net/mlx5e: Use termination table for VLAN push actions

HW does not support push VLAN action in the RX direction (packets
arriving from the wire). The FW works around this limitation by haripining
the packet. The hairpin workaround applies only when the push VLAN action
is specified in a termination table, assuring that there are no actions
following the haripin.

Instantiate termination table for push VLAN actions. Re-use identical
terminating tables for increased HW cache efficiency.
Signed-off-by: Oz Shlomo <ozsh@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Eli Britstein <elibr@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

drivers/net/ethernet/mellanox/mlx5/core/Makefile

@@ -38,7 +38,8 @@ mlx5_core-$(CONFIG_MLX5_ESWITCH)     += en_rep.o en_tc.o en/tc_tun.o lib/port_tu
 #
 # Core extra
 #
-mlx5_core-$(CONFIG_MLX5_ESWITCH)   += eswitch.o eswitch_offloads.o ecpf.o rdma.o
+mlx5_core-$(CONFIG_MLX5_ESWITCH)   += eswitch.o eswitch_offloads.o eswitch_offloads_termtbl.o \
+				      ecpf.o rdma.o
 mlx5_core-$(CONFIG_MLX5_MPFS)      += lib/mpfs.o
 mlx5_core-$(CONFIG_VXLAN)          += lib/vxlan.o
 mlx5_core-$(CONFIG_PTP_1588_CLOCK) += lib/clock.o

drivers/net/ethernet/mellanox/mlx5/core/eswitch.h

@@ -173,6 +173,8 @@ struct mlx5_esw_offload {
 	struct mutex peer_mutex;
 	DECLARE_HASHTABLE(encap_tbl, 8);
 	DECLARE_HASHTABLE(mod_hdr_tbl, 8);
+	DECLARE_HASHTABLE(termtbl_tbl, 8);
+	struct mutex termtbl_mutex; /* protects termtbl hash */
 	const struct mlx5_eswitch_rep_ops *rep_ops[NUM_REP_TYPES];
 	u8 inline_mode;
 	u64 num_flows;
@@ -269,6 +271,25 @@ void mlx5_eswitch_del_send_to_vport_rule(struct mlx5_flow_handle *rule);
 
 struct mlx5_flow_spec;
 struct mlx5_esw_flow_attr;
+struct mlx5_termtbl_handle;
+
+bool
+mlx5_eswitch_termtbl_required(struct mlx5_eswitch *esw,
+			      struct mlx5_flow_act *flow_act,
+			      struct mlx5_flow_spec *spec);
+
+struct mlx5_flow_handle *
+mlx5_eswitch_add_termtbl_rule(struct mlx5_eswitch *esw,
+			      struct mlx5_flow_table *ft,
+			      struct mlx5_flow_spec *spec,
+			      struct mlx5_esw_flow_attr *attr,
+			      struct mlx5_flow_act *flow_act,
+			      struct mlx5_flow_destination *dest,
+			      int num_dest);
+
+void
+mlx5_eswitch_termtbl_put(struct mlx5_eswitch *esw,
+			 struct mlx5_termtbl_handle *tt);
 
 struct mlx5_flow_handle *
 mlx5_eswitch_add_offloaded_rule(struct mlx5_eswitch *esw,
@@ -339,6 +360,7 @@ struct mlx5_esw_flow_attr {
 		struct mlx5_eswitch_rep *rep;
 		struct mlx5_core_dev *mdev;
 		u32 encap_id;
+		struct mlx5_termtbl_handle *termtbl;
 	} dests[MLX5_MAX_FLOW_FWD_VPORTS];
 	u32	mod_hdr_id;
 	u8	match_level;

drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c

@@ -192,6 +192,10 @@ mlx5_eswitch_add_offloaded_rule(struct mlx5_eswitch *esw,
 		goto err_esw_get;
 	}
 
+	if (mlx5_eswitch_termtbl_required(esw, &flow_act, spec))
+		rule = mlx5_eswitch_add_termtbl_rule(esw, fdb, spec, attr,
+						     &flow_act, dest, i);
+	else
 		rule = mlx5_add_flow_rules(fdb, spec, &flow_act, dest, i);
 	if (IS_ERR(rule))
 		goto err_add_rule;
@@ -294,8 +298,16 @@ __mlx5_eswitch_del_rule(struct mlx5_eswitch *esw,
 			bool fwd_rule)
 {
 	bool split = (attr->split_count > 0);
+	int i;
 
 	mlx5_del_flow_rules(rule);
+
+	/* unref the term table */
+	for (i = 0; i < MLX5_MAX_FLOW_FWD_VPORTS; i++) {
+		if (attr->dests[i].termtbl)
+			mlx5_eswitch_termtbl_put(esw, attr->dests[i].termtbl);
+	}
+
 	esw->offloads.num_flows--;
 
 	if (fwd_rule)  {
@@ -1870,6 +1882,7 @@ int esw_offloads_init(struct mlx5_eswitch *esw, int vf_nvports,
 		goto err_reps;
 
 	esw_offloads_devcom_init(esw);
+	mutex_init(&esw->offloads.termtbl_mutex);
 
 	esw_functions_changed_event_init(esw, vf_nvports);
 

drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads_termtbl.c

+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+// Copyright (c) 2019 Mellanox Technologies.
+
+#include <linux/mlx5/fs.h>
+#include "eswitch.h"
+
+struct mlx5_termtbl_handle {
+	struct hlist_node termtbl_hlist;
+
+	struct mlx5_flow_table *termtbl;
+	struct mlx5_flow_act flow_act;
+	struct mlx5_flow_destination dest;
+
+	struct mlx5_flow_handle *rule;
+	int ref_count;
+};
+
+static u32
+mlx5_eswitch_termtbl_hash(struct mlx5_flow_act *flow_act,
+			  struct mlx5_flow_destination *dest)
+{
+	u32 hash;
+
+	hash = jhash_1word(flow_act->action, 0);
+	hash = jhash((const void *)&flow_act->vlan,
+		     sizeof(flow_act->vlan), hash);
+	hash = jhash((const void *)&dest->vport.num,
+		     sizeof(dest->vport.num), hash);
+	hash = jhash((const void *)&dest->vport.vhca_id,
+		     sizeof(dest->vport.num), hash);
+	return hash;
+}
+
+static int
+mlx5_eswitch_termtbl_cmp(struct mlx5_flow_act *flow_act1,
+			 struct mlx5_flow_destination *dest1,
+			 struct mlx5_flow_act *flow_act2,
+			 struct mlx5_flow_destination *dest2)
+{
+	return flow_act1->action != flow_act2->action ||
+	       dest1->vport.num != dest2->vport.num ||
+	       dest1->vport.vhca_id != dest2->vport.vhca_id ||
+	       memcmp(&flow_act1->vlan, &flow_act2->vlan,
+		      sizeof(flow_act1->vlan));
+}
+
+static int
+mlx5_eswitch_termtbl_create(struct mlx5_core_dev *dev,
+			    struct mlx5_termtbl_handle *tt,
+			    struct mlx5_flow_act *flow_act)
+{
+	struct mlx5_flow_namespace *root_ns;
+	struct mlx5_flow_spec spec = {};
+	int prio, flags;
+	int err;
+
+	root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_FDB);
+	if (!root_ns) {
+		esw_warn(dev, "Failed to get FDB flow namespace\n");
+		return -EOPNOTSUPP;
+	}
+
+	/* As this is the terminating action then the termination table is the
+	 * same prio as the slow path
+	 */
+	prio = FDB_SLOW_PATH;
+	flags = MLX5_FLOW_TABLE_TERMINATION;
+	tt->termtbl = mlx5_create_auto_grouped_flow_table(root_ns, prio, 1, 1,
+							  0, flags);
+	if (IS_ERR(tt->termtbl)) {
+		esw_warn(dev, "Failed to create termination table\n");
+		return -EOPNOTSUPP;
+	}
+
+	tt->rule = mlx5_add_flow_rules(tt->termtbl, &spec, flow_act,
+				       &tt->dest, 1);
+
+	if (IS_ERR(tt->rule)) {
+		esw_warn(dev, "Failed to create termination table rule\n");
+		goto add_flow_err;
+	}
+	return 0;
+
+add_flow_err:
+	err = mlx5_destroy_flow_table(tt->termtbl);
+	if (err)
+		esw_warn(dev, "Failed to destroy termination table\n");
+
+	return -EOPNOTSUPP;
+}
+
+static struct mlx5_termtbl_handle *
+mlx5_eswitch_termtbl_get_create(struct mlx5_eswitch *esw,
+				struct mlx5_flow_act *flow_act,
+				struct mlx5_flow_destination *dest)
+{
+	struct mlx5_termtbl_handle *tt;
+	bool found = false;
+	u32 hash_key;
+	int err;
+
+	mutex_lock(&esw->offloads.termtbl_mutex);
+
+	hash_key = mlx5_eswitch_termtbl_hash(flow_act, dest);
+	hash_for_each_possible(esw->offloads.termtbl_tbl, tt,
+			       termtbl_hlist, hash_key) {
+		if (!mlx5_eswitch_termtbl_cmp(&tt->flow_act, &tt->dest,
+					      flow_act, dest)) {
+			found = true;
+			break;
+		}
+	}
+	if (found)
+		goto tt_add_ref;
+
+	tt = kzalloc(sizeof(*tt), GFP_KERNEL);
+	if (!tt) {
+		err = -ENOMEM;
+		goto tt_create_err;
+	}
+
+	tt->dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
+	tt->dest.vport.num = dest->vport.num;
+	tt->dest.vport.vhca_id = dest->vport.vhca_id;
+	memcpy(&tt->flow_act, flow_act, sizeof(*flow_act));
+
+	err = mlx5_eswitch_termtbl_create(esw->dev, tt, flow_act);
+	if (err) {
+		esw_warn(esw->dev, "Failed to create termination table\n");
+		goto tt_create_err;
+	}
+	hash_add(esw->offloads.termtbl_tbl, &tt->termtbl_hlist, hash_key);
+tt_add_ref:
+	tt->ref_count++;
+	mutex_unlock(&esw->offloads.termtbl_mutex);
+	return tt;
+tt_create_err:
+	kfree(tt);
+	mutex_unlock(&esw->offloads.termtbl_mutex);
+	return ERR_PTR(err);
+}
+
+void
+mlx5_eswitch_termtbl_put(struct mlx5_eswitch *esw,
+			 struct mlx5_termtbl_handle *tt)
+{
+	mutex_lock(&esw->offloads.termtbl_mutex);
+	if (--tt->ref_count == 0)
+		hash_del(&tt->termtbl_hlist);
+	mutex_unlock(&esw->offloads.termtbl_mutex);
+
+	if (!tt->ref_count) {
+		mlx5_del_flow_rules(tt->rule);
+		mlx5_destroy_flow_table(tt->termtbl);
+		kfree(tt);
+	}
+}
+
+static void
+mlx5_eswitch_termtbl_actions_move(struct mlx5_flow_act *src,
+				  struct mlx5_flow_act *dst)
+{
+	if (!(src->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH))
+		return;
+
+	src->action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
+	dst->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
+	memcpy(&dst->vlan[0], &src->vlan[0], sizeof(src->vlan[0]));
+	memset(&src->vlan[0], 0, sizeof(src->vlan[0]));
+
+	if (!(src->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
+		return;
+
+	src->action &= ~MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
+	dst->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2;
+	memcpy(&dst->vlan[1], &src->vlan[1], sizeof(src->vlan[1]));
+	memset(&src->vlan[1], 0, sizeof(src->vlan[1]));
+}
+
+bool
+mlx5_eswitch_termtbl_required(struct mlx5_eswitch *esw,
+			      struct mlx5_flow_act *flow_act,
+			      struct mlx5_flow_spec *spec)
+{
+	u32 port_mask = MLX5_GET(fte_match_param, spec->match_criteria,
+				 misc_parameters.source_port);
+	u32 port_value = MLX5_GET(fte_match_param, spec->match_value,
+				  misc_parameters.source_port);
+
+	if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, termination_table))
+		return false;
+
+	/* push vlan on RX */
+	return (flow_act->action & MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH) &&
+		((port_mask & port_value) == MLX5_VPORT_UPLINK);
+}
+
+struct mlx5_flow_handle *
+mlx5_eswitch_add_termtbl_rule(struct mlx5_eswitch *esw,
+			      struct mlx5_flow_table *fdb,
+			      struct mlx5_flow_spec *spec,
+			      struct mlx5_esw_flow_attr *attr,
+			      struct mlx5_flow_act *flow_act,
+			      struct mlx5_flow_destination *dest,
+			      int num_dest)
+{
+	struct mlx5_flow_act term_tbl_act = {};
+	struct mlx5_flow_handle *rule = NULL;
+	bool term_table_created = false;
+	int num_vport_dests = 0;
+	int i, curr_dest;
+
+	mlx5_eswitch_termtbl_actions_move(flow_act, &term_tbl_act);
+	term_tbl_act.action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
+
+	for (i = 0; i < num_dest; i++) {
+		struct mlx5_termtbl_handle *tt;
+
+		/* only vport destinations can be terminated */
+		if (dest[i].type != MLX5_FLOW_DESTINATION_TYPE_VPORT)
+			continue;
+
+		/* get the terminating table for the action list */
+		tt = mlx5_eswitch_termtbl_get_create(esw, &term_tbl_act,
+						     &dest[i]);
+		if (IS_ERR(tt)) {
+			esw_warn(esw->dev, "Failed to create termination table\n");
+			goto revert_changes;
+		}
+		attr->dests[num_vport_dests].termtbl = tt;
+		num_vport_dests++;
+
+		/* link the destination with the termination table */
+		dest[i].type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
+		dest[i].ft = tt->termtbl;
+		term_table_created = true;
+	}
+
+	/* at least one destination should reference a termination table */
+	if (!term_table_created)
+		goto revert_changes;
+
+	/* create the FTE */
+	rule = mlx5_add_flow_rules(fdb, spec, flow_act, dest, num_dest);
+	if (IS_ERR(rule))
+		goto revert_changes;
+
+	goto out;
+
+revert_changes:
+	/* revert the changes that were made to the original flow_act
+	 * and fall-back to the original rule actions
+	 */
+	mlx5_eswitch_termtbl_actions_move(&term_tbl_act, flow_act);
+
+	for (curr_dest = 0; curr_dest < num_vport_dests; curr_dest++) {
+		struct mlx5_termtbl_handle *tt = attr->dests[curr_dest].termtbl;
+
+		/* search for the destination associated with the
+		 * current term table
+		 */
+		for (i = 0; i < num_dest; i++) {
+			if (dest[i].ft != tt->termtbl)
+				continue;
+
+			memset(&dest[i], 0, sizeof(dest[i]));
+			dest[i].type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
+			dest[i].vport.num = tt->dest.vport.num;
+			dest[i].vport.vhca_id = tt->dest.vport.vhca_id;
+			mlx5_eswitch_termtbl_put(esw, tt);
+			break;
+		}
+	}
+	rule = mlx5_add_flow_rules(fdb, spec, flow_act, dest, num_dest);
+out:
+	return rule;
+}