--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -246,6 +246,9 @@ static const char * const mtk_clks_sourc
 	"top_netsys_warp_sel",
 };
 
+struct net_device *(*mtk_get_tnl_dev)(int tnl_idx) = NULL;
+EXPORT_SYMBOL(mtk_get_tnl_dev);
+
 void mtk_w32(struct mtk_eth *eth, u32 val, unsigned reg)
 {
 	__raw_writel(val, eth->base + reg);
@@ -2172,6 +2175,7 @@ static int mtk_poll_rx(struct napi_struc
 	u64 addr64 = 0;
 	u8 *data, *new_data;
 	struct mtk_rx_dma_v2 *rxd, trxd;
+	int tnl_idx = 0;
 	int done = 0;
 
 	if (unlikely(!ring))
@@ -2215,11 +2219,20 @@ static int mtk_poll_rx(struct napi_struc
 				      0 : RX_DMA_GET_SPORT(trxd.rxd4) - 1;
 		}
 
-		if (unlikely(mac < 0 || mac >= MTK_MAC_COUNT ||
-			     !eth->netdev[mac]))
-			goto release_desc;
+		tnl_idx = RX_DMA_GET_TOPS_CRSN(trxd.rxd6);
+		if (mtk_get_tnl_dev && tnl_idx) {
+			netdev = mtk_get_tnl_dev(tnl_idx);
+			if (unlikely(IS_ERR(netdev)))
+				netdev = NULL;
+		}
 
-		netdev = eth->netdev[mac];
+		if (!netdev) {
+			if (unlikely(mac < 0 || mac >= MTK_MAC_COUNT ||
+				     !eth->netdev[mac]))
+				goto release_desc;
+
+			netdev = eth->netdev[mac];
+		}
 
 		if (unlikely(test_bit(MTK_RESETTING, &eth->state)))
 			goto release_desc;
@@ -2304,6 +2317,8 @@ static int mtk_poll_rx(struct napi_struc
 		skb_hnat_alg(skb) = 0;
 		skb_hnat_filled(skb) = 0;
 		skb_hnat_magic_tag(skb) = HNAT_MAGIC_TAG;
+		skb_hnat_set_tops(skb, 0);
+		skb_hnat_set_is_decap(skb, 0);
 
 		if (skb_hnat_reason(skb) == HIT_BIND_FORCE_TO_CPU) {
 			trace_printk("[%s] reason=0x%x(force to CPU) from WAN to Ext\n",
--- a/drivers/net/ethernet/mediatek/mtk_hnat/hnat.c
+++ b/drivers/net/ethernet/mediatek/mtk_hnat/hnat.c
@@ -43,6 +43,12 @@ void (*ppe_dev_register_hook)(struct net
 EXPORT_SYMBOL(ppe_dev_register_hook);
 void (*ppe_dev_unregister_hook)(struct net_device *dev) = NULL;
 EXPORT_SYMBOL(ppe_dev_unregister_hook);
+int (*mtk_tnl_encap_offload)(struct sk_buff *skb) = NULL;
+EXPORT_SYMBOL(mtk_tnl_encap_offload);
+int (*mtk_tnl_decap_offload)(struct sk_buff *skb) = NULL;
+EXPORT_SYMBOL(mtk_tnl_decap_offload);
+bool (*mtk_tnl_decap_offloadable)(struct sk_buff *skb) = NULL;
+EXPORT_SYMBOL(mtk_tnl_decap_offloadable);
 
 static void hnat_sma_build_entry(struct timer_list *t)
 {
@@ -53,6 +59,16 @@ static void hnat_sma_build_entry(struct
 			     SMA, SMA_FWD_CPU_BUILD_ENTRY);
 }
 
+struct foe_entry *hnat_get_foe_entry(u32 ppe_id, u32 index)
+{
+	if (index == 0x7fff || index >= hnat_priv->foe_etry_num
+	    || ppe_id >= CFG_PPE_NUM)
+		return ERR_PTR(-EINVAL);
+
+	return &hnat_priv->foe_table_cpu[ppe_id][index];
+}
+EXPORT_SYMBOL(hnat_get_foe_entry);
+
 void hnat_cache_ebl(int enable)
 {
 	int i;
@@ -63,6 +79,7 @@ void hnat_cache_ebl(int enable)
 		cr_set_field(hnat_priv->ppe_base[i] + PPE_CAH_CTRL, CAH_EN, enable);
 	}
 }
+EXPORT_SYMBOL(hnat_cache_ebl);
 
 static void hnat_reset_timestamp(struct timer_list *t)
 {
--- a/drivers/net/ethernet/mediatek/mtk_hnat/hnat.h
+++ b/drivers/net/ethernet/mediatek/mtk_hnat/hnat.h
@@ -1087,6 +1087,8 @@ enum FoeIpAct {
 #define NR_WDMA1_PORT 9
 #define NR_WDMA2_PORT 13
 #define NR_GMAC3_PORT 15
+#define NR_TDMA_TPORT 4
+#define NR_TDMA_QDMA_TPORT 5
 #define LAN_DEV_NAME hnat_priv->lan
 #define LAN2_DEV_NAME hnat_priv->lan2
 #define IS_WAN(dev)                                                            \
@@ -1210,6 +1212,8 @@ static inline bool hnat_dsa_is_enable(st
 }
 #endif
 
+struct foe_entry *hnat_get_foe_entry(u32 ppe_id, u32 index);
+
 void hnat_deinit_debugfs(struct mtk_hnat *h);
 int hnat_init_debugfs(struct mtk_hnat *h);
 int hnat_register_nf_hooks(void);
@@ -1226,6 +1230,9 @@ extern int qos_ul_toggle;
 extern int hook_toggle;
 extern int mape_toggle;
 extern int qos_toggle;
+extern int (*mtk_tnl_encap_offload)(struct sk_buff *skb);
+extern int (*mtk_tnl_decap_offload)(struct sk_buff *skb);
+extern bool (*mtk_tnl_decap_offloadable)(struct sk_buff *skb);
 
 int ext_if_add(struct extdev_entry *ext_entry);
 int ext_if_del(struct extdev_entry *ext_entry);
--- a/drivers/net/ethernet/mediatek/mtk_hnat/hnat_nf_hook.c
+++ b/drivers/net/ethernet/mediatek/mtk_hnat/hnat_nf_hook.c
@@ -726,10 +726,14 @@ static unsigned int is_ppe_support_type(
 	case ETH_P_IP:
 		iph = ip_hdr(skb);
 
-		/* do not accelerate non tcp/udp traffic */
-		if ((iph->protocol == IPPROTO_TCP) ||
+		if (mtk_tnl_decap_offloadable && mtk_tnl_decap_offloadable(skb)) {
+			/* tunnel protocol is offloadable */
+			skb_hnat_set_is_decap(skb, 1);
+			return 1;
+		} else if ((iph->protocol == IPPROTO_TCP) ||
 		    (iph->protocol == IPPROTO_UDP) ||
 		    (iph->protocol == IPPROTO_IPV6)) {
+			/* do not accelerate non tcp/udp traffic */
 			return 1;
 		}
 
@@ -846,6 +850,13 @@ mtk_hnat_ipv4_nf_pre_routing(void *priv,
 
 	hnat_set_head_frags(state, skb, -1, hnat_set_iif);
 
+	if (skb_hnat_tops(skb) && skb_hnat_is_decap(skb)
+	    && is_magic_tag_valid(skb)
+	    && skb_hnat_iface(skb) == FOE_MAGIC_GE_VIRTUAL
+	    && mtk_tnl_decap_offload && mtk_tnl_decap_offload(skb)) {
+		return NF_ACCEPT;
+	}
+
 	/*
 	 * Avoid mistakenly binding of outer IP, ports in SW L2TP decap flow.
 	 * In pre-routing, if dev is virtual iface, TOPS module is not loaded,
@@ -922,6 +933,13 @@ mtk_hnat_br_nf_local_in(void *priv, stru
 
 	hnat_set_head_frags(state, skb, -1, hnat_set_iif);
 
+	if (skb_hnat_tops(skb) && skb_hnat_is_decap(skb)
+	    && is_magic_tag_valid(skb)
+	    && skb_hnat_iface(skb) == FOE_MAGIC_GE_VIRTUAL
+	    && mtk_tnl_decap_offload && mtk_tnl_decap_offload(skb)) {
+		return NF_ACCEPT;
+	}
+
 	pre_routing_print(skb, state->in, state->out, __func__);
 
 	if (unlikely(debug_level >= 7)) {
@@ -1074,9 +1092,22 @@ static unsigned int hnat_ipv4_get_nextho
 		return -1;
 	}
 
+	/*
+	 * if this packet is a tunnel packet and is about to construct
+	 * outer header, we must update its outer mac header pointer
+	 * before filling outer mac or it may screw up inner mac
+	 */
+	if (skb_hnat_tops(skb) && skb_hnat_is_encap(skb)) {
+		skb_push(skb, sizeof(struct ethhdr));
+		skb_reset_mac_header(skb);
+	}
+
 	memcpy(eth_hdr(skb)->h_dest, neigh->ha, ETH_ALEN);
 	memcpy(eth_hdr(skb)->h_source, out->dev_addr, ETH_ALEN);
 
+	if (skb_hnat_tops(skb) && skb_hnat_is_encap(skb))
+		skb_pull(skb, sizeof(struct ethhdr));
+
 	rcu_read_unlock_bh();
 
 	return 0;
@@ -1202,6 +1233,72 @@ static struct ethhdr *get_ipv6_ipip_ethh
 	return eth;
 }
 
+static inline void hnat_get_filled_unbind_entry(struct sk_buff *skb,
+						struct foe_entry *entry)
+{
+	if (unlikely(!skb || !entry))
+		return;
+
+	memcpy(entry,
+	       &hnat_priv->foe_table_cpu[skb_hnat_ppe(skb)][skb_hnat_entry(skb)],
+	       sizeof(*entry));
+}
+
+static inline void hnat_qos_tnl(u32 id, const struct net_device *dev)
+{
+	u32 cfg;
+	u32 max_man = 0;
+	u32 max_exp = 0;
+	const struct mtk_mac *mac;
+
+	if (!dev)
+		return;
+	mac = netdev_priv(dev);
+
+	switch (mac->speed) {
+	case SPEED_100:
+	case SPEED_1000:
+	case SPEED_2500:
+	case SPEED_5000:
+	case SPEED_10000:
+		max_man = mac->speed / SPEED_100;
+		max_exp = 5;
+		break;
+	default:
+		return;
+	}
+
+	cfg = QTX_SCH_MIN_RATE_EN | QTX_SCH_MAX_RATE_EN;
+	cfg |= (1 << QTX_SCH_MIN_RATE_MAN_OFFSET) |
+	       (4 << QTX_SCH_MIN_RATE_EXP_OFFSET) |
+	       (max_man << QTX_SCH_MAX_RATE_MAN_OFFSET) |
+	       (max_exp << QTX_SCH_MAX_RATE_EXP_OFFSET) |
+	       (4 << QTX_SCH_MAX_RATE_WGHT_OFFSET);
+	writel(cfg, hnat_priv->fe_base + QTX_SCH(id % NUM_OF_Q_PER_PAGE));
+}
+
+static inline void hnat_fill_offload_engine_entry(struct sk_buff *skb,
+						  struct foe_entry *entry,
+						  const struct net_device *dev)
+{
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+	if (skb_hnat_tops(skb) && skb_hnat_is_encap(skb)) {
+		/*
+		 * if skb_hnat_tops(skb) is setup for encapsulation,
+		 * we fill in hnat tport and tops_entry for tunnel encapsulation
+		 * offloading
+		 */
+		entry->ipv4_hnapt.tport_id = NR_TDMA_QDMA_TPORT;
+		entry->ipv4_hnapt.tops_entry = skb_hnat_tops(skb);
+	} else {
+		return;
+	}
+
+	entry->ipv4_hnapt.iblk2.qid = 12; /* offload engine use QID 12 */
+	hnat_qos_tnl(12, dev); /* set rate limit to line rate */
+#endif /* defined(CONFIG_MEDIATEK_NETSYS_V3) */
+}
+
 static unsigned int skb_to_hnat_info(struct sk_buff *skb,
 				     const struct net_device *dev,
 				     struct foe_entry *foe,
@@ -1238,6 +1335,11 @@ static unsigned int skb_to_hnat_info(str
 	if (whnat && is_hnat_pre_filled(foe))
 		return 0;
 
+	if (skb_hnat_tops(skb) && !(hw_path->flags & FLOW_OFFLOAD_PATH_TNL)) {
+		hnat_get_filled_unbind_entry(skb, &entry);
+		goto hnat_entry_bind;
+	}
+
 	entry.bfib1.pkt_type = foe->udib1.pkt_type; /* Get packte type state*/
 	entry.bfib1.state = foe->udib1.state;
 
@@ -1248,6 +1350,7 @@ static unsigned int skb_to_hnat_info(str
 	switch (ntohs(eth->h_proto)) {
 	case ETH_P_IP:
 		iph = ip_hdr(skb);
+
 		switch (iph->protocol) {
 		case IPPROTO_UDP:
 			udp = 1;
@@ -1629,6 +1732,10 @@ static unsigned int skb_to_hnat_info(str
 	/* Fill Layer2 Info.*/
 	entry = ppe_fill_L2_info(eth, entry, hw_path);
 
+	if (skb_hnat_tops(skb) && hw_path->flags & FLOW_OFFLOAD_PATH_TNL)
+		goto hnat_entry_skip_bind;
+
+hnat_entry_bind:
 	/* Fill Info Blk*/
 	entry = ppe_fill_info_blk(eth, entry, hw_path);
 
@@ -1827,7 +1934,20 @@ static unsigned int skb_to_hnat_info(str
 			entry.ipv6_5t_route.act_dp |= UDF_HNAT_PRE_FILLED;
 	}
 
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+	hnat_fill_offload_engine_entry(skb, &entry, dev);
+#endif
+
+hnat_entry_skip_bind:
 	wmb();
+
+	/*
+	 * final check before we write BIND info.
+	 * If this entry is already bound, we should not modify it right now
+	 */
+	if (entry_hnat_is_bound(foe))
+		return 0;
+
 	memcpy(foe, &entry, sizeof(entry));
 	/*reset statistic for this entry*/
 	if (hnat_priv->data->per_flow_accounting &&
@@ -1880,6 +2000,7 @@ int mtk_sw_nat_hook_tx(struct sk_buff *s
 		return NF_ACCEPT;
 
 	eth = eth_hdr(skb);
+
 	memcpy(&bfib1_tx, &entry->bfib1, sizeof(entry->bfib1));
 
 	/*not bind multicast if PPE mcast not enable*/
@@ -1899,6 +2020,12 @@ int mtk_sw_nat_hook_tx(struct sk_buff *s
 	switch ((int)bfib1_tx.pkt_type) {
 	case IPV4_HNAPT:
 	case IPV4_HNAT:
+		/*
+		 * skip if packet is an encap tnl packet or it may
+		 * screw up inner mac header
+		 */
+		if (skb_hnat_tops(skb) && skb_hnat_is_encap(skb))
+			break;
 		entry->ipv4_hnapt.smac_hi = swab32(*((u32 *)eth->h_source));
 		entry->ipv4_hnapt.smac_lo = swab16(*((u16 *)&eth->h_source[4]));
 		break;
@@ -2060,6 +2187,10 @@ int mtk_sw_nat_hook_tx(struct sk_buff *s
 		entry->ipv6_5t_route.iblk2.dp = gmac_no;
 	}
 
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+	hnat_fill_offload_engine_entry(skb, entry, NULL);
+#endif
+
 	bfib1_tx.ttl = 1;
 	bfib1_tx.state = BIND;
 	wmb();
@@ -2081,6 +2212,7 @@ int mtk_sw_nat_hook_rx(struct sk_buff *s
 	}
 
 	skb_hnat_alg(skb) = 0;
+	skb_hnat_set_tops(skb, 0);
 	skb_hnat_magic_tag(skb) = HNAT_MAGIC_TAG;
 
 	if (skb_hnat_iface(skb) == FOE_MAGIC_WED0)
@@ -2529,6 +2661,7 @@ static unsigned int mtk_hnat_nf_post_rou
 	struct flow_offload_hw_path hw_path = { .dev = (struct net_device*)out,
 						.virt_dev = (struct net_device*)out };
 	const struct net_device *arp_dev = out;
+	bool is_virt_dev = false;
 
 	if (xlat_toggle && !mtk_464xlat_post_process(skb, out))
 		return 0;
@@ -2549,10 +2682,18 @@ static unsigned int mtk_hnat_nf_post_rou
 
 	if (out->netdev_ops->ndo_flow_offload_check) {
 		out->netdev_ops->ndo_flow_offload_check(&hw_path);
+
 		out = (IS_GMAC1_MODE) ? hw_path.virt_dev : hw_path.dev;
+		if (hw_path.flags & FLOW_OFFLOAD_PATH_TNL && mtk_tnl_encap_offload)
+			skb_hnat_set_tops(skb, hw_path.tnl_type + 1);
 	}
 
 	if (!IS_LAN_GRP(out) && !IS_WAN(out) && !IS_EXT(out))
+		is_virt_dev = true;
+
+	if (is_virt_dev
+	    && !(skb_hnat_tops(skb) && skb_hnat_is_encap(skb)
+		 && (hw_path.flags & FLOW_OFFLOAD_PATH_TNL)))
 		return 0;
 
 	trace_printk("[%s] case hit, %x-->%s, reason=%x\n", __func__,
@@ -2572,9 +2713,18 @@ static unsigned int mtk_hnat_nf_post_rou
 		if (fn && !mtk_hnat_accel_type(skb))
 			break;
 
-		if (fn && fn(skb, arp_dev, &hw_path))
+		if (!is_virt_dev && fn && fn(skb, arp_dev, &hw_path))
 			break;
 
+		/* skb_hnat_tops(skb) is updated in mtk_tnl_offload() */
+		if (skb_hnat_tops(skb)) {
+			if (skb_hnat_is_encap(skb) && !is_virt_dev
+			    && mtk_tnl_encap_offload && mtk_tnl_encap_offload(skb))
+				break;
+			if (skb_hnat_is_decap(skb))
+				break;
+		}
+
 		skb_to_hnat_info(skb, out, entry, &hw_path);
 		break;
 	case HIT_BIND_KEEPALIVE_DUP_OLD_HDR:
@@ -2845,7 +2995,7 @@ mtk_hnat_ipv4_nf_local_out(void *priv, s
 	if (iph->protocol == IPPROTO_IPV6) {
 		entry->udib1.pkt_type = IPV6_6RD;
 		hnat_set_head_frags(state, skb, 0, hnat_set_alg);
-	} else {
+	} else if (!skb_hnat_tops(skb)) {
 		hnat_set_head_frags(state, skb, 1, hnat_set_alg);
 	}
 
--- a/drivers/net/ethernet/mediatek/mtk_hnat/nf_hnat_mtk.h
+++ b/drivers/net/ethernet/mediatek/mtk_hnat/nf_hnat_mtk.h
@@ -44,7 +44,9 @@ struct hnat_desc {
 	u32 is_sp : 1;
 	u32 hf : 1;
 	u32 amsdu : 1;
-	u32 resv3 : 19;
+	u32 tops : 6;
+	u32 is_decap : 1;
+	u32 resv3 : 12;
 	u32 magic_tag_protect : 16;
 } __packed;
 #elif defined(CONFIG_MEDIATEK_NETSYS_RX_V2)
@@ -91,6 +93,19 @@ struct hnat_desc {
 	((((skb_headroom(skb) >= FOE_INFO_LEN) ? 1 : 0)))
 
 #define skb_hnat_info(skb) ((struct hnat_desc *)(skb->head))
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+#define skb_hnat_tops(skb) (((struct hnat_desc *)((skb)->head))->tops)
+#define skb_hnat_is_decap(skb) (((struct hnat_desc *)((skb)->head))->is_decap)
+#define skb_hnat_is_encap(skb) (!skb_hnat_is_decap(skb))
+#define skb_hnat_set_tops(skb, tops) ((skb_hnat_tops(skb)) = (tops))
+#define skb_hnat_set_is_decap(skb, is_decap) ((skb_hnat_is_decap(skb)) = (is_decap))
+#else /* !defined(CONFIG_MEDIATEK_NETSYS_V3) */
+#define skb_hnat_tops(skb) (0)
+#define skb_hnat_is_decap(skb) (0)
+#define skb_hnat_is_encap(skb) (0)
+#define skb_hnat_set_tops(skb, tops)
+#define skb_hnat_set_is_decap(skb, is_decap)
+#endif /* defined(CONFIG_MEDIATEK_NETSYS_V3) */
 #define skb_hnat_magic(skb) (((struct hnat_desc *)(skb->head))->magic)
 #define skb_hnat_reason(skb) (((struct hnat_desc *)(skb->head))->crsn)
 #define skb_hnat_entry(skb) (((struct hnat_desc *)(skb->head))->entry)
--- a/include/net/netfilter/nf_flow_table.h
+++ b/include/net/netfilter/nf_flow_table.h
@@ -98,10 +98,22 @@ struct flow_offload {
 #define FLOW_OFFLOAD_PATH_6RD		BIT(5)
 #define FLOW_OFFLOAD_PATH_TNL		BIT(6)
 
+enum flow_offload_tnl {
+	FLOW_OFFLOAD_TNL_GRETAP,
+	FLOW_OFFLOAD_TNL_PPTP,
+	FLOW_OFFLOAD_TNL_IP_L2TP,
+	FLOW_OFFLOAD_TNL_UDP_L2TP_CTRL,
+	FLOW_OFFLOAD_TNL_UDP_L2TP_DATA,
+	FLOW_OFFLOAD_VXLAN,
+	FLOW_OFFLOAD_NATT,
+	__FLOW_OFFLOAD_MAX,
+};
+
 struct flow_offload_hw_path {
 	struct net_device *dev;
 	struct net_device *virt_dev;
 	u32 flags;
+	u32 tnl_type;
 
 	u8 eth_src[ETH_ALEN];
 	u8 eth_dest[ETH_ALEN];
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.h
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
@@ -1874,6 +1874,9 @@ extern const struct of_device_id of_mtk_
 extern u32 mtk_hwlro_stats_ebl;
 extern u32 dbg_show_level;
 
+/* tunnel offload related */
+extern struct net_device *(*mtk_get_tnl_dev)(int tnl_idx);
+
 /* read the hardware status register */
 void mtk_stats_update_mac(struct mtk_mac *mac);