autobuild_mac80211_release/mt7988_mt7996_mac80211/target/linux/mediatek/patches-5.4/999-3019-mtk-wed-add-wed3-support.patch

From d7b7f03e5ab30b0420d0c97d6112dc3fa6e64d71 Mon Sep 17 00:00:00 2001
From: Sujuan Chen <sujuan.chen@mediatek.com>
Date: Mon, 18 Sep 2023 13:21:15 +0800
Subject: [PATCH 20/22] mtk:wed:add wed3 support

---
 arch/arm64/boot/dts/mediatek/mt7988.dtsi      |  152 ++-
 .../dts/mediatek/mt7988a-dsa-10g-spim-nor.dts |   16 +-
 .../dts/mediatek/mt7988d-dsa-10g-spim-nor.dts |   16 +-
 drivers/net/ethernet/mediatek/mtk_eth_soc.c   |    3 +-
 drivers/net/ethernet/mediatek/mtk_eth_soc.h   |    5 +-
 drivers/net/ethernet/mediatek/mtk_ppe.c       |   17 +-
 drivers/net/ethernet/mediatek/mtk_ppe.h       |    2 +-
 .../net/ethernet/mediatek/mtk_ppe_offload.c   |   13 +-
 drivers/net/ethernet/mediatek/mtk_wed.c       | 1165 +++++++++++++----
 drivers/net/ethernet/mediatek/mtk_wed.h       |   25 +-
 .../net/ethernet/mediatek/mtk_wed_debugfs.c   |  584 ++++++++-
 drivers/net/ethernet/mediatek/mtk_wed_mcu.c   |   13 +-
 drivers/net/ethernet/mediatek/mtk_wed_mcu.h   |    5 +-
 drivers/net/ethernet/mediatek/mtk_wed_regs.h  |  338 ++++-
 include/linux/netdevice.h                     |    7 +
 include/linux/soc/mediatek/mtk_wed.h          |   83 +-
 16 files changed, 2056 insertions(+), 388 deletions(-)
 mode change 100755 => 100644 drivers/net/ethernet/mediatek/mtk_ppe.c

diff --git a/arch/arm64/boot/dts/mediatek/mt7988.dtsi b/arch/arm64/boot/dts/mediatek/mt7988.dtsi
index 561450e..8995ea3 100644
--- a/arch/arm64/boot/dts/mediatek/mt7988.dtsi
+++ b/arch/arm64/boot/dts/mediatek/mt7988.dtsi
@@ -205,44 +205,49 @@
 		status = "disabled";
 	};
 
-	wed: wed@15010000 {
-		compatible = "mediatek,wed";
-		wed_num = <3>;
-		/* add this property for wed get the pci slot number. */
-		pci_slot_map = <0>, <1>, <2>;
-		reg = <0 0x15010000 0 0x2000>,
-		      <0 0x15012000 0 0x2000>,
-		      <0 0x15014000 0 0x2000>;
+	wed0: wed@15010000 {
+		compatible = "mediatek,mt7988-wed",
+			     "syscon";
+		reg = <0 0x15010000 0 0x2000>;
 		interrupt-parent = <&gic>;
-		interrupts = <GIC_SPI 205 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 206 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 270 IRQ_TYPE_LEVEL_HIGH>;
-	};
-
-	wed2: wed2@15012000 {
-		compatible = "mediatek,wed2";
-		wed_num = <3>;
-		/* add this property for wed get the pci slot number. */
-		reg = <0 0x15010000 0 0x2000>,
-		      <0 0x15012000 0 0x2000>,
-		      <0 0x15014000 0 0x2000>;
+		interrupts = <GIC_SPI 205 IRQ_TYPE_LEVEL_HIGH>;
+		mediatek,wed_pcie = <&wed_pcie>;
+		mediatek,ap2woccif = <&ap2woccif0>;
+		mediatek,wocpu_ilm = <&wocpu0_ilm>;
+		mediatek,wocpu_dlm = <&wocpu0_dlm>;
+		mediatek,wocpu_boot = <&cpu0_boot>;
+		mediatek,wocpu_emi = <&wocpu0_emi>;
+		mediatek,wocpu_data = <&wocpu_data>;
+	};
+
+	wed1: wed@15012000 {
+		compatible = "mediatek,mt7988-wed",
+                             "syscon";
+		reg = <0 0x15012000 0 0x2000>;
 		interrupt-parent = <&gic>;
-		interrupts = <GIC_SPI 205 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 206 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 270 IRQ_TYPE_LEVEL_HIGH>;
-	};
-
-	wed3: wed3@15014000 {
-		compatible = "mediatek,wed3";
-		wed_num = <3>;
-		/* add this property for wed get the pci slot number. */
-		reg = <0 0x15010000 0 0x2000>,
-		      <0 0x15012000 0 0x2000>,
-		      <0 0x15014000 0 0x2000>;
+		interrupts = <GIC_SPI 206 IRQ_TYPE_LEVEL_HIGH>;
+		mediatek,wed_pcie = <&wed_pcie>;
+		mediatek,ap2woccif = <&ap2woccif1>;
+		mediatek,wocpu_ilm = <&wocpu1_ilm>;
+		mediatek,wocpu_dlm = <&wocpu1_dlm>;
+		mediatek,wocpu_boot = <&cpu1_boot>;
+		mediatek,wocpu_emi = <&wocpu1_emi>;
+		mediatek,wocpu_data = <&wocpu_data>;
+	};
+
+	wed2: wed@15014000 {
+		compatible = "mediatek,mt7988-wed",
+                             "syscon";
+		reg = <0 0x15014000 0 0x2000>;
 		interrupt-parent = <&gic>;
-		interrupts = <GIC_SPI 205 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 206 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 270 IRQ_TYPE_LEVEL_HIGH>;
+		interrupts = <GIC_SPI 207 IRQ_TYPE_LEVEL_HIGH>;
+		mediatek,wed_pcie = <&wed_pcie>;
+		mediatek,ap2woccif = <&ap2woccif2>;
+		mediatek,wocpu_ilm = <&wocpu2_ilm>;
+		mediatek,wocpu_dlm = <&wocpu2_dlm>;
+		mediatek,wocpu_boot = <&cpu2_boot>;
+		mediatek,wocpu_emi = <&wocpu2_emi>;
+		mediatek,wocpu_data = <&wocpu_data>;
 	};
 
 	wdma: wdma@15104800 {
@@ -252,15 +257,25 @@
 		      <0 0x15105000 0 0x400>;
 	};
 
-	ap2woccif: ap2woccif@151A5000 {
-		compatible = "mediatek,ap2woccif";
-		reg = <0 0x151A5000 0 0x1000>,
-		      <0 0x152A5000 0 0x1000>,
-		      <0 0x153A5000 0 0x1000>;
+	ap2woccif0: ap2woccif@151A5000 {
+		compatible = "mediatek,ap2woccif", "syscon";
+		reg = <0 0x151A5000 0 0x1000>;
+		interrupt-parent = <&gic>;
+		interrupts = <GIC_SPI 211 IRQ_TYPE_LEVEL_HIGH>;
+	};
+
+	ap2woccif1: ap2woccif@152A5000 {
+		compatible = "mediatek,ap2woccif", "syscon";
+		reg = <0 0x152A5000 0 0x1000>;
 		interrupt-parent = <&gic>;
-		interrupts = <GIC_SPI 211 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 212 IRQ_TYPE_LEVEL_HIGH>,
-			     <GIC_SPI 272 IRQ_TYPE_LEVEL_HIGH>;
+		interrupts = <GIC_SPI 212 IRQ_TYPE_LEVEL_HIGH>;
+	};
+
+	ap2woccif2: ap2woccif@153A5000 {
+		compatible = "mediatek,ap2woccif", "syscon";
+		reg = <0 0x153A5000 0 0x1000>;
+		interrupt-parent = <&gic>;
+		interrupts = <GIC_SPI 272 IRQ_TYPE_LEVEL_HIGH>;
 	};
 
 	wocpu0_ilm: wocpu0_ilm@151E0000 {
@@ -268,31 +283,53 @@
 		reg = <0 0x151E0000 0 0x8000>;
 	};
 
-	wocpu1_ilm: wocpu1_ilm@152E0000 {
-		compatible = "mediatek,wocpu1_ilm";
+	wocpu1_ilm: wocpu_ilm@152E0000 {
+		compatible = "mediatek,wocpu_ilm";
 		reg = <0 0x152E0000 0 0x8000>;
 	};
 
-	wocpu2_ilm: wocpu2_ilm@153E0000 {
-		compatible = "mediatek,wocpu2_ilm";
-		reg = <0 0x153E0000 0 0x8000>;
+	wocpu2_ilm: wocpu_ilm@153E0000 {
+                compatible = "mediatek,wocpu_ilm";
+                reg = <0 0x153E0000 0 0x8000>;
+    };
+
+	wocpu0_dlm: wocpu_dlm@151E8000 {
+		compatible = "mediatek,wocpu_dlm";
+		reg = <0 0x151E8000 0 0x2000>;
+
+		resets = <&ethsysrst 0>;
+		reset-names = "wocpu_rst";
+	};
+
+	wocpu1_dlm: wocpu_dlm@0x152E8000 {
+		compatible = "mediatek,wocpu_dlm";
+		reg = <0 0x152E8000 0 0x2000>;
+
+		resets = <&ethsysrst 0>;
+		reset-names = "wocpu_rst";
 	};
 
-	wocpu_dlm: wocpu_dlm@151E8000 {
+	wocpu2_dlm: wocpu_dlm@0x153E8000 {
 		compatible = "mediatek,wocpu_dlm";
-		reg = <0 0x151E8000 0 0x2000>,
-		      <0 0x152E8000 0 0x2000>,
-		      <0 0x153E8000 0 0x2000>;
+		reg = <0 0x153E8000 0 0x2000>;
 
 		resets = <&ethsysrst 0>;
 		reset-names = "wocpu_rst";
 	};
 
-	cpu_boot: wocpu_boot@15194000 {
-		compatible = "mediatek,wocpu_boot";
-		reg = <0 0x15194000 0 0x1000>,
-		      <0 0x15294000 0 0x1000>,
-		      <0 0x15394000 0 0x1000>;
+	cpu0_boot: wocpu_boot@15194000 {
+		compatible = "mediatek,wocpu0_boot";
+		reg = <0 0x15194000 0 0x1000>;
+	};
+
+	cpu1_boot: wocpu_boot@15294000 {
+		compatible = "mediatek,wocpu1_boot";
+		reg = <0 0x15294000 0 0x1000>;
+	};
+
+	cpu2_boot: wocpu_boot@15394000 {
+		compatible = "mediatek,wocpu2_boot";
+		reg = <0 0x15394000 0 0x1000>;
 	};
 
 	reserved-memory {
@@ -902,6 +939,7 @@
 					 <&topckgen CK_TOP_CB_SGM_325M>;
 		mediatek,ethsys = <&ethsys>;
 		mediatek,sgmiisys = <&sgmiisys0>, <&sgmiisys1>;
+		mediatek,wed = <&wed0>, <&wed1>, <&wed2>;
 		mediatek,usxgmiisys = <&usxgmiisys0>, <&usxgmiisys1>;
 		mediatek,xfi_pextp = <&xfi_pextp0>, <&xfi_pextp1>;
 		mediatek,xfi_pll = <&xfi_pll>;
diff --git a/arch/arm64/boot/dts/mediatek/mt7988a-dsa-10g-spim-nor.dts b/arch/arm64/boot/dts/mediatek/mt7988a-dsa-10g-spim-nor.dts
index 70a7554..bed27b4 100644
--- a/arch/arm64/boot/dts/mediatek/mt7988a-dsa-10g-spim-nor.dts
+++ b/arch/arm64/boot/dts/mediatek/mt7988a-dsa-10g-spim-nor.dts
@@ -369,9 +369,23 @@
 	status = "okay";
 };
 
-&wed {
+&wed0 {
 	dy_txbm_enable = "true";
 	dy_txbm_budge = <8>;
 	txbm_init_sz = <10>;
 	status = "okay";
 };
+
+&wed1 {
+	dy_txbm_enable = "true";
+	dy_txbm_budge = <8>;
+	txbm_init_sz = <10>;
+	status = "okay";
+};
+
+&wed2 {
+	dy_txbm_enable = "true";
+	dy_txbm_budge = <8>;
+	txbm_init_sz = <10>;
+	status = "okay";
+};
\ No newline at end of file
diff --git a/arch/arm64/boot/dts/mediatek/mt7988d-dsa-10g-spim-nor.dts b/arch/arm64/boot/dts/mediatek/mt7988d-dsa-10g-spim-nor.dts
index e8e3a69..5dd481b 100644
--- a/arch/arm64/boot/dts/mediatek/mt7988d-dsa-10g-spim-nor.dts
+++ b/arch/arm64/boot/dts/mediatek/mt7988d-dsa-10g-spim-nor.dts
@@ -379,9 +379,23 @@
 	status = "okay";
 };
 
-&wed {
+&wed0 {
 	dy_txbm_enable = "true";
 	dy_txbm_budge = <8>;
 	txbm_init_sz = <10>;
 	status = "okay";
 };
+
+&wed1 {
+	dy_txbm_enable = "true";
+	dy_txbm_budge = <8>;
+	txbm_init_sz = <10>;
+	status = "okay";
+};
+
+&wed2 {
+	dy_txbm_enable = "true";
+	dy_txbm_budge = <8>;
+	txbm_init_sz = <10>;
+	status = "okay";
+};
\ No newline at end of file
diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
index 8bd526a..dea66d7 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -5095,7 +5095,8 @@ static int mtk_probe(struct platform_device *pdev)
 							  "mediatek,wed", i);
 		static const u32 wdma_regs[] = {
 			MTK_WDMA0_BASE,
-			MTK_WDMA1_BASE
+			MTK_WDMA1_BASE,
+			MTK_WDMA2_BASE
 		};
 		void __iomem *wdma;
 		u32 wdma_phy;
diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.h b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
index ee89b4c..8656b5f 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.h
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
@@ -613,9 +613,12 @@
 #define RX_DMA_SPORT_MASK       0x7
 #define RX_DMA_SPORT_MASK_V2    0xf
 
-#if defined(CONFIG_MEDIATEK_NETSYS_V2)
+#if defined(CONFIG_MEDIATEK_NETSYS_V2) || defined(CONFIG_MEDIATEK_NETSYS_V3)
 #define MTK_WDMA0_BASE		0x4800
 #define MTK_WDMA1_BASE		0x4c00
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+#define MTK_WDMA2_BASE		0x5000
+#endif
 #else
 #define MTK_WDMA0_BASE		0x2800
 #define MTK_WDMA1_BASE		0x2c00
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe.c b/drivers/net/ethernet/mediatek/mtk_ppe.c
old mode 100755
new mode 100644
index 384e811..eda23c2
--- a/drivers/net/ethernet/mediatek/mtk_ppe.c
+++ b/drivers/net/ethernet/mediatek/mtk_ppe.c
@@ -9,6 +9,7 @@
 #include <linux/if_ether.h>
 #include <linux/if_vlan.h>
 #include <net/dsa.h>
+#include <net/route.h>
 #include "mtk_eth_soc.h"
 #include "mtk_ppe.h"
 #include "mtk_ppe_regs.h"
@@ -396,7 +397,7 @@ int mtk_foe_entry_set_pppoe(struct mtk_foe_entry *entry, int sid)
 }
 
 int mtk_foe_entry_set_wdma(struct mtk_foe_entry *entry, int wdma_idx, int txq,
-			   int bss, int wcid)
+			   int bss, int wcid, bool amsdu_en)
 {
 	struct mtk_foe_mac_info *l2 = mtk_foe_entry_l2(entry);
 	u32 *ib2 = mtk_foe_entry_ib2(entry);
@@ -408,6 +409,9 @@ int mtk_foe_entry_set_wdma(struct mtk_foe_entry *entry, int wdma_idx, int txq,
 
 	l2->winfo = FIELD_PREP(MTK_FOE_WINFO_WCID, wcid) |
 		    FIELD_PREP(MTK_FOE_WINFO_BSS, bss);
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+	l2->winfo_pao = FIELD_PREP(MTK_FOE_WINFO_PAO_AMSDU_EN, amsdu_en);
+#endif
 #else
 	if (wdma_idx)
 		*ib2 |= MTK_FOE_IB2_WDMA_DEVIDX;
@@ -443,6 +447,17 @@ int mtk_foe_entry_set_dscp(struct mtk_foe_entry *entry, int dscp)
 	*ib2 &= ~MTK_FOE_IB2_DSCP;
 	*ib2 |= FIELD_PREP(MTK_FOE_IB2_DSCP, dscp);
 
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+	struct mtk_foe_mac_info *l2 = mtk_foe_entry_l2(entry);
+
+	if (*ib2 & MTK_FOE_IB2_WDMA_WINFO &&
+	    l2->winfo_pao & MTK_FOE_WINFO_PAO_AMSDU_EN) {
+		u8 tid = rt_tos2priority(dscp) & 0xf;
+
+		l2->winfo_pao |= FIELD_PREP(MTK_FOE_WINFO_PAO_TID, tid);
+	}
+#endif
+
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe.h b/drivers/net/ethernet/mediatek/mtk_ppe.h
index 2a8b6ef..66c7f10 100644
--- a/drivers/net/ethernet/mediatek/mtk_ppe.h
+++ b/drivers/net/ethernet/mediatek/mtk_ppe.h
@@ -428,7 +428,7 @@ int mtk_foe_entry_set_dsa(struct mtk_foe_entry *entry, int port);
 int mtk_foe_entry_set_vlan(struct mtk_foe_entry *entry, int vid);
 int mtk_foe_entry_set_pppoe(struct mtk_foe_entry *entry, int sid);
 int mtk_foe_entry_set_wdma(struct mtk_foe_entry *entry, int wdma_idx, int txq,
-			   int bss, int wcid);
+			   int bss, int wcid, bool amsdu_en);
 int mtk_foe_entry_set_qid(struct mtk_foe_entry *entry, int qid);
 int mtk_foe_entry_set_dscp(struct mtk_foe_entry *entry, int dscp);
 bool mtk_foe_entry_match(struct mtk_foe_entry *entry, struct mtk_foe_entry *data);
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe_offload.c b/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
index 95174b7..eab9e9d 100644
--- a/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
+++ b/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
@@ -112,6 +112,7 @@ mtk_flow_get_wdma_info(struct net_device *dev, const u8 *addr, struct mtk_wdma_i
 	info->queue = path.mtk_wdma.queue;
 	info->bss = path.mtk_wdma.bss;
 	info->wcid = path.mtk_wdma.wcid;
+	info->amsdu_en = path.mtk_wdma.amsdu_en;
 
 	return 0;
 }
@@ -193,13 +194,15 @@ mtk_flow_set_output_device(struct mtk_eth *eth, struct mtk_foe_entry *foe,
 
 	if (mtk_flow_get_wdma_info(dev, dest_mac, &info) == 0) {
 		mtk_foe_entry_set_wdma(foe, info.wdma_idx, info.queue, info.bss,
-				       info.wcid);
+				       info.wcid, info.amsdu_en);
 		pse_port = PSE_PPE0_PORT;
 #if defined(CONFIG_MEDIATEK_NETSYS_V2) || defined(CONFIG_MEDIATEK_NETSYS_V3)
 		if (info.wdma_idx == 0)
 			pse_port = PSE_WDMA0_PORT;
 		else if (info.wdma_idx == 1)
 			pse_port = PSE_WDMA1_PORT;
+		else if (info.wdma_idx == 2)
+			pse_port = PSE_WDMA2_PORT;
 		else
 			return -EOPNOTSUPP;
 #endif
@@ -490,8 +493,8 @@ mtk_flow_offload_replace(struct mtk_eth *eth, struct flow_cls_offload *f)
 	if (err)
 		return err;
 
-	if (wed_index >= 0 && (err = mtk_wed_flow_add(wed_index)) < 0)
-		return err;
+	/*if (wed_index >= 0 && (err = mtk_wed_flow_add(wed_index)) < 0)
+		return err;*/
 
 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 	if (!entry)
@@ -516,8 +519,8 @@ mtk_flow_offload_replace(struct mtk_eth *eth, struct flow_cls_offload *f)
 	mtk_foe_entry_clear(eth->ppe[ppe_index], entry);
 free:
 	kfree(entry);
-	if (wed_index >= 0)
-	    mtk_wed_flow_remove(wed_index);
+	/*if (wed_index >= 0)
+	    mtk_wed_flow_remove(wed_index);*/
 	return err;
 }
 
diff --git a/drivers/net/ethernet/mediatek/mtk_wed.c b/drivers/net/ethernet/mediatek/mtk_wed.c
index 3e760f7..9047cb0 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed.c
@@ -28,7 +28,7 @@ struct wo_cmd_ring {
 	u32 cnt;
 	u32 unit;
 };
-static struct mtk_wed_hw *hw_list[2];
+static struct mtk_wed_hw *hw_list[3];
 static DEFINE_MUTEX(hw_lock);
 
 static void
@@ -73,6 +73,26 @@ mtk_wdma_read_reset(struct mtk_wed_device *dev)
 	return wdma_r32(dev, MTK_WDMA_GLO_CFG);
 }
 
+static u32
+mtk_wed_check_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
+{
+	if (wed_r32(dev, reg) & mask)
+		return true;
+
+	return false;
+}
+
+static int
+mtk_wed_poll_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
+{
+	int sleep = 1000;
+	int timeout = 100 * sleep;
+	u32 val;
+
+	return read_poll_timeout(mtk_wed_check_busy, val, !val, sleep,
+				 timeout, false, dev, reg, mask);
+}
+
 static int
 mtk_wdma_rx_reset(struct mtk_wed_device *dev)
 {
@@ -235,6 +255,8 @@ mtk_wed_assign(struct mtk_wed_device *dev)
 			continue;
 
 		hw->wed_dev = dev;
+		hw->pci_base = MTK_WED_PCIE_BASE;
+
 		return hw;
 	}
 
@@ -242,23 +264,84 @@ mtk_wed_assign(struct mtk_wed_device *dev)
 }
 
 static int
-mtk_wed_buffer_alloc(struct mtk_wed_device *dev)
+mtk_wed_pao_buffer_alloc(struct mtk_wed_device *dev)
+{
+	struct mtk_wed_pao *pao;
+	int i, j;
+
+	pao = kzalloc(sizeof(struct mtk_wed_pao), GFP_KERNEL);
+	if (!pao)
+		return -ENOMEM;
+
+	dev->hw->wed_pao = pao;
+
+	for (i = 0; i < 32; i++) {
+		/* each segment is 64K*/
+		pao->hif_txd[i] = (char *)__get_free_pages(GFP_ATOMIC |
+							   GFP_DMA32 |
+							   __GFP_ZERO, 4);
+		if (!pao->hif_txd[i])
+			goto err;
+
+		pao->hif_txd_phys[i] = dma_map_single(dev->hw->dev,
+						      pao->hif_txd[i],
+						      16 * PAGE_SIZE,
+						      DMA_TO_DEVICE);
+		if (unlikely(dma_mapping_error(dev->hw->dev,
+					       pao->hif_txd_phys[i])))
+			goto err;
+	}
+
+	return 0;
+
+err:
+	for (j = 0; j < i; j++)
+		dma_unmap_single(dev->hw->dev, pao->hif_txd_phys[j],
+			     16 * PAGE_SIZE, DMA_TO_DEVICE);
+
+	return -ENOMEM;
+}
+
+static int
+mtk_wed_pao_free_buffer(struct mtk_wed_device *dev)
+{
+	struct mtk_wed_pao *pao = dev->hw->wed_pao;
+	int i;
+
+	for (i = 0; i < 32; i++) {
+		dma_unmap_single(dev->hw->dev, pao->hif_txd_phys[i],
+				 16 * PAGE_SIZE, DMA_TO_DEVICE);
+		free_pages((unsigned long)pao->hif_txd[i], 4);
+	}
+
+	return 0;
+}
+
+static int
+mtk_wed_tx_buffer_alloc(struct mtk_wed_device *dev)
 {
 	struct mtk_wdma_desc *desc;
+	void *desc_ptr;
 	dma_addr_t desc_phys;
-	void **page_list;
+	struct dma_page_info *page_list;
 	u32 last_seg = MTK_WDMA_DESC_CTRL_LAST_SEG1;
 	int token = dev->wlan.token_start;
-	int ring_size, n_pages, page_idx;
-	int i;
-
+	int ring_size, pkt_nums, n_pages, page_idx;
+	int i, ret = 0;
 
 	if (dev->ver == MTK_WED_V1) {
 		ring_size = dev->wlan.nbuf & ~(MTK_WED_BUF_PER_PAGE - 1);
-	} else {
+		pkt_nums = ring_size;
+		dev->tx_buf_ring.desc_size = sizeof(struct mtk_wdma_desc);
+	} else if (dev->hw->version == 2) {
 		ring_size = MTK_WED_VLD_GROUP_SIZE * MTK_WED_PER_GROUP_PKT +
 			    MTK_WED_WDMA_RING_SIZE * 2;
 		last_seg = MTK_WDMA_DESC_CTRL_LAST_SEG0;
+		dev->tx_buf_ring.desc_size = sizeof(struct mtk_wdma_desc);
+	} else if (dev->hw->version == 3) {
+		ring_size = MTK_WED_TX_BM_DMA_SIZE;
+		pkt_nums = MTK_WED_TX_BM_PKT_CNT;
+		dev->tx_buf_ring.desc_size = sizeof(struct mtk_rxbm_desc);
 	}
 
 	n_pages = ring_size / MTK_WED_BUF_PER_PAGE;
@@ -267,18 +350,20 @@ mtk_wed_buffer_alloc(struct mtk_wed_device *dev)
 	if (!page_list)
 		return -ENOMEM;
 
-	dev->buf_ring.size = dev->wlan.nbuf & ~(MTK_WED_BUF_PER_PAGE - 1);
-	dev->buf_ring.pages = page_list;
+	dev->tx_buf_ring.size = ring_size;
+	dev->tx_buf_ring.pages = page_list;
+	dev->tx_buf_ring.pkt_nums = pkt_nums;
 
-	desc = dma_alloc_coherent(dev->hw->dev, ring_size * sizeof(*desc),
-				  &desc_phys, GFP_KERNEL);
-	if (!desc)
+	desc_ptr = dma_alloc_coherent(dev->hw->dev,
+				      ring_size * dev->tx_buf_ring.desc_size,
+				      &desc_phys, GFP_KERNEL);
+	if (!desc_ptr)
 		return -ENOMEM;
 
-	dev->buf_ring.desc = desc;
-	dev->buf_ring.desc_phys = desc_phys;
+	dev->tx_buf_ring.desc = desc_ptr;
+	dev->tx_buf_ring.desc_phys = desc_phys;
 
-	for (i = 0, page_idx = 0; i < ring_size; i += MTK_WED_BUF_PER_PAGE) {
+	for (i = 0, page_idx = 0; i < pkt_nums; i += MTK_WED_BUF_PER_PAGE) {
 		dma_addr_t page_phys, buf_phys;
 		struct page *page;
 		void *buf;
@@ -295,7 +380,10 @@ mtk_wed_buffer_alloc(struct mtk_wed_device *dev)
 			return -ENOMEM;
 		}
 
-		page_list[page_idx++] = page;
+		page_list[page_idx].addr = page;
+		page_list[page_idx].addr_phys = page_phys;
+		page_idx++;
+
 		dma_sync_single_for_cpu(dev->hw->dev, page_phys, PAGE_SIZE,
 					DMA_BIDIRECTIONAL);
 
@@ -303,19 +391,23 @@ mtk_wed_buffer_alloc(struct mtk_wed_device *dev)
 		buf_phys = page_phys;
 
 		for (s = 0; s < MTK_WED_BUF_PER_PAGE; s++) {
-			u32 txd_size;
-
-			txd_size = dev->wlan.init_buf(buf, buf_phys, token++);
-
+			desc = desc_ptr;
 			desc->buf0 = buf_phys;
-			desc->buf1 = buf_phys + txd_size;
-			desc->ctrl = FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN0,
-						txd_size) |
-				     FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN1,
-						MTK_WED_BUF_SIZE - txd_size) |
-						last_seg;
-			desc->info = 0;
-			desc++;
+			if (dev->hw->version < 3) {
+				u32 txd_size;
+
+				txd_size = dev->wlan.init_buf(buf, buf_phys, token++);
+				desc->buf1 = buf_phys + txd_size;
+				desc->ctrl = FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN0,
+							txd_size) |
+					     FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN1,
+							MTK_WED_BUF_SIZE - txd_size) |
+							last_seg;
+				desc->info = 0;
+			} else {
+				desc->ctrl = token << 16;
+			}
+			desc_ptr += dev->tx_buf_ring.desc_size;
 
 			buf += MTK_WED_BUF_SIZE;
 			buf_phys += MTK_WED_BUF_SIZE;
@@ -325,15 +417,18 @@ mtk_wed_buffer_alloc(struct mtk_wed_device *dev)
 					   DMA_BIDIRECTIONAL);
 	}
 
-	return 0;
+	if (dev->hw->version == 3)
+		ret = mtk_wed_pao_buffer_alloc(dev);
+
+	return ret;
 }
 
 static void
-mtk_wed_free_buffer(struct mtk_wed_device *dev)
+mtk_wed_free_tx_buffer(struct mtk_wed_device *dev)
 {
-	struct mtk_wdma_desc *desc = dev->buf_ring.desc;
-	void **page_list = dev->buf_ring.pages;
-	int ring_size, page_idx;
+	struct mtk_rxbm_desc *desc = dev->tx_buf_ring.desc;
+	struct dma_page_info *page_list = dev->tx_buf_ring.pages;
+	int ring_size, page_idx, pkt_nums;
 	int i;
 
 	if (!page_list)
@@ -342,33 +437,33 @@ mtk_wed_free_buffer(struct mtk_wed_device *dev)
 	if (!desc)
 		goto free_pagelist;
 
-	if (dev->ver == MTK_WED_V1) {
-		ring_size = dev->wlan.nbuf & ~(MTK_WED_BUF_PER_PAGE - 1);
-	} else {
-		ring_size = MTK_WED_VLD_GROUP_SIZE * MTK_WED_PER_GROUP_PKT +
-			    MTK_WED_WDMA_RING_SIZE * 2;
+	pkt_nums = ring_size = dev->tx_buf_ring.size;
+	if (dev->hw->version == 3) {
+		mtk_wed_pao_free_buffer(dev);
+		pkt_nums = dev->tx_buf_ring.pkt_nums;
 	}
 
-	for (i = 0, page_idx = 0; i < ring_size; i += MTK_WED_BUF_PER_PAGE) {
-		void *page = page_list[page_idx++];
+	for (i = 0, page_idx = 0; i < pkt_nums; i += MTK_WED_BUF_PER_PAGE) {
+		void *page = page_list[page_idx].addr;
 
 		if (!page)
 			break;
 
-		dma_unmap_page(dev->hw->dev, desc[i].buf0,
+		dma_unmap_page(dev->hw->dev, page_list[page_idx].addr_phys,
 			       PAGE_SIZE, DMA_BIDIRECTIONAL);
 		__free_page(page);
+		page_idx++;
 	}
 
-	dma_free_coherent(dev->hw->dev, ring_size * sizeof(*desc),
-			  desc, dev->buf_ring.desc_phys);
+	dma_free_coherent(dev->hw->dev, ring_size * dev->tx_buf_ring.desc_size,
+			  dev->tx_buf_ring.desc, dev->tx_buf_ring.desc_phys);
 
 free_pagelist:
 	kfree(page_list);
 }
 
 static int
-mtk_wed_rx_bm_alloc(struct mtk_wed_device *dev)
+mtk_wed_rx_buffer_alloc(struct mtk_wed_device *dev)
 {
 	struct mtk_rxbm_desc *desc;
 	dma_addr_t desc_phys;
@@ -389,7 +484,7 @@ mtk_wed_rx_bm_alloc(struct mtk_wed_device *dev)
 }
 
 static void
-mtk_wed_free_rx_bm(struct mtk_wed_device *dev)
+mtk_wed_free_rx_buffer(struct mtk_wed_device *dev)
 {
 	struct mtk_rxbm_desc *desc = dev->rx_buf_ring.desc;
 	int ring_size = dev->rx_buf_ring.size;
@@ -403,6 +498,113 @@ mtk_wed_free_rx_bm(struct mtk_wed_device *dev)
 			  desc, dev->rx_buf_ring.desc_phys);
 }
 
+/* TODO */
+static int
+mtk_wed_rx_page_buffer_alloc(struct mtk_wed_device *dev)
+{
+	int ring_size = dev->wlan.rx_nbuf, buf_num = MTK_WED_RX_PG_BM_CNT;
+	struct mtk_rxbm_desc *desc;
+	dma_addr_t desc_phys;
+	struct dma_page_info *page_list;
+	int n_pages, page_idx;
+	int i;
+
+	n_pages = buf_num / MTK_WED_RX_PAGE_BUF_PER_PAGE;
+
+	page_list = kcalloc(n_pages, sizeof(*page_list), GFP_KERNEL);
+	if (!page_list)
+		return -ENOMEM;
+
+	dev->rx_page_buf_ring.size = ring_size & ~(MTK_WED_BUF_PER_PAGE - 1);
+	dev->rx_page_buf_ring.pages = page_list;
+	dev->rx_page_buf_ring.pkt_nums = buf_num;
+
+	desc = dma_alloc_coherent(dev->hw->dev, ring_size * sizeof(*desc),
+	                         &desc_phys, GFP_KERNEL);
+	if (!desc)
+		return -ENOMEM;
+
+	dev->rx_page_buf_ring.desc = desc;
+	dev->rx_page_buf_ring.desc_phys = desc_phys;
+
+	for (i = 0, page_idx = 0; i < buf_num; i += MTK_WED_RX_PAGE_BUF_PER_PAGE) {
+		dma_addr_t page_phys, buf_phys;
+		struct page *page;
+		void *buf;
+		int s;
+
+		page = __dev_alloc_pages(GFP_KERNEL, 0);
+		if (!page)
+			return -ENOMEM;
+
+		page_phys = dma_map_page(dev->hw->dev, page, 0, PAGE_SIZE,
+		                        DMA_BIDIRECTIONAL);
+		if (dma_mapping_error(dev->hw->dev, page_phys)) {
+			__free_page(page);
+			return -ENOMEM;
+		}
+
+		page_list[page_idx].addr= page;
+		page_list[page_idx].addr_phys= page_phys;
+		page_idx++;
+
+		dma_sync_single_for_cpu(dev->hw->dev, page_phys, PAGE_SIZE,
+		                       DMA_BIDIRECTIONAL);
+
+		buf = page_to_virt(page);
+		buf_phys = page_phys;
+
+		for (s = 0; s < MTK_WED_RX_PAGE_BUF_PER_PAGE; s++) {
+
+			desc->buf0 = cpu_to_le32(buf_phys);
+			desc++;
+
+			buf += MTK_WED_PAGE_BUF_SIZE;
+			buf_phys += MTK_WED_PAGE_BUF_SIZE;
+		}
+
+		dma_sync_single_for_device(dev->hw->dev, page_phys, PAGE_SIZE,
+					   DMA_BIDIRECTIONAL);
+	}
+
+	return 0;
+}
+
+static void
+mtk_wed_rx_page_free_buffer(struct mtk_wed_device *dev)
+{
+	struct mtk_rxbm_desc *desc = dev->rx_page_buf_ring.desc;
+	struct dma_page_info *page_list = dev->rx_page_buf_ring.pages;
+	int ring_size, page_idx;
+	int i;
+
+	if (!page_list)
+		return;
+
+	if (!desc)
+		goto free_pagelist;
+
+	ring_size = dev->rx_page_buf_ring.pkt_nums;
+
+	for (i = 0, page_idx = 0; i < ring_size; i += MTK_WED_RX_PAGE_BUF_PER_PAGE) {
+		void *page = page_list[page_idx].addr;
+
+		if (!page)
+			break;
+
+		dma_unmap_page(dev->hw->dev, page_list[page_idx].addr_phys,
+                              PAGE_SIZE, DMA_BIDIRECTIONAL);
+		__free_page(page);
+		page_idx++;
+       }
+
+	dma_free_coherent(dev->hw->dev, dev->rx_page_buf_ring.size * sizeof(*desc),
+                         desc, dev->rx_page_buf_ring.desc_phys);
+
+free_pagelist:
+       kfree(page_list);
+}
+
 static void
 mtk_wed_free_ring(struct mtk_wed_device *dev, struct mtk_wed_ring *ring, int scale)
 {
@@ -416,19 +618,25 @@ mtk_wed_free_ring(struct mtk_wed_device *dev, struct mtk_wed_ring *ring, int sca
 static void
 mtk_wed_free_tx_rings(struct mtk_wed_device *dev)
 {
-	int i;
+	int i, scale = dev->hw->version > 1 ? 2 : 1;
 
 	for (i = 0; i < ARRAY_SIZE(dev->tx_ring); i++)
-		mtk_wed_free_ring(dev, &dev->tx_ring[i], 1);
+		if (!(dev->rx_ring[i].flags & MTK_WED_RING_CONFIGURED))
+			mtk_wed_free_ring(dev, &dev->tx_ring[i], 1);
+
 	for (i = 0; i < ARRAY_SIZE(dev->tx_wdma); i++)
-		mtk_wed_free_ring(dev, &dev->tx_wdma[i], dev->ver);
+		if ((dev->rx_ring[i].flags & MTK_WED_RING_CONFIGURED))
+			mtk_wed_free_ring(dev, &dev->tx_wdma[i], scale);
 }
 
 static void
 mtk_wed_free_rx_rings(struct mtk_wed_device *dev)
 {
-	mtk_wed_free_rx_bm(dev);
+	mtk_wed_free_rx_buffer(dev);
 	mtk_wed_free_ring(dev, &dev->rro.rro_ring, 1);
+
+	if (dev->wlan.hwrro)
+		mtk_wed_rx_page_free_buffer(dev);
 }
 
 static void
@@ -437,7 +645,7 @@ mtk_wed_set_int(struct mtk_wed_device *dev, u32 irq_mask)
 	u32 wdma_mask;
 
 	wdma_mask = FIELD_PREP(MTK_WDMA_INT_MASK_RX_DONE, GENMASK(1, 0));
-	if (dev->ver > MTK_WED_V1)
+	if (mtk_wed_get_rx_capa(dev))
 		wdma_mask |= FIELD_PREP(MTK_WDMA_INT_MASK_TX_DONE,
 					GENMASK(1, 0));
 	/* wed control cr set */
@@ -447,7 +655,7 @@ mtk_wed_set_int(struct mtk_wed_device *dev, u32 irq_mask)
 		MTK_WED_CTRL_WED_TX_BM_EN |
 		MTK_WED_CTRL_WED_TX_FREE_AGENT_EN);
 
-	if (dev->ver == MTK_WED_V1) {
+	if (dev->hw->version == 1) {
 		wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER,
 			MTK_WED_PCIE_INT_TRIGGER_STATUS);
 
@@ -458,6 +666,8 @@ mtk_wed_set_int(struct mtk_wed_device *dev, u32 irq_mask)
 		wed_set(dev, MTK_WED_WPDMA_INT_CTRL,
 			MTK_WED_WPDMA_INT_CTRL_SUBRT_ADV);
 	} else {
+		if (dev->hw->version == 3)
+			wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_TKID_ALI_EN);
 
 		wed_w32(dev, MTK_WED_WPDMA_INT_CTRL_TX,
 			MTK_WED_WPDMA_INT_CTRL_TX0_DONE_EN |
@@ -475,18 +685,20 @@ mtk_wed_set_int(struct mtk_wed_device *dev, u32 irq_mask)
 			FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_TX_FREE_DONE_TRIG,
 				    dev->wlan.txfree_tbit));
 
-		wed_w32(dev, MTK_WED_WPDMA_INT_CTRL_RX,
-			MTK_WED_WPDMA_INT_CTRL_RX0_EN |
-			MTK_WED_WPDMA_INT_CTRL_RX0_CLR |
-			MTK_WED_WPDMA_INT_CTRL_RX1_EN |
-			MTK_WED_WPDMA_INT_CTRL_RX1_CLR |
-			FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RX0_DONE_TRIG,
-				   dev->wlan.rx_tbit[0]) |
-			FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RX1_DONE_TRIG,
-				   dev->wlan.rx_tbit[1]));
+		if (mtk_wed_get_rx_capa(dev))
+			wed_w32(dev, MTK_WED_WPDMA_INT_CTRL_RX,
+				MTK_WED_WPDMA_INT_CTRL_RX0_EN |
+				MTK_WED_WPDMA_INT_CTRL_RX0_CLR |
+				MTK_WED_WPDMA_INT_CTRL_RX1_EN |
+				MTK_WED_WPDMA_INT_CTRL_RX1_CLR |
+				FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RX0_DONE_TRIG,
+					   dev->wlan.rx_tbit[0]) |
+				FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RX1_DONE_TRIG,
+					   dev->wlan.rx_tbit[1]));
 	}
+
 	wed_w32(dev, MTK_WED_WDMA_INT_TRIGGER, wdma_mask);
-	if (dev->ver == MTK_WED_V1) {
+	if (dev->hw->version == 1) {
 		wed_clr(dev, MTK_WED_WDMA_INT_CTRL, wdma_mask);
 	} else {
 		wed_w32(dev, MTK_WED_WDMA_INT_CLR, wdma_mask);
@@ -506,6 +718,21 @@ mtk_wed_set_ext_int(struct mtk_wed_device *dev, bool en)
 {
 	u32 mask = MTK_WED_EXT_INT_STATUS_ERROR_MASK;
 
+	switch (dev->hw->version) {
+	case 1:
+		mask |= MTK_WED_EXT_INT_STATUS_TX_DRV_R_RESP_ERR;
+		break;
+	case 2 :
+		mask |= MTK_WED_EXT_INT_STATUS_RX_FBUF_LO_TH2 |
+			MTK_WED_EXT_INT_STATUS_RX_FBUF_HI_TH2 |
+			MTK_WED_EXT_INT_STATUS_RX_DRV_COHERENT |
+			MTK_WED_EXT_INT_STATUS_TX_DMA_W_RESP_ERR;
+		break;
+	case 3:
+		mask = MTK_WED_EXT_INT_STATUS_RX_DRV_COHERENT;
+		break;
+	}
+
 	if (!dev->hw->num_flows)
 		mask &= ~MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD;
 
@@ -514,31 +741,86 @@ mtk_wed_set_ext_int(struct mtk_wed_device *dev, bool en)
 }
 
 static void
-mtk_wed_set_512_support(struct mtk_wed_device *dev, bool en)
+mtk_wed_pao_init(struct mtk_wed_device *dev)
 {
-	if (en) {
-		wed_w32(dev, MTK_WED_TXDP_CTRL, MTK_WED_TXDP_DW9_OVERWR);
-		wed_w32(dev, MTK_WED_TXP_DW1,
-			FIELD_PREP(MTK_WED_WPDMA_WRITE_TXP, 0x0103));
-	} else {
-		wed_w32(dev, MTK_WED_TXP_DW1,
-			FIELD_PREP(MTK_WED_WPDMA_WRITE_TXP, 0x0100));
-		wed_clr(dev, MTK_WED_TXDP_CTRL, MTK_WED_TXDP_DW9_OVERWR);
+	struct mtk_wed_pao *pao = dev->hw->wed_pao;
+	int i;
+
+	for (i = 0; i < 32; i++)
+		wed_w32(dev, MTK_WED_PAO_HIFTXD_BASE_L(i),
+			pao->hif_txd_phys[i]);
+
+	/* init all sta parameter */
+	wed_w32(dev, MTK_WED_PAO_STA_INFO_INIT, MTK_WED_PAO_STA_RMVL |
+		MTK_WED_PAO_STA_WTBL_HDRT_MODE |
+		FIELD_PREP(MTK_WED_PAO_STA_MAX_AMSDU_LEN,
+			   dev->wlan.max_amsdu_len >> 8) |
+		FIELD_PREP(MTK_WED_PAO_STA_MAX_AMSDU_NUM,
+			   dev->wlan.max_amsdu_nums));
+
+	wed_w32(dev, MTK_WED_PAO_STA_INFO, MTK_WED_PAO_STA_INFO_DO_INIT);
+
+	if (mtk_wed_poll_busy(dev, MTK_WED_PAO_STA_INFO,
+			      MTK_WED_PAO_STA_INFO_DO_INIT)) {
+		dev_err(dev->hw->dev, "mtk_wed%d: pao init failed!\n",
+			dev->hw->index);
+		return;
 	}
+
+	/* init pao txd src */
+	wed_set(dev, MTK_WED_PAO_HIFTXD_CFG,
+		FIELD_PREP(MTK_WED_PAO_HIFTXD_SRC, dev->hw->index));
+
+	/* init qmem */
+	wed_set(dev, MTK_WED_PAO_PSE, MTK_WED_PAO_PSE_RESET);
+	if (mtk_wed_poll_busy(dev, MTK_WED_PAO_MON_QMEM_STS1, BIT(29))) {
+		pr_info("%s: init pao qmem fail\n", __func__);
+		return;
+	}
+
+	/* eagle E1 PCIE1 tx ring 22 flow control issue */
+	if (dev->wlan.chip_id == 0x7991) {
+		wed_clr(dev, MTK_WED_PAO_AMSDU_FIFO,
+			MTK_WED_PAO_AMSDU_IS_PRIOR0_RING);
+	}
+
+	wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_PAO_EN);
+
+	return;
 }
 
-static void
-mtk_wed_check_wfdma_rx_fill(struct mtk_wed_device *dev, int idx)
+static int
+mtk_wed_hwrro_init(struct mtk_wed_device *dev)
 {
-#define MTK_WFMDA_RX_DMA_EN 	BIT(2)
+	if (!mtk_wed_get_rx_capa(dev))
+		return 0;
+
+	wed_set(dev, MTK_WED_RRO_PG_BM_RX_DMAM,
+		FIELD_PREP(MTK_WED_RRO_PG_BM_RX_SDL0, 128));
 
+	wed_w32(dev, MTK_WED_RRO_PG_BM_BASE,
+		dev->rx_page_buf_ring.desc_phys);
+
+	wed_w32(dev, MTK_WED_RRO_PG_BM_INIT_PTR,
+		MTK_WED_RRO_PG_BM_INIT_SW_TAIL_IDX |
+		FIELD_PREP(MTK_WED_RRO_PG_BM_SW_TAIL_IDX,
+			   MTK_WED_RX_PG_BM_CNT));
+
+	/* enable rx_page_bm to fetch dmad */
+	wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_WED_RX_PG_BM_EN);
+
+	return 0;
+}
+
+static int
+mtk_wed_check_wfdma_rx_fill(struct mtk_wed_device *dev,
+			   struct mtk_wed_ring *ring)
+{
 	int timeout = 3;
-	u32 cur_idx, regs;
+	u32 cur_idx;
 
 	do {
-		regs = MTK_WED_WPDMA_RING_RX_DATA(idx) +
-		       MTK_WED_RING_OFS_CPU_IDX;
-		cur_idx = wed_r32(dev, regs);
+		cur_idx = readl(ring->wpdma + MTK_WED_RING_OFS_CPU_IDX);
 		if (cur_idx == MTK_WED_RX_RING_SIZE - 1)
 			break;
 
@@ -546,70 +828,133 @@ mtk_wed_check_wfdma_rx_fill(struct mtk_wed_device *dev, int idx)
 		timeout--;
 	} while (timeout > 0);
 
-	if (timeout) {
-		unsigned int val;
+	return timeout;
+}
 
-		val = wifi_r32(dev, dev->wlan.wpdma_rx_glo -
-			       dev->wlan.phy_base);
-		val |= MTK_WFMDA_RX_DMA_EN;
 
-		wifi_w32(dev, dev->wlan.wpdma_rx_glo -
-			 dev->wlan.phy_base, val);
+static void
+mtk_wed_set_512_support(struct mtk_wed_device *dev, bool en)
+{
+	if (en) {
+		wed_w32(dev, MTK_WED_TXDP_CTRL, MTK_WED_TXDP_DW9_OVERWR);
+		wed_w32(dev, MTK_WED_TXP_DW1,
+			FIELD_PREP(MTK_WED_WPDMA_WRITE_TXP, 0x0103));
 	} else {
-		dev_err(dev->hw->dev, "mtk_wed%d: rx(%d) dma enable failed!\n",
-			       dev->hw->index, idx);
+		wed_w32(dev, MTK_WED_TXP_DW1,
+			FIELD_PREP(MTK_WED_WPDMA_WRITE_TXP, 0x0100));
+		wed_clr(dev, MTK_WED_TXDP_CTRL, MTK_WED_TXDP_DW9_OVERWR);
 	}
 }
 
 static void
 mtk_wed_dma_enable(struct mtk_wed_device *dev)
 {
-	wed_set(dev, MTK_WED_WPDMA_INT_CTRL,
-		MTK_WED_WPDMA_INT_CTRL_SUBRT_ADV);
+#define MTK_WFMDA_RX_DMA_EN 	BIT(2)
+
+	if (dev->hw->version == 1)
+		wed_set(dev, MTK_WED_WPDMA_INT_CTRL,
+			MTK_WED_WPDMA_INT_CTRL_SUBRT_ADV);
 
 	wed_set(dev, MTK_WED_GLO_CFG,
 		MTK_WED_GLO_CFG_TX_DMA_EN |
 		MTK_WED_GLO_CFG_RX_DMA_EN);
+
+	wed_set(dev, MTK_WED_WDMA_RX_PREF_CFG,
+		FIELD_PREP(MTK_WED_WDMA_RX_PREF_BURST_SIZE, 0x10) |
+		FIELD_PREP(MTK_WED_WDMA_RX_PREF_LOW_THRES, 0x8));
+	wed_clr(dev, MTK_WED_WDMA_RX_PREF_CFG,
+		MTK_WED_WDMA_RX_PREF_DDONE2_EN);
+
+	wed_set(dev, MTK_WED_WDMA_RX_PREF_CFG, MTK_WED_WDMA_RX_PREF_EN);
+
 	wed_set(dev, MTK_WED_WPDMA_GLO_CFG,
 		MTK_WED_WPDMA_GLO_CFG_TX_DRV_EN |
-		MTK_WED_WPDMA_GLO_CFG_RX_DRV_EN);
+		MTK_WED_WPDMA_GLO_CFG_RX_DRV_EN |
+		MTK_WED_WPDMA_GLO_CFG_RX_DDONE2_WR);
 	wed_set(dev, MTK_WED_WDMA_GLO_CFG,
 		MTK_WED_WDMA_GLO_CFG_RX_DRV_EN);
 
 	wdma_set(dev, MTK_WDMA_GLO_CFG,
-		 MTK_WDMA_GLO_CFG_TX_DMA_EN |
+		 MTK_WDMA_GLO_CFG_TX_DMA_EN /*|
 		 MTK_WDMA_GLO_CFG_RX_INFO1_PRERES |
-		 MTK_WDMA_GLO_CFG_RX_INFO2_PRERES);
+		 MTK_WDMA_GLO_CFG_RX_INFO2_PRERES*/);
 
-	if (dev->ver == MTK_WED_V1) {
+	if (dev->hw->version == 1) {
 		wdma_set(dev, MTK_WDMA_GLO_CFG,
 			 MTK_WDMA_GLO_CFG_RX_INFO3_PRERES);
 	} else {
 		int idx = 0;
 
-		wed_set(dev, MTK_WED_WPDMA_CTRL,
-			MTK_WED_WPDMA_CTRL_SDL1_FIXED);
-
-		wed_set(dev, MTK_WED_WDMA_GLO_CFG,
-			MTK_WED_WDMA_GLO_CFG_TX_DRV_EN |
-			MTK_WED_WDMA_GLO_CFG_TX_DDONE_CHK);
+		if (mtk_wed_get_rx_capa(dev))
+			wed_set(dev, MTK_WED_WDMA_GLO_CFG,
+				MTK_WED_WDMA_GLO_CFG_TX_DRV_EN |
+				MTK_WED_WDMA_GLO_CFG_TX_DDONE_CHK);
 
 		wed_set(dev, MTK_WED_WPDMA_GLO_CFG,
 			MTK_WED_WPDMA_GLO_CFG_RX_DRV_R0_PKT_PROC |
 			MTK_WED_WPDMA_GLO_CFG_RX_DRV_R0_CRX_SYNC);
 
+		if (dev->hw->version == 3) {
+			wed_clr(dev, MTK_WED_WPDMA_GLO_CFG,
+				MTK_WED_WPDMA_GLO_CFG_TX_DDONE_CHK_LAST);
+			wed_set(dev, MTK_WED_WPDMA_GLO_CFG,
+				MTK_WED_WPDMA_GLO_CFG_TX_DDONE_CHK |
+				MTK_WED_WPDMA_GLO_CFG_RX_DRV_EVENT_PKT_FMT_CHK |
+				MTK_WED_WPDMA_GLO_CFG_RX_DRV_UNS_VER_FORCE_4);
+
+			wdma_set(dev, MTK_WDMA_PREF_RX_CFG, MTK_WDMA_PREF_RX_CFG_PREF_EN);
+			//wdma_w32(dev, MTK_WDMA_WRBK_RX_CFG, MTK_WDMA_WRBK_RX_CFG_WRBK_EN);
+			if (mtk_wed_get_rx_capa(dev)) {
+				wed_set(dev, MTK_WED_WPDMA_RX_D_PREF_CFG,
+					MTK_WED_WPDMA_RX_D_PREF_EN |
+					FIELD_PREP(MTK_WED_WPDMA_RX_D_PREF_BURST_SIZE, 0x10) |
+					FIELD_PREP(MTK_WED_WPDMA_RX_D_PREF_LOW_THRES, 0x8));
+
+				wed_set(dev, MTK_WED_RRO_RX_D_CFG(2), MTK_WED_RRO_RX_D_DRV_EN);
+
+				wdma_set(dev, MTK_WDMA_PREF_TX_CFG, MTK_WDMA_PREF_TX_CFG_PREF_EN);
+
+				wdma_set(dev, MTK_WDMA_WRBK_TX_CFG, MTK_WDMA_WRBK_TX_CFG_WRBK_EN);
+			}
+		}
+
 		wed_clr(dev, MTK_WED_WPDMA_GLO_CFG,
 			MTK_WED_WPDMA_GLO_CFG_TX_TKID_KEEP |
 			MTK_WED_WPDMA_GLO_CFG_TX_DMAD_DW3_PREV);
 
+		if (!mtk_wed_get_rx_capa(dev))
+			return;
+
+		wed_clr(dev, MTK_WED_WPDMA_RX_D_GLO_CFG, MTK_WED_WPDMA_RX_D_RXD_READ_LEN);
 		wed_set(dev, MTK_WED_WPDMA_RX_D_GLO_CFG,
 			MTK_WED_WPDMA_RX_D_RX_DRV_EN |
 			FIELD_PREP(MTK_WED_WPDMA_RX_D_RXD_READ_LEN, 0x18) |
 			FIELD_PREP(MTK_WED_WPDMA_RX_D_INIT_PHASE_RXEN_SEL,
 				   0x2));
 
-		for (idx = 0; idx < dev->hw->ring_num; idx++)
-			mtk_wed_check_wfdma_rx_fill(dev, idx);
+		for (idx = 0; idx < dev->hw->ring_num; idx++) {
+			struct mtk_wed_ring *ring = &dev->rx_ring[idx];
+
+			if(!(ring->flags & MTK_WED_RING_CONFIGURED))
+				continue;
+
+			if(mtk_wed_check_wfdma_rx_fill(dev, ring)) {
+				unsigned int val;
+
+				val = wifi_r32(dev, dev->wlan.wpdma_rx_glo -
+					       dev->wlan.phy_base);
+				val |= MTK_WFMDA_RX_DMA_EN;
+
+				wifi_w32(dev, dev->wlan.wpdma_rx_glo -
+					 dev->wlan.phy_base, val);
+
+				dev_err(dev->hw->dev, "mtk_wed%d: rx(%d) dma enable successful!\n",
+						dev->hw->index, idx);
+			} else {
+				dev_err(dev->hw->dev, "mtk_wed%d: rx(%d) dma enable failed!\n",
+					dev->hw->index, idx);
+			}
+		}
 	}
 }
 
@@ -644,15 +989,20 @@ mtk_wed_dma_disable(struct mtk_wed_device *dev)
 			MTK_WED_WPDMA_RX_D_RX_DRV_EN);
 		wed_clr(dev, MTK_WED_WDMA_GLO_CFG,
 			MTK_WED_WDMA_GLO_CFG_TX_DDONE_CHK);
-	}
 
-	mtk_wed_set_512_support(dev, false);
+		if (dev->hw->version == 3 && mtk_wed_get_rx_capa(dev)) {
+			wdma_clr(dev, MTK_WDMA_PREF_TX_CFG,
+				 MTK_WDMA_PREF_TX_CFG_PREF_EN);
+			wdma_clr(dev, MTK_WDMA_PREF_RX_CFG,
+				 MTK_WDMA_PREF_RX_CFG_PREF_EN);
+		}
+	}
 }
 
 static void
 mtk_wed_stop(struct mtk_wed_device *dev)
 {
-	if (dev->ver > MTK_WED_V1) {
+	if (mtk_wed_get_rx_capa(dev)) {
 		wed_w32(dev, MTK_WED_EXT_INT_MASK1, 0);
 		wed_w32(dev, MTK_WED_EXT_INT_MASK2, 0);
 	}
@@ -677,13 +1027,21 @@ mtk_wed_deinit(struct mtk_wed_device *dev)
 		MTK_WED_CTRL_WED_TX_BM_EN |
 		MTK_WED_CTRL_WED_TX_FREE_AGENT_EN);
 
-	if (dev->hw->ver == 1)
+	if (dev->hw->version == 1)
 		return;
 
 	wed_clr(dev, MTK_WED_CTRL,
 		MTK_WED_CTRL_RX_ROUTE_QM_EN |
 		MTK_WED_CTRL_WED_RX_BM_EN |
 		MTK_WED_CTRL_RX_RRO_QM_EN);
+
+	if (dev->hw->version == 3) {
+		wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_PAO_EN);
+		wed_clr(dev, MTK_WED_RESET, MTK_WED_RESET_TX_PAO);
+		wed_clr(dev, MTK_WED_PCIE_INT_CTRL,
+			MTK_WED_PCIE_INT_CTRL_MSK_EN_POLA |
+			MTK_WED_PCIE_INT_CTRL_MSK_IRQ_FILTER);
+	}
 }
 
 static void
@@ -702,9 +1060,9 @@ mtk_wed_detach(struct mtk_wed_device *dev)
 
 	mtk_wdma_tx_reset(dev);
 
-	mtk_wed_free_buffer(dev);
+	mtk_wed_free_tx_buffer(dev);
 	mtk_wed_free_tx_rings(dev);
-	if (dev->ver > MTK_WED_V1) {
+	if (mtk_wed_get_rx_capa(dev)) {
 		mtk_wed_wo_reset(dev);
 		mtk_wed_free_rx_rings(dev);
 		mtk_wed_wo_exit(hw);
@@ -731,24 +1089,29 @@ mtk_wed_detach(struct mtk_wed_device *dev)
 static void
 mtk_wed_bus_init(struct mtk_wed_device *dev)
 {
-#define PCIE_BASE_ADDR0 0x11280000
+	switch (dev->wlan.bus_type) {
+	case MTK_WED_BUS_PCIE: {
+		struct device_node *np = dev->hw->eth->dev->of_node;
+		struct regmap *regs;
 
-	if (dev->wlan.bus_type == MTK_WED_BUS_PCIE) {
-		struct device_node *node;
-		void __iomem * base_addr;
-		u32 value = 0;
+		if (dev->hw->version == 2) {
+			regs = syscon_regmap_lookup_by_phandle(np,
+							       "mediatek,wed-pcie");
+			if (IS_ERR(regs))
+				break;
 
-		node = of_parse_phandle(dev->hw->node, "mediatek,wed_pcie", 0);
-		if (!node) {
-			pr_err("%s: no wed_pcie node\n", __func__);
-			return;
+			regmap_update_bits(regs, 0, BIT(0), BIT(0));
 		}
 
-		base_addr = of_iomap(node, 0);
-
-		value = readl(base_addr);
-		value |= BIT(0);
-		writel(value, base_addr);
+		if (dev->wlan.msi) {
+		     wed_w32(dev, MTK_WED_PCIE_CFG_INTM, dev->hw->pci_base| 0xc08);
+		     wed_w32(dev, MTK_WED_PCIE_CFG_BASE, dev->hw->pci_base | 0xc04);
+		     wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER, BIT(8));
+		} else {
+		     wed_w32(dev, MTK_WED_PCIE_CFG_INTM, dev->hw->pci_base | 0x180);
+		     wed_w32(dev, MTK_WED_PCIE_CFG_BASE, dev->hw->pci_base | 0x184);
+		     wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER, BIT(24));
+		}
 
 		wed_w32(dev, MTK_WED_PCIE_INT_CTRL,
 			FIELD_PREP(MTK_WED_PCIE_INT_CTRL_POLL_EN, 2));
@@ -756,45 +1119,53 @@ mtk_wed_bus_init(struct mtk_wed_device *dev)
 		/* pcie interrupt control: pola/source selection */
 		wed_set(dev, MTK_WED_PCIE_INT_CTRL,
 			MTK_WED_PCIE_INT_CTRL_MSK_EN_POLA |
-			FIELD_PREP(MTK_WED_PCIE_INT_CTRL_SRC_SEL, 1));
-		wed_r32(dev, MTK_WED_PCIE_INT_CTRL);
-
-		value = wed_r32(dev, MTK_WED_PCIE_CFG_INTM);
-		value = wed_r32(dev, MTK_WED_PCIE_CFG_BASE);
-		wed_w32(dev, MTK_WED_PCIE_CFG_INTM, PCIE_BASE_ADDR0 | 0x180);
-		wed_w32(dev, MTK_WED_PCIE_CFG_BASE, PCIE_BASE_ADDR0 | 0x184);
+			MTK_WED_PCIE_INT_CTRL_MSK_IRQ_FILTER |
+			FIELD_PREP(MTK_WED_PCIE_INT_CTRL_SRC_SEL, dev->hw->index));
 
-		value = wed_r32(dev, MTK_WED_PCIE_CFG_INTM);
-		value = wed_r32(dev, MTK_WED_PCIE_CFG_BASE);
-
-		wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER, BIT(24));
-		wed_r32(dev, MTK_WED_PCIE_INT_TRIGGER);
-
-		/* pola setting */
-		value = wed_r32(dev, MTK_WED_PCIE_INT_CTRL);
-		wed_set(dev, MTK_WED_PCIE_INT_CTRL,
-			MTK_WED_PCIE_INT_CTRL_MSK_EN_POLA);
-	} else if (dev->wlan.bus_type == MTK_WED_BUS_AXI) {
+		break;
+	}
+	case MTK_WED_BUS_AXI:
 		wed_set(dev, MTK_WED_WPDMA_INT_CTRL,
 			MTK_WED_WPDMA_INT_CTRL_SIG_SRC |
 			FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_SRC_SEL, 0));
+		break;
+	default:
+		break;
 	}
+
 	return;
 }
 
 static void
 mtk_wed_set_wpdma(struct mtk_wed_device *dev)
 {
-	if (dev->ver > MTK_WED_V1) {
+	if (dev->hw->version == 1) {
+		wed_w32(dev, MTK_WED_WPDMA_CFG_BASE,  dev->wlan.wpdma_phys);
+	} else {
+		mtk_wed_bus_init(dev);
+
 		wed_w32(dev, MTK_WED_WPDMA_CFG_BASE,  dev->wlan.wpdma_int);
 		wed_w32(dev, MTK_WED_WPDMA_CFG_INT_MASK,  dev->wlan.wpdma_mask);
-		wed_w32(dev, MTK_WED_WPDMA_CFG_TX,  dev->wlan.wpdma_tx);
+		wed_w32(dev, MTK_WED_WPDMA_CFG_TX, dev->wlan.wpdma_tx);
 		wed_w32(dev, MTK_WED_WPDMA_CFG_TX_FREE,  dev->wlan.wpdma_txfree);
 
-		wed_w32(dev, MTK_WED_WPDMA_RX_GLO_CFG,  dev->wlan.wpdma_rx_glo);
-		wed_w32(dev, MTK_WED_WPDMA_RX_RING,  dev->wlan.wpdma_rx);
-	} else {
-		wed_w32(dev, MTK_WED_WPDMA_CFG_BASE,  dev->wlan.wpdma_phys);
+		if (mtk_wed_get_rx_capa(dev)) {
+			int i;
+
+			wed_w32(dev, MTK_WED_WPDMA_RX_GLO_CFG,  dev->wlan.wpdma_rx_glo);
+			wed_w32(dev, MTK_WED_WPDMA_RX_RING0,  dev->wlan.wpdma_rx[0]);
+			if (dev->wlan.wpdma_rx[1])
+				wed_w32(dev, MTK_WED_WPDMA_RX_RING1,  dev->wlan.wpdma_rx[1]);
+
+			if (dev->wlan.hwrro) {
+				wed_w32(dev, MTK_WED_RRO_RX_D_CFG(0), dev->wlan.wpdma_rx_rro[0]);
+				wed_w32(dev, MTK_WED_RRO_RX_D_CFG(1), dev->wlan.wpdma_rx_rro[1]);
+				for (i = 0; i < MTK_WED_RX_PAGE_QUEUES; i++) {
+					wed_w32(dev, MTK_WED_RRO_MSDU_PG_RING_CFG(i),
+						dev->wlan.wpdma_rx_pg + i * 0x10);
+			       }
+			}
+		}
 	}
 }
 
@@ -806,21 +1177,25 @@ mtk_wed_hw_init_early(struct mtk_wed_device *dev)
 	mtk_wed_deinit(dev);
 	mtk_wed_reset(dev, MTK_WED_RESET_WED);
 
-	if (dev->ver > MTK_WED_V1)
-		mtk_wed_bus_init(dev);
-
 	mtk_wed_set_wpdma(dev);
 
-	mask = MTK_WED_WDMA_GLO_CFG_BT_SIZE |
-	       MTK_WED_WDMA_GLO_CFG_DYNAMIC_DMAD_RECYCLE |
-	       MTK_WED_WDMA_GLO_CFG_RX_DIS_FSM_AUTO_IDLE;
-	set = FIELD_PREP(MTK_WED_WDMA_GLO_CFG_BT_SIZE, 2) |
-	      MTK_WED_WDMA_GLO_CFG_DYNAMIC_SKIP_DMAD_PREP |
-	      MTK_WED_WDMA_GLO_CFG_IDLE_DMAD_SUPPLY;
+	if (dev->hw->version == 3) {
+		mask = MTK_WED_WDMA_GLO_CFG_BT_SIZE;
+		set = FIELD_PREP(MTK_WED_WDMA_GLO_CFG_BT_SIZE, 2);
+	} else {
+		mask = MTK_WED_WDMA_GLO_CFG_BT_SIZE |
+		       MTK_WED_WDMA_GLO_CFG_DYNAMIC_DMAD_RECYCLE |
+		       MTK_WED_WDMA_GLO_CFG_RX_DIS_FSM_AUTO_IDLE;
+		set = FIELD_PREP(MTK_WED_WDMA_GLO_CFG_BT_SIZE, 2) |
+		      MTK_WED_WDMA_GLO_CFG_DYNAMIC_SKIP_DMAD_PREP |
+		      MTK_WED_WDMA_GLO_CFG_IDLE_DMAD_SUPPLY;
+	}
+
 	wed_m32(dev, MTK_WED_WDMA_GLO_CFG, mask, set);
 
-	if (dev->ver == MTK_WED_V1) {
+	if (dev->hw->version == 1) {
 		u32 offset;
+
 		offset = dev->hw->index ? 0x04000400 : 0;
 		wed_w32(dev, MTK_WED_WDMA_OFFSET0, 0x2a042a20 + offset);
 		wed_w32(dev, MTK_WED_WDMA_OFFSET1, 0x29002800 + offset);
@@ -907,11 +1282,16 @@ mtk_wed_route_qm_hw_init(struct mtk_wed_device *dev)
 	} while (1);
 
 	/* configure RX_ROUTE_QM */
-	wed_clr(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_Q_RST);
-	wed_clr(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_TXDMAD_FPORT);
-	wed_set(dev, MTK_WED_RTQM_GLO_CFG,
-		FIELD_PREP(MTK_WED_RTQM_TXDMAD_FPORT, 0x3 + dev->hw->index));
-	wed_clr(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_Q_RST);
+	if (dev->hw->version == 2) {
+		wed_clr(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_Q_RST);
+		wed_clr(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_TXDMAD_FPORT);
+		wed_set(dev, MTK_WED_RTQM_GLO_CFG,
+			FIELD_PREP(MTK_WED_RTQM_TXDMAD_FPORT, 0x3 + dev->hw->index));
+		wed_clr(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_Q_RST);
+	} else {
+		wed_set(dev, MTK_WED_RTQM_ENQ_CFG0,
+			FIELD_PREP(MTK_WED_RTQM_ENQ_CFG_TXDMAD_FPORT, 0x3 + dev->hw->index));
+	}
 
 	/* enable RX_ROUTE_QM */
 	wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_RX_ROUTE_QM_EN);
@@ -920,23 +1300,45 @@ mtk_wed_route_qm_hw_init(struct mtk_wed_device *dev)
 static void
 mtk_wed_tx_hw_init(struct mtk_wed_device *dev)
 {
-	int size = dev->buf_ring.size;
+	int size = dev->wlan.nbuf;
 	int rev_size = MTK_WED_TX_RING_SIZE / 2;
-	int thr = 1;
+	int thr_lo = 1, thr_hi = 1;
 
-	if (dev->ver > MTK_WED_V1) {
+	if (dev->hw->version == 1) {
+		wed_w32(dev, MTK_WED_TX_BM_CTRL,
+			MTK_WED_TX_BM_CTRL_PAUSE |
+			FIELD_PREP(MTK_WED_TX_BM_CTRL_VLD_GRP_NUM, size / 128) |
+			FIELD_PREP(MTK_WED_TX_BM_CTRL_RSV_GRP_NUM, rev_size / 128));
+	} else {
 		size = MTK_WED_WDMA_RING_SIZE * ARRAY_SIZE(dev->tx_wdma) +
-		       dev->buf_ring.size;
+		       dev->tx_buf_ring.size;
 		rev_size = size;
-		thr = 0;
+		thr_lo = 0;
+		thr_hi = MTK_WED_TX_BM_DYN_THR_HI;
+
+		wed_w32(dev, MTK_WED_TX_TKID_CTRL,
+			MTK_WED_TX_TKID_CTRL_PAUSE |
+			FIELD_PREP(MTK_WED_TX_TKID_CTRL_VLD_GRP_NUM,
+				   size / 128) |
+			FIELD_PREP(MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM,
+				   size / 128));
+
+		/* return SKBID + SDP back to bm */
+		if (dev->ver == 3) {
+			wed_set(dev, MTK_WED_TX_TKID_CTRL,
+				MTK_WED_TX_TKID_CTRL_FREE_FORMAT);
+			 size = dev->wlan.nbuf;
+			 rev_size = size;
+		} else {
+			wed_w32(dev, MTK_WED_TX_TKID_DYN_THR,
+				FIELD_PREP(MTK_WED_TX_TKID_DYN_THR_LO, 0) |
+				MTK_WED_TX_TKID_DYN_THR_HI);
+		}
 	}
 
-	wed_w32(dev, MTK_WED_TX_BM_CTRL,
-		MTK_WED_TX_BM_CTRL_PAUSE |
-		FIELD_PREP(MTK_WED_TX_BM_CTRL_VLD_GRP_NUM, size / 128) |
-		FIELD_PREP(MTK_WED_TX_BM_CTRL_RSV_GRP_NUM, rev_size / 128));
+	mtk_wed_reset(dev, MTK_WED_RESET_TX_BM);
 
-	wed_w32(dev, MTK_WED_TX_BM_BASE, dev->buf_ring.desc_phys);
+	wed_w32(dev, MTK_WED_TX_BM_BASE, dev->tx_buf_ring.desc_phys);
 
 	wed_w32(dev, MTK_WED_TX_BM_TKID,
 		FIELD_PREP(MTK_WED_TX_BM_TKID_START,
@@ -946,25 +1348,44 @@ mtk_wed_tx_hw_init(struct mtk_wed_device *dev)
 
 	wed_w32(dev, MTK_WED_TX_BM_BUF_LEN, MTK_WED_PKT_SIZE);
 
-	wed_w32(dev, MTK_WED_TX_BM_DYN_THR,
-		FIELD_PREP(MTK_WED_TX_BM_DYN_THR_LO, thr) |
-		MTK_WED_TX_BM_DYN_THR_HI);
+	if (dev->hw->version < 3)
+		wed_w32(dev, MTK_WED_TX_BM_DYN_THR,
+			FIELD_PREP(MTK_WED_TX_BM_DYN_THR_LO, thr_lo) |
+			FIELD_PREP(MTK_WED_TX_BM_DYN_THR_LO, thr_hi));
+	else {
+		/* change to new bm */
+		wed_w32(dev, MTK_WED_TX_BM_INIT_PTR, dev->tx_buf_ring.pkt_nums |
+			MTK_WED_TX_BM_INIT_SW_TAIL_IDX);
+		wed_clr(dev, MTK_WED_TX_BM_CTRL, MTK_WED_TX_BM_CTRL_LEGACY_EN);
+	}
 
-	if (dev->ver > MTK_WED_V1) {
+	if (dev->hw->version != 1) {
 		wed_w32(dev, MTK_WED_TX_TKID_CTRL,
 			MTK_WED_TX_TKID_CTRL_PAUSE |
 			FIELD_PREP(MTK_WED_TX_TKID_CTRL_VLD_GRP_NUM,
-				   dev->buf_ring.size / 128) |
+				   size / 128) |
 			FIELD_PREP(MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM,
-				   dev->buf_ring.size / 128));
-		wed_w32(dev, MTK_WED_TX_TKID_DYN_THR,
-			FIELD_PREP(MTK_WED_TX_TKID_DYN_THR_LO, 0) |
-			MTK_WED_TX_TKID_DYN_THR_HI);
+				   size / 128));
+
+		/* return SKBID + SDP back to bm */
+		if (dev->ver == 3)
+			wed_set(dev, MTK_WED_TX_TKID_CTRL,
+				MTK_WED_TX_TKID_CTRL_FREE_FORMAT);
+		else
+			wed_w32(dev, MTK_WED_TX_TKID_DYN_THR,
+				FIELD_PREP(MTK_WED_TX_TKID_DYN_THR_LO, 0) |
+				MTK_WED_TX_TKID_DYN_THR_HI);
 	}
-	mtk_wed_reset(dev, MTK_WED_RESET_TX_BM);
+	wed_w32(dev, MTK_WED_TX_BM_TKID,
+		FIELD_PREP(MTK_WED_TX_BM_TKID_START,
+			   dev->wlan.token_start) |
+		FIELD_PREP(MTK_WED_TX_BM_TKID_END,
+			   dev->wlan.token_start + dev->wlan.nbuf - 1));
 
+	wed_w32(dev, MTK_WED_TX_BM_INIT_PTR, dev->tx_buf_ring.pkt_nums |
+		MTK_WED_TX_BM_INIT_SW_TAIL_IDX);
 	wed_clr(dev, MTK_WED_TX_BM_CTRL, MTK_WED_TX_BM_CTRL_PAUSE);
-	if (dev->ver > MTK_WED_V1)
+	if (dev->hw->version != 1)
 		wed_clr(dev, MTK_WED_TX_TKID_CTRL, MTK_WED_TX_TKID_CTRL_PAUSE);
 }
 
@@ -977,7 +1398,26 @@ mtk_wed_rx_hw_init(struct mtk_wed_device *dev)
 
 	wed_w32(dev, MTK_WED_WPDMA_RX_D_RST_IDX, 0);
 
+	/* reset prefetch index of ring */
+	wed_set(dev, MTK_WED_WPDMA_RX_D_PREF_RX0_SIDX,
+		MTK_WED_WPDMA_RX_D_PREF_SIDX_IDX_CLR);
+	wed_clr(dev, MTK_WED_WPDMA_RX_D_PREF_RX0_SIDX,
+		MTK_WED_WPDMA_RX_D_PREF_SIDX_IDX_CLR);
+
+	wed_set(dev, MTK_WED_WPDMA_RX_D_PREF_RX1_SIDX,
+		MTK_WED_WPDMA_RX_D_PREF_SIDX_IDX_CLR);
+	wed_clr(dev, MTK_WED_WPDMA_RX_D_PREF_RX1_SIDX,
+		MTK_WED_WPDMA_RX_D_PREF_SIDX_IDX_CLR);
+
+	/* reset prefetch FIFO of ring */
+	wed_set(dev, MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG,
+		MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG_R0_CLR |
+		MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG_R1_CLR);
+	wed_w32(dev, MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG, 0);
+
 	mtk_wed_rx_bm_hw_init(dev);
+	if (dev->wlan.hwrro)
+		mtk_wed_hwrro_init(dev);
 	mtk_wed_rro_hw_init(dev);
 	mtk_wed_route_qm_hw_init(dev);
 }
@@ -991,7 +1431,7 @@ mtk_wed_hw_init(struct mtk_wed_device *dev)
 	dev->init_done = true;
 	mtk_wed_set_ext_int(dev, false);
 	mtk_wed_tx_hw_init(dev);
-	if (dev->ver > MTK_WED_V1)
+	if (mtk_wed_get_rx_capa(dev))
 		mtk_wed_rx_hw_init(dev);
 }
 
@@ -1015,26 +1455,6 @@ mtk_wed_ring_reset(struct mtk_wdma_desc *desc, int size, int scale, bool tx)
 	}
 }
 
-static u32
-mtk_wed_check_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
-{
- 	if (wed_r32(dev, reg) & mask)
-		return true;
-
-	return false;
-}
-
-static int
-mtk_wed_poll_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
-{
-	int sleep = 1000;
-	int timeout = 100 * sleep;
-	u32 val;
-
-	return read_poll_timeout(mtk_wed_check_busy, val, !val, sleep,
-				 timeout, false, dev, reg, mask);
-}
-
 static void
 mtk_wed_rx_reset(struct mtk_wed_device *dev)
 {
@@ -1133,7 +1553,7 @@ mtk_wed_rx_reset(struct mtk_wed_device *dev)
 		mtk_wed_ring_reset(desc, MTK_WED_RX_RING_SIZE, 1, false);
 	}
 
-	mtk_wed_free_rx_bm(dev);
+	mtk_wed_free_rx_buffer(dev);
 }
 
 
@@ -1271,12 +1691,15 @@ mtk_wed_wdma_rx_ring_setup(struct mtk_wed_device *dev,
 		int idx, int size, bool reset)
 {
 	struct mtk_wed_ring *wdma = &dev->tx_wdma[idx];
+	int scale = dev->hw->version > 1 ? 2 : 1;
 
 	if(!reset)
 		if (mtk_wed_ring_alloc(dev, wdma, MTK_WED_WDMA_RING_SIZE,
-				       dev->ver, true))
+				       scale, true))
 			return -ENOMEM;
 
+	wdma->flags |= MTK_WED_RING_CONFIGURED;
+
 	wdma_w32(dev, MTK_WDMA_RING_RX(idx) + MTK_WED_RING_OFS_BASE,
 		 wdma->desc_phys);
 	wdma_w32(dev, MTK_WDMA_RING_RX(idx) + MTK_WED_RING_OFS_COUNT,
@@ -1296,12 +1719,31 @@ mtk_wed_wdma_tx_ring_setup(struct mtk_wed_device *dev,
 	int idx, int size, bool reset)
 {
 	struct mtk_wed_ring *wdma = &dev->rx_wdma[idx];
+	int scale = dev->hw->version > 1 ? 2 : 1;
 
 	if (!reset)
 		if (mtk_wed_ring_alloc(dev, wdma, MTK_WED_WDMA_RING_SIZE,
-				       dev->ver, true))
+				       scale, true))
 			return -ENOMEM;
 
+	if (dev->hw->version == 3) {
+		struct mtk_wdma_desc *desc = wdma->desc;
+		int i;
+
+		for (i = 0; i < MTK_WED_WDMA_RING_SIZE; i++) {
+			desc->buf0 = 0;
+			desc->ctrl = MTK_WDMA_DESC_CTRL_DMA_DONE;
+			desc->buf1 = 0;
+			desc->info = MTK_WDMA_TXD0_DESC_INFO_DMA_DONE;
+			desc++;
+			desc->buf0 = 0;
+			desc->ctrl = MTK_WDMA_DESC_CTRL_DMA_DONE;
+			desc->buf1 = 0;
+			desc->info = MTK_WDMA_TXD1_DESC_INFO_DMA_DONE;
+			desc++;
+		}
+	}
+
 	wdma_w32(dev, MTK_WDMA_RING_TX(idx) + MTK_WED_RING_OFS_BASE,
 		 wdma->desc_phys);
 	wdma_w32(dev, MTK_WDMA_RING_TX(idx) + MTK_WED_RING_OFS_COUNT,
@@ -1312,7 +1754,7 @@ mtk_wed_wdma_tx_ring_setup(struct mtk_wed_device *dev,
 		 MTK_WDMA_RING_TX(idx) + MTK_WED_RING_OFS_DMA_IDX, 0);
 	if (reset)
 		mtk_wed_ring_reset(wdma->desc, MTK_WED_WDMA_RING_SIZE,
-				   dev->ver, true);
+				   scale, true);
 	if (idx == 0)  {
 		wed_w32(dev, MTK_WED_WDMA_RING_TX
 			+ MTK_WED_RING_OFS_BASE, wdma->desc_phys);
@@ -1395,7 +1837,7 @@ mtk_wed_send_msg(struct mtk_wed_device *dev, int cmd_id, void *data, int len)
 {
 	struct mtk_wed_wo *wo = dev->hw->wed_wo;
 
-	if (dev->ver == MTK_WED_V1)
+	if (!mtk_wed_get_rx_capa(dev))
 		return 0;
 
 	return mtk_wed_mcu_send_msg(wo, MODULE_ID_WO, cmd_id, data, len, true);
@@ -1420,24 +1862,106 @@ mtk_wed_ppe_check(struct mtk_wed_device *dev, struct sk_buff *skb,
 	}
 }
 
+static void
+mtk_wed_start_hwrro(struct mtk_wed_device *dev, u32 irq_mask)
+{
+	int idx, ret;
+
+	wed_w32(dev, MTK_WED_WPDMA_INT_MASK, irq_mask);
+	wed_w32(dev, MTK_WED_INT_MASK, irq_mask);
+
+	if (!mtk_wed_get_rx_capa(dev) || !dev->wlan.hwrro)
+		return;
+
+	wed_set(dev, MTK_WED_RRO_RX_D_CFG(2), MTK_WED_RRO_MSDU_PG_DRV_CLR);
+	wed_w32(dev, MTK_WED_RRO_MSDU_PG_RING2_CFG, MTK_WED_RRO_MSDU_PG_DRV_CLR);
+
+	wed_w32(dev, MTK_WED_WPDMA_INT_CTRL_RRO_RX,
+		MTK_WED_WPDMA_INT_CTRL_RRO_RX0_EN |
+		MTK_WED_WPDMA_INT_CTRL_RRO_RX0_CLR |
+		MTK_WED_WPDMA_INT_CTRL_RRO_RX1_EN |
+		MTK_WED_WPDMA_INT_CTRL_RRO_RX1_CLR |
+		FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RRO_RX0_DONE_TRIG,
+			   dev->wlan.rro_rx_tbit[0]) |
+		FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RRO_RX1_DONE_TRIG,
+			   dev->wlan.rro_rx_tbit[1]));
+
+	wed_w32(dev, MTK_WED_WPDMA_INT_CTRL_RRO_MSDU_PG,
+		MTK_WED_WPDMA_INT_CTRL_RRO_PG0_EN |
+		MTK_WED_WPDMA_INT_CTRL_RRO_PG0_CLR |
+		MTK_WED_WPDMA_INT_CTRL_RRO_PG1_EN |
+		MTK_WED_WPDMA_INT_CTRL_RRO_PG1_CLR |
+		MTK_WED_WPDMA_INT_CTRL_RRO_PG2_EN |
+		MTK_WED_WPDMA_INT_CTRL_RRO_PG2_CLR |
+		FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RRO_PG0_DONE_TRIG,
+			   dev->wlan.rx_pg_tbit[0]) |
+		FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RRO_PG1_DONE_TRIG,
+			   dev->wlan.rx_pg_tbit[1])|
+		FIELD_PREP(MTK_WED_WPDMA_INT_CTRL_RRO_PG2_DONE_TRIG,
+			   dev->wlan.rx_pg_tbit[2]));
+
+	/*
+	 * RRO_MSDU_PG_RING2_CFG1_FLD_DRV_EN should be enabled after
+	 * WM FWDL completed, otherwise RRO_MSDU_PG ring may broken
+	 */
+	wed_set(dev, MTK_WED_RRO_MSDU_PG_RING2_CFG, MTK_WED_RRO_MSDU_PG_DRV_EN);
+
+	for (idx = 0; idx < MTK_WED_RX_QUEUES; idx++) {
+		struct mtk_wed_ring *ring = &dev->rx_rro_ring[idx];
+
+		if(!(ring->flags & MTK_WED_RING_CONFIGURED))
+			continue;
+
+		ret = mtk_wed_check_wfdma_rx_fill(dev, ring);
+		if (!ret)
+			dev_err(dev->hw->dev, "mtk_wed%d: rx_rro_ring(%d) init failed!\n",
+				dev->hw->index, idx);
+	}
+
+	for (idx = 0; idx < MTK_WED_RX_PAGE_QUEUES; idx++){
+		struct mtk_wed_ring *ring = &dev->rx_page_ring[idx];
+		if(!(ring->flags & MTK_WED_RING_CONFIGURED))
+			continue;
+
+		ret = mtk_wed_check_wfdma_rx_fill(dev, ring);
+		if (!ret)
+			dev_err(dev->hw->dev, "mtk_wed%d: rx_page_ring(%d) init failed!\n",
+				dev->hw->index, idx);
+	}
+}
+
 static void
 mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
 {
 	int i, ret;
 
-	if (dev->ver > MTK_WED_V1)
-		ret = mtk_wed_rx_bm_alloc(dev);
+	if (mtk_wed_get_rx_capa(dev)) {
+		ret = mtk_wed_rx_buffer_alloc(dev);
+		if (ret)
+			return;
+
+		if (dev->wlan.hwrro)
+			mtk_wed_rx_page_buffer_alloc(dev);
+	}
 
 	for (i = 0; i < ARRAY_SIZE(dev->tx_wdma); i++)
 		if (!dev->tx_wdma[i].desc)
 			mtk_wed_wdma_rx_ring_setup(dev, i, 16, false);
 
+	for (i = 0; i < ARRAY_SIZE(dev->rx_page_ring); i++) {
+		u32 count = MTK_WED_RRO_MSDU_PG_CTRL0(i) +
+			    MTK_WED_RING_OFS_COUNT;
+
+		if (!wed_r32(dev, count))
+			wed_w32(dev, count, 1);
+	}
+
 	mtk_wed_hw_init(dev);
 
 	mtk_wed_set_int(dev, irq_mask);
 	mtk_wed_set_ext_int(dev, true);
 
-	if (dev->ver == MTK_WED_V1) {
+	if (dev->hw->version == 1) {
 		u32 val;
 
 		val = dev->wlan.wpdma_phys |
@@ -1448,33 +1972,52 @@ mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
 			val |= BIT(1);
 		val |= BIT(0);
 		regmap_write(dev->hw->mirror, dev->hw->index * 4, val);
-	} else {
+	} else if (mtk_wed_get_rx_capa(dev)) {
 		/* driver set mid ready and only once */
 		wed_w32(dev, MTK_WED_EXT_INT_MASK1,
 			MTK_WED_EXT_INT_STATUS_WPDMA_MID_RDY);
 		wed_w32(dev, MTK_WED_EXT_INT_MASK2,
 			MTK_WED_EXT_INT_STATUS_WPDMA_MID_RDY);
+		if (dev->hw->version == 3)
+			wed_w32(dev, MTK_WED_EXT_INT_MASK3,
+				MTK_WED_EXT_INT_STATUS_WPDMA_MID_RDY);
 
 		wed_r32(dev, MTK_WED_EXT_INT_MASK1);
 		wed_r32(dev, MTK_WED_EXT_INT_MASK2);
+		if (dev->hw->version == 3)
+			wed_r32(dev, MTK_WED_EXT_INT_MASK3);
 
 		ret = mtk_wed_rro_cfg(dev);
 		if (ret)
 			return;
 	}
-	mtk_wed_set_512_support(dev, dev->wlan.wcid_512);
+
+	if (dev->hw->version == 2)
+		mtk_wed_set_512_support(dev, dev->wlan.wcid_512);
+	else if (dev->hw->version == 3)
+		mtk_wed_pao_init(dev);
 
 	mtk_wed_dma_enable(dev);
 	dev->running = true;
 }
 
+static int
+mtk_wed_get_pci_base(struct mtk_wed_device *dev)
+{
+	if (dev->hw->index == 0)
+		return MTK_WED_PCIE_BASE0;
+	else if (dev->hw->index == 1)
+		return MTK_WED_PCIE_BASE1;
+	else
+		return MTK_WED_PCIE_BASE2;
+}
+
 static int
 mtk_wed_attach(struct mtk_wed_device *dev)
 	__releases(RCU)
 {
 	struct mtk_wed_hw *hw;
 	struct device *device;
-	u16 ver;
 	int ret = 0;
 
 	RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
@@ -1494,34 +2037,30 @@ mtk_wed_attach(struct mtk_wed_device *dev)
 		goto out;
 	}
 
-	device = dev->wlan.bus_type == MTK_WED_BUS_PCIE
-	? &dev->wlan.pci_dev->dev
-	: &dev->wlan.platform_dev->dev;
+	device = dev->wlan.bus_type == MTK_WED_BUS_PCIE ?
+				       &dev->wlan.pci_dev->dev
+				       : &dev->wlan.platform_dev->dev;
 	dev_info(device, "attaching wed device %d version %d\n",
-		 hw->index, hw->ver);
+		 hw->index, hw->version);
 
 	dev->hw = hw;
 	dev->dev = hw->dev;
 	dev->irq = hw->irq;
 	dev->wdma_idx = hw->index;
+	dev->ver = hw->version;
+
+	if (dev->hw->version == 3)
+		dev->hw->pci_base = mtk_wed_get_pci_base(dev);
 
 	if (hw->eth->dma_dev == hw->eth->dev &&
 	    of_dma_is_coherent(hw->eth->dev->of_node))
 		mtk_eth_set_dma_device(hw->eth, hw->dev);
 
-	dev->ver = FIELD_GET(MTK_WED_REV_ID_MAJOR,
-			    wed_r32(dev, MTK_WED_REV_ID));
-	if (dev->ver > MTK_WED_V1)
-		ver = FIELD_GET(MTK_WED_REV_ID_MINOR,
-			    wed_r32(dev, MTK_WED_REV_ID));
-
-	dev->rev_id = ((dev->ver << 28) | ver << 16);
-
-	ret = mtk_wed_buffer_alloc(dev);
+	ret = mtk_wed_tx_buffer_alloc(dev);
 	if (ret)
 		goto error;
 
-	if (dev->ver > MTK_WED_V1) {
+	if (mtk_wed_get_rx_capa(dev)) {
 		ret = mtk_wed_rro_alloc(dev);
 		if (ret)
 			goto error;
@@ -1533,15 +2072,20 @@ mtk_wed_attach(struct mtk_wed_device *dev)
 	init_completion(&dev->wlan_reset_done);
 	atomic_set(&dev->fe_reset, 0);
 
-	if (dev->ver == MTK_WED_V1)
+	if (dev->hw->version != 1)
+		dev->rev_id = wed_r32(dev, MTK_WED_REV_ID);
+	else
 		regmap_update_bits(hw->hifsys, HIFSYS_DMA_AG_MAP,
 				   BIT(hw->index), 0);
-	else
+
+	if (mtk_wed_get_rx_capa(dev))
 		ret = mtk_wed_wo_init(hw);
 
 error:
-	if (ret)
+	if (ret) {
+		pr_info("%s: detach wed\n", __func__);
 		mtk_wed_detach(dev);
+	}
 out:
 	mutex_unlock(&hw_lock);
 
@@ -1576,8 +2120,26 @@ mtk_wed_tx_ring_setup(struct mtk_wed_device *dev, int idx,
 	if (mtk_wed_wdma_rx_ring_setup(dev, idx, MTK_WED_WDMA_RING_SIZE, reset))
 		return -ENOMEM;
 
+	if (dev->hw->version == 3 && idx == 1) {
+		/* reset prefetch index */
+		wed_set(dev, MTK_WED_WDMA_RX_PREF_CFG,
+		       MTK_WED_WDMA_RX_PREF_RX0_SIDX_CLR |
+		       MTK_WED_WDMA_RX_PREF_RX1_SIDX_CLR);
+
+		wed_clr(dev, MTK_WED_WDMA_RX_PREF_CFG,
+		       MTK_WED_WDMA_RX_PREF_RX0_SIDX_CLR |
+		       MTK_WED_WDMA_RX_PREF_RX1_SIDX_CLR);
+
+		/* reset prefetch FIFO */
+		wed_w32(dev, MTK_WED_WDMA_RX_PREF_FIFO_CFG,
+		       MTK_WED_WDMA_RX_PREF_FIFO_RX0_CLR |
+		       MTK_WED_WDMA_RX_PREF_FIFO_RX1_CLR);
+		wed_w32(dev, MTK_WED_WDMA_RX_PREF_FIFO_CFG, 0);
+	}
+
 	ring->reg_base = MTK_WED_RING_TX(idx);
 	ring->wpdma = regs;
+	ring->flags |= MTK_WED_RING_CONFIGURED;
 
 	/* WED -> WPDMA */
 	wpdma_tx_w32(dev, idx, MTK_WED_RING_OFS_BASE, ring->desc_phys);
@@ -1599,7 +2161,7 @@ mtk_wed_txfree_ring_setup(struct mtk_wed_device *dev, void __iomem *regs)
 	struct mtk_wed_ring *ring = &dev->txfree_ring;
 	int i, idx = 1;
 
-	if(dev->ver > MTK_WED_V1)
+	if(dev->hw->version > 1)
 		idx = 0;
 
 	/*
@@ -1652,6 +2214,129 @@ mtk_wed_rx_ring_setup(struct mtk_wed_device *dev,
 	return 0;
 }
 
+static int
+mtk_wed_rro_rx_ring_setup(struct mtk_wed_device *dev, int idx, void __iomem *regs)
+{
+	struct mtk_wed_ring *ring = &dev->rx_rro_ring[idx];
+
+	ring->wpdma = regs;
+
+	wed_w32(dev, MTK_WED_RRO_RX_D_RX(idx) + MTK_WED_RING_OFS_BASE,
+		readl(regs));
+	wed_w32(dev, MTK_WED_RRO_RX_D_RX(idx) + MTK_WED_RING_OFS_COUNT,
+		readl(regs + MTK_WED_RING_OFS_COUNT));
+
+	ring->flags |= MTK_WED_RING_CONFIGURED;
+
+	return 0;
+}
+
+static int
+mtk_wed_msdu_pg_rx_ring_setup(struct mtk_wed_device *dev, int idx, void __iomem *regs)
+{
+	struct mtk_wed_ring *ring = &dev->rx_page_ring[idx];
+
+	ring->wpdma = regs;
+
+	wed_w32(dev, MTK_WED_RRO_MSDU_PG_CTRL0(idx) + MTK_WED_RING_OFS_BASE,
+		readl(regs));
+	wed_w32(dev, MTK_WED_RRO_MSDU_PG_CTRL0(idx) + MTK_WED_RING_OFS_COUNT,
+		readl(regs + MTK_WED_RING_OFS_COUNT));
+
+	ring->flags |= MTK_WED_RING_CONFIGURED;
+
+	return 0;
+}
+
+static int
+mtk_wed_ind_rx_ring_setup(struct mtk_wed_device *dev, void __iomem *regs)
+{
+	struct mtk_wed_ring *ring = &dev->ind_cmd_ring;
+	u32 val = readl(regs + MTK_WED_RING_OFS_COUNT);
+	int i = 0, cnt = 0;
+
+	ring->wpdma = regs;
+
+	if (readl(regs) & 0xf)
+		pr_info("%s(): address is not 16-byte alignment\n", __func__);
+
+	wed_w32(dev, MTK_WED_IND_CMD_RX_CTRL1 + MTK_WED_RING_OFS_BASE,
+		readl(regs) & 0xfffffff0);
+
+	wed_w32(dev, MTK_WED_IND_CMD_RX_CTRL1 + MTK_WED_RING_OFS_COUNT,
+		readl(regs + MTK_WED_RING_OFS_COUNT));
+
+	/* ack sn cr */
+	wed_w32(dev, MTK_WED_RRO_CFG0, dev->wlan.phy_base +
+		dev->wlan.ind_cmd.ack_sn_addr);
+	wed_w32(dev, MTK_WED_RRO_CFG1,
+		FIELD_PREP(MTK_WED_RRO_CFG1_MAX_WIN_SZ,
+			   dev->wlan.ind_cmd.win_size) |
+		FIELD_PREP(MTK_WED_RRO_CFG1_PARTICL_SE_ID,
+			   dev->wlan.ind_cmd.particular_sid));
+
+	/* particular session addr element */
+	wed_w32(dev, MTK_WED_ADDR_ELEM_CFG0, dev->wlan.ind_cmd.particular_se_phys);
+
+	for (i = 0; i < dev->wlan.ind_cmd.se_group_nums; i++) {
+		wed_w32(dev, MTK_WED_RADDR_ELEM_TBL_WDATA,
+			dev->wlan.ind_cmd.addr_elem_phys[i] >> 4);
+		wed_w32(dev, MTK_WED_ADDR_ELEM_TBL_CFG,
+			MTK_WED_ADDR_ELEM_TBL_WR | (i & 0x7f));
+
+		val = wed_r32(dev, MTK_WED_ADDR_ELEM_TBL_CFG);
+		while (!(val & MTK_WED_ADDR_ELEM_TBL_WR_RDY) &&
+			 cnt < 100) {
+			val = wed_r32(dev, MTK_WED_ADDR_ELEM_TBL_CFG);
+			cnt++;
+		}
+		if (cnt >= 100) {
+			dev_err(dev->hw->dev, "mtk_wed%d: write ba session base failed!\n",
+				dev->hw->index);
+		}
+		/*if (mtk_wed_poll_busy(dev, MTK_WED_ADDR_ELEM_TBL_CFG,
+				      MTK_WED_ADDR_ELEM_TBL_WR_RDY)) {
+			dev_err(dev->hw->dev, "mtk_wed%d: write ba session base failed!\n",
+				dev->hw->index);
+			return -1;
+		}*/
+	}
+
+	/* pn check init */
+	for (i = 0; i < dev->wlan.ind_cmd.particular_sid; i++) {
+		wed_w32(dev, MTK_WED_PN_CHECK_WDATA_M,
+			MTK_WED_PN_CHECK_IS_FIRST);
+
+		wed_w32(dev, MTK_WED_PN_CHECK_CFG, MTK_WED_PN_CHECK_WR |
+			FIELD_PREP(MTK_WED_PN_CHECK_SE_ID, i));
+
+		cnt = 0;
+		val = wed_r32(dev, MTK_WED_PN_CHECK_CFG);
+		while (!(val & MTK_WED_PN_CHECK_WR_RDY) &&
+			 cnt < 100) {
+			val = wed_r32(dev, MTK_WED_PN_CHECK_CFG);
+			cnt++;
+		}
+		if (cnt >= 100) {
+			dev_err(dev->hw->dev, "mtk_wed%d: session(%d) init failed!\n",
+				dev->hw->index, i);
+		}
+		/*if (mtk_wed_poll_busy(dev, MTK_WED_PN_CHECK_CFG,
+				      MTK_WED_PN_CHECK_WR_RDY)) {
+			dev_err(dev->hw->dev, "mtk_wed%d: session(%d) init failed!\n",
+				dev->hw->index, i);
+			//return -1;
+		}*/
+	}
+
+	wed_w32(dev, MTK_WED_RX_IND_CMD_CNT0, MTK_WED_RX_IND_CMD_DBG_CNT_EN);
+
+	wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_WED_RX_IND_CMD_EN);
+
+	return 0;
+}
+
+
 static u32
 mtk_wed_irq_get(struct mtk_wed_device *dev, u32 mask)
 {
@@ -1659,9 +2344,13 @@ mtk_wed_irq_get(struct mtk_wed_device *dev, u32 mask)
 
 	val = wed_r32(dev, MTK_WED_EXT_INT_STATUS);
 	wed_w32(dev, MTK_WED_EXT_INT_STATUS, val);
-	val &= MTK_WED_EXT_INT_STATUS_ERROR_MASK;
-	if (!dev->hw->num_flows)
-		val &= ~MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD;
+	if (dev->hw->version == 3) {
+		val &= MTK_WED_EXT_INT_STATUS_RX_DRV_COHERENT;
+	} else {
+		val &= MTK_WED_EXT_INT_STATUS_ERROR_MASK;	
+		if (!dev->hw->num_flows)
+			val &= ~MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD;	
+	}
 	if (val && net_ratelimit())
 		pr_err("mtk_wed%d: error status=%08x\n", dev->hw->index, val);
 
@@ -1754,6 +2443,9 @@ void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 		.tx_ring_setup = mtk_wed_tx_ring_setup,
 		.txfree_ring_setup = mtk_wed_txfree_ring_setup,
 		.rx_ring_setup = mtk_wed_rx_ring_setup,
+		.rro_rx_ring_setup = mtk_wed_rro_rx_ring_setup,
+		.msdu_pg_rx_ring_setup = mtk_wed_msdu_pg_rx_ring_setup,
+		.ind_rx_ring_setup = mtk_wed_ind_rx_ring_setup,
 		.msg_update = mtk_wed_send_msg,
 		.start = mtk_wed_start,
 		.stop = mtk_wed_stop,
@@ -1765,6 +2457,7 @@ void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 		.detach = mtk_wed_detach,
 		.setup_tc = mtk_wed_eth_setup_tc,
 		.ppe_check = mtk_wed_ppe_check,
+		.start_hwrro = mtk_wed_start_hwrro,
 	};
 	struct device_node *eth_np = eth->dev->of_node;
 	struct platform_device *pdev;
@@ -1804,9 +2497,10 @@ void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 	hw->wdma_phy = wdma_phy;
 	hw->index = index;
 	hw->irq = irq;
-	hw->ver = MTK_HAS_CAPS(eth->soc->caps, MTK_NETSYS_V2) ? 2 : 1;
+	hw->version = MTK_HAS_CAPS(eth->soc->caps, MTK_NETSYS_V3) ?
+		      3 : MTK_HAS_CAPS(eth->soc->caps, MTK_NETSYS_V2) ? 2 : 1;
 
-	if (!MTK_HAS_CAPS(eth->soc->caps, MTK_NETSYS_V2)) {
+	if (hw->version == 1) {
 		hw->mirror = syscon_regmap_lookup_by_phandle(eth_np,
 							     "mediatek,pcie-mirror");
 		hw->hifsys = syscon_regmap_lookup_by_phandle(eth_np,
@@ -1821,7 +2515,6 @@ void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 			regmap_write(hw->mirror, 0, 0);
 			regmap_write(hw->mirror, 4, 0);
 		}
-		hw->ver = MTK_WED_V1;
 	}
 
 	mtk_wed_hw_add_debugfs(hw);
diff --git a/drivers/net/ethernet/mediatek/mtk_wed.h b/drivers/net/ethernet/mediatek/mtk_wed.h
index 490873c..fcf7bd0 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed.h
+++ b/drivers/net/ethernet/mediatek/mtk_wed.h
@@ -10,10 +10,13 @@
 #include <linux/netdevice.h>
 #define MTK_PCIE_BASE(n)		(0x1a143000 + (n) * 0x2000)
 
-#define MTK_WED_PKT_SIZE		1900
+#define MTK_WED_PKT_SIZE		1920//1900
 #define MTK_WED_BUF_SIZE		2048
+#define MTK_WED_PAGE_BUF_SIZE		128
 #define MTK_WED_BUF_PER_PAGE		(PAGE_SIZE / 2048)
+#define MTK_WED_RX_PAGE_BUF_PER_PAGE	(PAGE_SIZE / 128)
 #define MTK_WED_RX_RING_SIZE		1536
+#define MTK_WED_RX_PG_BM_CNT		8192
 
 #define MTK_WED_TX_RING_SIZE		2048
 #define MTK_WED_WDMA_RING_SIZE		512
@@ -27,6 +30,9 @@
 #define MTK_WED_RRO_QUE_CNT		8192
 #define MTK_WED_MIOD_ENTRY_CNT		128
 
+#define MTK_WED_TX_BM_DMA_SIZE		65536
+#define MTK_WED_TX_BM_PKT_CNT		32768
+
 #define MODULE_ID_WO		1
 
 struct mtk_eth;
@@ -43,6 +49,8 @@ struct mtk_wed_hw {
 	struct dentry *debugfs_dir;
 	struct mtk_wed_device *wed_dev;
 	struct mtk_wed_wo *wed_wo;
+	struct mtk_wed_pao *wed_pao;
+	u32 pci_base;
 	u32 debugfs_reg;
 	u32 num_flows;
 	u32 wdma_phy;
@@ -50,7 +58,8 @@ struct mtk_wed_hw {
 	int ring_num;
 	int irq;
 	int index;
-	u32 ver;
+	int token_id;
+	u32 version;
 };
 
 struct mtk_wdma_info {
@@ -58,6 +67,18 @@ struct mtk_wdma_info {
 	u8 queue;
 	u16 wcid;
 	u8 bss;
+	u32 usr_info;
+	u8 tid;
+	u8 is_fixedrate;
+	u8 is_prior;
+	u8 is_sp;
+	u8 hf;
+	u8 amsdu_en;
+};
+
+struct mtk_wed_pao {
+	char *hif_txd[32];
+	dma_addr_t hif_txd_phys[32];
 };
 
 #ifdef CONFIG_NET_MEDIATEK_SOC_WED
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c b/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c
index 4a9e684..51e3d7c 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c
@@ -11,9 +11,11 @@ struct reg_dump {
 	u16 offset;
 	u8 type;
 	u8 base;
+	u32 mask;
 };
 
 enum {
+	DUMP_TYPE_END,
 	DUMP_TYPE_STRING,
 	DUMP_TYPE_WED,
 	DUMP_TYPE_WDMA,
@@ -23,8 +25,11 @@ enum {
 	DUMP_TYPE_WED_RRO,
 };
 
+#define DUMP_END() { .type = DUMP_TYPE_END }
 #define DUMP_STR(_str) { _str, 0, DUMP_TYPE_STRING }
 #define DUMP_REG(_reg, ...) { #_reg, MTK_##_reg, __VA_ARGS__ }
+#define DUMP_REG_MASK(_reg, _mask) { #_mask, MTK_##_reg, DUMP_TYPE_WED, 0, MTK_##_mask }
+
 #define DUMP_RING(_prefix, _base, ...)				\
 	{ _prefix " BASE", _base, __VA_ARGS__ },		\
 	{ _prefix " CNT",  _base + 0x4, __VA_ARGS__ },	\
@@ -32,6 +37,7 @@ enum {
 	{ _prefix " DIDX", _base + 0xc, __VA_ARGS__ }
 
 #define DUMP_WED(_reg) DUMP_REG(_reg, DUMP_TYPE_WED)
+#define DUMP_WED_MASK(_reg, _mask) DUMP_REG_MASK(_reg, _mask)
 #define DUMP_WED_RING(_base) DUMP_RING(#_base, MTK_##_base, DUMP_TYPE_WED)
 
 #define DUMP_WDMA(_reg) DUMP_REG(_reg, DUMP_TYPE_WDMA)
@@ -52,36 +58,49 @@ print_reg_val(struct seq_file *s, const char *name, u32 val)
 
 static void
 dump_wed_regs(struct seq_file *s, struct mtk_wed_device *dev,
-	      const struct reg_dump *regs, int n_regs)
+	      const struct reg_dump **regs)
 {
-	const struct reg_dump *cur;
+	const struct reg_dump **cur_o = regs, *cur;
+	bool newline = false;
 	u32 val;
 
-	for (cur = regs; cur < &regs[n_regs]; cur++) {
-		switch (cur->type) {
-		case DUMP_TYPE_STRING:
-			seq_printf(s, "%s======== %s:\n",
-				   cur > regs ? "\n" : "",
-				   cur->name);
-			continue;
-		case DUMP_TYPE_WED:
-		case DUMP_TYPE_WED_RRO:
-			val = wed_r32(dev, cur->offset);
-			break;
-		case DUMP_TYPE_WDMA:
-			val = wdma_r32(dev, cur->offset);
-			break;
-		case DUMP_TYPE_WPDMA_TX:
-			val = wpdma_tx_r32(dev, cur->base, cur->offset);
-			break;
-		case DUMP_TYPE_WPDMA_TXFREE:
-			val = wpdma_txfree_r32(dev, cur->offset);
-			break;
-		case DUMP_TYPE_WPDMA_RX:
-			val = wpdma_rx_r32(dev, cur->base, cur->offset);
-			break;
+	while (*cur_o) {
+		cur = *cur_o;
+
+		while (cur->type != DUMP_TYPE_END) {
+			switch (cur->type) {
+			case DUMP_TYPE_STRING:
+				seq_printf(s, "%s======== %s:\n",
+					   newline ? "\n" : "",
+					   cur->name);
+				newline = true;
+				cur++;
+				continue;
+			case DUMP_TYPE_WED:
+			case DUMP_TYPE_WED_RRO:
+				val = wed_r32(dev, cur->offset);
+				break;
+			case DUMP_TYPE_WDMA:
+				val = wdma_r32(dev, cur->offset);
+				break;
+			case DUMP_TYPE_WPDMA_TX:
+				val = wpdma_tx_r32(dev, cur->base, cur->offset);
+				break;
+			case DUMP_TYPE_WPDMA_TXFREE:
+				val = wpdma_txfree_r32(dev, cur->offset);
+				break;
+			case DUMP_TYPE_WPDMA_RX:
+				val = wpdma_rx_r32(dev, cur->base, cur->offset);
+				break;
+			}
+
+			if (cur->mask)
+				val = (cur->mask & val) >> (ffs(cur->mask) - 1);
+
+			print_reg_val(s, cur->name, val);
+			cur++;
 		}
-		print_reg_val(s, cur->name, val);
+		cur_o++;
 	}
 }
 
@@ -89,7 +108,7 @@ dump_wed_regs(struct seq_file *s, struct mtk_wed_device *dev,
 static int
 wed_txinfo_show(struct seq_file *s, void *data)
 {
-	static const struct reg_dump regs[] = {
+	static const struct reg_dump regs_common[] = {
 		DUMP_STR("WED TX"),
 		DUMP_WED(WED_TX_MIB(0)),
 		DUMP_WED_RING(WED_RING_TX(0)),
@@ -128,16 +147,32 @@ wed_txinfo_show(struct seq_file *s, void *data)
 		DUMP_WDMA_RING(WDMA_RING_RX(0)),
 		DUMP_WDMA_RING(WDMA_RING_RX(1)),
 
-		DUMP_STR("TX FREE"),
+		DUMP_STR("WED TX FREE"),
 		DUMP_WED(WED_RX_MIB(0)),
+		DUMP_WED_RING(WED_RING_RX(0)),
+		DUMP_WED(WED_WPDMA_RX_COHERENT_MIB(0)),
+
+		DUMP_WED(WED_RX_MIB(1)),
+		DUMP_WED_RING(WED_RING_RX(1)),
+		DUMP_WED(WED_WPDMA_RX_COHERENT_MIB(1)),
+		DUMP_STR("WED_WPDMA TX FREE"),
+		DUMP_WED_RING(WED_WPDMA_RING_RX(0)),
+		DUMP_WED_RING(WED_WPDMA_RING_RX(1)),
+		DUMP_END(),
+	};
+
+	static const struct reg_dump *regs[] = {
+		&regs_common[0],
+		NULL,
 	};
+
 	struct mtk_wed_hw *hw = s->private;
 	struct mtk_wed_device *dev = hw->wed_dev;
 
 	if (!dev)
 		return 0;
 
-	dump_wed_regs(s, dev, regs, ARRAY_SIZE(regs));
+	dump_wed_regs(s, dev, regs);
 
 	return 0;
 }
@@ -146,7 +181,7 @@ DEFINE_SHOW_ATTRIBUTE(wed_txinfo);
 static int
 wed_rxinfo_show(struct seq_file *s, void *data)
 {
-	static const struct reg_dump regs[] = {
+	static const struct reg_dump regs_common[] = {
 		DUMP_STR("WPDMA RX"),
 		DUMP_WPDMA_RX_RING(0),
 		DUMP_WPDMA_RX_RING(1),
@@ -164,7 +199,7 @@ wed_rxinfo_show(struct seq_file *s, void *data)
 		DUMP_WED_RING(WED_RING_RX_DATA(0)),
 		DUMP_WED_RING(WED_RING_RX_DATA(1)),
 
-		DUMP_STR("WED RRO"),
+		DUMP_STR("WED WO RRO"),
 		DUMP_WED_RRO_RING(WED_RROQM_MIOD_CTRL0),
 		DUMP_WED(WED_RROQM_MID_MIB),
 		DUMP_WED(WED_RROQM_MOD_MIB),
@@ -175,16 +210,6 @@ wed_rxinfo_show(struct seq_file *s, void *data)
 		DUMP_WED(WED_RROQM_FDBK_ANC_MIB),
 		DUMP_WED(WED_RROQM_FDBK_ANC2H_MIB),
 
-		DUMP_STR("WED Route QM"),
-		DUMP_WED(WED_RTQM_R2H_MIB(0)),
-		DUMP_WED(WED_RTQM_R2Q_MIB(0)),
-		DUMP_WED(WED_RTQM_Q2H_MIB(0)),
-		DUMP_WED(WED_RTQM_R2H_MIB(1)),
-		DUMP_WED(WED_RTQM_R2Q_MIB(1)),
-		DUMP_WED(WED_RTQM_Q2H_MIB(1)),
-		DUMP_WED(WED_RTQM_Q2N_MIB),
-		DUMP_WED(WED_RTQM_Q2B_MIB),
-		DUMP_WED(WED_RTQM_PFDBK_MIB),
 
 		DUMP_STR("WED WDMA TX"),
 		DUMP_WED(WED_WDMA_TX_MIB),
@@ -205,15 +230,99 @@ wed_rxinfo_show(struct seq_file *s, void *data)
 		DUMP_WED(WED_RX_BM_INTF2),
 		DUMP_WED(WED_RX_BM_INTF),
 		DUMP_WED(WED_RX_BM_ERR_STS),
+		DUMP_END()
+	};
+
+	static const struct reg_dump regs_v2[] = {
+		DUMP_STR("WED Route QM"),
+		DUMP_WED(WED_RTQM_R2H_MIB(0)),
+		DUMP_WED(WED_RTQM_R2Q_MIB(0)),
+		DUMP_WED(WED_RTQM_Q2H_MIB(0)),
+		DUMP_WED(WED_RTQM_R2H_MIB(1)),
+		DUMP_WED(WED_RTQM_R2Q_MIB(1)),
+		DUMP_WED(WED_RTQM_Q2H_MIB(1)),
+		DUMP_WED(WED_RTQM_Q2N_MIB),
+		DUMP_WED(WED_RTQM_Q2B_MIB),
+		DUMP_WED(WED_RTQM_PFDBK_MIB),
+
+		DUMP_END()
+	};
+
+	static const struct reg_dump regs_v3[] = {
+		DUMP_STR("WED RX RRO DATA"),
+		DUMP_WED_RING(WED_RRO_RX_D_RX(0)),
+		DUMP_WED_RING(WED_RRO_RX_D_RX(1)),
+
+		DUMP_STR("WED RX MSDU PAGE"),
+		DUMP_WED_RING(WED_RRO_MSDU_PG_CTRL0(0)),
+		DUMP_WED_RING(WED_RRO_MSDU_PG_CTRL0(1)),
+		DUMP_WED_RING(WED_RRO_MSDU_PG_CTRL0(2)),
+
+		DUMP_STR("WED RX IND CMD"),
+		DUMP_WED(WED_IND_CMD_RX_CTRL1),
+		DUMP_WED_MASK(WED_IND_CMD_RX_CTRL2, WED_IND_CMD_MAX_CNT),
+		DUMP_WED_MASK(WED_IND_CMD_RX_CTRL0, WED_IND_CMD_PROC_IDX),
+		DUMP_WED_MASK(RRO_IND_CMD_SIGNATURE, RRO_IND_CMD_DMA_IDX),
+		DUMP_WED_MASK(WED_IND_CMD_RX_CTRL0, WED_IND_CMD_MAGIC_CNT),
+		DUMP_WED_MASK(RRO_IND_CMD_SIGNATURE, RRO_IND_CMD_MAGIC_CNT),
+		DUMP_WED_MASK(WED_IND_CMD_RX_CTRL0,
+			      WED_IND_CMD_PREFETCH_FREE_CNT),
+		DUMP_WED_MASK(WED_RRO_CFG1, WED_RRO_CFG1_PARTICL_SE_ID),
+
+		DUMP_STR("WED ADDR ELEM"),
+		DUMP_WED(WED_ADDR_ELEM_CFG0),
+		DUMP_WED_MASK(WED_ADDR_ELEM_CFG1,
+			      WED_ADDR_ELEM_PREFETCH_FREE_CNT),
+
+		DUMP_STR("WED Route QM"),
+		DUMP_WED(WED_RTQM_ENQ_I2Q_DMAD_CNT),
+		DUMP_WED(WED_RTQM_ENQ_I2N_DMAD_CNT),
+		DUMP_WED(WED_RTQM_ENQ_I2Q_PKT_CNT),
+		DUMP_WED(WED_RTQM_ENQ_I2N_PKT_CNT),
+		DUMP_WED(WED_RTQM_ENQ_USED_ENTRY_CNT),
+		DUMP_WED(WED_RTQM_ENQ_ERR_CNT),
+
+		DUMP_WED(WED_RTQM_DEQ_DMAD_CNT),
+		DUMP_WED(WED_RTQM_DEQ_Q2I_DMAD_CNT),
+		DUMP_WED(WED_RTQM_DEQ_PKT_CNT),
+		DUMP_WED(WED_RTQM_DEQ_Q2I_PKT_CNT),
+		DUMP_WED(WED_RTQM_DEQ_USED_PFDBK_CNT),
+		DUMP_WED(WED_RTQM_DEQ_ERR_CNT),
+
+		DUMP_END()
+	};
+
+	static const struct reg_dump *regs_new_v2[] = {
+		&regs_common[0],
+		&regs_v2[0],
+		NULL,
+	};
+
+	static const struct reg_dump *regs_new_v3[] = {
+		&regs_common[0],
+		&regs_v3[0],
+		NULL,
 	};
 
 	struct mtk_wed_hw *hw = s->private;
 	struct mtk_wed_device *dev = hw->wed_dev;
+	const struct reg_dump **regs;
 
 	if (!dev)
 		return 0;
 
-	dump_wed_regs(s, dev, regs, ARRAY_SIZE(regs));
+	switch(dev->hw->version) {
+	case 2:
+		regs = regs_new_v2;
+		break;
+	case 3:
+		regs = regs_new_v3;
+		break;
+	default:
+		return 0;
+	}
+
+	dump_wed_regs(s, dev, regs);
 
 	return 0;
 }
@@ -248,6 +357,383 @@ mtk_wed_reg_get(void *data, u64 *val)
 DEFINE_DEBUGFS_ATTRIBUTE(fops_regval, mtk_wed_reg_get, mtk_wed_reg_set,
              "0x%08llx\n");
 
+static int
+wed_token_txd_show(struct seq_file *s, void *data)
+{
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+	struct dma_page_info *page_list = dev->tx_buf_ring.pages;
+	int token = dev->wlan.token_start;
+	u32 val = hw->token_id, size = 1;
+	int page_idx = (val - token) / 2;
+	int i;
+
+	if (val < token) {
+		size = val;
+		page_idx = 0;
+	}
+
+	for (i = 0; i < size; i += MTK_WED_BUF_PER_PAGE) {
+		void *page = page_list[page_idx++].addr;
+		void *buf;
+		int j;
+
+		if (!page)
+			break;
+
+		buf = page_to_virt(page);
+
+		for (j = 0; j < MTK_WED_BUF_PER_PAGE; j++) {
+			printk("[TXD]:token id = %d\n", token + 2 * (page_idx - 1) + j);
+			print_hex_dump(KERN_ERR , "", DUMP_PREFIX_OFFSET, 16, 1, (u8 *)buf, 128, false);
+			seq_printf(s, "\n");
+
+			buf += MTK_WED_BUF_SIZE;
+		}
+	}
+
+	return 0;
+}
+
+DEFINE_SHOW_ATTRIBUTE(wed_token_txd);
+
+static int
+wed_pao_show(struct seq_file *s, void *data)
+{
+	static const struct reg_dump regs_common[] = {
+		DUMP_STR("PAO AMDSU INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_FIFO_DMAD),
+
+		DUMP_STR("PAO AMDSU ENG0 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(0)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(0)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(0)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(0)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(0)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(0),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(0),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(0),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(0),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(0),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG1 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(1)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(1)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(1)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(1)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(1)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(1),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(1),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(1),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(2),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(2),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG2 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(2)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(2)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(2)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(2)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(2)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(2),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(2),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(2),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(2),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(2),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG3 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(3)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(3)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(3)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(3)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(3)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(3),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(3),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(3),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(3),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(3),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG4 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(4)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(4)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(4)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(4)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(4)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(4),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(4),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(4),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(4),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(4),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG5 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(5)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(5)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(5)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(5)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(5)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(5),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(5),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(5),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(5),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(5),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG6 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(6)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(6)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(6)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(6)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(6)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(6),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(6),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(6),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(6),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(6),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG7 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(7)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(7)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(7)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(7)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(7)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(7),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(7),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(7),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(7),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(4),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO AMDSU ENG8 INFO"),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_DMAD(8)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QFPL(8)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENI(8)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_QENO(8)),
+		DUMP_WED(WED_PAO_MON_AMSDU_ENG_MERG(8)),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(8),
+			      WED_PAO_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT8(8),
+			      WED_PAO_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(8),
+			      WED_PAO_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(8),
+			      WED_PAO_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_PAO_MON_AMSDU_ENG_CNT9(8),
+			      WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("PAO QMEM INFO"),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(0), WED_PAO_QMEM_FQ_CNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(0), WED_PAO_QMEM_SP_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(1), WED_PAO_QMEM_TID0_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(1), WED_PAO_QMEM_TID1_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(2), WED_PAO_QMEM_TID2_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(2), WED_PAO_QMEM_TID3_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(3), WED_PAO_QMEM_TID4_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(3), WED_PAO_QMEM_TID5_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(4), WED_PAO_QMEM_TID6_QCNT),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_CNT(4), WED_PAO_QMEM_TID7_QCNT),
+
+
+		DUMP_STR("PAO QMEM HEAD INFO"),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(0), WED_PAO_QMEM_FQ_HEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(0), WED_PAO_QMEM_SP_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(1), WED_PAO_QMEM_TID0_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(1), WED_PAO_QMEM_TID1_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(2), WED_PAO_QMEM_TID2_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(2), WED_PAO_QMEM_TID3_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(3), WED_PAO_QMEM_TID4_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(3), WED_PAO_QMEM_TID5_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(4), WED_PAO_QMEM_TID6_QHEAD),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(4), WED_PAO_QMEM_TID7_QHEAD),
+
+		DUMP_STR("PAO QMEM TAIL INFO"),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(5), WED_PAO_QMEM_FQ_TAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(5), WED_PAO_QMEM_SP_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(6), WED_PAO_QMEM_TID0_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(6), WED_PAO_QMEM_TID1_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(7), WED_PAO_QMEM_TID2_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(7), WED_PAO_QMEM_TID3_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(8), WED_PAO_QMEM_TID4_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(8), WED_PAO_QMEM_TID5_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(9), WED_PAO_QMEM_TID6_QTAIL),
+		DUMP_WED_MASK(WED_PAO_MON_QMEM_PTR(9), WED_PAO_QMEM_TID7_QTAIL),
+
+		DUMP_STR("PAO HIFTXD MSDU INFO"),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(1)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(2)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(3)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(4)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(5)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(6)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(7)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(8)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(9)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(10)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(11)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(12)),
+		DUMP_WED(WED_PAO_MON_HIFTXD_FETCH_MSDU(13)),
+		DUMP_END()
+	};
+
+	static const struct reg_dump *regs[] = {
+		&regs_common[0],
+		NULL,
+	};
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+
+	if (!dev)
+		return 0;
+
+	dump_wed_regs(s, dev, regs);
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(wed_pao);
+
+static int
+wed_rtqm_show(struct seq_file *s, void *data)
+{
+	static const struct reg_dump regs_common[] = {
+		DUMP_STR("WED Route QM IGRS0(N2H + Recycle)"),
+		DUMP_WED(WED_RTQM_IGRS0_I2HW_DMAD_CNT),
+		DUMP_WED(WED_RTQM_IGRS0_I2H_DMAD_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS0_I2H_DMAD_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS0_I2HW_PKT_CNT),
+		DUMP_WED(WED_RTQM_IGRS0_I2H_PKT_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS0_I2H_PKT_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS0_FDROP_CNT),
+
+
+		DUMP_STR("WED Route QM IGRS1(Legacy)"),
+		DUMP_WED(WED_RTQM_IGRS1_I2HW_DMAD_CNT),
+		DUMP_WED(WED_RTQM_IGRS1_I2H_DMAD_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS1_I2H_DMAD_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS1_I2HW_PKT_CNT),
+		DUMP_WED(WED_RTQM_IGRS1_I2H_PKT_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS1_I2H_PKT_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS1_FDROP_CNT),
+
+		DUMP_STR("WED Route QM IGRS2(RRO3.0)"),
+		DUMP_WED(WED_RTQM_IGRS2_I2HW_DMAD_CNT),
+		DUMP_WED(WED_RTQM_IGRS2_I2H_DMAD_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS2_I2H_DMAD_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS2_I2HW_PKT_CNT),
+		DUMP_WED(WED_RTQM_IGRS2_I2H_PKT_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS2_I2H_PKT_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS2_FDROP_CNT),
+
+		DUMP_STR("WED Route QM IGRS3(DEBUG)"),
+		DUMP_WED(WED_RTQM_IGRS2_I2HW_DMAD_CNT),
+		DUMP_WED(WED_RTQM_IGRS3_I2H_DMAD_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS3_I2H_DMAD_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS3_I2HW_PKT_CNT),
+		DUMP_WED(WED_RTQM_IGRS3_I2H_PKT_CNT(0)),
+		DUMP_WED(WED_RTQM_IGRS3_I2H_PKT_CNT(1)),
+		DUMP_WED(WED_RTQM_IGRS3_FDROP_CNT),
+
+		DUMP_END()
+	};
+
+	static const struct reg_dump *regs[] = {
+		&regs_common[0],
+		NULL,
+	};
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+
+	if (!dev)
+		return 0;
+
+	dump_wed_regs(s, dev, regs);
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(wed_rtqm);
+
+
+static int
+wed_rro_show(struct seq_file *s, void *data)
+{
+	static const struct reg_dump regs_common[] = {
+		DUMP_STR("RRO/IND CMD CNT"),
+		DUMP_WED(WED_RX_IND_CMD_CNT(1)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(2)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(3)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(4)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(5)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(6)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(7)),
+		DUMP_WED(WED_RX_IND_CMD_CNT(8)),
+		DUMP_WED_MASK(WED_RX_IND_CMD_CNT(9),
+			      WED_IND_CMD_MAGIC_CNT_FAIL_CNT),
+
+		DUMP_WED(WED_RX_ADDR_ELEM_CNT(0)),
+		DUMP_WED_MASK(WED_RX_ADDR_ELEM_CNT(1),
+			      WED_ADDR_ELEM_SIG_FAIL_CNT),
+		DUMP_WED(WED_RX_MSDU_PG_CNT(1)),
+		DUMP_WED(WED_RX_MSDU_PG_CNT(2)),
+		DUMP_WED(WED_RX_MSDU_PG_CNT(3)),
+		DUMP_WED(WED_RX_MSDU_PG_CNT(4)),
+		DUMP_WED(WED_RX_MSDU_PG_CNT(5)),
+		DUMP_WED_MASK(WED_RX_PN_CHK_CNT,
+			      WED_PN_CHK_FAIL_CNT),
+
+		DUMP_END()
+	};
+
+	static const struct reg_dump *regs[] = {
+		&regs_common[0],
+		NULL,
+	};
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+
+	if (!dev)
+		return 0;
+
+	dump_wed_regs(s, dev, regs);
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(wed_rro);
+
 void mtk_wed_hw_add_debugfs(struct mtk_wed_hw *hw)
 {
 	struct dentry *dir;
@@ -261,8 +747,18 @@ void mtk_wed_hw_add_debugfs(struct mtk_wed_hw *hw)
 	debugfs_create_u32("regidx", 0600, dir, &hw->debugfs_reg);
 	debugfs_create_file_unsafe("regval", 0600, dir, hw, &fops_regval);
 	debugfs_create_file_unsafe("txinfo", 0400, dir, hw, &wed_txinfo_fops);
-	debugfs_create_file_unsafe("rxinfo", 0400, dir, hw, &wed_rxinfo_fops);
-	if (hw->ver != MTK_WED_V1) {
+	debugfs_create_u32("token_id", 0600, dir, &hw->token_id);
+	debugfs_create_file_unsafe("token_txd", 0600, dir, hw, &wed_token_txd_fops);
+
+	if (hw->version == 3)
+		debugfs_create_file_unsafe("pao", 0400, dir, hw, &wed_pao_fops);
+
+	if (hw->version != 1) {
+		debugfs_create_file_unsafe("rxinfo", 0400, dir, hw, &wed_rxinfo_fops);
+		if (hw->version == 3) {
+			debugfs_create_file_unsafe("rtqm", 0400, dir, hw, &wed_rtqm_fops);
+			debugfs_create_file_unsafe("rro", 0400, dir, hw, &wed_rro_fops);
+		}
 		wed_wo_mcu_debugfs(hw, dir);
 	}
 }
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_mcu.c b/drivers/net/ethernet/mediatek/mtk_wed_mcu.c
index 96e30a3..055594d 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_mcu.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed_mcu.c
@@ -242,7 +242,7 @@ mtk_wed_load_firmware(struct mtk_wed_wo *wo)
 	u32 ofs = 0;
 	u32 boot_cr, val;
 
-	mcu = wo->hw->index ? MT7986_FIRMWARE_WO_2 : MT7986_FIRMWARE_WO_1;
+	mcu = wo->hw->index ? MTK_FIRMWARE_WO_1 : MTK_FIRMWARE_WO_0;
 
 	ret = request_firmware(&fw, mcu, wo->hw->dev);
 	if (ret)
@@ -289,8 +289,12 @@ mtk_wed_load_firmware(struct mtk_wed_wo *wo)
 	}
 
 	/* write the start address */
-	boot_cr = wo->hw->index ?
-		WOX_MCU_CFG_LS_WA_BOOT_ADDR_ADDR : WOX_MCU_CFG_LS_WM_BOOT_ADDR_ADDR;
+	if (wo->hw->version == 3)
+		boot_cr = WOX_MCU_CFG_LS_WM_BOOT_ADDR_ADDR;
+	else
+		boot_cr = wo->hw->index ?
+			WOX_MCU_CFG_LS_WA_BOOT_ADDR_ADDR : WOX_MCU_CFG_LS_WM_BOOT_ADDR_ADDR;
+
 	wo_w32(wo, boot_cr, (wo->region[WO_REGION_EMI].addr_pa >> 16));
 
 	/* wo firmware reset */
@@ -298,8 +302,7 @@ mtk_wed_load_firmware(struct mtk_wed_wo *wo)
 
 	val = wo_r32(wo, WOX_MCU_CFG_LS_WF_MCU_CFG_WM_WA_ADDR);
 
-	val |= wo->hw->index ? WOX_MCU_CFG_LS_WF_MCU_CFG_WM_WA_WA_CPU_RSTB_MASK :
-		WOX_MCU_CFG_LS_WF_MCU_CFG_WM_WA_WM_CPU_RSTB_MASK;
+	val |= WOX_MCU_CFG_LS_WF_MCU_CFG_WM_WA_WM_CPU_RSTB_MASK;
 
 	wo_w32(wo, WOX_MCU_CFG_LS_WF_MCU_CFG_WM_WA_ADDR, val);
 
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_mcu.h b/drivers/net/ethernet/mediatek/mtk_wed_mcu.h
index 19e1199..c07bdb6 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_mcu.h
+++ b/drivers/net/ethernet/mediatek/mtk_wed_mcu.h
@@ -16,8 +16,9 @@
 #define WARP_OK_STATUS (0)
 #define WARP_ALREADY_DONE_STATUS (1)
 
-#define MT7986_FIRMWARE_WO_1		"mediatek/mt7986_wo_0.bin"
-#define MT7986_FIRMWARE_WO_2		"mediatek/mt7986_wo_1.bin"
+#define MTK_FIRMWARE_WO_0		"mediatek/mtk_wo_0.bin"
+#define MTK_FIRMWARE_WO_1		"mediatek/mtk_wo_1.bin"
+#define MTK_FIRMWARE_WO_2		"mediatek/mtk_wo_2.bin"
 
 #define WOCPU_EMI_DEV_NODE		"mediatek,wocpu_emi"
 #define WOCPU_ILM_DEV_NODE		"mediatek,wocpu_ilm"
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_regs.h b/drivers/net/ethernet/mediatek/mtk_wed_regs.h
index 403a36b..25be547 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_regs.h
+++ b/drivers/net/ethernet/mediatek/mtk_wed_regs.h
@@ -20,6 +20,9 @@
 #define MTK_WDMA_DESC_CTRL_DMA_DONE		BIT(31)
 #define MTK_WED_RX_BM_TOKEN			GENMASK(31, 16)
 
+#define MTK_WDMA_TXD0_DESC_INFO_DMA_DONE	BIT(29)
+#define MTK_WDMA_TXD1_DESC_INFO_DMA_DONE	BIT(31)
+
 struct mtk_wdma_desc {
 	__le32 buf0;
 	__le32 ctrl;
@@ -51,6 +54,7 @@ struct mtk_wdma_desc {
 #define MTK_WED_RESET_WDMA_INT_AGENT			BIT(19)
 #define MTK_WED_RESET_RX_RRO_QM				BIT(20)
 #define MTK_WED_RESET_RX_ROUTE_QM			BIT(21)
+#define MTK_WED_RESET_TX_PAO				BIT(22)
 #define MTK_WED_RESET_WED				BIT(31)
 
 #define MTK_WED_CTRL					0x00c
@@ -58,6 +62,9 @@ struct mtk_wdma_desc {
 #define MTK_WED_CTRL_WPDMA_INT_AGENT_BUSY		BIT(1)
 #define MTK_WED_CTRL_WDMA_INT_AGENT_EN			BIT(2)
 #define MTK_WED_CTRL_WDMA_INT_AGENT_BUSY		BIT(3)
+#define MTK_WED_CTRL_WED_RX_IND_CMD_EN			BIT(5)
+#define MTK_WED_CTRL_WED_RX_PG_BM_EN			BIT(6)
+#define MTK_WED_CTRL_WED_RX_PG_BM_BUSU			BIT(7)
 #define MTK_WED_CTRL_WED_TX_BM_EN			BIT(8)
 #define MTK_WED_CTRL_WED_TX_BM_BUSY			BIT(9)
 #define MTK_WED_CTRL_WED_TX_FREE_AGENT_EN		BIT(10)
@@ -68,9 +75,14 @@ struct mtk_wdma_desc {
 #define MTK_WED_CTRL_RX_RRO_QM_BUSY			BIT(15)
 #define MTK_WED_CTRL_RX_ROUTE_QM_EN			BIT(16)
 #define MTK_WED_CTRL_RX_ROUTE_QM_BUSY			BIT(17)
+#define MTK_WED_CTRL_TX_TKID_ALI_EN			BIT(20)
+#define MTK_WED_CTRL_TX_TKID_ALI_BUSY			BIT(21)
+#define MTK_WED_CTRL_TX_PAO_EN				BIT(22)
+#define MTK_WED_CTRL_TX_PAO_BUSY			BIT(23)
 #define MTK_WED_CTRL_FINAL_DIDX_READ			BIT(24)
 #define MTK_WED_CTRL_ETH_DMAD_FMT			BIT(25)
 #define MTK_WED_CTRL_MIB_READ_CLEAR			BIT(28)
+#define MTK_WED_CTRL_FLD_MIB_RD_CLR			BIT(28)
 
 #define MTK_WED_EXT_INT_STATUS				0x020
 #define MTK_WED_EXT_INT_STATUS_TF_LEN_ERR		BIT(0)
@@ -78,12 +90,10 @@ struct mtk_wdma_desc {
 #define MTK_WED_EXT_INT_STATUS_TKID_TITO_INVALID	BIT(4)
 #define MTK_WED_EXT_INT_STATUS_TX_FBUF_LO_TH		BIT(8)
 #define MTK_WED_EXT_INT_STATUS_TX_FBUF_HI_TH		BIT(9)
-#if defined(CONFIG_MEDIATEK_NETSYS_V2)
-#define MTK_WED_EXT_INT_STATUS_TX_TKID_LO_TH		BIT(10)
-#define MTK_WED_EXT_INT_STATUS_TX_TKID_HI_TH		BIT(11)
-#endif
-#define MTK_WED_EXT_INT_STATUS_RX_FREE_AT_EMPTY		BIT(12)
-#define MTK_WED_EXT_INT_STATUS_RX_FBUF_DMAD_ER		BIT(13)
+#define MTK_WED_EXT_INT_STATUS_RX_FBUF_LO_TH2		BIT(10)
+#define MTK_WED_EXT_INT_STATUS_RX_FBUF_HI_TH2		BIT(11)
+#define MTK_WED_EXT_INT_STATUS_RX_FBUF_LO_TH		BIT(12)
+#define MTK_WED_EXT_INT_STATUS_RX_FBUF_HI_TH		BIT(13)
 #define MTK_WED_EXT_INT_STATUS_RX_DRV_R_RESP_ERR	BIT(16)
 #define MTK_WED_EXT_INT_STATUS_RX_DRV_W_RESP_ERR	BIT(17)
 #define MTK_WED_EXT_INT_STATUS_RX_DRV_COHERENT		BIT(18)
@@ -100,17 +110,15 @@ struct mtk_wdma_desc {
 #define MTK_WED_EXT_INT_STATUS_ERROR_MASK		(MTK_WED_EXT_INT_STATUS_TF_LEN_ERR | \
 							 MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD | \
 							 MTK_WED_EXT_INT_STATUS_TKID_TITO_INVALID | \
-							 MTK_WED_EXT_INT_STATUS_RX_FREE_AT_EMPTY | \
-							 MTK_WED_EXT_INT_STATUS_RX_FBUF_DMAD_ER | \
 							 MTK_WED_EXT_INT_STATUS_RX_DRV_R_RESP_ERR | \
 							 MTK_WED_EXT_INT_STATUS_RX_DRV_W_RESP_ERR | \
 							 MTK_WED_EXT_INT_STATUS_RX_DRV_INIT_WDMA_EN | \
-							 MTK_WED_EXT_INT_STATUS_TX_DMA_R_RESP_ERR | \
-							 MTK_WED_EXT_INT_STATUS_TX_DMA_W_RESP_ERR)
+							 MTK_WED_EXT_INT_STATUS_TX_DMA_R_RESP_ERR)
 
 #define MTK_WED_EXT_INT_MASK				0x028
 #define MTK_WED_EXT_INT_MASK1				0x02c
 #define MTK_WED_EXT_INT_MASK2				0x030
+#define MTK_WED_EXT_INT_MASK3				0x034
 
 #define MTK_WED_STATUS					0x060
 #define MTK_WED_STATUS_TX				GENMASK(15, 8)
@@ -118,9 +126,14 @@ struct mtk_wdma_desc {
 #define MTK_WED_TX_BM_CTRL				0x080
 #define MTK_WED_TX_BM_CTRL_VLD_GRP_NUM			GENMASK(6, 0)
 #define MTK_WED_TX_BM_CTRL_RSV_GRP_NUM			GENMASK(22, 16)
+#define MTK_WED_TX_BM_CTRL_LEGACY_EN			BIT(26)
+#define MTK_WED_TX_TKID_CTRL_FREE_FORMAT		BIT(27)
 #define MTK_WED_TX_BM_CTRL_PAUSE			BIT(28)
 
 #define MTK_WED_TX_BM_BASE				0x084
+#define MTK_WED_TX_BM_INIT_PTR				0x088
+#define MTK_WED_TX_BM_SW_TAIL_IDX			GENMASK(16, 0)
+#define MTK_WED_TX_BM_INIT_SW_TAIL_IDX			BIT(16)
 
 #define MTK_WED_TX_BM_BUF_LEN				0x08c
 
@@ -134,22 +147,24 @@ struct mtk_wdma_desc {
 #if defined(CONFIG_MEDIATEK_NETSYS_V2)
 #define MTK_WED_TX_BM_DYN_THR_LO			GENMASK(8, 0)
 #define MTK_WED_TX_BM_DYN_THR_HI			GENMASK(24, 16)
-
-#define MTK_WED_TX_BM_TKID				0x0c8
-#define MTK_WED_TX_BM_TKID_START			GENMASK(15, 0)
-#define MTK_WED_TX_BM_TKID_END				GENMASK(31, 16)
 #else
 #define MTK_WED_TX_BM_DYN_THR_LO			GENMASK(6, 0)
 #define MTK_WED_TX_BM_DYN_THR_HI			GENMASK(22, 16)
+#endif
 
-#define MTK_WED_TX_BM_TKID				0x088
+#define MTK_WED_TX_BM_TKID				0x0c8
 #define MTK_WED_TX_BM_TKID_START			GENMASK(15, 0)
 #define MTK_WED_TX_BM_TKID_END				GENMASK(31, 16)
-#endif
 
 #define MTK_WED_TX_TKID_CTRL				0x0c0
+#if defined(CONFIG_MEDIATEK_NETSYS_V3)
+#define MTK_WED_TX_TKID_CTRL_VLD_GRP_NUM		GENMASK(7, 0)
+#define MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM		GENMASK(23, 16)
+#else
 #define MTK_WED_TX_TKID_CTRL_VLD_GRP_NUM		GENMASK(6, 0)
 #define MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM		GENMASK(22, 16)
+#endif
+
 #define MTK_WED_TX_TKID_CTRL_PAUSE			BIT(28)
 
 #define MTK_WED_TX_TKID_DYN_THR				0x0e0
@@ -220,12 +235,15 @@ struct mtk_wdma_desc {
 #define MTK_WED_WPDMA_GLO_CFG_RX_DRV_R1_PKT_PROC	BIT(5)
 #define MTK_WED_WPDMA_GLO_CFG_RX_DRV_R0_CRX_SYNC	BIT(6)
 #define MTK_WED_WPDMA_GLO_CFG_RX_DRV_R1_CRX_SYNC	BIT(7)
-#define MTK_WED_WPDMA_GLO_CFG_RX_DRV_EVENT_PKT_FMT_VER	GENMASK(18, 16)
+#define MTK_WED_WPDMA_GLO_CFG_RX_DRV_EVENT_PKT_FMT_VER	GENMASK(15, 12)
+#define MTK_WED_WPDMA_GLO_CFG_RX_DRV_UNS_VER_FORCE_4	BIT(18)
 #define MTK_WED_WPDMA_GLO_CFG_RX_DRV_UNSUPPORT_FMT	BIT(19)
-#define MTK_WED_WPDMA_GLO_CFG_RX_DRV_UEVENT_PKT_FMT_CHK BIT(20)
+#define MTK_WED_WPDMA_GLO_CFG_RX_DRV_EVENT_PKT_FMT_CHK	BIT(20)
 #define MTK_WED_WPDMA_GLO_CFG_RX_DDONE2_WR		BIT(21)
 #define MTK_WED_WPDMA_GLO_CFG_TX_TKID_KEEP		BIT(24)
+#define MTK_WED_WPDMA_GLO_CFG_TX_DDONE_CHK_LAST		BIT(25)
 #define MTK_WED_WPDMA_GLO_CFG_TX_DMAD_DW3_PREV		BIT(28)
+#define MTK_WED_WPDMA_GLO_CFG_TX_DDONE_CHK		BIT(30)
 
 /* CONFIG_MEDIATEK_NETSYS_V1 */
 #define MTK_WED_WPDMA_GLO_CFG_RX_BT_SIZE		GENMASK(5, 4)
@@ -288,9 +306,11 @@ struct mtk_wdma_desc {
 #define MTK_WED_PCIE_INT_TRIGGER_STATUS			BIT(16)
 
 #define MTK_WED_PCIE_INT_CTRL				0x57c
-#define MTK_WED_PCIE_INT_CTRL_MSK_EN_POLA		BIT(20)
-#define MTK_WED_PCIE_INT_CTRL_SRC_SEL			GENMASK(17, 16)
 #define MTK_WED_PCIE_INT_CTRL_POLL_EN 			GENMASK(13, 12)
+#define MTK_WED_PCIE_INT_CTRL_SRC_SEL			GENMASK(17, 16)
+#define MTK_WED_PCIE_INT_CTRL_MSK_EN_POLA		BIT(20)
+#define MTK_WED_PCIE_INT_CTRL_MSK_IRQ_FILTER		BIT(21)
+
 #define MTK_WED_WPDMA_CFG_BASE				0x580
 #define MTK_WED_WPDMA_CFG_INT_MASK			0x584
 #define MTK_WED_WPDMA_CFG_TX				0x588
@@ -319,20 +339,50 @@ struct mtk_wdma_desc {
 #define MTK_WED_WPDMA_RX_D_RST_DRV_IDX			GENMASK(25, 24)
 
 #define MTK_WED_WPDMA_RX_GLO_CFG			0x76c
-#define MTK_WED_WPDMA_RX_RING				0x770
+#if defined(CONFIG_MEDIATEK_NETSYS_V2)
+#define MTK_WED_WPDMA_RX_RING0				0x770
+#else
+#define MTK_WED_WPDMA_RX_RING0				0x7d0
+#endif
+#define MTK_WED_WPDMA_RX_RING1				0x7d8
 
 #define MTK_WED_WPDMA_RX_D_MIB(_n)			(0x774 + (_n) * 4)
 #define MTK_WED_WPDMA_RX_D_PROCESSED_MIB(_n)		(0x784 + (_n) * 4)
 #define MTK_WED_WPDMA_RX_D_COHERENT_MIB			0x78c
 
+#define MTK_WED_WPDMA_RX_D_PREF_CFG			0x7b4
+#define MTK_WED_WPDMA_RX_D_PREF_EN			BIT(0)
+#define MTK_WED_WPDMA_RX_D_PREF_BURST_SIZE		GENMASK(12, 8)
+#define MTK_WED_WPDMA_RX_D_PREF_LOW_THRES		GENMASK(21, 16)
+
+#define MTK_WED_WPDMA_RX_D_PREF_RX0_SIDX		0x7b8
+#define MTK_WED_WPDMA_RX_D_PREF_SIDX_IDX_CLR		BIT(15)
+
+#define MTK_WED_WPDMA_RX_D_PREF_RX1_SIDX		0x7bc
+
+#define MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG		0x7c0
+#define MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG_R0_CLR		BIT(0)
+#define MTK_WED_WPDMA_RX_D_PREF_FIFO_CFG_R1_CLR		BIT(16)
+
 #define MTK_WED_WDMA_RING_TX				0x800
 
 #define MTK_WED_WDMA_TX_MIB				0x810
 
-
 #define MTK_WED_WDMA_RING_RX(_n)			(0x900 + (_n) * 0x10)
 #define MTK_WED_WDMA_RX_THRES(_n)			(0x940 + (_n) * 0x4)
 
+#define MTK_WED_WDMA_RX_PREF_CFG			0x950
+#define MTK_WED_WDMA_RX_PREF_EN				BIT(0)
+#define MTK_WED_WDMA_RX_PREF_BURST_SIZE			GENMASK(12, 8)
+#define MTK_WED_WDMA_RX_PREF_LOW_THRES			GENMASK(21, 16)
+#define MTK_WED_WDMA_RX_PREF_RX0_SIDX_CLR		BIT(24)
+#define MTK_WED_WDMA_RX_PREF_RX1_SIDX_CLR		BIT(25)
+#define MTK_WED_WDMA_RX_PREF_DDONE2_EN			BIT(26)
+
+#define MTK_WED_WDMA_RX_PREF_FIFO_CFG			0x95C
+#define MTK_WED_WDMA_RX_PREF_FIFO_RX0_CLR		BIT(0)
+#define MTK_WED_WDMA_RX_PREF_FIFO_RX1_CLR		BIT(16)
+
 #define MTK_WED_WDMA_GLO_CFG				0xa04
 #define MTK_WED_WDMA_GLO_CFG_TX_DRV_EN			BIT(0)
 #define MTK_WED_WDMA_GLO_CFG_TX_DDONE_CHK		BIT(1)
@@ -365,6 +415,7 @@ struct mtk_wdma_desc {
 #define MTK_WED_WDMA_INT_TRIGGER_RX_DONE		GENMASK(17, 16)
 
 #define MTK_WED_WDMA_INT_CTRL				0xa2c
+#define MTK_WED_WDMA_INT_POLL_PRD			GENMASK(7, 0)
 #define MTK_WED_WDMA_INT_POLL_SRC_SEL			GENMASK(17, 16)
 
 #define MTK_WED_WDMA_CFG_BASE				0xaa0
@@ -426,6 +477,18 @@ struct mtk_wdma_desc {
 #define MTK_WDMA_INT_GRP1				0x250
 #define MTK_WDMA_INT_GRP2				0x254
 
+#define MTK_WDMA_PREF_TX_CFG				0x2d0
+#define MTK_WDMA_PREF_TX_CFG_PREF_EN			BIT(0)
+
+#define MTK_WDMA_PREF_RX_CFG				0x2dc
+#define MTK_WDMA_PREF_RX_CFG_PREF_EN			BIT(0)
+
+#define MTK_WDMA_WRBK_TX_CFG				0x300
+#define MTK_WDMA_WRBK_TX_CFG_WRBK_EN			BIT(30)
+
+#define MTK_WDMA_WRBK_RX_CFG				0x344
+#define MTK_WDMA_WRBK_RX_CFG_WRBK_EN			BIT(30)
+
 #define MTK_PCIE_MIRROR_MAP(n)				((n) ? 0x4 : 0x0)
 #define MTK_PCIE_MIRROR_MAP_EN				BIT(0)
 #define MTK_PCIE_MIRROR_MAP_WED_ID			BIT(1)
@@ -439,6 +502,31 @@ struct mtk_wdma_desc {
 #define MTK_WED_RTQM_Q_DBG_BYPASS			BIT(5)
 #define MTK_WED_RTQM_TXDMAD_FPORT			GENMASK(23, 20)
 
+#define MTK_WED_RTQM_IGRS0_I2HW_DMAD_CNT		0xb1c
+#define MTK_WED_RTQM_IGRS0_I2H_DMAD_CNT(_n)		(0xb20 + (_n) * 0x4)
+#define	MTK_WED_RTQM_IGRS0_I2HW_PKT_CNT			0xb28
+#define MTK_WED_RTQM_IGRS0_I2H_PKT_CNT(_n)		(0xb2c + (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS0_FDROP_CNT			0xb34
+
+
+#define MTK_WED_RTQM_IGRS1_I2HW_DMAD_CNT		0xb44
+#define MTK_WED_RTQM_IGRS1_I2H_DMAD_CNT(_n)		(0xb48 + (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS1_I2HW_PKT_CNT			0xb50
+#define MTK_WED_RTQM_IGRS1_I2H_PKT_CNT(_n)		(0xb54+ (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS1_FDROP_CNT			0xb5c
+
+#define MTK_WED_RTQM_IGRS2_I2HW_DMAD_CNT		0xb6c
+#define MTK_WED_RTQM_IGRS2_I2H_DMAD_CNT(_n)		(0xb70 + (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS2_I2HW_PKT_CNT			0xb78
+#define MTK_WED_RTQM_IGRS2_I2H_PKT_CNT(_n)		(0xb7c+ (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS2_FDROP_CNT			0xb84
+
+#define MTK_WED_RTQM_IGRS3_I2HW_DMAD_CNT		0xb94
+#define MTK_WED_RTQM_IGRS3_I2H_DMAD_CNT(_n)		(0xb98 + (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS3_I2HW_PKT_CNT			0xba0
+#define MTK_WED_RTQM_IGRS3_I2H_PKT_CNT(_n)		(0xba4+ (_n) * 0x4)
+#define MTK_WED_RTQM_IGRS3_FDROP_CNT			0xbac
+
 #define MTK_WED_RTQM_R2H_MIB(_n)			(0xb70 + (_n) * 0x4)
 #define MTK_WED_RTQM_R2Q_MIB(_n)			(0xb78 + (_n) * 0x4)
 #define MTK_WED_RTQM_Q2N_MIB				0xb80
@@ -447,6 +535,24 @@ struct mtk_wdma_desc {
 #define MTK_WED_RTQM_Q2B_MIB				0xb8c
 #define MTK_WED_RTQM_PFDBK_MIB				0xb90
 
+#define MTK_WED_RTQM_ENQ_CFG0				0xbb8
+#define MTK_WED_RTQM_ENQ_CFG_TXDMAD_FPORT		GENMASK(15, 12)
+
+#define MTK_WED_RTQM_FDROP_MIB				0xb84
+#define MTK_WED_RTQM_ENQ_I2Q_DMAD_CNT			0xbbc
+#define MTK_WED_RTQM_ENQ_I2N_DMAD_CNT			0xbc0
+#define MTK_WED_RTQM_ENQ_I2Q_PKT_CNT			0xbc4
+#define MTK_WED_RTQM_ENQ_I2N_PKT_CNT			0xbc8
+#define MTK_WED_RTQM_ENQ_USED_ENTRY_CNT			0xbcc
+#define MTK_WED_RTQM_ENQ_ERR_CNT			0xbd0
+
+#define MTK_WED_RTQM_DEQ_DMAD_CNT			0xbd8
+#define MTK_WED_RTQM_DEQ_Q2I_DMAD_CNT			0xbdc
+#define MTK_WED_RTQM_DEQ_PKT_CNT			0xbe0
+#define MTK_WED_RTQM_DEQ_Q2I_PKT_CNT			0xbe4
+#define MTK_WED_RTQM_DEQ_USED_PFDBK_CNT			0xbe8
+#define MTK_WED_RTQM_DEQ_ERR_CNT			0xbec
+
 #define MTK_WED_RROQM_GLO_CFG				0xc04
 #define MTK_WED_RROQM_RST_IDX				0xc08
 #define MTK_WED_RROQM_RST_IDX_MIOD 			BIT(0)
@@ -487,8 +593,8 @@ struct mtk_wdma_desc {
 #define MTK_WED_RX_BM_BASE				0xd84
 #define MTK_WED_RX_BM_INIT_PTR				0xd88
 #define MTK_WED_RX_BM_PTR	      			0xd8c
-#define MTK_WED_RX_BM_PTR_HEAD				GENMASK(32, 16)
 #define MTK_WED_RX_BM_PTR_TAIL				GENMASK(15, 0)
+#define MTK_WED_RX_BM_PTR_HEAD				GENMASK(32, 16)
 
 #define MTK_WED_RX_BM_BLEN	      			0xd90
 #define MTK_WED_RX_BM_STS				0xd94
@@ -496,7 +602,193 @@ struct mtk_wdma_desc {
 #define MTK_WED_RX_BM_INTF				0xd9c
 #define MTK_WED_RX_BM_ERR_STS				0xda8
 
+#define MTK_RRO_IND_CMD_SIGNATURE			0xe00
+#define MTK_RRO_IND_CMD_DMA_IDX				GENMASK(11, 0)
+#define MTK_RRO_IND_CMD_MAGIC_CNT			GENMASK(30, 28)
+
+#define MTK_WED_IND_CMD_RX_CTRL0			0xe04
+#define MTK_WED_IND_CMD_PROC_IDX			GENMASK(11, 0)
+#define MTK_WED_IND_CMD_PREFETCH_FREE_CNT		GENMASK(19, 16)
+#define MTK_WED_IND_CMD_MAGIC_CNT			GENMASK(30, 28)
+
+#define MTK_WED_IND_CMD_RX_CTRL1			0xe08
+#define MTK_WED_IND_CMD_RX_CTRL2			0xe0c
+#define MTK_WED_IND_CMD_MAX_CNT				GENMASK(11, 0)
+#define MTK_WED_IND_CMD_BASE_M				GENMASK(19, 16)
+
+#define MTK_WED_RRO_CFG0				0xe10
+#define MTK_WED_RRO_CFG1				0xe14
+#define MTK_WED_RRO_CFG1_MAX_WIN_SZ			GENMASK(31, 29)
+#define MTK_WED_RRO_CFG1_ACK_SN_BASE_M			GENMASK(19, 16)
+#define MTK_WED_RRO_CFG1_PARTICL_SE_ID			GENMASK(11, 0)
+
+#define MTK_WED_ADDR_ELEM_CFG0				0xe18
+#define MTK_WED_ADDR_ELEM_CFG1				0xe1c
+#define MTK_WED_ADDR_ELEM_PREFETCH_FREE_CNT		GENMASK(19, 16)
+
+#define MTK_WED_ADDR_ELEM_TBL_CFG 			0xe20
+#define MTK_WED_ADDR_ELEM_TBL_OFFSET			GENMASK(6, 0)
+#define MTK_WED_ADDR_ELEM_TBL_RD_RDY			BIT(28)
+#define MTK_WED_ADDR_ELEM_TBL_WR_RDY			BIT(29)
+#define MTK_WED_ADDR_ELEM_TBL_RD			BIT(30)
+#define MTK_WED_ADDR_ELEM_TBL_WR			BIT(31)
+
+#define MTK_WED_RADDR_ELEM_TBL_WDATA 			0xe24
+#define MTK_WED_RADDR_ELEM_TBL_RDATA 			0xe28
+
+#define MTK_WED_PN_CHECK_CFG 				0xe30
+#define MTK_WED_PN_CHECK_SE_ID				GENMASK(11, 0)
+#define MTK_WED_PN_CHECK_RD_RDY				BIT(28)
+#define MTK_WED_PN_CHECK_WR_RDY				BIT(29)
+#define MTK_WED_PN_CHECK_RD				BIT(30)
+#define MTK_WED_PN_CHECK_WR				BIT(31)
+
+#define MTK_WED_PN_CHECK_WDATA_M 			0xe38
+#define MTK_WED_PN_CHECK_IS_FIRST			BIT(17)
+
+#define MTK_WED_RRO_MSDU_PG_RING_CFG(_n)		(0xe44 + (_n) * 0x8)
+
+#define MTK_WED_RRO_MSDU_PG_RING2_CFG			0xe58
+#define MTK_WED_RRO_MSDU_PG_DRV_CLR			BIT(26)
+#define MTK_WED_RRO_MSDU_PG_DRV_EN			BIT(31)
+
+#define MTK_WED_RRO_MSDU_PG_CTRL0(_n)			(0xe5c + (_n) * 0xc)
+#define MTK_WED_RRO_MSDU_PG_CTRL1(_n)			(0xe60 + (_n) * 0xc)
+#define MTK_WED_RRO_MSDU_PG_CTRL2(_n)			(0xe64 + (_n) * 0xc)
+
+#define MTK_WED_RRO_RX_D_RX(_n)				(0xe80 + (_n) * 0x10)
+
+#define MTK_WED_RRO_RX_MAGIC_CNT			BIT(13)
+
+#define MTK_WED_RRO_RX_D_CFG(_n)			(0xea0 + (_n) * 0x4)
+#define MTK_WED_RRO_RX_D_DRV_CLR			BIT(26)
+#define MTK_WED_RRO_RX_D_DRV_EN				BIT(31)
+
+#define MTK_WED_RRO_PG_BM_RX_DMAM			0xeb0
+#define MTK_WED_RRO_PG_BM_RX_SDL0			GENMASK(13, 0)
+
+#define MTK_WED_RRO_PG_BM_BASE				0xeb4
+#define MTK_WED_RRO_PG_BM_INIT_PTR			0xeb8
+#define MTK_WED_RRO_PG_BM_SW_TAIL_IDX			GENMASK(15, 0)
+#define MTK_WED_RRO_PG_BM_INIT_SW_TAIL_IDX		BIT(16)
+
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX			0xeec
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX0_EN		BIT(0)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX0_CLR		BIT(1)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX0_DONE_TRIG	GENMASK(6, 2)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX1_EN		BIT(8)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX1_CLR		BIT(9)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_RX1_DONE_TRIG	GENMASK(14, 10)
+
+#define MTK_WED_WPDMA_INT_CTRL_RRO_MSDU_PG		0xef4
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG0_EN		BIT(0)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG0_CLR		BIT(1)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG0_DONE_TRIG	GENMASK(6, 2)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG1_EN		BIT(8)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG1_CLR		BIT(9)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG1_DONE_TRIG	GENMASK(14, 10)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG2_EN		BIT(16)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG2_CLR		BIT(17)
+#define MTK_WED_WPDMA_INT_CTRL_RRO_PG2_DONE_TRIG	GENMASK(22, 18)
+
+#define MTK_WED_RX_IND_CMD_CNT0				0xf20
+#define MTK_WED_RX_IND_CMD_DBG_CNT_EN			BIT(31)
+
+#define MTK_WED_RX_IND_CMD_CNT(_n)			(0xf20 + (_n) * 0x4)
+#define MTK_WED_IND_CMD_MAGIC_CNT_FAIL_CNT		GENMASK(15, 0)
+
+#define MTK_WED_RX_ADDR_ELEM_CNT(_n)			(0xf48 + (_n) * 0x4)
+#define MTK_WED_ADDR_ELEM_SIG_FAIL_CNT			GENMASK(15, 0)
+#define MTK_WED_ADDR_ELEM_FIRST_SIG_FAIL_CNT		GENMASK(31, 16)
+#define MTK_WED_ADDR_ELEM_ACKSN_CNT			GENMASK(27, 0)
+
+#define MTK_WED_RX_MSDU_PG_CNT(_n)			(0xf5c + (_n) * 0x4)
+
+#define MTK_WED_RX_PN_CHK_CNT 				0xf70
+#define MTK_WED_PN_CHK_FAIL_CNT				GENMASK(15, 0)
+
 #define MTK_WED_WOCPU_VIEW_MIOD_BASE		 	0x8000
 #define MTK_WED_PCIE_INT_MASK				0x0
 
+#define MTK_WED_PAO_AMSDU_FIFO				0x1800
+#define MTK_WED_PAO_AMSDU_IS_PRIOR0_RING		BIT(10)
+
+#define MTK_WED_PAO_STA_INFO				0x01810
+#define MTK_WED_PAO_STA_INFO_DO_INIT			BIT(0)
+#define MTK_WED_PAO_STA_INFO_SET_INIT			BIT(1)
+
+#define MTK_WED_PAO_STA_INFO_INIT			0x01814
+#define MTK_WED_PAO_STA_WTBL_HDRT_MODE			BIT(0)
+#define MTK_WED_PAO_STA_RMVL				BIT(1)
+#define MTK_WED_PAO_STA_MAX_AMSDU_LEN			GENMASK(7, 2)
+#define MTK_WED_PAO_STA_MAX_AMSDU_NUM			GENMASK(11, 8)
+
+#define MTK_WED_PAO_HIFTXD_BASE_L(_n)			(0x1980 + (_n) * 0x4)
+
+#define MTK_WED_PAO_PSE					0x1910
+#define MTK_WED_PAO_PSE_RESET				BIT(16)
+
+#define MTK_WED_PAO_HIFTXD_CFG				0x1968
+#define MTK_WED_PAO_HIFTXD_SRC				GENMASK(16, 15)
+
+#define MTK_WED_PAO_MON_AMSDU_FIFO_DMAD			0x1a34
+
+#define MTK_WED_PAO_MON_AMSDU_ENG_DMAD(_n)		(0x1a80 + (_n) * 0x50)
+#define MTK_WED_PAO_MON_AMSDU_ENG_QFPL(_n)		(0x1a84 + (_n) * 0x50)
+#define MTK_WED_PAO_MON_AMSDU_ENG_QENI(_n)		(0x1a88 + (_n) * 0x50)
+#define MTK_WED_PAO_MON_AMSDU_ENG_QENO(_n)		(0x1a8c + (_n) * 0x50)
+#define MTK_WED_PAO_MON_AMSDU_ENG_MERG(_n)		(0x1a90 + (_n) * 0x50)
+
+#define MTK_WED_PAO_MON_AMSDU_ENG_CNT8(_n)		(0x1a94 + (_n) * 0x50)
+#define MTK_WED_PAO_AMSDU_ENG_MAX_QGPP_CNT		GENMASK(10, 0)
+#define MTK_WED_PAO_AMSDU_ENG_MAX_PL_CNT		GENMASK(27, 16)
+
+#define MTK_WED_PAO_MON_AMSDU_ENG_CNT9(_n)		(0x1a98 + (_n) * 0x50)
+#define MTK_WED_PAO_AMSDU_ENG_CUR_ENTRY			GENMASK(10, 0)
+#define MTK_WED_PAO_AMSDU_ENG_MAX_BUF_MERGED		GENMASK(20, 16)
+#define MTK_WED_PAO_AMSDU_ENG_MAX_MSDU_MERGED		GENMASK(28, 24)
+
+#define MTK_WED_PAO_MON_QMEM_STS1			0x1e04
+
+#define MTK_WED_PAO_MON_QMEM_CNT(_n)			(0x1e0c + (_n) * 0x4)
+#define MTK_WED_PAO_QMEM_FQ_CNT				GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_SP_QCNT			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID0_QCNT			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID1_QCNT			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID2_QCNT			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID3_QCNT			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID4_QCNT			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID5_QCNT			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID6_QCNT			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID7_QCNT			GENMASK(11, 0)
+
+#define MTK_WED_PAO_MON_QMEM_PTR(_n)			(0x1e20 + (_n) * 0x4)
+#define MTK_WED_PAO_QMEM_FQ_HEAD				GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_SP_QHEAD			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID0_QHEAD			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID1_QHEAD			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID2_QHEAD			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID3_QHEAD			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID4_QHEAD			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID5_QHEAD			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID6_QHEAD			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID7_QHEAD			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_FQ_TAIL			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_SP_QTAIL			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID0_QTAIL			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID1_QTAIL			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID2_QTAIL			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID3_QTAIL			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID4_QTAIL			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID5_QTAIL			GENMASK(11, 0)
+#define MTK_WED_PAO_QMEM_TID6_QTAIL			GENMASK(27, 16)
+#define MTK_WED_PAO_QMEM_TID7_QTAIL			GENMASK(11, 0)
+
+#define MTK_WED_PAO_MON_HIFTXD_FETCH_MSDU(_n)		(0x1ec4 + (_n) * 0x4)
+
+#define MTK_WED_PCIE_BASE			0x11280000
+
+#define MTK_WED_PCIE_BASE0			0x11300000
+#define MTK_WED_PCIE_BASE1			0x11310000
+#define MTK_WED_PCIE_BASE2			0x11290000
 #endif
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 0967dc2..3211f34 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -875,6 +875,13 @@ struct net_device_path {
 			u8 queue;
 			u16 wcid;
 			u8 bss;
+			u32 usr_info;
+			u8 tid;
+			u8 is_fixedrate;
+			u8 is_prior;
+			u8 is_sp;
+			u8 hf;
+			u8 amsdu_en;
 		} mtk_wdma;
 	};
 };
diff --git a/include/linux/soc/mediatek/mtk_wed.h b/include/linux/soc/mediatek/mtk_wed.h
index 27cf284..92df4ba 100644
--- a/include/linux/soc/mediatek/mtk_wed.h
+++ b/include/linux/soc/mediatek/mtk_wed.h
@@ -5,11 +5,14 @@
 #include <linux/rcupdate.h>
 #include <linux/regmap.h>
 #include <linux/pci.h>
+#include <linux/skbuff.h>
+#include <linux/iopoll.h>
 
 #define WED_WO_STA_REC			0x6
 
 #define MTK_WED_TX_QUEUES		2
 #define MTK_WED_RX_QUEUES		2
+#define MTK_WED_RX_PAGE_QUEUES         3
 
 enum mtk_wed_wo_cmd {
 	MTK_WED_WO_CMD_WED_CFG,
@@ -55,10 +58,13 @@ enum mtk_wed_bus_tye {
 struct mtk_wed_hw;
 struct mtk_wdma_desc;
 
+#define MTK_WED_RING_CONFIGURED		BIT(0)
+
 struct mtk_wed_ring {
 	struct mtk_wdma_desc *desc;
 	dma_addr_t desc_phys;
 	int size;
+	u32 flags;
 
 	u32 reg_base;
 	void __iomem *wpdma;
@@ -69,11 +75,18 @@ struct mtk_rxbm_desc {
 	__le32 token;
 } __packed __aligned(4);
 
+struct dma_page_info {
+	void *addr;
+	dma_addr_t addr_phys;
+};
+
 struct dma_buf {
 	int size;
-	void **pages;
-	struct mtk_wdma_desc *desc;
+	int pkt_nums;
+	void *desc;
+	int desc_size;
 	dma_addr_t desc_phys;
+	struct dma_page_info *pages;
 };
 
 struct dma_entry {
@@ -97,6 +110,7 @@ struct mtk_wed_device {
 	struct device *dev;
 	struct mtk_wed_hw *hw;
 	bool init_done, running;
+	bool wdma_init_done;
 	int wdma_idx;
 	int irq;
 	u8 ver;
@@ -108,7 +122,11 @@ struct mtk_wed_device {
 	struct mtk_wed_ring rx_ring[MTK_WED_RX_QUEUES];
 	struct mtk_wed_ring rx_wdma[MTK_WED_RX_QUEUES];
 
-	struct dma_buf buf_ring;
+	struct mtk_wed_ring rx_rro_ring[MTK_WED_RX_QUEUES];
+	struct mtk_wed_ring rx_page_ring[MTK_WED_RX_PAGE_QUEUES];
+	struct mtk_wed_ring ind_cmd_ring;
+
+	struct dma_buf tx_buf_ring;
 
 	struct {
 		int size;
@@ -117,6 +135,8 @@ struct mtk_wed_device {
 		dma_addr_t desc_phys;
 	} rx_buf_ring;
 
+	struct dma_buf rx_page_buf_ring;
+
 	struct {
 		struct mtk_wed_ring rro_ring;
 		void __iomem *rro_desc;
@@ -131,8 +151,9 @@ struct mtk_wed_device {
 			struct platform_device *platform_dev;
 			struct pci_dev *pci_dev;
 		};
+		enum mtk_wed_bus_tye bus_type;
 		void __iomem *base;
-		u32 bus_type;
+		void __iomem *regs;
 		u32 phy_base;
 
 		u32 wpdma_phys;
@@ -141,10 +162,14 @@ struct mtk_wed_device {
 		u32 wpdma_tx;
 		u32 wpdma_txfree;
 		u32 wpdma_rx_glo;
-		u32 wpdma_rx;
+		u32 wpdma_rx[MTK_WED_RX_QUEUES];
+		u32 wpdma_rx_rro[MTK_WED_RX_QUEUES];
+		u32 wpdma_rx_pg;
 
 		u8 tx_tbit[MTK_WED_TX_QUEUES];
 		u8 rx_tbit[MTK_WED_RX_QUEUES];
+		u8 rro_rx_tbit[MTK_WED_RX_QUEUES];
+		u8 rx_pg_tbit[MTK_WED_RX_PAGE_QUEUES];
 		u8 txfree_tbit;
 
 		u16 token_start;
@@ -154,12 +179,26 @@ struct mtk_wed_device {
 		unsigned int rx_size;
 
 		bool wcid_512;
-
+		bool hwrro;
+		bool msi;
+
+		u8 max_amsdu_nums;
+		u32 max_amsdu_len;
+
+		struct {
+			u8 se_group_nums;
+			u16 win_size;
+			u16 particular_sid;
+			u32 ack_sn_addr;
+			dma_addr_t particular_se_phys;
+			dma_addr_t addr_elem_phys[1024];
+		} ind_cmd;
+
+		u32 chip_id;
 		u32 (*init_buf)(void *ptr, dma_addr_t phys, int token_id);
 		int (*offload_enable)(struct mtk_wed_device *wed);
 		void (*offload_disable)(struct mtk_wed_device *wed);
-		u32 (*init_rx_buf)(struct mtk_wed_device *wed,
-				   int pkt_num);
+		u32 (*init_rx_buf)(struct mtk_wed_device *wed, int size);
 		void (*release_rx_buf)(struct mtk_wed_device *wed);
 		void (*update_wo_rx_stats)(struct mtk_wed_device *wed,
 					   struct mtk_wed_wo_rx_stats *stats);
@@ -180,6 +219,11 @@ struct mtk_wed_ops {
 				 void __iomem *regs);
 	int (*rx_ring_setup)(struct mtk_wed_device *dev, int ring,
 			     void __iomem *regs, bool reset);
+	int (*rro_rx_ring_setup)(struct mtk_wed_device *dev, int ring,
+			     void __iomem *regs);
+	int (*msdu_pg_rx_ring_setup)(struct mtk_wed_device *dev, int ring,
+			     void __iomem *regs);
+	int (*ind_rx_ring_setup)(struct mtk_wed_device *dev, void __iomem *regs);
 	int (*msg_update)(struct mtk_wed_device *dev, int cmd_id,
 			  void *data, int len);
 	void (*detach)(struct mtk_wed_device *dev);
@@ -196,6 +240,7 @@ struct mtk_wed_ops {
 	void (*irq_set_mask)(struct mtk_wed_device *dev, u32 mask);
 	void (*ppe_check)(struct mtk_wed_device *dev, struct sk_buff *skb,
 			  u32 reason, u32 hash);
+	void (*start_hwrro)(struct mtk_wed_device *dev, u32 irq_mask);
 };
 
 extern const struct mtk_wed_ops __rcu *mtk_soc_wed_ops;
@@ -224,12 +269,21 @@ static inline bool
 mtk_wed_get_rx_capa(struct mtk_wed_device *dev)
 {
 #ifdef CONFIG_NET_MEDIATEK_SOC_WED
+	if (dev->ver == 3 && !dev->wlan.hwrro)
+		return false;
+
 	return dev->ver != 1;
 #else
 	return false;
 #endif
 }
 
+static inline bool
+mtk_wed_device_support_pao(struct mtk_wed_device *dev)
+{
+	return dev->ver == 3;
+}
+
 #ifdef CONFIG_NET_MEDIATEK_SOC_WED
 #define mtk_wed_device_active(_dev) !!(_dev)->ops
 #define mtk_wed_device_detach(_dev) (_dev)->ops->detach(_dev)
@@ -243,6 +297,12 @@ mtk_wed_get_rx_capa(struct mtk_wed_device *dev)
 	(_dev)->ops->txfree_ring_setup(_dev, _regs)
 #define mtk_wed_device_rx_ring_setup(_dev, _ring, _regs, _reset) \
 	(_dev)->ops->rx_ring_setup(_dev, _ring, _regs, _reset)
+#define mtk_wed_device_rro_rx_ring_setup(_dev, _ring, _regs) \
+	(_dev)->ops->rro_rx_ring_setup(_dev, _ring, _regs)
+#define mtk_wed_device_msdu_pg_rx_ring_setup(_dev, _ring, _regs) \
+	(_dev)->ops->msdu_pg_rx_ring_setup(_dev, _ring, _regs)
+#define mtk_wed_device_ind_rx_ring_setup(_dev, _regs) \
+	(_dev)->ops->ind_rx_ring_setup(_dev, _regs)
 #define mtk_wed_device_update_msg(_dev, _id, _msg, _len) \
 	(_dev)->ops->msg_update(_dev, _id, _msg, _len)
 #define mtk_wed_device_reg_read(_dev, _reg) \
@@ -257,6 +317,9 @@ mtk_wed_get_rx_capa(struct mtk_wed_device *dev)
 	(_dev)->ops->reset_dma(_dev)
 #define mtk_wed_device_ppe_check(_dev, _skb, _reason, _hash) \
 	(_dev)->ops->ppe_check(_dev, _skb, _reason, _hash)
+#define mtk_wed_device_start_hwrro(_dev, _mask) \
+	(_dev)->ops->start_hwrro(_dev, _mask)
+
 #else
 static inline bool mtk_wed_device_active(struct mtk_wed_device *dev)
 {
@@ -268,6 +331,9 @@ static inline bool mtk_wed_device_active(struct mtk_wed_device *dev)
 #define mtk_wed_device_tx_ring_setup(_dev, _ring, _regs, _reset) -ENODEV
 #define mtk_wed_device_txfree_ring_setup(_dev, _ring, _regs) -ENODEV
 #define mtk_wed_device_rx_ring_setup(_dev, _ring, _regs, _reset) -ENODEV
+#define mtk_wed_device_rro_rx_ring_setup(_dev, _ring, _regs) -ENODEV
+#define mtk_wed_device_msdu_pg_rx_ring_setup(_dev, _ring, _regs)  -ENODEV
+#define mtk_wed_device_ind_rx_ring_setup(_dev, _regs)  -ENODEV
 #define mtk_wed_device_reg_read(_dev, _reg) 0
 #define mtk_wed_device_reg_write(_dev, _reg, _val) do {} while (0)
 #define mtk_wed_device_irq_get(_dev, _mask) 0
@@ -275,6 +341,7 @@ static inline bool mtk_wed_device_active(struct mtk_wed_device *dev)
 #define mtk_wed_device_dma_reset(_dev) do {} while (0)
 #define mtk_wed_device_setup_tc(_dev, _ndev, _type, _data) do {} while (0)
 #define mtk_wed_device_ppe_check(_dev, _hash)  do {} while (0)
+#define mtk_wed_device_start_hwrro(_dev, _mask) do {} while (0)
 #endif
 
 #endif
-- 
2.18.0