[PATCH v2 1/3] mt76: improve tx queue stop/wake

[PATCH v2 1/3] mt76: improve tx queue stop/wake

[Felix Fietkau]

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 <nbd@nbd.name>
---
 drivers/net/wireless/mediatek/mt76/dma.c  | 10 ----------
 drivers/net/wireless/mediatek/mt76/mt76.h |  1 +
 drivers/net/wireless/mediatek/mt76/tx.c   | 23 ++++++++++++-----------
 3 files changed, 13 insertions(+), 21 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/dma.c b/drivers/net/wireless/mediatek/mt76/dma.c
index bab961ebdf1c..6255f4f0a455 100644
--- a/drivers/net/wireless/mediatek/mt76/dma.c
+++ b/drivers/net/wireless/mediatek/mt76/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 *
diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 15d52af24d12..ea80ae188dd6 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -136,6 +136,7 @@ struct mt76_queue {
 	int queued;
 	int buf_size;
 	bool stopped;
+	bool blocked;
 
 	u8 buf_offset;
 	u8 hw_idx;
diff --git a/drivers/net/wireless/mediatek/mt76/tx.c b/drivers/net/wireless/mediatek/mt76/tx.c
index 65360067b8fa..4026e5f4f3f3 100644
--- a/drivers/net/wireless/mediatek/mt76/tx.c
+++ b/drivers/net/wireless/mediatek/mt76/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);
@@ -381,6 +375,13 @@ mt76_release_buffered_frames(struct ieee80211_hw *hw, struct ieee80211_sta *sta,
 }
 EXPORT_SYMBOL_GPL(mt76_release_buffered_frames);
 
+static bool
+mt76_txq_stopped(struct mt76_queue *q)
+{
+	return q->stopped || q->blocked ||
+	       q->queued + MT_TXQ_FREE_THR >= q->ndesc;
+}
+
 static int
 mt76_txq_send_burst(struct mt76_phy *phy, struct mt76_queue *q,
 		    struct mt76_txq *mtxq)
@@ -419,10 +420,7 @@ mt76_txq_send_burst(struct mt76_phy *phy, struct mt76_queue *q,
 		    test_bit(MT76_RESET, &phy->state))
 			return -EBUSY;
 
-		if (stop)
-			break;
-
-		if (q->queued + MT_TXQ_FREE_THR >= q->ndesc)
+		if (stop || mt76_txq_stopped(q))
 			break;
 
 		skb = mt76_txq_dequeue(phy, mtxq);
@@ -463,7 +461,7 @@ mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 			break;
 		}
 
-		if (q->queued + MT_TXQ_FREE_THR >= q->ndesc)
+		if (mt76_txq_stopped(q))
 			break;
 
 		txq = ieee80211_next_txq(phy->hw, qid);
@@ -498,11 +496,14 @@ 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_queue *q;
 	int len;
 
 	if (qid >= 4)
 		return;
 
+	q = phy->q_tx[qid];
+
 	rcu_read_lock();
 
 	do {
-- 
2.28.0

[PATCH v2 2/3] mt76: mt7915: stop queues when running out of tx tokens

[Felix Fietkau]

Avoids packet drops under load with lots of stations

Signed-off-by: Felix Fietkau <nbd@nbd.name>
---
 .../net/wireless/mediatek/mt76/mt7915/init.c  |  1 +
 .../net/wireless/mediatek/mt76/mt7915/mac.c   | 39 +++++++++++++++++++
 .../wireless/mediatek/mt76/mt7915/mt7915.h    |  2 +
 3 files changed, 42 insertions(+)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/init.c b/drivers/net/wireless/mediatek/mt76/mt7915/init.c
index ff29a8090739..ed4635bd151a 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/init.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/init.c
@@ -690,6 +690,7 @@ void mt7915_unregister_device(struct mt7915_dev *dev)
 			ieee80211_free_txskb(hw, txwi->skb);
 		}
 		mt76_put_txwi(&dev->mt76, txwi);
+		dev->token_count--;
 	}
 	spin_unlock_bh(&dev->token_lock);
 	idr_destroy(&dev->token);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
index a7fa6fffffff..71714b22188a 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mac.c
@@ -918,6 +918,26 @@ void mt7915_mac_write_txwi(struct mt7915_dev *dev, __le32 *txwi,
 		mt7915_mac_write_txwi_tm(dev, mphy, txwi, skb);
 }
 
+static void
+mt7915_set_tx_blocked(struct mt7915_dev *dev, bool blocked)
+{
+	struct mt76_phy *mphy = &dev->mphy, *mphy2 = dev->mt76.phy2;
+	struct mt76_queue *q, *q2 = NULL;
+
+	q = mphy->q_tx[0];
+	if (blocked == q->blocked)
+		return;
+
+	q->blocked = blocked;
+	if (mphy2) {
+		q2 = mphy2->q_tx[0];
+		q2->blocked = blocked;
+	}
+
+	if (!blocked)
+		mt76_worker_schedule(&dev->mt76.tx_worker);
+}
+
 int mt7915_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
 			  enum mt76_txq_id qid, struct mt76_wcid *wcid,
 			  struct ieee80211_sta *sta,
@@ -974,7 +994,13 @@ int mt7915_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
 
 	spin_lock_bh(&dev->token_lock);
 	id = idr_alloc(&dev->token, t, 0, MT7915_TOKEN_SIZE, GFP_ATOMIC);
+	if (id >= 0)
+		dev->token_count++;
+
+	if (dev->token_count >= MT7915_TOKEN_SIZE - MT7915_TOKEN_FREE_THR)
+		mt7915_set_tx_blocked(dev, true);
 	spin_unlock_bh(&dev->token_lock);
+
 	if (id < 0)
 		return id;
 
@@ -1083,6 +1109,7 @@ void mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 	LIST_HEAD(free_list);
 	struct sk_buff *tmp;
 	u8 i, count;
+	bool wake = false;
 
 	/* clean DMA queues and unmap buffers first */
 	mt76_queue_tx_cleanup(dev, dev->mphy.q_tx[MT_TXQ_PSD], false);
@@ -1135,6 +1162,11 @@ void mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 
 		spin_lock_bh(&dev->token_lock);
 		txwi = idr_remove(&dev->token, msdu);
+		if (txwi)
+			dev->token_count--;
+		if (dev->token_count < MT7915_TOKEN_SIZE - MT7915_TOKEN_FREE_THR &&
+		    dev->mphy.q_tx[0]->blocked)
+			wake = true;
 		spin_unlock_bh(&dev->token_lock);
 
 		if (!txwi)
@@ -1165,6 +1197,13 @@ void mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 	}
 
 	mt7915_mac_sta_poll(dev);
+
+	if (wake) {
+		spin_lock_bh(&dev->token_lock);
+		mt7915_set_tx_blocked(dev, false);
+		spin_unlock_bh(&dev->token_lock);
+	}
+
 	mt76_worker_schedule(&dev->mt76.tx_worker);
 
 	napi_consume_skb(skb, 1);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h b/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
index 30e53a0f01fb..0339abf360d3 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7915/mt7915.h
@@ -32,6 +32,7 @@
 
 #define MT7915_EEPROM_SIZE		3584
 #define MT7915_TOKEN_SIZE		8192
+#define MT7915_TOKEN_FREE_THR		64
 
 #define MT7915_CFEND_RATE_DEFAULT	0x49	/* OFDM 24M */
 #define MT7915_CFEND_RATE_11B		0x03	/* 11B LP, 11M */
@@ -161,6 +162,7 @@ struct mt7915_dev {
 	u32 hw_pattern;
 
 	spinlock_t token_lock;
+	int token_count;
 	struct idr token;
 
 	s8 **rate_power; /* TODO: use mt76_rate_power */
-- 
2.28.0

[PATCH v2 3/3] mt76: attempt to free up more room when filling the tx queue

[Felix Fietkau]

Run dma cleanup immediately if the queue is almost full, instead of waiting
for the tx interrupt

Signed-off-by: Felix Fietkau <nbd@nbd.name>
---
 drivers/net/wireless/mediatek/mt76/dma.c  | 3 +++
 drivers/net/wireless/mediatek/mt76/mt76.h | 1 +
 drivers/net/wireless/mediatek/mt76/tx.c   | 8 ++++++++
 3 files changed, 12 insertions(+)

diff --git a/drivers/net/wireless/mediatek/mt76/dma.c b/drivers/net/wireless/mediatek/mt76/dma.c
index 6255f4f0a455..73eeb00d5aa6 100644
--- a/drivers/net/wireless/mediatek/mt76/dma.c
+++ b/drivers/net/wireless/mediatek/mt76/dma.c
@@ -88,6 +88,7 @@ mt76_dma_alloc_queue(struct mt76_dev *dev, struct mt76_queue *q,
 	int i;
 
 	spin_lock_init(&q->lock);
+	spin_lock_init(&q->cleanup_lock);
 
 	q->regs = dev->mmio.regs + ring_base + idx * MT_RING_SIZE;
 	q->ndesc = n_desc;
@@ -225,6 +226,7 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
 	if (!q)
 		return;
 
+	spin_lock_bh(&q->cleanup_lock);
 	if (flush)
 		last = -1;
 	else
@@ -243,6 +245,7 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
 			last = readl(&q->regs->dma_idx);
 
 	}
+	spin_unlock_bh(&q->cleanup_lock);
 
 	if (flush) {
 		spin_lock_bh(&q->lock);
diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index ea80ae188dd6..3e496a188bf0 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -126,6 +126,7 @@ struct mt76_queue {
 	struct mt76_queue_regs __iomem *regs;
 
 	spinlock_t lock;
+	spinlock_t cleanup_lock;
 	struct mt76_queue_entry *entry;
 	struct mt76_desc *desc;
 
diff --git a/drivers/net/wireless/mediatek/mt76/tx.c b/drivers/net/wireless/mediatek/mt76/tx.c
index 4026e5f4f3f3..1e20afb70fc1 100644
--- a/drivers/net/wireless/mediatek/mt76/tx.c
+++ b/drivers/net/wireless/mediatek/mt76/tx.c
@@ -448,6 +448,7 @@ static int
 mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 {
 	struct mt76_queue *q = phy->q_tx[qid];
+	struct mt76_dev *dev = phy->dev;
 	struct ieee80211_txq *txq;
 	struct mt76_txq *mtxq;
 	struct mt76_wcid *wcid;
@@ -461,6 +462,13 @@ mt76_txq_schedule_list(struct mt76_phy *phy, enum mt76_txq_id qid)
 			break;
 		}
 
+		if (dev->queue_ops->tx_cleanup &&
+		    q->queued + 2 * MT_TXQ_FREE_THR >= q->ndesc) {
+			spin_unlock_bh(&q->lock);
+			dev->queue_ops->tx_cleanup(dev, q, false);
+			spin_lock_bh(&q->lock);
+		}
+
 		if (mt76_txq_stopped(q))
 			break;
 
-- 
2.28.0