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>

debugfs.c

@@ -31,15 +31,14 @@ int mt76_queues_read(struct seq_file *s, void *data)
 	int i;
 
 	for (i = 0; i < ARRAY_SIZE(dev->q_tx); i++) {
-		struct mt76_sw_queue *q = &dev->q_tx[i];
+		struct mt76_queue *q = dev->q_tx[i].q;
 
-		if (!q->q)
+		if (!q)
 			continue;
 
 		seq_printf(s,
-			   "%d:	queued=%d head=%d tail=%d swq_queued=%d\n",
-			   i, q->q->queued, q->q->head, q->q->tail,
-			   q->swq_queued);
+			   "%d:	queued=%d head=%d tail=%d\n",
+			   i, q->queued, q->head, q->tail);
 	}
 
 	return 0;

mt76.h

@@ -22,6 +22,9 @@
 #define MT_RX_BUF_SIZE      2048
 #define MT_SKB_HEAD_LEN     128
 
+#define MT_MAX_NON_AQL_PKT  16
+#define MT_TXQ_FREE_THR     32
+
 struct mt76_dev;
 struct mt76_phy;
 struct mt76_wcid;
@@ -100,10 +103,9 @@ struct mt76_queue_entry {
 		struct urb *urb;
 		int buf_sz;
 	};
-	enum mt76_txq_id qid;
+	u16 wcid;
 	bool skip_buf0:1;
 	bool skip_buf1:1;
-	bool schedule:1;
 	bool done:1;
 };
 
@@ -139,8 +141,6 @@ struct mt76_queue {
 
 struct mt76_sw_queue {
 	struct mt76_queue *q;
-
-	int swq_queued;
 };
 
 struct mt76_mcu_ops {
@@ -206,6 +206,7 @@ DECLARE_EWMA(signal, 10, 8);
 struct mt76_wcid {
 	struct mt76_rx_tid __rcu *aggr[IEEE80211_NUM_TIDS];
 
+	atomic_t non_aql_packets;
 	unsigned long flags;
 
 	struct ewma_signal rssi;
@@ -953,7 +954,7 @@ struct sk_buff *mt76_tx_status_skb_get(struct mt76_dev *dev,
 				       struct sk_buff_head *list);
 void mt76_tx_status_skb_done(struct mt76_dev *dev, struct sk_buff *skb,
 			     struct sk_buff_head *list);
-void mt76_tx_complete_skb(struct mt76_dev *dev, struct sk_buff *skb);
+void mt76_tx_complete_skb(struct mt76_dev *dev, u16 wcid, struct sk_buff *skb);
 void mt76_tx_status_check(struct mt76_dev *dev, struct mt76_wcid *wcid,
 			  bool flush);
 int mt76_sta_state(struct ieee80211_hw *hw, struct ieee80211_vif *vif,

mt7603/mac.c

@@ -1282,7 +1282,7 @@ void mt7603_tx_complete_skb(struct mt76_dev *mdev, struct mt76_queue_entry *e)
 	}
 
 	dev->tx_hang_check = 0;
-	mt76_tx_complete_skb(mdev, skb);
+	mt76_tx_complete_skb(mdev, e->wcid, skb);
 }
 
 static bool

mt7615/mac.c

@@ -1400,6 +1400,9 @@ mt7615_mac_tx_free_token(struct mt7615_dev *dev, u16 token)
 {
 	struct mt76_dev *mdev = &dev->mt76;
 	struct mt76_txwi_cache *txwi;
+	__le32 *txwi_data;
+	u32 val;
+	u8 wcid;
 
 	trace_mac_tx_free(dev, token);
 
@@ -1410,9 +1413,13 @@ mt7615_mac_tx_free_token(struct mt7615_dev *dev, u16 token)
 	if (!txwi)
 		return;
 
+	txwi_data = (__le32 *)mt76_get_txwi_ptr(mdev, txwi);
+	val = le32_to_cpu(txwi_data[1]);
+	wcid = FIELD_GET(MT_TXD1_WLAN_IDX, val);
+
 	mt7615_txp_skb_unmap(mdev, txwi);
 	if (txwi->skb) {
-		mt76_tx_complete_skb(mdev, txwi->skb);
+		mt76_tx_complete_skb(mdev, wcid, txwi->skb);
 		txwi->skb = NULL;
 	}
 

mt7615/pci_mac.c

@@ -44,7 +44,7 @@ void mt7615_tx_complete_skb(struct mt76_dev *mdev, struct mt76_queue_entry *e)
 	}
 
 	if (e->skb)
-		mt76_tx_complete_skb(mdev, e->skb);
+		mt76_tx_complete_skb(mdev, e->wcid, e->skb);
 }
 
 static void

mt7615/usb_sdio.c

@@ -234,7 +234,7 @@ void mt7663_usb_sdio_tx_complete_skb(struct mt76_dev *mdev,
 		headroom += MT_USB_HDR_SIZE;
 	skb_pull(e->skb, headroom);
 
-	mt76_tx_complete_skb(mdev, e->skb);
+	mt76_tx_complete_skb(mdev, e->wcid, e->skb);
 }
 EXPORT_SYMBOL_GPL(mt7663_usb_sdio_tx_complete_skb);
 

mt76x02_mac.c

@@ -917,7 +917,7 @@ void mt76x02_tx_complete_skb(struct mt76_dev *mdev, struct mt76_queue_entry *e)
 	txwi = (struct mt76x02_txwi *)txwi_ptr;
 	trace_mac_txdone(mdev, txwi->wcid, txwi->pktid);
 
-	mt76_tx_complete_skb(mdev, e->skb);
+	mt76_tx_complete_skb(mdev, e->wcid, e->skb);
 }
 EXPORT_SYMBOL_GPL(mt76x02_tx_complete_skb);
 

mt76x02_usb_core.c

@@ -18,7 +18,7 @@ static void mt76x02u_remove_dma_hdr(struct sk_buff *skb)
 void mt76x02u_tx_complete_skb(struct mt76_dev *mdev, struct mt76_queue_entry *e)
 {
 	mt76x02u_remove_dma_hdr(e->skb);
-	mt76_tx_complete_skb(mdev, e->skb);
+	mt76_tx_complete_skb(mdev, e->wcid, e->skb);
 }
 EXPORT_SYMBOL_GPL(mt76x02u_tx_complete_skb);
 

mt7915/mac.c

@@ -997,10 +997,21 @@ void mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 
 		mt7915_txp_skb_unmap(mdev, txwi);
 		if (txwi->skb) {
+			struct ieee80211_tx_info *info = IEEE80211_SKB_CB(txwi->skb);
 			void *txwi_ptr = mt76_get_txwi_ptr(mdev, txwi);
 
 			if (likely(txwi->skb->protocol != cpu_to_be16(ETH_P_PAE)))
 				mt7915_tx_check_aggr(sta, txwi_ptr);
+
+			if (sta && !info->tx_time_est) {
+				struct mt76_wcid *wcid = (struct mt76_wcid *)sta->drv_priv;
+				int pending;
+
+				pending = atomic_dec_return(&wcid->non_aql_packets);
+				if (pending < 0)
+					atomic_cmpxchg(&wcid->non_aql_packets, pending, 0);
+			}
+
 			mt7915_tx_complete_status(mdev, txwi->skb, sta, stat);
 			txwi->skb = NULL;
 		}

sdio.c

@@ -144,7 +144,6 @@ static void mt76s_process_tx_queue(struct mt76_dev *dev, enum mt76_txq_id qid)
 
 		entry = q->entry[q->tail];
 		q->entry[q->tail].done = false;
-		q->entry[q->tail].schedule = false;
 
 		if (qid == MT_TXQ_MCU) {
 			dev_kfree_skb(entry.skb);