diff --git a/dma.c b/dma.c
index 6dd6a61c..d8f61e54 100644
--- a/dma.c
+++ b/dma.c
@@ -310,6 +310,8 @@ mt76_dma_tx_queue_skb(struct mt76_dev *dev, enum mt76_txq_id qid,
 	txwi = mt76_get_txwi_ptr(dev, t);
 
 	skb->prev = skb->next = NULL;
+	if (dev->drv->tx_aligned4_skbs)
+		mt76_insert_hdr_pad(skb);
 
 	len = skb_headlen(skb);
 	addr = dma_map_single(dev->dev, skb->data, len, DMA_TO_DEVICE);
diff --git a/mt76.h b/mt76.h
index 6301f1af..00ae1d68 100644
--- a/mt76.h
+++ b/mt76.h
@@ -292,6 +292,7 @@ struct mt76_hw_cap {
 #define MT_TXWI_NO_FREE			BIT(0)
 
 struct mt76_driver_ops {
+	bool tx_aligned4_skbs;
 	u32 txwi_flags;
 	u16 txwi_size;
 
@@ -675,6 +676,20 @@ static inline struct mt76_tx_cb *mt76_tx_skb_cb(struct sk_buff *skb)
 	return ((void *) IEEE80211_SKB_CB(skb)->status.status_driver_data);
 }
 
+static inline void mt76_insert_hdr_pad(struct sk_buff *skb)
+{
+	int len = ieee80211_get_hdrlen_from_skb(skb);
+
+	if (len % 4 == 0)
+		return;
+
+	skb_push(skb, 2);
+	memmove(skb->data, skb->data + 2, len);
+
+	skb->data[len] = 0;
+	skb->data[len + 1] = 0;
+}
+
 static inline bool mt76_is_skb_pktid(u8 pktid)
 {
 	if (pktid & MT_PACKET_ID_HAS_RATE)
diff --git a/mt76x0/pci.c b/mt76x0/pci.c
index 246c0fd5..4585e1b7 100644
--- a/mt76x0/pci.c
+++ b/mt76x0/pci.c
@@ -153,6 +153,7 @@ mt76x0e_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	static const struct mt76_driver_ops drv_ops = {
 		.txwi_size = sizeof(struct mt76x02_txwi),
+		.tx_aligned4_skbs = true,
 		.update_survey = mt76x02_update_channel,
 		.tx_prepare_skb = mt76x02_tx_prepare_skb,
 		.tx_complete_skb = mt76x02_tx_complete_skb,
diff --git a/mt76x02_txrx.c b/mt76x02_txrx.c
index 40a1e8b7..82acebb7 100644
--- a/mt76x02_txrx.c
+++ b/mt76x02_txrx.c
@@ -154,19 +154,17 @@ int mt76x02_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
 	struct mt76x02_dev *dev = container_of(mdev, struct mt76x02_dev, mt76);
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)tx_info->skb->data;
 	struct mt76x02_txwi *txwi = txwi_ptr;
-	struct sk_buff *skb = tx_info->skb;
-	int hdrlen, pad, len, pid, qsel = MT_QSEL_EDCA;
-	bool ampdu = IEEE80211_SKB_CB(skb)->flags & IEEE80211_TX_CTL_AMPDU;
+	bool ampdu = IEEE80211_SKB_CB(tx_info->skb)->flags & IEEE80211_TX_CTL_AMPDU;
+	int hdrlen, len, pid, qsel = MT_QSEL_EDCA;
 
 	if (qid == MT_TXQ_PSD && wcid && wcid->idx < 128)
 		mt76x02_mac_wcid_set_drop(dev, wcid->idx, false);
 
 	hdrlen = ieee80211_hdrlen(hdr->frame_control);
-	pad = skb->len > hdrlen ? (hdrlen & 2) : 0;
-	len = skb->len - pad;
-	mt76x02_mac_write_txwi(dev, txwi, skb, wcid, sta, len);
+	len = tx_info->skb->len - (hdrlen & 2);
+	mt76x02_mac_write_txwi(dev, txwi, tx_info->skb, wcid, sta, len);
 
-	pid = mt76_tx_status_skb_add(mdev, wcid, skb);
+	pid = mt76_tx_status_skb_add(mdev, wcid, tx_info->skb);
 
 	/* encode packet rate for no-skb packet id to fix up status reporting */
 	if (pid == MT_PACKET_ID_NO_SKB)
diff --git a/mt76x02_usb_core.c b/mt76x02_usb_core.c
index 2cc190a9..f3696afc 100644
--- a/mt76x02_usb_core.c
+++ b/mt76x02_usb_core.c
@@ -77,21 +77,19 @@ int mt76x02u_tx_prepare_skb(struct mt76_dev *mdev, void *data,
 			    struct mt76_tx_info *tx_info)
 {
 	struct mt76x02_dev *dev = container_of(mdev, struct mt76x02_dev, mt76);
-	struct sk_buff *skb = tx_info->skb;
-	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
-	int pid, hdrlen, len = skb->len, ep = q2ep(mdev->q_tx[qid].q->hw_idx);
-	bool ampdu = IEEE80211_SKB_CB(skb)->flags & IEEE80211_TX_CTL_AMPDU;
+	int pid, len = tx_info->skb->len, ep = q2ep(mdev->q_tx[qid].q->hw_idx);
 	struct mt76x02_txwi *txwi;
+	bool ampdu = IEEE80211_SKB_CB(tx_info->skb)->flags & IEEE80211_TX_CTL_AMPDU;
 	enum mt76_qsel qsel;
 	u32 flags;
 
-	hdrlen = ieee80211_hdrlen(hdr->frame_control);
-	len -= skb->len > hdrlen ? (hdrlen & 2) : 0;
-	txwi = (struct mt76x02_txwi *)(skb->data - sizeof(struct mt76x02_txwi));
-	mt76x02_mac_write_txwi(dev, txwi, skb, wcid, sta, len);
-	skb_push(skb, sizeof(struct mt76x02_txwi));
+	mt76_insert_hdr_pad(tx_info->skb);
 
-	pid = mt76_tx_status_skb_add(mdev, wcid, skb);
+	txwi = (struct mt76x02_txwi *)(tx_info->skb->data - sizeof(*txwi));
+	mt76x02_mac_write_txwi(dev, txwi, tx_info->skb, wcid, sta, len);
+	skb_push(tx_info->skb, sizeof(*txwi));
+
+	pid = mt76_tx_status_skb_add(mdev, wcid, tx_info->skb);
 
 	/* encode packet rate for no-skb packet id to fix up status reporting */
 	if (pid == MT_PACKET_ID_NO_SKB)
diff --git a/mt76x02_util.c b/mt76x02_util.c
index 55e43e1f..ad532344 100644
--- a/mt76x02_util.c
+++ b/mt76x02_util.c
@@ -138,6 +138,7 @@ void mt76x02_init_device(struct mt76x02_dev *dev)
 	hw->max_rates = 1;
 	hw->max_report_rates = 7;
 	hw->max_rate_tries = 1;
+	hw->extra_tx_headroom = 2;
 
 	wiphy->interface_modes =
 		BIT(NL80211_IFTYPE_STATION) |
@@ -173,7 +174,6 @@ void mt76x02_init_device(struct mt76x02_dev *dev)
 	hw->sta_data_size = sizeof(struct mt76x02_sta);
 	hw->vif_data_size = sizeof(struct mt76x02_vif);
 
-	ieee80211_hw_set(hw, TX_NEEDS_ALIGNED4_SKBS);
 	ieee80211_hw_set(hw, SUPPORTS_HT_CCK_RATES);
 	ieee80211_hw_set(hw, SUPPORTS_REORDERING_BUFFER);
 
diff --git a/mt76x2/pci.c b/mt76x2/pci.c
index cab00d0b..e84d5c59 100644
--- a/mt76x2/pci.c
+++ b/mt76x2/pci.c
@@ -32,6 +32,7 @@ mt76pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	static const struct mt76_driver_ops drv_ops = {
 		.txwi_size = sizeof(struct mt76x02_txwi),
+		.tx_aligned4_skbs = true,
 		.update_survey = mt76x02_update_channel,
 		.tx_prepare_skb = mt76x02_tx_prepare_skb,
 		.tx_complete_skb = mt76x02_tx_complete_skb,