[PATCH net-next v7 0/9] net: marvell: prestera: add nexthop routes offloading

[PATCH net-next v7 0/9] net: marvell: prestera: add nexthop routes offloading

[Yevhen Orlov]

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,
  747c14307214 ("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

Yevhen Orlov (9):
  net: marvell: prestera: Add router nexthops ABI
  net: marvell: prestera: Add cleanup of allocated fib_nodes
  net: marvell: prestera: Add strict cleanup of fib arbiter
  net: marvell: prestera: add delayed wq and flush wq on deinit
  net: marvell: prestera: Add length macros for prestera_ip_addr
  net: marvell: prestera: Add heplers to interact with fib_notifier_info
  net: marvell: prestera: add stub handler neighbour events
  net: marvell: prestera: Add neighbour cache accounting
  net: marvell: prestera: Propagate nh state from hw to kernel

 .../net/ethernet/marvell/prestera/prestera.h  |   12 +
 .../ethernet/marvell/prestera/prestera_hw.c   |  130 ++
 .../ethernet/marvell/prestera/prestera_hw.h   |   11 +
 .../ethernet/marvell/prestera/prestera_main.c |   11 +
 .../marvell/prestera/prestera_router.c        | 1119 ++++++++++++++++-
 .../marvell/prestera/prestera_router_hw.c     |  366 +++++-
 .../marvell/prestera/prestera_router_hw.h     |   76 +-
 7 files changed, 1683 insertions(+), 42 deletions(-)

-- 
2.17.1

[PATCH net-next v7 1/9] net: marvell: prestera: Add router nexthops ABI

[Yevhen Orlov]

- Add functions to allocate/delete/set nexthop group
  - NOTE: non-ECMP nexthop is nexthop group with allocated size = 1
- Add function to read state of HW nh (if packets going through it)

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>
---
 .../net/ethernet/marvell/prestera/prestera.h  |   5 +
 .../ethernet/marvell/prestera/prestera_hw.c   | 130 +++++++
 .../ethernet/marvell/prestera/prestera_hw.h   |  11 +
 .../marvell/prestera/prestera_router.c        |  16 +-
 .../marvell/prestera/prestera_router_hw.c     | 354 +++++++++++++++++-
 .../marvell/prestera/prestera_router_hw.h     |  74 +++-
 6 files changed, 582 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera.h b/drivers/net/ethernet/marvell/prestera/prestera.h
index e5a4381a88b3..903e2e13e687 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera.h
@@ -306,17 +306,22 @@ 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_fib_cache_ht;
 	struct notifier_block inetaddr_nb;
 	struct notifier_block inetaddr_valid_nb;
 	struct notifier_block fib_nb;
+	u8 *nhgrp_hw_state_cache; /* Bitmap cached hw state of nhs */
+	unsigned long nhgrp_hw_cache_kick; /* jiffies */
 };
 
 struct prestera_rxtx_params {
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_hw.c b/drivers/net/ethernet/marvell/prestera/prestera_hw.c
index 5803a28050e1..fc6f7d2746e8 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_hw.c
+++ b/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)
 {
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_hw.h b/drivers/net/ethernet/marvell/prestera/prestera_hw.h
index 21078a2256b2..0a929279e1ce 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_hw.h
+++ b/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,
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index 58f4e44d5ad7..a8548b9f9cf1 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -191,7 +191,7 @@ 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, NULL);
 
 	if (!fib_node) {
 		dev_err(sw->dev->dev, "fib_node=NULL %pI4n/%d kern_tb_id = %d",
@@ -220,6 +220,8 @@ static int __prestera_k_arb_fc_apply(struct prestera_switch *sw,
 	}
 
 	switch (fc->lpm_info.fib_type) {
+	case PRESTERA_FIB_TYPE_UC_NH:
+		break;
 	case PRESTERA_FIB_TYPE_TRAP:
 		__prestera_k_arb_fib_lpm_offload_set(sw, fc, false,
 						     false, fc->reachable);
@@ -537,7 +539,7 @@ static int __prestera_router_fib_event(struct notifier_block *nb,
 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 +557,13 @@ int prestera_router_init(struct prestera_switch *sw)
 	if (err)
 		goto err_kern_fib_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;
+	}
+
 	router->inetaddr_valid_nb.notifier_call = __prestera_inetaddr_valid_cb;
 	err = register_inetaddr_validator_notifier(&router->inetaddr_valid_nb);
 	if (err)
@@ -578,6 +587,8 @@ int prestera_router_init(struct prestera_switch *sw)
 err_register_inetaddr_notifier:
 	unregister_inetaddr_validator_notifier(&router->inetaddr_valid_nb);
 err_register_inetaddr_validator_notifier:
+	kfree(router->nhgrp_hw_state_cache);
+err_nh_state_cache_alloc:
 	rhashtable_destroy(&router->kern_fib_cache_ht);
 err_kern_fib_cache_ht_init:
 	prestera_router_hw_fini(sw);
@@ -591,6 +602,7 @@ void prestera_router_fini(struct prestera_switch *sw)
 	unregister_fib_notifier(&init_net, &sw->router->fib_nb);
 	unregister_inetaddr_notifier(&sw->router->inetaddr_nb);
 	unregister_inetaddr_validator_notifier(&sw->router->inetaddr_valid_nb);
+	kfree(sw->router->nhgrp_hw_state_cache);
 	rhashtable_destroy(&sw->router->kern_fib_cache_ht);
 	prestera_router_hw_fini(sw);
 	kfree(sw->router);
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c
index 5b0cf3be9a9e..db9d2e9d9904 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c
+++ b/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 +39,44 @@ 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);
+
+/* 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)
@@ -43,7 +85,13 @@ 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;
 }
 
@@ -52,6 +100,8 @@ void prestera_router_hw_fini(struct prestera_switch *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 +282,286 @@ 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 +581,9 @@ 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:
@@ -275,7 +608,8 @@ 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)
 {
 	struct prestera_fib_node *fib_node;
 	u32 grp_id;
@@ -302,6 +636,14 @@ 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 +665,8 @@ 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:
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h
index 67dbb49c8bd4..43bad23f38ec 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h
@@ -33,6 +33,61 @@ struct prestera_ip_addr {
 	} v;
 };
 
+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 {
 	struct prestera_ip_addr addr;
 	u32 prefix_len;
@@ -44,12 +99,16 @@ 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 +126,18 @@ 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 +145,8 @@ 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);
 
-- 
2.17.1

[PATCH net-next v7 2/9] net: marvell: prestera: Add cleanup of allocated fib_nodes

[Yevhen Orlov]

Do explicity cleanup on router_hw_fini, to ensure, that all allocated
objects cleaned. This will be used in cases,
when upper layer (cache) is not mapped to router_hw layer.

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>
---
 .../ethernet/marvell/prestera/prestera_router_hw.c   | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c
index db9d2e9d9904..4f65df0ae5e8 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.c
@@ -56,6 +56,7 @@ static int prestera_nexthop_group_set(struct prestera_switch *sw,
 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)
@@ -97,6 +98,8 @@ int prestera_router_hw_init(struct prestera_switch *sw)
 
 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);
@@ -605,6 +608,15 @@ 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,
-- 
2.17.1

[PATCH net-next v7 3/9] net: marvell: prestera: Add strict cleanup of fib arbiter

[Yevhen Orlov]

This will, ensure, that there is no more, preciously allocated fib_cache
entries left after deinit.
Will be used to free allocated resources of nexthop routes, that points
to "not our" port (e.g. eth0).

Signed-off-by: Yevhen Orlov <yevhen.orlov@plvision.eu>
---
 .../marvell/prestera/prestera_router.c        | 42 ++++++++++++++++++-
 1 file changed, 41 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index a8548b9f9cf1..b4fd8276bbce 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -72,14 +72,21 @@ 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)
+{
+	fib_info_put(fib_cache->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);
 }
 
@@ -336,6 +343,36 @@ prestera_k_arb_fib_evt(struct prestera_switch *sw,
 	return 0;
 }
 
+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);
+	/* 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);
+}
+
 static int __prestera_inetaddr_port_event(struct net_device *port_dev,
 					  unsigned long event,
 					  struct netlink_ext_ack *extack)
@@ -602,6 +639,9 @@ void prestera_router_fini(struct prestera_switch *sw)
 	unregister_fib_notifier(&init_net, &sw->router->fib_nb);
 	unregister_inetaddr_notifier(&sw->router->inetaddr_nb);
 	unregister_inetaddr_validator_notifier(&sw->router->inetaddr_valid_nb);
+
+	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);
-- 
2.17.1

[PATCH net-next v7 4/9] net: marvell: prestera: add delayed wq and flush wq on deinit

[Yevhen Orlov]

Flushing workqueues ensures, that no more pending works, related to just
unregistered or deinitialized notifiers. After that we can free memory.

Delayed wq will be used for neighbours in next patches.

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>
---
 drivers/net/ethernet/marvell/prestera/prestera.h      |  2 ++
 drivers/net/ethernet/marvell/prestera/prestera_main.c | 11 +++++++++++
 .../net/ethernet/marvell/prestera/prestera_router.c   |  1 +
 3 files changed, 14 insertions(+)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera.h b/drivers/net/ethernet/marvell/prestera/prestera.h
index 903e2e13e687..fe0d6001a6b6 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera.h
@@ -367,6 +367,8 @@ 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);
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_main.c b/drivers/net/ethernet/marvell/prestera/prestera_main.c
index 3956d6d5df3c..c0794603a733 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_main.c
+++ b/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);
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index b4fd8276bbce..9625c5870847 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -639,6 +639,7 @@ void prestera_router_fini(struct prestera_switch *sw)
 	unregister_fib_notifier(&init_net, &sw->router->fib_nb);
 	unregister_inetaddr_notifier(&sw->router->inetaddr_nb);
 	unregister_inetaddr_validator_notifier(&sw->router->inetaddr_valid_nb);
+	prestera_queue_drain();
 
 	prestera_k_arb_abort(sw);
 
-- 
2.17.1

[PATCH net-next v7 5/9] net: marvell: prestera: Add length macros for prestera_ip_addr

[Yevhen Orlov]

Add macros to determine IP address length (internal driver types).
This will be used in next patches for nexthops logic.

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>
---
 drivers/net/ethernet/marvell/prestera/prestera_router_hw.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h
index 43bad23f38ec..9ca97919c863 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router_hw.h
@@ -31,6 +31,8 @@ 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 {
-- 
2.17.1

[PATCH net-next v7 6/9] net: marvell: prestera: Add heplers to interact with fib_notifier_info

[Yevhen Orlov]

This will be used to implement nexthops related logic in next patches.
Also try to keep ipv4/6 abstraction to be able to reuse helpers for ipv6
in the future.

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>
---
 .../marvell/prestera/prestera_router.c        | 99 ++++++++++++-------
 1 file changed, 65 insertions(+), 34 deletions(-)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index 9625c5870847..607efd481782 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -7,6 +7,7 @@
 #include <net/inet_dscp.h>
 #include <net/switchdev.h>
 #include <linux/rhashtable.h>
+#include <net/nexthop.h>
 
 #include "prestera.h"
 #include "prestera_router_hw.h"
@@ -26,9 +27,10 @@ struct prestera_kern_fib_cache {
 	} 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;
+	union {
+		struct fib_notifier_info info; /* point to any of 4/6 */
+		struct fib_entry_notifier_info fen4_info;
+	};
 	bool reachable;
 };
 
@@ -51,15 +53,41 @@ 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 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;
+}
+
 static struct prestera_kern_fib_cache *
 prestera_kern_fib_cache_find(struct prestera_switch *sw,
 			     struct prestera_kern_fib_cache_key *key)
@@ -76,7 +104,7 @@ static void
 __prestera_kern_fib_cache_destruct(struct prestera_switch *sw,
 				   struct prestera_kern_fib_cache *fib_cache)
 {
-	fib_info_put(fib_cache->fi);
+	fib_info_put(fib_cache->fen4_info.fi);
 }
 
 static void
@@ -96,8 +124,10 @@ prestera_kern_fib_cache_destroy(struct prestera_switch *sw,
 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;
 
@@ -106,10 +136,8 @@ 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,
@@ -120,7 +148,7 @@ prestera_kern_fib_cache_create(struct prestera_switch *sw,
 	return fib_cache;
 
 err_ht_insert:
-	fib_info_put(fi);
+	fib_info_put(fen_info->fi);
 	kfree(fib_cache);
 err_kzalloc:
 	return NULL;
@@ -133,21 +161,25 @@ __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;
-
-	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);
+	case PRESTERA_IPV6:
+		/* TODO */
+		return;
+	}
 }
 
 static int
@@ -156,7 +188,7 @@ __prestera_pr_k_arb_fc_lpm_info_calc(struct prestera_switch *sw,
 {
 	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;
 		break;
@@ -283,14 +315,14 @@ __prestera_k_arb_util_fib_overlapped(struct prestera_switch *sw,
 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;
@@ -313,10 +345,7 @@ 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;
@@ -508,13 +537,15 @@ 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;
 
-- 
2.17.1

[PATCH net-next v7 7/9] net: marvell: prestera: add stub handler neighbour events

[Yevhen Orlov]

Actual handler will be added in next patches

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>
---
 .../net/ethernet/marvell/prestera/prestera.h  |  1 +
 .../marvell/prestera/prestera_router.c        | 59 +++++++++++++++++++
 2 files changed, 60 insertions(+)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera.h b/drivers/net/ethernet/marvell/prestera/prestera.h
index fe0d6001a6b6..2f2f80e7e358 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera.h
@@ -320,6 +320,7 @@ struct prestera_router {
 	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 */
 };
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index 607efd481782..d31dd1fe6633 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -8,6 +8,7 @@
 #include <net/switchdev.h>
 #include <linux/rhashtable.h>
 #include <net/nexthop.h>
+#include <net/netevent.h>
 
 #include "prestera.h"
 #include "prestera_router_hw.h"
@@ -604,6 +605,56 @@ 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 neighbour *n = net_work->n;
+
+	/* neigh - its not hw related object. It stored only in kernel. So... */
+	rtnl_lock();
+
+	/* TODO: handler */
+
+	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;
+}
+
 int prestera_router_init(struct prestera_switch *sw)
 {
 	struct prestera_router *router;
@@ -642,6 +693,11 @@ 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);
@@ -651,6 +707,8 @@ 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);
@@ -668,6 +726,7 @@ 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_queue_drain();
-- 
2.17.1

[PATCH net-next v7 8/9] net: marvell: prestera: Add neighbour cache accounting

[Yevhen Orlov]

Move forward and use new PRESTERA_FIB_TYPE_UC_NH to provide basic
nexthop routes support.
Provide deinitialization sequence for all created router objects.

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

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>
---
 .../net/ethernet/marvell/prestera/prestera.h  |   1 +
 .../marvell/prestera/prestera_router.c        | 799 +++++++++++++++++-
 2 files changed, 797 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera.h b/drivers/net/ethernet/marvell/prestera/prestera.h
index 2f2f80e7e358..540a36069b79 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera.h
@@ -316,6 +316,7 @@ struct prestera_router {
 	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;
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index d31dd1fe6633..af7d24390d2e 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -8,11 +8,30 @@
 #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"
 
+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;
@@ -25,9 +44,15 @@ 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 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;
@@ -35,6 +60,13 @@ struct prestera_kern_fib_cache {
 	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),
@@ -67,6 +99,268 @@ prestera_util_fen_info2fib_cache_key(struct fib_notifier_info *info,
 	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)
 {
@@ -89,6 +383,153 @@ prestera_kern_fib_info_type(struct fib_notifier_info *info)
 	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)
@@ -105,6 +546,17 @@ 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);
 }
 
@@ -119,6 +571,41 @@ prestera_kern_fib_cache_destroy(struct prestera_switch *sw,
 	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.
  */
@@ -146,6 +633,12 @@ 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:
@@ -155,6 +648,46 @@ prestera_kern_fib_cache_create(struct prestera_switch *sw,
 	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,
@@ -183,15 +716,187 @@ __prestera_k_arb_fib_lpm_offload_set(struct prestera_switch *sw,
 	}
 }
 
+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;
+
+		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 (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:
@@ -231,7 +936,8 @@ 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, NULL);
+					    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",
@@ -261,6 +967,8 @@ 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,
@@ -313,6 +1021,57 @@ __prestera_k_arb_util_fib_overlapped(struct prestera_switch *sw,
 	return rfc;
 }
 
+/* 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 */
@@ -370,9 +1129,30 @@ 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;
@@ -381,6 +1161,8 @@ static void __prestera_k_arb_abort_fib_ht_cb(void *ptr, void *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
 	 */
@@ -401,6 +1183,9 @@ static void prestera_k_arb_abort(struct prestera_switch *sw)
 	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,
@@ -615,12 +1400,13 @@ 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();
 
-	/* TODO: handler */
+	prestera_k_arb_n_evt(sw, n);
 
 	neigh_release(n);
 	rtnl_unlock();
@@ -676,6 +1462,11 @@ 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) {
@@ -715,6 +1506,8 @@ int prestera_router_init(struct prestera_switch *sw)
 err_register_inetaddr_validator_notifier:
 	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);
-- 
2.17.1

[PATCH net-next v7 9/9] net: marvell: prestera: Propagate nh state from hw to kernel

[Yevhen Orlov]

We poll nexthops in HW and call for each active nexthop appropriate
neighbour.

Also we provide implicity neighbour resolving.
For example, user have added nexthop route:
  # ip route add 5.5.5.5 via 1.1.1.2
But neighbour 1.1.1.2 doesn't exist. In this case we will try to call
neigh_event_send, even if there is no traffic.
This is useful, when you have add route, which will be used after some
time but with a lot of traffic (burst). So, we has prepared, offloaded
route in advance.

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>
---
 .../net/ethernet/marvell/prestera/prestera.h  |   3 +
 .../marvell/prestera/prestera_router.c        | 111 ++++++++++++++++++
 2 files changed, 114 insertions(+)

diff --git a/drivers/net/ethernet/marvell/prestera/prestera.h b/drivers/net/ethernet/marvell/prestera/prestera.h
index 540a36069b79..35554ee805cd 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera.h
+++ b/drivers/net/ethernet/marvell/prestera/prestera.h
@@ -324,6 +324,9 @@ struct prestera_router {
 	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 {
diff --git a/drivers/net/ethernet/marvell/prestera/prestera_router.c b/drivers/net/ethernet/marvell/prestera/prestera_router.c
index af7d24390d2e..4046be0e86ff 100644
--- a/drivers/net/ethernet/marvell/prestera/prestera_router.c
+++ b/drivers/net/ethernet/marvell/prestera/prestera_router.c
@@ -16,6 +16,9 @@
 #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;
@@ -32,6 +35,7 @@ struct prestera_kern_neigh_cache {
 	/* 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;
@@ -1021,6 +1025,78 @@ __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)
@@ -1441,6 +1517,34 @@ static int prestera_router_netevent_event(struct notifier_block *nb,
 	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;
@@ -1474,6 +1578,10 @@ int prestera_router_init(struct prestera_switch *sw)
 		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)
@@ -1504,6 +1612,8 @@ int prestera_router_init(struct prestera_switch *sw)
 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);
@@ -1522,6 +1632,7 @@ void prestera_router_fini(struct prestera_switch *sw)
 	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);
-- 
2.17.1

Re: [PATCH net-next v7 0/9] net: marvell: prestera: add nexthop routes offloading

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net-next.git (master)
by Jakub Kicinski <kuba@kernel.org>:

On Sat,  1 Oct 2022 12:34:08 +0300 you wrote:
> 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
> 
> [...]

Here is the summary with links:
  - [net-next,v7,1/9] net: marvell: prestera: Add router nexthops ABI
    https://git.kernel.org/netdev/net-next/c/0a23ae237171
  - [net-next,v7,2/9] net: marvell: prestera: Add cleanup of allocated fib_nodes
    https://git.kernel.org/netdev/net-next/c/1e7313e83ef7
  - [net-next,v7,3/9] net: marvell: prestera: Add strict cleanup of fib arbiter
    https://git.kernel.org/netdev/net-next/c/333fe4d033fa
  - [net-next,v7,4/9] net: marvell: prestera: add delayed wq and flush wq on deinit
    https://git.kernel.org/netdev/net-next/c/90b6f9c09851
  - [net-next,v7,5/9] net: marvell: prestera: Add length macros for prestera_ip_addr
    https://git.kernel.org/netdev/net-next/c/59b44ea8aa56
  - [net-next,v7,6/9] net: marvell: prestera: Add heplers to interact with fib_notifier_info
    https://git.kernel.org/netdev/net-next/c/04f24a1e6de6
  - [net-next,v7,7/9] net: marvell: prestera: add stub handler neighbour events
    https://git.kernel.org/netdev/net-next/c/8b1ef4911a41
  - [net-next,v7,8/9] net: marvell: prestera: Add neighbour cache accounting
    https://git.kernel.org/netdev/net-next/c/396b80cb5cc8
  - [net-next,v7,9/9] net: marvell: prestera: Propagate nh state from hw to kernel
    https://git.kernel.org/netdev/net-next/c/ae15ed6e40c9

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html