mlx5e_post_act *post_act);
 void
 mlx5_tc_ct_clean(struct mlx5_tc_ct_priv *ct_priv);
 
@@ -132,7 +133,8 @@ mlx5e_tc_ct_restore_flow(struct mlx5_tc_ct_priv *ct_priv,
 static inline struct mlx5_tc_ct_priv *
 mlx5_tc_ct_init(struct mlx5e_priv *priv, struct mlx5_fs_chains *chains,
                struct mod_hdr_tbl *mod_hdr,
-               enum mlx5_flow_namespace_type ns_type)
+               enum mlx5_flow_namespace_type ns_type,
+               struct mlx5e_post_act *post_act)
 {
        return NULL;
 }
index e46698b..48a203a 100644 (file)
@@ -60,6 +60,7 @@ struct mlx5e_neigh_update_table {
 struct mlx5_tc_ct_priv;
 struct mlx5e_rep_bond;
 struct mlx5e_tc_tun_encap;
+struct mlx5e_post_act;
 
 struct mlx5_rep_uplink_priv {
        /* Filters DB - instantiated by the uplink representor and shared by
@@ -88,6 +89,7 @@ struct mlx5_rep_uplink_priv {
        /* maps tun_enc_opts to a unique id*/
        struct mapping_ctx *tunnel_enc_opts_mapping;
 
+       struct mlx5e_post_act *post_act;
        struct mlx5_tc_ct_priv *ct_priv;
        struct mlx5e_tc_psample *tc_psample;
 
index 040acef..8049c4c 100644 (file)
@@ -47,6 +47,7 @@
 #include <net/bareudp.h>
 #include <net/bonding.h>
 #include "en.h"
+#include "en/tc/post_act.h"
 #include "en_rep.h"
 #include "en/rep/tc.h"
 #include "en/rep/neigh.h"
@@ -4895,8 +4896,9 @@ int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
                goto err_chains;
        }
 
+       tc->post_act = mlx5e_tc_post_act_init(priv, tc->chains, MLX5_FLOW_NAMESPACE_KERNEL);
        tc->ct = mlx5_tc_ct_init(priv, tc->chains, &priv->fs.tc.mod_hdr,
-                                MLX5_FLOW_NAMESPACE_KERNEL);
+                                MLX5_FLOW_NAMESPACE_KERNEL, tc->post_act);
 
        tc->netdevice_nb.notifier_call = mlx5e_tc_netdev_event;
        err = register_netdevice_notifier_dev_net(priv->netdev,
@@ -4912,6 +4914,7 @@ int mlx5e_tc_nic_init(struct mlx5e_priv *priv)
 
 err_reg:
        mlx5_tc_ct_clean(tc->ct);
+       mlx5e_tc_post_act_destroy(tc->post_act);
        mlx5_chains_destroy(tc->chains);
 err_chains:
        mapping_destroy(chains_mapping);
@@ -4950,6 +4953,7 @@ void mlx5e_tc_nic_cleanup(struct mlx5e_priv *priv)
        mutex_destroy(&tc->t_lock);
 
        mlx5_tc_ct_clean(tc->ct);
+       mlx5e_tc_post_act_destroy(tc->post_act);
        mapping_destroy(tc->mapping);
        mlx5_chains_destroy(tc->chains);
 }
@@ -4970,10 +4974,13 @@ int mlx5e_tc_esw_init(struct rhashtable *tc_ht)
        priv = netdev_priv(rpriv->netdev);
        esw = priv->mdev->priv.eswitch;
 
+       uplink_priv->post_act = mlx5e_tc_post_act_init(priv, esw_chains(esw),
+                                                      MLX5_FLOW_NAMESPACE_FDB);
        uplink_priv->ct_priv = mlx5_tc_ct_init(netdev_priv(priv->netdev),
                                               esw_chains(esw),
                                               &esw->offloads.mod_hdr,
-                                              MLX5_FLOW_NAMESPACE_FDB);
+                                              MLX5_FLOW_NAMESPACE_FDB,
+                                              uplink_priv->post_act);
 
 #if IS_ENABLED(CONFIG_MLX5_TC_SAMPLE)
        uplink_priv->tc_psample = mlx5e_tc_sample_init(esw);
@@ -5027,6 +5034,7 @@ err_tun_mapping:
        mlx5_tc_ct_clean(uplink_priv->ct_priv);
        netdev_warn(priv->netdev,
                    "Failed to initialize tc (eswitch), err: %d", err);
+       mlx5e_tc_post_act_destroy(uplink_priv->post_act);
        return err;
 }
 
@@ -5046,6 +5054,7 @@ void mlx5e_tc_esw_cleanup(struct rhashtable *tc_ht)
        mlx5e_tc_sample_cleanup(uplink_priv->tc_psample);
 #endif
        mlx5_tc_ct_clean(uplink_priv->ct_priv);
+       mlx5e_tc_post_act_destroy(uplink_priv->post_act);
 }
 
 int mlx5e_tc_num_filters(struct mlx5e_priv *priv, unsigned long flags)