Merge branch 'net-marvell-prestera-add-nexthop-routes-offloading'

Yevhen Orlov says:

====================
net: marvell: prestera: add nexthop routes offloading

Add support for nexthop routes for Marvell Prestera driver.
Subscribe on NEIGH_UPDATE events.

Add features:
 - Support connected route adding
   e.g.: "ip address add 1.1.1.1/24 dev sw1p1"
   e.g.: "ip route add 6.6.6/24 dev sw1p1"
 - Support nexthop route adding
   e.g.: "ip route add 5.5.5/24 via 1.1.1.2"
 - Support ECMP route adding
   e.g.: "ip route add 5.5.5/24 nexthop via 1.1.1.2 nexthop via 1.1.1.3"
 - Support "offload" and "trap" flags per each nexthop
 - Support "offload" flag for neighbours

Limitations:
 - Only "local" and "main" tables supported
 - Only generic interfaces supported for router (no bridges or vlans)

Flags meaning:
  ip route add 5.5.5/24 nexthop via 2.2.2.2 nexthop via 2.2.2.3
  ip route show
  ...
  5.5.5.0/24 rt_offload
        nexthop via 2.2.2.2 dev sw1p31 weight 1 trap
        nexthop via 2.2.2.3 dev sw1p31 weight 1 trap
  ...
  # When you just add route - lpm entry became occupied
  # in HW ("rt_offload" flag), but related to nexthops neighbours
  # still not resolved ("trap" flag).
  #
  # After some time...
  ip route show
  ...
  5.5.5.0/24 rt_offload
        nexthop via 2.2.2.2 dev sw1p31 weight 1 offload
        nexthop via 2.2.2.3 dev sw1p31 weight 1 offload
  ...
  # You will see, that appropriate neighbours was resolved and nexthop
  # entries occupied in HW too ("offload" flag)
Co-developed-by: Taras Chornyi <tchornyi@marvell.com>
Signed-off-by: Taras Chornyi <tchornyi@marvell.com>
Co-developed-by: Oleksandr Mazur <oleksandr.mazur@plvision.eu>
Signed-off-by: Oleksandr Mazur <oleksandr.mazur@plvision.eu>
Signed-off-by: Yevhen Orlov <yevhen.orlov@plvision.eu>

Changes for v2:
* Add more reviewers in CC
* Check if route nexthop or direct with fib_nh_gw_family instead of fib_nh_scope
  This is needed after,
  747c1430 ("ip: fix dflt addr selection for connected nexthop"),
  because direct route is now with the same scope as nexthop (RT_SCOPE_LINK)

Changes for v3:
* Resolve "unused functions" warnings, after
  patch ("net: marvell: prestera: Add heplers to interact ... "), and before
  patch ("net: marvell: prestera: Add neighbour cache accounting")

Changes for v4:
* Rebase to the latest master to resolve patch applying issues

Changes for v5:
* Repack structures to prevent holes
* Remove unused variables
* Fix misspeling issues

Changes for v6:
* Rebase on top of master
* Fix smatch warnings

Changes for v7:
* Rebase on top of master
* Refactor: use "fib_lookup" instead of "fib_new_table"+"fib_table_lookup",
  according to Paolo Abeni suggestion
* Refactor: use "rhashtable_free_and_destroy" instead of rhashtable
  walk, according to Paolo Abeni suggestion
====================

Link: https://lore.kernel.org/r/20221001093417.22388-1-yevhen.orlov@plvision.euSigned-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/marvell/prestera/prestera.h

@@ -306,17 +306,27 @@ struct prestera_switch {
 	struct prestera_counter *counter;
 	u8 lag_member_max;
 	u8 lag_max;
+	u32 size_tbl_router_nexthop;
 };
 
 struct prestera_router {
 	struct prestera_switch *sw;
 	struct list_head vr_list;
 	struct list_head rif_entry_list;
+	struct rhashtable nh_neigh_ht;
+	struct rhashtable nexthop_group_ht;
 	struct rhashtable fib_ht;
+	struct rhashtable kern_neigh_cache_ht;
 	struct rhashtable kern_fib_cache_ht;
 	struct notifier_block inetaddr_nb;
 	struct notifier_block inetaddr_valid_nb;
 	struct notifier_block fib_nb;
+	struct notifier_block netevent_nb;
+	u8 *nhgrp_hw_state_cache; /* Bitmap cached hw state of nhs */
+	unsigned long nhgrp_hw_cache_kick; /* jiffies */
+	struct {
+		struct delayed_work dw;
+	} neighs_update;
 };
 
 struct prestera_rxtx_params {
@@ -362,6 +372,8 @@ int prestera_port_cfg_mac_write(struct prestera_port *port,
 struct prestera_port *prestera_port_dev_lower_find(struct net_device *dev);
 
 void prestera_queue_work(struct work_struct *work);
+void prestera_queue_delayed_work(struct delayed_work *work, unsigned long delay);
+void prestera_queue_drain(void);
 
 int prestera_port_learning_set(struct prestera_port *port, bool learn_enable);
 int prestera_port_uc_flood_set(struct prestera_port *port, bool flood);

drivers/net/ethernet/marvell/prestera/prestera_hw.c

@@ -10,11 +10,14 @@
 #include "prestera_hw.h"
 #include "prestera_acl.h"
 #include "prestera_counter.h"
+#include "prestera_router_hw.h"
 
 #define PRESTERA_SWITCH_INIT_TIMEOUT_MS (30 * 1000)
 
 #define PRESTERA_MIN_MTU 64
 
+#define PRESTERA_MSG_CHUNK_SIZE 1024
+
 enum prestera_cmd_type_t {
 	PRESTERA_CMD_TYPE_SWITCH_INIT = 0x1,
 	PRESTERA_CMD_TYPE_SWITCH_ATTR_SET = 0x2,
@@ -57,6 +60,10 @@ enum prestera_cmd_type_t {
 	PRESTERA_CMD_TYPE_ROUTER_RIF_DELETE = 0x601,
 	PRESTERA_CMD_TYPE_ROUTER_LPM_ADD = 0x610,
 	PRESTERA_CMD_TYPE_ROUTER_LPM_DELETE = 0x611,
+	PRESTERA_CMD_TYPE_ROUTER_NH_GRP_SET = 0x622,
+	PRESTERA_CMD_TYPE_ROUTER_NH_GRP_BLK_GET = 0x645,
+	PRESTERA_CMD_TYPE_ROUTER_NH_GRP_ADD = 0x623,
+	PRESTERA_CMD_TYPE_ROUTER_NH_GRP_DELETE = 0x624,
 	PRESTERA_CMD_TYPE_ROUTER_VR_CREATE = 0x630,
 	PRESTERA_CMD_TYPE_ROUTER_VR_DELETE = 0x631,
 
@@ -542,6 +549,14 @@ struct prestera_msg_ip_addr {
 	u8 __pad[3];
 };
 
+struct prestera_msg_nh {
+	struct prestera_msg_iface oif;
+	__le32 hw_id;
+	u8 mac[ETH_ALEN];
+	u8 is_active;
+	u8 pad;
+};
+
 struct prestera_msg_rif_req {
 	struct prestera_msg_cmd cmd;
 	struct prestera_msg_iface iif;
@@ -567,6 +582,34 @@ struct prestera_msg_lpm_req {
 	u8 __pad[2];
 };
 
+struct prestera_msg_nh_req {
+	struct prestera_msg_cmd cmd;
+	struct prestera_msg_nh nh[PRESTERA_NHGR_SIZE_MAX];
+	__le32 size;
+	__le32 grp_id;
+};
+
+struct prestera_msg_nh_chunk_req {
+	struct prestera_msg_cmd cmd;
+	__le32 offset;
+};
+
+struct prestera_msg_nh_chunk_resp {
+	struct prestera_msg_ret ret;
+	u8 hw_state[PRESTERA_MSG_CHUNK_SIZE];
+};
+
+struct prestera_msg_nh_grp_req {
+	struct prestera_msg_cmd cmd;
+	__le32 grp_id;
+	__le32 size;
+};
+
+struct prestera_msg_nh_grp_resp {
+	struct prestera_msg_ret ret;
+	__le32 grp_id;
+};
+
 struct prestera_msg_vr_req {
 	struct prestera_msg_cmd cmd;
 	__le16 vr_id;
@@ -729,11 +772,15 @@ static void prestera_hw_build_tests(void)
 	BUILD_BUG_ON(sizeof(struct prestera_msg_flood_domain_ports_reset_req) != 8);
 	BUILD_BUG_ON(sizeof(struct prestera_msg_mdb_create_req) != 16);
 	BUILD_BUG_ON(sizeof(struct prestera_msg_mdb_destroy_req) != 16);
+	BUILD_BUG_ON(sizeof(struct prestera_msg_nh_req) != 124);
+	BUILD_BUG_ON(sizeof(struct prestera_msg_nh_chunk_req) != 8);
+	BUILD_BUG_ON(sizeof(struct prestera_msg_nh_grp_req) != 12);
 
 	/*  structure that are part of req/resp fw messages */
 	BUILD_BUG_ON(sizeof(struct prestera_msg_iface) != 16);
 	BUILD_BUG_ON(sizeof(struct prestera_msg_ip_addr) != 20);
 	BUILD_BUG_ON(sizeof(struct prestera_msg_flood_domain_port) != 12);
+	BUILD_BUG_ON(sizeof(struct prestera_msg_nh) != 28);
 
 	/* check responses */
 	BUILD_BUG_ON(sizeof(struct prestera_msg_common_resp) != 8);
@@ -750,6 +797,8 @@ static void prestera_hw_build_tests(void)
 	BUILD_BUG_ON(sizeof(struct prestera_msg_vr_resp) != 12);
 	BUILD_BUG_ON(sizeof(struct prestera_msg_policer_resp) != 12);
 	BUILD_BUG_ON(sizeof(struct prestera_msg_flood_domain_create_resp) != 12);
+	BUILD_BUG_ON(sizeof(struct prestera_msg_nh_chunk_resp) != 1032);
+	BUILD_BUG_ON(sizeof(struct prestera_msg_nh_grp_resp) != 12);
 
 	/* check events */
 	BUILD_BUG_ON(sizeof(struct prestera_msg_event_port) != 20);
@@ -1027,6 +1076,8 @@ int prestera_hw_switch_init(struct prestera_switch *sw)
 	sw->id = resp.switch_id;
 	sw->lag_member_max = resp.lag_member_max;
 	sw->lag_max = resp.lag_max;
+	sw->size_tbl_router_nexthop =
+		__le32_to_cpu(resp.size_tbl_router_nexthop);
 
 	return 0;
 }
@@ -2037,6 +2088,85 @@ int prestera_hw_lpm_del(struct prestera_switch *sw, u16 vr_id,
 			    sizeof(req));
 }
 
+int prestera_hw_nh_entries_set(struct prestera_switch *sw, int count,
+			       struct prestera_neigh_info *nhs, u32 grp_id)
+{
+	struct prestera_msg_nh_req req = { .size = __cpu_to_le32((u32)count),
+			.grp_id = __cpu_to_le32(grp_id) };
+	int i, err;
+
+	for (i = 0; i < count; i++) {
+		req.nh[i].is_active = nhs[i].connected;
+		memcpy(&req.nh[i].mac, nhs[i].ha, ETH_ALEN);
+		err = prestera_iface_to_msg(&nhs[i].iface, &req.nh[i].oif);
+		if (err)
+			return err;
+	}
+
+	return prestera_cmd(sw, PRESTERA_CMD_TYPE_ROUTER_NH_GRP_SET, &req.cmd,
+			    sizeof(req));
+}
+
+int prestera_hw_nhgrp_blk_get(struct prestera_switch *sw,
+			      u8 *hw_state, u32 buf_size /* Buffer in bytes */)
+{
+	static struct prestera_msg_nh_chunk_resp resp;
+	struct prestera_msg_nh_chunk_req req;
+	u32 buf_offset;
+	int err;
+
+	memset(&hw_state[0], 0, buf_size);
+	buf_offset = 0;
+	while (1) {
+		if (buf_offset >= buf_size)
+			break;
+
+		memset(&req, 0, sizeof(req));
+		req.offset = __cpu_to_le32(buf_offset * 8); /* 8 bits in u8 */
+		err = prestera_cmd_ret(sw,
+				       PRESTERA_CMD_TYPE_ROUTER_NH_GRP_BLK_GET,
+				       &req.cmd, sizeof(req), &resp.ret,
+				       sizeof(resp));
+		if (err)
+			return err;
+
+		memcpy(&hw_state[buf_offset], &resp.hw_state[0],
+		       buf_offset + PRESTERA_MSG_CHUNK_SIZE > buf_size ?
+			buf_size - buf_offset : PRESTERA_MSG_CHUNK_SIZE);
+		buf_offset += PRESTERA_MSG_CHUNK_SIZE;
+	}
+
+	return 0;
+}
+
+int prestera_hw_nh_group_create(struct prestera_switch *sw, u16 nh_count,
+				u32 *grp_id)
+{
+	struct prestera_msg_nh_grp_req req = { .size = __cpu_to_le32((u32)nh_count) };
+	struct prestera_msg_nh_grp_resp resp;
+	int err;
+
+	err = prestera_cmd_ret(sw, PRESTERA_CMD_TYPE_ROUTER_NH_GRP_ADD,
+			       &req.cmd, sizeof(req), &resp.ret, sizeof(resp));
+	if (err)
+		return err;
+
+	*grp_id = __le32_to_cpu(resp.grp_id);
+	return err;
+}
+
+int prestera_hw_nh_group_delete(struct prestera_switch *sw, u16 nh_count,
+				u32 grp_id)
+{
+	struct prestera_msg_nh_grp_req req = {
+	    .grp_id = __cpu_to_le32(grp_id),
+	    .size = __cpu_to_le32(nh_count)
+	};
+
+	return prestera_cmd(sw, PRESTERA_CMD_TYPE_ROUTER_NH_GRP_DELETE,
+			    &req.cmd, sizeof(req));
+}
+
 int prestera_hw_rxtx_init(struct prestera_switch *sw,
 			  struct prestera_rxtx_params *params)
 {

drivers/net/ethernet/marvell/prestera/prestera_hw.h

@@ -146,6 +146,7 @@ struct prestera_counter_stats;
 struct prestera_iface;
 struct prestera_flood_domain;
 struct prestera_mdb_entry;
+struct prestera_neigh_info;
 
 /* Switch API */
 int prestera_hw_switch_init(struct prestera_switch *sw);
@@ -266,6 +267,16 @@ int prestera_hw_lpm_add(struct prestera_switch *sw, u16 vr_id,
 int prestera_hw_lpm_del(struct prestera_switch *sw, u16 vr_id,
 			__be32 dst, u32 dst_len);
 
+/* NH API */
+int prestera_hw_nh_entries_set(struct prestera_switch *sw, int count,
+			       struct prestera_neigh_info *nhs, u32 grp_id);
+int prestera_hw_nhgrp_blk_get(struct prestera_switch *sw,
+			      u8 *hw_state, u32 buf_size /* Buffer in bytes */);
+int prestera_hw_nh_group_create(struct prestera_switch *sw, u16 nh_count,
+				u32 *grp_id);
+int prestera_hw_nh_group_delete(struct prestera_switch *sw, u16 nh_count,
+				u32 grp_id);
+
 /* Event handlers */
 int prestera_hw_event_handler_register(struct prestera_switch *sw,
 				       enum prestera_event_type type,

drivers/net/ethernet/marvell/prestera/prestera_main.c

@@ -36,6 +36,17 @@ void prestera_queue_work(struct work_struct *work)
 	queue_work(prestera_owq, work);
 }
 
+void prestera_queue_delayed_work(struct delayed_work *work, unsigned long delay)
+{
+	queue_delayed_work(prestera_wq, work, delay);
+}
+
+void prestera_queue_drain(void)
+{
+	drain_workqueue(prestera_wq);
+	drain_workqueue(prestera_owq);
+}
+
 int prestera_port_learning_set(struct prestera_port *port, bool learn)
 {
 	return prestera_hw_port_learning_set(port, learn);

drivers/net/ethernet/marvell/prestera/prestera_router.c

@@ -7,10 +7,35 @@
 #include <net/inet_dscp.h>
 #include <net/switchdev.h>
 #include <linux/rhashtable.h>
+#include <net/nexthop.h>
+#include <net/arp.h>
+#include <linux/if_vlan.h>
+#include <linux/if_macvlan.h>
+#include <net/netevent.h>
 
 #include "prestera.h"
 #include "prestera_router_hw.h"
 
+#define PRESTERA_IMPLICITY_RESOLVE_DEAD_NEIGH
+#define PRESTERA_NH_PROBE_INTERVAL 5000 /* ms */
+
+struct prestera_kern_neigh_cache_key {
+	struct prestera_ip_addr addr;
+	struct net_device *dev;
+};
+
+struct prestera_kern_neigh_cache {
+	struct prestera_kern_neigh_cache_key key;
+	struct rhash_head ht_node;
+	struct list_head kern_fib_cache_list;
+	/* Hold prepared nh_neigh info if is in_kernel */
+	struct prestera_neigh_info nh_neigh_info;
+	/* Indicate if neighbour is reachable by direct route */
+	bool reachable;
+	/* Lock cache if neigh is present in kernel */
+	bool in_kernel;
+};
+
 struct prestera_kern_fib_cache_key {
 	struct prestera_ip_addr addr;
 	u32 prefix_len;
@@ -23,15 +48,29 @@ struct prestera_kern_fib_cache {
 	struct {
 		struct prestera_fib_key fib_key;
 		enum prestera_fib_type fib_type;
+		struct prestera_nexthop_group_key nh_grp_key;
 	} lpm_info; /* hold prepared lpm info */
 	/* Indicate if route is not overlapped by another table */
 	struct rhash_head ht_node; /* node of prestera_router */
-	struct fib_info *fi;
-	dscp_t kern_dscp;
-	u8 kern_type;
+	struct prestera_kern_neigh_cache_head {
+		struct prestera_kern_fib_cache *this;
+		struct list_head head;
+		struct prestera_kern_neigh_cache *n_cache;
+	} kern_neigh_cache_head[PRESTERA_NHGR_SIZE_MAX];
+	union {
+		struct fib_notifier_info info; /* point to any of 4/6 */
+		struct fib_entry_notifier_info fen4_info;
+	};
 	bool reachable;
 };
 
+static const struct rhashtable_params __prestera_kern_neigh_cache_ht_params = {
+	.key_offset  = offsetof(struct prestera_kern_neigh_cache, key),
+	.head_offset = offsetof(struct prestera_kern_neigh_cache, ht_node),
+	.key_len     = sizeof(struct prestera_kern_neigh_cache_key),
+	.automatic_shrinking = true,
+};
+
 static const struct rhashtable_params __prestera_kern_fib_cache_ht_params = {
 	.key_offset  = offsetof(struct prestera_kern_fib_cache, key),
 	.head_offset = offsetof(struct prestera_kern_fib_cache, ht_node),
@@ -51,15 +90,450 @@ static u32 prestera_fix_tb_id(u32 tb_id)
 }
 
 static void
-prestera_util_fen_info2fib_cache_key(struct fib_entry_notifier_info *fen_info,
+prestera_util_fen_info2fib_cache_key(struct fib_notifier_info *info,
 				     struct prestera_kern_fib_cache_key *key)
 {
+	struct fib_entry_notifier_info *fen_info =
+		container_of(info, struct fib_entry_notifier_info, info);
+
 	memset(key, 0, sizeof(*key));
+	key->addr.v = PRESTERA_IPV4;
 	key->addr.u.ipv4 = cpu_to_be32(fen_info->dst);
 	key->prefix_len = fen_info->dst_len;
 	key->kern_tb_id = fen_info->tb_id;
 }
 
+static int prestera_util_nhc2nc_key(struct prestera_switch *sw,
+				    struct fib_nh_common *nhc,
+				    struct prestera_kern_neigh_cache_key *nk)
+{
+	memset(nk, 0, sizeof(*nk));
+	if (nhc->nhc_gw_family == AF_INET) {
+		nk->addr.v = PRESTERA_IPV4;
+		nk->addr.u.ipv4 = nhc->nhc_gw.ipv4;
+	} else {
+		nk->addr.v = PRESTERA_IPV6;
+		nk->addr.u.ipv6 = nhc->nhc_gw.ipv6;
+	}
+
+	nk->dev = nhc->nhc_dev;
+	return 0;
+}
+
+static void
+prestera_util_nc_key2nh_key(struct prestera_kern_neigh_cache_key *ck,
+			    struct prestera_nh_neigh_key *nk)
+{
+	memset(nk, 0, sizeof(*nk));
+	nk->addr = ck->addr;
+	nk->rif = (void *)ck->dev;
+}
+
+static bool
+prestera_util_nhc_eq_n_cache_key(struct prestera_switch *sw,
+				 struct fib_nh_common *nhc,
+				 struct prestera_kern_neigh_cache_key *nk)
+{
+	struct prestera_kern_neigh_cache_key tk;
+	int err;
+
+	err = prestera_util_nhc2nc_key(sw, nhc, &tk);
+	if (err)
+		return false;
+
+	if (memcmp(&tk, nk, sizeof(tk)))
+		return false;
+
+	return true;
+}
+
+static int
+prestera_util_neigh2nc_key(struct prestera_switch *sw, struct neighbour *n,
+			   struct prestera_kern_neigh_cache_key *key)
+{
+	memset(key, 0, sizeof(*key));
+	if (n->tbl->family == AF_INET) {
+		key->addr.v = PRESTERA_IPV4;
+		key->addr.u.ipv4 = *(__be32 *)n->primary_key;
+	} else {
+		return -ENOENT;
+	}
+
+	key->dev = n->dev;
+
+	return 0;
+}
+
+static bool __prestera_fi_is_direct(struct fib_info *fi)
+{
+	struct fib_nh *fib_nh;
+
+	if (fib_info_num_path(fi) == 1) {
+		fib_nh = fib_info_nh(fi, 0);
+		if (fib_nh->fib_nh_gw_family == AF_UNSPEC)
+			return true;
+	}
+
+	return false;
+}
+
+static bool prestera_fi_is_direct(struct fib_info *fi)
+{
+	if (fi->fib_type != RTN_UNICAST)
+		return false;
+
+	return __prestera_fi_is_direct(fi);
+}
+
+static bool prestera_fi_is_nh(struct fib_info *fi)
+{
+	if (fi->fib_type != RTN_UNICAST)
+		return false;
+
+	return !__prestera_fi_is_direct(fi);
+}
+
+static bool __prestera_fi6_is_direct(struct fib6_info *fi)
+{
+	if (!fi->fib6_nh->nh_common.nhc_gw_family)
+		return true;
+
+	return false;
+}
+
+static bool prestera_fi6_is_direct(struct fib6_info *fi)
+{
+	if (fi->fib6_type != RTN_UNICAST)
+		return false;
+
+	return __prestera_fi6_is_direct(fi);
+}
+
+static bool prestera_fi6_is_nh(struct fib6_info *fi)
+{
+	if (fi->fib6_type != RTN_UNICAST)
+		return false;
+
+	return !__prestera_fi6_is_direct(fi);
+}
+
+static bool prestera_fib_info_is_direct(struct fib_notifier_info *info)
+{
+	struct fib6_entry_notifier_info *fen6_info =
+		container_of(info, struct fib6_entry_notifier_info, info);
+	struct fib_entry_notifier_info *fen_info =
+		container_of(info, struct fib_entry_notifier_info, info);
+
+	if (info->family == AF_INET)
+		return prestera_fi_is_direct(fen_info->fi);
+	else
+		return prestera_fi6_is_direct(fen6_info->rt);
+}
+
+static bool prestera_fib_info_is_nh(struct fib_notifier_info *info)
+{
+	struct fib6_entry_notifier_info *fen6_info =
+		container_of(info, struct fib6_entry_notifier_info, info);
+	struct fib_entry_notifier_info *fen_info =
+		container_of(info, struct fib_entry_notifier_info, info);
+
+	if (info->family == AF_INET)
+		return prestera_fi_is_nh(fen_info->fi);
+	else
+		return prestera_fi6_is_nh(fen6_info->rt);
+}
+
+/* must be called with rcu_read_lock() */
+static int prestera_util_kern_get_route(struct fib_result *res, u32 tb_id,
+					__be32 *addr)
+{
+	struct flowi4 fl4;
+
+	/* TODO: walkthrough appropriate tables in kernel
+	 * to know if the same prefix exists in several tables
+	 */
+	memset(&fl4, 0, sizeof(fl4));
+	fl4.daddr = *addr;
+	return fib_lookup(&init_net, &fl4, res, 0 /* FIB_LOOKUP_NOREF */);
+}
+
+static bool
+__prestera_util_kern_n_is_reachable_v4(u32 tb_id, __be32 *addr,
+				       struct net_device *dev)
+{
+	struct fib_nh *fib_nh;
+	struct fib_result res;
+	bool reachable;
+
+	reachable = false;
+
+	if (!prestera_util_kern_get_route(&res, tb_id, addr))
+		if (prestera_fi_is_direct(res.fi)) {
+			fib_nh = fib_info_nh(res.fi, 0);
+			if (dev == fib_nh->fib_nh_dev)
+				reachable = true;
+		}
+
+	return reachable;
+}
+
+/* Check if neigh route is reachable */
+static bool
+prestera_util_kern_n_is_reachable(u32 tb_id,
+				  struct prestera_ip_addr *addr,
+				  struct net_device *dev)
+{
+	if (addr->v == PRESTERA_IPV4)
+		return __prestera_util_kern_n_is_reachable_v4(tb_id,
+							      &addr->u.ipv4,
+							      dev);
+	else
+		return false;
+}
+
+static void prestera_util_kern_set_neigh_offload(struct neighbour *n,
+						 bool offloaded)
+{
+	if (offloaded)
+		n->flags |= NTF_OFFLOADED;
+	else
+		n->flags &= ~NTF_OFFLOADED;
+}
+
+static void
+prestera_util_kern_set_nh_offload(struct fib_nh_common *nhc, bool offloaded, bool trap)
+{
+		if (offloaded)
+			nhc->nhc_flags |= RTNH_F_OFFLOAD;
+		else
+			nhc->nhc_flags &= ~RTNH_F_OFFLOAD;
+
+		if (trap)
+			nhc->nhc_flags |= RTNH_F_TRAP;
+		else
+			nhc->nhc_flags &= ~RTNH_F_TRAP;
+}
+
+static struct fib_nh_common *
+prestera_kern_fib_info_nhc(struct fib_notifier_info *info, int n)
+{
+	struct fib6_entry_notifier_info *fen6_info;
+	struct fib_entry_notifier_info *fen4_info;
+	struct fib6_info *iter;
+
+	if (info->family == AF_INET) {
+		fen4_info = container_of(info, struct fib_entry_notifier_info,
+					 info);
+		return &fib_info_nh(fen4_info->fi, n)->nh_common;
+	} else if (info->family == AF_INET6) {
+		fen6_info = container_of(info, struct fib6_entry_notifier_info,
+					 info);
+		if (!n)
+			return &fen6_info->rt->fib6_nh->nh_common;
+
+		list_for_each_entry(iter, &fen6_info->rt->fib6_siblings,
+				    fib6_siblings) {
+			if (!--n)
+				return &iter->fib6_nh->nh_common;
+		}
+	}
+
+	/* if family is incorrect - than upper functions has BUG */
+	/* if doesn't find requested index - there is alsi bug, because
+	 * valid index must be produced by nhs, which checks list length
+	 */
+	WARN(1, "Invalid parameters passed to %s n=%d i=%p",
+	     __func__, n, info);
+	return NULL;
+}
+
+static int prestera_kern_fib_info_nhs(struct fib_notifier_info *info)
+{
+	struct fib6_entry_notifier_info *fen6_info;
+	struct fib_entry_notifier_info *fen4_info;
+
+	if (info->family == AF_INET) {
+		fen4_info = container_of(info, struct fib_entry_notifier_info,
+					 info);
+		return fib_info_num_path(fen4_info->fi);
+	} else if (info->family == AF_INET6) {
+		fen6_info = container_of(info, struct fib6_entry_notifier_info,
+					 info);
+		return fen6_info->rt->fib6_nsiblings + 1;
+	}
+
+	return 0;
+}
+
+static unsigned char
+prestera_kern_fib_info_type(struct fib_notifier_info *info)
+{
+	struct fib6_entry_notifier_info *fen6_info;
+	struct fib_entry_notifier_info *fen4_info;
+
+	if (info->family == AF_INET) {
+		fen4_info = container_of(info, struct fib_entry_notifier_info,
+					 info);
+		return fen4_info->fi->fib_type;
+	} else if (info->family == AF_INET6) {
+		fen6_info = container_of(info, struct fib6_entry_notifier_info,
+					 info);
+		/* TODO: ECMP in ipv6 is several routes.
+		 * Every route has single nh.
+		 */
+		return fen6_info->rt->fib6_type;
+	}
+
+	return RTN_UNSPEC;
+}
+
+/* Decided, that uc_nh route with key==nh is obviously neighbour route */
+static bool
+prestera_fib_node_util_is_neighbour(struct prestera_fib_node *fib_node)
+{
+	if (fib_node->info.type != PRESTERA_FIB_TYPE_UC_NH)
+		return false;
+
+	if (fib_node->info.nh_grp->nh_neigh_head[1].neigh)
+		return false;
+
+	if (!fib_node->info.nh_grp->nh_neigh_head[0].neigh)
+		return false;
+
+	if (memcmp(&fib_node->info.nh_grp->nh_neigh_head[0].neigh->key.addr,
+		   &fib_node->key.addr, sizeof(struct prestera_ip_addr)))
+		return false;
+
+	return true;
+}
+
+static int prestera_dev_if_type(const struct net_device *dev)
+{
+	struct macvlan_dev *vlan;
+
+	if (is_vlan_dev(dev) &&
+	    netif_is_bridge_master(vlan_dev_real_dev(dev))) {
+		return PRESTERA_IF_VID_E;
+	} else if (netif_is_bridge_master(dev)) {
+		return PRESTERA_IF_VID_E;
+	} else if (netif_is_lag_master(dev)) {
+		return PRESTERA_IF_LAG_E;
+	} else if (netif_is_macvlan(dev)) {
+		vlan = netdev_priv(dev);
+		return prestera_dev_if_type(vlan->lowerdev);
+	} else {
+		return PRESTERA_IF_PORT_E;
+	}
+}
+
+static int
+prestera_neigh_iface_init(struct prestera_switch *sw,
+			  struct prestera_iface *iface,
+			  struct neighbour *n)
+{
+	struct prestera_port *port;
+
+	iface->vlan_id = 0; /* TODO: vlan egress */
+	iface->type = prestera_dev_if_type(n->dev);
+	if (iface->type != PRESTERA_IF_PORT_E)
+		return -EINVAL;
+
+	if (!prestera_netdev_check(n->dev))
+		return -EINVAL;
+
+	port = netdev_priv(n->dev);
+	iface->dev_port.hw_dev_num = port->dev_id;
+	iface->dev_port.port_num = port->hw_id;
+
+	return 0;
+}
+
+static struct prestera_kern_neigh_cache *
+prestera_kern_neigh_cache_find(struct prestera_switch *sw,
+			       struct prestera_kern_neigh_cache_key *key)
+{
+	struct prestera_kern_neigh_cache *n_cache;
+
+	n_cache =
+	 rhashtable_lookup_fast(&sw->router->kern_neigh_cache_ht, key,
+				__prestera_kern_neigh_cache_ht_params);
+	return IS_ERR(n_cache) ? NULL : n_cache;
+}
+
+static void
+__prestera_kern_neigh_cache_destruct(struct prestera_switch *sw,
+				     struct prestera_kern_neigh_cache *n_cache)
+{
+	dev_put(n_cache->key.dev);
+}
+
+static void
+__prestera_kern_neigh_cache_destroy(struct prestera_switch *sw,
+				    struct prestera_kern_neigh_cache *n_cache)
+{
+	rhashtable_remove_fast(&sw->router->kern_neigh_cache_ht,
+			       &n_cache->ht_node,
+			       __prestera_kern_neigh_cache_ht_params);
+	__prestera_kern_neigh_cache_destruct(sw, n_cache);
+	kfree(n_cache);
+}
+
+static struct prestera_kern_neigh_cache *
+__prestera_kern_neigh_cache_create(struct prestera_switch *sw,
+				   struct prestera_kern_neigh_cache_key *key)
+{
+	struct prestera_kern_neigh_cache *n_cache;
+	int err;
+
+	n_cache = kzalloc(sizeof(*n_cache), GFP_KERNEL);
+	if (!n_cache)
+		goto err_kzalloc;
+
+	memcpy(&n_cache->key, key, sizeof(*key));
+	dev_hold(n_cache->key.dev);
+
+	INIT_LIST_HEAD(&n_cache->kern_fib_cache_list);
+	err = rhashtable_insert_fast(&sw->router->kern_neigh_cache_ht,
+				     &n_cache->ht_node,
+				     __prestera_kern_neigh_cache_ht_params);
+	if (err)
+		goto err_ht_insert;
+
+	return n_cache;
+
+err_ht_insert:
+	dev_put(n_cache->key.dev);
+	kfree(n_cache);
+err_kzalloc:
+	return NULL;
+}
+
+static struct prestera_kern_neigh_cache *
+prestera_kern_neigh_cache_get(struct prestera_switch *sw,
+			      struct prestera_kern_neigh_cache_key *key)
+{
+	struct prestera_kern_neigh_cache *n_cache;
+
+	n_cache = prestera_kern_neigh_cache_find(sw, key);
+	if (!n_cache)
+		n_cache = __prestera_kern_neigh_cache_create(sw, key);
+
+	return n_cache;
+}
+
+static struct prestera_kern_neigh_cache *
+prestera_kern_neigh_cache_put(struct prestera_switch *sw,
+			      struct prestera_kern_neigh_cache *n_cache)
+{
+	if (!n_cache->in_kernel &&
+	    list_empty(&n_cache->kern_fib_cache_list)) {
+		__prestera_kern_neigh_cache_destroy(sw, n_cache);
+		return NULL;
+	}
+
+	return n_cache;
+}
+
 static struct prestera_kern_fib_cache *
 prestera_kern_fib_cache_find(struct prestera_switch *sw,
 			     struct prestera_kern_fib_cache_key *key)
@@ -72,25 +546,80 @@ prestera_kern_fib_cache_find(struct prestera_switch *sw,
 	return fib_cache;
 }
 
+static void
+__prestera_kern_fib_cache_destruct(struct prestera_switch *sw,
+				   struct prestera_kern_fib_cache *fib_cache)
+{
+	struct prestera_kern_neigh_cache *n_cache;
+	int i;
+
+	for (i = 0; i < PRESTERA_NHGR_SIZE_MAX; i++) {
+		n_cache = fib_cache->kern_neigh_cache_head[i].n_cache;
+		if (n_cache) {
+			list_del(&fib_cache->kern_neigh_cache_head[i].head);
+			prestera_kern_neigh_cache_put(sw, n_cache);
+		}
+	}
+
+	fib_info_put(fib_cache->fen4_info.fi);
+}
+
 static void
 prestera_kern_fib_cache_destroy(struct prestera_switch *sw,
 				struct prestera_kern_fib_cache *fib_cache)
 {
-	fib_info_put(fib_cache->fi);
 	rhashtable_remove_fast(&sw->router->kern_fib_cache_ht,
 			       &fib_cache->ht_node,
 			       __prestera_kern_fib_cache_ht_params);
+	__prestera_kern_fib_cache_destruct(sw, fib_cache);
 	kfree(fib_cache);
 }
 
+static int
+__prestera_kern_fib_cache_create_nhs(struct prestera_switch *sw,
+				     struct prestera_kern_fib_cache *fc)
+{
+	struct prestera_kern_neigh_cache_key nc_key;
+	struct prestera_kern_neigh_cache *n_cache;
+	struct fib_nh_common *nhc;
+	int i, nhs, err;
+
+	if (!prestera_fib_info_is_nh(&fc->info))
+		return 0;
+
+	nhs = prestera_kern_fib_info_nhs(&fc->info);
+	if (nhs > PRESTERA_NHGR_SIZE_MAX)
+		return 0;
+
+	for (i = 0; i < nhs; i++) {
+		nhc = prestera_kern_fib_info_nhc(&fc->fen4_info.info, i);
+		err = prestera_util_nhc2nc_key(sw, nhc, &nc_key);
+		if (err)
+			return 0;
+
+		n_cache = prestera_kern_neigh_cache_get(sw, &nc_key);
+		if (!n_cache)
+			return 0;
+
+		fc->kern_neigh_cache_head[i].this = fc;
+		fc->kern_neigh_cache_head[i].n_cache = n_cache;
+		list_add(&fc->kern_neigh_cache_head[i].head,
+			 &n_cache->kern_fib_cache_list);
+	}
+
+	return 0;
+}
+
 /* Operations on fi (offload, etc) must be wrapped in utils.
  * This function just create storage.
  */
 static struct prestera_kern_fib_cache *
 prestera_kern_fib_cache_create(struct prestera_switch *sw,
 			       struct prestera_kern_fib_cache_key *key,
-			       struct fib_info *fi, dscp_t dscp, u8 type)
+			       struct fib_notifier_info *info)
 {
+	struct fib_entry_notifier_info *fen_info =
+		container_of(info, struct fib_entry_notifier_info, info);
 	struct prestera_kern_fib_cache *fib_cache;
 	int err;
 
@@ -99,10 +628,8 @@ prestera_kern_fib_cache_create(struct prestera_switch *sw,
 		goto err_kzalloc;
 
 	memcpy(&fib_cache->key, key, sizeof(*key));
-	fib_info_hold(fi);
-	fib_cache->fi = fi;
-	fib_cache->kern_dscp = dscp;
-	fib_cache->kern_type = type;
+	fib_info_hold(fen_info->fi);
+	memcpy(&fib_cache->fen4_info, fen_info, sizeof(*fen_info));
 
 	err = rhashtable_insert_fast(&sw->router->kern_fib_cache_ht,
 				     &fib_cache->ht_node,
@@ -110,15 +637,61 @@ prestera_kern_fib_cache_create(struct prestera_switch *sw,
 	if (err)
 		goto err_ht_insert;
 
+	/* Handle nexthops */
+	err = __prestera_kern_fib_cache_create_nhs(sw, fib_cache);
+	if (err)
+		goto out; /* Not critical */
+
+out:
 	return fib_cache;
 
 err_ht_insert:
-	fib_info_put(fi);
+	fib_info_put(fen_info->fi);
 	kfree(fib_cache);
 err_kzalloc:
 	return NULL;
 }
 
+static void
+__prestera_k_arb_fib_nh_offload_set(struct prestera_switch *sw,
+				    struct prestera_kern_fib_cache *fibc,
+				    struct prestera_kern_neigh_cache *nc,
+				    bool offloaded, bool trap)
+{
+	struct fib_nh_common *nhc;
+	int i, nhs;
+
+	nhs = prestera_kern_fib_info_nhs(&fibc->info);
+	for (i = 0; i < nhs; i++) {
+		nhc = prestera_kern_fib_info_nhc(&fibc->info, i);
+		if (!nc) {
+			prestera_util_kern_set_nh_offload(nhc, offloaded, trap);
+			continue;
+		}
+
+		if (prestera_util_nhc_eq_n_cache_key(sw, nhc, &nc->key)) {
+			prestera_util_kern_set_nh_offload(nhc, offloaded, trap);
+			break;
+		}
+	}
+}
+
+static void
+__prestera_k_arb_n_offload_set(struct prestera_switch *sw,
+			       struct prestera_kern_neigh_cache *nc,
+			       bool offloaded)
+{
+	struct neighbour *n;
+
+	n = neigh_lookup(&arp_tbl, &nc->key.addr.u.ipv4,
+			 nc->key.dev);
+	if (!n)
+		return;
+
+	prestera_util_kern_set_neigh_offload(n, offloaded);
+	neigh_release(n);
+}
+
 static void
 __prestera_k_arb_fib_lpm_offload_set(struct prestera_switch *sw,
 				     struct prestera_kern_fib_cache *fc,
@@ -126,32 +699,208 @@ __prestera_k_arb_fib_lpm_offload_set(struct prestera_switch *sw,
 {
 	struct fib_rt_info fri;
 
-	if (fc->key.addr.v != PRESTERA_IPV4)
+	switch (fc->key.addr.v) {
+	case PRESTERA_IPV4:
+		fri.fi = fc->fen4_info.fi;
+		fri.tb_id = fc->key.kern_tb_id;
+		fri.dst = fc->key.addr.u.ipv4;
+		fri.dst_len = fc->key.prefix_len;
+		fri.dscp = fc->fen4_info.dscp;
+		fri.type = fc->fen4_info.type;
+		/* flags begin */
+		fri.offload = offload;
+		fri.trap = trap;
+		fri.offload_failed = fail;
+		/* flags end */
+		fib_alias_hw_flags_set(&init_net, &fri);
+		return;
+	case PRESTERA_IPV6:
+		/* TODO */
 		return;
+	}
+}
+
+static void
+__prestera_k_arb_n_lpm_set(struct prestera_switch *sw,
+			   struct prestera_kern_neigh_cache *n_cache,
+			   bool enabled)
+{
+	struct prestera_nexthop_group_key nh_grp_key;
+	struct prestera_kern_fib_cache_key fc_key;
+	struct prestera_kern_fib_cache *fib_cache;
+	struct prestera_fib_node *fib_node;
+	struct prestera_fib_key fib_key;
+
+	/* Exception for fc with prefix 32: LPM entry is already used by fib */
+	memset(&fc_key, 0, sizeof(fc_key));
+	fc_key.addr = n_cache->key.addr;
+	fc_key.prefix_len = PRESTERA_IP_ADDR_PLEN(n_cache->key.addr.v);
+	/* But better to use tb_id of route, which pointed to this neighbour. */
+	/* We take it from rif, because rif inconsistent.
+	 * Must be separated in_rif and out_rif.
+	 * Also note: for each fib pointed to this neigh should be separated
+	 *            neigh lpm entry (for each ingress vr)
+	 */
+	fc_key.kern_tb_id = l3mdev_fib_table(n_cache->key.dev);
+	fib_cache = prestera_kern_fib_cache_find(sw, &fc_key);
+	memset(&fib_key, 0, sizeof(fib_key));
+	fib_key.addr = n_cache->key.addr;
+	fib_key.prefix_len = PRESTERA_IP_ADDR_PLEN(n_cache->key.addr.v);
+	fib_key.tb_id = prestera_fix_tb_id(fc_key.kern_tb_id);
+	fib_node = prestera_fib_node_find(sw, &fib_key);
+	if (!fib_cache || !fib_cache->reachable) {
+		if (!enabled && fib_node) {
+			if (prestera_fib_node_util_is_neighbour(fib_node))
+				prestera_fib_node_destroy(sw, fib_node);
+			return;
+		}
+	}
+
+	if (enabled && !fib_node) {
+		memset(&nh_grp_key, 0, sizeof(nh_grp_key));
+		prestera_util_nc_key2nh_key(&n_cache->key,
+					    &nh_grp_key.neigh[0]);
+		fib_node = prestera_fib_node_create(sw, &fib_key,
+						    PRESTERA_FIB_TYPE_UC_NH,
+						    &nh_grp_key);
+		if (!fib_node)
+			pr_err("%s failed ip=%pI4n", "prestera_fib_node_create",
+			       &fib_key.addr.u.ipv4);
+		return;
+	}
+}
+
+static void
+__prestera_k_arb_nc_kern_fib_fetch(struct prestera_switch *sw,
+				   struct prestera_kern_neigh_cache *nc)
+{
+	if (prestera_util_kern_n_is_reachable(l3mdev_fib_table(nc->key.dev),
+					      &nc->key.addr, nc->key.dev))
+		nc->reachable = true;
+	else
+		nc->reachable = false;
+}
+
+/* Kernel neighbour -> neigh_cache info */
+static void
+__prestera_k_arb_nc_kern_n_fetch(struct prestera_switch *sw,
+				 struct prestera_kern_neigh_cache *nc)
+{
+	struct neighbour *n;
+	int err;
+
+	memset(&nc->nh_neigh_info, 0, sizeof(nc->nh_neigh_info));
+	n = neigh_lookup(&arp_tbl, &nc->key.addr.u.ipv4, nc->key.dev);
+	if (!n)
+		goto out;
+
+	read_lock_bh(&n->lock);
+	if (n->nud_state & NUD_VALID && !n->dead) {
+		err = prestera_neigh_iface_init(sw, &nc->nh_neigh_info.iface,
+						n);
+		if (err)
+			goto n_read_out;
 
-	fri.fi = fc->fi;
-	fri.tb_id = fc->key.kern_tb_id;
-	fri.dst = fc->key.addr.u.ipv4;
-	fri.dst_len = fc->key.prefix_len;
-	fri.dscp = fc->kern_dscp;
-	fri.type = fc->kern_type;
-	/* flags begin */
-	fri.offload = offload;
-	fri.trap = trap;
-	fri.offload_failed = fail;
-	/* flags end */
-	fib_alias_hw_flags_set(&init_net, &fri);
+		memcpy(&nc->nh_neigh_info.ha[0], &n->ha[0], ETH_ALEN);
+		nc->nh_neigh_info.connected = true;
+	}
+n_read_out:
+	read_unlock_bh(&n->lock);
+out:
+	nc->in_kernel = nc->nh_neigh_info.connected;
+	if (n)
+		neigh_release(n);
+}
+
+/* neigh_cache info -> lpm update */
+static void
+__prestera_k_arb_nc_apply(struct prestera_switch *sw,
+			  struct prestera_kern_neigh_cache *nc)
+{
+	struct prestera_kern_neigh_cache_head *nhead;
+	struct prestera_nh_neigh_key nh_key;
+	struct prestera_nh_neigh *nh_neigh;
+	int err;
+
+	__prestera_k_arb_n_lpm_set(sw, nc, nc->reachable && nc->in_kernel);
+	__prestera_k_arb_n_offload_set(sw, nc, nc->reachable && nc->in_kernel);
+
+	prestera_util_nc_key2nh_key(&nc->key, &nh_key);
+	nh_neigh = prestera_nh_neigh_find(sw, &nh_key);
+	if (!nh_neigh)
+		goto out;
+
+	/* Do hw update only if something changed to prevent nh flap */
+	if (memcmp(&nc->nh_neigh_info, &nh_neigh->info,
+		   sizeof(nh_neigh->info))) {
+		memcpy(&nh_neigh->info, &nc->nh_neigh_info,
+		       sizeof(nh_neigh->info));
+		err = prestera_nh_neigh_set(sw, nh_neigh);
+		if (err) {
+			pr_err("%s failed with err=%d ip=%pI4n mac=%pM",
+			       "prestera_nh_neigh_set", err,
+			       &nh_neigh->key.addr.u.ipv4,
+			       &nh_neigh->info.ha[0]);
+			goto out;
+		}
+	}
+
+out:
+	list_for_each_entry(nhead, &nc->kern_fib_cache_list, head) {
+		__prestera_k_arb_fib_nh_offload_set(sw, nhead->this, nc,
+						    nc->in_kernel,
+						    !nc->in_kernel);
+	}
 }
 
 static int
 __prestera_pr_k_arb_fc_lpm_info_calc(struct prestera_switch *sw,
 				     struct prestera_kern_fib_cache *fc)
 {
+	struct fib_nh_common *nhc;
+	int nh_cnt;
+
 	memset(&fc->lpm_info, 0, sizeof(fc->lpm_info));
 
-	switch (fc->fi->fib_type) {
+	switch (prestera_kern_fib_info_type(&fc->info)) {
 	case RTN_UNICAST:
-		fc->lpm_info.fib_type = PRESTERA_FIB_TYPE_TRAP;
+		if (prestera_fib_info_is_direct(&fc->info) &&
+		    fc->key.prefix_len ==
+			PRESTERA_IP_ADDR_PLEN(fc->key.addr.v)) {
+			/* This is special case.
+			 * When prefix is 32. Than we will have conflict in lpm
+			 * for direct route - once TRAP added, there is no
+			 * place for neighbour entry. So represent direct route
+			 * with prefix 32, as NH. So neighbour will be resolved
+			 * as nexthop of this route.
+			 */
+			nhc = prestera_kern_fib_info_nhc(&fc->info, 0);
+			fc->lpm_info.fib_type = PRESTERA_FIB_TYPE_UC_NH;
+			fc->lpm_info.nh_grp_key.neigh[0].addr =
+				fc->key.addr;
+			fc->lpm_info.nh_grp_key.neigh[0].rif =
+				nhc->nhc_dev;
+
+			break;
+		}
+
+		/* We can also get nh_grp_key from fi. This will be correct to
+		 * because cache not always represent, what actually written to
+		 * lpm. But we use nh cache, as well for now (for this case).
+		 */
+		for (nh_cnt = 0; nh_cnt < PRESTERA_NHGR_SIZE_MAX; nh_cnt++) {
+			if (!fc->kern_neigh_cache_head[nh_cnt].n_cache)
+				break;
+
+			fc->lpm_info.nh_grp_key.neigh[nh_cnt].addr =
+				fc->kern_neigh_cache_head[nh_cnt].n_cache->key.addr;
+			fc->lpm_info.nh_grp_key.neigh[nh_cnt].rif =
+				fc->kern_neigh_cache_head[nh_cnt].n_cache->key.dev;
+		}
+
+		fc->lpm_info.fib_type = nh_cnt ?
+					PRESTERA_FIB_TYPE_UC_NH :
+					PRESTERA_FIB_TYPE_TRAP;
 		break;
 	/* Unsupported. Leave it for kernel: */
 	case RTN_BROADCAST:
@@ -191,7 +940,8 @@ static int __prestera_k_arb_f_lpm_set(struct prestera_switch *sw,
 		return 0;
 
 	fib_node = prestera_fib_node_create(sw, &fc->lpm_info.fib_key,
-					    fc->lpm_info.fib_type);
+					    fc->lpm_info.fib_type,
+					    &fc->lpm_info.nh_grp_key);
 
 	if (!fib_node) {
 		dev_err(sw->dev->dev, "fib_node=NULL %pI4n/%d kern_tb_id = %d",
@@ -220,6 +970,10 @@ static int __prestera_k_arb_fc_apply(struct prestera_switch *sw,
 	}
 
 	switch (fc->lpm_info.fib_type) {
+	case PRESTERA_FIB_TYPE_UC_NH:
+		__prestera_k_arb_fib_lpm_offload_set(sw, fc, false,
+						     fc->reachable, false);
+		break;
 	case PRESTERA_FIB_TYPE_TRAP:
 		__prestera_k_arb_fib_lpm_offload_set(sw, fc, false,
 						     false, fc->reachable);
@@ -271,17 +1025,140 @@ __prestera_k_arb_util_fib_overlapped(struct prestera_switch *sw,
 	return rfc;
 }
 
+static void __prestera_k_arb_hw_state_upd(struct prestera_switch *sw,
+					  struct prestera_kern_neigh_cache *nc)
+{
+	struct prestera_nh_neigh_key nh_key;
+	struct prestera_nh_neigh *nh_neigh;
+	struct neighbour *n;
+	bool hw_active;
+
+	prestera_util_nc_key2nh_key(&nc->key, &nh_key);
+	nh_neigh = prestera_nh_neigh_find(sw, &nh_key);
+	if (!nh_neigh) {
+		pr_err("Cannot find nh_neigh for cached %pI4n",
+		       &nc->key.addr.u.ipv4);
+		return;
+	}
+
+	hw_active = prestera_nh_neigh_util_hw_state(sw, nh_neigh);
+
+#ifdef PRESTERA_IMPLICITY_RESOLVE_DEAD_NEIGH
+	if (!hw_active && nc->in_kernel)
+		goto out;
+#else /* PRESTERA_IMPLICITY_RESOLVE_DEAD_NEIGH */
+	if (!hw_active)
+		goto out;
+#endif /* PRESTERA_IMPLICITY_RESOLVE_DEAD_NEIGH */
+
+	if (nc->key.addr.v == PRESTERA_IPV4) {
+		n = neigh_lookup(&arp_tbl, &nc->key.addr.u.ipv4,
+				 nc->key.dev);
+		if (!n)
+			n = neigh_create(&arp_tbl, &nc->key.addr.u.ipv4,
+					 nc->key.dev);
+	} else {
+		n = NULL;
+	}
+
+	if (!IS_ERR(n) && n) {
+		neigh_event_send(n, NULL);
+		neigh_release(n);
+	} else {
+		pr_err("Cannot create neighbour %pI4n", &nc->key.addr.u.ipv4);
+	}
+
+out:
+	return;
+}
+
+/* Propagate hw state to kernel */
+static void prestera_k_arb_hw_evt(struct prestera_switch *sw)
+{
+	struct prestera_kern_neigh_cache *n_cache;
+	struct rhashtable_iter iter;
+
+	rhashtable_walk_enter(&sw->router->kern_neigh_cache_ht, &iter);
+	rhashtable_walk_start(&iter);
+	while (1) {
+		n_cache = rhashtable_walk_next(&iter);
+
+		if (!n_cache)
+			break;
+
+		if (IS_ERR(n_cache))
+			continue;
+
+		rhashtable_walk_stop(&iter);
+		__prestera_k_arb_hw_state_upd(sw, n_cache);
+		rhashtable_walk_start(&iter);
+	}
+	rhashtable_walk_stop(&iter);
+	rhashtable_walk_exit(&iter);
+}
+
+/* Propagate kernel event to hw */
+static void prestera_k_arb_n_evt(struct prestera_switch *sw,
+				 struct neighbour *n)
+{
+	struct prestera_kern_neigh_cache_key n_key;
+	struct prestera_kern_neigh_cache *n_cache;
+	int err;
+
+	err = prestera_util_neigh2nc_key(sw, n, &n_key);
+	if (err)
+		return;
+
+	n_cache = prestera_kern_neigh_cache_find(sw, &n_key);
+	if (!n_cache) {
+		n_cache = prestera_kern_neigh_cache_get(sw, &n_key);
+		if (!n_cache)
+			return;
+		__prestera_k_arb_nc_kern_fib_fetch(sw, n_cache);
+	}
+
+	__prestera_k_arb_nc_kern_n_fetch(sw, n_cache);
+	__prestera_k_arb_nc_apply(sw, n_cache);
+
+	prestera_kern_neigh_cache_put(sw, n_cache);
+}
+
+static void __prestera_k_arb_fib_evt2nc(struct prestera_switch *sw)
+{
+	struct prestera_kern_neigh_cache *n_cache;
+	struct rhashtable_iter iter;
+
+	rhashtable_walk_enter(&sw->router->kern_neigh_cache_ht, &iter);
+	rhashtable_walk_start(&iter);
+	while (1) {
+		n_cache = rhashtable_walk_next(&iter);
+
+		if (!n_cache)
+			break;
+
+		if (IS_ERR(n_cache))
+			continue;
+
+		rhashtable_walk_stop(&iter);
+		__prestera_k_arb_nc_kern_fib_fetch(sw, n_cache);
+		__prestera_k_arb_nc_apply(sw, n_cache);
+		rhashtable_walk_start(&iter);
+	}
+	rhashtable_walk_stop(&iter);
+	rhashtable_walk_exit(&iter);
+}
+
 static int
 prestera_k_arb_fib_evt(struct prestera_switch *sw,
 		       bool replace, /* replace or del */
-		       struct fib_entry_notifier_info *fen_info)
+		       struct fib_notifier_info *info)
 {
 	struct prestera_kern_fib_cache *tfib_cache, *bfib_cache; /* top/btm */
 	struct prestera_kern_fib_cache_key fc_key;
 	struct prestera_kern_fib_cache *fib_cache;
 	int err;
 
-	prestera_util_fen_info2fib_cache_key(fen_info, &fc_key);
+	prestera_util_fen_info2fib_cache_key(info, &fc_key);
 	fib_cache = prestera_kern_fib_cache_find(sw, &fc_key);
 	if (fib_cache) {
 		fib_cache->reachable = false;
@@ -304,10 +1181,7 @@ prestera_k_arb_fib_evt(struct prestera_switch *sw,
 	}
 
 	if (replace) {
-		fib_cache = prestera_kern_fib_cache_create(sw, &fc_key,
-							   fen_info->fi,
-							   fen_info->dscp,
-							   fen_info->type);
+		fib_cache = prestera_kern_fib_cache_create(sw, &fc_key, info);
 		if (!fib_cache) {
 			dev_err(sw->dev->dev, "fib_cache == NULL");
 			return -ENOENT;
@@ -331,9 +1205,65 @@ prestera_k_arb_fib_evt(struct prestera_switch *sw,
 			dev_err(sw->dev->dev, "Applying fib_cache failed");
 	}
 
+	/* Update all neighs to resolve overlapped and apply related */
+	__prestera_k_arb_fib_evt2nc(sw);
+
 	return 0;
 }
 
+static void __prestera_k_arb_abort_neigh_ht_cb(void *ptr, void *arg)
+{
+	struct prestera_kern_neigh_cache *n_cache = ptr;
+	struct prestera_switch *sw = arg;
+
+	if (!list_empty(&n_cache->kern_fib_cache_list)) {
+		WARN_ON(1); /* BUG */
+		return;
+	}
+	__prestera_k_arb_n_offload_set(sw, n_cache, false);
+	n_cache->in_kernel = false;
+	/* No need to destroy lpm.
+	 * It will be aborted by destroy_ht
+	 */
+	__prestera_kern_neigh_cache_destruct(sw, n_cache);
+	kfree(n_cache);
+}
+
+static void __prestera_k_arb_abort_fib_ht_cb(void *ptr, void *arg)
+{
+	struct prestera_kern_fib_cache *fib_cache = ptr;
+	struct prestera_switch *sw = arg;
+
+	__prestera_k_arb_fib_lpm_offload_set(sw, fib_cache,
+					     false, false,
+					     false);
+	__prestera_k_arb_fib_nh_offload_set(sw, fib_cache, NULL,
+					    false, false);
+	/* No need to destroy lpm.
+	 * It will be aborted by destroy_ht
+	 */
+	__prestera_kern_fib_cache_destruct(sw, fib_cache);
+	kfree(fib_cache);
+}
+
+static void prestera_k_arb_abort(struct prestera_switch *sw)
+{
+	/* Function to remove all arbiter entries and related hw objects. */
+	/* Sequence:
+	 *   1) Clear arbiter tables, but don't touch hw
+	 *   2) Clear hw
+	 * We use such approach, because arbiter object is not directly mapped
+	 * to hw. So deletion of one arbiter object may even lead to creation of
+	 * hw object (e.g. in case of overlapped routes).
+	 */
+	rhashtable_free_and_destroy(&sw->router->kern_fib_cache_ht,
+				    __prestera_k_arb_abort_fib_ht_cb,
+				    sw);
+	rhashtable_free_and_destroy(&sw->router->kern_neigh_cache_ht,
+				    __prestera_k_arb_abort_neigh_ht_cb,
+				    sw);
+}
+
 static int __prestera_inetaddr_port_event(struct net_device *port_dev,
 					  unsigned long event,
 					  struct netlink_ext_ack *extack)
@@ -469,13 +1399,15 @@ static void __prestera_router_fib_event_work(struct work_struct *work)
 
 	switch (fib_work->event) {
 	case FIB_EVENT_ENTRY_REPLACE:
-		err = prestera_k_arb_fib_evt(sw, true, &fib_work->fen_info);
+		err = prestera_k_arb_fib_evt(sw, true,
+					     &fib_work->fen_info.info);
 		if (err)
 			goto err_out;
 
 		break;
 	case FIB_EVENT_ENTRY_DEL:
-		err = prestera_k_arb_fib_evt(sw, false, &fib_work->fen_info);
+		err = prestera_k_arb_fib_evt(sw, false,
+					     &fib_work->fen_info.info);
 		if (err)
 			goto err_out;
 
@@ -534,10 +1466,89 @@ static int __prestera_router_fib_event(struct notifier_block *nb,
 	return NOTIFY_DONE;
 }
 
+struct prestera_netevent_work {
+	struct work_struct work;
+	struct prestera_switch *sw;
+	struct neighbour *n;
+};
+
+static void prestera_router_neigh_event_work(struct work_struct *work)
+{
+	struct prestera_netevent_work *net_work =
+		container_of(work, struct prestera_netevent_work, work);
+	struct prestera_switch *sw = net_work->sw;
+	struct neighbour *n = net_work->n;
+
+	/* neigh - its not hw related object. It stored only in kernel. So... */
+	rtnl_lock();
+
+	prestera_k_arb_n_evt(sw, n);
+
+	neigh_release(n);
+	rtnl_unlock();
+	kfree(net_work);
+}
+
+static int prestera_router_netevent_event(struct notifier_block *nb,
+					  unsigned long event, void *ptr)
+{
+	struct prestera_netevent_work *net_work;
+	struct prestera_router *router;
+	struct neighbour *n = ptr;
+
+	router = container_of(nb, struct prestera_router, netevent_nb);
+
+	switch (event) {
+	case NETEVENT_NEIGH_UPDATE:
+		if (n->tbl->family != AF_INET)
+			return NOTIFY_DONE;
+
+		net_work = kzalloc(sizeof(*net_work), GFP_ATOMIC);
+		if (WARN_ON(!net_work))
+			return NOTIFY_BAD;
+
+		neigh_clone(n);
+		net_work->n = n;
+		net_work->sw = router->sw;
+		INIT_WORK(&net_work->work, prestera_router_neigh_event_work);
+		prestera_queue_work(&net_work->work);
+	}
+
+	return NOTIFY_DONE;
+}
+
+static void prestera_router_update_neighs_work(struct work_struct *work)
+{
+	struct prestera_router *router;
+
+	router = container_of(work, struct prestera_router,
+			      neighs_update.dw.work);
+	rtnl_lock();
+
+	prestera_k_arb_hw_evt(router->sw);
+
+	rtnl_unlock();
+	prestera_queue_delayed_work(&router->neighs_update.dw,
+				    msecs_to_jiffies(PRESTERA_NH_PROBE_INTERVAL));
+}
+
+static int prestera_neigh_work_init(struct prestera_switch *sw)
+{
+	INIT_DELAYED_WORK(&sw->router->neighs_update.dw,
+			  prestera_router_update_neighs_work);
+	prestera_queue_delayed_work(&sw->router->neighs_update.dw, 0);
+	return 0;
+}
+
+static void prestera_neigh_work_fini(struct prestera_switch *sw)
+{
+	cancel_delayed_work_sync(&sw->router->neighs_update.dw);
+}
+
 int prestera_router_init(struct prestera_switch *sw)
 {
 	struct prestera_router *router;
-	int err;
+	int err, nhgrp_cache_bytes;
 
 	router = kzalloc(sizeof(*sw->router), GFP_KERNEL);
 	if (!router)
@@ -555,6 +1566,22 @@ int prestera_router_init(struct prestera_switch *sw)
 	if (err)
 		goto err_kern_fib_cache_ht_init;
 
+	err = rhashtable_init(&router->kern_neigh_cache_ht,
+			      &__prestera_kern_neigh_cache_ht_params);
+	if (err)
+		goto err_kern_neigh_cache_ht_init;
+
+	nhgrp_cache_bytes = sw->size_tbl_router_nexthop / 8 + 1;
+	router->nhgrp_hw_state_cache = kzalloc(nhgrp_cache_bytes, GFP_KERNEL);
+	if (!router->nhgrp_hw_state_cache) {
+		err = -ENOMEM;
+		goto err_nh_state_cache_alloc;
+	}
+
+	err = prestera_neigh_work_init(sw);
+	if (err)
+		goto err_neigh_work_init;
+
 	router->inetaddr_valid_nb.notifier_call = __prestera_inetaddr_valid_cb;
 	err = register_inetaddr_validator_notifier(&router->inetaddr_valid_nb);
 	if (err)
@@ -565,6 +1592,11 @@ int prestera_router_init(struct prestera_switch *sw)
 	if (err)
 		goto err_register_inetaddr_notifier;
 
+	router->netevent_nb.notifier_call = prestera_router_netevent_event;
+	err = register_netevent_notifier(&router->netevent_nb);
+	if (err)
+		goto err_register_netevent_notifier;
+
 	router->fib_nb.notifier_call = __prestera_router_fib_event;
 	err = register_fib_notifier(&init_net, &router->fib_nb,
 				    /* TODO: flush fib entries */ NULL, NULL);
@@ -574,10 +1606,18 @@ int prestera_router_init(struct prestera_switch *sw)
 	return 0;
 
 err_register_fib_notifier:
+	unregister_netevent_notifier(&router->netevent_nb);
+err_register_netevent_notifier:
 	unregister_inetaddr_notifier(&router->inetaddr_nb);
 err_register_inetaddr_notifier:
 	unregister_inetaddr_validator_notifier(&router->inetaddr_valid_nb);
 err_register_inetaddr_validator_notifier:
+	prestera_neigh_work_fini(sw);
+err_neigh_work_init:
+	kfree(router->nhgrp_hw_state_cache);
+err_nh_state_cache_alloc:
+	rhashtable_destroy(&router->kern_neigh_cache_ht);
+err_kern_neigh_cache_ht_init:
 	rhashtable_destroy(&router->kern_fib_cache_ht);
 err_kern_fib_cache_ht_init:
 	prestera_router_hw_fini(sw);
@@ -589,8 +1629,15 @@ int prestera_router_init(struct prestera_switch *sw)
 void prestera_router_fini(struct prestera_switch *sw)
 {
 	unregister_fib_notifier(&init_net, &sw->router->fib_nb);
+	unregister_netevent_notifier(&sw->router->netevent_nb);
 	unregister_inetaddr_notifier(&sw->router->inetaddr_nb);
 	unregister_inetaddr_validator_notifier(&sw->router->inetaddr_valid_nb);
+	prestera_neigh_work_fini(sw);
+	prestera_queue_drain();
+
+	prestera_k_arb_abort(sw);
+
+	kfree(sw->router->nhgrp_hw_state_cache);
 	rhashtable_destroy(&sw->router->kern_fib_cache_ht);
 	prestera_router_hw_fini(sw);
 	kfree(sw->router);

drivers/net/ethernet/marvell/prestera/prestera_router_hw.c

@@ -8,10 +8,16 @@
 #include "prestera_router_hw.h"
 #include "prestera_acl.h"
 
-/*            +--+
- *   +------->|vr|<-+
- *   |        +--+  |
- *   |              |
+/*                                Nexthop is pointed
+ *                                to port (not rif)
+ *                                +-------+
+ *                              +>|nexthop|
+ *                              | +-------+
+ *                              |
+ *            +--+        +-----++
+ *   +------->|vr|<-+   +>|nh_grp|
+ *   |        +--+  |   | +------+
+ *   |              |   |
  * +-+-------+   +--+---+-+
  * |rif_entry|   |fib_node|
  * +---------+   +--------+
@@ -23,6 +29,8 @@
 
 #define PRESTERA_NHGR_UNUSED (0)
 #define PRESTERA_NHGR_DROP (0xFFFFFFFF)
+/* Need to merge it with router_manager */
+#define PRESTERA_NH_ACTIVE_JIFFER_FILTER 3000 /* ms */
 
 static const struct rhashtable_params __prestera_fib_ht_params = {
 	.key_offset  = offsetof(struct prestera_fib_node, key),
@@ -31,10 +39,45 @@ static const struct rhashtable_params __prestera_fib_ht_params = {
 	.automatic_shrinking = true,
 };
 
+static const struct rhashtable_params __prestera_nh_neigh_ht_params = {
+	.key_offset  = offsetof(struct prestera_nh_neigh, key),
+	.key_len     = sizeof(struct prestera_nh_neigh_key),
+	.head_offset = offsetof(struct prestera_nh_neigh, ht_node),
+};
+
+static const struct rhashtable_params __prestera_nexthop_group_ht_params = {
+	.key_offset  = offsetof(struct prestera_nexthop_group, key),
+	.key_len     = sizeof(struct prestera_nexthop_group_key),
+	.head_offset = offsetof(struct prestera_nexthop_group, ht_node),
+};
+
+static int prestera_nexthop_group_set(struct prestera_switch *sw,
+				      struct prestera_nexthop_group *nh_grp);
+static bool
+prestera_nexthop_group_util_hw_state(struct prestera_switch *sw,
+				     struct prestera_nexthop_group *nh_grp);
+static void prestera_fib_node_destroy_ht_cb(void *ptr, void *arg);
+
+/* TODO: move to router.h as macros */
+static bool prestera_nh_neigh_key_is_valid(struct prestera_nh_neigh_key *key)
+{
+	return memchr_inv(key, 0, sizeof(*key)) ? true : false;
+}
+
 int prestera_router_hw_init(struct prestera_switch *sw)
 {
 	int err;
 
+	err = rhashtable_init(&sw->router->nh_neigh_ht,
+			      &__prestera_nh_neigh_ht_params);
+	if (err)
+		goto err_nh_neigh_ht_init;
+
+	err = rhashtable_init(&sw->router->nexthop_group_ht,
+			      &__prestera_nexthop_group_ht_params);
+	if (err)
+		goto err_nexthop_grp_ht_init;
+
 	err = rhashtable_init(&sw->router->fib_ht,
 			      &__prestera_fib_ht_params);
 	if (err)
@@ -43,15 +86,25 @@ int prestera_router_hw_init(struct prestera_switch *sw)
 	INIT_LIST_HEAD(&sw->router->vr_list);
 	INIT_LIST_HEAD(&sw->router->rif_entry_list);
 
+	return 0;
+
 err_fib_ht_init:
+	rhashtable_destroy(&sw->router->nexthop_group_ht);
+err_nexthop_grp_ht_init:
+	rhashtable_destroy(&sw->router->nh_neigh_ht);
+err_nh_neigh_ht_init:
 	return 0;
 }
 
 void prestera_router_hw_fini(struct prestera_switch *sw)
 {
+	rhashtable_free_and_destroy(&sw->router->fib_ht,
+				    prestera_fib_node_destroy_ht_cb, sw);
 	WARN_ON(!list_empty(&sw->router->vr_list));
 	WARN_ON(!list_empty(&sw->router->rif_entry_list));
 	rhashtable_destroy(&sw->router->fib_ht);
+	rhashtable_destroy(&sw->router->nexthop_group_ht);
+	rhashtable_destroy(&sw->router->nh_neigh_ht);
 }
 
 static struct prestera_vr *__prestera_vr_find(struct prestera_switch *sw,
@@ -232,6 +285,286 @@ prestera_rif_entry_create(struct prestera_switch *sw,
 	return NULL;
 }
 
+static void __prestera_nh_neigh_destroy(struct prestera_switch *sw,
+					struct prestera_nh_neigh *neigh)
+{
+	rhashtable_remove_fast(&sw->router->nh_neigh_ht,
+			       &neigh->ht_node,
+			       __prestera_nh_neigh_ht_params);
+	kfree(neigh);
+}
+
+static struct prestera_nh_neigh *
+__prestera_nh_neigh_create(struct prestera_switch *sw,
+			   struct prestera_nh_neigh_key *key)
+{
+	struct prestera_nh_neigh *neigh;
+	int err;
+
+	neigh = kzalloc(sizeof(*neigh), GFP_KERNEL);
+	if (!neigh)
+		goto err_kzalloc;
+
+	memcpy(&neigh->key, key, sizeof(*key));
+	neigh->info.connected = false;
+	INIT_LIST_HEAD(&neigh->nexthop_group_list);
+	err = rhashtable_insert_fast(&sw->router->nh_neigh_ht,
+				     &neigh->ht_node,
+				     __prestera_nh_neigh_ht_params);
+	if (err)
+		goto err_rhashtable_insert;
+
+	return neigh;
+
+err_rhashtable_insert:
+	kfree(neigh);
+err_kzalloc:
+	return NULL;
+}
+
+struct prestera_nh_neigh *
+prestera_nh_neigh_find(struct prestera_switch *sw,
+		       struct prestera_nh_neigh_key *key)
+{
+	struct prestera_nh_neigh *nh_neigh;
+
+	nh_neigh = rhashtable_lookup_fast(&sw->router->nh_neigh_ht,
+					  key, __prestera_nh_neigh_ht_params);
+	return IS_ERR(nh_neigh) ? NULL : nh_neigh;
+}
+
+struct prestera_nh_neigh *
+prestera_nh_neigh_get(struct prestera_switch *sw,
+		      struct prestera_nh_neigh_key *key)
+{
+	struct prestera_nh_neigh *neigh;
+
+	neigh = prestera_nh_neigh_find(sw, key);
+	if (!neigh)
+		return __prestera_nh_neigh_create(sw, key);
+
+	return neigh;
+}
+
+void prestera_nh_neigh_put(struct prestera_switch *sw,
+			   struct prestera_nh_neigh *neigh)
+{
+	if (list_empty(&neigh->nexthop_group_list))
+		__prestera_nh_neigh_destroy(sw, neigh);
+}
+
+/* Updates new prestera_neigh_info */
+int prestera_nh_neigh_set(struct prestera_switch *sw,
+			  struct prestera_nh_neigh *neigh)
+{
+	struct prestera_nh_neigh_head *nh_head;
+	struct prestera_nexthop_group *nh_grp;
+	int err;
+
+	list_for_each_entry(nh_head, &neigh->nexthop_group_list, head) {
+		nh_grp = nh_head->this;
+		err = prestera_nexthop_group_set(sw, nh_grp);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+bool prestera_nh_neigh_util_hw_state(struct prestera_switch *sw,
+				     struct prestera_nh_neigh *nh_neigh)
+{
+	bool state;
+	struct prestera_nh_neigh_head *nh_head, *tmp;
+
+	state = false;
+	list_for_each_entry_safe(nh_head, tmp,
+				 &nh_neigh->nexthop_group_list, head) {
+		state = prestera_nexthop_group_util_hw_state(sw, nh_head->this);
+		if (state)
+			goto out;
+	}
+
+out:
+	return state;
+}
+
+static struct prestera_nexthop_group *
+__prestera_nexthop_group_create(struct prestera_switch *sw,
+				struct prestera_nexthop_group_key *key)
+{
+	struct prestera_nexthop_group *nh_grp;
+	struct prestera_nh_neigh *nh_neigh;
+	int nh_cnt, err, gid;
+
+	nh_grp = kzalloc(sizeof(*nh_grp), GFP_KERNEL);
+	if (!nh_grp)
+		goto err_kzalloc;
+
+	memcpy(&nh_grp->key, key, sizeof(*key));
+	for (nh_cnt = 0; nh_cnt < PRESTERA_NHGR_SIZE_MAX; nh_cnt++) {
+		if (!prestera_nh_neigh_key_is_valid(&nh_grp->key.neigh[nh_cnt]))
+			break;
+
+		nh_neigh = prestera_nh_neigh_get(sw,
+						 &nh_grp->key.neigh[nh_cnt]);
+		if (!nh_neigh)
+			goto err_nh_neigh_get;
+
+		nh_grp->nh_neigh_head[nh_cnt].neigh = nh_neigh;
+		nh_grp->nh_neigh_head[nh_cnt].this = nh_grp;
+		list_add(&nh_grp->nh_neigh_head[nh_cnt].head,
+			 &nh_neigh->nexthop_group_list);
+	}
+
+	err = prestera_hw_nh_group_create(sw, nh_cnt, &nh_grp->grp_id);
+	if (err)
+		goto err_nh_group_create;
+
+	err = prestera_nexthop_group_set(sw, nh_grp);
+	if (err)
+		goto err_nexthop_group_set;
+
+	err = rhashtable_insert_fast(&sw->router->nexthop_group_ht,
+				     &nh_grp->ht_node,
+				     __prestera_nexthop_group_ht_params);
+	if (err)
+		goto err_ht_insert;
+
+	/* reset cache for created group */
+	gid = nh_grp->grp_id;
+	sw->router->nhgrp_hw_state_cache[gid / 8] &= ~BIT(gid % 8);
+
+	return nh_grp;
+
+err_ht_insert:
+err_nexthop_group_set:
+	prestera_hw_nh_group_delete(sw, nh_cnt, nh_grp->grp_id);
+err_nh_group_create:
+err_nh_neigh_get:
+	for (nh_cnt--; nh_cnt >= 0; nh_cnt--) {
+		list_del(&nh_grp->nh_neigh_head[nh_cnt].head);
+		prestera_nh_neigh_put(sw, nh_grp->nh_neigh_head[nh_cnt].neigh);
+	}
+
+	kfree(nh_grp);
+err_kzalloc:
+	return NULL;
+}
+
+static void
+__prestera_nexthop_group_destroy(struct prestera_switch *sw,
+				 struct prestera_nexthop_group *nh_grp)
+{
+	struct prestera_nh_neigh *nh_neigh;
+	int nh_cnt;
+
+	rhashtable_remove_fast(&sw->router->nexthop_group_ht,
+			       &nh_grp->ht_node,
+			       __prestera_nexthop_group_ht_params);
+
+	for (nh_cnt = 0; nh_cnt < PRESTERA_NHGR_SIZE_MAX; nh_cnt++) {
+		nh_neigh = nh_grp->nh_neigh_head[nh_cnt].neigh;
+		if (!nh_neigh)
+			break;
+
+		list_del(&nh_grp->nh_neigh_head[nh_cnt].head);
+		prestera_nh_neigh_put(sw, nh_neigh);
+	}
+
+	prestera_hw_nh_group_delete(sw, nh_cnt, nh_grp->grp_id);
+	kfree(nh_grp);
+}
+
+static struct prestera_nexthop_group *
+__prestera_nexthop_group_find(struct prestera_switch *sw,
+			      struct prestera_nexthop_group_key *key)
+{
+	struct prestera_nexthop_group *nh_grp;
+
+	nh_grp = rhashtable_lookup_fast(&sw->router->nexthop_group_ht,
+					key, __prestera_nexthop_group_ht_params);
+	return IS_ERR(nh_grp) ? NULL : nh_grp;
+}
+
+static struct prestera_nexthop_group *
+prestera_nexthop_group_get(struct prestera_switch *sw,
+			   struct prestera_nexthop_group_key *key)
+{
+	struct prestera_nexthop_group *nh_grp;
+
+	nh_grp = __prestera_nexthop_group_find(sw, key);
+	if (nh_grp) {
+		refcount_inc(&nh_grp->refcount);
+	} else {
+		nh_grp = __prestera_nexthop_group_create(sw, key);
+		if (IS_ERR(nh_grp))
+			return ERR_CAST(nh_grp);
+
+		refcount_set(&nh_grp->refcount, 1);
+	}
+
+	return nh_grp;
+}
+
+static void prestera_nexthop_group_put(struct prestera_switch *sw,
+				       struct prestera_nexthop_group *nh_grp)
+{
+	if (refcount_dec_and_test(&nh_grp->refcount))
+		__prestera_nexthop_group_destroy(sw, nh_grp);
+}
+
+/* Updates with new nh_neigh's info */
+static int prestera_nexthop_group_set(struct prestera_switch *sw,
+				      struct prestera_nexthop_group *nh_grp)
+{
+	struct prestera_neigh_info info[PRESTERA_NHGR_SIZE_MAX];
+	struct prestera_nh_neigh *neigh;
+	int nh_cnt;
+
+	memset(&info[0], 0, sizeof(info));
+	for (nh_cnt = 0; nh_cnt < PRESTERA_NHGR_SIZE_MAX; nh_cnt++) {
+		neigh = nh_grp->nh_neigh_head[nh_cnt].neigh;
+		if (!neigh)
+			break;
+
+		memcpy(&info[nh_cnt], &neigh->info, sizeof(neigh->info));
+	}
+
+	return prestera_hw_nh_entries_set(sw, nh_cnt, &info[0], nh_grp->grp_id);
+}
+
+static bool
+prestera_nexthop_group_util_hw_state(struct prestera_switch *sw,
+				     struct prestera_nexthop_group *nh_grp)
+{
+	int err;
+	u32 buf_size = sw->size_tbl_router_nexthop / 8 + 1;
+	u32 gid = nh_grp->grp_id;
+	u8 *cache = sw->router->nhgrp_hw_state_cache;
+
+	/* Antijitter
+	 * Prevent situation, when we read state of nh_grp twice in short time,
+	 * and state bit is still cleared on second call. So just stuck active
+	 * state for PRESTERA_NH_ACTIVE_JIFFER_FILTER, after last occurred.
+	 */
+	if (!time_before(jiffies, sw->router->nhgrp_hw_cache_kick +
+			msecs_to_jiffies(PRESTERA_NH_ACTIVE_JIFFER_FILTER))) {
+		err = prestera_hw_nhgrp_blk_get(sw, cache, buf_size);
+		if (err) {
+			pr_err("Failed to get hw state nh_grp's");
+			return false;
+		}
+
+		sw->router->nhgrp_hw_cache_kick = jiffies;
+	}
+
+	if (cache[gid / 8] & BIT(gid % 8))
+		return true;
+
+	return false;
+}
+
 struct prestera_fib_node *
 prestera_fib_node_find(struct prestera_switch *sw, struct prestera_fib_key *key)
 {
@@ -251,6 +584,9 @@ static void __prestera_fib_node_destruct(struct prestera_switch *sw,
 	prestera_hw_lpm_del(sw, vr->hw_vr_id, fib_node->key.addr.u.ipv4,
 			    fib_node->key.prefix_len);
 	switch (fib_node->info.type) {
+	case PRESTERA_FIB_TYPE_UC_NH:
+		prestera_nexthop_group_put(sw, fib_node->info.nh_grp);
+		break;
 	case PRESTERA_FIB_TYPE_TRAP:
 		break;
 	case PRESTERA_FIB_TYPE_DROP:
@@ -272,10 +608,20 @@ void prestera_fib_node_destroy(struct prestera_switch *sw,
 	kfree(fib_node);
 }
 
+static void prestera_fib_node_destroy_ht_cb(void *ptr, void *arg)
+{
+	struct prestera_fib_node *node = ptr;
+	struct prestera_switch *sw = arg;
+
+	__prestera_fib_node_destruct(sw, node);
+	kfree(node);
+}
+
 struct prestera_fib_node *
 prestera_fib_node_create(struct prestera_switch *sw,
 			 struct prestera_fib_key *key,
-			 enum prestera_fib_type fib_type)
+			 enum prestera_fib_type fib_type,
+			 struct prestera_nexthop_group_key *nh_grp_key)
 {
 	struct prestera_fib_node *fib_node;
 	u32 grp_id;
@@ -302,6 +648,14 @@ prestera_fib_node_create(struct prestera_switch *sw,
 	case PRESTERA_FIB_TYPE_DROP:
 		grp_id = PRESTERA_NHGR_DROP;
 		break;
+	case PRESTERA_FIB_TYPE_UC_NH:
+		fib_node->info.nh_grp = prestera_nexthop_group_get(sw,
+								   nh_grp_key);
+		if (!fib_node->info.nh_grp)
+			goto err_nh_grp_get;
+
+		grp_id = fib_node->info.nh_grp->grp_id;
+		break;
 	default:
 		pr_err("Unsupported fib_type %d", fib_type);
 		goto err_nh_grp_get;
@@ -323,6 +677,8 @@ prestera_fib_node_create(struct prestera_switch *sw,
 	prestera_hw_lpm_del(sw, vr->hw_vr_id, key->addr.u.ipv4,
 			    key->prefix_len);
 err_lpm_add:
+	if (fib_type == PRESTERA_FIB_TYPE_UC_NH)
+		prestera_nexthop_group_put(sw, fib_node->info.nh_grp);
 err_nh_grp_get:
 	prestera_vr_put(sw, vr);
 err_vr_get:

drivers/net/ethernet/marvell/prestera/prestera_router_hw.h

@@ -31,6 +31,63 @@ struct prestera_ip_addr {
 		PRESTERA_IPV4 = 0,
 		PRESTERA_IPV6
 	} v;
+#define PRESTERA_IP_ADDR_PLEN(V) ((V) == PRESTERA_IPV4 ? 32 : \
+				  /* (V) == PRESTERA_IPV6 ? */ 128 /* : 0 */)
+};
+
+struct prestera_nh_neigh_key {
+	struct prestera_ip_addr addr;
+	/* Seems like rif is obsolete, because there is iface in info ?
+	 * Key can contain functional fields, or fields, which is used to
+	 * filter duplicate objects on logical level (before you pass it to
+	 * HW)... also key can be used to cover hardware restrictions.
+	 * In our case rif - is logical interface (even can be VLAN), which
+	 * is used in combination with IP address (which is also not related to
+	 * hardware nexthop) to provide logical compression of created nexthops.
+	 * You even can imagine, that rif+IPaddr is just cookie.
+	 */
+	/* struct prestera_rif *rif; */
+	/* Use just as cookie, to divide ARP domains (in order with addr) */
+	void *rif;
+};
+
+/* Used for hw call */
+struct prestera_neigh_info {
+	struct prestera_iface iface;
+	unsigned char ha[ETH_ALEN];
+	u8 connected; /* bool. indicate, if mac/oif valid */
+	u8 __pad[1];
+};
+
+/* Used to notify nh about neigh change */
+struct prestera_nh_neigh {
+	struct prestera_nh_neigh_key key;
+	struct prestera_neigh_info info;
+	struct rhash_head ht_node; /* node of prestera_vr */
+	struct list_head nexthop_group_list;
+};
+
+#define PRESTERA_NHGR_SIZE_MAX 4
+
+struct prestera_nexthop_group {
+	struct prestera_nexthop_group_key {
+		struct prestera_nh_neigh_key neigh[PRESTERA_NHGR_SIZE_MAX];
+	} key;
+	/* Store intermediate object here.
+	 * This prevent overhead kzalloc call.
+	 */
+	/* nh_neigh is used only to notify nexthop_group */
+	struct prestera_nh_neigh_head {
+		struct prestera_nexthop_group *this;
+		struct list_head head;
+		/* ptr to neigh is not necessary.
+		 * It used to prevent lookup of nh_neigh by key (n) on destroy
+		 */
+		struct prestera_nh_neigh *neigh;
+	} nh_neigh_head[PRESTERA_NHGR_SIZE_MAX];
+	struct rhash_head ht_node; /* node of prestera_vr */
+	refcount_t refcount;
+	u32 grp_id; /* hw */
 };
 
 struct prestera_fib_key {
@@ -44,12 +101,16 @@ struct prestera_fib_info {
 	struct list_head vr_node;
 	enum prestera_fib_type {
 		PRESTERA_FIB_TYPE_INVALID = 0,
+		/* must be pointer to nh_grp id */
+		PRESTERA_FIB_TYPE_UC_NH,
 		/* It can be connected route
 		 * and will be overlapped with neighbours
 		 */
 		PRESTERA_FIB_TYPE_TRAP,
 		PRESTERA_FIB_TYPE_DROP
 	} type;
+	/* Valid only if type = UC_NH*/
+	struct prestera_nexthop_group *nh_grp;
 };
 
 struct prestera_fib_node {
@@ -67,6 +128,18 @@ struct prestera_rif_entry *
 prestera_rif_entry_create(struct prestera_switch *sw,
 			  struct prestera_rif_entry_key *k,
 			  u32 tb_id, const unsigned char *addr);
+struct prestera_nh_neigh *
+prestera_nh_neigh_find(struct prestera_switch *sw,
+		       struct prestera_nh_neigh_key *key);
+struct prestera_nh_neigh *
+prestera_nh_neigh_get(struct prestera_switch *sw,
+		      struct prestera_nh_neigh_key *key);
+void prestera_nh_neigh_put(struct prestera_switch *sw,
+			   struct prestera_nh_neigh *neigh);
+int prestera_nh_neigh_set(struct prestera_switch *sw,
+			  struct prestera_nh_neigh *neigh);
+bool prestera_nh_neigh_util_hw_state(struct prestera_switch *sw,
+				     struct prestera_nh_neigh *nh_neigh);
 struct prestera_fib_node *prestera_fib_node_find(struct prestera_switch *sw,
 						 struct prestera_fib_key *key);
 void prestera_fib_node_destroy(struct prestera_switch *sw,
@@ -74,7 +147,8 @@ void prestera_fib_node_destroy(struct prestera_switch *sw,
 struct prestera_fib_node *
 prestera_fib_node_create(struct prestera_switch *sw,
 			 struct prestera_fib_key *key,
-			 enum prestera_fib_type fib_type);
+			 enum prestera_fib_type fib_type,
+			 struct prestera_nexthop_group_key *nh_grp_key);
 int prestera_router_hw_init(struct prestera_switch *sw);
 void prestera_router_hw_fini(struct prestera_switch *sw);