nfp: flower: monitor and offload LAG groups

Monitor LAG events via the NETDEV_CHANGEUPPER/NETDEV_CHANGELOWERSTATE
notifiers to maintain a list of offloadable groups. Sync these groups with
HW via a delayed workqueue to prevent excessive re-configuration. When the
workqueue is triggered it may generate multiple control messages for
different groups. These messages are linked via a batch ID and flags to
indicate a new batch and the end of a batch.

Update private data in each repr to track their LAG lower state flags. The
state of a repr is used to determine the active netdevs that can be
offloaded. For example, in active-backup mode, we only offload the netdev
currently active.

Signed-off-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Pieter Jansen van Vuuren <pieter.jansenvanvuuren@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 646 insertions, 3 deletions

diff --git a/drivers/net/ethernet/netronome/nfp/Makefile b/drivers/net/ethernet/netronome/nfp/Makefile
index 6373f56205fd..4afb10375397 100644
--- a/drivers/net/ethernet/netronome/nfp/Makefile
+++ b/drivers/net/ethernet/netronome/nfp/Makefile
@@ -37,6 +37,7 @@ ifeq ($(CONFIG_NFP_APP_FLOWER),y)
 nfp-objs += \
 	    flower/action.o \
 	    flower/cmsg.o \
+	    flower/lag_conf.o \
 	    flower/main.o \
 	    flower/match.o \
 	    flower/metadata.o \
diff --git a/drivers/net/ethernet/netronome/nfp/flower/lag_conf.c b/drivers/net/ethernet/netronome/nfp/flower/lag_conf.c
new file mode 100644
index 000000000000..35a700b879d7
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/flower/lag_conf.c
@@ -0,0 +1,589 @@
+/*
+ * Copyright (C) 2018 Netronome Systems, Inc.
+ *
+ * This software is dual licensed under the GNU General License Version 2,
+ * June 1991 as shown in the file COPYING in the top-level directory of this
+ * source tree or the BSD 2-Clause License provided below.  You have the
+ * option to license this software under the complete terms of either license.
+ *
+ * The BSD 2-Clause License:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      1. Redistributions of source code must retain the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer.
+ *
+ *      2. Redistributions in binary form must reproduce the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer in the documentation and/or other materials
+ *         provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "main.h"
+
+/* LAG group config flags. */
+#define NFP_FL_LAG_LAST			BIT(1)
+#define NFP_FL_LAG_FIRST		BIT(2)
+#define NFP_FL_LAG_SWITCH		BIT(6)
+#define NFP_FL_LAG_RESET		BIT(7)
+
+/* LAG port state flags. */
+#define NFP_PORT_LAG_LINK_UP		BIT(0)
+#define NFP_PORT_LAG_TX_ENABLED		BIT(1)
+#define NFP_PORT_LAG_CHANGED		BIT(2)
+
+enum nfp_fl_lag_batch {
+	NFP_FL_LAG_BATCH_FIRST,
+	NFP_FL_LAG_BATCH_MEMBER,
+	NFP_FL_LAG_BATCH_FINISHED
+};
+
+/**
+ * struct nfp_flower_cmsg_lag_config - control message payload for LAG config
+ * @ctrl_flags:	Configuration flags
+ * @reserved:	Reserved for future use
+ * @ttl:	Time to live of packet - host always sets to 0xff
+ * @pkt_number:	Config message packet number - increment for each message
+ * @batch_ver:	Batch version of messages - increment for each batch of messages
+ * @group_id:	Group ID applicable
+ * @group_inst:	Group instance number - increment when group is reused
+ * @members:	Array of 32-bit words listing all active group members
+ */
+struct nfp_flower_cmsg_lag_config {
+	u8 ctrl_flags;
+	u8 reserved[2];
+	u8 ttl;
+	__be32 pkt_number;
+	__be32 batch_ver;
+	__be32 group_id;
+	__be32 group_inst;
+	__be32 members[];
+};
+
+/**
+ * struct nfp_fl_lag_group - list entry for each LAG group
+ * @group_id:		Assigned group ID for host/kernel sync
+ * @group_inst:		Group instance in case of ID reuse
+ * @list:		List entry
+ * @master_ndev:	Group master Netdev
+ * @dirty:		Marked if the group needs synced to HW
+ * @offloaded:		Marked if the group is currently offloaded to NIC
+ * @to_remove:		Marked if the group should be removed from NIC
+ * @to_destroy:		Marked if the group should be removed from driver
+ * @slave_cnt:		Number of slaves in group
+ */
+struct nfp_fl_lag_group {
+	unsigned int group_id;
+	u8 group_inst;
+	struct list_head list;
+	struct net_device *master_ndev;
+	bool dirty;
+	bool offloaded;
+	bool to_remove;
+	bool to_destroy;
+	unsigned int slave_cnt;
+};
+
+#define NFP_FL_LAG_PKT_NUMBER_MASK	GENMASK(30, 0)
+#define NFP_FL_LAG_VERSION_MASK		GENMASK(22, 0)
+#define NFP_FL_LAG_HOST_TTL		0xff
+
+/* Use this ID with zero members to ack a batch config */
+#define NFP_FL_LAG_SYNC_ID		0
+#define NFP_FL_LAG_GROUP_MIN		1 /* ID 0 reserved */
+#define NFP_FL_LAG_GROUP_MAX		32 /* IDs 1 to 31 are valid */
+
+/* wait for more config */
+#define NFP_FL_LAG_DELAY		(msecs_to_jiffies(2))
+
+static unsigned int nfp_fl_get_next_pkt_number(struct nfp_fl_lag *lag)
+{
+	lag->pkt_num++;
+	lag->pkt_num &= NFP_FL_LAG_PKT_NUMBER_MASK;
+
+	return lag->pkt_num;
+}
+
+static void nfp_fl_increment_version(struct nfp_fl_lag *lag)
+{
+	/* LSB is not considered by firmware so add 2 for each increment. */
+	lag->batch_ver += 2;
+	lag->batch_ver &= NFP_FL_LAG_VERSION_MASK;
+
+	/* Zero is reserved by firmware. */
+	if (!lag->batch_ver)
+		lag->batch_ver += 2;
+}
+
+static struct nfp_fl_lag_group *
+nfp_fl_lag_group_create(struct nfp_fl_lag *lag, struct net_device *master)
+{
+	struct nfp_fl_lag_group *group;
+	struct nfp_flower_priv *priv;
+	int id;
+
+	priv = container_of(lag, struct nfp_flower_priv, nfp_lag);
+
+	id = ida_simple_get(&lag->ida_handle, NFP_FL_LAG_GROUP_MIN,
+			    NFP_FL_LAG_GROUP_MAX, GFP_KERNEL);
+	if (id < 0) {
+		nfp_flower_cmsg_warn(priv->app,
+				     "No more bonding groups available\n");
+		return ERR_PTR(id);
+	}
+
+	group = kmalloc(sizeof(*group), GFP_KERNEL);
+	if (!group) {
+		ida_simple_remove(&lag->ida_handle, id);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	group->group_id = id;
+	group->master_ndev = master;
+	group->dirty = true;
+	group->offloaded = false;
+	group->to_remove = false;
+	group->to_destroy = false;
+	group->slave_cnt = 0;
+	group->group_inst = ++lag->global_inst;
+	list_add_tail(&group->list, &lag->group_list);
+
+	return group;
+}
+
+static struct nfp_fl_lag_group *
+nfp_fl_lag_find_group_for_master_with_lag(struct nfp_fl_lag *lag,
+					  struct net_device *master)
+{
+	struct nfp_fl_lag_group *entry;
+
+	if (!master)
+		return NULL;
+
+	list_for_each_entry(entry, &lag->group_list, list)
+		if (entry->master_ndev == master)
+			return entry;
+
+	return NULL;
+}
+
+static int
+nfp_fl_lag_config_group(struct nfp_fl_lag *lag, struct nfp_fl_lag_group *group,
+			struct net_device **active_members,
+			unsigned int member_cnt, enum nfp_fl_lag_batch *batch)
+{
+	struct nfp_flower_cmsg_lag_config *cmsg_payload;
+	struct nfp_flower_priv *priv;
+	unsigned long int flags;
+	unsigned int size, i;
+	struct sk_buff *skb;
+
+	priv = container_of(lag, struct nfp_flower_priv, nfp_lag);
+	size = sizeof(*cmsg_payload) + sizeof(__be32) * member_cnt;
+	skb = nfp_flower_cmsg_alloc(priv->app, size,
+				    NFP_FLOWER_CMSG_TYPE_LAG_CONFIG,
+				    GFP_KERNEL);
+	if (!skb)
+		return -ENOMEM;
+
+	cmsg_payload = nfp_flower_cmsg_get_data(skb);
+	flags = 0;
+
+	/* Increment batch version for each new batch of config messages. */
+	if (*batch == NFP_FL_LAG_BATCH_FIRST) {
+		flags |= NFP_FL_LAG_FIRST;
+		nfp_fl_increment_version(lag);
+		*batch = NFP_FL_LAG_BATCH_MEMBER;
+	}
+
+	/* If it is a reset msg then it is also the end of the batch. */
+	if (lag->rst_cfg) {
+		flags |= NFP_FL_LAG_RESET;
+		*batch = NFP_FL_LAG_BATCH_FINISHED;
+	}
+
+	/* To signal the end of a batch, both the switch and last flags are set
+	 * and the the reserved SYNC group ID is used.
+	 */
+	if (*batch == NFP_FL_LAG_BATCH_FINISHED) {
+		flags |= NFP_FL_LAG_SWITCH | NFP_FL_LAG_LAST;
+		lag->rst_cfg = false;
+		cmsg_payload->group_id = cpu_to_be32(NFP_FL_LAG_SYNC_ID);
+		cmsg_payload->group_inst = 0;
+	} else {
+		cmsg_payload->group_id = cpu_to_be32(group->group_id);
+		cmsg_payload->group_inst = cpu_to_be32(group->group_inst);
+	}
+
+	cmsg_payload->reserved[0] = 0;
+	cmsg_payload->reserved[1] = 0;
+	cmsg_payload->ttl = NFP_FL_LAG_HOST_TTL;
+	cmsg_payload->ctrl_flags = flags;
+	cmsg_payload->batch_ver = cpu_to_be32(lag->batch_ver);
+	cmsg_payload->pkt_number = cpu_to_be32(nfp_fl_get_next_pkt_number(lag));
+
+	for (i = 0; i < member_cnt; i++)
+		cmsg_payload->members[i] =
+			cpu_to_be32(nfp_repr_get_port_id(active_members[i]));
+
+	nfp_ctrl_tx(priv->app->ctrl, skb);
+	return 0;
+}
+
+static void nfp_fl_lag_do_work(struct work_struct *work)
+{
+	enum nfp_fl_lag_batch batch = NFP_FL_LAG_BATCH_FIRST;
+	struct nfp_fl_lag_group *entry, *storage;
+	struct delayed_work *delayed_work;
+	struct nfp_flower_priv *priv;
+	struct nfp_fl_lag *lag;
+	int err;
+
+	delayed_work = to_delayed_work(work);
+	lag = container_of(delayed_work, struct nfp_fl_lag, work);
+	priv = container_of(lag, struct nfp_flower_priv, nfp_lag);
+
+	mutex_lock(&lag->lock);
+	list_for_each_entry_safe(entry, storage, &lag->group_list, list) {
+		struct net_device *iter_netdev, **acti_netdevs;
+		struct nfp_flower_repr_priv *repr_priv;
+		int active_count = 0, slaves = 0;
+		struct nfp_repr *repr;
+		unsigned long *flags;
+
+		if (entry->to_remove) {
+			/* Active count of 0 deletes group on hw. */
+			err = nfp_fl_lag_config_group(lag, entry, NULL, 0,
+						      &batch);
+			if (!err) {
+				entry->to_remove = false;
+				entry->offloaded = false;
+			} else {
+				nfp_flower_cmsg_warn(priv->app,
+						     "group delete failed\n");
+				schedule_delayed_work(&lag->work,
+						      NFP_FL_LAG_DELAY);
+				continue;
+			}
+
+			if (entry->to_destroy) {
+				ida_simple_remove(&lag->ida_handle,
+						  entry->group_id);
+				list_del(&entry->list);
+				kfree(entry);
+			}
+			continue;
+		}
+
+		acti_netdevs = kmalloc_array(entry->slave_cnt,
+					     sizeof(*acti_netdevs), GFP_KERNEL);
+
+		/* Include sanity check in the loop. It may be that a bond has
+		 * changed between processing the last notification and the
+		 * work queue triggering. If the number of slaves has changed
+		 * or it now contains netdevs that cannot be offloaded, ignore
+		 * the group until pending notifications are processed.
+		 */
+		rcu_read_lock();
+		for_each_netdev_in_bond_rcu(entry->master_ndev, iter_netdev) {
+			if (!nfp_netdev_is_nfp_repr(iter_netdev)) {
+				slaves = 0;
+				break;
+			}
+
+			repr = netdev_priv(iter_netdev);
+
+			if (repr->app != priv->app) {
+				slaves = 0;
+				break;
+			}
+
+			slaves++;
+			if (slaves > entry->slave_cnt)
+				break;
+
+			/* Check the ports for state changes. */
+			repr_priv = repr->app_priv;
+			flags = &repr_priv->lag_port_flags;
+
+			if (*flags & NFP_PORT_LAG_CHANGED) {
+				*flags &= ~NFP_PORT_LAG_CHANGED;
+				entry->dirty = true;
+			}
+
+			if ((*flags & NFP_PORT_LAG_TX_ENABLED) &&
+			    (*flags & NFP_PORT_LAG_LINK_UP))
+				acti_netdevs[active_count++] = iter_netdev;
+		}
+		rcu_read_unlock();
+
+		if (slaves != entry->slave_cnt || !entry->dirty) {
+			kfree(acti_netdevs);
+			continue;
+		}
+
+		err = nfp_fl_lag_config_group(lag, entry, acti_netdevs,
+					      active_count, &batch);
+		if (!err) {
+			entry->offloaded = true;
+			entry->dirty = false;
+		} else {
+			nfp_flower_cmsg_warn(priv->app,
+					     "group offload failed\n");
+			schedule_delayed_work(&lag->work, NFP_FL_LAG_DELAY);
+		}
+
+		kfree(acti_netdevs);
+	}
+
+	/* End the config batch if at least one packet has been batched. */
+	if (batch == NFP_FL_LAG_BATCH_MEMBER) {
+		batch = NFP_FL_LAG_BATCH_FINISHED;
+		err = nfp_fl_lag_config_group(lag, NULL, NULL, 0, &batch);
+		if (err)
+			nfp_flower_cmsg_warn(priv->app,
+					     "group batch end cmsg failed\n");
+	}
+
+	mutex_unlock(&lag->lock);
+}
+
+static void
+nfp_fl_lag_schedule_group_remove(struct nfp_fl_lag *lag,
+				 struct nfp_fl_lag_group *group)
+{
+	group->to_remove = true;
+
+	schedule_delayed_work(&lag->work, NFP_FL_LAG_DELAY);
+}
+
+static int
+nfp_fl_lag_schedule_group_delete(struct nfp_fl_lag *lag,
+				 struct net_device *master)
+{
+	struct nfp_fl_lag_group *group;
+
+	mutex_lock(&lag->lock);
+	group = nfp_fl_lag_find_group_for_master_with_lag(lag, master);
+	if (!group) {
+		mutex_unlock(&lag->lock);
+		return -ENOENT;
+	}
+
+	group->to_remove = true;
+	group->to_destroy = true;
+	mutex_unlock(&lag->lock);
+
+	schedule_delayed_work(&lag->work, NFP_FL_LAG_DELAY);
+	return 0;
+}
+
+static int
+nfp_fl_lag_changeupper_event(struct nfp_fl_lag *lag,
+			     struct netdev_notifier_changeupper_info *info)
+{
+	struct net_device *upper = info->upper_dev, *iter_netdev;
+	struct netdev_lag_upper_info *lag_upper_info;
+	struct nfp_fl_lag_group *group;
+	struct nfp_flower_priv *priv;
+	unsigned int slave_count = 0;
+	bool can_offload = true;
+	struct nfp_repr *repr;
+
+	if (!netif_is_lag_master(upper))
+		return 0;
+
+	priv = container_of(lag, struct nfp_flower_priv, nfp_lag);
+
+	rcu_read_lock();
+	for_each_netdev_in_bond_rcu(upper, iter_netdev) {
+		if (!nfp_netdev_is_nfp_repr(iter_netdev)) {
+			can_offload = false;
+			break;
+		}
+		repr = netdev_priv(iter_netdev);
+
+		/* Ensure all ports are created by the same app/on same card. */
+		if (repr->app != priv->app) {
+			can_offload = false;
+			break;
+		}
+
+		slave_count++;
+	}
+	rcu_read_unlock();
+
+	lag_upper_info = info->upper_info;
+
+	/* Firmware supports active/backup and L3/L4 hash bonds. */
+	if (lag_upper_info &&
+	    lag_upper_info->tx_type != NETDEV_LAG_TX_TYPE_ACTIVEBACKUP &&
+	    (lag_upper_info->tx_type != NETDEV_LAG_TX_TYPE_HASH ||
+	    (lag_upper_info->hash_type != NETDEV_LAG_HASH_L34 &&
+	    lag_upper_info->hash_type != NETDEV_LAG_HASH_E34))) {
+		can_offload = false;
+		nfp_flower_cmsg_warn(priv->app,
+				     "Unable to offload tx_type %u hash %u\n",
+				     lag_upper_info->tx_type,
+				     lag_upper_info->hash_type);
+	}
+
+	mutex_lock(&lag->lock);
+	group = nfp_fl_lag_find_group_for_master_with_lag(lag, upper);
+
+	if (slave_count == 0 || !can_offload) {
+		/* Cannot offload the group - remove if previously offloaded. */
+		if (group && group->offloaded)
+			nfp_fl_lag_schedule_group_remove(lag, group);
+
+		mutex_unlock(&lag->lock);
+		return 0;
+	}
+
+	if (!group) {
+		group = nfp_fl_lag_group_create(lag, upper);
+		if (IS_ERR(group)) {
+			mutex_unlock(&lag->lock);
+			return PTR_ERR(group);
+		}
+	}
+
+	group->dirty = true;
+	group->slave_cnt = slave_count;
+
+	/* Group may have been on queue for removal but is now offfloable. */
+	group->to_remove = false;
+	mutex_unlock(&lag->lock);
+
+	schedule_delayed_work(&lag->work, NFP_FL_LAG_DELAY);
+	return 0;
+}
+
+static int
+nfp_fl_lag_changels_event(struct nfp_fl_lag *lag, struct net_device *netdev,
+			  struct netdev_notifier_changelowerstate_info *info)
+{
+	struct netdev_lag_lower_state_info *lag_lower_info;
+	struct nfp_flower_repr_priv *repr_priv;
+	struct nfp_flower_priv *priv;
+	struct nfp_repr *repr;
+	unsigned long *flags;
+
+	if (!netif_is_lag_port(netdev) || !nfp_netdev_is_nfp_repr(netdev))
+		return 0;
+
+	lag_lower_info = info->lower_state_info;
+	if (!lag_lower_info)
+		return 0;
+
+	priv = container_of(lag, struct nfp_flower_priv, nfp_lag);
+	repr = netdev_priv(netdev);
+
+	/* Verify that the repr is associated with this app. */
+	if (repr->app != priv->app)
+		return 0;
+
+	repr_priv = repr->app_priv;
+	flags = &repr_priv->lag_port_flags;
+
+	mutex_lock(&lag->lock);
+	if (lag_lower_info->link_up)
+		*flags |= NFP_PORT_LAG_LINK_UP;
+	else
+		*flags &= ~NFP_PORT_LAG_LINK_UP;
+
+	if (lag_lower_info->tx_enabled)
+		*flags |= NFP_PORT_LAG_TX_ENABLED;
+	else
+		*flags &= ~NFP_PORT_LAG_TX_ENABLED;
+
+	*flags |= NFP_PORT_LAG_CHANGED;
+	mutex_unlock(&lag->lock);
+
+	schedule_delayed_work(&lag->work, NFP_FL_LAG_DELAY);
+	return 0;
+}
+
+static int
+nfp_fl_lag_netdev_event(struct notifier_block *nb, unsigned long event,
+			void *ptr)
+{
+	struct net_device *netdev;
+	struct nfp_fl_lag *lag;
+	int err;
+
+	netdev = netdev_notifier_info_to_dev(ptr);
+	lag = container_of(nb, struct nfp_fl_lag, lag_nb);
+
+	switch (event) {
+	case NETDEV_CHANGEUPPER:
+		err = nfp_fl_lag_changeupper_event(lag, ptr);
+		if (err)
+			return NOTIFY_BAD;
+		return NOTIFY_OK;
+	case NETDEV_CHANGELOWERSTATE:
+		err = nfp_fl_lag_changels_event(lag, netdev, ptr);
+		if (err)
+			return NOTIFY_BAD;
+		return NOTIFY_OK;
+	case NETDEV_UNREGISTER:
+		if (netif_is_bond_master(netdev)) {
+			err = nfp_fl_lag_schedule_group_delete(lag, netdev);
+			if (err)
+				return NOTIFY_BAD;
+			return NOTIFY_OK;
+		}
+	}
+
+	return NOTIFY_DONE;
+}
+
+int nfp_flower_lag_reset(struct nfp_fl_lag *lag)
+{
+	enum nfp_fl_lag_batch batch = NFP_FL_LAG_BATCH_FIRST;
+
+	lag->rst_cfg = true;
+	return nfp_fl_lag_config_group(lag, NULL, NULL, 0, &batch);
+}
+
+void nfp_flower_lag_init(struct nfp_fl_lag *lag)
+{
+	INIT_DELAYED_WORK(&lag->work, nfp_fl_lag_do_work);
+	INIT_LIST_HEAD(&lag->group_list);
+	mutex_init(&lag->lock);
+	ida_init(&lag->ida_handle);
+
+	/* 0 is a reserved batch version so increment to first valid value. */
+	nfp_fl_increment_version(lag);
+
+	lag->lag_nb.notifier_call = nfp_fl_lag_netdev_event;
+}
+
+void nfp_flower_lag_cleanup(struct nfp_fl_lag *lag)
+{
+	struct nfp_fl_lag_group *entry, *storage;
+
+	cancel_delayed_work_sync(&lag->work);
+
+	/* Remove all groups. */
+	mutex_lock(&lag->lock);
+	list_for_each_entry_safe(entry, storage, &lag->group_list, list) {
+		list_del(&entry->list);
+		kfree(entry);
+	}
+	mutex_unlock(&lag->lock);
+	mutex_destroy(&lag->lock);
+	ida_destroy(&lag->ida_handle);
+}
diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.c b/drivers/net/ethernet/netronome/nfp/flower/main.c
index 202284b42fd9..19cfa162ac65 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.c
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.c
@@ -575,12 +575,14 @@ static int nfp_flower_init(struct nfp_app *app)
 	/* Tell the firmware that the driver supports lag. */
 	err = nfp_rtsym_write_le(app->pf->rtbl,
 				 "_abi_flower_balance_sync_enable", 1);
-	if (!err)
+	if (!err) {
 		app_priv->flower_ext_feats |= NFP_FL_FEATS_LAG;
-	else if (err == -ENOENT)
+		nfp_flower_lag_init(&app_priv->nfp_lag);
+	} else if (err == -ENOENT) {
 		nfp_warn(app->cpp, "LAG not supported by FW.\n");
-	else
+	} else {
 		goto err_cleanup_metadata;
+	}
 
 	return 0;
 
@@ -599,6 +601,9 @@ static void nfp_flower_clean(struct nfp_app *app)
 	skb_queue_purge(&app_priv->cmsg_skbs_low);
 	flush_work(&app_priv->cmsg_work);
 
+	if (app_priv->flower_ext_feats & NFP_FL_FEATS_LAG)
+		nfp_flower_lag_cleanup(&app_priv->nfp_lag);
+
 	nfp_flower_metadata_cleanup(app);
 	vfree(app->priv);
 	app->priv = NULL;
@@ -665,11 +670,29 @@ nfp_flower_repr_change_mtu(struct nfp_app *app, struct net_device *netdev,
 
 static int nfp_flower_start(struct nfp_app *app)
 {
+	struct nfp_flower_priv *app_priv = app->priv;
+	int err;
+
+	if (app_priv->flower_ext_feats & NFP_FL_FEATS_LAG) {
+		err = nfp_flower_lag_reset(&app_priv->nfp_lag);
+		if (err)
+			return err;
+
+		err = register_netdevice_notifier(&app_priv->nfp_lag.lag_nb);
+		if (err)
+			return err;
+	}
+
 	return nfp_tunnel_config_start(app);
 }
 
 static void nfp_flower_stop(struct nfp_app *app)
 {
+	struct nfp_flower_priv *app_priv = app->priv;
+
+	if (app_priv->flower_ext_feats & NFP_FL_FEATS_LAG)
+		unregister_netdevice_notifier(&app_priv->nfp_lag.lag_nb);
+
 	nfp_tunnel_config_stop(app);
 }
 
diff --git a/drivers/net/ethernet/netronome/nfp/flower/main.h b/drivers/net/ethernet/netronome/nfp/flower/main.h
index 7ce255705446..e03efb034948 100644
--- a/drivers/net/ethernet/netronome/nfp/flower/main.h
+++ b/drivers/net/ethernet/netronome/nfp/flower/main.h
@@ -43,6 +43,7 @@
 #include <net/pkt_cls.h>
 #include <net/tcp.h>
 #include <linux/workqueue.h>
+#include <linux/idr.h>
 
 struct net_device;
 struct nfp_app;
@@ -98,6 +99,30 @@ struct nfp_mtu_conf {
 };
 
 /**
+ * struct nfp_fl_lag - Flower APP priv data for link aggregation
+ * @lag_nb:		Notifier to track master/slave events
+ * @work:		Work queue for writing configs to the HW
+ * @lock:		Lock to protect lag_group_list
+ * @group_list:		List of all master/slave groups offloaded
+ * @ida_handle:		IDA to handle group ids
+ * @pkt_num:		Incremented for each config packet sent
+ * @batch_ver:		Incremented for each batch of config packets
+ * @global_inst:	Instance allocator for groups
+ * @rst_cfg:		Marker to reset HW LAG config
+ */
+struct nfp_fl_lag {
+	struct notifier_block lag_nb;
+	struct delayed_work work;
+	struct mutex lock;
+	struct list_head group_list;
+	struct ida ida_handle;
+	unsigned int pkt_num;
+	unsigned int batch_ver;
+	u8 global_inst;
+	bool rst_cfg;
+};
+
+/**
  * struct nfp_flower_priv - Flower APP per-vNIC priv data
  * @app:		Back pointer to app
  * @nn:			Pointer to vNIC
@@ -129,6 +154,7 @@ struct nfp_mtu_conf {
  *			from firmware for repr reify
  * @reify_wait_queue:	wait queue for repr reify response counting
  * @mtu_conf:		Configuration of repr MTU value
+ * @nfp_lag:		Link aggregation data block
  */
 struct nfp_flower_priv {
 	struct nfp_app *app;
@@ -158,6 +184,7 @@ struct nfp_flower_priv {
 	atomic_t reify_replies;
 	wait_queue_head_t reify_wait_queue;
 	struct nfp_mtu_conf mtu_conf;
+	struct nfp_fl_lag nfp_lag;
 };
 
 /**
@@ -250,5 +277,8 @@ void nfp_tunnel_request_route(struct nfp_app *app, struct sk_buff *skb);
 void nfp_tunnel_keep_alive(struct nfp_app *app, struct sk_buff *skb);
 int nfp_flower_setup_tc_egress_cb(enum tc_setup_type type, void *type_data,
 				  void *cb_priv);
+void nfp_flower_lag_init(struct nfp_fl_lag *lag);
+void nfp_flower_lag_cleanup(struct nfp_fl_lag *lag);
+int nfp_flower_lag_reset(struct nfp_fl_lag *lag);
 
 #endif