mac80211: skip netdev queue control with software queuing

Qdiscs are designed with no regard to 802.11
aggregation requirements and hand out
packet-by-packet with no guarantee they are
destined to the same tid. This does more bad than
good no matter how fairly a given qdisc may behave
on an ethernet interface.

Software queuing used per-AC netdev subqueue
congestion control whenever a global AC limit was
hit. This meant in practice a single station or
tid queue could starve others rather easily. This
could resonate with qdiscs in a bad way or could
just end up with poor aggregation performance.
Increasing the AC limit would increase induced
latency which is also bad.

Disabling qdiscs by default and performing
taildrop instead of netdev subqueue congestion
control on the other hand makes it possible for
tid queues to fill up "in the meantime" while
preventing stations starving each other.

This increases aggregation opportunities and
should allow software queuing based drivers
achieve better performance by utilizing airtime
more efficiently with big aggregates.

Signed-off-by: Michal Kazior <michal.kazior@tieto.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
diff --git a/include/net/mac80211.h b/include/net/mac80211.h
index be30b05..a8683ae 100644
--- a/include/net/mac80211.h
+++ b/include/net/mac80211.h
@@ -2147,9 +2147,6 @@
  * @n_cipher_schemes: a size of an array of cipher schemes definitions.
  * @cipher_schemes: a pointer to an array of cipher scheme definitions
  *	supported by HW.
- *
- * @txq_ac_max_pending: maximum number of frames per AC pending in all txq
- *	entries for a vif.
  */
 struct ieee80211_hw {
 	struct ieee80211_conf conf;
@@ -2180,7 +2177,6 @@
 	u8 uapsd_max_sp_len;
 	u8 n_cipher_schemes;
 	const struct ieee80211_cipher_scheme *cipher_schemes;
-	int txq_ac_max_pending;
 };
 
 static inline bool _ieee80211_hw_check(struct ieee80211_hw *hw,
diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h
index 9438c94..6346033 100644
--- a/net/mac80211/ieee80211_i.h
+++ b/net/mac80211/ieee80211_i.h
@@ -856,7 +856,7 @@
 	bool control_port_no_encrypt;
 	int encrypt_headroom;
 
-	atomic_t txqs_len[IEEE80211_NUM_ACS];
+	atomic_t num_tx_queued;
 	struct ieee80211_tx_queue_params tx_conf[IEEE80211_NUM_ACS];
 	struct mac80211_qos_map __rcu *qos_map;
 
diff --git a/net/mac80211/iface.c b/net/mac80211/iface.c
index c59af3e..609c517 100644
--- a/net/mac80211/iface.c
+++ b/net/mac80211/iface.c
@@ -976,13 +976,13 @@
 
 	if (sdata->vif.txq) {
 		struct txq_info *txqi = to_txq_info(sdata->vif.txq);
+		int n = skb_queue_len(&txqi->queue);
 
 		spin_lock_bh(&txqi->queue.lock);
 		ieee80211_purge_tx_queue(&local->hw, &txqi->queue);
+		atomic_sub(n, &sdata->num_tx_queued);
 		txqi->byte_cnt = 0;
 		spin_unlock_bh(&txqi->queue.lock);
-
-		atomic_set(&sdata->txqs_len[txqi->txq.ac], 0);
 	}
 
 	if (local->open_count == 0)
@@ -1198,6 +1198,12 @@
 	dev->destructor = ieee80211_if_free;
 }
 
+static void ieee80211_if_setup_no_queue(struct net_device *dev)
+{
+	ieee80211_if_setup(dev);
+	dev->priv_flags |= IFF_NO_QUEUE;
+}
+
 static void ieee80211_iface_work(struct work_struct *work)
 {
 	struct ieee80211_sub_if_data *sdata =
@@ -1707,6 +1713,7 @@
 	struct net_device *ndev = NULL;
 	struct ieee80211_sub_if_data *sdata = NULL;
 	struct txq_info *txqi;
+	void (*if_setup)(struct net_device *dev);
 	int ret, i;
 	int txqs = 1;
 
@@ -1734,12 +1741,17 @@
 			txq_size += sizeof(struct txq_info) +
 				    local->hw.txq_data_size;
 
+		if (local->ops->wake_tx_queue)
+			if_setup = ieee80211_if_setup_no_queue;
+		else
+			if_setup = ieee80211_if_setup;
+
 		if (local->hw.queues >= IEEE80211_NUM_ACS)
 			txqs = IEEE80211_NUM_ACS;
 
 		ndev = alloc_netdev_mqs(size + txq_size,
 					name, name_assign_type,
-					ieee80211_if_setup, txqs, 1);
+					if_setup, txqs, 1);
 		if (!ndev)
 			return -ENOMEM;
 		dev_net_set(ndev, wiphy_net(local->hw.wiphy));
diff --git a/net/mac80211/main.c b/net/mac80211/main.c
index 7ee91d6..160ac6b 100644
--- a/net/mac80211/main.c
+++ b/net/mac80211/main.c
@@ -1055,9 +1055,6 @@
 
 	local->dynamic_ps_forced_timeout = -1;
 
-	if (!local->hw.txq_ac_max_pending)
-		local->hw.txq_ac_max_pending = 64;
-
 	result = ieee80211_wep_init(local);
 	if (result < 0)
 		wiphy_debug(local->hw.wiphy, "Failed to initialize wep: %d\n",
diff --git a/net/mac80211/sta_info.c b/net/mac80211/sta_info.c
index 5ccfdbd..177cc6c 100644
--- a/net/mac80211/sta_info.c
+++ b/net/mac80211/sta_info.c
@@ -116,7 +116,7 @@
 			int n = skb_queue_len(&txqi->queue);
 
 			ieee80211_purge_tx_queue(&local->hw, &txqi->queue);
-			atomic_sub(n, &sdata->txqs_len[txqi->txq.ac]);
+			atomic_sub(n, &sdata->num_tx_queued);
 			txqi->byte_cnt = 0;
 		}
 	}
diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
index 2030443..3e77da1 100644
--- a/net/mac80211/tx.c
+++ b/net/mac80211/tx.c
@@ -1236,27 +1236,21 @@
 	return TX_CONTINUE;
 }
 
-static void ieee80211_drv_tx(struct ieee80211_local *local,
-			     struct ieee80211_vif *vif,
-			     struct ieee80211_sta *pubsta,
-			     struct sk_buff *skb)
+static struct txq_info *ieee80211_get_txq(struct ieee80211_local *local,
+					  struct ieee80211_vif *vif,
+					  struct ieee80211_sta *pubsta,
+					  struct sk_buff *skb)
 {
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data;
-	struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif);
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
-	struct ieee80211_tx_control control = {
-		.sta = pubsta,
-	};
 	struct ieee80211_txq *txq = NULL;
-	struct txq_info *txqi;
-	u8 ac;
 
 	if ((info->flags & IEEE80211_TX_CTL_SEND_AFTER_DTIM) ||
 	    (info->control.flags & IEEE80211_TX_CTRL_PS_RESPONSE))
-		goto tx_normal;
+		return NULL;
 
 	if (!ieee80211_is_data(hdr->frame_control))
-		goto tx_normal;
+		return NULL;
 
 	if (pubsta) {
 		u8 tid = skb->priority & IEEE80211_QOS_CTL_TID_MASK;
@@ -1267,25 +1261,28 @@
 	}
 
 	if (!txq)
-		goto tx_normal;
+		return NULL;
 
-	ac = txq->ac;
-	txqi = to_txq_info(txq);
-	atomic_inc(&sdata->txqs_len[ac]);
-	if (atomic_read(&sdata->txqs_len[ac]) >= local->hw.txq_ac_max_pending)
-		netif_stop_subqueue(sdata->dev, ac);
+	return to_txq_info(txq);
+}
 
-	spin_lock_bh(&txqi->queue.lock);
+static void ieee80211_txq_enqueue(struct ieee80211_local *local,
+				  struct txq_info *txqi,
+				  struct sk_buff *skb)
+{
+	struct ieee80211_sub_if_data *sdata = vif_to_sdata(txqi->txq.vif);
+
+	lockdep_assert_held(&txqi->queue.lock);
+
+	if (atomic_read(&sdata->num_tx_queued) >= TOTAL_MAX_TX_BUFFER ||
+	    txqi->queue.qlen >= STA_MAX_TX_BUFFER) {
+		ieee80211_free_txskb(&local->hw, skb);
+		return;
+	}
+
+	atomic_inc(&sdata->num_tx_queued);
 	txqi->byte_cnt += skb->len;
 	__skb_queue_tail(&txqi->queue, skb);
-	spin_unlock_bh(&txqi->queue.lock);
-
-	drv_wake_tx_queue(local, txqi);
-
-	return;
-
-tx_normal:
-	drv_tx(local, &control, skb);
 }
 
 struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,
@@ -1296,7 +1293,6 @@
 	struct txq_info *txqi = container_of(txq, struct txq_info, txq);
 	struct ieee80211_hdr *hdr;
 	struct sk_buff *skb = NULL;
-	u8 ac = txq->ac;
 
 	spin_lock_bh(&txqi->queue.lock);
 
@@ -1307,12 +1303,9 @@
 	if (!skb)
 		goto out;
 
+	atomic_dec(&sdata->num_tx_queued);
 	txqi->byte_cnt -= skb->len;
 
-	atomic_dec(&sdata->txqs_len[ac]);
-	if (__netif_subqueue_stopped(sdata->dev, ac))
-		ieee80211_propagate_queue_wake(local, sdata->vif.hw_queue[ac]);
-
 	hdr = (struct ieee80211_hdr *)skb->data;
 	if (txq->sta && ieee80211_is_data_qos(hdr->frame_control)) {
 		struct sta_info *sta = container_of(txq->sta, struct sta_info,
@@ -1343,7 +1336,9 @@
 			       struct sk_buff_head *skbs,
 			       bool txpending)
 {
+	struct ieee80211_tx_control control = {};
 	struct sk_buff *skb, *tmp;
+	struct txq_info *txqi;
 	unsigned long flags;
 
 	skb_queue_walk_safe(skbs, skb, tmp) {
@@ -1358,6 +1353,21 @@
 		}
 #endif
 
+		txqi = ieee80211_get_txq(local, vif, sta, skb);
+		if (txqi) {
+			info->control.vif = vif;
+
+			__skb_unlink(skb, skbs);
+
+			spin_lock_bh(&txqi->queue.lock);
+			ieee80211_txq_enqueue(local, txqi, skb);
+			spin_unlock_bh(&txqi->queue.lock);
+
+			drv_wake_tx_queue(local, txqi);
+
+			continue;
+		}
+
 		spin_lock_irqsave(&local->queue_stop_reason_lock, flags);
 		if (local->queue_stop_reasons[q] ||
 		    (!txpending && !skb_queue_empty(&local->pending[q]))) {
@@ -1400,9 +1410,10 @@
 		spin_unlock_irqrestore(&local->queue_stop_reason_lock, flags);
 
 		info->control.vif = vif;
+		control.sta = sta;
 
 		__skb_unlink(skb, skbs);
-		ieee80211_drv_tx(local, vif, sta, skb);
+		drv_tx(local, &control, skb);
 	}
 
 	return true;
diff --git a/net/mac80211/util.c b/net/mac80211/util.c
index 905003f..0db4644 100644
--- a/net/mac80211/util.c
+++ b/net/mac80211/util.c
@@ -244,6 +244,9 @@
 	struct ieee80211_sub_if_data *sdata;
 	int n_acs = IEEE80211_NUM_ACS;
 
+	if (local->ops->wake_tx_queue)
+		return;
+
 	if (local->hw.queues < IEEE80211_NUM_ACS)
 		n_acs = 1;
 
@@ -260,11 +263,6 @@
 		for (ac = 0; ac < n_acs; ac++) {
 			int ac_queue = sdata->vif.hw_queue[ac];
 
-			if (local->ops->wake_tx_queue &&
-			    (atomic_read(&sdata->txqs_len[ac]) >
-			     local->hw.txq_ac_max_pending))
-				continue;
-
 			if (ac_queue == queue ||
 			    (sdata->vif.cab_queue == queue &&
 			     local->queue_stop_reasons[ac_queue] == 0 &&
@@ -352,6 +350,9 @@
 	if (__test_and_set_bit(reason, &local->queue_stop_reasons[queue]))
 		return;
 
+	if (local->ops->wake_tx_queue)
+		return;
+
 	if (local->hw.queues < IEEE80211_NUM_ACS)
 		n_acs = 1;