diff --git a/dma.c b/dma.c
index 2d23467b..b22dbd46 100644
--- a/dma.c
+++ b/dma.c
@@ -134,6 +134,7 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, enum mt76_txq_id qid, bool flush)
 {
 	struct mt76_queue *q = &dev->q_tx[qid];
 	struct mt76_queue_entry entry;
+	bool wake = false;
 	int last;
 
 	if (!q->ndesc)
@@ -153,8 +154,10 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, enum mt76_txq_id qid, bool flush)
 		if (entry.skb)
 			dev->drv->tx_complete_skb(dev, q, &entry, flush);
 
-		if (entry.txwi)
+		if (entry.txwi) {
 			mt76_put_txwi(dev, entry.txwi);
+			wake = true;
+		}
 
 		q->tail = (q->tail + 1) % q->ndesc;
 		q->queued--;
@@ -168,7 +171,11 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, enum mt76_txq_id qid, bool flush)
 	else
 		mt76_dma_sync_idx(dev, q);
 
+	wake = wake && qid < IEEE80211_NUM_ACS && q->queued < q->ndesc - 8;
 	spin_unlock_bh(&q->lock);
+
+	if (wake)
+		ieee80211_wake_queue(dev->hw, qid);
 }
 
 static void *
diff --git a/mt76x2_tx.c b/mt76x2_tx.c
index 309e7dad..8d9f134f 100644
--- a/mt76x2_tx.c
+++ b/mt76x2_tx.c
@@ -41,8 +41,6 @@ void mt76x2_tx(struct ieee80211_hw *hw, struct ieee80211_tx_control *control,
 void mt76x2_tx_complete(struct mt76x2_dev *dev, struct sk_buff *skb)
 {
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
-	struct mt76_queue *q;
-	int qid = skb_get_queue_mapping(skb);
 
 	if (info->flags & IEEE80211_TX_CTL_AMPDU) {
 		ieee80211_free_txskb(mt76_hw(dev), skb);
@@ -52,10 +50,6 @@ void mt76x2_tx_complete(struct mt76x2_dev *dev, struct sk_buff *skb)
 		info->flags |= IEEE80211_TX_STAT_ACK;
 		ieee80211_tx_status(mt76_hw(dev), skb);
 	}
-
-	q = &dev->mt76.q_tx[qid];
-	if (q->queued < q->ndesc - 8)
-		ieee80211_wake_queue(mt76_hw(dev), qid);
 }
 
 int mt76x2_tx_prepare_skb(struct mt76_dev *mdev, void *txwi,