[pull request][net-next 00/13] Mellanox, mlx5 VF LAG 2018-12-14

@@ -15,7 +15,7 @@ mlx5_core-y :=	main.o cmd.o debugfs.o fw.o eq.o uar.o pagealloc.o \
 		health.o mcg.o cq.o alloc.o qp.o port.o mr.o pd.o \
 		mad.o transobj.o vport.o sriov.o fs_cmd.o fs_core.o \
 		fs_counters.o rl.o lag.o dev.o events.o wq.o lib/gid.o \
-		diag/fs_tracepoint.o diag/fw_tracer.o
+		lib/devcom.o diag/fs_tracepoint.o diag/fw_tracer.o
 
 #
 # Netdev basic

@@ -0,0 +1,255 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/* Copyright (c) 2018 Mellanox Technologies */
+
+#include <linux/mlx5/vport.h>
+#include "lib/devcom.h"
+
+static LIST_HEAD(devcom_list);
+
+#define devcom_for_each_component(priv, comp, iter) \
+	for (iter = 0; \
+	     comp = &(priv)->components[iter], iter < MLX5_DEVCOM_NUM_COMPONENTS; \
+	     iter++)
+
+struct mlx5_devcom_component {
+	struct {
+		void *data;
+	} device[MLX5_MAX_PORTS];
+
+	mlx5_devcom_event_handler_t handler;
+	struct rw_semaphore sem;
+	bool paired;
+};
+
+struct mlx5_devcom_list {
+	struct list_head list;
+
+	struct mlx5_devcom_component components[MLX5_DEVCOM_NUM_COMPONENTS];
+	struct mlx5_core_dev *devs[MLX5_MAX_PORTS];
+};
+
+struct mlx5_devcom {
+	struct mlx5_devcom_list *priv;
+	int idx;
+};
+
+static struct mlx5_devcom_list *mlx5_devcom_list_alloc(void)
+{
+	struct mlx5_devcom_component *comp;
+	struct mlx5_devcom_list *priv;
+	int i;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return NULL;
+
+	devcom_for_each_component(priv, comp, i)
+		init_rwsem(&comp->sem);
+
+	return priv;
+}
+
+static struct mlx5_devcom *mlx5_devcom_alloc(struct mlx5_devcom_list *priv,
+					     u8 idx)
+{
+	struct mlx5_devcom *devcom;
+
+	devcom = kzalloc(sizeof(*devcom), GFP_KERNEL);
+	if (!devcom)
+		return NULL;
+
+	devcom->priv = priv;
+	devcom->idx = idx;
+	return devcom;
+}
+
+/* Must be called with intf_mutex held */
+struct mlx5_devcom *mlx5_devcom_register_device(struct mlx5_core_dev *dev)
+{
+	struct mlx5_devcom_list *priv = NULL, *iter;
+	struct mlx5_devcom *devcom = NULL;
+	bool new_priv = false;
+	u64 sguid0, sguid1;
+	int idx, i;
+
+	if (!mlx5_core_is_pf(dev))
+		return NULL;
+
+	sguid0 = mlx5_query_nic_system_image_guid(dev);
+	list_for_each_entry(iter, &devcom_list, list) {
+		struct mlx5_core_dev *tmp_dev = NULL;
+
+		idx = -1;
+		for (i = 0; i < MLX5_MAX_PORTS; i++) {
+			if (iter->devs[i])
+				tmp_dev = iter->devs[i];
+			else
+				idx = i;
+		}
+
+		if (idx == -1)
+			continue;
+
+		sguid1 = mlx5_query_nic_system_image_guid(tmp_dev);
+		if (sguid0 != sguid1)
+			continue;
+
+		priv = iter;
+		break;
+	}
+
+	if (!priv) {
+		priv = mlx5_devcom_list_alloc();
+		if (!priv)
+			return ERR_PTR(-ENOMEM);
+
+		idx = 0;
+		new_priv = true;
+	}
+
+	priv->devs[idx] = dev;
+	devcom = mlx5_devcom_alloc(priv, idx);
+	if (!devcom) {
+		kfree(priv);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	if (new_priv)
+		list_add(&priv->list, &devcom_list);
+
+	return devcom;
+}
+
+/* Must be called with intf_mutex held */
+void mlx5_devcom_unregister_device(struct mlx5_devcom *devcom)
+{
+	struct mlx5_devcom_list *priv;
+	int i;
+
+	if (IS_ERR_OR_NULL(devcom))
+		return;
+
+	priv = devcom->priv;
+	priv->devs[devcom->idx] = NULL;
+
+	kfree(devcom);
+
+	for (i = 0; i < MLX5_MAX_PORTS; i++)
+		if (priv->devs[i])
+			break;
+
+	if (i != MLX5_MAX_PORTS)
+		return;
+
+	list_del(&priv->list);
+	kfree(priv);
+}
+
+void mlx5_devcom_register_component(struct mlx5_devcom *devcom,
+				    enum mlx5_devcom_components id,
+				    mlx5_devcom_event_handler_t handler,
+				    void *data)
+{
+	struct mlx5_devcom_component *comp;
+
+	if (IS_ERR_OR_NULL(devcom))
+		return;
+
+	WARN_ON(!data);
+
+	comp = &devcom->priv->components[id];
+	down_write(&comp->sem);
+	comp->handler = handler;
+	comp->device[devcom->idx].data = data;
+	up_write(&comp->sem);
+}
+
+void mlx5_devcom_unregister_component(struct mlx5_devcom *devcom,
+				      enum mlx5_devcom_components id)
+{
+	struct mlx5_devcom_component *comp;
+
+	if (IS_ERR_OR_NULL(devcom))
+		return;
+
+	comp = &devcom->priv->components[id];
+	down_write(&comp->sem);
+	comp->device[devcom->idx].data = NULL;
+	up_write(&comp->sem);
+}
+
+int mlx5_devcom_send_event(struct mlx5_devcom *devcom,
+			   enum mlx5_devcom_components id,
+			   int event,
+			   void *event_data)
+{
+	struct mlx5_devcom_component *comp;
+	int err = -ENODEV, i;
+
+	if (IS_ERR_OR_NULL(devcom))
+		return err;
+
+	comp = &devcom->priv->components[id];
+	down_write(&comp->sem);
+	for (i = 0; i < MLX5_MAX_PORTS; i++)
+		if (i != devcom->idx && comp->device[i].data) {
+			err = comp->handler(event, comp->device[i].data,
+					    event_data);
+			break;
+		}
+
+	up_write(&comp->sem);
+	return err;
+}
+
+void mlx5_devcom_set_paired(struct mlx5_devcom *devcom,
+			    enum mlx5_devcom_components id,
+			    bool paired)
+{
+	struct mlx5_devcom_component *comp;
+
+	comp = &devcom->priv->components[id];
+	WARN_ON(!rwsem_is_locked(&comp->sem));
+
+	comp->paired = paired;
+}
+
+bool mlx5_devcom_is_paired(struct mlx5_devcom *devcom,
+			   enum mlx5_devcom_components id)
+{
+	if (IS_ERR_OR_NULL(devcom))
+		return false;
+
+	return devcom->priv->components[id].paired;
+}
+
+void *mlx5_devcom_get_peer_data(struct mlx5_devcom *devcom,
+				enum mlx5_devcom_components id)
+{
+	struct mlx5_devcom_component *comp;
+	int i;
+
+	if (IS_ERR_OR_NULL(devcom))
+		return NULL;
+
+	comp = &devcom->priv->components[id];
+	down_read(&comp->sem);
+	if (!comp->paired) {
+		up_read(&comp->sem);
+		return NULL;
+	}
+
+	for (i = 0; i < MLX5_MAX_PORTS; i++)
+		if (i != devcom->idx)
+			break;
+
+	return comp->device[i].data;
+}
+
+void mlx5_devcom_release_peer_data(struct mlx5_devcom *devcom,
+				   enum mlx5_devcom_components id)
+{
+	struct mlx5_devcom_component *comp = &devcom->priv->components[id];
+
+	up_read(&comp->sem);
+}

@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/* Copyright (c) 2018 Mellanox Technologies */
+
+#ifndef __LIB_MLX5_DEVCOM_H__
+#define __LIB_MLX5_DEVCOM_H__
+
+#include <linux/mlx5/driver.h>
+
+enum mlx5_devcom_components {
+	MLX5_DEVCOM_NUM_COMPONENTS,
+};
+
+typedef int (*mlx5_devcom_event_handler_t)(int event,
+					   void *my_data,
+					   void *event_data);
+
+struct mlx5_devcom *mlx5_devcom_register_device(struct mlx5_core_dev *dev);
+void mlx5_devcom_unregister_device(struct mlx5_devcom *devcom);
+
+void mlx5_devcom_register_component(struct mlx5_devcom *devcom,
+				    enum mlx5_devcom_components id,
+				    mlx5_devcom_event_handler_t handler,
+				    void *data);
+void mlx5_devcom_unregister_component(struct mlx5_devcom *devcom,
+				      enum mlx5_devcom_components id);
+
+int mlx5_devcom_send_event(struct mlx5_devcom *devcom,
+			   enum mlx5_devcom_components id,
+			   int event,
+			   void *event_data);
+
+void mlx5_devcom_set_paired(struct mlx5_devcom *devcom,
+			    enum mlx5_devcom_components id,
+			    bool paired);
+bool mlx5_devcom_is_paired(struct mlx5_devcom *devcom,
+			   enum mlx5_devcom_components id);
+
+void *mlx5_devcom_get_peer_data(struct mlx5_devcom *devcom,
+				enum mlx5_devcom_components id);
+void mlx5_devcom_release_peer_data(struct mlx5_devcom *devcom,
+				   enum mlx5_devcom_components id);
+
+#endif
+

@@ -63,6 +63,7 @@
 #include "accel/tls.h"
 #include "lib/clock.h"
 #include "lib/vxlan.h"
+#include "lib/devcom.h"
 #include "diag/fw_tracer.h"
 
 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");

@@ -722,16 +723,21 @@ static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
 	struct pci_dev *pdev = dev->pdev;
 	int err;
 
+	priv->devcom = mlx5_devcom_register_device(dev);
+	if (IS_ERR(priv->devcom))
+		dev_err(&pdev->dev, "failed to register with devcom (0x%p)\n",
+			priv->devcom);
+
 	err = mlx5_query_board_id(dev);
 	if (err) {
 		dev_err(&pdev->dev, "query board id failed\n");
-		goto out;
+		goto err_devcom;
 	}
 
 	err = mlx5_eq_table_init(dev);
 	if (err) {
 		dev_err(&pdev->dev, "failed to initialize eq\n");
-		goto out;
+		goto err_devcom;
 	}
 
 	err = mlx5_events_init(dev);

@@ -807,8 +813,9 @@ static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
 	mlx5_events_cleanup(dev);
 err_eq_cleanup:
 	mlx5_eq_table_cleanup(dev);
+err_devcom:
+	mlx5_devcom_unregister_device(dev->priv.devcom);
 
-out:
 	return err;
 }
 

@@ -828,6 +835,7 @@ static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
 	mlx5_cq_debugfs_cleanup(dev);
 	mlx5_events_cleanup(dev);
 	mlx5_eq_table_cleanup(dev);
+	mlx5_devcom_unregister_device(dev->priv.devcom);
 }
 
 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,

@@ -486,6 +486,7 @@ struct mlx5_events;
 struct mlx5_mpfs;
 struct mlx5_eswitch;
 struct mlx5_lag;
+struct mlx5_devcom;
 struct mlx5_eq_table;
 
 struct mlx5_rate_limit {

@@ -560,6 +561,7 @@ struct mlx5_priv {
 	struct mlx5_eswitch     *eswitch;
 	struct mlx5_core_sriov	sriov;
 	struct mlx5_lag		*lag;
+	struct mlx5_devcom	*devcom;
 	unsigned long		pci_dev_data;
 	struct mlx5_fc_stats		fc_stats;
 	struct mlx5_rl_table            rl_table;

@@ -143,6 +143,8 @@ struct mlx5_eswitch_fdb {
 		struct offloads_fdb {
 			struct mlx5_flow_table *slow_fdb;
 			struct mlx5_flow_group *send_to_vport_grp;
+			struct mlx5_flow_group *peer_miss_grp;
+			struct mlx5_flow_handle **peer_miss_rules;
 			struct mlx5_flow_group *miss_grp;
 			struct mlx5_flow_handle *miss_rule_uni;
 			struct mlx5_flow_handle *miss_rule_multi;

@@ -39,6 +39,7 @@
 #include "eswitch.h"
 #include "en.h"
 #include "fs_core.h"
+#include "lib/devcom.h"
 
 enum {
 	FDB_FAST_PATH = 0,

@@ -541,6 +542,98 @@ void mlx5_eswitch_del_send_to_vport_rule(struct mlx5_flow_handle *rule)
 	mlx5_del_flow_rules(rule);
 }
 
+static void peer_miss_rules_setup(struct mlx5_core_dev *peer_dev,
+				  struct mlx5_flow_spec *spec,
+				  struct mlx5_flow_destination *dest)
+{
+	void *misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
+				  misc_parameters);
+
+	MLX5_SET(fte_match_set_misc, misc, source_eswitch_owner_vhca_id,
+		 MLX5_CAP_GEN(peer_dev, vhca_id));
+
+	spec->match_criteria_enable = MLX5_MATCH_MISC_PARAMETERS;
+
+	misc = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
+			    misc_parameters);
+	MLX5_SET_TO_ONES(fte_match_set_misc, misc, source_port);
+	MLX5_SET_TO_ONES(fte_match_set_misc, misc,
+			 source_eswitch_owner_vhca_id);
+
+	dest->type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
+	dest->vport.num = 0;
+	dest->vport.vhca_id = MLX5_CAP_GEN(peer_dev, vhca_id);
+	dest->vport.vhca_id_valid = 1;
+}
+
+static int esw_add_fdb_peer_miss_rules(struct mlx5_eswitch *esw,
+				       struct mlx5_core_dev *peer_dev)
+{
+	struct mlx5_flow_destination dest = {};
+	struct mlx5_flow_act flow_act = {0};
+	struct mlx5_flow_handle **flows;
+	struct mlx5_flow_handle *flow;
+	struct mlx5_flow_spec *spec;
+	/* total vports is the same for both e-switches */
+	int nvports = esw->total_vports;
+	void *misc;
+	int err, i;
+
+	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
+	if (!spec)
+		return -ENOMEM;
+
+	peer_miss_rules_setup(peer_dev, spec, &dest);
+
+	flows = kvzalloc(nvports * sizeof(*flows), GFP_KERNEL);
+	if (!flows) {
+		err = -ENOMEM;
+		goto alloc_flows_err;
+	}
+
+	flow_act.action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
+	misc = MLX5_ADDR_OF(fte_match_param, spec->match_value,
+			    misc_parameters);
+
+	for (i = 1; i < nvports; i++) {
+		MLX5_SET(fte_match_set_misc, misc, source_port, i);
+		flow = mlx5_add_flow_rules(esw->fdb_table.offloads.slow_fdb,
+					   spec, &flow_act, &dest, 1);
+		if (IS_ERR(flow)) {
+			err = PTR_ERR(flow);
+			esw_warn(esw->dev, "FDB: Failed to add peer miss flow rule err %d\n", err);
+			goto add_flow_err;
+		}
+		flows[i] = flow;
+	}
+
+	esw->fdb_table.offloads.peer_miss_rules = flows;
+
+	kvfree(spec);
+	return 0;
+
+add_flow_err:
+	for (i--; i > 0; i--)
+		mlx5_del_flow_rules(flows[i]);
+	kvfree(flows);
+alloc_flows_err:
+	kvfree(spec);
+	return err;
+}
+
+static void esw_del_fdb_peer_miss_rules(struct mlx5_eswitch *esw)
+{
+	struct mlx5_flow_handle **flows;
+	int i;
+
+	flows = esw->fdb_table.offloads.peer_miss_rules;
+
+	for (i = 1; i < esw->total_vports; i++)
+		mlx5_del_flow_rules(flows[i]);
+
+	kvfree(flows);
+}
+
 static int esw_add_fdb_miss_rule(struct mlx5_eswitch *esw)
 {
 	struct mlx5_flow_act flow_act = {0};

@@ -811,7 +904,8 @@ static int esw_create_offloads_fdb_tables(struct mlx5_eswitch *esw, int nvports)
 		esw->fdb_table.offloads.fdb_left[i] =
 			ESW_POOLS[i] <= fdb_max ? ESW_SIZE / ESW_POOLS[i] : 0;
 
-	table_size = nvports * MAX_SQ_NVPORTS + MAX_PF_SQ + 2;
+	table_size = nvports * MAX_SQ_NVPORTS + MAX_PF_SQ + 2 +
+		esw->total_vports;
 
 	/* create the slow path fdb with encap set, so further table instances
 	 * can be created at run time while VFs are probed if the FW allows that.

@@ -866,6 +960,34 @@ static int esw_create_offloads_fdb_tables(struct mlx5_eswitch *esw, int nvports)
 	}
 	esw->fdb_table.offloads.send_to_vport_grp = g;
 
+	/* create peer esw miss group */
+	memset(flow_group_in, 0, inlen);
+	MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,
+		 MLX5_MATCH_MISC_PARAMETERS);
+
+	match_criteria = MLX5_ADDR_OF(create_flow_group_in, flow_group_in,
+				      match_criteria);
+
+	MLX5_SET_TO_ONES(fte_match_param, match_criteria,
+			 misc_parameters.source_port);
+	MLX5_SET_TO_ONES(fte_match_param, match_criteria,
+			 misc_parameters.source_eswitch_owner_vhca_id);
+
+	MLX5_SET(create_flow_group_in, flow_group_in,
+		 source_eswitch_owner_vhca_id_valid, 1);
+	MLX5_SET(create_flow_group_in, flow_group_in, start_flow_index, ix);
+	MLX5_SET(create_flow_group_in, flow_group_in, end_flow_index,
+		 ix + esw->total_vports - 1);
+	ix += esw->total_vports;
+
+	g = mlx5_create_flow_group(fdb, flow_group_in);
+	if (IS_ERR(g)) {
+		err = PTR_ERR(g);
+		esw_warn(dev, "Failed to create peer miss flow group err(%d)\n", err);
+		goto peer_miss_err;
+	}
+	esw->fdb_table.offloads.peer_miss_grp = g;
+
 	/* create miss group */
 	memset(flow_group_in, 0, inlen);
 	MLX5_SET(create_flow_group_in, flow_group_in, match_criteria_enable,

@@ -898,6 +1020,8 @@ static int esw_create_offloads_fdb_tables(struct mlx5_eswitch *esw, int nvports)
 miss_rule_err:
 	mlx5_destroy_flow_group(esw->fdb_table.offloads.miss_grp);
 miss_err:
+	mlx5_destroy_flow_group(esw->fdb_table.offloads.peer_miss_grp);
+peer_miss_err:
 	mlx5_destroy_flow_group(esw->fdb_table.offloads.send_to_vport_grp);
 send_vport_err:
 	esw_destroy_offloads_fast_fdb_tables(esw);

@@ -917,6 +1041,7 @@ static void esw_destroy_offloads_fdb_tables(struct mlx5_eswitch *esw)
 	mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_multi);
 	mlx5_del_flow_rules(esw->fdb_table.offloads.miss_rule_uni);
 	mlx5_destroy_flow_group(esw->fdb_table.offloads.send_to_vport_grp);
+	mlx5_destroy_flow_group(esw->fdb_table.offloads.peer_miss_grp);
 	mlx5_destroy_flow_group(esw->fdb_table.offloads.miss_grp);
 
 	mlx5_destroy_flow_table(esw->fdb_table.offloads.slow_fdb);

@@ -1173,6 +1298,99 @@ static int esw_offloads_load_reps(struct mlx5_eswitch *esw, int nvports)
 	return err;
 }
 
+#define ESW_OFFLOADS_DEVCOM_PAIR	(0)
+#define ESW_OFFLOADS_DEVCOM_UNPAIR	(1)
+
+static int mlx5_esw_offloads_pair(struct mlx5_eswitch *esw,
+				  struct mlx5_eswitch *peer_esw)
+{
+	int err;
+
+	err = esw_add_fdb_peer_miss_rules(esw, peer_esw->dev);
+	if (err)
+		return err;
+
+	return 0;
+}
+
+static void mlx5_esw_offloads_unpair(struct mlx5_eswitch *esw)
+{
+	esw_del_fdb_peer_miss_rules(esw);
+}
+
+static int mlx5_esw_offloads_devcom_event(int event,
+					  void *my_data,
+					  void *event_data)
+{
+	struct mlx5_eswitch *esw = my_data;
+	struct mlx5_eswitch *peer_esw = event_data;
+	struct mlx5_devcom *devcom = esw->dev->priv.devcom;
+	int err;
+
+	switch (event) {
+	case ESW_OFFLOADS_DEVCOM_PAIR:
+		err = mlx5_esw_offloads_pair(esw, peer_esw);
+		if (err)
+			goto err_out;
+
+		err = mlx5_esw_offloads_pair(peer_esw, esw);
+		if (err)
+			goto err_pair;
+
+		mlx5_devcom_set_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS, true);
+		break;
+
+	case ESW_OFFLOADS_DEVCOM_UNPAIR:
+		if (!mlx5_devcom_is_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS))
+			break;
+
+		mlx5_devcom_set_paired(devcom, MLX5_DEVCOM_ESW_OFFLOADS, false);
+		mlx5_esw_offloads_unpair(peer_esw);
+		mlx5_esw_offloads_unpair(esw);
+		break;
+	}
+
+	return 0;
+
+err_pair:
+	mlx5_esw_offloads_unpair(esw);
+
+err_out:
+	mlx5_core_err(esw->dev, "esw offloads devcom event failure, event %u err %d",
+		      event, err);
+	return err;
+}
+
+static void esw_offloads_devcom_init(struct mlx5_eswitch *esw)
+{
+	struct mlx5_devcom *devcom = esw->dev->priv.devcom;
+
+	if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
+		return;
+
+	mlx5_devcom_register_component(devcom,
+				       MLX5_DEVCOM_ESW_OFFLOADS,
+				       mlx5_esw_offloads_devcom_event,
+				       esw);
+
+	mlx5_devcom_send_event(devcom,
+			       MLX5_DEVCOM_ESW_OFFLOADS,
+			       ESW_OFFLOADS_DEVCOM_PAIR, esw);
+}
+
+static void esw_offloads_devcom_cleanup(struct mlx5_eswitch *esw)
+{
+	struct mlx5_devcom *devcom = esw->dev->priv.devcom;
+
+	if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
+		return;
+
+	mlx5_devcom_send_event(devcom, MLX5_DEVCOM_ESW_OFFLOADS,
+			       ESW_OFFLOADS_DEVCOM_UNPAIR, esw);
+
+	mlx5_devcom_unregister_component(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+}
+
 int esw_offloads_init(struct mlx5_eswitch *esw, int nvports)
 {
 	int err;

@@ -1195,6 +1413,7 @@ int esw_offloads_init(struct mlx5_eswitch *esw, int nvports)
 	if (err)
 		goto err_reps;
 
+	esw_offloads_devcom_init(esw);
 	return 0;
 
 err_reps:

@@ -1233,6 +1452,7 @@ static int esw_offloads_stop(struct mlx5_eswitch *esw,
 
 void esw_offloads_cleanup(struct mlx5_eswitch *esw, int nvports)
 {
+	esw_offloads_devcom_cleanup(esw);
 	esw_offloads_unload_reps(esw, nvports);
 	esw_destroy_vport_rx_group(esw);
 	esw_destroy_offloads_table(esw);

@@ -7,6 +7,8 @@
 #include <linux/mlx5/driver.h>
 
 enum mlx5_devcom_components {
+	MLX5_DEVCOM_ESW_OFFLOADS,
+
 	MLX5_DEVCOM_NUM_COMPONENTS,
 };
 

@@ -52,6 +52,7 @@
 #include "fs_core.h"
 #include "en/port.h"
 #include "en/tc_tun.h"
+#include "lib/devcom.h"
 
 struct mlx5_nic_flow_attr {
 	u32 action;

@@ -74,6 +75,7 @@ enum {
 	MLX5E_TC_FLOW_HAIRPIN	= BIT(MLX5E_TC_FLOW_BASE + 3),
 	MLX5E_TC_FLOW_HAIRPIN_RSS = BIT(MLX5E_TC_FLOW_BASE + 4),
 	MLX5E_TC_FLOW_SLOW	  = BIT(MLX5E_TC_FLOW_BASE + 5),
+	MLX5E_TC_FLOW_DUP         = BIT(MLX5E_TC_FLOW_BASE + 6),
 };
 
 #define MLX5E_TC_MAX_SPLITS 1

@@ -111,8 +113,10 @@ struct mlx5e_tc_flow {
 	 * destinations.
 	 */
 	struct encap_flow_item encaps[MLX5_MAX_FLOW_FWD_VPORTS];
+	struct mlx5e_tc_flow    *peer_flow;
 	struct list_head	mod_hdr; /* flows sharing the same mod hdr ID */
 	struct list_head	hairpin; /* flows sharing the same hairpin */
+	struct list_head	peer;    /* flows with peer flow */
 	union {
 		struct mlx5_esw_flow_attr esw_attr[0];
 		struct mlx5_nic_flow_attr nic_attr[0];

@@ -1249,13 +1253,48 @@ static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 	}
 }
 
+static void __mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
+{
+	struct mlx5_eswitch *esw = flow->priv->mdev->priv.eswitch;
+
+	if (!(flow->flags & MLX5E_TC_FLOW_ESWITCH) ||
+	    !(flow->flags & MLX5E_TC_FLOW_DUP))
+		return;
+
+	mutex_lock(&esw->offloads.peer_mutex);
+	list_del(&flow->peer);
+	mutex_unlock(&esw->offloads.peer_mutex);
+
+	flow->flags &= ~MLX5E_TC_FLOW_DUP;
+
+	mlx5e_tc_del_fdb_flow(flow->peer_flow->priv, flow->peer_flow);
+	kvfree(flow->peer_flow);
+	flow->peer_flow = NULL;
+}
+
+static void mlx5e_tc_del_fdb_peer_flow(struct mlx5e_tc_flow *flow)
+{
+	struct mlx5_core_dev *dev = flow->priv->mdev;
+	struct mlx5_devcom *devcom = dev->priv.devcom;
+	struct mlx5_eswitch *peer_esw;
+
+	peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+	if (!peer_esw)
+		return;
+
+	__mlx5e_tc_del_fdb_peer_flow(flow);
+	mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+}
+
 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
 			      struct mlx5e_tc_flow *flow)
 {
-	if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
+	if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
+		mlx5e_tc_del_fdb_peer_flow(flow);
 		mlx5e_tc_del_fdb_flow(priv, flow);
-	else
+	} else {
 		mlx5e_tc_del_nic_flow(priv, flow);
+	}
 }
 
 

@@ -2660,6 +2699,11 @@ static struct rhashtable *get_tc_ht(struct mlx5e_priv *priv)
 		return &priv->fs.tc.ht;
 }
 
+static bool is_peer_flow_needed(struct mlx5e_tc_flow *flow)
+{
+	return false;
+}
+
 static int
 mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
 		 struct tc_cls_flower_offload *f, u16 flow_flags,

@@ -2693,11 +2737,13 @@ mlx5e_alloc_flow(struct mlx5e_priv *priv, int attr_size,
 }
 
 static int
-mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
-		   struct tc_cls_flower_offload *f,
-		   u16 flow_flags,
-		   struct net_device *filter_dev,
-		   struct mlx5e_tc_flow **__flow)
+__mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
+		     struct tc_cls_flower_offload *f,
+		     u16 flow_flags,
+		     struct net_device *filter_dev,
+		     struct mlx5_eswitch_rep *in_rep,
+		     struct mlx5_core_dev *in_mdev,
+		     struct mlx5e_tc_flow **__flow)
 {
 	struct netlink_ext_ack *extack = f->common.extack;
 	struct mlx5e_tc_flow_parse_attr *parse_attr;

@@ -2723,6 +2769,8 @@ mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
 	if (err)
 		goto err_free;
 
+	flow->esw_attr->in_rep = in_rep;
+	flow->esw_attr->in_mdev = in_mdev;
 	err = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow, extack);
 	if (err)
 		goto err_free;

@@ -2738,6 +2786,87 @@ mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
 	return err;
 }
 
+static int mlx5e_tc_add_fdb_peer_flow(struct tc_cls_flower_offload *f,
+				      struct mlx5e_tc_flow *flow)
+{
+	struct mlx5e_priv *priv = flow->priv, *peer_priv;
+	struct mlx5_eswitch *esw = priv->mdev->priv.eswitch, *peer_esw;
+	struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
+	struct mlx5e_tc_flow_parse_attr *parse_attr;
+	struct mlx5e_rep_priv *peer_urpriv;
+	struct mlx5e_tc_flow *peer_flow;
+	struct mlx5_core_dev *in_mdev;
+	int err = 0;
+
+	peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+	if (!peer_esw)
+		return -ENODEV;
+
+	peer_urpriv = mlx5_eswitch_get_uplink_priv(peer_esw, REP_ETH);
+	peer_priv = netdev_priv(peer_urpriv->netdev);
+
+	/* in_mdev is assigned of which the packet originated from.
+	 * So packets redirected to uplink use the same mdev of the
+	 * original flow and packets redirected from uplink use the
+	 * peer mdev.
+	 */
+	if (flow->esw_attr->in_rep->vport == FDB_UPLINK_VPORT)
+		in_mdev = peer_priv->mdev;
+	else
+		in_mdev = priv->mdev;
+
+	parse_attr = flow->esw_attr->parse_attr;
+	err = __mlx5e_add_fdb_flow(peer_priv, f, flow->flags,
+				   parse_attr->filter_dev,
+				   flow->esw_attr->in_rep, in_mdev, &peer_flow);
+	if (err)
+		goto out;
+
+	flow->peer_flow = peer_flow;
+	flow->flags |= MLX5E_TC_FLOW_DUP;
+	mutex_lock(&esw->offloads.peer_mutex);
+	list_add_tail(&flow->peer, &esw->offloads.peer_flows);
+	mutex_unlock(&esw->offloads.peer_mutex);
+
+out:
+	mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+	return err;
+}
+
+static int
+mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
+		   struct tc_cls_flower_offload *f,
+		   u16 flow_flags,
+		   struct net_device *filter_dev,
+		   struct mlx5e_tc_flow **__flow)
+{
+	struct mlx5e_rep_priv *rpriv = priv->ppriv;
+	struct mlx5_eswitch_rep *in_rep = rpriv->rep;
+	struct mlx5_core_dev *in_mdev = priv->mdev;
+	struct mlx5e_tc_flow *flow;
+	int err;
+
+	err = __mlx5e_add_fdb_flow(priv, f, flow_flags, filter_dev, in_rep,
+				   in_mdev, &flow);
+	if (err)
+		goto out;
+
+	if (is_peer_flow_needed(flow)) {
+		err = mlx5e_tc_add_fdb_peer_flow(f, flow);
+		if (err) {
+			mlx5e_tc_del_fdb_flow(priv, flow);
+			goto out;
+		}
+	}
+
+	*__flow = flow;
+
+	return 0;
+
+out:
+	return err;
+}
+
 static int
 mlx5e_add_nic_flow(struct mlx5e_priv *priv,
 		   struct tc_cls_flower_offload *f,

@@ -2882,7 +3011,9 @@ int mlx5e_delete_flower(struct net_device *dev, struct mlx5e_priv *priv,
 int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
 		       struct tc_cls_flower_offload *f, int flags)
 {
+	struct mlx5_devcom *devcom = priv->mdev->priv.devcom;
 	struct rhashtable *tc_ht = get_tc_ht(priv);
+	struct mlx5_eswitch *peer_esw;
 	struct mlx5e_tc_flow *flow;
 	struct mlx5_fc *counter;
 	u64 bytes;

@@ -2902,6 +3033,27 @@ int mlx5e_stats_flower(struct net_device *dev, struct mlx5e_priv *priv,
 
 	mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
 
+	peer_esw = mlx5_devcom_get_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+	if (!peer_esw)
+		goto out;
+
+	if ((flow->flags & MLX5E_TC_FLOW_DUP) &&
+	    (flow->peer_flow->flags & MLX5E_TC_FLOW_OFFLOADED)) {
+		u64 bytes2;
+		u64 packets2;
+		u64 lastuse2;
+
+		counter = mlx5e_tc_get_counter(flow->peer_flow);
+		mlx5_fc_query_cached(counter, &bytes2, &packets2, &lastuse2);
+
+		bytes += bytes2;
+		packets += packets2;
+		lastuse = max_t(u64, lastuse, lastuse2);
+	}
+
+	mlx5_devcom_release_peer_data(devcom, MLX5_DEVCOM_ESW_OFFLOADS);
+
+out:
 	tcf_exts_stats_update(f->exts, bytes, packets, lastuse);
 
 	return 0;

@@ -3014,3 +3166,11 @@ int mlx5e_tc_num_filters(struct mlx5e_priv *priv)
 
 	return atomic_read(&tc_ht->nelems);
 }
+
+void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw)
+{
+	struct mlx5e_tc_flow *flow, *tmp;
+
+	list_for_each_entry_safe(flow, tmp, &esw->offloads.peer_flows, peer)
+		__mlx5e_tc_del_fdb_peer_flow(flow);
+}

@@ -167,6 +167,8 @@ struct mlx5_esw_offload {
 	struct mlx5_flow_table *ft_offloads;
 	struct mlx5_flow_group *vport_rx_group;
 	struct mlx5_eswitch_rep *vport_reps;
+	struct list_head peer_flows;
+	struct mutex peer_mutex;
 	DECLARE_HASHTABLE(encap_tbl, 8);
 	DECLARE_HASHTABLE(mod_hdr_tbl, 8);
 	u8 inline_mode;

@@ -563,7 +563,7 @@ static void peer_miss_rules_setup(struct mlx5_core_dev *peer_dev,
 	dest->type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
 	dest->vport.num = 0;
 	dest->vport.vhca_id = MLX5_CAP_GEN(peer_dev, vhca_id);
-	dest->vport.vhca_id_valid = 1;
+	dest->vport.flags |= MLX5_FLOW_DEST_VPORT_VHCA_ID;
 }
 
 static int esw_add_fdb_peer_miss_rules(struct mlx5_eswitch *esw,

@@ -1313,8 +1313,11 @@ static int mlx5_esw_offloads_pair(struct mlx5_eswitch *esw,
 	return 0;
 }
 
+void mlx5e_tc_clean_fdb_peer_flows(struct mlx5_eswitch *esw);
+
 static void mlx5_esw_offloads_unpair(struct mlx5_eswitch *esw)
 {
+	mlx5e_tc_clean_fdb_peer_flows(esw);
 	esw_del_fdb_peer_miss_rules(esw);
 }
 

@@ -1365,6 +1368,9 @@ static void esw_offloads_devcom_init(struct mlx5_eswitch *esw)
 {
 	struct mlx5_devcom *devcom = esw->dev->priv.devcom;
 
+	INIT_LIST_HEAD(&esw->offloads.peer_flows);
+	mutex_init(&esw->offloads.peer_mutex);
+
 	if (!MLX5_CAP_ESW(esw->dev, merged_eswitch))
 		return;
 

@@ -994,7 +994,7 @@ mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
 	}
 
 	if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
-		counter = mlx5_fc_create(esw->dev, true);
+		counter = mlx5_fc_create(attr->counter_dev, true);
 		if (IS_ERR(counter)) {
 			err = PTR_ERR(counter);
 			goto err_create_counter;

@@ -1023,7 +1023,7 @@ mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
 	return 0;
 
 err_add_rule:
-	mlx5_fc_destroy(esw->dev, counter);
+	mlx5_fc_destroy(attr->counter_dev, counter);
 err_create_counter:
 	if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
 		mlx5e_detach_mod_hdr(priv, flow);

@@ -1064,7 +1064,7 @@ static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
 		mlx5e_detach_mod_hdr(priv, flow);
 
 	if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
-		mlx5_fc_destroy(esw->dev, attr->counter);
+		mlx5_fc_destroy(attr->counter_dev, attr->counter);
 }
 
 void mlx5e_tc_encap_flows_add(struct mlx5e_priv *priv,

@@ -2746,6 +2746,7 @@ __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
 		     struct mlx5e_tc_flow **__flow)
 {
 	struct netlink_ext_ack *extack = f->common.extack;
+	struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 	struct mlx5e_tc_flow_parse_attr *parse_attr;
 	struct mlx5e_tc_flow *flow;
 	int attr_size, err;

@@ -2771,6 +2772,13 @@ __mlx5e_add_fdb_flow(struct mlx5e_priv *priv,
 
 	flow->esw_attr->in_rep = in_rep;
 	flow->esw_attr->in_mdev = in_mdev;
+
+	if (MLX5_CAP_ESW(esw->dev, counter_eswitch_affinity) ==
+	    MLX5_COUNTER_SOURCE_ESWITCH)
+		flow->esw_attr->counter_dev = in_mdev;
+	else
+		flow->esw_attr->counter_dev = priv->mdev;
+
 	err = mlx5e_tc_add_fdb_flow(priv, parse_attr, flow, extack);
 	if (err)
 		goto err_free;

@@ -293,6 +293,7 @@ enum {
 struct mlx5_esw_flow_attr {
 	struct mlx5_eswitch_rep *in_rep;
 	struct mlx5_core_dev	*in_mdev;
+	struct mlx5_core_dev    *counter_dev;
 
 	int split_count;
 	int out_count;