diff --git a/dma.c b/dma.c
index 0db35fdb..f8ca17fa 100644
--- a/dma.c
+++ b/dma.c
@@ -236,17 +236,6 @@ mt76_dma_rx_cleanup(struct mt76_dev *dev, struct mt76_queue *q)
 	spin_unlock_bh(&q->lock);
 }
 
-static int
-mt76_dma_init(struct mt76_dev *dev)
-{
-	int i;
-
-	for (i = 0; i < ARRAY_SIZE(dev->q_rx); i++)
-		mt76_dma_rx_fill(dev, &dev->q_rx[i], false);
-
-	return 0;
-}
-
 static void
 mt76_add_fragment(struct mt76_dev *dev, struct mt76_queue *q, void *data,
 		    int len, bool more)
@@ -272,7 +261,6 @@ mt76_dma_rx_process(struct mt76_dev *dev, struct mt76_queue *q, int budget)
 	unsigned char *data;
 	int len;
 	int done = 0;
-	bool napi = q == &dev->q_rx[MT_RXQ_MAIN];
 	bool more;
 
 	while (done < budget) {
@@ -315,17 +303,50 @@ mt76_dma_rx_process(struct mt76_dev *dev, struct mt76_queue *q, int budget)
 		dev->drv->rx_skb(dev, q - dev->q_rx, skb);
 	}
 
-	mt76_dma_rx_fill(dev, q, napi);
+	mt76_dma_rx_fill(dev, q, true);
 	return done;
 }
 
+static int
+mt76_dma_rx_poll(struct napi_struct *napi, int budget)
+{
+	struct mt76_dev *dev;
+	int qid, done;
+
+	dev = container_of(napi->dev, struct mt76_dev, napi_dev);
+	qid = napi - dev->napi;
+
+	done = mt76_dma_rx_process(dev, &dev->q_rx[qid], budget);
+	if (done < budget) {
+		napi_complete(napi);
+		dev->drv->rx_poll_complete(dev, qid);
+	}
+
+	return done;
+}
+
+static int
+mt76_dma_init(struct mt76_dev *dev)
+{
+	int i;
+
+	init_dummy_netdev(&dev->napi_dev);
+
+	for (i = 0; i < ARRAY_SIZE(dev->q_rx); i++) {
+		netif_napi_add(&dev->napi_dev, &dev->napi[i], mt76_dma_rx_poll, 64);
+		mt76_dma_rx_fill(dev, &dev->q_rx[i], false);
+		napi_enable(&dev->napi[i]);
+	}
+
+	return 0;
+}
+
 static const struct mt76_queue_ops mt76_dma_ops = {
 	.init = mt76_dma_init,
 	.alloc = mt76_dma_alloc_queue,
 	.add_buf = mt76_dma_add_buf,
 	.tx_cleanup = mt76_dma_tx_cleanup,
 	.kick = mt76_dma_kick_queue,
-	.rx_process = mt76_dma_rx_process,
 };
 
 int mt76_dma_attach(struct mt76_dev *dev)
@@ -352,7 +373,10 @@ void mt76_dma_cleanup(struct mt76_dev *dev)
 	for (i = 0; i < ARRAY_SIZE(dev->q_tx); i++)
 		mt76_dma_tx_cleanup(dev, &dev->q_tx[i], true,
 				    mt76_tx_cleanup_entry);
-	for (i = 0; i < ARRAY_SIZE(dev->q_rx); i++)
+
+	for (i = 0; i < ARRAY_SIZE(dev->q_rx); i++) {
+		netif_napi_del(&dev->napi[i]);
 		mt76_dma_rx_cleanup(dev, &dev->q_rx[i]);
+	}
 }
 EXPORT_SYMBOL_GPL(mt76_dma_cleanup);
diff --git a/mt76.h b/mt76.h
index f2fae9c2..38184ff1 100644
--- a/mt76.h
+++ b/mt76.h
@@ -105,8 +105,6 @@ struct mt76_queue_ops {
 			   void (*done)(struct mt76_dev *dev, struct mt76_queue *q,
 					struct mt76_queue_entry *e));
 
-	int (*rx_process)(struct mt76_dev *dev, struct mt76_queue *q, int budget);
-
 	void (*kick)(struct mt76_dev *dev, struct mt76_queue *q);
 };
 
@@ -159,6 +157,8 @@ struct mt76_driver_ops {
 
 	void (*rx_skb)(struct mt76_dev *dev, enum mt76_rxq_id q,
 		       struct sk_buff *skb);
+
+	void (*rx_poll_complete)(struct mt76_dev *dev, enum mt76_rxq_id q);
 };
 
 struct mt76_dev {
@@ -170,6 +170,9 @@ struct mt76_dev {
 	void __iomem *regs;
 	struct device *dev;
 
+	struct net_device napi_dev;
+	struct napi_struct napi[__MT_RXQ_MAX];
+
 	struct list_head txwi_cache;
 	struct mt76_queue q_tx[__MT_TXQ_MAX];
 	struct mt76_queue q_rx[__MT_RXQ_MAX];
@@ -234,7 +237,6 @@ static inline u16 mt76_rev(struct mt76_dev *dev)
 #define mt76_queue_alloc(dev, ...)	(dev)->mt76.queue_ops->alloc(&((dev)->mt76), __VA_ARGS__)
 #define mt76_queue_add_buf(dev, ...)	(dev)->mt76.queue_ops->add_buf(&((dev)->mt76), __VA_ARGS__)
 #define mt76_queue_tx_cleanup(dev, ...)	(dev)->mt76.queue_ops->tx_cleanup(&((dev)->mt76), __VA_ARGS__)
-#define mt76_queue_rx_process(dev, ...)	(dev)->mt76.queue_ops->rx_process(&((dev)->mt76), __VA_ARGS__)
 #define mt76_queue_kick(dev, ...)	(dev)->mt76.queue_ops->kick(&((dev)->mt76), __VA_ARGS__)
 
 int mt76_register_device(struct mt76_dev *dev, bool vht,
diff --git a/mt7603.h b/mt7603.h
index dc16f133..d2ff8ee3 100644
--- a/mt7603.h
+++ b/mt7603.h
@@ -84,11 +84,7 @@ struct mt7603_dev {
 	struct mt7603_mcu mcu;
 	struct mt76_queue q_rx;
 
-	struct net_device napi_dev;
-	struct napi_struct napi;
-
 	struct tasklet_struct tx_tasklet;
-	struct tasklet_struct rx_tasklet;
 	struct tasklet_struct pre_tbtt_tasklet;
 };
 
@@ -145,5 +141,6 @@ int mt7603_mac_write_txwi(struct mt76_dev *mdev, void *txwi_ptr,
 
 void mt7603_queue_rx_skb(struct mt76_dev *dev, enum mt76_rxq_id q,
 			 struct sk_buff *skb);
+void mt7603_rx_poll_complete(struct mt76_dev *mdev, enum mt76_rxq_id q);
 
 #endif
diff --git a/mt7603_core.c b/mt7603_core.c
index 76e8093e..e0257c44 100644
--- a/mt7603_core.c
+++ b/mt7603_core.c
@@ -25,6 +25,12 @@ void mt7603_set_irq_mask(struct mt7603_dev *dev, u32 clear, u32 set)
 	spin_unlock_irqrestore(&dev->irq_lock, flags);
 }
 
+void mt7603_rx_poll_complete(struct mt76_dev *mdev, enum mt76_rxq_id q)
+{
+	struct mt7603_dev *dev = container_of(mdev, struct mt7603_dev, mt76);
+	mt7603_irq_enable(dev, MT_INT_RX_DONE(q));
+}
+
 irqreturn_t mt7603_irq_handler(int irq, void *dev_instance)
 {
 	struct mt7603_dev *dev = dev_instance;
@@ -45,12 +51,12 @@ irqreturn_t mt7603_irq_handler(int irq, void *dev_instance)
 
 	if (intr & MT_INT_RX_DONE(0)) {
 		mt7603_irq_disable(dev, MT_INT_RX_DONE(0));
-		napi_schedule(&dev->napi);
+		napi_schedule(&dev->mt76.napi[0]);
 	}
 
 	if (intr & MT_INT_RX_DONE(1)) {
 		mt7603_irq_disable(dev, MT_INT_RX_DONE(1));
-		tasklet_schedule(&dev->rx_tasklet);
+		napi_schedule(&dev->mt76.napi[1]);
 	}
 
 #if 0
diff --git a/mt7603_dma.c b/mt7603_dma.c
index 2b8da2b3..e6f7a51e 100644
--- a/mt7603_dma.c
+++ b/mt7603_dma.c
@@ -155,32 +155,6 @@ mt7603_tx_tasklet(unsigned long data)
 	mt7603_irq_enable(dev, MT_INT_TX_DONE_ALL);
 }
 
-static int
-mt7603_dma_rx_poll(struct napi_struct *napi, int budget)
-{
-	struct mt7603_dev *dev = container_of(napi, struct mt7603_dev, napi);
-	int done;
-
-	done = mt76_queue_rx_process(dev, &dev->mt76.q_rx[MT_RXQ_MAIN], budget);
-
-	if (done < budget) {
-		napi_complete(napi);
-		mt7603_irq_enable(dev, MT_INT_RX_DONE(0));
-	}
-
-	return done;
-}
-
-static void
-mt7603_rx_tasklet(unsigned long data)
-{
-	struct mt7603_dev *dev = (struct mt7603_dev *) data;
-	struct mt76_queue *q = &dev->mt76.q_rx[MT_RXQ_MCU];
-
-	mt76_queue_rx_process(dev, q, q->ndesc);
-	mt7603_irq_enable(dev, MT_INT_RX_DONE(1));
-}
-
 int mt7603_dma_init(struct mt7603_dev *dev)
 {
 	static const u8 wmm_queue_map[] = {
@@ -197,12 +171,7 @@ int mt7603_dma_init(struct mt7603_dev *dev)
 	init_waitqueue_head(&dev->mcu.wait);
 	skb_queue_head_init(&dev->mcu.res_q);
 
-	init_dummy_netdev(&dev->napi_dev);
-	netif_napi_add(&dev->napi_dev, &dev->napi, mt7603_dma_rx_poll, 64);
-	napi_enable(&dev->napi);
-
 	tasklet_init(&dev->tx_tasklet, mt7603_tx_tasklet, (unsigned long) dev);
-	tasklet_init(&dev->rx_tasklet, mt7603_rx_tasklet, (unsigned long) dev);
 
 	mt76_clear(dev, MT_WPDMA_GLO_CFG,
 		   MT_WPDMA_GLO_CFG_TX_DMA_EN |
@@ -261,6 +230,5 @@ void mt7603_dma_cleanup(struct mt7603_dev *dev)
 		   MT_WPDMA_GLO_CFG_TX_WRITEBACK_DONE);
 
 	tasklet_kill(&dev->tx_tasklet);
-	tasklet_kill(&dev->rx_tasklet);
 	mt76_dma_cleanup(&dev->mt76);
 }
diff --git a/mt7603_init.c b/mt7603_init.c
index 88c7dcfe..3ffb5a57 100644
--- a/mt7603_init.c
+++ b/mt7603_init.c
@@ -22,6 +22,7 @@ struct mt7603_dev *mt7603_alloc_device(struct device *pdev)
 		.fill_txwi = mt7603_mac_write_txwi,
 		.tx_queue_skb = mt7603_tx_queue_skb,
 		.rx_skb = mt7603_queue_rx_skb,
+		.rx_poll_complete = mt7603_rx_poll_complete,
 	};
 	struct ieee80211_hw *hw;
 	struct mt7603_dev *dev;
diff --git a/mt76x2.h b/mt76x2.h
index 1347b39a..f77d8449 100644
--- a/mt76x2.h
+++ b/mt76x2.h
@@ -105,11 +105,7 @@ struct mt76x2_dev {
 	struct mt76x2_mcu mcu;
 	struct sk_buff *rx_head;
 
-	struct net_device napi_dev;
-	struct napi_struct napi;
-
 	struct tasklet_struct tx_tasklet;
-	struct tasklet_struct rx_tasklet;
 	struct tasklet_struct pre_tbtt_tasklet;
 	struct delayed_work cal_work;
 	struct delayed_work mac_work;
@@ -227,6 +223,8 @@ void mt76x2_pre_tbtt_tasklet(unsigned long data);
 
 void mt76x2_txq_init(struct mt76x2_dev *dev, struct ieee80211_txq *txq);
 
+
+void mt76x2_rx_poll_complete(struct mt76_dev *dev, enum mt76_rxq_id q);
 void mt76x2_queue_rx_skb(struct mt76_dev *mdev, enum mt76_rxq_id q,
 			 struct sk_buff *skb);
 
diff --git a/mt76x2_core.c b/mt76x2_core.c
index 9274cac5..7c0bf2d5 100644
--- a/mt76x2_core.c
+++ b/mt76x2_core.c
@@ -36,6 +36,12 @@ void mt76x2_set_irq_mask(struct mt76x2_dev *dev, u32 clear, u32 set)
 	spin_unlock_irqrestore(&dev->irq_lock, flags);
 }
 
+void mt76x2_rx_poll_complete(struct mt76_dev *mdev, enum mt76_rxq_id q)
+{
+	struct mt76x2_dev *dev = container_of(mdev, struct mt76x2_dev, mt76);
+	mt76x2_irq_enable(dev, MT_INT_RX_DONE(q));
+}
+
 irqreturn_t mt76x2_irq_handler(int irq, void *dev_instance)
 {
 	struct mt76x2_dev *dev = dev_instance;
@@ -58,12 +64,12 @@ irqreturn_t mt76x2_irq_handler(int irq, void *dev_instance)
 
 	if (intr & MT_INT_RX_DONE(0)) {
 		mt76x2_irq_disable(dev, MT_INT_RX_DONE(0));
-		napi_schedule(&dev->napi);
+		napi_schedule(&dev->mt76.napi[0]);
 	}
 
 	if (intr & MT_INT_RX_DONE(1)) {
 		mt76x2_irq_disable(dev, MT_INT_RX_DONE(1));
-		tasklet_schedule(&dev->rx_tasklet);
+		napi_schedule(&dev->mt76.napi[1]);
 	}
 
 	if (intr & MT_INT_PRE_TBTT)
diff --git a/mt76x2_dma.c b/mt76x2_dma.c
index 1a56a4ad..95d47aa2 100644
--- a/mt76x2_dma.c
+++ b/mt76x2_dma.c
@@ -177,33 +177,6 @@ mt76x2_tx_tasklet(unsigned long data)
 	mt76x2_irq_enable(dev, MT_INT_TX_DONE_ALL);
 }
 
-static int
-mt76x2_dma_rx_poll(struct napi_struct *napi, int budget)
-{
-	struct mt76x2_dev *dev = container_of(napi, struct mt76x2_dev, napi);
-	int done;
-
-	done = mt76_queue_rx_process(dev, &dev->mt76.q_rx[MT_RXQ_MAIN], budget);
-
-	if (done < budget) {
-		napi_complete(napi);
-		mt76x2_irq_enable(dev, MT_INT_RX_DONE(0));
-	}
-
-	return done;
-}
-
-static void
-mt76x2_rx_tasklet(unsigned long data)
-{
-	struct mt76x2_dev *dev = (struct mt76x2_dev *) data;
-	struct mt76_queue *q = &dev->mt76.q_rx[MT_RXQ_MCU];
-
-	mt76_queue_rx_process(dev, q, q->ndesc);
-
-	mt76x2_irq_enable(dev, MT_INT_RX_DONE(1));
-}
-
 int mt76x2_dma_init(struct mt76x2_dev *dev)
 {
 	static const u8 wmm_queue_map[] = {
@@ -225,11 +198,7 @@ int mt76x2_dma_init(struct mt76x2_dev *dev)
 	init_waitqueue_head(&dev->mcu.wait);
 	skb_queue_head_init(&dev->mcu.res_q);
 
-	init_dummy_netdev(&dev->napi_dev);
-	netif_napi_add(&dev->napi_dev, &dev->napi, mt76x2_dma_rx_poll, 64);
-
 	tasklet_init(&dev->tx_tasklet, mt76x2_tx_tasklet, (unsigned long) dev);
-	tasklet_init(&dev->rx_tasklet, mt76x2_rx_tasklet, (unsigned long) dev);
 
 	mt76_wr(dev, MT_WPDMA_RST_IDX, ~0);
 
@@ -268,6 +237,5 @@ int mt76x2_dma_init(struct mt76x2_dev *dev)
 void mt76x2_dma_cleanup(struct mt76x2_dev *dev)
 {
 	tasklet_kill(&dev->tx_tasklet);
-	tasklet_kill(&dev->rx_tasklet);
 	mt76_dma_cleanup(&dev->mt76);
 }
diff --git a/mt76x2_init.c b/mt76x2_init.c
index b160e4e9..00e785a0 100644
--- a/mt76x2_init.c
+++ b/mt76x2_init.c
@@ -592,6 +592,7 @@ struct mt76x2_dev *mt76x2_alloc_device(struct device *pdev)
 		.fill_txwi = mt76x2_mac_write_txwi,
 		.tx_queue_skb = mt76x2_tx_queue_skb,
 		.rx_skb = mt76x2_queue_rx_skb,
+		.rx_poll_complete = mt76x2_rx_poll_complete,
 	};
 	struct ieee80211_hw *hw;
 	struct mt76x2_dev *dev;
diff --git a/mt76x2_main.c b/mt76x2_main.c
index 9ae94156..81e9a454 100644
--- a/mt76x2_main.c
+++ b/mt76x2_main.c
@@ -31,7 +31,6 @@ mt76x2_start(struct ieee80211_hw *hw)
 
 	ieee80211_queue_delayed_work(mt76_hw(dev), &dev->mac_work,
 				     MT_CALIBRATE_INTERVAL);
-	napi_enable(&dev->napi);
 
 	set_bit(MT76_STATE_RUNNING, &dev->mt76.state);
 
@@ -46,7 +45,6 @@ mt76x2_stop(struct ieee80211_hw *hw)
 	struct mt76x2_dev *dev = hw->priv;
 
 	mutex_lock(&dev->mutex);
-	napi_disable(&dev->napi);
 	clear_bit(MT76_STATE_RUNNING, &dev->mt76.state);
 	mt76x2_stop_hardware(dev);
 	mutex_unlock(&dev->mutex);