mt76: improve tx queue stop/wake

Instead of stopping and waking only a single queue, handle all phy tx queues
mapped ot the same hardware queue.
Also allow the driver to block tx queues

Signed-off-by: Felix Fietkau <[email protected]>

dma.c

@@ -220,7 +220,6 @@ static void
 mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
 {
 	struct mt76_queue_entry entry;
-	bool wake = false;
 	int last;
 
 	if (!q)
@@ -238,7 +237,6 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
 		if (entry.txwi) {
 			if (!(dev->drv->drv_flags & MT_DRV_TXWI_NO_FREE))
 				mt76_put_txwi(dev, entry.txwi);
-			wake = !flush;
 		}
 
 		if (!flush && q->tail == last)
@@ -253,16 +251,8 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
 		spin_unlock_bh(&q->lock);
 	}
 
-	wake = wake && q->stopped &&
-	       q->qid < IEEE80211_NUM_ACS && q->queued < q->ndesc - 8;
-	if (wake)
-		q->stopped = false;
-
 	if (!q->queued)
 		wake_up(&dev->tx_wait);
-
-	if (wake)
-		ieee80211_wake_queue(dev->hw, q->qid);
 }
 
 static void *

mt76.h

@@ -136,6 +136,7 @@ struct mt76_queue {
 	int queued;
 	int buf_size;
 	bool stopped;
+	bool blocked;
 
 	u8 buf_offset;
 	u8 hw_idx;

tx.c

@@ -291,12 +291,6 @@ mt76_tx(struct mt76_phy *phy, struct ieee80211_sta *sta,
 	spin_lock_bh(&q->lock);
 	__mt76_tx_queue_skb(phy, qid, skb, wcid, sta, NULL);
 	dev->queue_ops->kick(dev, q);
-
-	if (q->queued > q->ndesc - 8 && !q->stopped) {
-		ieee80211_stop_queue(phy->hw, skb_get_queue_mapping(skb));
-		q->stopped = true;
-	}
-
 	spin_unlock_bh(&q->lock);
 }
 EXPORT_SYMBOL_GPL(mt76_tx);
@@ -463,6 +457,9 @@ mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 			break;
 		}
 
+		if (q->stopped || q->blocked)
+			break;
+
 		if (q->queued + MT_TXQ_FREE_THR >= q->ndesc)
 			break;
 
@@ -496,13 +493,60 @@ mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 	return ret;
 }
 
+
+static void
+__mt76_tx_check_hwq_stop(struct mt76_phy *phy, struct mt76_queue *q,
+			 bool stopped)
+{
+	int i;
+
+	for (i = 0; i < IEEE80211_NUM_ACS; i++) {
+		if (phy->q_tx[i] != q)
+			continue;
+
+		if (stopped)
+			ieee80211_stop_queue(phy->hw, i);
+		else
+			ieee80211_wake_queue(phy->hw, i);
+	}
+}
+
+static bool
+mt76_tx_check_hwq_stop(struct mt76_dev *dev, struct mt76_queue *q)
+{
+	 return q->blocked || q->queued >= q->ndesc - MT_TXQ_FREE_THR;
+}
+
+static void
+mt76_tx_update_hwq_stop(struct mt76_dev *dev, struct mt76_queue *q)
+{
+	bool stopped, prev_stopped;
+
+	stopped = mt76_tx_check_hwq_stop(dev, q);
+	prev_stopped = q->stopped;
+	q->stopped = stopped;
+
+	if (stopped == prev_stopped)
+		return;
+
+	__mt76_tx_check_hwq_stop(&dev->phy, q, stopped);
+	if (dev->phy2)
+		__mt76_tx_check_hwq_stop(dev->phy2, q, stopped);
+}
+
 void mt76_txq_schedule(struct mt76_phy *phy, enum mt76_txq_id qid)
 {
+	struct mt76_queue *q;
 	int len;
 
 	if (qid >= 4)
 		return;
 
+	q = phy->q_tx[qid];
+	spin_lock_bh(&q->lock);
+	mt76_tx_update_hwq_stop(phy->dev, q);
+	spin_unlock_bh(&q->lock);
+
 	rcu_read_lock();
 
 	do {
@@ -623,6 +667,10 @@ void mt76_queue_tx_complete(struct mt76_dev *dev, struct mt76_queue *q,
 	spin_lock_bh(&q->lock);
 	q->tail = (q->tail + 1) % q->ndesc;
 	q->queued--;
+
+	if (q->stopped && !mt76_tx_check_hwq_stop(dev, q))
+		mt76_worker_schedule(&dev->tx_worker);
+
 	spin_unlock_bh(&q->lock);
 }
 EXPORT_SYMBOL_GPL(mt76_queue_tx_complete);