mt76: rely on AQL for burst size limits on tx queueing

Now that AQL works well on all mt76 drivers, completely replace the arbitrary
burst sizing and number of bursts logic for tx scheduling.
For the short period of time in which AQL does not work well yet, limit each
stations to 16 in-flight packets that have no estimated tx time.
This should avoid filling the queue if a station connects and queues up a
large number of packets before rate control information is available, especially
with hardware rate control

Signed-off-by: Felix Fietkau <nbd@nbd.name>

1 files changed, 77 insertions, 55 deletions

diff --git a/drivers/net/wireless/mediatek/mt76/tx.c b/drivers/net/wireless/mediatek/mt76/tx.c
index 3dac8f7dbff6..702fbfd2e1b5 100644
--- a/drivers/net/wireless/mediatek/mt76/tx.c
+++ b/drivers/net/wireless/mediatek/mt76/tx.c
@@ -241,7 +241,32 @@ mt76_tx_status_check(struct mt76_dev *dev, struct mt76_wcid *wcid, bool flush)
 }
 EXPORT_SYMBOL_GPL(mt76_tx_status_check);
 
-void mt76_tx_complete_skb(struct mt76_dev *dev, struct sk_buff *skb)
+static void
+mt76_tx_check_non_aql(struct mt76_dev *dev, u16 wcid_idx, struct sk_buff *skb)
+{
+	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+	struct mt76_wcid *wcid;
+	int pending;
+
+	if (info->tx_time_est)
+		return;
+
+	if (wcid_idx >= ARRAY_SIZE(dev->wcid))
+		return;
+
+	rcu_read_lock();
+
+	wcid = rcu_dereference(dev->wcid[wcid_idx]);
+	if (wcid) {
+		pending = atomic_dec_return(&wcid->non_aql_packets);
+		if (pending < 0)
+			atomic_cmpxchg(&wcid->non_aql_packets, pending, 0);
+	}
+
+	rcu_read_unlock();
+}
+
+void mt76_tx_complete_skb(struct mt76_dev *dev, u16 wcid_idx, struct sk_buff *skb)
 {
 	struct ieee80211_hw *hw;
 	struct sk_buff_head list;
@@ -254,6 +279,8 @@ void mt76_tx_complete_skb(struct mt76_dev *dev, struct sk_buff *skb)
 	}
 #endif
 
+	mt76_tx_check_non_aql(dev, wcid_idx, skb);
+
 	if (!skb->prev) {
 		hw = mt76_tx_status_get_hw(dev, skb);
 		ieee80211_free_txskb(hw, skb);
@@ -266,6 +293,32 @@ void mt76_tx_complete_skb(struct mt76_dev *dev, struct sk_buff *skb)
 }
 EXPORT_SYMBOL_GPL(mt76_tx_complete_skb);
 
+static int
+__mt76_tx_queue_skb(struct mt76_dev *dev, int qid, struct sk_buff *skb,
+		    struct mt76_wcid *wcid, struct ieee80211_sta *sta,
+		    bool *stop)
+{
+	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
+	struct mt76_queue *q;
+	bool non_aql;
+	int pending;
+	int idx;
+
+	non_aql = !info->tx_time_est;
+	idx = dev->queue_ops->tx_queue_skb(dev, qid, skb, wcid, sta);
+	if (idx < 0 || !sta || !non_aql)
+		return idx;
+
+	wcid = (struct mt76_wcid *)sta->drv_priv;
+	q = dev->q_tx[qid].q;
+	q->entry[idx].wcid = wcid->idx;
+	pending = atomic_inc_return(&wcid->non_aql_packets);
+	if (stop && pending >= MT_MAX_NON_AQL_PKT)
+		*stop = true;
+
+	return idx;
+}
+
 void
 mt76_tx(struct mt76_phy *phy, struct ieee80211_sta *sta,
 	struct mt76_wcid *wcid, struct sk_buff *skb)
@@ -304,7 +357,7 @@ mt76_tx(struct mt76_phy *phy, struct ieee80211_sta *sta,
 	q = dev->q_tx[qid].q;
 
 	spin_lock_bh(&q->lock);
-	dev->queue_ops->tx_queue_skb(dev, qid, skb, wcid, sta);
+	__mt76_tx_queue_skb(dev, qid, skb, wcid, sta, NULL);
 	dev->queue_ops->kick(dev, q);
 
 	if (q->queued > q->ndesc - 8 && !q->stopped) {
@@ -358,7 +411,7 @@ mt76_queue_ps_skb(struct mt76_dev *dev, struct ieee80211_sta *sta,
 			       IEEE80211_TX_CTL_REQ_TX_STATUS;
 
 	mt76_skb_set_moredata(skb, !last);
-	dev->queue_ops->tx_queue_skb(dev, MT_TXQ_PSD, skb, wcid, sta);
+	__mt76_tx_queue_skb(dev, MT_TXQ_PSD, skb, wcid, sta, NULL);
 }
 
 void
@@ -417,15 +470,16 @@ mt76_txq_send_burst(struct mt76_phy *phy, struct mt76_sw_queue *sq,
 	struct mt76_queue *hwq = sq->q;
 	struct ieee80211_tx_info *info;
 	struct sk_buff *skb;
-	int n_frames = 1, limit;
-	struct ieee80211_tx_rate tx_rate;
-	bool ampdu;
-	bool probe;
+	int n_frames = 1;
+	bool stop = false;
 	int idx;
 
 	if (test_bit(MT_WCID_FLAG_PS, &wcid->flags))
 		return 0;
 
+	if (atomic_read(&wcid->non_aql_packets) >= MT_MAX_NON_AQL_PKT)
+		return 0;
+
 	skb = mt76_txq_dequeue(phy, mtxq, false);
 	if (!skb)
 		return 0;
@@ -434,55 +488,37 @@ mt76_txq_send_burst(struct mt76_phy *phy, struct mt76_sw_queue *sq,
 	if (!(wcid->tx_info & MT_WCID_TX_INFO_SET))
 		ieee80211_get_tx_rates(txq->vif, txq->sta, skb,
 				       info->control.rates, 1);
-	tx_rate = info->control.rates[0];
-
-	probe = (info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE);
-	ampdu = IEEE80211_SKB_CB(skb)->flags & IEEE80211_TX_CTL_AMPDU;
-	limit = ampdu ? 16 : 3;
-
-	idx = dev->queue_ops->tx_queue_skb(dev, qid, skb, wcid, txq->sta);
 
+	idx = __mt76_tx_queue_skb(dev, qid, skb, wcid, txq->sta, &stop);
 	if (idx < 0)
 		return idx;
 
 	do {
-		bool cur_ampdu;
-
-		if (probe)
-			break;
-
 		if (test_bit(MT76_STATE_PM, &phy->state) ||
 		    test_bit(MT76_RESET, &phy->state))
 			return -EBUSY;
 
-		skb = mt76_txq_dequeue(phy, mtxq, false);
-		if (!skb)
+		if (stop)
 			break;
 
-		info = IEEE80211_SKB_CB(skb);
-		cur_ampdu = info->flags & IEEE80211_TX_CTL_AMPDU;
+		if (q->queued + MT_TXQ_FREE_THR >= q->ndesc)
+			break;
 
-		if (ampdu != cur_ampdu ||
-		    (info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE)) {
-			skb_queue_tail(&mtxq->retry_q, skb);
+		skb = mt76_txq_dequeue(phy, mtxq, false);
+		if (!skb)
 			break;
-		}
 
-		info->control.rates[0] = tx_rate;
+		info = IEEE80211_SKB_CB(skb);
+		if (!(wcid->tx_info & MT_WCID_TX_INFO_SET))
+			ieee80211_get_tx_rates(txq->vif, txq->sta, skb,
+					       info->control.rates, 1);
 
-		idx = dev->queue_ops->tx_queue_skb(dev, qid, skb, wcid,
-						   txq->sta);
+		idx = __mt76_tx_queue_skb(dev, qid, skb, wcid, txq->sta, &stop);
 		if (idx < 0)
-			return idx;
+			break;
 
 		n_frames++;
-	} while (n_frames < limit);
-
-	if (!probe) {
-		hwq->entry[idx].qid = sq - dev->q_tx;
-		hwq->entry[idx].schedule = true;
-		sq->swq_queued++;
-	}
+	} while (1);
 
 	dev->queue_ops->kick(dev, hwq);
 
@@ -502,15 +538,15 @@ mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 
 	spin_lock_bh(&hwq->lock);
 	while (1) {
-		if (sq->swq_queued >= 4)
-			break;
-
 		if (test_bit(MT76_STATE_PM, &phy->state) ||
 		    test_bit(MT76_RESET, &phy->state)) {
 			ret = -EBUSY;
 			break;
 		}
 
+		if (q->queued + MT_TXQ_FREE_THR >= q->ndesc)
+			break;
+
 		txq = ieee80211_next_txq(phy->hw, qid);
 		if (!txq)
 			break;
@@ -544,16 +580,11 @@ mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 
 void mt76_txq_schedule(struct mt76_phy *phy, enum mt76_txq_id qid)
 {
-	struct mt76_dev *dev = phy->dev;
-	struct mt76_sw_queue *sq = &dev->q_tx[qid];
 	int len;
 
 	if (qid >= 4)
 		return;
 
-	if (sq->swq_queued >= 4)
-		return;
-
 	rcu_read_lock();
 
 	do {
@@ -700,21 +731,12 @@ EXPORT_SYMBOL_GPL(mt76_skb_adjust_pad);
 void mt76_queue_tx_complete(struct mt76_dev *dev, struct mt76_queue *q,
 			    struct mt76_queue_entry *e)
 {
-	enum mt76_txq_id qid = e->qid % 4;
-	bool ext_phy = e->qid >= 4;
-
 	if (e->skb)
 		dev->drv->tx_complete_skb(dev, e);
 
 	spin_lock_bh(&q->lock);
 	q->tail = (q->tail + 1) % q->ndesc;
 	q->queued--;
-
-	if (ext_phy)
-		qid += __MT_TXQ_MAX;
-
-	if (e->schedule)
-		dev->q_tx[qid].swq_queued--;
 	spin_unlock_bh(&q->lock);
 }
 EXPORT_SYMBOL_GPL(mt76_queue_tx_complete);