Merge branch 'devlink-eswitch-inactive-mode'

Saeed Mahameed says:

====================
devlink eswitch inactive mode

Before having traffic flow through an eswitch, a user may want to have the
ability to block traffic towards the FDB until FDB is fully programmed and the
user is ready to send traffic to it. For example: when two eswitches are present
for vports in a multi-PF setup, one eswitch may take over the traffic from the
other when the user chooses. Before this take over, a user may want to first
program the inactive eswitch and then once ready redirect traffic to this new
eswitch.

This series introduces a user-configurable mode for an eswitch that allows
dynamically switching between active and inactive modes. When inactive, traffic
does not flow through the eswitch. While inactive, steering pipeline
configuration can be done (e.g. adding TC rules, discovering representors,
enabling the desired SDN modes such as bridge/OVS/DPDK/etc). Once configuration
is completed, a user can set the eswitch mode to active and have traffic flow
through. This allows admins to upgrade forwarding pipeline rules with very
minimal downtime and packet drops.

A user can start the eswitch in switchdev or switchdev_inactive mode.

Active: Traffic is enabled on this eswitch FDB.
Inactive: Traffic is ignored/dropped on this eswitch FDB.

An example use case:
$ devlink dev eswitch set pci/0000:08:00.1 mode switchdev_inactive
Setup FDB pipeline and netdev representors
...
Once ready to start receiving traffic
$ devlink dev eswitch set pci/0000:08:00.1 mode switchdev

v2: https://lore.kernel.org/all/20251107000831.157375-1-saeed@kernel.org/
v1: https://lore.kernel.org/all/20251016013618.2030940-1-saeed@kernel.org/
====================

Link: https://patch.msgid.link/20251108070404.1551708-1-saeed@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Documentation/netlink/specs/devlink.yaml

@@ -99,6 +99,8 @@ definitions:
         name: legacy
       -
         name: switchdev
+      -
+        name: switchdev-inactive
   -
     type: enum
     name: eswitch-inline-mode

Documentation/networking/devlink/devlink-eswitch-attr.rst

@@ -39,6 +39,10 @@ The following is a list of E-Switch attributes.
          rules.
        * ``switchdev`` allows for more advanced offloading capabilities of
          the E-Switch to hardware.
+       * ``switchdev_inactive`` switchdev mode but starts inactive, doesn't allow traffic
+         until explicitly activated. This mode is useful for orchestrators that
+         want to prepare the device in switchdev mode but only activate it when
+         all configurations are done.
    * - ``inline-mode``
      - enum
      - Some HWs need the VF driver to put part of the packet
@@ -74,3 +78,12 @@ Example Usage
 
     # enable encap-mode with legacy mode
     $ devlink dev eswitch set pci/0000:08:00.0 mode legacy inline-mode none encap-mode basic
+
+    # start switchdev mode in inactive state
+    $ devlink dev eswitch set pci/0000:08:00.0 mode switchdev_inactive
+
+    # setup switchdev configurations, representors, FDB entries, etc..
+    ...
+
+    # activate switchdev mode to allow traffic
+    $ devlink dev eswitch set pci/0000:08:00.0 mode switchdev

drivers/net/ethernet/mellanox/mlx5/core/esw/adj_vport.c

@@ -4,13 +4,8 @@
 #include "fs_core.h"
 #include "eswitch.h"
 
-enum {
-	MLX5_ADJ_VPORT_DISCONNECT = 0x0,
-	MLX5_ADJ_VPORT_CONNECT = 0x1,
-};
-
-static int mlx5_esw_adj_vport_modify(struct mlx5_core_dev *dev,
-				     u16 vport, bool connect)
+int mlx5_esw_adj_vport_modify(struct mlx5_core_dev *dev, u16 vport,
+			      bool connect)
 {
 	u32 in[MLX5_ST_SZ_DW(modify_vport_state_in)] = {};
 
@@ -24,7 +19,7 @@ static int mlx5_esw_adj_vport_modify(struct mlx5_core_dev *dev,
 	MLX5_SET(modify_vport_state_in, in, egress_connect_valid, 1);
 	MLX5_SET(modify_vport_state_in, in, ingress_connect, connect);
 	MLX5_SET(modify_vport_state_in, in, egress_connect, connect);
-
+	MLX5_SET(modify_vport_state_in, in, admin_state, connect);
 	return mlx5_cmd_exec_in(dev, modify_vport_state, in);
 }
 
@@ -96,7 +91,6 @@ static int mlx5_esw_adj_vport_create(struct mlx5_eswitch *esw, u16 vhca_id,
 	if (err)
 		goto acl_ns_remove;
 
-	mlx5_esw_adj_vport_modify(esw->dev, vport_num, MLX5_ADJ_VPORT_CONNECT);
 	return 0;
 
 acl_ns_remove:
@@ -117,8 +111,7 @@ static void mlx5_esw_adj_vport_destroy(struct mlx5_eswitch *esw,
 
 	esw_debug(esw->dev, "Destroying adjacent vport %d for vhca_id 0x%x\n",
 		  vport_num, vport->vhca_id);
-	mlx5_esw_adj_vport_modify(esw->dev, vport_num,
-				  MLX5_ADJ_VPORT_DISCONNECT);
+
 	mlx5_esw_offloads_rep_remove(esw, vport);
 	mlx5_fs_vport_egress_acl_ns_remove(esw->dev->priv.steering,
 					   vport->index);

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

@@ -264,6 +264,9 @@ struct mlx5_eswitch_fdb {
 
 		struct offloads_fdb {
 			struct mlx5_flow_namespace *ns;
+			struct mlx5_flow_table *drop_root;
+			struct mlx5_flow_handle *drop_root_rule;
+			struct mlx5_fc *drop_root_fc;
 			struct mlx5_flow_table *tc_miss_table;
 			struct mlx5_flow_table *slow_fdb;
 			struct mlx5_flow_group *send_to_vport_grp;
@@ -392,6 +395,7 @@ struct mlx5_eswitch {
 	struct mlx5_esw_offload offloads;
 	u32 last_vport_idx;
 	int                     mode;
+	bool                    offloads_inactive;
 	u16                     manager_vport;
 	u16                     first_host_vport;
 	u8			num_peers;
@@ -634,6 +638,8 @@ const u32 *mlx5_esw_query_functions(struct mlx5_core_dev *dev);
 
 void mlx5_esw_adjacent_vhcas_setup(struct mlx5_eswitch *esw);
 void mlx5_esw_adjacent_vhcas_cleanup(struct mlx5_eswitch *esw);
+int mlx5_esw_adj_vport_modify(struct mlx5_core_dev *dev, u16 vport,
+			      bool connect);
 
 #define MLX5_DEBUG_ESWITCH_MASK BIT(3)
 

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

@@ -1577,6 +1577,7 @@ esw_chains_create(struct mlx5_eswitch *esw, struct mlx5_flow_table *miss_fdb)
 	attr.max_grp_num = esw->params.large_group_num;
 	attr.default_ft = miss_fdb;
 	attr.mapping = esw->offloads.reg_c0_obj_pool;
+	attr.fs_base_prio = FDB_BYPASS_PATH;
 
 	chains = mlx5_chains_create(dev, &attr);
 	if (IS_ERR(chains)) {
@@ -2355,6 +2356,131 @@ static void esw_mode_change(struct mlx5_eswitch *esw, u16 mode)
 	mlx5_devcom_comp_unlock(esw->dev->priv.hca_devcom_comp);
 }
 
+static void mlx5_esw_fdb_drop_destroy(struct mlx5_eswitch *esw)
+{
+	if (!esw->fdb_table.offloads.drop_root)
+		return;
+
+	esw_debug(esw->dev, "Destroying FDB drop root table %#x fc %#x\n",
+		  esw->fdb_table.offloads.drop_root->id,
+		  esw->fdb_table.offloads.drop_root_fc->id);
+	mlx5_del_flow_rules(esw->fdb_table.offloads.drop_root_rule);
+	/* Don't free flow counter here, can be reused on a later activation */
+	mlx5_destroy_flow_table(esw->fdb_table.offloads.drop_root);
+	esw->fdb_table.offloads.drop_root_rule = NULL;
+	esw->fdb_table.offloads.drop_root = NULL;
+}
+
+static int mlx5_esw_fdb_drop_create(struct mlx5_eswitch *esw)
+{
+	struct mlx5_flow_destination drop_fc_dst = {};
+	struct mlx5_flow_table_attr ft_attr = {};
+	struct mlx5_flow_destination *dst = NULL;
+	struct mlx5_core_dev *dev = esw->dev;
+	struct mlx5_flow_namespace *root_ns;
+	struct mlx5_flow_act flow_act = {};
+	struct mlx5_flow_handle *flow_rule;
+	struct mlx5_flow_table *table;
+	int err = 0, dst_num = 0;
+
+	if (esw->fdb_table.offloads.drop_root)
+		return 0;
+
+	root_ns = esw->fdb_table.offloads.ns;
+
+	ft_attr.prio = FDB_DROP_ROOT;
+	ft_attr.max_fte = 1;
+	ft_attr.autogroup.max_num_groups = 1;
+	table = mlx5_create_auto_grouped_flow_table(root_ns, &ft_attr);
+	if (IS_ERR(table)) {
+		esw_warn(dev, "Failed to create fdb drop root table, err %pe\n",
+			 table);
+		return PTR_ERR(table);
+	}
+
+	/* Drop FC reusable, create once on first deactivation of FDB */
+	if (!esw->fdb_table.offloads.drop_root_fc) {
+		struct mlx5_fc *counter = mlx5_fc_create(dev, 0);
+
+		err = PTR_ERR_OR_ZERO(counter);
+		if (err)
+			esw_warn(esw->dev, "create fdb drop fc err %d\n", err);
+		else
+			esw->fdb_table.offloads.drop_root_fc = counter;
+	}
+
+	flow_act.action = MLX5_FLOW_CONTEXT_ACTION_DROP;
+
+	if (esw->fdb_table.offloads.drop_root_fc) {
+		flow_act.action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
+		drop_fc_dst.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
+		drop_fc_dst.counter = esw->fdb_table.offloads.drop_root_fc;
+		dst = &drop_fc_dst;
+		dst_num++;
+	}
+
+	flow_rule = mlx5_add_flow_rules(table, NULL, &flow_act, dst, dst_num);
+	err = PTR_ERR_OR_ZERO(flow_rule);
+	if (err) {
+		esw_warn(esw->dev,
+			 "fs offloads: Failed to add vport rx drop rule err %d\n",
+			 err);
+		goto err_flow_rule;
+	}
+
+	esw->fdb_table.offloads.drop_root = table;
+	esw->fdb_table.offloads.drop_root_rule = flow_rule;
+	esw_debug(esw->dev, "Created FDB drop root table %#x fc %#x\n",
+		  table->id, dst ? dst->counter->id : 0);
+	return 0;
+
+err_flow_rule:
+	/* no need to free drop fc, esw_offloads_steering_cleanup will do it */
+	mlx5_destroy_flow_table(table);
+	return err;
+}
+
+static void mlx5_esw_fdb_active(struct mlx5_eswitch *esw)
+{
+	struct mlx5_vport *vport;
+	unsigned long i;
+
+	mlx5_esw_fdb_drop_destroy(esw);
+	mlx5_mpfs_enable(esw->dev);
+
+	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
+		if (!vport->adjacent)
+			continue;
+		esw_debug(esw->dev, "Connecting vport %d to eswitch\n",
+			  vport->vport);
+		mlx5_esw_adj_vport_modify(esw->dev, vport->vport, true);
+	}
+
+	esw->offloads_inactive = false;
+	esw_warn(esw->dev, "MPFS/FDB active\n");
+}
+
+static void mlx5_esw_fdb_inactive(struct mlx5_eswitch *esw)
+{
+	struct mlx5_vport *vport;
+	unsigned long i;
+
+	mlx5_mpfs_disable(esw->dev);
+	mlx5_esw_fdb_drop_create(esw);
+
+	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
+		if (!vport->adjacent)
+			continue;
+		esw_debug(esw->dev, "Disconnecting vport %u from eswitch\n",
+			  vport->vport);
+
+		mlx5_esw_adj_vport_modify(esw->dev, vport->vport, false);
+	}
+
+	esw->offloads_inactive = true;
+	esw_warn(esw->dev, "MPFS/FDB inactive\n");
+}
+
 static int esw_offloads_start(struct mlx5_eswitch *esw,
 			      struct netlink_ext_ack *extack)
 {
@@ -3438,6 +3564,10 @@ create_indir_err:
 
 static void esw_offloads_steering_cleanup(struct mlx5_eswitch *esw)
 {
+	mlx5_esw_fdb_drop_destroy(esw);
+	if (esw->fdb_table.offloads.drop_root_fc)
+		mlx5_fc_destroy(esw->dev, esw->fdb_table.offloads.drop_root_fc);
+	esw->fdb_table.offloads.drop_root_fc = NULL;
 	esw_destroy_vport_rx_drop_rule(esw);
 	esw_destroy_vport_rx_drop_group(esw);
 	esw_destroy_vport_rx_group(esw);
@@ -3600,6 +3730,11 @@ int esw_offloads_enable(struct mlx5_eswitch *esw)
 	if (err)
 		goto err_steering_init;
 
+	if (esw->offloads_inactive)
+		mlx5_esw_fdb_inactive(esw);
+	else
+		mlx5_esw_fdb_active(esw);
+
 	/* Representor will control the vport link state */
 	mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs)
 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_DOWN;
@@ -3666,6 +3801,9 @@ void esw_offloads_disable(struct mlx5_eswitch *esw)
 	esw_offloads_metadata_uninit(esw);
 	mlx5_rdma_disable_roce(esw->dev);
 	mlx5_esw_adjacent_vhcas_cleanup(esw);
+	/* must be done after vhcas cleanup to avoid adjacent vports connect */
+	if (esw->offloads_inactive)
+		mlx5_esw_fdb_active(esw); /* legacy mode always active */
 	mutex_destroy(&esw->offloads.termtbl_mutex);
 }
 
@@ -3676,6 +3814,7 @@ static int esw_mode_from_devlink(u16 mode, u16 *mlx5_mode)
 		*mlx5_mode = MLX5_ESWITCH_LEGACY;
 		break;
 	case DEVLINK_ESWITCH_MODE_SWITCHDEV:
+	case DEVLINK_ESWITCH_MODE_SWITCHDEV_INACTIVE:
 		*mlx5_mode = MLX5_ESWITCH_OFFLOADS;
 		break;
 	default:
@@ -3685,14 +3824,17 @@ static int esw_mode_from_devlink(u16 mode, u16 *mlx5_mode)
 	return 0;
 }
 
-static int esw_mode_to_devlink(u16 mlx5_mode, u16 *mode)
+static int esw_mode_to_devlink(struct mlx5_eswitch *esw, u16 *mode)
 {
-	switch (mlx5_mode) {
+	switch (esw->mode) {
 	case MLX5_ESWITCH_LEGACY:
 		*mode = DEVLINK_ESWITCH_MODE_LEGACY;
 		break;
 	case MLX5_ESWITCH_OFFLOADS:
-		*mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
+		if (esw->offloads_inactive)
+			*mode = DEVLINK_ESWITCH_MODE_SWITCHDEV_INACTIVE;
+		else
+			*mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
 		break;
 	default:
 		return -EINVAL;
@@ -3798,6 +3940,45 @@ static bool mlx5_devlink_netdev_netns_immutable_set(struct devlink *devlink,
 	return ret;
 }
 
+/* Returns true when only changing between active and inactive switchdev mode */
+static bool mlx5_devlink_switchdev_active_mode_change(struct mlx5_eswitch *esw,
+						      u16 devlink_mode)
+{
+	/* current mode is not switchdev */
+	if (esw->mode != MLX5_ESWITCH_OFFLOADS)
+		return false;
+
+	/* new mode is not switchdev */
+	if (devlink_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV &&
+	    devlink_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV_INACTIVE)
+		return false;
+
+	/* already inactive: no change in current state */
+	if (devlink_mode == DEVLINK_ESWITCH_MODE_SWITCHDEV_INACTIVE &&
+	    esw->offloads_inactive)
+		return false;
+
+	/* already active: no change in current state */
+	if (devlink_mode == DEVLINK_ESWITCH_MODE_SWITCHDEV &&
+	    !esw->offloads_inactive)
+		return false;
+
+	down_write(&esw->mode_lock);
+	esw->offloads_inactive = !esw->offloads_inactive;
+	esw->eswitch_operation_in_progress = true;
+	up_write(&esw->mode_lock);
+
+	if (esw->offloads_inactive)
+		mlx5_esw_fdb_inactive(esw);
+	else
+		mlx5_esw_fdb_active(esw);
+
+	down_write(&esw->mode_lock);
+	esw->eswitch_operation_in_progress = false;
+	up_write(&esw->mode_lock);
+	return true;
+}
+
 int mlx5_devlink_eswitch_mode_set(struct devlink *devlink, u16 mode,
 				  struct netlink_ext_ack *extack)
 {
@@ -3812,12 +3993,16 @@ int mlx5_devlink_eswitch_mode_set(struct devlink *devlink, u16 mode,
 	if (esw_mode_from_devlink(mode, &mlx5_mode))
 		return -EINVAL;
 
-	if (mode == DEVLINK_ESWITCH_MODE_SWITCHDEV && mlx5_get_sd(esw->dev)) {
+	if (mlx5_mode == MLX5_ESWITCH_OFFLOADS && mlx5_get_sd(esw->dev)) {
 		NL_SET_ERR_MSG_MOD(extack,
 				   "Can't change E-Switch mode to switchdev when multi-PF netdev (Socket Direct) is configured.");
 		return -EPERM;
 	}
 
+	/* Avoid try_lock, active/inactive mode change is not restricted */
+	if (mlx5_devlink_switchdev_active_mode_change(esw, mode))
+		return 0;
+
 	mlx5_lag_disable_change(esw->dev);
 	err = mlx5_esw_try_lock(esw);
 	if (err < 0) {
@@ -3840,7 +4025,7 @@ int mlx5_devlink_eswitch_mode_set(struct devlink *devlink, u16 mode,
 	esw->eswitch_operation_in_progress = true;
 	up_write(&esw->mode_lock);
 
-	if (mode == DEVLINK_ESWITCH_MODE_SWITCHDEV &&
+	if (mlx5_mode == MLX5_ESWITCH_OFFLOADS &&
 	    !mlx5_devlink_netdev_netns_immutable_set(devlink, true)) {
 		NL_SET_ERR_MSG_MOD(extack,
 				   "Can't change E-Switch mode to switchdev when netdev net namespace has diverged from the devlink's.");
@@ -3848,25 +4033,27 @@ int mlx5_devlink_eswitch_mode_set(struct devlink *devlink, u16 mode,
 		goto skip;
 	}
 
-	if (mode == DEVLINK_ESWITCH_MODE_LEGACY)
+	if (mlx5_mode == MLX5_ESWITCH_LEGACY)
 		esw->dev->priv.flags |= MLX5_PRIV_FLAGS_SWITCH_LEGACY;
 	mlx5_eswitch_disable_locked(esw);
-	if (mode == DEVLINK_ESWITCH_MODE_SWITCHDEV) {
+	if (mlx5_mode == MLX5_ESWITCH_OFFLOADS) {
 		if (mlx5_devlink_trap_get_num_active(esw->dev)) {
 			NL_SET_ERR_MSG_MOD(extack,
 					   "Can't change mode while devlink traps are active");
 			err = -EOPNOTSUPP;
 			goto skip;
 		}
+		esw->offloads_inactive =
+			(mode == DEVLINK_ESWITCH_MODE_SWITCHDEV_INACTIVE);
 		err = esw_offloads_start(esw, extack);
-	} else if (mode == DEVLINK_ESWITCH_MODE_LEGACY) {
+	} else if (mlx5_mode == MLX5_ESWITCH_LEGACY) {
 		err = esw_offloads_stop(esw, extack);
 	} else {
 		err = -EINVAL;
 	}
 
 skip:
-	if (mode == DEVLINK_ESWITCH_MODE_SWITCHDEV && err)
+	if (mlx5_mode == MLX5_ESWITCH_OFFLOADS && err)
 		mlx5_devlink_netdev_netns_immutable_set(devlink, false);
 	down_write(&esw->mode_lock);
 	esw->eswitch_operation_in_progress = false;
@@ -3885,7 +4072,7 @@ int mlx5_devlink_eswitch_mode_get(struct devlink *devlink, u16 *mode)
 	if (IS_ERR(esw))
 		return PTR_ERR(esw);
 
-	return esw_mode_to_devlink(esw->mode, mode);
+	return esw_mode_to_devlink(esw, mode);
 }
 
 static int mlx5_esw_vports_inline_set(struct mlx5_eswitch *esw, u8 mlx5_mode,

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

@@ -3520,6 +3520,11 @@ static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
 	if (!steering->fdb_root_ns)
 		return -ENOMEM;
 
+	maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_DROP_ROOT, 1);
+	err = PTR_ERR_OR_ZERO(maj_prio);
+	if (err)
+		goto out_err;
+
 	err = create_fdb_bypass(steering);
 	if (err)
 		goto out_err;

drivers/net/ethernet/mellanox/mlx5/core/lib/mpfs.c

@@ -65,13 +65,14 @@ static int del_l2table_entry_cmd(struct mlx5_core_dev *dev, u32 index)
 /* UC L2 table hash node */
 struct l2table_node {
 	struct l2addr_node node;
-	u32                index; /* index in HW l2 table */
+	int                index; /* index in HW l2 table */
 	int                ref_count;
 };
 
 struct mlx5_mpfs {
 	struct hlist_head    hash[MLX5_L2_ADDR_HASH_SIZE];
 	struct mutex         lock; /* Synchronize l2 table access */
+	bool                 enabled;
 	u32                  size;
 	unsigned long        *bitmap;
 };
@@ -114,6 +115,8 @@ int mlx5_mpfs_init(struct mlx5_core_dev *dev)
 		return -ENOMEM;
 	}
 
+	mpfs->enabled = true;
+
 	dev->priv.mpfs = mpfs;
 	return 0;
 }
@@ -135,7 +138,7 @@ int mlx5_mpfs_add_mac(struct mlx5_core_dev *dev, u8 *mac)
 	struct mlx5_mpfs *mpfs = dev->priv.mpfs;
 	struct l2table_node *l2addr;
 	int err = 0;
-	u32 index;
+	int index;
 
 	if (!mpfs)
 		return 0;
@@ -148,30 +151,34 @@ int mlx5_mpfs_add_mac(struct mlx5_core_dev *dev, u8 *mac)
 		goto out;
 	}
 
-	err = alloc_l2table_index(mpfs, &index);
-	if (err)
-		goto out;
-
 	l2addr = l2addr_hash_add(mpfs->hash, mac, struct l2table_node, GFP_KERNEL);
 	if (!l2addr) {
 		err = -ENOMEM;
-		goto hash_add_err;
+		goto out;
 	}
 
-	err = set_l2table_entry_cmd(dev, index, mac);
-	if (err)
-		goto set_table_entry_err;
+	index = -1;
+
+	if (mpfs->enabled) {
+		err = alloc_l2table_index(mpfs, &index);
+		if (err)
+			goto hash_del;
+		err = set_l2table_entry_cmd(dev, index, mac);
+		if (err)
+			goto free_l2table_index;
+		mlx5_core_dbg(dev, "MPFS entry %pM, set @index (%d)\n",
+			      l2addr->node.addr, index);
+	}
 
 	l2addr->index = index;
 	l2addr->ref_count = 1;
 
 	mlx5_core_dbg(dev, "MPFS mac added %pM, index (%d)\n", mac, index);
 	goto out;
-
-set_table_entry_err:
-	l2addr_hash_del(l2addr);
-hash_add_err:
+free_l2table_index:
 	free_l2table_index(mpfs, index);
+hash_del:
+	l2addr_hash_del(l2addr);
 out:
 	mutex_unlock(&mpfs->lock);
 	return err;
@@ -183,7 +190,7 @@ int mlx5_mpfs_del_mac(struct mlx5_core_dev *dev, u8 *mac)
 	struct mlx5_mpfs *mpfs = dev->priv.mpfs;
 	struct l2table_node *l2addr;
 	int err = 0;
-	u32 index;
+	int index;
 
 	if (!mpfs)
 		return 0;
@@ -200,12 +207,87 @@ int mlx5_mpfs_del_mac(struct mlx5_core_dev *dev, u8 *mac)
 		goto unlock;
 
 	index = l2addr->index;
-	del_l2table_entry_cmd(dev, index);
+	if (index >= 0) {
+		del_l2table_entry_cmd(dev, index);
+		free_l2table_index(mpfs, index);
+		mlx5_core_dbg(dev, "MPFS entry %pM, deleted @index (%d)\n",
+			      mac, index);
+	}
 	l2addr_hash_del(l2addr);
-	free_l2table_index(mpfs, index);
 	mlx5_core_dbg(dev, "MPFS mac deleted %pM, index (%d)\n", mac, index);
 unlock:
 	mutex_unlock(&mpfs->lock);
 	return err;
 }
 EXPORT_SYMBOL(mlx5_mpfs_del_mac);
+
+int mlx5_mpfs_enable(struct mlx5_core_dev *dev)
+{
+	struct mlx5_mpfs *mpfs = dev->priv.mpfs;
+	struct l2table_node *l2addr;
+	struct hlist_node *n;
+	int err = 0, i;
+
+	if (!mpfs)
+		return -ENODEV;
+
+	mutex_lock(&mpfs->lock);
+	if (mpfs->enabled)
+		goto out;
+	mpfs->enabled = true;
+	mlx5_core_dbg(dev, "MPFS enabling mpfs\n");
+
+	mlx5_mpfs_foreach(l2addr, n, mpfs, i) {
+		u32 index;
+
+		err = alloc_l2table_index(mpfs, &index);
+		if (err) {
+			mlx5_core_err(dev, "Failed to allocated MPFS index for %pM, err(%d)\n",
+				      l2addr->node.addr, err);
+			goto out;
+		}
+
+		err = set_l2table_entry_cmd(dev, index, l2addr->node.addr);
+		if (err) {
+			mlx5_core_err(dev, "Failed to set MPFS l2table entry for %pM index=%d, err(%d)\n",
+				      l2addr->node.addr, index, err);
+			free_l2table_index(mpfs, index);
+			goto out;
+		}
+
+		l2addr->index = index;
+		mlx5_core_dbg(dev, "MPFS entry %pM, set @index (%d)\n",
+			      l2addr->node.addr, l2addr->index);
+	}
+out:
+	mutex_unlock(&mpfs->lock);
+	return err;
+}
+
+void mlx5_mpfs_disable(struct mlx5_core_dev *dev)
+{
+	struct mlx5_mpfs *mpfs = dev->priv.mpfs;
+	struct l2table_node *l2addr;
+	struct hlist_node *n;
+	int i;
+
+	if (!mpfs)
+		return;
+
+	mutex_lock(&mpfs->lock);
+	if (!mpfs->enabled)
+		goto unlock;
+	mlx5_mpfs_foreach(l2addr, n, mpfs, i) {
+		if (l2addr->index < 0)
+			continue;
+		del_l2table_entry_cmd(dev, l2addr->index);
+		free_l2table_index(mpfs, l2addr->index);
+		mlx5_core_dbg(dev, "MPFS entry %pM, deleted @index (%d)\n",
+			      l2addr->node.addr, l2addr->index);
+		l2addr->index = -1;
+	}
+	mpfs->enabled = false;
+	mlx5_core_dbg(dev, "MPFS disabled\n");
+unlock:
+	mutex_unlock(&mpfs->lock);
+}

drivers/net/ethernet/mellanox/mlx5/core/lib/mpfs.h

@@ -45,6 +45,10 @@ struct l2addr_node {
 	u8                addr[ETH_ALEN];
 };
 
+#define mlx5_mpfs_foreach(hs, tmp, mpfs, i) \
+	for (i = 0; i < MLX5_L2_ADDR_HASH_SIZE; i++) \
+		hlist_for_each_entry_safe(hs, tmp, &(mpfs)->hash[i], node.hlist)
+
 #define for_each_l2hash_node(hn, tmp, hash, i) \
 	for (i = 0; i < MLX5_L2_ADDR_HASH_SIZE; i++) \
 		hlist_for_each_entry_safe(hn, tmp, &(hash)[i], hlist)
@@ -82,11 +86,16 @@ struct l2addr_node {
 })
 
 #ifdef CONFIG_MLX5_MPFS
+struct mlx5_core_dev;
 int  mlx5_mpfs_init(struct mlx5_core_dev *dev);
 void mlx5_mpfs_cleanup(struct mlx5_core_dev *dev);
+int mlx5_mpfs_enable(struct mlx5_core_dev *dev);
+void mlx5_mpfs_disable(struct mlx5_core_dev *dev);
 #else /* #ifndef CONFIG_MLX5_MPFS */
 static inline int  mlx5_mpfs_init(struct mlx5_core_dev *dev) { return 0; }
 static inline void mlx5_mpfs_cleanup(struct mlx5_core_dev *dev) {}
+static inline int mlx5_mpfs_enable(struct mlx5_core_dev *dev) { return 0; }
+static inline void mlx5_mpfs_disable(struct mlx5_core_dev *dev) {}
 #endif
 
 #endif

include/linux/mlx5/fs.h

@@ -116,6 +116,7 @@ enum mlx5_flow_namespace_type {
 };
 
 enum {
+	FDB_DROP_ROOT,
 	FDB_BYPASS_PATH,
 	FDB_CRYPTO_INGRESS,
 	FDB_TC_OFFLOAD,

include/uapi/linux/devlink.h

@@ -181,6 +181,7 @@ enum devlink_sb_threshold_type {
 enum devlink_eswitch_mode {
 	DEVLINK_ESWITCH_MODE_LEGACY,
 	DEVLINK_ESWITCH_MODE_SWITCHDEV,
+	DEVLINK_ESWITCH_MODE_SWITCHDEV_INACTIVE,
 };
 
 enum devlink_eswitch_inline_mode {

net/devlink/netlink_gen.c

@@ -229,7 +229,7 @@ static const struct nla_policy devlink_eswitch_get_nl_policy[DEVLINK_ATTR_DEV_NA
 static const struct nla_policy devlink_eswitch_set_nl_policy[DEVLINK_ATTR_ESWITCH_ENCAP_MODE + 1] = {
 	[DEVLINK_ATTR_BUS_NAME] = { .type = NLA_NUL_STRING, },
 	[DEVLINK_ATTR_DEV_NAME] = { .type = NLA_NUL_STRING, },
-	[DEVLINK_ATTR_ESWITCH_MODE] = NLA_POLICY_MAX(NLA_U16, 1),
+	[DEVLINK_ATTR_ESWITCH_MODE] = NLA_POLICY_MAX(NLA_U16, 2),
 	[DEVLINK_ATTR_ESWITCH_INLINE_MODE] = NLA_POLICY_MAX(NLA_U8, 3),
 	[DEVLINK_ATTR_ESWITCH_ENCAP_MODE] = NLA_POLICY_MAX(NLA_U8, 1),
 };