net/mlx5: Introduce mlx5_flow_steering structure
authorMaor Gottlieb <maorg@mellanox.com>
Mon, 4 Jul 2016 14:23:06 +0000 (17:23 +0300)
committerDavid S. Miller <davem@davemloft.net>
Tue, 5 Jul 2016 07:06:02 +0000 (00:06 -0700)
Instead of having all steering private name spaces and
steering module fields flat in mlx5_core_priv, we wrap
them in mlx5_flow_steering for better modularity and
API exposure.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
drivers/net/ethernet/mellanox/mlx5/core/fs_core.h
include/linux/mlx5/driver.h

index 7fcdae1cf05391d2806d506a6d56e0409efd9e69..0f969cb2291ac4144ab7b8e9704f54f9462d1068 100644 (file)
@@ -1363,12 +1363,13 @@ void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
                                                    enum mlx5_flow_namespace_type type)
 {
-       struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
+       struct mlx5_flow_steering *steering = dev->priv.steering;
+       struct mlx5_flow_root_namespace *root_ns;
        int prio;
        struct fs_prio *fs_prio;
        struct mlx5_flow_namespace *ns;
 
-       if (!root_ns)
+       if (!steering)
                return NULL;
 
        switch (type) {
@@ -1380,24 +1381,28 @@ struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
                prio = type;
                break;
        case MLX5_FLOW_NAMESPACE_FDB:
-               if (dev->priv.fdb_root_ns)
-                       return &dev->priv.fdb_root_ns->ns;
+               if (steering->fdb_root_ns)
+                       return &steering->fdb_root_ns->ns;
                else
                        return NULL;
        case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
-               if (dev->priv.esw_egress_root_ns)
-                       return &dev->priv.esw_egress_root_ns->ns;
+               if (steering->esw_egress_root_ns)
+                       return &steering->esw_egress_root_ns->ns;
                else
                        return NULL;
        case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
-               if (dev->priv.esw_ingress_root_ns)
-                       return &dev->priv.esw_ingress_root_ns->ns;
+               if (steering->esw_ingress_root_ns)
+                       return &steering->esw_ingress_root_ns->ns;
                else
                        return NULL;
        default:
                return NULL;
        }
 
+       root_ns = steering->root_ns;
+       if (!root_ns)
+               return NULL;
+
        fs_prio = find_prio(&root_ns->ns, prio);
        if (!fs_prio)
                return NULL;
@@ -1483,13 +1488,13 @@ static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
        return true;
 }
 
-static int init_root_tree_recursive(struct mlx5_core_dev *dev,
+static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
                                    struct init_tree_node *init_node,
                                    struct fs_node *fs_parent_node,
                                    struct init_tree_node *init_parent_node,
                                    int prio)
 {
-       int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
+       int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
                                              flow_table_properties_nic_receive.
                                              max_ft_level);
        struct mlx5_flow_namespace *fs_ns;
@@ -1500,7 +1505,7 @@ static int init_root_tree_recursive(struct mlx5_core_dev *dev,
 
        if (init_node->type == FS_TYPE_PRIO) {
                if ((init_node->min_ft_level > max_ft_level) ||
-                   !has_required_caps(dev, &init_node->caps))
+                   !has_required_caps(steering->dev, &init_node->caps))
                        return 0;
 
                fs_get_obj(fs_ns, fs_parent_node);
@@ -1521,7 +1526,7 @@ static int init_root_tree_recursive(struct mlx5_core_dev *dev,
        }
        prio = 0;
        for (i = 0; i < init_node->ar_size; i++) {
-               err = init_root_tree_recursive(dev, &init_node->children[i],
+               err = init_root_tree_recursive(steering, &init_node->children[i],
                                               base, init_node, prio);
                if (err)
                        return err;
@@ -1534,7 +1539,7 @@ static int init_root_tree_recursive(struct mlx5_core_dev *dev,
        return 0;
 }
 
-static int init_root_tree(struct mlx5_core_dev *dev,
+static int init_root_tree(struct mlx5_flow_steering *steering,
                          struct init_tree_node *init_node,
                          struct fs_node *fs_parent_node)
 {
@@ -1544,7 +1549,7 @@ static int init_root_tree(struct mlx5_core_dev *dev,
 
        fs_get_obj(fs_ns, fs_parent_node);
        for (i = 0; i < init_node->ar_size; i++) {
-               err = init_root_tree_recursive(dev, &init_node->children[i],
+               err = init_root_tree_recursive(steering, &init_node->children[i],
                                               &fs_ns->node,
                                               init_node, i);
                if (err)
@@ -1553,7 +1558,7 @@ static int init_root_tree(struct mlx5_core_dev *dev,
        return 0;
 }
 
-static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
+static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
                                                       enum fs_flow_table_type
                                                       table_type)
 {
@@ -1565,7 +1570,7 @@ static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev
        if (!root_ns)
                return NULL;
 
-       root_ns->dev = dev;
+       root_ns->dev = steering->dev;
        root_ns->table_type = table_type;
 
        ns = &root_ns->ns;
@@ -1620,46 +1625,45 @@ static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
 #define ANCHOR_PRIO 0
 #define ANCHOR_SIZE 1
 #define ANCHOR_LEVEL 0
-static int create_anchor_flow_table(struct mlx5_core_dev
-                                                       *dev)
+static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
 {
        struct mlx5_flow_namespace *ns = NULL;
        struct mlx5_flow_table *ft;
 
-       ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ANCHOR);
+       ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
        if (!ns)
                return -EINVAL;
        ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
        if (IS_ERR(ft)) {
-               mlx5_core_err(dev, "Failed to create last anchor flow table");
+               mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
                return PTR_ERR(ft);
        }
        return 0;
 }
 
-static int init_root_ns(struct mlx5_core_dev *dev)
+static int init_root_ns(struct mlx5_flow_steering *steering)
 {
 
-       dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
-       if (IS_ERR_OR_NULL(dev->priv.root_ns))
+       steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
+       if (IS_ERR_OR_NULL(steering->root_ns))
                goto cleanup;
 
-       if (init_root_tree(dev, &root_fs, &dev->priv.root_ns->ns.node))
+       if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
                goto cleanup;
 
-       set_prio_attrs(dev->priv.root_ns);
+       set_prio_attrs(steering->root_ns);
 
-       if (create_anchor_flow_table(dev))
+       if (create_anchor_flow_table(steering))
                goto cleanup;
 
        return 0;
 
 cleanup:
-       mlx5_cleanup_fs(dev);
+       mlx5_cleanup_fs(steering->dev);
        return -ENOMEM;
 }
 
-static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
+static void cleanup_single_prio_root_ns(struct mlx5_flow_steering *steering,
                                        struct mlx5_flow_root_namespace *root_ns)
 {
        struct fs_node *prio;
@@ -1672,11 +1676,11 @@ static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
                                        struct fs_node,
                                 list);
                if (tree_remove_node(prio))
-                       mlx5_core_warn(dev,
+                       mlx5_core_warn(steering->dev,
                                       "Flow steering priority wasn't destroyed, refcount > 1\n");
        }
        if (tree_remove_node(&root_ns->ns.node))
-               mlx5_core_warn(dev,
+               mlx5_core_warn(steering->dev,
                               "Flow steering namespace wasn't destroyed, refcount > 1\n");
        root_ns = NULL;
 }
@@ -1690,12 +1694,12 @@ static void destroy_flow_tables(struct fs_prio *prio)
                mlx5_destroy_flow_table(iter);
 }
 
-static void cleanup_root_ns(struct mlx5_core_dev *dev)
+static void cleanup_root_ns(struct mlx5_flow_steering *steering)
 {
-       struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
+       struct mlx5_flow_root_namespace *root_ns = steering->root_ns;
        struct fs_prio *iter_prio;
 
-       if (!MLX5_CAP_GEN(dev, nic_flow_table))
+       if (!MLX5_CAP_GEN(steering->dev, nic_flow_table))
                return;
 
        if (!root_ns)
@@ -1720,7 +1724,7 @@ static void cleanup_root_ns(struct mlx5_core_dev *dev)
                                fs_get_obj(obj_iter_prio2, iter_prio2);
                                destroy_flow_tables(obj_iter_prio2);
                                if (tree_remove_node(iter_prio2)) {
-                                       mlx5_core_warn(dev,
+                                       mlx5_core_warn(steering->dev,
                                                       "Priority %d wasn't destroyed, refcount > 1\n",
                                                       obj_iter_prio2->prio);
                                        return;
@@ -1737,7 +1741,7 @@ static void cleanup_root_ns(struct mlx5_core_dev *dev)
                                                 struct fs_node,
                                                 list);
                        if (tree_remove_node(iter_ns)) {
-                               mlx5_core_warn(dev,
+                               mlx5_core_warn(steering->dev,
                                               "Namespace wasn't destroyed, refcount > 1\n");
                                return;
                        }
@@ -1754,7 +1758,7 @@ static void cleanup_root_ns(struct mlx5_core_dev *dev)
 
                fs_get_obj(obj_prio_node, prio_node);
                if (tree_remove_node(prio_node)) {
-                       mlx5_core_warn(dev,
+                       mlx5_core_warn(steering->dev,
                                       "Priority %d wasn't destroyed, refcount > 1\n",
                                       obj_prio_node->prio);
                        return;
@@ -1762,70 +1766,75 @@ static void cleanup_root_ns(struct mlx5_core_dev *dev)
        }
 
        if (tree_remove_node(&root_ns->ns.node)) {
-               mlx5_core_warn(dev,
+               mlx5_core_warn(steering->dev,
                               "root namespace wasn't destroyed, refcount > 1\n");
                return;
        }
 
-       dev->priv.root_ns = NULL;
+       steering->root_ns = NULL;
 }
 
 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
 {
+       struct mlx5_flow_steering *steering = dev->priv.steering;
+
        if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
                return;
 
-       cleanup_root_ns(dev);
-       cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
-       cleanup_single_prio_root_ns(dev, dev->priv.esw_egress_root_ns);
-       cleanup_single_prio_root_ns(dev, dev->priv.esw_ingress_root_ns);
+       cleanup_root_ns(steering);
+       cleanup_single_prio_root_ns(steering, steering->esw_egress_root_ns);
+       cleanup_single_prio_root_ns(steering, steering->esw_ingress_root_ns);
+       cleanup_single_prio_root_ns(steering, steering->fdb_root_ns);
        mlx5_cleanup_fc_stats(dev);
+       kfree(steering);
 }
 
-static int init_fdb_root_ns(struct mlx5_core_dev *dev)
+static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
 {
        struct fs_prio *prio;
 
-       dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
-       if (!dev->priv.fdb_root_ns)
+       steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
+       if (!steering->fdb_root_ns)
                return -ENOMEM;
 
        /* Create single prio */
-       prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1);
+       prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
        if (IS_ERR(prio)) {
-               cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
+               cleanup_single_prio_root_ns(steering, steering->fdb_root_ns);
                return PTR_ERR(prio);
        } else {
                return 0;
        }
 }
 
-static int init_egress_acl_root_ns(struct mlx5_core_dev *dev)
+static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
 {
        struct fs_prio *prio;
 
-       dev->priv.esw_egress_root_ns = create_root_ns(dev, FS_FT_ESW_EGRESS_ACL);
-       if (!dev->priv.esw_egress_root_ns)
+       steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
+       if (!steering->esw_egress_root_ns)
                return -ENOMEM;
 
        /* create 1 prio*/
-       prio = fs_create_prio(&dev->priv.esw_egress_root_ns->ns, 0, MLX5_TOTAL_VPORTS(dev));
+       prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
+                             MLX5_TOTAL_VPORTS(steering->dev));
        if (IS_ERR(prio))
                return PTR_ERR(prio);
        else
                return 0;
 }
 
-static int init_ingress_acl_root_ns(struct mlx5_core_dev *dev)
+static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
 {
        struct fs_prio *prio;
 
-       dev->priv.esw_ingress_root_ns = create_root_ns(dev, FS_FT_ESW_INGRESS_ACL);
-       if (!dev->priv.esw_ingress_root_ns)
+       steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
+       if (!steering->esw_ingress_root_ns)
                return -ENOMEM;
 
        /* create 1 prio*/
-       prio = fs_create_prio(&dev->priv.esw_ingress_root_ns->ns, 0, MLX5_TOTAL_VPORTS(dev));
+       prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
+                             MLX5_TOTAL_VPORTS(steering->dev));
        if (IS_ERR(prio))
                return PTR_ERR(prio);
        else
@@ -1834,6 +1843,7 @@ static int init_ingress_acl_root_ns(struct mlx5_core_dev *dev)
 
 int mlx5_init_fs(struct mlx5_core_dev *dev)
 {
+       struct mlx5_flow_steering *steering;
        int err = 0;
 
        if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
@@ -1843,26 +1853,32 @@ int mlx5_init_fs(struct mlx5_core_dev *dev)
        if (err)
                return err;
 
+       steering = kzalloc(sizeof(*steering), GFP_KERNEL);
+       if (!steering)
+               return -ENOMEM;
+       steering->dev = dev;
+       dev->priv.steering = steering;
+
        if (MLX5_CAP_GEN(dev, nic_flow_table) &&
            MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
-               err = init_root_ns(dev);
+               err = init_root_ns(steering);
                if (err)
                        goto err;
        }
 
        if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
                if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
-                       err = init_fdb_root_ns(dev);
+                       err = init_fdb_root_ns(steering);
                        if (err)
                                goto err;
                }
                if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
-                       err = init_egress_acl_root_ns(dev);
+                       err = init_egress_acl_root_ns(steering);
                        if (err)
                                goto err;
                }
                if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
-                       err = init_ingress_acl_root_ns(dev);
+                       err = init_ingress_acl_root_ns(steering);
                        if (err)
                                goto err;
                }
index aa41a7314691977e5457bdb2f159334d1dcf76aa..d7ba91a1eea2b8c5fd5c46b4401da8d6c6d6d15c 100644 (file)
@@ -55,6 +55,14 @@ enum fs_fte_status {
        FS_FTE_STATUS_EXISTING = 1UL << 0,
 };
 
+struct mlx5_flow_steering {
+       struct mlx5_core_dev *dev;
+       struct mlx5_flow_root_namespace *root_ns;
+       struct mlx5_flow_root_namespace *fdb_root_ns;
+       struct mlx5_flow_root_namespace *esw_egress_root_ns;
+       struct mlx5_flow_root_namespace *esw_ingress_root_ns;
+};
+
 struct fs_node {
        struct list_head        list;
        struct list_head        children;
index e22b3456b2eef2fd1924a7f1d1bf6ed544f2c35a..f21c459418872405651b82138f20536cd8808f79 100644 (file)
@@ -550,14 +550,10 @@ struct mlx5_priv {
        struct list_head        ctx_list;
        spinlock_t              ctx_lock;
 
+       struct mlx5_flow_steering *steering;
        struct mlx5_eswitch     *eswitch;
        struct mlx5_core_sriov  sriov;
        unsigned long           pci_dev_data;
-       struct mlx5_flow_root_namespace *root_ns;
-       struct mlx5_flow_root_namespace *fdb_root_ns;
-       struct mlx5_flow_root_namespace *esw_egress_root_ns;
-       struct mlx5_flow_root_namespace *esw_ingress_root_ns;
-
        struct mlx5_fc_stats            fc_stats;
        struct mlx5_rl_table            rl_table;
 };