[PATCH v8 1/9] mt76: add hash lookup for skb on TXS status_list

[PATCH v8 1/9] mt76: add hash lookup for skb on TXS status_list

[greearb]

From: Ben Greear <greearb@candelatech.com>

This improves performance when sending lots of frames that
are requesting being mapped to a TXS callback.

Add comments to help next person understood the tx path
better.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---

v8:  Re-do the txs hash per Felix's suggestions.
   Remove the tx-overrides patch from this series per Kalle.

 drivers/net/wireless/mediatek/mt76/mt76.h     | 48 +++++++---
 .../net/wireless/mediatek/mt76/mt7603/mac.c   |  2 +-
 .../net/wireless/mediatek/mt76/mt7615/mac.c   |  2 +-
 .../net/wireless/mediatek/mt76/mt76x02_mac.c  |  2 +-
 .../net/wireless/mediatek/mt76/mt7915/mac.c   |  4 +-
 .../net/wireless/mediatek/mt76/mt7921/mac.c   |  9 +-
 drivers/net/wireless/mediatek/mt76/tx.c       | 94 ++++++++++++++++---
 7 files changed, 130 insertions(+), 31 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 436bf2b8e2cd..3ea4936753c2 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -235,6 +235,16 @@ DECLARE_EWMA(signal, 10, 8);
 #define MT_WCID_TX_INFO_TXPWR_ADJ	GENMASK(25, 18)
 #define MT_WCID_TX_INFO_SET		BIT(31)
 
+#define MT_PACKET_ID_MASK		GENMASK(6, 0)
+#define MT_PACKET_ID_NO_ACK		0
+/* Request TXS, but don't try to match with skb. */
+#define MT_PACKET_ID_NO_SKB		1
+#define MT_PACKET_ID_FIRST		2
+#define MT_PACKET_ID_HAS_RATE		BIT(7)
+#define MT_PACKET_ID_MAX		(GENMASK(7, 0) - 1)
+#define MT_NUM_STATUS_PACKETS \
+	(MT_PACKET_ID_MAX + 1 - MT_PACKET_ID_FIRST)
+
 struct mt76_wcid {
 	struct mt76_rx_tid __rcu *aggr[IEEE80211_NUM_TIDS];
 
@@ -246,6 +256,8 @@ struct mt76_wcid {
 
 	struct rate_info rate;
 
+	struct sk_buff *skb_status_array[MT_NUM_STATUS_PACKETS];
+
 	u16 idx;
 	u8 hw_key_idx;
 	u8 hw_key_idx2;
@@ -302,16 +314,16 @@ 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(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
+/* This is timer for when to give up when waiting for TXS callback,
+ * with starting time being the time at which the DMA_DONE callback
+ * was seen (so, we know packet was processed then, it should not take
+ * long after that for firmware to send the TXS callback if it is going
+ * to do so.)
+ */
+#define MT_TX_STATUS_SKB_TIMEOUT	(HZ / 4)
 
 struct mt76_tx_cb {
-	unsigned long jiffies;
+	unsigned long jiffies; /* for when DMA_DONE was reported. */
 	u16 wcid;
 	u8 pktid;
 	u8 flags;
@@ -651,6 +663,7 @@ struct mt76_dev {
 	spinlock_t cc_lock;
 
 	u32 cur_cc_bss_rx;
+	unsigned long next_status_jiffies;
 
 	struct mt76_rx_status rx_ampdu_status;
 	u32 rx_ampdu_len;
@@ -1090,7 +1103,7 @@ struct sk_buff *mt76_tx_status_skb_get(struct mt76_dev *dev,
 				       struct mt76_wcid *wcid, int pktid,
 				       struct sk_buff_head *list);
 void mt76_tx_status_skb_done(struct mt76_dev *dev, struct sk_buff *skb,
-			     struct sk_buff_head *list);
+			     struct sk_buff_head *list, struct mt76_wcid *wcid);
 void __mt76_tx_complete_skb(struct mt76_dev *dev, u16 wcid, struct sk_buff *skb,
 			    struct list_head *free_list);
 static inline void
@@ -1297,13 +1310,24 @@ mt76_token_put(struct mt76_dev *dev, int token)
 }
 
 static inline int
-mt76_get_next_pkt_id(struct mt76_wcid *wcid)
+mt76_get_next_pkt_id(struct mt76_dev *dev, struct mt76_wcid *wcid,
+		     struct sk_buff *skb)
 {
+	struct sk_buff *qskb;
+
+	lockdep_assert_held(&dev->status_list.lock);
+
 	wcid->packet_id = (wcid->packet_id + 1) & MT_PACKET_ID_MASK;
-	if (wcid->packet_id == MT_PACKET_ID_NO_ACK ||
-	    wcid->packet_id == MT_PACKET_ID_NO_SKB)
+	if (wcid->packet_id < MT_PACKET_ID_FIRST)
 		wcid->packet_id = MT_PACKET_ID_FIRST;
 
+	qskb = wcid->skb_status_array[wcid->packet_id - MT_PACKET_ID_FIRST];
+	if (qskb)
+		/* bummer, already waiting on this pid.  Other logic
+		 * will clean up stale entries.
+		 */
+		return MT_PACKET_ID_NO_SKB;
+
 	return wcid->packet_id;
 }
 #endif
diff --git a/drivers/net/wireless/mediatek/mt76/mt7603/mac.c b/drivers/net/wireless/mediatek/mt76/mt7603/mac.c
index 3972c56136a2..2f268eb7c1e6 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7603/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7603/mac.c
@@ -1230,7 +1230,7 @@ mt7603_mac_add_txs_skb(struct mt7603_dev *dev, struct mt7603_sta *sta, int pid,
 			info->status.rates[0].idx = -1;
 		}
 
-		mt76_tx_status_skb_done(mdev, skb, &list);
+		mt76_tx_status_skb_done(mdev, skb, &list, &sta->wcid);
 	}
 	mt76_tx_status_unlock(mdev, &list);
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/mac.c b/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
index ff3f85e4087c..381a998817d4 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
@@ -1433,7 +1433,7 @@ static bool mt7615_mac_add_txs_skb(struct mt7615_dev *dev,
 			info->status.rates[0].idx = -1;
 		}
 
-		mt76_tx_status_skb_done(mdev, skb, &list);
+		mt76_tx_status_skb_done(mdev, skb, &list, &sta->wcid);
 	}
 	mt76_tx_status_unlock(mdev, &list);
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt76x02_mac.c b/drivers/net/wireless/mediatek/mt76/mt76x02_mac.c
index c32e6dc68773..fce020e64678 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76x02_mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt76x02_mac.c
@@ -622,7 +622,7 @@ void mt76x02_send_tx_status(struct mt76x02_dev *dev,
 		info = *status.info;
 		len = status.skb->len;
 		ac = skb_get_queue_mapping(status.skb);
-		mt76_tx_status_skb_done(mdev, status.skb, &list);
+		mt76_tx_status_skb_done(mdev, status.skb, &list, wcid);
 	} else if (msta) {
 		len = status.info->status.ampdu_len * ewma_pktlen_read(&msta->pktlen);
 		ac = FIELD_GET(MT_PKTID_AC, cur_pktid);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index d9f52e2611a7..c8432054a63e 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -1318,6 +1318,7 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 
 	mt76_tx_status_lock(mdev, &list);
 	skb = mt76_tx_status_skb_get(mdev, wcid, pid, &list);
+
 	if (!skb)
 		goto out;
 
@@ -1417,10 +1418,11 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 		stats->tx_bw[0]++;
 		break;
 	}
+
 	wcid->rate = rate;
 
 out:
-	mt76_tx_status_skb_done(mdev, skb, &list);
+	mt76_tx_status_skb_done(mdev, skb, &list, wcid);
 	mt76_tx_status_unlock(mdev, &list);
 
 	return !!skb;
diff --git a/drivers/net/wireless/mediatek/mt76/mt7921/mac.c b/drivers/net/wireless/mediatek/mt76/mt7921/mac.c
index 76985a6b3be5..219c17d77e46 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7921/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7921/mac.c
@@ -732,7 +732,9 @@ mt7921_mac_write_txwi_80211(struct mt7921_dev *dev, __le32 *txwi,
 	txwi[7] |= cpu_to_le32(val);
 }
 
-static void mt7921_update_txs(struct mt76_wcid *wcid, __le32 *txwi)
+static void mt7921_update_txs(struct mt7921_dev *dev,
+			      struct mt76_wcid *wcid, __le32 *txwi,
+			      struct sk_buff *skb)
 {
 	struct mt7921_sta *msta = container_of(wcid, struct mt7921_sta, wcid);
 	u32 pid, frame_type = FIELD_GET(MT_TXD2_FRAME_TYPE, txwi[2]);
@@ -744,7 +746,7 @@ static void mt7921_update_txs(struct mt76_wcid *wcid, __le32 *txwi)
 		return;
 
 	msta->next_txs_ts = jiffies + msecs_to_jiffies(250);
-	pid = mt76_get_next_pkt_id(wcid);
+	pid = mt76_get_next_pkt_id(&dev->mt76, wcid, skb);
 	txwi[5] |= cpu_to_le32(MT_TXD5_TX_STATUS_MCU |
 			       FIELD_PREP(MT_TXD5_PID, pid));
 }
@@ -771,7 +773,6 @@ void mt7921_mac_write_txwi(struct mt7921_dev *dev, __le32 *txwi,
 {
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
 	struct ieee80211_vif *vif = info->control.vif;
-	struct mt76_phy *mphy = &dev->mphy;
 	u8 p_fmt, q_idx, omac_idx = 0, wmm_idx = 0;
 	bool is_8023 = info->flags & IEEE80211_TX_CTL_HW_80211_ENCAP;
 	u16 tx_count = 15;
@@ -839,7 +840,7 @@ void mt7921_mac_write_txwi(struct mt7921_dev *dev, __le32 *txwi,
 		txwi[3] |= cpu_to_le32(MT_TXD3_BA_DISABLE);
 	}
 
-	mt7921_update_txs(wcid, txwi);
+	mt7921_update_txs(dev, wcid, txwi, skb);
 }
 
 static void
diff --git a/drivers/net/wireless/mediatek/mt76/tx.c b/drivers/net/wireless/mediatek/mt76/tx.c
index 6f302acb6e69..7b281cd062c9 100644
--- a/drivers/net/wireless/mediatek/mt76/tx.c
+++ b/drivers/net/wireless/mediatek/mt76/tx.c
@@ -36,6 +36,7 @@ mt76_tx_check_agg_ssn(struct ieee80211_sta *sta, struct sk_buff *skb)
 }
 EXPORT_SYMBOL_GPL(mt76_tx_check_agg_ssn);
 
+/* Lock list, and initialize the timed-out-skb list object. */
 void
 mt76_tx_status_lock(struct mt76_dev *dev, struct sk_buff_head *list)
 		   __acquires(&dev->status_list.lock)
@@ -45,6 +46,9 @@ mt76_tx_status_lock(struct mt76_dev *dev, struct sk_buff_head *list)
 }
 EXPORT_SYMBOL_GPL(mt76_tx_status_lock);
 
+/* Unlock list, and use last-received status for any skbs that
+ * timed out getting TXS callback (they are on the list passed in
+ */
 void
 mt76_tx_status_unlock(struct mt76_dev *dev, struct sk_buff_head *list)
 		      __releases(&dev->status_list.lock)
@@ -80,20 +84,44 @@ EXPORT_SYMBOL_GPL(mt76_tx_status_unlock);
 
 static void
 __mt76_tx_status_skb_done(struct mt76_dev *dev, struct sk_buff *skb, u8 flags,
-			  struct sk_buff_head *list)
+			  struct sk_buff_head *list, struct mt76_wcid *wcid)
 {
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
 	struct mt76_tx_cb *cb = mt76_tx_skb_cb(skb);
 	u8 done = MT_TX_CB_DMA_DONE | MT_TX_CB_TXS_DONE;
 
+	lockdep_assert_held(&dev->status_list.lock);
+
+	/* Start timer for TXS callback expiration. */
+	if (flags & MT_TX_CB_DMA_DONE)
+		cb->jiffies = jiffies;
+
 	flags |= cb->flags;
 	cb->flags = flags;
 
+	/* Only process skb with TXS status has been received and also
+	 * the txfree (DMA_DONE) callback has happened.
+	 */
 	if ((flags & done) != done)
 		return;
 
 	__skb_unlink(skb, &dev->status_list);
 
+	rcu_read_lock();
+	/* calling code may not know wcid, for instance in the tx_status_check
+	 * path, look it up in that case.
+	 */
+	if (!wcid)
+		wcid = rcu_dereference(dev->wcid[cb->wcid]);
+
+	/* Make sure we clear any cached skb. */
+	if (wcid) {
+		if (!(WARN_ON_ONCE(cb->pktid - MT_PACKET_ID_FIRST
+				   >= ARRAY_SIZE(wcid->skb_status_array))))
+			wcid->skb_status_array[cb->pktid - MT_PACKET_ID_FIRST] = NULL;
+	}
+	rcu_read_unlock();
+
 	/* Tx status can be unreliable. if it fails, mark the frame as ACKed */
 	if (flags & MT_TX_CB_TXS_FAILED) {
 		info->status.rates[0].count = 0;
@@ -106,9 +134,9 @@ __mt76_tx_status_skb_done(struct mt76_dev *dev, struct sk_buff *skb, u8 flags,
 
 void
 mt76_tx_status_skb_done(struct mt76_dev *dev, struct sk_buff *skb,
-			struct sk_buff_head *list)
+			struct sk_buff_head *list, struct mt76_wcid *wcid)
 {
-	__mt76_tx_status_skb_done(dev, skb, MT_TX_CB_TXS_DONE, list);
+	__mt76_tx_status_skb_done(dev, skb, MT_TX_CB_TXS_DONE, list, wcid);
 }
 EXPORT_SYMBOL_GPL(mt76_tx_status_skb_done);
 
@@ -133,12 +161,16 @@ mt76_tx_status_skb_add(struct mt76_dev *dev, struct mt76_wcid *wcid,
 	spin_lock_bh(&dev->status_list.lock);
 
 	memset(cb, 0, sizeof(*cb));
-	pid = mt76_get_next_pkt_id(wcid);
+	pid = mt76_get_next_pkt_id(dev, wcid, skb);
 	cb->wcid = wcid->idx;
 	cb->pktid = pid;
-	cb->jiffies = jiffies;
 
-	__skb_queue_tail(&dev->status_list, skb);
+	if (cb->pktid != MT_PACKET_ID_NO_SKB) {
+		__skb_queue_tail(&dev->status_list, skb);
+		/* cache this skb for fast lookup by packet-id */
+		wcid->skb_status_array[pid - MT_PACKET_ID_FIRST] = skb;
+	}
+
 	spin_unlock_bh(&dev->status_list.lock);
 
 	return pid;
@@ -150,25 +182,65 @@ mt76_tx_status_skb_get(struct mt76_dev *dev, struct mt76_wcid *wcid, int pktid,
 		       struct sk_buff_head *list)
 {
 	struct sk_buff *skb, *tmp;
+	struct sk_buff *rvskb = NULL;
 
+	/* If pktid is < first-valid-id, then it is not something we requested
+	 * TXS for, so we will not find SKB.  Bail out early in that case,
+	 * unless we need to walk due to stale-skb-reaper timeout.
+	 */
+	if (pktid < MT_PACKET_ID_FIRST) {
+		if (!time_after(jiffies, dev->next_status_jiffies))
+			return NULL;
+		goto check_list;
+	}
+
+	if (wcid) {
+		lockdep_assert_held(&dev->status_list.lock);
+		if (WARN_ON_ONCE(pktid - MT_PACKET_ID_FIRST
+				 >= ARRAY_SIZE(wcid->skb_status_array))) {
+			dev_err(dev->dev, "invalid pktid: %d  status-array-size: %d\n",
+				pktid, (int)(ARRAY_SIZE(wcid->skb_status_array)));
+			WARN_ON_ONCE(true);
+			goto check_list;
+		}
+
+		skb = wcid->skb_status_array[pktid - MT_PACKET_ID_FIRST];
+
+		if (skb && !time_after(jiffies, dev->next_status_jiffies))
+			return skb;
+	}
+
+check_list:
 	skb_queue_walk_safe(&dev->status_list, skb, tmp) {
 		struct mt76_tx_cb *cb = mt76_tx_skb_cb(skb);
 
 		if (wcid && cb->wcid != wcid->idx)
 			continue;
 
-		if (cb->pktid == pktid)
-			return skb;
+		if (cb->pktid == pktid) {
+			/* Found our skb, but check for timeouts too */
+			rvskb = skb;
+			continue;
+		}
+
+		/* Cannot time out anything that does not have DMA completed. */
+		if (!(cb->flags & MT_TX_CB_DMA_DONE))
+			continue;
 
+		/* Check if it has been too long since DMA_DONE was reported. */
 		if (pktid >= 0 && !time_after(jiffies, cb->jiffies +
 					      MT_TX_STATUS_SKB_TIMEOUT))
 			continue;
 
+		/* It has been too long since DMA_DONE, time out this packet
+		 * and stop waiting for TXS callback.
+		 */
 		__mt76_tx_status_skb_done(dev, skb, MT_TX_CB_TXS_FAILED |
-						    MT_TX_CB_TXS_DONE, list);
+					  MT_TX_CB_TXS_DONE, list, wcid);
 	}
+	dev->next_status_jiffies = jiffies + MT_TX_STATUS_SKB_TIMEOUT + 1;
 
-	return NULL;
+	return rvskb;
 }
 EXPORT_SYMBOL_GPL(mt76_tx_status_skb_get);
 
@@ -238,7 +310,7 @@ void __mt76_tx_complete_skb(struct mt76_dev *dev, u16 wcid_idx, struct sk_buff *
 	}
 
 	mt76_tx_status_lock(dev, &list);
-	__mt76_tx_status_skb_done(dev, skb, MT_TX_CB_DMA_DONE, &list);
+	__mt76_tx_status_skb_done(dev, skb, MT_TX_CB_DMA_DONE, &list, wcid);
 	mt76_tx_status_unlock(dev, &list);
 
 out:
-- 
2.20.1

[PATCH v8 2/9] mt76: mt7915: fix potential NPE in TXS processing

[greearb]

From: Ben Greear <greearb@candelatech.com>

If skb is null, then don't try to dereference it.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 drivers/net/wireless/mediatek/mt76/mt7915/mac.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index c8432054a63e..05c80a8dd82a 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -1320,7 +1320,7 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 	skb = mt76_tx_status_skb_get(mdev, wcid, pid, &list);
 
 	if (!skb)
-		goto out;
+		goto out_no_skb;
 
 	txs = le32_to_cpu(txs_data[0]);
 
@@ -1423,6 +1423,8 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 
 out:
 	mt76_tx_status_skb_done(mdev, skb, &list, wcid);
+
+out_no_skb:
 	mt76_tx_status_unlock(mdev, &list);
 
 	return !!skb;
-- 
2.20.1

[PATCH v8 3/9] mt76: mt7915: move TXS parsing to its own method

[greearb]

From: Ben Greear <greearb@candelatech.com>

This will allow us to also parse TXS when we have no skb on
the status callback list in future patches.

Explicitly cache short_gi and he_gi in wcid, don't try to store
it in the wcid.rate object.  Slightly less confusing and less fragile
when TXS starts parsing lots of frames.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 drivers/net/wireless/mediatek/mt76/mt76.h     |   5 +
 .../net/wireless/mediatek/mt76/mt7915/mac.c   | 120 ++++++++++--------
 2 files changed, 75 insertions(+), 50 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 3ea4936753c2..9f0d6449fe9b 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -254,7 +254,12 @@ struct mt76_wcid {
 	struct ewma_signal rssi;
 	int inactive_count;
 
+	/* cached rate, updated from mac_sta_poll() and from TXS callback logic, in 7915
+	 * at least.
+	 */
 	struct rate_info rate;
+	bool rate_short_gi; /* cached HT/VHT short_gi, from mac_sta_poll() */
+	u8 rate_he_gi; /* cached HE GI, from mac_sta_poll() */
 
 	struct sk_buff *skb_status_array[MT_NUM_STATUS_PACKETS];
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index 05c80a8dd82a..6f92e207680f 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -205,12 +205,17 @@ static void mt7915_mac_sta_poll(struct mt7915_dev *dev)
 			u8 offs = 24 + 2 * bw;
 
 			rate->he_gi = (val & (0x3 << offs)) >> offs;
+			msta->wcid.rate_he_gi = rate->he_gi; /* cache for later */
 		} else if (rate->flags & RATE_INFO_FLAGS_VHT_MCS) {
-			if (val & BIT(12 + bw))
+			if (val & BIT(12 + bw)) {
 				rate->flags |= RATE_INFO_FLAGS_SHORT_GI;
-			else
+				msta->wcid.rate_short_gi = 1;
+			} else {
 				rate->flags &= ~RATE_INFO_FLAGS_SHORT_GI;
+				msta->wcid.rate_short_gi = 0;
+			}
 		}
+		/* TODO:  Deal with HT_MCS */
 	}
 
 	rcu_read_unlock();
@@ -1302,29 +1307,19 @@ mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 	}
 }
 
-static bool
-mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
-		       __le32 *txs_data, struct mt7915_sta_stats *stats)
+static void
+mt7915_mac_parse_txs(struct mt7915_dev *dev, struct mt76_wcid *wcid,
+		     __le32 *txs_data, struct mt7915_sta_stats *stats,
+		     struct rate_info *rate,
+		     struct ieee80211_tx_info *info)
 {
 	struct ieee80211_supported_band *sband;
-	struct mt76_dev *mdev = &dev->mt76;
 	struct mt76_phy *mphy;
-	struct ieee80211_tx_info *info;
-	struct sk_buff_head list;
-	struct rate_info rate = {};
-	struct sk_buff *skb;
-	bool cck = false;
 	u32 txrate, txs, mode;
-
-	mt76_tx_status_lock(mdev, &list);
-	skb = mt76_tx_status_skb_get(mdev, wcid, pid, &list);
-
-	if (!skb)
-		goto out_no_skb;
+	bool cck = false;
 
 	txs = le32_to_cpu(txs_data[0]);
 
-	info = IEEE80211_SKB_CB(skb);
 	if (!(txs & MT_TXS0_ACK_ERROR_MASK))
 		info->flags |= IEEE80211_TX_STAT_ACK;
 
@@ -1336,18 +1331,18 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 
 	txrate = FIELD_GET(MT_TXS0_TX_RATE, txs);
 
-	rate.mcs = FIELD_GET(MT_TX_RATE_IDX, txrate);
-	rate.nss = FIELD_GET(MT_TX_RATE_NSS, txrate) + 1;
+	rate->mcs = FIELD_GET(MT_TX_RATE_IDX, txrate);
+	rate->nss = FIELD_GET(MT_TX_RATE_NSS, txrate) + 1;
 
-	stats->tx_nss[rate.nss - 1]++;
+	stats->tx_nss[rate->nss - 1]++;
 	/* It appears that rate.mcs even for HT may be small, considering in HT
 	 * code below it is multiplied... but not certain on that,
 	 * so code safely.
 	 */
-	if (rate.mcs >= ARRAY_SIZE(stats->tx_mcs))
+	if (rate->mcs >= ARRAY_SIZE(stats->tx_mcs))
 		stats->tx_mcs[ARRAY_SIZE(stats->tx_mcs) - 1]++;
 	else
-		stats->tx_mcs[rate.mcs]++;
+		stats->tx_mcs[rate->mcs]++;
 
 	mode = FIELD_GET(MT_TX_RATE_MODE, txrate);
 	switch (mode) {
@@ -1364,70 +1359,95 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 		else
 			sband = &mphy->sband_2g.sband;
 
-		rate.mcs = mt76_get_rate(mphy->dev, sband, rate.mcs, cck);
-		rate.legacy = sband->bitrates[rate.mcs].bitrate;
+		rate->mcs = mt76_get_rate(mphy->dev, sband, rate->mcs, cck);
+		rate->legacy = sband->bitrates[rate->mcs].bitrate;
 		break;
 	case MT_PHY_TYPE_HT:
 	case MT_PHY_TYPE_HT_GF:
-		rate.mcs += (rate.nss - 1) * 8;
-		if (rate.mcs > 31)
-			goto out;
+		rate->mcs += (rate->nss - 1) * 8;
+		if (rate->mcs > 31)
+			break;
 
-		rate.flags = RATE_INFO_FLAGS_MCS;
-		if (wcid->rate.flags & RATE_INFO_FLAGS_SHORT_GI)
-			rate.flags |= RATE_INFO_FLAGS_SHORT_GI;
+		rate->flags = RATE_INFO_FLAGS_MCS;
+		if (wcid->rate_short_gi)
+			rate->flags |= RATE_INFO_FLAGS_SHORT_GI;
 		break;
 	case MT_PHY_TYPE_VHT:
-		if (rate.mcs > 9)
-			goto out;
+		if (rate->mcs > 9)
+			break;
 
-		rate.flags = RATE_INFO_FLAGS_VHT_MCS;
+		rate->flags = RATE_INFO_FLAGS_VHT_MCS;
+		if (wcid->rate_short_gi)
+			rate->flags |= RATE_INFO_FLAGS_SHORT_GI;
 		break;
 	case MT_PHY_TYPE_HE_SU:
 	case MT_PHY_TYPE_HE_EXT_SU:
 	case MT_PHY_TYPE_HE_TB:
 	case MT_PHY_TYPE_HE_MU:
-		if (rate.mcs > 11)
-			goto out;
+		if (rate->mcs > 11)
+			break;
 
-		rate.he_gi = wcid->rate.he_gi;
-		rate.he_dcm = FIELD_GET(MT_TX_RATE_DCM, txrate);
-		rate.flags = RATE_INFO_FLAGS_HE_MCS;
+		rate->he_gi = wcid->rate_he_gi;
+		rate->he_dcm = FIELD_GET(MT_TX_RATE_DCM, txrate);
+		rate->flags = RATE_INFO_FLAGS_HE_MCS;
 		break;
 	default:
-		goto out;
+		WARN_ON_ONCE(true);
+		mode = 0;
 	}
 
 	stats->tx_mode[mode]++;
 
 	switch (FIELD_GET(MT_TXS0_BW, txs)) {
 	case IEEE80211_STA_RX_BW_160:
-		rate.bw = RATE_INFO_BW_160;
+		rate->bw = RATE_INFO_BW_160;
 		stats->tx_bw[3]++;
 		break;
 	case IEEE80211_STA_RX_BW_80:
-		rate.bw = RATE_INFO_BW_80;
+		rate->bw = RATE_INFO_BW_80;
 		stats->tx_bw[2]++;
 		break;
 	case IEEE80211_STA_RX_BW_40:
-		rate.bw = RATE_INFO_BW_40;
+		rate->bw = RATE_INFO_BW_40;
 		stats->tx_bw[1]++;
 		break;
 	default:
-		rate.bw = RATE_INFO_BW_20;
+		rate->bw = RATE_INFO_BW_20;
 		stats->tx_bw[0]++;
 		break;
 	}
+}
 
-	wcid->rate = rate;
+static void
+mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
+		       __le32 *txs_data, struct mt7915_sta_stats *stats)
+{
+	struct mt76_dev *mdev = &dev->mt76;
+	struct ieee80211_tx_info *info;
+	struct ieee80211_tx_info info_stack;
+	struct sk_buff_head list;
+	/* rate is cached in wcid->rate for skbs that do not request to be
+	 * paired with TXS data.  This is normal datapath.
+	 */
+	struct rate_info *rate = &wcid->rate;
+	struct sk_buff *skb;
 
-out:
-	mt76_tx_status_skb_done(mdev, skb, &list, wcid);
+	mt76_tx_status_lock(mdev, &list);
+	skb = mt76_tx_status_skb_get(mdev, wcid, pid, &list);
 
-out_no_skb:
-	mt76_tx_status_unlock(mdev, &list);
+	memset(rate, 0, sizeof(*rate));
 
-	return !!skb;
+	if (skb)
+		info = IEEE80211_SKB_CB(skb);
+	else
+		info = &info_stack;
+
+	mt7915_mac_parse_txs(dev, wcid, txs_data, stats, rate, info);
+
+	if (skb)
+		mt76_tx_status_skb_done(mdev, skb, &list, wcid);
+
+	mt76_tx_status_unlock(mdev, &list);
 }
 
 static void mt7915_mac_add_txs(struct mt7915_dev *dev, void *data)
-- 
2.20.1

[PATCH v8 4/9] mt76: mt7915: allow processing TXS for 'NO_SKB' pkt-ids

[greearb]

From: Ben Greear <greearb@candelatech.com>

This will let us update stats and wcid.rate for every TXS
callback we receive for a particular wcid.

For now, the TXS is not requested for NO_SKB frames, however.
That will be allowed in next patch.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 drivers/net/wireless/mediatek/mt76/mt7915/mac.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index 6f92e207680f..f6ed5d5febb7 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -1430,10 +1430,16 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 	 * paired with TXS data.  This is normal datapath.
 	 */
 	struct rate_info *rate = &wcid->rate;
-	struct sk_buff *skb;
+	struct sk_buff *skb = NULL;
+	bool check_status;
 
-	mt76_tx_status_lock(mdev, &list);
-	skb = mt76_tx_status_skb_get(mdev, wcid, pid, &list);
+	check_status = ((pid >= MT_PACKET_ID_FIRST) ||
+			time_after(jiffies, mdev->next_status_jiffies));
+
+	if (check_status) {
+		mt76_tx_status_lock(mdev, &list);
+		skb = mt76_tx_status_skb_get(mdev, wcid, pid, &list);
+	}
 
 	memset(rate, 0, sizeof(*rate));
 
@@ -1447,7 +1453,8 @@ mt7915_mac_add_txs_skb(struct mt7915_dev *dev, struct mt76_wcid *wcid, int pid,
 	if (skb)
 		mt76_tx_status_skb_done(mdev, skb, &list, wcid);
 
-	mt76_tx_status_unlock(mdev, &list);
+	if (check_status)
+		mt76_tx_status_unlock(mdev, &list);
 }
 
 static void mt7915_mac_add_txs(struct mt7915_dev *dev, void *data)
@@ -1469,7 +1476,7 @@ static void mt7915_mac_add_txs(struct mt7915_dev *dev, void *data)
 	txs = le32_to_cpu(txs_data[3]);
 	pid = FIELD_GET(MT_TXS3_PID, txs);
 
-	if (pid < MT_PACKET_ID_FIRST)
+	if (pid < MT_PACKET_ID_NO_SKB)
 		return;
 
 	if (wcidx >= MT7915_WTBL_SIZE)
-- 
2.20.1

[PATCH v8 5/9] mt76: mt7915: debugfs hook to enable TXS for NO_SKB pkt-ids

[greearb]

From: Ben Greear <greearb@candelatech.com>

This lets user turn on/off this feature.  Enabling gives better
tx-rate related stats, but will cause extra driver and (maybe)
firmware work.  Not sure if it actually affects performance or
not.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 .../wireless/mediatek/mt76/mt7915/debugfs.c   | 24 +++++++++++++++++++
 .../net/wireless/mediatek/mt76/mt7915/mac.c   |  3 ++-
 .../wireless/mediatek/mt76/mt7915/mt7915.h    |  5 ++++
 3 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/debugfs.c b/drivers/net/wireless/mediatek/mt76/mt7915/debugfs.c
index 91664ac63a8d..b2df42802868 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/debugfs.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/debugfs.c
@@ -109,6 +109,29 @@ mt7915_fw_debug_get(void *data, u64 *val)
 DEFINE_DEBUGFS_ATTRIBUTE(fops_fw_debug, mt7915_fw_debug_get,
 			 mt7915_fw_debug_set, "%lld\n");
 
+static int
+mt7915_txs_for_no_skb_set(void *data, u64 val)
+{
+	struct mt7915_dev *dev = data;
+
+	dev->txs_for_no_skb_enabled = !!val;
+
+	return 0;
+}
+
+static int
+mt7915_txs_for_no_skb_get(void *data, u64 *val)
+{
+	struct mt7915_dev *dev = data;
+
+	*val = dev->txs_for_no_skb_enabled;
+
+	return 0;
+}
+
+DEFINE_DEBUGFS_ATTRIBUTE(fops_txs_for_no_skb, mt7915_txs_for_no_skb_get,
+			 mt7915_txs_for_no_skb_set, "%lld\n");
+
 static void
 mt7915_ampdu_stat_read_phy(struct mt7915_phy *phy,
 			   struct seq_file *file)
@@ -344,6 +367,7 @@ int mt7915_init_debugfs(struct mt7915_dev *dev)
 				    mt7915_queues_acq);
 	debugfs_create_file("tx_stats", 0400, dir, dev, &mt7915_tx_stats_fops);
 	debugfs_create_file("fw_debug", 0600, dir, dev, &fops_fw_debug);
+	debugfs_create_file("force_txs", 0600, dir, dev, &fops_txs_for_no_skb);
 	debugfs_create_file("implicit_txbf", 0600, dir, dev,
 			    &fops_implicit_txbf);
 	debugfs_create_u32("dfs_hw_pattern", 0400, dir, &dev->hw_pattern);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index f6ed5d5febb7..39d2175862bd 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -1047,7 +1047,8 @@ void mt7915_mac_write_txwi(struct mt7915_dev *dev, __le32 *txwi,
 	txwi[4] = 0;
 
 	val = FIELD_PREP(MT_TXD5_PID, pid);
-	if (pid >= MT_PACKET_ID_FIRST)
+	if (pid >= MT_PACKET_ID_FIRST ||
+	    (pid == MT_PACKET_ID_NO_SKB && dev->txs_for_no_skb_enabled))
 		val |= MT_TXD5_TX_STATUS_HOST;
 	txwi[5] = cpu_to_le32(val);
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h b/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
index ea6da6b4e92d..1c0a1bf91c1c 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
@@ -223,6 +223,11 @@ struct mt7915_dev {
 
 	u16 chainmask;
 	u32 hif_idx;
+	/* Should we request TXS for MT_PACKET_ID_NO_SKB?  Doing so gives better
+	 * costs but causes a great deal more TXS packet processing by driver and
+	 * creation by firmware, so may be a performance drag.
+	 */
+	bool txs_for_no_skb_enabled;
 
 	struct work_struct init_work;
 	struct work_struct rc_work;
-- 
2.20.1

[PATCH v8 6/9] mt76: mt7915: add note about TXSFM 0x2

[greearb]

From: Ben Greear <greearb@candelatech.com>

This format is not supported by any current hardware,
add note to mention this.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 drivers/net/wireless/mediatek/mt76/mt7915/mac.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.h b/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
index ccce994dc07a..65dba1061376 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
@@ -347,6 +347,10 @@ struct mt7915_tx_free {
 
 #define MT_TXS4_TIMESTAMP		GENMASK(31, 0)
 
+/* The F0 variables are for TXSFM 0x0 and 0x1.  The F1 variables
+ * are for TXSFM 0x2 aka PPDU format.  TXSFM 0x2 not supported
+ * at this time.
+ */
 #define MT_TXS5_F0_FINAL_MPDU		BIT(31)
 #define MT_TXS5_F0_QOS			BIT(30)
 #define MT_TXS5_F0_TX_COUNT		GENMASK(29, 25)
-- 
2.20.1

[PATCH v8 7/9] mt76: mt7915: txfree status to show txcount instead of latency

[greearb]

From: Ben Greear <greearb@candelatech.com>

Latency is not obviously that useful, but txcount can let us deduce
retries, which may be more interesting.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 drivers/net/wireless/mediatek/mt76/mt7915/init.c | 3 +++
 drivers/net/wireless/mediatek/mt76/mt7915/mac.h  | 8 +++++---
 drivers/net/wireless/mediatek/mt76/mt7915/regs.h | 4 ++++
 3 files changed, 12 insertions(+), 3 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/init.c b/drivers/net/wireless/mediatek/mt76/mt7915/init.c
index f174cf219724..a0d282771d77 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/init.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/init.c
@@ -304,6 +304,9 @@ static void mt7915_mac_init(struct mt7915_dev *dev)
 	/* enable hardware de-agg */
 	mt76_set(dev, MT_MDP_DCR0, MT_MDP_DCR0_DAMSDU_EN);
 
+	/* disable Tx latency report to enable Tx count in txfree path */
+	mt76_clear(dev, MT_PLE_HOST_RPT0, MT_PLE_HOST_RPT0_TX_LATENCY);
+
 	for (i = 0; i < MT7915_WTBL_SIZE; i++)
 		mt7915_mac_wtbl_update(dev, i,
 				       MT_WTBL_UPDATE_ADM_COUNT_CLEAR);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.h b/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
index 65dba1061376..3f5a80158866 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
@@ -304,13 +304,15 @@ struct mt7915_tx_free {
 
 #define MT_TX_FREE_MSDU_CNT		GENMASK(9, 0)
 #define MT_TX_FREE_WLAN_ID		GENMASK(23, 14)
+/* when configured for txfree latency mode.  See MT_PLE_HOST_RPT0_TX_LATENCY
+ * Not enabled by default now.
+ */
 #define MT_TX_FREE_LATENCY		GENMASK(12, 0)
-/* 0: success, others: dropped */
+/* when configured for txcount mode.  See MT_PLE_HOST_RPT0_TX_LATENCY. */
+#define MT_TX_FREE_TXCNT		GENMASK(12, 0)
 #define MT_TX_FREE_STATUS		GENMASK(14, 13)
 #define MT_TX_FREE_MSDU_ID		GENMASK(30, 16)
 #define MT_TX_FREE_PAIR			BIT(31)
-/* will support this field in further revision */
-#define MT_TX_FREE_RATE			GENMASK(13, 0)
 
 #define MT_TXS0_FIXED_RATE		BIT(31)
 #define MT_TXS0_BW			GENMASK(30, 29)
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/regs.h b/drivers/net/wireless/mediatek/mt76/mt7915/regs.h
index 62cc32a098fc..ac4d233b8cf2 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/regs.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/regs.h
@@ -22,6 +22,10 @@
 #define MT_PLE_BASE			0x8000
 #define MT_PLE(ofs)			(MT_PLE_BASE + (ofs))
 
+/* Modify whether txfree struct returns latency or txcount. */
+#define MT_PLE_HOST_RPT0               MT_PLE(0x030)
+#define MT_PLE_HOST_RPT0_TX_LATENCY    BIT(3)
+
 #define MT_PLE_FL_Q0_CTRL		MT_PLE(0x1b0)
 #define MT_PLE_FL_Q1_CTRL		MT_PLE(0x1b4)
 #define MT_PLE_FL_Q2_CTRL		MT_PLE(0x1b8)
-- 
2.20.1

[PATCH v8 8/9] mt76: mt7915: report tx-retries

[greearb]

From: Ben Greear <greearb@candelatech.com>

mac80211 stack will only report tx-status for skb claiming to be ampdu heads,
so lie a bit in mt7915 and set the flag so that mac80211 will record status
for each skb.

mt7915 appears to report retry status on an individual per-skb manner,
so that method above seems to work.

Re-constitute the txinfo status rate info so that the rix and flags
is also at least close to correct.  No direct way to report HE
rates that way, so mac80211 might could use some tweaking in
the ieee80211_tx_status_ext to take both info and status->rate
into account.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
 .../net/wireless/mediatek/mt76/mt7915/init.c  |  1 +
 .../net/wireless/mediatek/mt76/mt7915/mac.c   | 93 ++++++++++++++++++-
 .../net/wireless/mediatek/mt76/mt7915/mac.h   |  4 +-
 .../net/wireless/mediatek/mt76/mt7915/main.c  |  8 ++
 .../wireless/mediatek/mt76/mt7915/mt7915.h    |  4 +
 drivers/net/wireless/mediatek/mt76/tx.c       |  6 +-
 6 files changed, 109 insertions(+), 7 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/init.c b/drivers/net/wireless/mediatek/mt76/mt7915/init.c
index a0d282771d77..9dc7a67dd76f 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/init.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/init.c
@@ -217,6 +217,7 @@ mt7915_init_wiphy(struct ieee80211_hw *hw)
 	struct wiphy *wiphy = hw->wiphy;
 
 	hw->queues = 4;
+	hw->max_report_rates = 1;
 	hw->max_rx_aggregation_subframes = IEEE80211_MAX_AMPDU_BUF;
 	hw->max_tx_aggregation_subframes = IEEE80211_MAX_AMPDU_BUF;
 	hw->netdev_features = NETIF_F_RXCSUM;
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index 39d2175862bd..f5332e2c5e4e 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -1202,17 +1202,22 @@ mt7915_txp_skb_unmap(struct mt76_dev *dev, struct mt76_txwi_cache *t)
 
 static void
 mt7915_txwi_free(struct mt7915_dev *dev, struct mt76_txwi_cache *t,
-		 struct ieee80211_sta *sta, struct list_head *free_list)
+		 struct ieee80211_sta *sta, struct list_head *free_list,
+		 u32 tx_cnt, u32 tx_status, u32 ampdu)
 {
 	struct mt76_dev *mdev = &dev->mt76;
 	struct mt76_wcid *wcid;
 	__le32 *txwi;
 	u16 wcid_idx;
+	struct ieee80211_tx_info *info;
+	struct ieee80211_tx_rate *rate;
 
 	mt7915_txp_skb_unmap(mdev, t);
 	if (!t->skb)
 		goto out;
 
+	rcu_read_lock(); /* protect wcid access */
+
 	txwi = (__le32 *)mt76_get_txwi_ptr(mdev, t);
 	if (sta) {
 		wcid = (struct mt76_wcid *)sta->drv_priv;
@@ -1222,6 +1227,75 @@ mt7915_txwi_free(struct mt7915_dev *dev, struct mt76_txwi_cache *t,
 			mt7915_tx_check_aggr(sta, txwi);
 	} else {
 		wcid_idx = FIELD_GET(MT_TXD1_WLAN_IDX, le32_to_cpu(txwi[1]));
+		wcid = rcu_dereference(mdev->wcid[wcid_idx]);
+	}
+
+	info = IEEE80211_SKB_CB(t->skb);
+
+	/* Cannot clear all of info->status, we need the driver private
+	 * status intact.
+	 */
+	info->status.is_valid_ack_signal = 0;
+
+	rate = &info->status.rates[0];
+	rate->idx = -1; /* will over-write below if we found wcid */
+	info->status.rates[1].idx = -1; /* terminate rate list */
+
+	/* force TX_STAT_AMPDU to be set, or mac80211 will ignore status */
+	if (ampdu || (info->flags & IEEE80211_TX_CTL_AMPDU)) {
+		info->flags |= IEEE80211_TX_STAT_AMPDU | IEEE80211_TX_CTL_AMPDU;
+		info->status.ampdu_len = 1;
+	}
+
+	/* update info status based on cached wcid rate info since
+	 * txfree path doesn't give us a lot of info.
+	 */
+	if (wcid) {
+		struct mt7915_sta *msta = container_of(wcid, struct mt7915_sta, wcid);
+		struct mt7915_sta_stats *stats = &msta->stats;
+
+		if (wcid->rate.flags & RATE_INFO_FLAGS_MCS) {
+			rate->flags |= IEEE80211_TX_RC_MCS;
+			rate->idx = wcid->rate.mcs + wcid->rate.nss * 8;
+		} else if (wcid->rate.flags & RATE_INFO_FLAGS_VHT_MCS) {
+			rate->flags |= IEEE80211_TX_RC_VHT_MCS;
+			rate->idx = (wcid->rate.nss << 4) | wcid->rate.mcs;
+		} else if (wcid->rate.flags & RATE_INFO_FLAGS_HE_MCS) {
+			rate->idx = (wcid->rate.nss << 4) | wcid->rate.mcs;
+		} else {
+			rate->idx = wcid->rate.mcs;
+		}
+
+		switch (wcid->rate.bw) {
+		case RATE_INFO_BW_160:
+			rate->flags |= IEEE80211_TX_RC_160_MHZ_WIDTH;
+			break;
+		case RATE_INFO_BW_80:
+			rate->flags |= IEEE80211_TX_RC_80_MHZ_WIDTH;
+			break;
+		case RATE_INFO_BW_40:
+			rate->flags |= IEEE80211_TX_RC_40_MHZ_WIDTH;
+			break;
+		}
+
+		stats->tx_mpdu_attempts += tx_cnt;
+		stats->tx_mpdu_retry += tx_cnt - 1;
+
+		if (tx_status == 0)
+			stats->tx_mpdu_ok++;
+		else
+			stats->tx_mpdu_fail++;
+	}
+
+	rcu_read_unlock();
+
+	/* Apply the values that this txfree path reports */
+	rate->count = tx_cnt;
+	if (tx_status == 0) {
+		info->flags |= IEEE80211_TX_STAT_ACK;
+		info->status.ampdu_ack_len = 1;
+	} else {
+		info->flags &= ~IEEE80211_TX_STAT_ACK;
 	}
 
 	__mt76_tx_complete_skb(mdev, wcid_idx, t->skb, free_list);
@@ -1241,7 +1315,8 @@ mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 	struct ieee80211_sta *sta = NULL;
 	LIST_HEAD(free_list);
 	struct sk_buff *tmp;
-	u8 i, count;
+	u8 i;
+	u16 count;
 	bool wake = false;
 
 	/* clean DMA queues and unmap buffers first */
@@ -1257,9 +1332,12 @@ mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 	 * to the time ack is received or dropped by hw (air + hw queue time).
 	 * Should avoid accessing WTBL to get Tx airtime, and use it instead.
 	 */
+	/* free->ctrl is high u16 of first DW in the txfree struct */
 	count = FIELD_GET(MT_TX_FREE_MSDU_CNT, le16_to_cpu(free->ctrl));
 	for (i = 0; i < count; i++) {
-		u32 msdu, info = le32_to_cpu(free->info[i]);
+		u32 msdu, tx_cnt, tx_status;
+		u32 info = le32_to_cpu(free->info[i]); /* DW3+ */
+		u32 ampdu;
 
 		/*
 		 * 1'b1: new wcid pair.
@@ -1290,7 +1368,12 @@ mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 		if (!txwi)
 			continue;
 
-		mt7915_txwi_free(dev, txwi, sta, &free_list);
+		tx_cnt = FIELD_GET(MT_TX_FREE_TXCNT, info);
+		/* 0 = success, 1 dropped-by-hw, 2 dropped-by-cpu */
+		tx_status = FIELD_GET(MT_TX_FREE_STATUS, info);
+		ampdu = FIELD_GET(MT_TX_FREE_HEAD_OF_PAGE, info);
+
+		mt7915_txwi_free(dev, txwi, sta, &free_list, tx_cnt, tx_status, ampdu);
 	}
 
 	mt7915_mac_sta_poll(dev);
@@ -1817,7 +1900,7 @@ void mt7915_tx_token_put(struct mt7915_dev *dev)
 
 	spin_lock_bh(&dev->mt76.token_lock);
 	idr_for_each_entry(&dev->mt76.token, txwi, id) {
-		mt7915_txwi_free(dev, txwi, NULL, NULL);
+		mt7915_txwi_free(dev, txwi, NULL, NULL, 0, 1, 0);
 		dev->mt76.token_count--;
 	}
 	spin_unlock_bh(&dev->mt76.token_lock);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.h b/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
index 3f5a80158866..c53eca81476f 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.h
@@ -299,7 +299,7 @@ struct mt7915_tx_free {
 	__le16 ctrl;
 	u8 txd_cnt;
 	u8 rsv[3];
-	__le32 info[];
+	__le32 info[]; /* DW3+ */
 } __packed __aligned(4);
 
 #define MT_TX_FREE_MSDU_CNT		GENMASK(9, 0)
@@ -311,6 +311,8 @@ struct mt7915_tx_free {
 /* when configured for txcount mode.  See MT_PLE_HOST_RPT0_TX_LATENCY. */
 #define MT_TX_FREE_TXCNT		GENMASK(12, 0)
 #define MT_TX_FREE_STATUS		GENMASK(14, 13)
+/* 0:  not MPDU, 1:  MSDU is head pkt of TXD page (MPDU) */
+#define MT_TX_FREE_HEAD_OF_PAGE		BIT(15)
 #define MT_TX_FREE_MSDU_ID		GENMASK(30, 16)
 #define MT_TX_FREE_PAIR			BIT(31)
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/main.c b/drivers/net/wireless/mediatek/mt76/mt7915/main.c
index 4a37c4dc49e4..ff945540c460 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/main.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/main.c
@@ -1093,6 +1093,10 @@ static const char mt7915_gstrings_stats[][ETH_GSTRING_LEN] = {
 	"rx_ba_cnt",
 
 	/* per vif counters */
+	"v_tx_mpdu_attempts", /* counting any retries */
+	"v_tx_mpdu_fail",  /* frames that failed even after retry */
+	"v_tx_mpdu_retry", /* number of times frames were retried */
+	"v_tx_mpdu_ok", /* frames that succeeded, perhaps after retry */
 	"v_tx_mode_cck",
 	"v_tx_mode_ofdm",
 	"v_tx_mode_ht",
@@ -1165,6 +1169,10 @@ static void mt7915_ethtool_worker(void *wi_data, struct ieee80211_sta *sta)
 
 	wi->sta_count++;
 
+	data[ei++] += mstats->tx_mpdu_attempts;
+	data[ei++] += mstats->tx_mpdu_fail;
+	data[ei++] += mstats->tx_mpdu_retry;
+	data[ei++] += mstats->tx_mpdu_ok;
 	data[ei++] += mstats->tx_mode[MT_PHY_TYPE_CCK];
 	data[ei++] += mstats->tx_mode[MT_PHY_TYPE_OFDM];
 	data[ei++] += mstats->tx_mode[MT_PHY_TYPE_HT];
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h b/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
index 1c0a1bf91c1c..b41c6a093354 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
@@ -65,6 +65,10 @@ enum mt7915_rxq_id {
 };
 
 struct mt7915_sta_stats {
+	unsigned long tx_mpdu_attempts; /* counting any retries */
+	unsigned long tx_mpdu_fail; /* frames that failed even after retry */
+	unsigned long tx_mpdu_ok; /* frames that succeeded, perhaps after retry */
+	unsigned long tx_mpdu_retry; /* number of times frames were retried */
 	unsigned long tx_mode[MT_PHY_TYPE_HE_LAST]; /* See mt76_phy_type */
 	unsigned long tx_bw[4]; /* 20, 40, 80, 160 */
 	unsigned long tx_nss[4]; /* 1, 2, 3, 4 */
diff --git a/drivers/net/wireless/mediatek/mt76/tx.c b/drivers/net/wireless/mediatek/mt76/tx.c
index 7b281cd062c9..9fec9613c67f 100644
--- a/drivers/net/wireless/mediatek/mt76/tx.c
+++ b/drivers/net/wireless/mediatek/mt76/tx.c
@@ -276,6 +276,7 @@ void __mt76_tx_complete_skb(struct mt76_dev *dev, u16 wcid_idx, struct sk_buff *
 	struct ieee80211_tx_status status = {
 		.skb = skb,
 		.free_list = free_list,
+		.info = IEEE80211_SKB_CB(skb),
 	};
 	struct mt76_wcid *wcid = NULL;
 	struct ieee80211_hw *hw;
@@ -283,8 +284,11 @@ void __mt76_tx_complete_skb(struct mt76_dev *dev, u16 wcid_idx, struct sk_buff *
 
 	rcu_read_lock();
 
-	if (wcid_idx < ARRAY_SIZE(dev->wcid))
+	if (wcid_idx < ARRAY_SIZE(dev->wcid)) {
 		wcid = rcu_dereference(dev->wcid[wcid_idx]);
+		if (wcid)
+			status.rate = &wcid->rate;
+	}
 
 	mt76_tx_check_non_aql(dev, wcid, skb);
 
-- 
2.20.1

[PATCH v8 9/9] mt76: mt7915: add a missing HT flag for GI parsing

[greearb]

From: Ryder Lee <ryder.lee@mediatek.com>

The previous commit missed a HT flag.

Fixes: 14b220c07188 ("mt76: mt7915: report tx rate directly from tx status")
Signed-off-by: Ryder Lee <ryder.lee@mediatek.com>
---
 drivers/net/wireless/mediatek/mt76/mt7915/mac.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index f5332e2c5e4e..10f4001d8d92 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -206,7 +206,8 @@ static void mt7915_mac_sta_poll(struct mt7915_dev *dev)
 
 			rate->he_gi = (val & (0x3 << offs)) >> offs;
 			msta->wcid.rate_he_gi = rate->he_gi; /* cache for later */
-		} else if (rate->flags & RATE_INFO_FLAGS_VHT_MCS) {
+		} else if (rate->flags &
+			   (RATE_INFO_FLAGS_VHT_MCS | RATE_INFO_FLAGS_MCS)) {
 			if (val & BIT(12 + bw)) {
 				rate->flags |= RATE_INFO_FLAGS_SHORT_GI;
 				msta->wcid.rate_short_gi = 1;
@@ -215,7 +216,6 @@ static void mt7915_mac_sta_poll(struct mt7915_dev *dev)
 				msta->wcid.rate_short_gi = 0;
 			}
 		}
-		/* TODO:  Deal with HT_MCS */
 	}
 
 	rcu_read_unlock();
-- 
2.20.1