diff --git a/mt76.h b/mt76.h
index eb72e4bf..91cc6ef7 100644
--- a/mt76.h
+++ b/mt76.h
@@ -252,10 +252,11 @@ struct mt76_rx_tid {
 #define MT_TX_CB_TXS_DONE		BIT(1)
 #define MT_TX_CB_TXS_FAILED		BIT(2)
 
-#define MT_PACKET_ID_MASK		GENMASK(7, 0)
+#define MT_PACKET_ID_MASK		GENMASK(6, 0)
 #define MT_PACKET_ID_NO_ACK		0
 #define MT_PACKET_ID_NO_SKB		1
 #define MT_PACKET_ID_FIRST		2
+#define MT_PACKET_ID_HAS_RATE		BIT(7)
 
 #define MT_TX_STATUS_SKB_TIMEOUT	HZ
 
@@ -647,10 +648,17 @@ 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 bool mt76_is_skb_pktid(u8 pktid)
+{
+	if (pktid & MT_PACKET_ID_HAS_RATE)
+		return false;
+
+	return pktid >= MT_PACKET_ID_FIRST;
+}
+
 int mt76_dma_tx_queue_skb(struct mt76_dev *dev, struct mt76_queue *q,
 			  struct sk_buff *skb, struct mt76_wcid *wcid,
 			  struct ieee80211_sta *sta);
-
 void mt76_rx(struct mt76_dev *dev, enum mt76_rxq_id q, struct sk_buff *skb);
 void mt76_tx(struct mt76_dev *dev, struct ieee80211_sta *sta,
 	     struct mt76_wcid *wcid, struct sk_buff *skb);
diff --git a/mt76x02_mac.c b/mt76x02_mac.c
index 9ed231ab..dd050416 100644
--- a/mt76x02_mac.c
+++ b/mt76x02_mac.c
@@ -411,31 +411,87 @@ void mt76x02_mac_write_txwi(struct mt76x02_dev *dev, struct mt76x02_txwi *txwi,
 }
 EXPORT_SYMBOL_GPL(mt76x02_mac_write_txwi);
 
+static void
+mt76x02_tx_rate_fallback(struct ieee80211_tx_rate *rates, int idx, int phy)
+{
+	u8 mcs, nss;
+
+	if (!idx)
+		return;
+
+	rates += idx - 1;
+	rates[1] = rates[0];
+	switch (phy) {
+	case MT_PHY_TYPE_VHT:
+		mcs = ieee80211_rate_get_vht_mcs(rates);
+		nss = ieee80211_rate_get_vht_nss(rates);
+
+		if (mcs == 0)
+			nss = max_t(int, nss - 1, 1);
+		else
+			mcs--;
+
+		ieee80211_rate_set_vht(rates + 1, mcs, nss);
+		break;
+	case MT_PHY_TYPE_HT_GF:
+	case MT_PHY_TYPE_HT:
+		/* MCS 8 falls back to MCS 0 */
+		if (rates[0].idx == 8) {
+		    rates[1].idx = 0;
+		    break;
+		}
+		/* fall through */
+	default:
+		rates[1].idx = max_t(int, rates[0].idx - 1, 0);
+		break;
+	}
+}
+
 static void
 mt76x02_mac_fill_tx_status(struct mt76x02_dev *dev,
 			   struct ieee80211_tx_info *info,
 			   struct mt76x02_tx_status *st, int n_frames)
 {
 	struct ieee80211_tx_rate *rate = info->status.rates;
-	int cur_idx, last_rate;
+	struct ieee80211_tx_rate last_rate;
+	u16 first_rate;
+	int retry = st->retry;
+	int phy;
 	int i;
 
 	if (!n_frames)
 		return;
 
-	last_rate = min_t(int, st->retry, IEEE80211_TX_MAX_RATES - 1);
-	mt76x02_mac_process_tx_rate(&rate[last_rate], st->rate,
-				    dev->mt76.chandef.chan->band);
-	if (last_rate < IEEE80211_TX_MAX_RATES - 1)
-		rate[last_rate + 1].idx = -1;
+	phy = FIELD_GET(MT_RXWI_RATE_PHY, st->rate);
 
-	cur_idx = rate[last_rate].idx + last_rate;
-	for (i = 0; i <= last_rate; i++) {
-		rate[i].flags = rate[last_rate].flags;
-		rate[i].idx = max_t(int, 0, cur_idx - i);
-		rate[i].count = 1;
+	if (st->pktid & MT_PACKET_ID_HAS_RATE) {
+		first_rate = st->rate & ~MT_RXWI_RATE_INDEX;
+		first_rate |= st->pktid & MT_RXWI_RATE_INDEX;
+
+		mt76x02_mac_process_tx_rate(&rate[0], first_rate,
+					    dev->mt76.chandef.chan->band);
+	}
+
+	mt76x02_mac_process_tx_rate(&last_rate, st->rate,
+				    dev->mt76.chandef.chan->band);
+
+	for (i = 0; i < ARRAY_SIZE(info->status.rates); i++) {
+		retry--;
+		if (i + 1 == ARRAY_SIZE(info->status.rates)) {
+			info->status.rates[i] = last_rate;
+			info->status.rates[i].count = max_t(int, retry, 1);
+			break;
+		}
+
+		mt76x02_tx_rate_fallback(info->status.rates, i, phy);
+		if (info->status.rates[i].idx == last_rate.idx)
+			break;
+	}
+
+	if (i + 1 < ARRAY_SIZE(info->status.rates)) {
+		info->status.rates[i + 1].idx = -1;
+		info->status.rates[i + 1].count = 0;
 	}
-	rate[last_rate].count = st->retry + 1 - last_rate;
 
 	info->status.ampdu_len = n_frames;
 	info->status.ampdu_ack_len = st->success ? n_frames : 0;
@@ -480,13 +536,19 @@ void mt76x02_send_tx_status(struct mt76x02_dev *dev,
 	}
 
 	if (wcid) {
-		if (stat->pktid >= MT_PACKET_ID_FIRST)
+		if (mt76_is_skb_pktid(stat->pktid))
 			status.skb = mt76_tx_status_skb_get(mdev, wcid,
 							    stat->pktid, &list);
 		if (status.skb)
 			status.info = IEEE80211_SKB_CB(status.skb);
 	}
 
+	if (!status.skb && !(stat->pktid & MT_PACKET_ID_HAS_RATE)) {
+		mt76_tx_status_unlock(mdev, &list);
+		rcu_read_unlock();
+		return;
+	}
+
 	if (msta && stat->aggr && !status.skb) {
 		u32 stat_val, stat_cache;
 
diff --git a/mt76x02_txrx.c b/mt76x02_txrx.c
index 94f47248..a842e0aa 100644
--- a/mt76x02_txrx.c
+++ b/mt76x02_txrx.c
@@ -163,13 +163,18 @@ int mt76x02_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
 	mt76x02_mac_write_txwi(dev, txwi, skb, wcid, sta, skb->len);
 
 	pid = mt76_tx_status_skb_add(mdev, wcid, skb);
+
+	/* encode packet rate for no-skb packet id to fix up status reporting */
+	if (pid == MT_PACKET_ID_NO_SKB)
+		pid = MT_PACKET_ID_HAS_RATE | (txwi->rate & MT_RXWI_RATE_INDEX);
+
 	txwi->pktid = pid;
 
 	ret = mt76x02_insert_hdr_pad(skb);
 	if (ret < 0)
 		return ret;
 
-	if (pid >= MT_PACKET_ID_FIRST)
+	if (mt76_is_skb_pktid(pid))
 		qsel = MT_QSEL_MGMT;
 
 	*tx_info = FIELD_PREP(MT_TXD_INFO_QSEL, qsel) |
diff --git a/mt76x02_usb_core.c b/mt76x02_usb_core.c
index 6fb52b59..b5deecd1 100644
--- a/mt76x02_usb_core.c
+++ b/mt76x02_usb_core.c
@@ -90,9 +90,14 @@ int mt76x02u_tx_prepare_skb(struct mt76_dev *mdev, void *data,
 	skb_push(skb, sizeof(struct mt76x02_txwi));
 
 	pid = mt76_tx_status_skb_add(mdev, wcid, skb);
+
+	/* encode packet rate for no-skb packet id to fix up status reporting */
+	if (pid == MT_PACKET_ID_NO_SKB)
+		pid = MT_PACKET_ID_HAS_RATE | (txwi->rate & MT_RXWI_RATE_INDEX);
+
 	txwi->pktid = pid;
 
-	if (pid >= MT_PACKET_ID_FIRST || q2ep(q->hw_idx) == MT_EP_OUT_HCCA)
+	if (mt76_is_skb_pktid(pid) || q2ep(q->hw_idx) == MT_EP_OUT_HCCA)
 		qsel = MT_QSEL_MGMT;
 	else
 		qsel = MT_QSEL_EDCA;