Merging upstream version 6.7.7.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

11 files changed, 2090 insertions, 370 deletions

diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
index 20afe79f38..d2c039f830 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -197,6 +197,7 @@ static const struct mtk_reg_map mt7988_reg_map = {
 	.wdma_base = {
 		[0]		= 0x4800,
 		[1]		= 0x4c00,
+		[2]		= 0x5000,
 	},
 	.pse_iq_sta		= 0x0180,
 	.pse_oq_sta		= 0x01a0,
@@ -2210,7 +2211,7 @@ rx_done:
 	net_dim(&eth->rx_dim, dim_sample);
 
 	if (xdp_flush)
-		xdp_do_flush_map();
+		xdp_do_flush();
 
 	return done;
 }
@@ -3328,7 +3329,7 @@ static int mtk_device_event(struct notifier_block *n, unsigned long event, void
 	return NOTIFY_DONE;
 
 found:
-	if (!dsa_slave_dev_check(dev))
+	if (!dsa_user_dev_check(dev))
 		return NOTIFY_DONE;
 
 	if (__ethtool_get_link_ksettings(dev, &s))
@@ -4757,7 +4758,10 @@ static int mtk_probe(struct platform_device *pdev)
 	}
 
 	if (MTK_HAS_CAPS(eth->soc->caps, MTK_36BIT_DMA)) {
-		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(36));
+		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(36));
+		if (!err)
+			err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
+
 		if (err) {
 			dev_err(&pdev->dev, "Wrong DMA config\n");
 			return -EINVAL;
@@ -5002,7 +5006,7 @@ err_destroy_sgmii:
 	return err;
 }
 
-static int mtk_remove(struct platform_device *pdev)
+static void mtk_remove(struct platform_device *pdev)
 {
 	struct mtk_eth *eth = platform_get_drvdata(pdev);
 	struct mtk_mac *mac;
@@ -5024,8 +5028,6 @@ static int mtk_remove(struct platform_device *pdev)
 	netif_napi_del(&eth->rx_napi);
 	mtk_cleanup(eth);
 	mtk_mdio_cleanup(eth);
-
-	return 0;
 }
 
 static const struct mtk_soc_data mt2701_data = {
@@ -5226,7 +5228,7 @@ MODULE_DEVICE_TABLE(of, of_mtk_match);
 
 static struct platform_driver mtk_driver = {
 	.probe = mtk_probe,
-	.remove = mtk_remove,
+	.remove_new = mtk_remove,
 	.driver = {
 		.name = "mtk_soc_eth",
 		.of_match_table = of_mtk_match,
diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.h b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
index 403219d987..9ae3b8a71d 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.h
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
@@ -1132,7 +1132,7 @@ struct mtk_reg_map {
 	u32	gdm1_cnt;
 	u32	gdma_to_ppe;
 	u32	ppe_base;
-	u32	wdma_base[2];
+	u32	wdma_base[3];
 	u32	pse_iq_sta;
 	u32	pse_oq_sta;
 };
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe.c b/drivers/net/ethernet/mediatek/mtk_ppe.c
index 86f32f4860..b2a5d9c373 100644
--- a/drivers/net/ethernet/mediatek/mtk_ppe.c
+++ b/drivers/net/ethernet/mediatek/mtk_ppe.c
@@ -425,7 +425,8 @@ int mtk_foe_entry_set_pppoe(struct mtk_eth *eth, struct mtk_foe_entry *entry,
 }
 
 int mtk_foe_entry_set_wdma(struct mtk_eth *eth, struct mtk_foe_entry *entry,
-			   int wdma_idx, int txq, int bss, int wcid)
+			   int wdma_idx, int txq, int bss, int wcid,
+			   bool amsdu_en)
 {
 	struct mtk_foe_mac_info *l2 = mtk_foe_entry_l2(eth, entry);
 	u32 *ib2 = mtk_foe_entry_ib2(eth, entry);
@@ -437,6 +438,7 @@ int mtk_foe_entry_set_wdma(struct mtk_eth *eth, struct mtk_foe_entry *entry,
 			 MTK_FOE_IB2_WDMA_WINFO_V2;
 		l2->w3info = FIELD_PREP(MTK_FOE_WINFO_WCID_V3, wcid) |
 			     FIELD_PREP(MTK_FOE_WINFO_BSS_V3, bss);
+		l2->amsdu = FIELD_PREP(MTK_FOE_WINFO_AMSDU_EN, amsdu_en);
 		break;
 	case 2:
 		*ib2 &= ~MTK_FOE_IB2_PORT_MG_V2;
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe.h b/drivers/net/ethernet/mediatek/mtk_ppe.h
index e3d0ec72bc..691806bca3 100644
--- a/drivers/net/ethernet/mediatek/mtk_ppe.h
+++ b/drivers/net/ethernet/mediatek/mtk_ppe.h
@@ -88,13 +88,13 @@ enum {
 #define MTK_FOE_WINFO_BSS_V3		GENMASK(23, 16)
 #define MTK_FOE_WINFO_WCID_V3		GENMASK(15, 0)
 
-#define MTK_FOE_WINFO_PAO_USR_INFO	GENMASK(15, 0)
-#define MTK_FOE_WINFO_PAO_TID		GENMASK(19, 16)
-#define MTK_FOE_WINFO_PAO_IS_FIXEDRATE	BIT(20)
-#define MTK_FOE_WINFO_PAO_IS_PRIOR	BIT(21)
-#define MTK_FOE_WINFO_PAO_IS_SP		BIT(22)
-#define MTK_FOE_WINFO_PAO_HF		BIT(23)
-#define MTK_FOE_WINFO_PAO_AMSDU_EN	BIT(24)
+#define MTK_FOE_WINFO_AMSDU_USR_INFO	GENMASK(15, 0)
+#define MTK_FOE_WINFO_AMSDU_TID		GENMASK(19, 16)
+#define MTK_FOE_WINFO_AMSDU_IS_FIXEDRATE	BIT(20)
+#define MTK_FOE_WINFO_AMSDU_IS_PRIOR	BIT(21)
+#define MTK_FOE_WINFO_AMSDU_IS_SP	BIT(22)
+#define MTK_FOE_WINFO_AMSDU_HF		BIT(23)
+#define MTK_FOE_WINFO_AMSDU_EN		BIT(24)
 
 enum {
 	MTK_FOE_STATE_INVALID,
@@ -123,7 +123,7 @@ struct mtk_foe_mac_info {
 
 	/* netsys_v3 */
 	u32 w3info;
-	u32 wpao;
+	u32 amsdu;
 };
 
 /* software-only entry type */
@@ -392,7 +392,8 @@ int mtk_foe_entry_set_vlan(struct mtk_eth *eth, struct mtk_foe_entry *entry,
 int mtk_foe_entry_set_pppoe(struct mtk_eth *eth, struct mtk_foe_entry *entry,
 			    int sid);
 int mtk_foe_entry_set_wdma(struct mtk_eth *eth, struct mtk_foe_entry *entry,
-			   int wdma_idx, int txq, int bss, int wcid);
+			   int wdma_idx, int txq, int bss, int wcid,
+			   bool amsdu_en);
 int mtk_foe_entry_set_queue(struct mtk_eth *eth, struct mtk_foe_entry *entry,
 			    unsigned int queue);
 int mtk_foe_entry_commit(struct mtk_ppe *ppe, struct mtk_flow_entry *entry);
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe_offload.c b/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
index a4efbeb162..fbb5e9d5af 100644
--- a/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
+++ b/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
@@ -111,6 +111,7 @@ 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 = path->mtk_wdma.amsdu;
 
 	return 0;
 }
@@ -174,7 +175,7 @@ mtk_flow_get_dsa_port(struct net_device **dev)
 	if (dp->cpu_dp->tag_ops->proto != DSA_TAG_PROTO_MTK)
 		return -ENODEV;
 
-	*dev = dsa_port_to_master(dp);
+	*dev = dsa_port_to_conduit(dp);
 
 	return dp->index;
 #else
@@ -192,14 +193,17 @@ 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(eth, foe, info.wdma_idx, info.queue,
-				       info.bss, info.wcid);
+				       info.bss, info.wcid, info.amsdu);
 		if (mtk_is_netsys_v2_or_greater(eth)) {
 			switch (info.wdma_idx) {
 			case 0:
-				pse_port = 8;
+				pse_port = PSE_WDMA0_PORT;
 				break;
 			case 1:
-				pse_port = 9;
+				pse_port = PSE_WDMA1_PORT;
+				break;
+			case 2:
+				pse_port = PSE_WDMA2_PORT;
 				break;
 			default:
 				return -EINVAL;
diff --git a/drivers/net/ethernet/mediatek/mtk_wed.c b/drivers/net/ethernet/mediatek/mtk_wed.c
index 94376aa2b3..9a6744c0d4 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed.c
@@ -17,17 +17,21 @@
 #include <net/flow_offload.h>
 #include <net/pkt_cls.h>
 #include "mtk_eth_soc.h"
-#include "mtk_wed_regs.h"
 #include "mtk_wed.h"
 #include "mtk_ppe.h"
 #include "mtk_wed_wo.h"
 
 #define MTK_PCIE_BASE(n)		(0x1a143000 + (n) * 0x2000)
 
-#define MTK_WED_PKT_SIZE		1900
+#define MTK_WED_PKT_SIZE		1920
 #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_BUF_PER_PAGE		(PAGE_SIZE / MTK_WED_PAGE_BUF_SIZE)
 #define MTK_WED_RX_RING_SIZE		1536
+#define MTK_WED_RX_PG_BM_CNT		8192
+#define MTK_WED_AMSDU_BUF_SIZE		(PAGE_SIZE << 4)
+#define MTK_WED_AMSDU_NPAGES		32
 
 #define MTK_WED_TX_RING_SIZE		2048
 #define MTK_WED_WDMA_RING_SIZE		1024
@@ -41,7 +45,10 @@
 #define MTK_WED_RRO_QUE_CNT		8192
 #define MTK_WED_MIOD_ENTRY_CNT		128
 
-static struct mtk_wed_hw *hw_list[2];
+#define MTK_WED_TX_BM_DMA_SIZE		65536
+#define MTK_WED_TX_BM_PKT_CNT		32768
+
+static struct mtk_wed_hw *hw_list[3];
 static DEFINE_MUTEX(hw_lock);
 
 struct mtk_wed_flow_block_priv {
@@ -49,6 +56,39 @@ struct mtk_wed_flow_block_priv {
 	struct net_device *dev;
 };
 
+static const struct mtk_wed_soc_data mt7622_data = {
+	.regmap = {
+		.tx_bm_tkid		= 0x088,
+		.wpdma_rx_ring0		= 0x770,
+		.reset_idx_tx_mask	= GENMASK(3, 0),
+		.reset_idx_rx_mask	= GENMASK(17, 16),
+	},
+	.tx_ring_desc_size = sizeof(struct mtk_wdma_desc),
+	.wdma_desc_size = sizeof(struct mtk_wdma_desc),
+};
+
+static const struct mtk_wed_soc_data mt7986_data = {
+	.regmap = {
+		.tx_bm_tkid		= 0x0c8,
+		.wpdma_rx_ring0		= 0x770,
+		.reset_idx_tx_mask	= GENMASK(1, 0),
+		.reset_idx_rx_mask	= GENMASK(7, 6),
+	},
+	.tx_ring_desc_size = sizeof(struct mtk_wdma_desc),
+	.wdma_desc_size = 2 * sizeof(struct mtk_wdma_desc),
+};
+
+static const struct mtk_wed_soc_data mt7988_data = {
+	.regmap = {
+		.tx_bm_tkid		= 0x0c8,
+		.wpdma_rx_ring0		= 0x7d0,
+		.reset_idx_tx_mask	= GENMASK(1, 0),
+		.reset_idx_rx_mask	= GENMASK(7, 6),
+	},
+	.tx_ring_desc_size = sizeof(struct mtk_wed_bm_desc),
+	.wdma_desc_size = 2 * sizeof(struct mtk_wdma_desc),
+};
+
 static void
 wed_m32(struct mtk_wed_device *dev, u32 reg, u32 mask, u32 val)
 {
@@ -109,6 +149,90 @@ mtk_wdma_read_reset(struct mtk_wed_device *dev)
 	return wdma_r32(dev, MTK_WDMA_GLO_CFG);
 }
 
+static void
+mtk_wdma_v3_rx_reset(struct mtk_wed_device *dev)
+{
+	u32 status;
+
+	if (!mtk_wed_is_v3_or_greater(dev->hw))
+		return;
+
+	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);
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_PREF_TX_CFG_PREF_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_PREF_TX_CFG))
+		dev_err(dev->hw->dev, "rx reset failed\n");
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_PREF_RX_CFG_PREF_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_PREF_RX_CFG))
+		dev_err(dev->hw->dev, "rx reset failed\n");
+
+	wdma_clr(dev, MTK_WDMA_WRBK_TX_CFG, MTK_WDMA_WRBK_TX_CFG_WRBK_EN);
+	wdma_clr(dev, MTK_WDMA_WRBK_RX_CFG, MTK_WDMA_WRBK_RX_CFG_WRBK_EN);
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_WRBK_TX_CFG_WRBK_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_WRBK_TX_CFG))
+		dev_err(dev->hw->dev, "rx reset failed\n");
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_WRBK_RX_CFG_WRBK_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_WRBK_RX_CFG))
+		dev_err(dev->hw->dev, "rx reset failed\n");
+
+	/* prefetch FIFO */
+	wdma_w32(dev, MTK_WDMA_PREF_RX_FIFO_CFG,
+		 MTK_WDMA_PREF_RX_FIFO_CFG_RING0_CLEAR |
+		 MTK_WDMA_PREF_RX_FIFO_CFG_RING1_CLEAR);
+	wdma_clr(dev, MTK_WDMA_PREF_RX_FIFO_CFG,
+		 MTK_WDMA_PREF_RX_FIFO_CFG_RING0_CLEAR |
+		 MTK_WDMA_PREF_RX_FIFO_CFG_RING1_CLEAR);
+
+	/* core FIFO */
+	wdma_w32(dev, MTK_WDMA_XDMA_RX_FIFO_CFG,
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_PAR_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_CMD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_DMAD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_ARR_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_LEN_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_WID_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_BID_FIFO_CLEAR);
+	wdma_clr(dev, MTK_WDMA_XDMA_RX_FIFO_CFG,
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_PAR_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_CMD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_DMAD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_ARR_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_LEN_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_WID_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_RX_FIFO_CFG_RX_BID_FIFO_CLEAR);
+
+	/* writeback FIFO */
+	wdma_w32(dev, MTK_WDMA_WRBK_RX_FIFO_CFG(0),
+		 MTK_WDMA_WRBK_RX_FIFO_CFG_RING_CLEAR);
+	wdma_w32(dev, MTK_WDMA_WRBK_RX_FIFO_CFG(1),
+		 MTK_WDMA_WRBK_RX_FIFO_CFG_RING_CLEAR);
+
+	wdma_clr(dev, MTK_WDMA_WRBK_RX_FIFO_CFG(0),
+		 MTK_WDMA_WRBK_RX_FIFO_CFG_RING_CLEAR);
+	wdma_clr(dev, MTK_WDMA_WRBK_RX_FIFO_CFG(1),
+		 MTK_WDMA_WRBK_RX_FIFO_CFG_RING_CLEAR);
+
+	/* prefetch ring status */
+	wdma_w32(dev, MTK_WDMA_PREF_SIDX_CFG,
+		 MTK_WDMA_PREF_SIDX_CFG_RX_RING_CLEAR);
+	wdma_clr(dev, MTK_WDMA_PREF_SIDX_CFG,
+		 MTK_WDMA_PREF_SIDX_CFG_RX_RING_CLEAR);
+
+	/* writeback ring status */
+	wdma_w32(dev, MTK_WDMA_WRBK_SIDX_CFG,
+		 MTK_WDMA_WRBK_SIDX_CFG_RX_RING_CLEAR);
+	wdma_clr(dev, MTK_WDMA_WRBK_SIDX_CFG,
+		 MTK_WDMA_WRBK_SIDX_CFG_RX_RING_CLEAR);
+}
+
 static int
 mtk_wdma_rx_reset(struct mtk_wed_device *dev)
 {
@@ -121,6 +245,7 @@ mtk_wdma_rx_reset(struct mtk_wed_device *dev)
 	if (ret)
 		dev_err(dev->hw->dev, "rx reset failed\n");
 
+	mtk_wdma_v3_rx_reset(dev);
 	wdma_w32(dev, MTK_WDMA_RESET_IDX, MTK_WDMA_RESET_IDX_RX);
 	wdma_w32(dev, MTK_WDMA_RESET_IDX, 0);
 
@@ -135,6 +260,101 @@ mtk_wdma_rx_reset(struct mtk_wed_device *dev)
 	return ret;
 }
 
+static u32
+mtk_wed_check_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
+{
+	return !!(wed_r32(dev, reg) & mask);
+}
+
+static int
+mtk_wed_poll_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
+{
+	int sleep = 15000;
+	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_wdma_v3_tx_reset(struct mtk_wed_device *dev)
+{
+	u32 status;
+
+	if (!mtk_wed_is_v3_or_greater(dev->hw))
+		return;
+
+	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);
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_PREF_TX_CFG_PREF_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_PREF_TX_CFG))
+		dev_err(dev->hw->dev, "tx reset failed\n");
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_PREF_RX_CFG_PREF_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_PREF_RX_CFG))
+		dev_err(dev->hw->dev, "tx reset failed\n");
+
+	wdma_clr(dev, MTK_WDMA_WRBK_TX_CFG, MTK_WDMA_WRBK_TX_CFG_WRBK_EN);
+	wdma_clr(dev, MTK_WDMA_WRBK_RX_CFG, MTK_WDMA_WRBK_RX_CFG_WRBK_EN);
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_WRBK_TX_CFG_WRBK_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_WRBK_TX_CFG))
+		dev_err(dev->hw->dev, "tx reset failed\n");
+
+	if (read_poll_timeout(wdma_r32, status,
+			      !(status & MTK_WDMA_WRBK_RX_CFG_WRBK_BUSY),
+			      0, 10000, false, dev, MTK_WDMA_WRBK_RX_CFG))
+		dev_err(dev->hw->dev, "tx reset failed\n");
+
+	/* prefetch FIFO */
+	wdma_w32(dev, MTK_WDMA_PREF_TX_FIFO_CFG,
+		 MTK_WDMA_PREF_TX_FIFO_CFG_RING0_CLEAR |
+		 MTK_WDMA_PREF_TX_FIFO_CFG_RING1_CLEAR);
+	wdma_clr(dev, MTK_WDMA_PREF_TX_FIFO_CFG,
+		 MTK_WDMA_PREF_TX_FIFO_CFG_RING0_CLEAR |
+		 MTK_WDMA_PREF_TX_FIFO_CFG_RING1_CLEAR);
+
+	/* core FIFO */
+	wdma_w32(dev, MTK_WDMA_XDMA_TX_FIFO_CFG,
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_PAR_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_CMD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_DMAD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_ARR_FIFO_CLEAR);
+	wdma_clr(dev, MTK_WDMA_XDMA_TX_FIFO_CFG,
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_PAR_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_CMD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_DMAD_FIFO_CLEAR |
+		 MTK_WDMA_XDMA_TX_FIFO_CFG_TX_ARR_FIFO_CLEAR);
+
+	/* writeback FIFO */
+	wdma_w32(dev, MTK_WDMA_WRBK_TX_FIFO_CFG(0),
+		 MTK_WDMA_WRBK_TX_FIFO_CFG_RING_CLEAR);
+	wdma_w32(dev, MTK_WDMA_WRBK_TX_FIFO_CFG(1),
+		 MTK_WDMA_WRBK_TX_FIFO_CFG_RING_CLEAR);
+
+	wdma_clr(dev, MTK_WDMA_WRBK_TX_FIFO_CFG(0),
+		 MTK_WDMA_WRBK_TX_FIFO_CFG_RING_CLEAR);
+	wdma_clr(dev, MTK_WDMA_WRBK_TX_FIFO_CFG(1),
+		 MTK_WDMA_WRBK_TX_FIFO_CFG_RING_CLEAR);
+
+	/* prefetch ring status */
+	wdma_w32(dev, MTK_WDMA_PREF_SIDX_CFG,
+		 MTK_WDMA_PREF_SIDX_CFG_TX_RING_CLEAR);
+	wdma_clr(dev, MTK_WDMA_PREF_SIDX_CFG,
+		 MTK_WDMA_PREF_SIDX_CFG_TX_RING_CLEAR);
+
+	/* writeback ring status */
+	wdma_w32(dev, MTK_WDMA_WRBK_SIDX_CFG,
+		 MTK_WDMA_WRBK_SIDX_CFG_TX_RING_CLEAR);
+	wdma_clr(dev, MTK_WDMA_WRBK_SIDX_CFG,
+		 MTK_WDMA_WRBK_SIDX_CFG_TX_RING_CLEAR);
+}
+
 static void
 mtk_wdma_tx_reset(struct mtk_wed_device *dev)
 {
@@ -146,6 +366,7 @@ mtk_wdma_tx_reset(struct mtk_wed_device *dev)
 			       !(status & mask), 0, 10000))
 		dev_err(dev->hw->dev, "tx reset failed\n");
 
+	mtk_wdma_v3_tx_reset(dev);
 	wdma_w32(dev, MTK_WDMA_RESET_IDX, MTK_WDMA_RESET_IDX_TX);
 	wdma_w32(dev, MTK_WDMA_RESET_IDX, 0);
 
@@ -278,7 +499,7 @@ mtk_wed_assign(struct mtk_wed_device *dev)
 		if (!hw->wed_dev)
 			goto out;
 
-		if (hw->version == 1)
+		if (mtk_wed_is_v1(hw))
 			return NULL;
 
 		/* MT7986 WED devices do not have any pcie slot restrictions */
@@ -298,35 +519,152 @@ out:
 }
 
 static int
+mtk_wed_amsdu_buffer_alloc(struct mtk_wed_device *dev)
+{
+	struct mtk_wed_hw *hw = dev->hw;
+	struct mtk_wed_amsdu *wed_amsdu;
+	int i;
+
+	if (!mtk_wed_is_v3_or_greater(hw))
+		return 0;
+
+	wed_amsdu = devm_kcalloc(hw->dev, MTK_WED_AMSDU_NPAGES,
+				 sizeof(*wed_amsdu), GFP_KERNEL);
+	if (!wed_amsdu)
+		return -ENOMEM;
+
+	for (i = 0; i < MTK_WED_AMSDU_NPAGES; i++) {
+		void *ptr;
+
+		/* each segment is 64K */
+		ptr = (void *)__get_free_pages(GFP_KERNEL | __GFP_NOWARN |
+					       __GFP_ZERO | __GFP_COMP |
+					       GFP_DMA32,
+					       get_order(MTK_WED_AMSDU_BUF_SIZE));
+		if (!ptr)
+			goto error;
+
+		wed_amsdu[i].txd = ptr;
+		wed_amsdu[i].txd_phy = dma_map_single(hw->dev, ptr,
+						      MTK_WED_AMSDU_BUF_SIZE,
+						      DMA_TO_DEVICE);
+		if (dma_mapping_error(hw->dev, wed_amsdu[i].txd_phy))
+			goto error;
+	}
+	dev->hw->wed_amsdu = wed_amsdu;
+
+	return 0;
+
+error:
+	for (i--; i >= 0; i--)
+		dma_unmap_single(hw->dev, wed_amsdu[i].txd_phy,
+				 MTK_WED_AMSDU_BUF_SIZE, DMA_TO_DEVICE);
+	return -ENOMEM;
+}
+
+static void
+mtk_wed_amsdu_free_buffer(struct mtk_wed_device *dev)
+{
+	struct mtk_wed_amsdu *wed_amsdu = dev->hw->wed_amsdu;
+	int i;
+
+	if (!wed_amsdu)
+		return;
+
+	for (i = 0; i < MTK_WED_AMSDU_NPAGES; i++) {
+		dma_unmap_single(dev->hw->dev, wed_amsdu[i].txd_phy,
+				 MTK_WED_AMSDU_BUF_SIZE, DMA_TO_DEVICE);
+		free_pages((unsigned long)wed_amsdu[i].txd,
+			   get_order(MTK_WED_AMSDU_BUF_SIZE));
+	}
+}
+
+static int
+mtk_wed_amsdu_init(struct mtk_wed_device *dev)
+{
+	struct mtk_wed_amsdu *wed_amsdu = dev->hw->wed_amsdu;
+	int i, ret;
+
+	if (!wed_amsdu)
+		return 0;
+
+	for (i = 0; i < MTK_WED_AMSDU_NPAGES; i++)
+		wed_w32(dev, MTK_WED_AMSDU_HIFTXD_BASE_L(i),
+			wed_amsdu[i].txd_phy);
+
+	/* init all sta parameter */
+	wed_w32(dev, MTK_WED_AMSDU_STA_INFO_INIT, MTK_WED_AMSDU_STA_RMVL |
+		MTK_WED_AMSDU_STA_WTBL_HDRT_MODE |
+		FIELD_PREP(MTK_WED_AMSDU_STA_MAX_AMSDU_LEN,
+			   dev->wlan.amsdu_max_len >> 8) |
+		FIELD_PREP(MTK_WED_AMSDU_STA_MAX_AMSDU_NUM,
+			   dev->wlan.amsdu_max_subframes));
+
+	wed_w32(dev, MTK_WED_AMSDU_STA_INFO, MTK_WED_AMSDU_STA_INFO_DO_INIT);
+
+	ret = mtk_wed_poll_busy(dev, MTK_WED_AMSDU_STA_INFO,
+				MTK_WED_AMSDU_STA_INFO_DO_INIT);
+	if (ret) {
+		dev_err(dev->hw->dev, "amsdu initialization failed\n");
+		return ret;
+	}
+
+	/* init partial amsdu offload txd src */
+	wed_set(dev, MTK_WED_AMSDU_HIFTXD_CFG,
+		FIELD_PREP(MTK_WED_AMSDU_HIFTXD_SRC, dev->hw->index));
+
+	/* init qmem */
+	wed_set(dev, MTK_WED_AMSDU_PSE, MTK_WED_AMSDU_PSE_RESET);
+	ret = mtk_wed_poll_busy(dev, MTK_WED_MON_AMSDU_QMEM_STS1, BIT(29));
+	if (ret) {
+		pr_info("%s: amsdu qmem initialization failed\n", __func__);
+		return ret;
+	}
+
+	/* eagle E1 PCIE1 tx ring 22 flow control issue */
+	if (dev->wlan.id == 0x7991)
+		wed_clr(dev, MTK_WED_AMSDU_FIFO, MTK_WED_AMSDU_IS_PRIOR0_RING);
+
+	wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_AMSDU_EN);
+
+	return 0;
+}
+
+static int
 mtk_wed_tx_buffer_alloc(struct mtk_wed_device *dev)
 {
-	struct mtk_wdma_desc *desc;
-	dma_addr_t desc_phys;
-	void **page_list;
+	u32 desc_size = dev->hw->soc->tx_ring_desc_size;
+	int i, page_idx = 0, n_pages, ring_size;
 	int token = dev->wlan.token_start;
-	int ring_size;
-	int n_pages;
-	int i, page_idx;
+	struct mtk_wed_buf *page_list;
+	dma_addr_t desc_phys;
+	void *desc_ptr;
 
-	ring_size = dev->wlan.nbuf & ~(MTK_WED_BUF_PER_PAGE - 1);
-	n_pages = ring_size / MTK_WED_BUF_PER_PAGE;
+	if (!mtk_wed_is_v3_or_greater(dev->hw)) {
+		ring_size = dev->wlan.nbuf & ~(MTK_WED_BUF_PER_PAGE - 1);
+		dev->tx_buf_ring.size = ring_size;
+	} else {
+		dev->tx_buf_ring.size = MTK_WED_TX_BM_DMA_SIZE;
+		ring_size = MTK_WED_TX_BM_PKT_CNT;
+	}
+	n_pages = dev->tx_buf_ring.size / MTK_WED_BUF_PER_PAGE;
 
 	page_list = kcalloc(n_pages, sizeof(*page_list), GFP_KERNEL);
 	if (!page_list)
 		return -ENOMEM;
 
-	dev->tx_buf_ring.size = ring_size;
 	dev->tx_buf_ring.pages = page_list;
 
-	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,
+				      dev->tx_buf_ring.size * desc_size,
+				      &desc_phys, GFP_KERNEL);
+	if (!desc_ptr)
 		return -ENOMEM;
 
-	dev->tx_buf_ring.desc = desc;
+	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; i < ring_size; i += MTK_WED_BUF_PER_PAGE) {
 		dma_addr_t page_phys, buf_phys;
 		struct page *page;
 		void *buf;
@@ -343,7 +681,8 @@ mtk_wed_tx_buffer_alloc(struct mtk_wed_device *dev)
 			return -ENOMEM;
 		}
 
-		page_list[page_idx++] = page;
+		page_list[page_idx].p = page;
+		page_list[page_idx++].phy_addr = page_phys;
 		dma_sync_single_for_cpu(dev->hw->dev, page_phys, PAGE_SIZE,
 					DMA_BIDIRECTIONAL);
 
@@ -351,28 +690,31 @@ mtk_wed_tx_buffer_alloc(struct mtk_wed_device *dev)
 		buf_phys = page_phys;
 
 		for (s = 0; s < MTK_WED_BUF_PER_PAGE; s++) {
-			u32 txd_size;
-			u32 ctrl;
-
-			txd_size = dev->wlan.init_buf(buf, buf_phys, token++);
+			struct mtk_wdma_desc *desc = desc_ptr;
 
 			desc->buf0 = cpu_to_le32(buf_phys);
-			desc->buf1 = cpu_to_le32(buf_phys + txd_size);
-
-			if (dev->hw->version == 1)
-				ctrl = FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN0, txd_size) |
-				       FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN1,
-						  MTK_WED_BUF_SIZE - txd_size) |
-				       MTK_WDMA_DESC_CTRL_LAST_SEG1;
-			else
-				ctrl = FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN0, txd_size) |
-				       FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN1_V2,
-						  MTK_WED_BUF_SIZE - txd_size) |
-				       MTK_WDMA_DESC_CTRL_LAST_SEG0;
-			desc->ctrl = cpu_to_le32(ctrl);
-			desc->info = 0;
-			desc++;
-
+			if (!mtk_wed_is_v3_or_greater(dev->hw)) {
+				u32 txd_size, ctrl;
+
+				txd_size = dev->wlan.init_buf(buf, buf_phys,
+							      token++);
+				desc->buf1 = cpu_to_le32(buf_phys + txd_size);
+				ctrl = FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN0, txd_size);
+				if (mtk_wed_is_v1(dev->hw))
+					ctrl |= MTK_WDMA_DESC_CTRL_LAST_SEG1 |
+						FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN1,
+							   MTK_WED_BUF_SIZE - txd_size);
+				else
+					ctrl |= MTK_WDMA_DESC_CTRL_LAST_SEG0 |
+						FIELD_PREP(MTK_WDMA_DESC_CTRL_LEN1_V2,
+							   MTK_WED_BUF_SIZE - txd_size);
+				desc->ctrl = cpu_to_le32(ctrl);
+				desc->info = 0;
+			} else {
+				desc->ctrl = cpu_to_le32(token << 16);
+			}
+
+			desc_ptr += desc_size;
 			buf += MTK_WED_BUF_SIZE;
 			buf_phys += MTK_WED_BUF_SIZE;
 		}
@@ -387,42 +729,103 @@ mtk_wed_tx_buffer_alloc(struct mtk_wed_device *dev)
 static void
 mtk_wed_free_tx_buffer(struct mtk_wed_device *dev)
 {
-	struct mtk_wdma_desc *desc = dev->tx_buf_ring.desc;
-	void **page_list = dev->tx_buf_ring.pages;
-	int page_idx;
-	int i;
+	struct mtk_wed_buf *page_list = dev->tx_buf_ring.pages;
+	struct mtk_wed_hw *hw = dev->hw;
+	int i, page_idx = 0;
 
 	if (!page_list)
 		return;
 
-	if (!desc)
+	if (!dev->tx_buf_ring.desc)
 		goto free_pagelist;
 
-	for (i = 0, page_idx = 0; i < dev->tx_buf_ring.size;
-	     i += MTK_WED_BUF_PER_PAGE) {
-		void *page = page_list[page_idx++];
-		dma_addr_t buf_addr;
+	for (i = 0; i < dev->tx_buf_ring.size; i += MTK_WED_BUF_PER_PAGE) {
+		dma_addr_t page_phy = page_list[page_idx].phy_addr;
+		void *page = page_list[page_idx++].p;
 
 		if (!page)
 			break;
 
-		buf_addr = le32_to_cpu(desc[i].buf0);
-		dma_unmap_page(dev->hw->dev, buf_addr, PAGE_SIZE,
+		dma_unmap_page(dev->hw->dev, page_phy, PAGE_SIZE,
 			       DMA_BIDIRECTIONAL);
 		__free_page(page);
 	}
 
-	dma_free_coherent(dev->hw->dev, dev->tx_buf_ring.size * sizeof(*desc),
-			  desc, dev->tx_buf_ring.desc_phys);
+	dma_free_coherent(dev->hw->dev,
+			  dev->tx_buf_ring.size * hw->soc->tx_ring_desc_size,
+			  dev->tx_buf_ring.desc,
+			  dev->tx_buf_ring.desc_phys);
 
 free_pagelist:
 	kfree(page_list);
 }
 
 static int
+mtk_wed_hwrro_buffer_alloc(struct mtk_wed_device *dev)
+{
+	int n_pages = MTK_WED_RX_PG_BM_CNT / MTK_WED_RX_BUF_PER_PAGE;
+	struct mtk_wed_buf *page_list;
+	struct mtk_wed_bm_desc *desc;
+	dma_addr_t desc_phys;
+	int i, page_idx = 0;
+
+	if (!dev->wlan.hw_rro)
+		return 0;
+
+	page_list = kcalloc(n_pages, sizeof(*page_list), GFP_KERNEL);
+	if (!page_list)
+		return -ENOMEM;
+
+	dev->hw_rro.size = dev->wlan.rx_nbuf & ~(MTK_WED_BUF_PER_PAGE - 1);
+	dev->hw_rro.pages = page_list;
+	desc = dma_alloc_coherent(dev->hw->dev,
+				  dev->wlan.rx_nbuf * sizeof(*desc),
+				  &desc_phys, GFP_KERNEL);
+	if (!desc)
+		return -ENOMEM;
+
+	dev->hw_rro.desc = desc;
+	dev->hw_rro.desc_phys = desc_phys;
+
+	for (i = 0; i < MTK_WED_RX_PG_BM_CNT; i += MTK_WED_RX_BUF_PER_PAGE) {
+		dma_addr_t page_phys, buf_phys;
+		struct page *page;
+		int s;
+
+		page = __dev_alloc_page(GFP_KERNEL);
+		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].p = page;
+		page_list[page_idx++].phy_addr = page_phys;
+		dma_sync_single_for_cpu(dev->hw->dev, page_phys, PAGE_SIZE,
+					DMA_BIDIRECTIONAL);
+
+		buf_phys = page_phys;
+		for (s = 0; s < MTK_WED_RX_BUF_PER_PAGE; s++) {
+			desc->buf0 = cpu_to_le32(buf_phys);
+			buf_phys += MTK_WED_PAGE_BUF_SIZE;
+			desc++;
+		}
+
+		dma_sync_single_for_device(dev->hw->dev, page_phys, PAGE_SIZE,
+					   DMA_BIDIRECTIONAL);
+	}
+
+	return 0;
+}
+
+static int
 mtk_wed_rx_buffer_alloc(struct mtk_wed_device *dev)
 {
-	struct mtk_rxbm_desc *desc;
+	struct mtk_wed_bm_desc *desc;
 	dma_addr_t desc_phys;
 
 	dev->rx_buf_ring.size = dev->wlan.rx_nbuf;
@@ -436,13 +839,48 @@ mtk_wed_rx_buffer_alloc(struct mtk_wed_device *dev)
 	dev->rx_buf_ring.desc_phys = desc_phys;
 	dev->wlan.init_rx_buf(dev, dev->wlan.rx_npkt);
 
-	return 0;
+	return mtk_wed_hwrro_buffer_alloc(dev);
+}
+
+static void
+mtk_wed_hwrro_free_buffer(struct mtk_wed_device *dev)
+{
+	struct mtk_wed_buf *page_list = dev->hw_rro.pages;
+	struct mtk_wed_bm_desc *desc = dev->hw_rro.desc;
+	int i, page_idx = 0;
+
+	if (!dev->wlan.hw_rro)
+		return;
+
+	if (!page_list)
+		return;
+
+	if (!desc)
+		goto free_pagelist;
+
+	for (i = 0; i < MTK_WED_RX_PG_BM_CNT; i += MTK_WED_RX_BUF_PER_PAGE) {
+		dma_addr_t buf_addr = page_list[page_idx].phy_addr;
+		void *page = page_list[page_idx++].p;
+
+		if (!page)
+			break;
+
+		dma_unmap_page(dev->hw->dev, buf_addr, PAGE_SIZE,
+			       DMA_BIDIRECTIONAL);
+		__free_page(page);
+	}
+
+	dma_free_coherent(dev->hw->dev, dev->hw_rro.size * sizeof(*desc),
+			  desc, dev->hw_rro.desc_phys);
+
+free_pagelist:
+	kfree(page_list);
 }
 
 static void
 mtk_wed_free_rx_buffer(struct mtk_wed_device *dev)
 {
-	struct mtk_rxbm_desc *desc = dev->rx_buf_ring.desc;
+	struct mtk_wed_bm_desc *desc = dev->rx_buf_ring.desc;
 
 	if (!desc)
 		return;
@@ -450,6 +888,28 @@ mtk_wed_free_rx_buffer(struct mtk_wed_device *dev)
 	dev->wlan.release_rx_buf(dev);
 	dma_free_coherent(dev->hw->dev, dev->rx_buf_ring.size * sizeof(*desc),
 			  desc, dev->rx_buf_ring.desc_phys);
+
+	mtk_wed_hwrro_free_buffer(dev);
+}
+
+static void
+mtk_wed_hwrro_init(struct mtk_wed_device *dev)
+{
+	if (!mtk_wed_get_rx_capa(dev) || !dev->wlan.hw_rro)
+		return;
+
+	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->hw_rro.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);
 }
 
 static void
@@ -463,6 +923,8 @@ mtk_wed_rx_buffer_hw_init(struct mtk_wed_device *dev)
 	wed_w32(dev, MTK_WED_RX_BM_DYN_ALLOC_TH,
 		FIELD_PREP(MTK_WED_RX_BM_DYN_ALLOC_TH_H, 0xffff));
 	wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_WED_RX_BM_EN);
+
+	mtk_wed_hwrro_init(dev);
 }
 
 static void
@@ -498,13 +960,23 @@ mtk_wed_set_ext_int(struct mtk_wed_device *dev, bool en)
 {
 	u32 mask = MTK_WED_EXT_INT_STATUS_ERROR_MASK;
 
-	if (dev->hw->version == 1)
+	switch (dev->hw->version) {
+	case 1:
 		mask |= MTK_WED_EXT_INT_STATUS_TX_DRV_R_RESP_ERR;
-	else
+		break;
+	case 2:
 		mask |= MTK_WED_EXT_INT_STATUS_RX_FBUF_LO_TH |
 			MTK_WED_EXT_INT_STATUS_RX_FBUF_HI_TH |
 			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 |
+		       MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD;
+		break;
+	default:
+		break;
+	}
 
 	if (!dev->hw->num_flows)
 		mask &= ~MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD;
@@ -516,6 +988,9 @@ 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 enable)
 {
+	if (!mtk_wed_is_v2(dev->hw))
+		return;
+
 	if (enable) {
 		wed_w32(dev, MTK_WED_TXDP_CTRL, MTK_WED_TXDP_DW9_OVERWR);
 		wed_w32(dev, MTK_WED_TXP_DW1,
@@ -527,22 +1002,15 @@ mtk_wed_set_512_support(struct mtk_wed_device *dev, bool enable)
 	}
 }
 
-#define MTK_WFMDA_RX_DMA_EN	BIT(2)
-static void
-mtk_wed_check_wfdma_rx_fill(struct mtk_wed_device *dev, int idx)
+static int
+mtk_wed_check_wfdma_rx_fill(struct mtk_wed_device *dev,
+			    struct mtk_wed_ring *ring)
 {
-	u32 val;
 	int i;
 
-	if (!(dev->rx_ring[idx].flags & MTK_WED_RING_CONFIGURED))
-		return; /* queue is not configured by mt76 */
-
 	for (i = 0; i < 3; i++) {
-		u32 cur_idx;
+		u32 cur_idx = readl(ring->wpdma + MTK_WED_RING_OFS_CPU_IDX);
 
-		cur_idx = wed_r32(dev,
-				  MTK_WED_WPDMA_RING_RX_DATA(idx) +
-				  MTK_WED_RING_OFS_CPU_IDX);
 		if (cur_idx == MTK_WED_RX_RING_SIZE - 1)
 			break;
 
@@ -551,12 +1019,10 @@ mtk_wed_check_wfdma_rx_fill(struct mtk_wed_device *dev, int idx)
 
 	if (i == 3) {
 		dev_err(dev->hw->dev, "rx dma enable failed\n");
-		return;
+		return -ETIMEDOUT;
 	}
 
-	val = wifi_r32(dev, dev->wlan.wpdma_rx_glo - dev->wlan.phy_base) |
-	      MTK_WFMDA_RX_DMA_EN;
-	wifi_w32(dev, dev->wlan.wpdma_rx_glo - dev->wlan.phy_base, val);
+	return 0;
 }
 
 static void
@@ -577,7 +1043,7 @@ mtk_wed_dma_disable(struct mtk_wed_device *dev)
 		 MTK_WDMA_GLO_CFG_RX_INFO1_PRERES |
 		 MTK_WDMA_GLO_CFG_RX_INFO2_PRERES);
 
-	if (dev->hw->version == 1) {
+	if (mtk_wed_is_v1(dev->hw)) {
 		regmap_write(dev->hw->mirror, dev->hw->index * 4, 0);
 		wdma_clr(dev, MTK_WDMA_GLO_CFG,
 			 MTK_WDMA_GLO_CFG_RX_INFO3_PRERES);
@@ -590,6 +1056,14 @@ 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);
+
+		if (mtk_wed_is_v3_or_greater(dev->hw) &&
+		    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);
+		}
 	}
 
 	mtk_wed_set_512_support(dev, false);
@@ -606,7 +1080,7 @@ mtk_wed_stop(struct mtk_wed_device *dev)
 	wdma_w32(dev, MTK_WDMA_INT_GRP2, 0);
 	wed_w32(dev, MTK_WED_WPDMA_INT_MASK, 0);
 
-	if (dev->hw->version == 1)
+	if (!mtk_wed_get_rx_capa(dev))
 		return;
 
 	wed_w32(dev, MTK_WED_EXT_INT_MASK1, 0);
@@ -625,13 +1099,21 @@ 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->version == 1)
+	if (mtk_wed_is_v1(dev->hw))
 		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 (mtk_wed_is_v3_or_greater(dev->hw)) {
+		wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_AMSDU_EN);
+		wed_clr(dev, MTK_WED_RESET, MTK_WED_RESET_TX_AMSDU);
+		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
@@ -643,6 +1125,7 @@ __mtk_wed_detach(struct mtk_wed_device *dev)
 
 	mtk_wdma_rx_reset(dev);
 	mtk_wed_reset(dev, MTK_WED_RESET_WED);
+	mtk_wed_amsdu_free_buffer(dev);
 	mtk_wed_free_tx_buffer(dev);
 	mtk_wed_free_tx_rings(dev);
 
@@ -681,21 +1164,37 @@ mtk_wed_detach(struct mtk_wed_device *dev)
 	mutex_unlock(&hw_lock);
 }
 
-#define PCIE_BASE_ADDR0		0x11280000
 static void
 mtk_wed_bus_init(struct mtk_wed_device *dev)
 {
 	switch (dev->wlan.bus_type) {
 	case MTK_WED_BUS_PCIE: {
 		struct device_node *np = dev->hw->eth->dev->of_node;
-		struct regmap *regs;
 
-		regs = syscon_regmap_lookup_by_phandle(np,
-						       "mediatek,wed-pcie");
-		if (IS_ERR(regs))
-			break;
+		if (mtk_wed_is_v2(dev->hw)) {
+			struct regmap *regs;
+
+			regs = syscon_regmap_lookup_by_phandle(np,
+							       "mediatek,wed-pcie");
+			if (IS_ERR(regs))
+				break;
 
-		regmap_update_bits(regs, 0, BIT(0), BIT(0));
+			regmap_update_bits(regs, 0, BIT(0), BIT(0));
+		}
+
+		if (dev->wlan.msi) {
+			wed_w32(dev, MTK_WED_PCIE_CFG_INTM,
+				dev->hw->pcie_base | 0xc08);
+			wed_w32(dev, MTK_WED_PCIE_CFG_BASE,
+				dev->hw->pcie_base | 0xc04);
+			wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER, BIT(8));
+		} else {
+			wed_w32(dev, MTK_WED_PCIE_CFG_INTM,
+				dev->hw->pcie_base | 0x180);
+			wed_w32(dev, MTK_WED_PCIE_CFG_BASE,
+				dev->hw->pcie_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));
@@ -703,19 +1202,9 @@ 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);
-
-		wed_w32(dev, MTK_WED_PCIE_CFG_INTM, PCIE_BASE_ADDR0 | 0x180);
-		wed_w32(dev, MTK_WED_PCIE_CFG_BASE, PCIE_BASE_ADDR0 | 0x184);
-
-		/* pcie interrupt status trigger register */
-		wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER, BIT(24));
-		wed_r32(dev, MTK_WED_PCIE_INT_TRIGGER);
-
-		/* pola setting */
-		wed_set(dev, MTK_WED_PCIE_INT_CTRL,
-			MTK_WED_PCIE_INT_CTRL_MSK_EN_POLA);
+			MTK_WED_PCIE_INT_CTRL_MSK_IRQ_FILTER  |
+			FIELD_PREP(MTK_WED_PCIE_INT_CTRL_SRC_SEL,
+				   dev->hw->index));
 		break;
 	}
 	case MTK_WED_BUS_AXI:
@@ -731,38 +1220,55 @@ mtk_wed_bus_init(struct mtk_wed_device *dev)
 static void
 mtk_wed_set_wpdma(struct mtk_wed_device *dev)
 {
-	if (dev->hw->version == 1) {
-		wed_w32(dev, MTK_WED_WPDMA_CFG_BASE,  dev->wlan.wpdma_phys);
-	} else {
-		mtk_wed_bus_init(dev);
+	int i;
 
-		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_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);
+	if (mtk_wed_is_v1(dev->hw)) {
+		wed_w32(dev, MTK_WED_WPDMA_CFG_BASE,  dev->wlan.wpdma_phys);
+		return;
 	}
+
+	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_FREE, dev->wlan.wpdma_txfree);
+
+	if (!mtk_wed_get_rx_capa(dev))
+		return;
+
+	wed_w32(dev, MTK_WED_WPDMA_RX_GLO_CFG, dev->wlan.wpdma_rx_glo);
+	wed_w32(dev, dev->hw->soc->regmap.wpdma_rx_ring0, dev->wlan.wpdma_rx);
+
+	if (!dev->wlan.hw_rro)
+		return;
+
+	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);
 }
 
 static void
 mtk_wed_hw_init_early(struct mtk_wed_device *dev)
 {
-	u32 mask, set;
+	u32 set = FIELD_PREP(MTK_WED_WDMA_GLO_CFG_BT_SIZE, 2);
+	u32 mask = MTK_WED_WDMA_GLO_CFG_BT_SIZE;
 
 	mtk_wed_deinit(dev);
 	mtk_wed_reset(dev, MTK_WED_RESET_WED);
 	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 (!mtk_wed_is_v3_or_greater(dev->hw)) {
+		mask |= MTK_WED_WDMA_GLO_CFG_DYNAMIC_DMAD_RECYCLE |
+			MTK_WED_WDMA_GLO_CFG_RX_DIS_FSM_AUTO_IDLE;
+		set |= 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->hw->version == 1) {
+	if (mtk_wed_is_v1(dev->hw)) {
 		u32 offset = dev->hw->index ? 0x04000400 : 0;
 
 		wdma_set(dev, MTK_WDMA_GLO_CFG,
@@ -907,11 +1413,18 @@ mtk_wed_route_qm_hw_init(struct mtk_wed_device *dev)
 	}
 
 	/* 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 (mtk_wed_is_v2(dev->hw)) {
+		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);
 }
@@ -924,34 +1437,30 @@ mtk_wed_hw_init(struct mtk_wed_device *dev)
 
 	dev->init_done = true;
 	mtk_wed_set_ext_int(dev, false);
-	wed_w32(dev, MTK_WED_TX_BM_CTRL,
-		MTK_WED_TX_BM_CTRL_PAUSE |
-		FIELD_PREP(MTK_WED_TX_BM_CTRL_VLD_GRP_NUM,
-			   dev->tx_buf_ring.size / 128) |
-		FIELD_PREP(MTK_WED_TX_BM_CTRL_RSV_GRP_NUM,
-			   MTK_WED_TX_RING_SIZE / 256));
 
 	wed_w32(dev, MTK_WED_TX_BM_BASE, dev->tx_buf_ring.desc_phys);
-
 	wed_w32(dev, MTK_WED_TX_BM_BUF_LEN, MTK_WED_PKT_SIZE);
 
-	if (dev->hw->version == 1) {
-		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));
+	if (mtk_wed_is_v1(dev->hw)) {
+		wed_w32(dev, MTK_WED_TX_BM_CTRL,
+			MTK_WED_TX_BM_CTRL_PAUSE |
+			FIELD_PREP(MTK_WED_TX_BM_CTRL_VLD_GRP_NUM,
+				   dev->tx_buf_ring.size / 128) |
+			FIELD_PREP(MTK_WED_TX_BM_CTRL_RSV_GRP_NUM,
+				   MTK_WED_TX_RING_SIZE / 256));
 		wed_w32(dev, MTK_WED_TX_BM_DYN_THR,
 			FIELD_PREP(MTK_WED_TX_BM_DYN_THR_LO, 1) |
 			MTK_WED_TX_BM_DYN_THR_HI);
-	} else {
-		wed_w32(dev, MTK_WED_TX_BM_TKID_V2,
-			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));
+	} else if (mtk_wed_is_v2(dev->hw)) {
+		wed_w32(dev, MTK_WED_TX_BM_CTRL,
+			MTK_WED_TX_BM_CTRL_PAUSE |
+			FIELD_PREP(MTK_WED_TX_BM_CTRL_VLD_GRP_NUM,
+				   dev->tx_buf_ring.size / 128) |
+			FIELD_PREP(MTK_WED_TX_BM_CTRL_RSV_GRP_NUM,
+				   MTK_WED_TX_RING_SIZE / 256));
+		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_DYN_THR,
 			FIELD_PREP(MTK_WED_TX_BM_DYN_THR_LO_V2, 0) |
 			MTK_WED_TX_BM_DYN_THR_HI_V2);
@@ -961,31 +1470,71 @@ mtk_wed_hw_init(struct mtk_wed_device *dev)
 				   dev->tx_buf_ring.size / 128) |
 			FIELD_PREP(MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM,
 				   dev->tx_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);
 	}
 
+	wed_w32(dev, dev->hw->soc->regmap.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));
+
 	mtk_wed_reset(dev, MTK_WED_RESET_TX_BM);
 
-	if (dev->hw->version == 1) {
+	if (mtk_wed_is_v3_or_greater(dev->hw)) {
+		/* switch to new bm architecture */
+		wed_clr(dev, MTK_WED_TX_BM_CTRL,
+			MTK_WED_TX_BM_CTRL_LEGACY_EN);
+
+		wed_w32(dev, MTK_WED_TX_TKID_CTRL,
+			MTK_WED_TX_TKID_CTRL_PAUSE |
+			FIELD_PREP(MTK_WED_TX_TKID_CTRL_VLD_GRP_NUM_V3,
+				   dev->wlan.nbuf / 128) |
+			FIELD_PREP(MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM_V3,
+				   dev->wlan.nbuf / 128));
+		/* return SKBID + SDP back to bm */
+		wed_set(dev, MTK_WED_TX_TKID_CTRL,
+			MTK_WED_TX_TKID_CTRL_FREE_FORMAT);
+
+		wed_w32(dev, MTK_WED_TX_BM_INIT_PTR,
+			MTK_WED_TX_BM_PKT_CNT |
+			MTK_WED_TX_BM_INIT_SW_TAIL_IDX);
+	}
+
+	if (mtk_wed_is_v1(dev->hw)) {
 		wed_set(dev, MTK_WED_CTRL,
 			MTK_WED_CTRL_WED_TX_BM_EN |
 			MTK_WED_CTRL_WED_TX_FREE_AGENT_EN);
-	} else {
-		wed_clr(dev, MTK_WED_TX_TKID_CTRL, MTK_WED_TX_TKID_CTRL_PAUSE);
+	} else if (mtk_wed_get_rx_capa(dev)) {
 		/* rx hw init */
 		wed_w32(dev, MTK_WED_WPDMA_RX_D_RST_IDX,
 			MTK_WED_WPDMA_RX_D_RST_CRX_IDX |
 			MTK_WED_WPDMA_RX_D_RST_DRV_IDX);
 		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_buffer_hw_init(dev);
 		mtk_wed_rro_hw_init(dev);
 		mtk_wed_route_qm_hw_init(dev);
 	}
 
 	wed_clr(dev, MTK_WED_TX_BM_CTRL, MTK_WED_TX_BM_CTRL_PAUSE);
+	if (!mtk_wed_is_v1(dev->hw))
+		wed_clr(dev, MTK_WED_TX_TKID_CTRL, MTK_WED_TX_TKID_CTRL_PAUSE);
 }
 
 static void
@@ -1008,23 +1557,6 @@ mtk_wed_ring_reset(struct mtk_wed_ring *ring, int size, bool tx)
 	}
 }
 
-static u32
-mtk_wed_check_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
-{
-	return !!(wed_r32(dev, reg) & mask);
-}
-
-static int
-mtk_wed_poll_busy(struct mtk_wed_device *dev, u32 reg, u32 mask)
-{
-	int sleep = 15000;
-	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_wed_rx_reset(struct mtk_wed_device *dev)
 {
@@ -1038,13 +1570,33 @@ mtk_wed_rx_reset(struct mtk_wed_device *dev)
 	if (ret)
 		return ret;
 
+	if (dev->wlan.hw_rro) {
+		wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_WED_RX_IND_CMD_EN);
+		mtk_wed_poll_busy(dev, MTK_WED_RRO_RX_HW_STS,
+				  MTK_WED_RX_IND_CMD_BUSY);
+		mtk_wed_reset(dev, MTK_WED_RESET_RRO_RX_TO_PG);
+	}
+
 	wed_clr(dev, MTK_WED_WPDMA_RX_D_GLO_CFG, MTK_WED_WPDMA_RX_D_RX_DRV_EN);
 	ret = mtk_wed_poll_busy(dev, MTK_WED_WPDMA_RX_D_GLO_CFG,
 				MTK_WED_WPDMA_RX_D_RX_DRV_BUSY);
+	if (!ret && mtk_wed_is_v3_or_greater(dev->hw))
+		ret = mtk_wed_poll_busy(dev, MTK_WED_WPDMA_RX_D_PREF_CFG,
+					MTK_WED_WPDMA_RX_D_PREF_BUSY);
 	if (ret) {
 		mtk_wed_reset(dev, MTK_WED_RESET_WPDMA_INT_AGENT);
 		mtk_wed_reset(dev, MTK_WED_RESET_WPDMA_RX_D_DRV);
 	} else {
+		if (mtk_wed_is_v3_or_greater(dev->hw)) {
+			/* 1.a. disable prefetch HW */
+			wed_clr(dev, MTK_WED_WPDMA_RX_D_PREF_CFG,
+				MTK_WED_WPDMA_RX_D_PREF_EN);
+			mtk_wed_poll_busy(dev, MTK_WED_WPDMA_RX_D_PREF_CFG,
+					  MTK_WED_WPDMA_RX_D_PREF_BUSY);
+			wed_w32(dev, MTK_WED_WPDMA_RX_D_RST_IDX,
+				MTK_WED_WPDMA_RX_D_RST_DRV_IDX_ALL);
+		}
+
 		wed_w32(dev, MTK_WED_WPDMA_RX_D_RST_IDX,
 			MTK_WED_WPDMA_RX_D_RST_CRX_IDX |
 			MTK_WED_WPDMA_RX_D_RST_DRV_IDX);
@@ -1072,23 +1624,52 @@ mtk_wed_rx_reset(struct mtk_wed_device *dev)
 		wed_w32(dev, MTK_WED_RROQM_RST_IDX, 0);
 	}
 
+	if (dev->wlan.hw_rro) {
+		/* disable rro msdu page drv */
+		wed_clr(dev, MTK_WED_RRO_MSDU_PG_RING2_CFG,
+			MTK_WED_RRO_MSDU_PG_DRV_EN);
+
+		/* disable rro data drv */
+		wed_clr(dev, MTK_WED_RRO_RX_D_CFG(2), MTK_WED_RRO_RX_D_DRV_EN);
+
+		/* rro msdu page drv reset */
+		wed_w32(dev, MTK_WED_RRO_MSDU_PG_RING2_CFG,
+			MTK_WED_RRO_MSDU_PG_DRV_CLR);
+		mtk_wed_poll_busy(dev, MTK_WED_RRO_MSDU_PG_RING2_CFG,
+				  MTK_WED_RRO_MSDU_PG_DRV_CLR);
+
+		/* rro data drv reset */
+		wed_w32(dev, MTK_WED_RRO_RX_D_CFG(2),
+			MTK_WED_RRO_RX_D_DRV_CLR);
+		mtk_wed_poll_busy(dev, MTK_WED_RRO_RX_D_CFG(2),
+				  MTK_WED_RRO_RX_D_DRV_CLR);
+	}
+
 	/* reset route qm */
 	wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_RX_ROUTE_QM_EN);
 	ret = mtk_wed_poll_busy(dev, MTK_WED_CTRL,
 				MTK_WED_CTRL_RX_ROUTE_QM_BUSY);
-	if (ret)
+	if (ret) {
 		mtk_wed_reset(dev, MTK_WED_RESET_RX_ROUTE_QM);
-	else
-		wed_set(dev, MTK_WED_RTQM_GLO_CFG,
-			MTK_WED_RTQM_Q_RST);
+	} else if (mtk_wed_is_v3_or_greater(dev->hw)) {
+		wed_set(dev, MTK_WED_RTQM_RST, BIT(0));
+		wed_clr(dev, MTK_WED_RTQM_RST, BIT(0));
+		mtk_wed_reset(dev, MTK_WED_RESET_RX_ROUTE_QM);
+	} else {
+		wed_set(dev, MTK_WED_RTQM_GLO_CFG, MTK_WED_RTQM_Q_RST);
+	}
 
 	/* reset tx wdma */
 	mtk_wdma_tx_reset(dev);
 
 	/* reset tx wdma drv */
 	wed_clr(dev, MTK_WED_WDMA_GLO_CFG, MTK_WED_WDMA_GLO_CFG_TX_DRV_EN);
-	mtk_wed_poll_busy(dev, MTK_WED_CTRL,
-			  MTK_WED_CTRL_WDMA_INT_AGENT_BUSY);
+	if (mtk_wed_is_v3_or_greater(dev->hw))
+		mtk_wed_poll_busy(dev, MTK_WED_WPDMA_STATUS,
+				  MTK_WED_WPDMA_STATUS_TX_DRV);
+	else
+		mtk_wed_poll_busy(dev, MTK_WED_CTRL,
+				  MTK_WED_CTRL_WDMA_INT_AGENT_BUSY);
 	mtk_wed_reset(dev, MTK_WED_RESET_WDMA_TX_DRV);
 
 	/* reset wed rx dma */
@@ -1098,13 +1679,8 @@ mtk_wed_rx_reset(struct mtk_wed_device *dev)
 	if (ret) {
 		mtk_wed_reset(dev, MTK_WED_RESET_WED_RX_DMA);
 	} else {
-		struct mtk_eth *eth = dev->hw->eth;
-
-		if (mtk_is_netsys_v2_or_greater(eth))
-			wed_set(dev, MTK_WED_RESET_IDX,
-				MTK_WED_RESET_IDX_RX_V2);
-		else
-			wed_set(dev, MTK_WED_RESET_IDX, MTK_WED_RESET_IDX_RX);
+		wed_set(dev, MTK_WED_RESET_IDX,
+			dev->hw->soc->regmap.reset_idx_rx_mask);
 		wed_w32(dev, MTK_WED_RESET_IDX, 0);
 	}
 
@@ -1114,6 +1690,14 @@ mtk_wed_rx_reset(struct mtk_wed_device *dev)
 			  MTK_WED_CTRL_WED_RX_BM_BUSY);
 	mtk_wed_reset(dev, MTK_WED_RESET_RX_BM);
 
+	if (dev->wlan.hw_rro) {
+		wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_WED_RX_PG_BM_EN);
+		mtk_wed_poll_busy(dev, MTK_WED_CTRL,
+				  MTK_WED_CTRL_WED_RX_PG_BM_BUSY);
+		wed_set(dev, MTK_WED_RESET, MTK_WED_RESET_RX_PG_BM);
+		wed_clr(dev, MTK_WED_RESET, MTK_WED_RESET_RX_PG_BM);
+	}
+
 	/* wo change to enable state */
 	val = MTK_WED_WO_STATE_ENABLE;
 	ret = mtk_wed_mcu_send_msg(wo, MTK_WED_MODULE_ID_WO,
@@ -1131,6 +1715,7 @@ mtk_wed_rx_reset(struct mtk_wed_device *dev)
 				   false);
 	}
 	mtk_wed_free_rx_buffer(dev);
+	mtk_wed_hwrro_free_buffer(dev);
 
 	return 0;
 }
@@ -1157,21 +1742,48 @@ mtk_wed_reset_dma(struct mtk_wed_device *dev)
 	if (busy) {
 		mtk_wed_reset(dev, MTK_WED_RESET_WED_TX_DMA);
 	} else {
-		wed_w32(dev, MTK_WED_RESET_IDX, MTK_WED_RESET_IDX_TX);
+		wed_w32(dev, MTK_WED_RESET_IDX,
+			dev->hw->soc->regmap.reset_idx_tx_mask);
 		wed_w32(dev, MTK_WED_RESET_IDX, 0);
 	}
 
 	/* 2. reset WDMA rx DMA */
 	busy = !!mtk_wdma_rx_reset(dev);
-	wed_clr(dev, MTK_WED_WDMA_GLO_CFG, MTK_WED_WDMA_GLO_CFG_RX_DRV_EN);
+	if (mtk_wed_is_v3_or_greater(dev->hw)) {
+		val = MTK_WED_WDMA_GLO_CFG_RX_DIS_FSM_AUTO_IDLE |
+		      wed_r32(dev, MTK_WED_WDMA_GLO_CFG);
+		val &= ~MTK_WED_WDMA_GLO_CFG_RX_DRV_EN;
+		wed_w32(dev, MTK_WED_WDMA_GLO_CFG, val);
+	} else {
+		wed_clr(dev, MTK_WED_WDMA_GLO_CFG,
+			MTK_WED_WDMA_GLO_CFG_RX_DRV_EN);
+	}
+
 	if (!busy)
 		busy = mtk_wed_poll_busy(dev, MTK_WED_WDMA_GLO_CFG,
 					 MTK_WED_WDMA_GLO_CFG_RX_DRV_BUSY);
+	if (!busy && mtk_wed_is_v3_or_greater(dev->hw))
+		busy = mtk_wed_poll_busy(dev, MTK_WED_WDMA_RX_PREF_CFG,
+					 MTK_WED_WDMA_RX_PREF_BUSY);
 
 	if (busy) {
 		mtk_wed_reset(dev, MTK_WED_RESET_WDMA_INT_AGENT);
 		mtk_wed_reset(dev, MTK_WED_RESET_WDMA_RX_DRV);
 	} else {
+		if (mtk_wed_is_v3_or_greater(dev->hw)) {
+			/* 1.a. disable prefetch HW */
+			wed_clr(dev, MTK_WED_WDMA_RX_PREF_CFG,
+				MTK_WED_WDMA_RX_PREF_EN);
+			mtk_wed_poll_busy(dev, MTK_WED_WDMA_RX_PREF_CFG,
+					  MTK_WED_WDMA_RX_PREF_BUSY);
+			wed_clr(dev, MTK_WED_WDMA_RX_PREF_CFG,
+				MTK_WED_WDMA_RX_PREF_DDONE2_EN);
+
+			/* 2. Reset dma index */
+			wed_w32(dev, MTK_WED_WDMA_RESET_IDX,
+				MTK_WED_WDMA_RESET_IDX_RX_ALL);
+		}
+
 		wed_w32(dev, MTK_WED_WDMA_RESET_IDX,
 			MTK_WED_WDMA_RESET_IDX_RX | MTK_WED_WDMA_RESET_IDX_DRV);
 		wed_w32(dev, MTK_WED_WDMA_RESET_IDX, 0);
@@ -1187,8 +1799,13 @@ mtk_wed_reset_dma(struct mtk_wed_device *dev)
 	wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_WED_TX_FREE_AGENT_EN);
 
 	for (i = 0; i < 100; i++) {
-		val = wed_r32(dev, MTK_WED_TX_BM_INTF);
-		if (FIELD_GET(MTK_WED_TX_BM_INTF_TKFIFO_FDEP, val) == 0x40)
+		if (mtk_wed_is_v1(dev->hw))
+			val = FIELD_GET(MTK_WED_TX_BM_INTF_TKFIFO_FDEP,
+					wed_r32(dev, MTK_WED_TX_BM_INTF));
+		else
+			val = FIELD_GET(MTK_WED_TX_TKID_INTF_TKFIFO_FDEP,
+					wed_r32(dev, MTK_WED_TX_TKID_INTF));
+		if (val == 0x40)
 			break;
 	}
 
@@ -1210,6 +1827,8 @@ mtk_wed_reset_dma(struct mtk_wed_device *dev)
 		mtk_wed_reset(dev, MTK_WED_RESET_WPDMA_INT_AGENT);
 		mtk_wed_reset(dev, MTK_WED_RESET_WPDMA_TX_DRV);
 		mtk_wed_reset(dev, MTK_WED_RESET_WPDMA_RX_DRV);
+		if (mtk_wed_is_v3_or_greater(dev->hw))
+			wed_w32(dev, MTK_WED_RX1_CTRL2, 0);
 	} else {
 		wed_w32(dev, MTK_WED_WPDMA_RESET_IDX,
 			MTK_WED_WPDMA_RESET_IDX_TX |
@@ -1218,7 +1837,7 @@ mtk_wed_reset_dma(struct mtk_wed_device *dev)
 	}
 
 	dev->init_done = false;
-	if (dev->hw->version == 1)
+	if (mtk_wed_is_v1(dev->hw))
 		return;
 
 	if (!busy) {
@@ -1226,7 +1845,14 @@ mtk_wed_reset_dma(struct mtk_wed_device *dev)
 		wed_w32(dev, MTK_WED_RESET_IDX, 0);
 	}
 
-	mtk_wed_rx_reset(dev);
+	if (mtk_wed_is_v3_or_greater(dev->hw)) {
+		/* reset amsdu engine */
+		wed_clr(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_AMSDU_EN);
+		mtk_wed_reset(dev, MTK_WED_RESET_TX_AMSDU);
+	}
+
+	if (mtk_wed_get_rx_capa(dev))
+		mtk_wed_rx_reset(dev);
 }
 
 static int
@@ -1249,7 +1875,6 @@ static int
 mtk_wed_wdma_rx_ring_setup(struct mtk_wed_device *dev, int idx, int size,
 			   bool reset)
 {
-	u32 desc_size = sizeof(struct mtk_wdma_desc) * dev->hw->version;
 	struct mtk_wed_ring *wdma;
 
 	if (idx >= ARRAY_SIZE(dev->rx_wdma))
@@ -1257,7 +1882,7 @@ mtk_wed_wdma_rx_ring_setup(struct mtk_wed_device *dev, int idx, int size,
 
 	wdma = &dev->rx_wdma[idx];
 	if (!reset && mtk_wed_ring_alloc(dev, wdma, MTK_WED_WDMA_RING_SIZE,
-					 desc_size, true))
+					 dev->hw->soc->wdma_desc_size, true))
 		return -ENOMEM;
 
 	wdma_w32(dev, MTK_WDMA_RING_RX(idx) + MTK_WED_RING_OFS_BASE,
@@ -1278,7 +1903,6 @@ static int
 mtk_wed_wdma_tx_ring_setup(struct mtk_wed_device *dev, int idx, int size,
 			   bool reset)
 {
-	u32 desc_size = sizeof(struct mtk_wdma_desc) * dev->hw->version;
 	struct mtk_wed_ring *wdma;
 
 	if (idx >= ARRAY_SIZE(dev->tx_wdma))
@@ -1286,9 +1910,27 @@ mtk_wed_wdma_tx_ring_setup(struct mtk_wed_device *dev, int idx, int size,
 
 	wdma = &dev->tx_wdma[idx];
 	if (!reset && mtk_wed_ring_alloc(dev, wdma, MTK_WED_WDMA_RING_SIZE,
-					 desc_size, true))
+					 dev->hw->soc->wdma_desc_size, true))
 		return -ENOMEM;
 
+	if (mtk_wed_is_v3_or_greater(dev->hw)) {
+		struct mtk_wdma_desc *desc = wdma->desc;
+		int i;
+
+		for (i = 0; i < MTK_WED_WDMA_RING_SIZE; i++) {
+			desc->buf0 = 0;
+			desc->ctrl = cpu_to_le32(MTK_WDMA_DESC_CTRL_DMA_DONE);
+			desc->buf1 = 0;
+			desc->info = cpu_to_le32(MTK_WDMA_TXD0_DESC_INFO_DMA_DONE);
+			desc++;
+			desc->buf0 = 0;
+			desc->ctrl = cpu_to_le32(MTK_WDMA_DESC_CTRL_DMA_DONE);
+			desc->buf1 = 0;
+			desc->info = cpu_to_le32(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,
@@ -1344,7 +1986,7 @@ mtk_wed_configure_irq(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->hw->version == 1) {
+	if (mtk_wed_is_v1(dev->hw)) {
 		wed_w32(dev, MTK_WED_PCIE_INT_TRIGGER,
 			MTK_WED_PCIE_INT_TRIGGER_STATUS);
 
@@ -1354,8 +1996,9 @@ mtk_wed_configure_irq(struct mtk_wed_device *dev, u32 irq_mask)
 
 		wed_clr(dev, MTK_WED_WDMA_INT_CTRL, wdma_mask);
 	} else {
-		wdma_mask |= FIELD_PREP(MTK_WDMA_INT_MASK_TX_DONE,
-					GENMASK(1, 0));
+		if (mtk_wed_is_v3_or_greater(dev->hw))
+			wed_set(dev, MTK_WED_CTRL, MTK_WED_CTRL_TX_TKID_ALI_EN);
+
 		/* initail tx interrupt trigger */
 		wed_w32(dev, MTK_WED_WPDMA_INT_CTRL_TX,
 			MTK_WED_WPDMA_INT_CTRL_TX0_DONE_EN |
@@ -1374,15 +2017,20 @@ mtk_wed_configure_irq(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]));
+
+			wdma_mask |= FIELD_PREP(MTK_WDMA_INT_MASK_TX_DONE,
+						GENMASK(1, 0));
+		}
 
 		wed_w32(dev, MTK_WED_WDMA_INT_CLR, wdma_mask);
 		wed_set(dev, MTK_WED_WDMA_INT_CTRL,
@@ -1398,55 +2046,280 @@ mtk_wed_configure_irq(struct mtk_wed_device *dev, u32 irq_mask)
 	wed_w32(dev, MTK_WED_INT_MASK, irq_mask);
 }
 
+#define MTK_WFMDA_RX_DMA_EN	BIT(2)
 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);
+	int i;
+
+	if (!mtk_wed_is_v3_or_greater(dev->hw)) {
+		wed_set(dev, MTK_WED_WPDMA_INT_CTRL,
+			MTK_WED_WPDMA_INT_CTRL_SUBRT_ADV);
+		wed_set(dev, MTK_WED_WPDMA_GLO_CFG,
+			MTK_WED_WPDMA_GLO_CFG_TX_DRV_EN |
+			MTK_WED_WPDMA_GLO_CFG_RX_DRV_EN);
+		wdma_set(dev, MTK_WDMA_GLO_CFG,
+			 MTK_WDMA_GLO_CFG_TX_DMA_EN |
+			 MTK_WDMA_GLO_CFG_RX_INFO1_PRERES |
+			 MTK_WDMA_GLO_CFG_RX_INFO2_PRERES);
+		wed_set(dev, MTK_WED_WPDMA_CTRL, MTK_WED_WPDMA_CTRL_SDL1_FIXED);
+	} else {
+		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_DDONE2_WR);
+		wdma_set(dev, MTK_WDMA_GLO_CFG, MTK_WDMA_GLO_CFG_TX_DMA_EN);
+	}
 
 	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_WPDMA_GLO_CFG,
-		MTK_WED_WPDMA_GLO_CFG_TX_DRV_EN |
-		MTK_WED_WPDMA_GLO_CFG_RX_DRV_EN);
+
 	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_RX_INFO1_PRERES |
-		 MTK_WDMA_GLO_CFG_RX_INFO2_PRERES);
-
-	if (dev->hw->version == 1) {
+	if (mtk_wed_is_v1(dev->hw)) {
 		wdma_set(dev, MTK_WDMA_GLO_CFG,
 			 MTK_WDMA_GLO_CFG_RX_INFO3_PRERES);
-	} else {
-		int i;
+		return;
+	}
 
-		wed_set(dev, MTK_WED_WPDMA_CTRL,
-			MTK_WED_WPDMA_CTRL_SDL1_FIXED);
+	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);
 
-		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_is_v3_or_greater(dev->hw)) {
+		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_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_RX_DRV_R0_PKT_PROC |
-			MTK_WED_WPDMA_GLO_CFG_RX_DRV_R0_CRX_SYNC);
+			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);
 
-		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);
+		wdma_set(dev, MTK_WDMA_PREF_RX_CFG, MTK_WDMA_PREF_RX_CFG_PREF_EN);
+		wdma_set(dev, MTK_WDMA_WRBK_RX_CFG, MTK_WDMA_WRBK_RX_CFG_WRBK_EN);
+	}
 
-		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));
+	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_set(dev, MTK_WED_WDMA_GLO_CFG,
+		MTK_WED_WDMA_GLO_CFG_TX_DRV_EN |
+		MTK_WED_WDMA_GLO_CFG_TX_DDONE_CHK);
+
+	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));
+
+	if (mtk_wed_is_v3_or_greater(dev->hw)) {
+		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);
+	}
+
+	for (i = 0; i < MTK_WED_RX_QUEUES; i++) {
+		struct mtk_wed_ring *ring = &dev->rx_ring[i];
+		u32 val;
+
+		if (!(ring->flags & MTK_WED_RING_CONFIGURED))
+			continue; /* queue is not configured by mt76 */
+
+		if (mtk_wed_check_wfdma_rx_fill(dev, ring)) {
+			dev_err(dev->hw->dev,
+				"rx_ring(%d) dma enable failed\n", i);
+			continue;
+		}
+
+		val = wifi_r32(dev,
+			       dev->wlan.wpdma_rx_glo -
+			       dev->wlan.phy_base) | MTK_WFMDA_RX_DMA_EN;
+		wifi_w32(dev,
+			 dev->wlan.wpdma_rx_glo - dev->wlan.phy_base,
+			 val);
+	}
+}
+
+static void
+mtk_wed_start_hw_rro(struct mtk_wed_device *dev, u32 irq_mask, bool reset)
+{
+	int i;
+
+	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.hw_rro)
+		return;
+
+	if (reset) {
+		wed_set(dev, MTK_WED_RRO_MSDU_PG_RING2_CFG,
+			MTK_WED_RRO_MSDU_PG_DRV_EN);
+		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 (i = 0; i < MTK_WED_RX_QUEUES; i++) {
+		struct mtk_wed_ring *ring = &dev->rx_rro_ring[i];
+
+		if (!(ring->flags & MTK_WED_RING_CONFIGURED))
+			continue;
+
+		if (mtk_wed_check_wfdma_rx_fill(dev, ring))
+			dev_err(dev->hw->dev,
+				"rx_rro_ring(%d) initialization failed\n", i);
+	}
+
+	for (i = 0; i < MTK_WED_RX_PAGE_QUEUES; i++) {
+		struct mtk_wed_ring *ring = &dev->rx_page_ring[i];
+
+		if (!(ring->flags & MTK_WED_RING_CONFIGURED))
+			continue;
+
+		if (mtk_wed_check_wfdma_rx_fill(dev, ring))
+			dev_err(dev->hw->dev,
+				"rx_page_ring(%d) initialization failed\n", i);
+	}
+}
+
+static void
+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;
+}
+
+static void
+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;
+}
+
+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, count = 0;
 
-		for (i = 0; i < MTK_WED_RX_QUEUES; i++)
-			mtk_wed_check_wfdma_rx_fill(dev, i);
+	ring->wpdma = regs;
+	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) && count++ < 100)
+			val = wed_r32(dev, MTK_WED_ADDR_ELEM_TBL_CFG);
+		if (count >= 100)
+			dev_err(dev->hw->dev,
+				"write ba session base failed\n");
+	}
+
+	/* 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));
+
+		count = 0;
+		val = wed_r32(dev, MTK_WED_PN_CHECK_CFG);
+		while (!(val & MTK_WED_PN_CHECK_WR_RDY) && count++ < 100)
+			val = wed_r32(dev, MTK_WED_PN_CHECK_CFG);
+		if (count >= 100)
+			dev_err(dev->hw->dev,
+				"session(%d) initialization failed\n", i);
 	}
+
+	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 void
@@ -1466,14 +2339,14 @@ mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
 
 	mtk_wed_set_ext_int(dev, true);
 
-	if (dev->hw->version == 1) {
+	if (mtk_wed_is_v1(dev->hw)) {
 		u32 val = dev->wlan.wpdma_phys | MTK_PCIE_MIRROR_MAP_EN |
 			  FIELD_PREP(MTK_PCIE_MIRROR_MAP_WED_ID,
 				     dev->hw->index);
 
 		val |= BIT(0) | (BIT(1) * !!dev->hw->index);
 		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);
@@ -1483,12 +2356,18 @@ mtk_wed_start(struct mtk_wed_device *dev, u32 irq_mask)
 		wed_r32(dev, MTK_WED_EXT_INT_MASK1);
 		wed_r32(dev, MTK_WED_EXT_INT_MASK2);
 
+		if (mtk_wed_is_v3_or_greater(dev->hw)) {
+			wed_w32(dev, MTK_WED_EXT_INT_MASK3,
+				MTK_WED_EXT_INT_STATUS_WPDMA_MID_RDY);
+			wed_r32(dev, MTK_WED_EXT_INT_MASK3);
+		}
+
 		if (mtk_wed_rro_cfg(dev))
 			return;
-
 	}
 
 	mtk_wed_set_512_support(dev, dev->wlan.wcid_512);
+	mtk_wed_amsdu_init(dev);
 
 	mtk_wed_dma_enable(dev);
 	dev->running = true;
@@ -1535,6 +2414,7 @@ mtk_wed_attach(struct mtk_wed_device *dev)
 	dev->irq = hw->irq;
 	dev->wdma_idx = hw->index;
 	dev->version = hw->version;
+	dev->hw->pcie_base = mtk_wed_get_pcie_base(dev);
 
 	if (hw->eth->dma_dev == hw->eth->dev &&
 	    of_dma_is_coherent(hw->eth->dev->of_node))
@@ -1544,6 +2424,10 @@ mtk_wed_attach(struct mtk_wed_device *dev)
 	if (ret)
 		goto out;
 
+	ret = mtk_wed_amsdu_buffer_alloc(dev);
+	if (ret)
+		goto out;
+
 	if (mtk_wed_get_rx_capa(dev)) {
 		ret = mtk_wed_rro_alloc(dev);
 		if (ret)
@@ -1551,13 +2435,14 @@ mtk_wed_attach(struct mtk_wed_device *dev)
 	}
 
 	mtk_wed_hw_init_early(dev);
-	if (hw->version == 1) {
+	if (mtk_wed_is_v1(hw))
 		regmap_update_bits(hw->hifsys, HIFSYS_DMA_AG_MAP,
 				   BIT(hw->index), 0);
-	} else {
+	else
 		dev->rev_id = wed_r32(dev, MTK_WED_REV_ID);
+
+	if (mtk_wed_get_rx_capa(dev))
 		ret = mtk_wed_wo_init(hw);
-	}
 out:
 	if (ret) {
 		dev_err(dev->hw->dev, "failed to attach wed device\n");
@@ -1601,6 +2486,23 @@ mtk_wed_tx_ring_setup(struct mtk_wed_device *dev, int idx, void __iomem *regs,
 	ring->reg_base = MTK_WED_RING_TX(idx);
 	ring->wpdma = regs;
 
+	if (mtk_wed_is_v3_or_greater(dev->hw) && 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);
+	}
+
 	/* WED -> WPDMA */
 	wpdma_tx_w32(dev, idx, MTK_WED_RING_OFS_BASE, ring->desc_phys);
 	wpdma_tx_w32(dev, idx, MTK_WED_RING_OFS_COUNT, MTK_WED_TX_RING_SIZE);
@@ -1619,7 +2521,7 @@ static int
 mtk_wed_txfree_ring_setup(struct mtk_wed_device *dev, void __iomem *regs)
 {
 	struct mtk_wed_ring *ring = &dev->txfree_ring;
-	int i, index = dev->hw->version == 1;
+	int i, index = mtk_wed_is_v1(dev->hw);
 
 	/*
 	 * For txfree event handling, the same DMA ring is shared between WED
@@ -1675,15 +2577,13 @@ mtk_wed_rx_ring_setup(struct mtk_wed_device *dev, int idx, void __iomem *regs,
 static u32
 mtk_wed_irq_get(struct mtk_wed_device *dev, u32 mask)
 {
-	u32 val, ext_mask = MTK_WED_EXT_INT_STATUS_ERROR_MASK;
+	u32 val, ext_mask;
 
-	if (dev->hw->version == 1)
-		ext_mask |= MTK_WED_EXT_INT_STATUS_TX_DRV_R_RESP_ERR;
+	if (mtk_wed_is_v3_or_greater(dev->hw))
+		ext_mask = MTK_WED_EXT_INT_STATUS_RX_DRV_COHERENT |
+			   MTK_WED_EXT_INT_STATUS_TKID_WO_PYLD;
 	else
-		ext_mask |= MTK_WED_EXT_INT_STATUS_RX_FBUF_LO_TH |
-			    MTK_WED_EXT_INT_STATUS_RX_FBUF_HI_TH |
-			    MTK_WED_EXT_INT_STATUS_RX_DRV_COHERENT |
-			    MTK_WED_EXT_INT_STATUS_TX_DMA_W_RESP_ERR;
+		ext_mask = MTK_WED_EXT_INT_STATUS_ERROR_MASK;
 
 	val = wed_r32(dev, MTK_WED_EXT_INT_STATUS);
 	wed_w32(dev, MTK_WED_EXT_INT_STATUS, val);
@@ -1713,19 +2613,20 @@ mtk_wed_irq_set_mask(struct mtk_wed_device *dev, u32 mask)
 int mtk_wed_flow_add(int index)
 {
 	struct mtk_wed_hw *hw = hw_list[index];
-	int ret;
+	int ret = 0;
 
-	if (!hw || !hw->wed_dev)
-		return -ENODEV;
+	mutex_lock(&hw_lock);
 
-	if (hw->num_flows) {
-		hw->num_flows++;
-		return 0;
+	if (!hw || !hw->wed_dev) {
+		ret = -ENODEV;
+		goto out;
 	}
 
-	mutex_lock(&hw_lock);
-	if (!hw->wed_dev) {
-		ret = -ENODEV;
+	if (!hw->wed_dev->wlan.offload_enable)
+		goto out;
+
+	if (hw->num_flows) {
+		hw->num_flows++;
 		goto out;
 	}
 
@@ -1744,14 +2645,15 @@ void mtk_wed_flow_remove(int index)
 {
 	struct mtk_wed_hw *hw = hw_list[index];
 
-	if (!hw)
-		return;
+	mutex_lock(&hw_lock);
 
-	if (--hw->num_flows)
-		return;
+	if (!hw || !hw->wed_dev)
+		goto out;
 
-	mutex_lock(&hw_lock);
-	if (!hw->wed_dev)
+	if (!hw->wed_dev->wlan.offload_disable)
+		goto out;
+
+	if (--hw->num_flows)
 		goto out;
 
 	hw->wed_dev->wlan.offload_disable(hw->wed_dev);
@@ -1842,7 +2744,7 @@ mtk_wed_setup_tc(struct mtk_wed_device *wed, struct net_device *dev,
 {
 	struct mtk_wed_hw *hw = wed->hw;
 
-	if (hw->version < 2)
+	if (mtk_wed_is_v1(hw))
 		return -EOPNOTSUPP;
 
 	switch (type) {
@@ -1874,6 +2776,10 @@ void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 		.detach = mtk_wed_detach,
 		.ppe_check = mtk_wed_ppe_check,
 		.setup_tc = mtk_wed_setup_tc,
+		.start_hw_rro = mtk_wed_start_hw_rro,
+		.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,
 	};
 	struct device_node *eth_np = eth->dev->of_node;
 	struct platform_device *pdev;
@@ -1916,9 +2822,17 @@ void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 	hw->wdma = wdma;
 	hw->index = index;
 	hw->irq = irq;
-	hw->version = mtk_is_netsys_v1(eth) ? 1 : 2;
+	hw->version = eth->soc->version;
 
-	if (hw->version == 1) {
+	switch (hw->version) {
+	case 2:
+		hw->soc = &mt7986_data;
+		break;
+	case 3:
+		hw->soc = &mt7988_data;
+		break;
+	default:
+	case 1:
 		hw->mirror = syscon_regmap_lookup_by_phandle(eth_np,
 				"mediatek,pcie-mirror");
 		hw->hifsys = syscon_regmap_lookup_by_phandle(eth_np,
@@ -1932,6 +2846,8 @@ 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->soc = &mt7622_data;
+		break;
 	}
 
 	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 43ab77eaf6..c1f0479d7a 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed.h
+++ b/drivers/net/ethernet/mediatek/mtk_wed.h
@@ -9,10 +9,29 @@
 #include <linux/regmap.h>
 #include <linux/netdevice.h>
 
+#include "mtk_wed_regs.h"
+
 struct mtk_eth;
 struct mtk_wed_wo;
 
+struct mtk_wed_soc_data {
+	struct {
+		u32 tx_bm_tkid;
+		u32 wpdma_rx_ring0;
+		u32 reset_idx_tx_mask;
+		u32 reset_idx_rx_mask;
+	} regmap;
+	u32 tx_ring_desc_size;
+	u32 wdma_desc_size;
+};
+
+struct mtk_wed_amsdu {
+	void *txd;
+	dma_addr_t txd_phy;
+};
+
 struct mtk_wed_hw {
+	const struct mtk_wed_soc_data *soc;
 	struct device_node *node;
 	struct mtk_eth *eth;
 	struct regmap *regs;
@@ -24,6 +43,8 @@ struct mtk_wed_hw {
 	struct dentry *debugfs_dir;
 	struct mtk_wed_device *wed_dev;
 	struct mtk_wed_wo *wed_wo;
+	struct mtk_wed_amsdu *wed_amsdu;
+	u32 pcie_base;
 	u32 debugfs_reg;
 	u32 num_flows;
 	u8 version;
@@ -37,9 +58,30 @@ struct mtk_wdma_info {
 	u8 queue;
 	u16 wcid;
 	u8 bss;
+	u8 amsdu;
 };
 
 #ifdef CONFIG_NET_MEDIATEK_SOC_WED
+static inline bool mtk_wed_is_v1(struct mtk_wed_hw *hw)
+{
+	return hw->version == 1;
+}
+
+static inline bool mtk_wed_is_v2(struct mtk_wed_hw *hw)
+{
+	return hw->version == 2;
+}
+
+static inline bool mtk_wed_is_v3(struct mtk_wed_hw *hw)
+{
+	return hw->version == 3;
+}
+
+static inline bool mtk_wed_is_v3_or_greater(struct mtk_wed_hw *hw)
+{
+	return hw->version > 2;
+}
+
 static inline void
 wed_w32(struct mtk_wed_device *dev, u32 reg, u32 val)
 {
@@ -122,6 +164,21 @@ wpdma_txfree_w32(struct mtk_wed_device *dev, u32 reg, u32 val)
 	writel(val, dev->txfree_ring.wpdma + reg);
 }
 
+static inline u32 mtk_wed_get_pcie_base(struct mtk_wed_device *dev)
+{
+	if (!mtk_wed_is_v3_or_greater(dev->hw))
+		return MTK_WED_PCIE_BASE;
+
+	switch (dev->hw->index) {
+	case 1:
+		return MTK_WED_PCIE_BASE1;
+	case 2:
+		return MTK_WED_PCIE_BASE2;
+	default:
+		return MTK_WED_PCIE_BASE0;
+	}
+}
+
 void mtk_wed_add_hw(struct device_node *np, struct mtk_eth *eth,
 		    void __iomem *wdma, phys_addr_t wdma_phy,
 		    int index);
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c b/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c
index e24afeaea0..781c691473 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed_debugfs.c
@@ -11,6 +11,7 @@ struct reg_dump {
 	u16 offset;
 	u8 type;
 	u8 base;
+	u32 mask;
 };
 
 enum {
@@ -25,6 +26,8 @@ enum {
 
 #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 +35,7 @@ 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)
@@ -151,7 +155,7 @@ 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),
@@ -169,7 +173,7 @@ 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),
@@ -180,17 +184,6 @@ 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),
 		DUMP_WED_RING(WED_WDMA_RING_TX),
@@ -211,6 +204,287 @@ wed_rxinfo_show(struct seq_file *s, void *data)
 		DUMP_WED(WED_RX_BM_INTF),
 		DUMP_WED(WED_RX_BM_ERR_STS),
 	};
+	static const struct reg_dump regs_wed_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),
+	};
+	static const struct reg_dump regs_wed_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),
+	};
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+
+	if (dev) {
+		dump_wed_regs(s, dev, regs_common, ARRAY_SIZE(regs_common));
+		if (mtk_wed_is_v2(hw))
+			dump_wed_regs(s, dev,
+				      regs_wed_v2, ARRAY_SIZE(regs_wed_v2));
+		else
+			dump_wed_regs(s, dev,
+				      regs_wed_v3, ARRAY_SIZE(regs_wed_v3));
+	}
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(wed_rxinfo);
+
+static int
+wed_amsdu_show(struct seq_file *s, void *data)
+{
+	static const struct reg_dump regs[] = {
+		DUMP_STR("WED AMDSU INFO"),
+		DUMP_WED(WED_MON_AMSDU_FIFO_DMAD),
+
+		DUMP_STR("WED AMDSU ENG0 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(0)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(0)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(0)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(0)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(0)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(0),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(0),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(0),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(0),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(0),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG1 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(1)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(1)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(1)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(1)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(1)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(1),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(1),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(1),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(2),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(2),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG2 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(2)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(2)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(2)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(2)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(2)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(2),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(2),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(2),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(2),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(2),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG3 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(3)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(3)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(3)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(3)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(3)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(3),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(3),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(3),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(3),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(3),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG4 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(4)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(4)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(4)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(4)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(4)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(4),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(4),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(4),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(4),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(4),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG5 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(5)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(5)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(5)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(5)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(5)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(5),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(5),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(5),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(5),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(5),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG6 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(6)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(6)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(6)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(6)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(6)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(6),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(6),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(6),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(6),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(6),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG7 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(7)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(7)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(7)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(7)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(7)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(7),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(7),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(7),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(7),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(4),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED AMDSU ENG8 INFO"),
+		DUMP_WED(WED_MON_AMSDU_ENG_DMAD(8)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QFPL(8)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENI(8)),
+		DUMP_WED(WED_MON_AMSDU_ENG_QENO(8)),
+		DUMP_WED(WED_MON_AMSDU_ENG_MERG(8)),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(8),
+			      WED_AMSDU_ENG_MAX_PL_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT8(8),
+			      WED_AMSDU_ENG_MAX_QGPP_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(8),
+			      WED_AMSDU_ENG_CUR_ENTRY),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(8),
+			      WED_AMSDU_ENG_MAX_BUF_MERGED),
+		DUMP_WED_MASK(WED_MON_AMSDU_ENG_CNT9(8),
+			      WED_AMSDU_ENG_MAX_MSDU_MERGED),
+
+		DUMP_STR("WED QMEM INFO"),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(0), WED_AMSDU_QMEM_FQ_CNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(0), WED_AMSDU_QMEM_SP_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(1), WED_AMSDU_QMEM_TID0_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(1), WED_AMSDU_QMEM_TID1_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(2), WED_AMSDU_QMEM_TID2_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(2), WED_AMSDU_QMEM_TID3_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(3), WED_AMSDU_QMEM_TID4_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(3), WED_AMSDU_QMEM_TID5_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(4), WED_AMSDU_QMEM_TID6_QCNT),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_CNT(4), WED_AMSDU_QMEM_TID7_QCNT),
+
+		DUMP_STR("WED QMEM HEAD INFO"),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(0), WED_AMSDU_QMEM_FQ_HEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(0), WED_AMSDU_QMEM_SP_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(1), WED_AMSDU_QMEM_TID0_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(1), WED_AMSDU_QMEM_TID1_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(2), WED_AMSDU_QMEM_TID2_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(2), WED_AMSDU_QMEM_TID3_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(3), WED_AMSDU_QMEM_TID4_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(3), WED_AMSDU_QMEM_TID5_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(4), WED_AMSDU_QMEM_TID6_QHEAD),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(4), WED_AMSDU_QMEM_TID7_QHEAD),
+
+		DUMP_STR("WED QMEM TAIL INFO"),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(5), WED_AMSDU_QMEM_FQ_TAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(5), WED_AMSDU_QMEM_SP_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(6), WED_AMSDU_QMEM_TID0_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(6), WED_AMSDU_QMEM_TID1_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(7), WED_AMSDU_QMEM_TID2_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(7), WED_AMSDU_QMEM_TID3_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(8), WED_AMSDU_QMEM_TID4_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(8), WED_AMSDU_QMEM_TID5_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(9), WED_AMSDU_QMEM_TID6_QTAIL),
+		DUMP_WED_MASK(WED_MON_AMSDU_QMEM_PTR(9), WED_AMSDU_QMEM_TID7_QTAIL),
+
+		DUMP_STR("WED HIFTXD MSDU INFO"),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(1)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(2)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(3)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(4)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(5)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(6)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(7)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(8)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(9)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(10)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(11)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(12)),
+		DUMP_WED(WED_MON_AMSDU_HIFTXD_FETCH_MSDU(13)),
+	};
 	struct mtk_wed_hw *hw = s->private;
 	struct mtk_wed_device *dev = hw->wed_dev;
 
@@ -219,7 +493,94 @@ wed_rxinfo_show(struct seq_file *s, void *data)
 
 	return 0;
 }
-DEFINE_SHOW_ATTRIBUTE(wed_rxinfo);
+DEFINE_SHOW_ATTRIBUTE(wed_amsdu);
+
+static int
+wed_rtqm_show(struct seq_file *s, void *data)
+{
+	static const struct reg_dump regs[] = {
+		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),
+	};
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+
+	if (dev)
+		dump_wed_regs(s, dev, regs, ARRAY_SIZE(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[] = {
+		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),
+	};
+	struct mtk_wed_hw *hw = s->private;
+	struct mtk_wed_device *dev = hw->wed_dev;
+
+	if (dev)
+		dump_wed_regs(s, dev, regs, ARRAY_SIZE(regs));
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(wed_rro);
 
 static int
 mtk_wed_reg_set(void *data, u64 val)
@@ -261,7 +622,16 @@ 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);
-	if (hw->version != 1)
+	if (!mtk_wed_is_v1(hw)) {
 		debugfs_create_file_unsafe("rxinfo", 0400, dir, hw,
 					   &wed_rxinfo_fops);
+		if (mtk_wed_is_v3_or_greater(hw)) {
+			debugfs_create_file_unsafe("amsdu", 0400, dir, hw,
+						   &wed_amsdu_fops);
+			debugfs_create_file_unsafe("rtqm", 0400, dir, hw,
+						   &wed_rtqm_fops);
+			debugfs_create_file_unsafe("rro", 0400, dir, hw,
+						   &wed_rro_fops);
+		}
+	}
 }
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_mcu.c b/drivers/net/ethernet/mediatek/mtk_wed_mcu.c
index 071ed3dea8..ea0884186d 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_mcu.c
+++ b/drivers/net/ethernet/mediatek/mtk_wed_mcu.c
@@ -16,14 +16,30 @@
 #include "mtk_wed_wo.h"
 #include "mtk_wed.h"
 
-static u32 wo_r32(struct mtk_wed_wo *wo, u32 reg)
+static struct mtk_wed_wo_memory_region mem_region[] = {
+	[MTK_WED_WO_REGION_EMI] = {
+		.name = "wo-emi",
+	},
+	[MTK_WED_WO_REGION_ILM] = {
+		.name = "wo-ilm",
+	},
+	[MTK_WED_WO_REGION_DATA] = {
+		.name = "wo-data",
+		.shared = true,
+	},
+	[MTK_WED_WO_REGION_BOOT] = {
+		.name = "wo-boot",
+	},
+};
+
+static u32 wo_r32(u32 reg)
 {
-	return readl(wo->boot.addr + reg);
+	return readl(mem_region[MTK_WED_WO_REGION_BOOT].addr + reg);
 }
 
-static void wo_w32(struct mtk_wed_wo *wo, u32 reg, u32 val)
+static void wo_w32(u32 reg, u32 val)
 {
-	writel(val, wo->boot.addr + reg);
+	writel(val, mem_region[MTK_WED_WO_REGION_BOOT].addr + reg);
 }
 
 static struct sk_buff *
@@ -68,6 +84,9 @@ mtk_wed_update_rx_stats(struct mtk_wed_device *wed, struct sk_buff *skb)
 	struct mtk_wed_wo_rx_stats *stats;
 	int i;
 
+	if (!wed->wlan.update_wo_rx_stats)
+		return;
+
 	if (count * sizeof(*stats) > skb->len - sizeof(u32))
 		return;
 
@@ -204,7 +223,7 @@ int mtk_wed_mcu_msg_update(struct mtk_wed_device *dev, int id, void *data,
 {
 	struct mtk_wed_wo *wo = dev->hw->wed_wo;
 
-	if (dev->hw->version == 1)
+	if (!mtk_wed_get_rx_capa(dev))
 		return 0;
 
 	if (WARN_ON(!wo))
@@ -215,19 +234,13 @@ int mtk_wed_mcu_msg_update(struct mtk_wed_device *dev, int id, void *data,
 }
 
 static int
-mtk_wed_get_memory_region(struct mtk_wed_wo *wo,
+mtk_wed_get_memory_region(struct mtk_wed_hw *hw, int index,
 			  struct mtk_wed_wo_memory_region *region)
 {
 	struct reserved_mem *rmem;
 	struct device_node *np;
-	int index;
 
-	index = of_property_match_string(wo->hw->node, "memory-region-names",
-					 region->name);
-	if (index < 0)
-		return index;
-
-	np = of_parse_phandle(wo->hw->node, "memory-region", index);
+	np = of_parse_phandle(hw->node, "memory-region", index);
 	if (!np)
 		return -ENODEV;
 
@@ -239,14 +252,13 @@ mtk_wed_get_memory_region(struct mtk_wed_wo *wo,
 
 	region->phy_addr = rmem->base;
 	region->size = rmem->size;
-	region->addr = devm_ioremap(wo->hw->dev, region->phy_addr, region->size);
+	region->addr = devm_ioremap(hw->dev, region->phy_addr, region->size);
 
 	return !region->addr ? -EINVAL : 0;
 }
 
 static int
-mtk_wed_mcu_run_firmware(struct mtk_wed_wo *wo, const struct firmware *fw,
-			 struct mtk_wed_wo_memory_region *region)
+mtk_wed_mcu_run_firmware(struct mtk_wed_wo *wo, const struct firmware *fw)
 {
 	const u8 *first_region_ptr, *region_ptr, *trailer_ptr, *ptr = fw->data;
 	const struct mtk_wed_fw_trailer *trailer;
@@ -259,50 +271,46 @@ mtk_wed_mcu_run_firmware(struct mtk_wed_wo *wo, const struct firmware *fw,
 
 	while (region_ptr < trailer_ptr) {
 		u32 length;
+		int i;
 
 		fw_region = (const struct mtk_wed_fw_region *)region_ptr;
 		length = le32_to_cpu(fw_region->len);
-
-		if (region->phy_addr != le32_to_cpu(fw_region->addr))
+		if (first_region_ptr < ptr + length)
 			goto next;
 
-		if (region->size < length)
-			goto next;
+		for (i = 0; i < ARRAY_SIZE(mem_region); i++) {
+			struct mtk_wed_wo_memory_region *region;
 
-		if (first_region_ptr < ptr + length)
-			goto next;
+			region = &mem_region[i];
+			if (region->phy_addr != le32_to_cpu(fw_region->addr))
+				continue;
+
+			if (region->size < length)
+				continue;
 
-		if (region->shared && region->consumed)
-			return 0;
+			if (region->shared && region->consumed)
+				break;
 
-		if (!region->shared || !region->consumed) {
-			memcpy_toio(region->addr, ptr, length);
-			region->consumed = true;
-			return 0;
+			if (!region->shared || !region->consumed) {
+				memcpy_toio(region->addr, ptr, length);
+				region->consumed = true;
+				break;
+			}
 		}
+
+		if (i == ARRAY_SIZE(mem_region))
+			return -EINVAL;
 next:
 		region_ptr += sizeof(*fw_region);
 		ptr += length;
 	}
 
-	return -EINVAL;
+	return 0;
 }
 
 static int
 mtk_wed_mcu_load_firmware(struct mtk_wed_wo *wo)
 {
-	static struct mtk_wed_wo_memory_region mem_region[] = {
-		[MTK_WED_WO_REGION_EMI] = {
-			.name = "wo-emi",
-		},
-		[MTK_WED_WO_REGION_ILM] = {
-			.name = "wo-ilm",
-		},
-		[MTK_WED_WO_REGION_DATA] = {
-			.name = "wo-data",
-			.shared = true,
-		},
-	};
 	const struct mtk_wed_fw_trailer *trailer;
 	const struct firmware *fw;
 	const char *fw_name;
@@ -311,25 +319,38 @@ mtk_wed_mcu_load_firmware(struct mtk_wed_wo *wo)
 
 	/* load firmware region metadata */
 	for (i = 0; i < ARRAY_SIZE(mem_region); i++) {
-		ret = mtk_wed_get_memory_region(wo, &mem_region[i]);
+		int index = of_property_match_string(wo->hw->node,
+						     "memory-region-names",
+						     mem_region[i].name);
+		if (index < 0)
+			continue;
+
+		ret = mtk_wed_get_memory_region(wo->hw, index, &mem_region[i]);
 		if (ret)
 			return ret;
 	}
 
-	wo->boot.name = "wo-boot";
-	ret = mtk_wed_get_memory_region(wo, &wo->boot);
-	if (ret)
-		return ret;
-
 	/* set dummy cr */
 	wed_w32(wo->hw->wed_dev, MTK_WED_SCR0 + 4 * MTK_WED_DUMMY_CR_FWDL,
 		wo->hw->index + 1);
 
 	/* load firmware */
-	if (of_device_is_compatible(wo->hw->node, "mediatek,mt7981-wed"))
-		fw_name = MT7981_FIRMWARE_WO;
-	else
-		fw_name = wo->hw->index ? MT7986_FIRMWARE_WO1 : MT7986_FIRMWARE_WO0;
+	switch (wo->hw->version) {
+	case 2:
+		if (of_device_is_compatible(wo->hw->node,
+					    "mediatek,mt7981-wed"))
+			fw_name = MT7981_FIRMWARE_WO;
+		else
+			fw_name = wo->hw->index ? MT7986_FIRMWARE_WO1
+						: MT7986_FIRMWARE_WO0;
+		break;
+	case 3:
+		fw_name = wo->hw->index ? MT7988_FIRMWARE_WO1
+					: MT7988_FIRMWARE_WO0;
+		break;
+	default:
+		return -EINVAL;
+	}
 
 	ret = request_firmware(&fw, fw_name, wo->hw->dev);
 	if (ret)
@@ -343,23 +364,22 @@ mtk_wed_mcu_load_firmware(struct mtk_wed_wo *wo)
 	dev_info(wo->hw->dev, "MTK WED WO Chip ID %02x Region %d\n",
 		 trailer->chip_id, trailer->num_region);
 
-	for (i = 0; i < ARRAY_SIZE(mem_region); i++) {
-		ret = mtk_wed_mcu_run_firmware(wo, fw, &mem_region[i]);
-		if (ret)
-			goto out;
-	}
+	ret = mtk_wed_mcu_run_firmware(wo, fw);
+	if (ret)
+		goto out;
 
 	/* set the start address */
-	boot_cr = wo->hw->index ? MTK_WO_MCU_CFG_LS_WA_BOOT_ADDR_ADDR
-				: MTK_WO_MCU_CFG_LS_WM_BOOT_ADDR_ADDR;
-	wo_w32(wo, boot_cr, mem_region[MTK_WED_WO_REGION_EMI].phy_addr >> 16);
+	if (!mtk_wed_is_v3_or_greater(wo->hw) && wo->hw->index)
+		boot_cr = MTK_WO_MCU_CFG_LS_WA_BOOT_ADDR_ADDR;
+	else
+		boot_cr = MTK_WO_MCU_CFG_LS_WM_BOOT_ADDR_ADDR;
+	wo_w32(boot_cr, mem_region[MTK_WED_WO_REGION_EMI].phy_addr >> 16);
 	/* wo firmware reset */
-	wo_w32(wo, MTK_WO_MCU_CFG_LS_WF_MCCR_CLR_ADDR, 0xc00);
+	wo_w32(MTK_WO_MCU_CFG_LS_WF_MCCR_CLR_ADDR, 0xc00);
 
-	val = wo_r32(wo, MTK_WO_MCU_CFG_LS_WF_MCU_CFG_WM_WA_ADDR);
-	val |= wo->hw->index ? MTK_WO_MCU_CFG_LS_WF_WM_WA_WA_CPU_RSTB_MASK
-			     : MTK_WO_MCU_CFG_LS_WF_WM_WA_WM_CPU_RSTB_MASK;
-	wo_w32(wo, MTK_WO_MCU_CFG_LS_WF_MCU_CFG_WM_WA_ADDR, val);
+	val = wo_r32(MTK_WO_MCU_CFG_LS_WF_MCU_CFG_WM_WA_ADDR) |
+	      MTK_WO_MCU_CFG_LS_WF_WM_WA_WM_CPU_RSTB_MASK;
+	wo_w32(MTK_WO_MCU_CFG_LS_WF_MCU_CFG_WM_WA_ADDR, val);
 out:
 	release_firmware(fw);
 
@@ -393,3 +413,5 @@ int mtk_wed_mcu_init(struct mtk_wed_wo *wo)
 MODULE_FIRMWARE(MT7981_FIRMWARE_WO);
 MODULE_FIRMWARE(MT7986_FIRMWARE_WO0);
 MODULE_FIRMWARE(MT7986_FIRMWARE_WO1);
+MODULE_FIRMWARE(MT7988_FIRMWARE_WO0);
+MODULE_FIRMWARE(MT7988_FIRMWARE_WO1);
diff --git a/drivers/net/ethernet/mediatek/mtk_wed_regs.h b/drivers/net/ethernet/mediatek/mtk_wed_regs.h
index f87ab9b8a5..c711909248 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_regs.h
+++ b/drivers/net/ethernet/mediatek/mtk_wed_regs.h
@@ -13,6 +13,9 @@
 #define MTK_WDMA_DESC_CTRL_LAST_SEG0		BIT(30)
 #define MTK_WDMA_DESC_CTRL_DMA_DONE		BIT(31)
 
+#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;
@@ -25,6 +28,8 @@ struct mtk_wdma_desc {
 #define MTK_WED_RESET					0x008
 #define MTK_WED_RESET_TX_BM				BIT(0)
 #define MTK_WED_RESET_RX_BM				BIT(1)
+#define MTK_WED_RESET_RX_PG_BM				BIT(2)
+#define MTK_WED_RESET_RRO_RX_TO_PG			BIT(3)
 #define MTK_WED_RESET_TX_FREE_AGENT			BIT(4)
 #define MTK_WED_RESET_WPDMA_TX_DRV			BIT(8)
 #define MTK_WED_RESET_WPDMA_RX_DRV			BIT(9)
@@ -37,6 +42,7 @@ 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_AMSDU				BIT(22)
 #define MTK_WED_RESET_WED				BIT(31)
 
 #define MTK_WED_CTRL					0x00c
@@ -44,6 +50,9 @@ 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_BUSY			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)
@@ -54,9 +63,14 @@ 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_AMSDU_EN			BIT(22)
+#define MTK_WED_CTRL_TX_AMSDU_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)
@@ -89,19 +103,26 @@ struct mtk_wdma_desc {
 #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)
 
+#define MTK_WED_WPDMA_STATUS				0x068
+#define MTK_WED_WPDMA_STATUS_TX_DRV			GENMASK(15, 8)
+
 #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_TKID				0x088
-#define MTK_WED_TX_BM_TKID_V2				0x0c8
 #define MTK_WED_TX_BM_TKID_START			GENMASK(15, 0)
 #define MTK_WED_TX_BM_TKID_END				GENMASK(31, 16)
 
@@ -124,6 +145,12 @@ struct mtk_wdma_desc {
 #define MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM		GENMASK(22, 16)
 #define MTK_WED_TX_TKID_CTRL_PAUSE			BIT(28)
 
+#define MTK_WED_TX_TKID_INTF				0x0dc
+#define MTK_WED_TX_TKID_INTF_TKFIFO_FDEP		GENMASK(25, 16)
+
+#define MTK_WED_TX_TKID_CTRL_VLD_GRP_NUM_V3		GENMASK(7, 0)
+#define MTK_WED_TX_TKID_CTRL_RSV_GRP_NUM_V3		GENMASK(23, 16)
+
 #define MTK_WED_TX_TKID_DYN_THR				0x0e0
 #define MTK_WED_TX_TKID_DYN_THR_LO			GENMASK(6, 0)
 #define MTK_WED_TX_TKID_DYN_THR_HI			GENMASK(22, 16)
@@ -160,9 +187,6 @@ struct mtk_wdma_desc {
 #define MTK_WED_GLO_CFG_RX_2B_OFFSET			BIT(31)
 
 #define MTK_WED_RESET_IDX				0x20c
-#define MTK_WED_RESET_IDX_TX				GENMASK(3, 0)
-#define MTK_WED_RESET_IDX_RX				GENMASK(17, 16)
-#define MTK_WED_RESET_IDX_RX_V2				GENMASK(7, 6)
 #define MTK_WED_RESET_WPDMA_IDX_RX			GENMASK(31, 30)
 
 #define MTK_WED_TX_MIB(_n)				(0x2a0 + (_n) * 4)
@@ -174,6 +198,7 @@ struct mtk_wdma_desc {
 #define MTK_WED_RING_RX_DATA(_n)			(0x420 + (_n) * 0x10)
 
 #define MTK_WED_SCR0					0x3c0
+#define MTK_WED_RX1_CTRL2				0x418
 #define MTK_WED_WPDMA_INT_TRIGGER			0x504
 #define MTK_WED_WPDMA_INT_TRIGGER_RX_DONE		BIT(1)
 #define MTK_WED_WPDMA_INT_TRIGGER_TX_DONE		GENMASK(5, 4)
@@ -204,12 +229,15 @@ 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)
 
 #define MTK_WED_WPDMA_RESET_IDX				0x50c
 #define MTK_WED_WPDMA_RESET_IDX_TX			GENMASK(3, 0)
@@ -255,9 +283,10 @@ 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
@@ -283,15 +312,30 @@ struct mtk_wdma_desc {
 
 #define MTK_WED_WPDMA_RX_D_RST_IDX			0x760
 #define MTK_WED_WPDMA_RX_D_RST_CRX_IDX			GENMASK(17, 16)
+#define MTK_WED_WPDMA_RX_D_RST_DRV_IDX_ALL		BIT(20)
 #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
 
 #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_BUSY			BIT(1)
+#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
@@ -299,6 +343,20 @@ struct mtk_wdma_desc {
 #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_BUSY			BIT(1)
+#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_DDONE2_BUSY		BIT(27)
+
+#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)
@@ -322,6 +380,7 @@ struct mtk_wdma_desc {
 
 #define MTK_WED_WDMA_RESET_IDX				0xa08
 #define MTK_WED_WDMA_RESET_IDX_RX			GENMASK(17, 16)
+#define MTK_WED_WDMA_RESET_IDX_RX_ALL			BIT(20)
 #define MTK_WED_WDMA_RESET_IDX_DRV			GENMASK(25, 24)
 
 #define MTK_WED_WDMA_INT_CLR				0xa24
@@ -331,6 +390,7 @@ 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_CTRL_POLL_SRC_SEL		GENMASK(17, 16)
 
 #define MTK_WED_WDMA_CFG_BASE				0xaa0
@@ -391,9 +451,62 @@ struct mtk_wdma_desc {
 #define MTK_WDMA_INT_MASK_RX_DELAY			BIT(30)
 #define MTK_WDMA_INT_MASK_RX_COHERENT			BIT(31)
 
+#define MTK_WDMA_XDMA_TX_FIFO_CFG			0x238
+#define MTK_WDMA_XDMA_TX_FIFO_CFG_TX_PAR_FIFO_CLEAR	BIT(0)
+#define MTK_WDMA_XDMA_TX_FIFO_CFG_TX_CMD_FIFO_CLEAR	BIT(4)
+#define MTK_WDMA_XDMA_TX_FIFO_CFG_TX_DMAD_FIFO_CLEAR	BIT(8)
+#define MTK_WDMA_XDMA_TX_FIFO_CFG_TX_ARR_FIFO_CLEAR	BIT(12)
+
+#define MTK_WDMA_XDMA_RX_FIFO_CFG			0x23c
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_PAR_FIFO_CLEAR	BIT(0)
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_CMD_FIFO_CLEAR	BIT(4)
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_DMAD_FIFO_CLEAR	BIT(8)
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_ARR_FIFO_CLEAR	BIT(12)
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_LEN_FIFO_CLEAR	BIT(15)
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_WID_FIFO_CLEAR	BIT(18)
+#define MTK_WDMA_XDMA_RX_FIFO_CFG_RX_BID_FIFO_CLEAR	BIT(21)
+
 #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_TX_CFG_PREF_BUSY			BIT(1)
+
+#define MTK_WDMA_PREF_RX_CFG				0x2dc
+#define MTK_WDMA_PREF_RX_CFG_PREF_EN			BIT(0)
+#define MTK_WDMA_PREF_RX_CFG_PREF_BUSY			BIT(1)
+
+#define MTK_WDMA_PREF_RX_FIFO_CFG			0x2e0
+#define MTK_WDMA_PREF_RX_FIFO_CFG_RING0_CLEAR		BIT(0)
+#define MTK_WDMA_PREF_RX_FIFO_CFG_RING1_CLEAR		BIT(16)
+
+#define MTK_WDMA_PREF_TX_FIFO_CFG			0x2d4
+#define MTK_WDMA_PREF_TX_FIFO_CFG_RING0_CLEAR		BIT(0)
+#define MTK_WDMA_PREF_TX_FIFO_CFG_RING1_CLEAR		BIT(16)
+
+#define MTK_WDMA_PREF_SIDX_CFG				0x2e4
+#define MTK_WDMA_PREF_SIDX_CFG_TX_RING_CLEAR		GENMASK(3, 0)
+#define MTK_WDMA_PREF_SIDX_CFG_RX_RING_CLEAR		GENMASK(5, 4)
+
+#define MTK_WDMA_WRBK_TX_CFG				0x300
+#define MTK_WDMA_WRBK_TX_CFG_WRBK_BUSY			BIT(0)
+#define MTK_WDMA_WRBK_TX_CFG_WRBK_EN			BIT(30)
+
+#define MTK_WDMA_WRBK_TX_FIFO_CFG(_n)			(0x304 + (_n) * 0x4)
+#define MTK_WDMA_WRBK_TX_FIFO_CFG_RING_CLEAR		BIT(0)
+
+#define MTK_WDMA_WRBK_RX_CFG				0x344
+#define MTK_WDMA_WRBK_RX_CFG_WRBK_BUSY			BIT(0)
+#define MTK_WDMA_WRBK_RX_CFG_WRBK_EN			BIT(30)
+
+#define MTK_WDMA_WRBK_RX_FIFO_CFG(_n)			(0x348 + (_n) * 0x4)
+#define MTK_WDMA_WRBK_RX_FIFO_CFG_RING_CLEAR		BIT(0)
+
+#define MTK_WDMA_WRBK_SIDX_CFG				0x388
+#define MTK_WDMA_WRBK_SIDX_CFG_TX_RING_CLEAR		GENMASK(3, 0)
+#define MTK_WDMA_WRBK_SIDX_CFG_RX_RING_CLEAR		GENMASK(5, 4)
+
 #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)
@@ -407,6 +520,32 @@ 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_RST				0xb04
+
+#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
@@ -415,6 +554,24 @@ 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)
@@ -464,7 +621,195 @@ 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_RRO_RX_HW_STS				0xf00
+#define MTK_WED_RX_IND_CMD_BUSY				GENMASK(31, 0)
+
+#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_AMSDU_FIFO				0x1800
+#define MTK_WED_AMSDU_IS_PRIOR0_RING			BIT(10)
+
+#define MTK_WED_AMSDU_STA_INFO				0x01810
+#define MTK_WED_AMSDU_STA_INFO_DO_INIT			BIT(0)
+#define MTK_WED_AMSDU_STA_INFO_SET_INIT			BIT(1)
+
+#define MTK_WED_AMSDU_STA_INFO_INIT			0x01814
+#define MTK_WED_AMSDU_STA_WTBL_HDRT_MODE		BIT(0)
+#define MTK_WED_AMSDU_STA_RMVL				BIT(1)
+#define MTK_WED_AMSDU_STA_MAX_AMSDU_LEN			GENMASK(7, 2)
+#define MTK_WED_AMSDU_STA_MAX_AMSDU_NUM			GENMASK(11, 8)
+
+#define MTK_WED_AMSDU_HIFTXD_BASE_L(_n)			(0x1980 + (_n) * 0x4)
+
+#define MTK_WED_AMSDU_PSE				0x1910
+#define MTK_WED_AMSDU_PSE_RESET				BIT(16)
+
+#define MTK_WED_AMSDU_HIFTXD_CFG			0x1968
+#define MTK_WED_AMSDU_HIFTXD_SRC			GENMASK(16, 15)
+
+#define MTK_WED_MON_AMSDU_FIFO_DMAD			0x1a34
+
+#define MTK_WED_MON_AMSDU_ENG_DMAD(_n)			(0x1a80 + (_n) * 0x50)
+#define MTK_WED_MON_AMSDU_ENG_QFPL(_n)			(0x1a84 + (_n) * 0x50)
+#define MTK_WED_MON_AMSDU_ENG_QENI(_n)			(0x1a88 + (_n) * 0x50)
+#define MTK_WED_MON_AMSDU_ENG_QENO(_n)			(0x1a8c + (_n) * 0x50)
+#define MTK_WED_MON_AMSDU_ENG_MERG(_n)			(0x1a90 + (_n) * 0x50)
+
+#define MTK_WED_MON_AMSDU_ENG_CNT8(_n)			(0x1a94 + (_n) * 0x50)
+#define MTK_WED_AMSDU_ENG_MAX_QGPP_CNT			GENMASK(10, 0)
+#define MTK_WED_AMSDU_ENG_MAX_PL_CNT			GENMASK(27, 16)
+
+#define MTK_WED_MON_AMSDU_ENG_CNT9(_n)			(0x1a98 + (_n) * 0x50)
+#define MTK_WED_AMSDU_ENG_CUR_ENTRY			GENMASK(10, 0)
+#define MTK_WED_AMSDU_ENG_MAX_BUF_MERGED		GENMASK(20, 16)
+#define MTK_WED_AMSDU_ENG_MAX_MSDU_MERGED		GENMASK(28, 24)
+
+#define MTK_WED_MON_AMSDU_QMEM_STS1			0x1e04
+
+#define MTK_WED_MON_AMSDU_QMEM_CNT(_n)			(0x1e0c + (_n) * 0x4)
+#define MTK_WED_AMSDU_QMEM_FQ_CNT			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_SP_QCNT			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID0_QCNT			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID1_QCNT			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID2_QCNT			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID3_QCNT			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID4_QCNT			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID5_QCNT			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID6_QCNT			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID7_QCNT			GENMASK(11, 0)
+
+#define MTK_WED_MON_AMSDU_QMEM_PTR(_n)			(0x1e20 + (_n) * 0x4)
+#define MTK_WED_AMSDU_QMEM_FQ_HEAD			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_SP_QHEAD			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID0_QHEAD			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID1_QHEAD			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID2_QHEAD			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID3_QHEAD			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID4_QHEAD			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID5_QHEAD			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID6_QHEAD			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID7_QHEAD			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_FQ_TAIL			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_SP_QTAIL			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID0_QTAIL			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID1_QTAIL			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID2_QTAIL			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID3_QTAIL			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID4_QTAIL			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID5_QTAIL			GENMASK(11, 0)
+#define MTK_WED_AMSDU_QMEM_TID6_QTAIL			GENMASK(27, 16)
+#define MTK_WED_AMSDU_QMEM_TID7_QTAIL			GENMASK(11, 0)
+
+#define MTK_WED_MON_AMSDU_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/drivers/net/ethernet/mediatek/mtk_wed_wo.h b/drivers/net/ethernet/mediatek/mtk_wed_wo.h
index 7a1a2a28f1..87a67fa386 100644
--- a/drivers/net/ethernet/mediatek/mtk_wed_wo.h
+++ b/drivers/net/ethernet/mediatek/mtk_wed_wo.h
@@ -91,6 +91,8 @@ enum mtk_wed_dummy_cr_idx {
 #define MT7981_FIRMWARE_WO	"mediatek/mt7981_wo.bin"
 #define MT7986_FIRMWARE_WO0	"mediatek/mt7986_wo_0.bin"
 #define MT7986_FIRMWARE_WO1	"mediatek/mt7986_wo_1.bin"
+#define MT7988_FIRMWARE_WO0	"mediatek/mt7988_wo_0.bin"
+#define MT7988_FIRMWARE_WO1	"mediatek/mt7988_wo_1.bin"
 
 #define MTK_WO_MCU_CFG_LS_BASE				0
 #define MTK_WO_MCU_CFG_LS_HW_VER_ADDR			(MTK_WO_MCU_CFG_LS_BASE + 0x000)
@@ -228,7 +230,6 @@ struct mtk_wed_wo_queue {
 
 struct mtk_wed_wo {
 	struct mtk_wed_hw *hw;
-	struct mtk_wed_wo_memory_region boot;
 
 	struct mtk_wed_wo_queue q_tx;
 	struct mtk_wed_wo_queue q_rx;