[PATCH 0/4] mt7663s: move tx/rx processing in a dedicated wq

[PATCH 0/4] mt7663s: move tx/rx processing in a dedicated wq

[Lorenzo Bianconi]

Introduce mt76s_txrx_wq workqueue and move tx/rx processing from
kthreads to dedicated works.
Split tx/rx status processing in two separated works.
This series improves device throughput increasing tx/rx parallelization

Lorenzo Bianconi (4):
  mt76: mt76s: move tx processing in a dedicated wq
  mt76: mt7663s: move rx processing in txrx wq
  mt76: mt76s: move status processing in txrx wq
  mt76: mt76s: move tx/rx processing in 2 separate works

 drivers/net/wireless/mediatek/mt76/mt76.h     |  12 +-
 .../wireless/mediatek/mt76/mt7615/mt7615.h    |   3 +-
 .../net/wireless/mediatek/mt76/mt7615/sdio.c  |  10 +-
 .../wireless/mediatek/mt76/mt7615/sdio_txrx.c | 134 ++++++++++--------
 drivers/net/wireless/mediatek/mt76/sdio.c     |  75 +++++-----
 5 files changed, 127 insertions(+), 107 deletions(-)

-- 
2.26.2

[PATCH 1/4] mt76: mt76s: move tx processing in a dedicated wq

[Lorenzo Bianconi]

Introduce mt76s_txrx_wq workqueue and move tx processing from kthread to
a dedicated work. This is preliminary patch to improve mt7663s throughput

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/wireless/mediatek/mt76/mt76.h     |  4 +-
 .../wireless/mediatek/mt76/mt7615/mt7615.h    |  2 +-
 .../net/wireless/mediatek/mt76/mt7615/sdio.c  |  9 +--
 .../wireless/mediatek/mt76/mt7615/sdio_txrx.c | 56 +++++++------------
 drivers/net/wireless/mediatek/mt76/sdio.c     | 16 +++++-
 5 files changed, 39 insertions(+), 48 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 4482eeb91a64..4570770e9b1d 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -447,10 +447,12 @@ struct mt76_usb {
 };
 
 struct mt76_sdio {
-	struct task_struct *tx_kthread;
 	struct task_struct *kthread;
 	struct work_struct stat_work;
 
+	struct workqueue_struct *txrx_wq;
+	struct work_struct tx_work;
+
 	unsigned long state;
 
 	struct sdio_func *func;
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h b/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h
index e93f87af3d2a..18b0b7b614ef 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h
@@ -678,7 +678,7 @@ u32 mt7663s_read_pcr(struct mt7615_dev *dev);
 int mt7663s_mcu_init(struct mt7615_dev *dev);
 int mt7663s_driver_own(struct mt7615_dev *dev);
 int mt7663s_firmware_own(struct mt7615_dev *dev);
-int mt7663s_kthread_run(void *data);
+void mt7663s_tx_work(struct work_struct *work);
 void mt7663s_sdio_irq(struct sdio_func *func);
 
 #endif
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
index dabce51117b0..3ab033e0f440 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
@@ -364,18 +364,15 @@ static int mt7663s_probe(struct sdio_func *func,
 	dev->ops = ops;
 	sdio_set_drvdata(func, dev);
 
-	mdev->sdio.tx_kthread = kthread_create(mt7663s_kthread_run, dev,
-					       "mt7663s_tx");
-	if (IS_ERR(mdev->sdio.tx_kthread))
-		return PTR_ERR(mdev->sdio.tx_kthread);
-
 	ret = mt76s_init(mdev, func, &mt7663s_ops);
 	if (ret < 0)
 		goto err_free;
 
+	INIT_WORK(&mdev->sdio.tx_work, mt7663s_tx_work);
+
 	ret = mt7663s_hw_init(dev, func);
 	if (ret)
-		goto err_free;
+		goto err_deinit;
 
 	mdev->rev = (mt76_rr(dev, MT_HW_CHIPID) << 16) |
 		    (mt76_rr(dev, MT_HW_REV) & 0xff);
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
index 443a4ecdad3a..9340d1570a78 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
@@ -116,12 +116,12 @@ static int mt7663s_rx_run_queue(struct mt7615_dev *dev, enum mt76_rxq_id qid,
 	return err;
 }
 
-static int mt7663s_tx_update_sched(struct mt7615_dev *dev,
+static int mt7663s_tx_update_sched(struct mt76_dev *dev,
 				   struct mt76_queue_entry *e,
 				   bool mcu)
 {
-	struct mt76_sdio *sdio = &dev->mt76.sdio;
-	struct mt76_phy *mphy = &dev->mt76.phy;
+	struct mt76_sdio *sdio = &dev->sdio;
+	struct mt76_phy *mphy = &dev->phy;
 	struct ieee80211_hdr *hdr;
 	int size, ret = -EBUSY;
 
@@ -157,10 +157,10 @@ static int mt7663s_tx_update_sched(struct mt7615_dev *dev,
 	return ret;
 }
 
-static int mt7663s_tx_run_queue(struct mt7615_dev *dev, struct mt76_queue *q)
+static int mt7663s_tx_run_queue(struct mt76_dev *dev, struct mt76_queue *q)
 {
-	bool mcu = q == dev->mt76.q_tx[MT_TXQ_MCU].q;
-	struct mt76_sdio *sdio = &dev->mt76.sdio;
+	bool mcu = q == dev->q_tx[MT_TXQ_MCU].q;
+	struct mt76_sdio *sdio = &dev->sdio;
 	int nframes = 0;
 
 	while (q->first != q->tail) {
@@ -174,9 +174,12 @@ static int mt7663s_tx_run_queue(struct mt7615_dev *dev, struct mt76_queue *q)
 			len = roundup(len, sdio->func->cur_blksize);
 
 		/* TODO: skb_walk_frags and then write to SDIO port */
+		sdio_claim_host(sdio->func);
 		err = sdio_writesb(sdio->func, MCR_WTDR1, e->skb->data, len);
+		sdio_release_host(sdio->func);
+
 		if (err) {
-			dev_err(dev->mt76.dev, "sdio write failed: %d\n", err);
+			dev_err(dev->dev, "sdio write failed: %d\n", err);
 			return -EIO;
 		}
 
@@ -188,46 +191,25 @@ static int mt7663s_tx_run_queue(struct mt7615_dev *dev, struct mt76_queue *q)
 	return nframes;
 }
 
-static int mt7663s_tx_run_queues(struct mt7615_dev *dev)
+void mt7663s_tx_work(struct work_struct *work)
 {
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio, tx_work);
+	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
 	int i, nframes = 0;
 
 	for (i = 0; i < MT_TXQ_MCU_WA; i++) {
 		int ret;
 
-		ret = mt7663s_tx_run_queue(dev, dev->mt76.q_tx[i].q);
+		ret = mt7663s_tx_run_queue(dev, dev->q_tx[i].q);
 		if (ret < 0)
-			return ret;
+			break;
 
 		nframes += ret;
 	}
+	if (nframes)
+		queue_work(sdio->txrx_wq, &sdio->tx_work);
 
-	return nframes;
-}
-
-int mt7663s_kthread_run(void *data)
-{
-	struct mt7615_dev *dev = data;
-	struct mt76_phy *mphy = &dev->mt76.phy;
-
-	while (!kthread_should_stop()) {
-		int ret;
-
-		cond_resched();
-
-		sdio_claim_host(dev->mt76.sdio.func);
-		ret = mt7663s_tx_run_queues(dev);
-		sdio_release_host(dev->mt76.sdio.func);
-
-		if (ret <= 0 || !test_bit(MT76_STATE_RUNNING, &mphy->state)) {
-			set_current_state(TASK_INTERRUPTIBLE);
-			schedule();
-		} else {
-			wake_up_process(dev->mt76.sdio.kthread);
-		}
-	}
-
-	return 0;
+	wake_up_process(sdio->kthread);
 }
 
 void mt7663s_sdio_irq(struct sdio_func *func)
@@ -258,7 +240,7 @@ void mt7663s_sdio_irq(struct sdio_func *func)
 
 		if (intr.isr & WHIER_TX_DONE_INT_EN) {
 			mt7663s_refill_sched_quota(dev, intr.tx.wtqcr);
-			mt7663s_tx_run_queues(dev);
+			queue_work(sdio->txrx_wq, &sdio->tx_work);
 			wake_up_process(sdio->kthread);
 		}
 	} while (intr.isr);
diff --git a/drivers/net/wireless/mediatek/mt76/sdio.c b/drivers/net/wireless/mediatek/mt76/sdio.c
index 5d8353026aaf..4a233e0e9d25 100644
--- a/drivers/net/wireless/mediatek/mt76/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/sdio.c
@@ -68,6 +68,7 @@ void mt76s_stop_txrx(struct mt76_dev *dev)
 {
 	struct mt76_sdio *sdio = &dev->sdio;
 
+	cancel_work_sync(&sdio->tx_work);
 	cancel_work_sync(&sdio->stat_work);
 	clear_bit(MT76_READING_STATS, &dev->phy.state);
 
@@ -179,7 +180,6 @@ static int mt76s_process_tx_queue(struct mt76_dev *dev, enum mt76_txq_id qid)
 	if (wake)
 		ieee80211_wake_queue(dev->hw, qid);
 
-	wake_up_process(dev->sdio.tx_kthread);
 out:
 	return n_dequeued;
 }
@@ -272,7 +272,7 @@ static void mt76s_tx_kick(struct mt76_dev *dev, struct mt76_queue *q)
 {
 	struct mt76_sdio *sdio = &dev->sdio;
 
-	wake_up_process(sdio->tx_kthread);
+	queue_work(sdio->txrx_wq, &sdio->tx_work);
 }
 
 static const struct mt76_queue_ops sdio_queue_ops = {
@@ -324,9 +324,13 @@ void mt76s_deinit(struct mt76_dev *dev)
 	int i;
 
 	kthread_stop(sdio->kthread);
-	kthread_stop(sdio->tx_kthread);
 	mt76s_stop_txrx(dev);
 
+	if (sdio->txrx_wq) {
+		destroy_workqueue(sdio->txrx_wq);
+		sdio->txrx_wq = NULL;
+	}
+
 	sdio_claim_host(sdio->func);
 	sdio_release_irq(sdio->func);
 	sdio_release_host(sdio->func);
@@ -353,6 +357,12 @@ int mt76s_init(struct mt76_dev *dev, struct sdio_func *func,
 {
 	struct mt76_sdio *sdio = &dev->sdio;
 
+	sdio->txrx_wq = alloc_workqueue("mt76s_txrx_wq",
+					WQ_UNBOUND | WQ_HIGHPRI,
+					WQ_UNBOUND_MAX_ACTIVE);
+	if (!sdio->txrx_wq)
+		return -ENOMEM;
+
 	sdio->kthread = kthread_create(mt76s_kthread_run, dev, "mt76s");
 	if (IS_ERR(sdio->kthread))
 		return PTR_ERR(sdio->kthread);
-- 
2.26.2

[PATCH 2/4] mt76: mt7663s: move rx processing in txrx wq

[Lorenzo Bianconi]

Move rx processing to mt76s_txrx_wq in order to minimize the interval when
the sdio bus is locked during rx

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/wireless/mediatek/mt76/mt76.h     |  1 +
 .../wireless/mediatek/mt76/mt7615/mt7615.h    |  1 +
 .../net/wireless/mediatek/mt76/mt7615/sdio.c  |  1 +
 .../wireless/mediatek/mt76/mt7615/sdio_txrx.c | 82 ++++++++++++-------
 drivers/net/wireless/mediatek/mt76/sdio.c     |  1 +
 5 files changed, 57 insertions(+), 29 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 4570770e9b1d..6079f44287ea 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -452,6 +452,7 @@ struct mt76_sdio {
 
 	struct workqueue_struct *txrx_wq;
 	struct work_struct tx_work;
+	struct work_struct rx_work;
 
 	unsigned long state;
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h b/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h
index 18b0b7b614ef..4892a3d6d6c3 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/mt7615.h
@@ -679,6 +679,7 @@ int mt7663s_mcu_init(struct mt7615_dev *dev);
 int mt7663s_driver_own(struct mt7615_dev *dev);
 int mt7663s_firmware_own(struct mt7615_dev *dev);
 void mt7663s_tx_work(struct work_struct *work);
+void mt7663s_rx_work(struct work_struct *work);
 void mt7663s_sdio_irq(struct sdio_func *func);
 
 #endif
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
index 3ab033e0f440..0cc3f0aca70b 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
@@ -369,6 +369,7 @@ static int mt7663s_probe(struct sdio_func *func,
 		goto err_free;
 
 	INIT_WORK(&mdev->sdio.tx_work, mt7663s_tx_work);
+	INIT_WORK(&mdev->sdio.rx_work, mt7663s_rx_work);
 
 	ret = mt7663s_hw_init(dev, func);
 	if (ret)
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
index 9340d1570a78..c214960504bf 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
@@ -19,9 +19,9 @@
 #include "sdio.h"
 #include "mac.h"
 
-static void mt7663s_refill_sched_quota(struct mt7615_dev *dev, u32 *data)
+static void mt7663s_refill_sched_quota(struct mt76_dev *dev, u32 *data)
 {
-	struct mt76_sdio *sdio = &dev->mt76.sdio;
+	struct mt76_sdio *sdio = &dev->sdio;
 
 	mutex_lock(&sdio->sched.lock);
 	sdio->sched.pse_data_quota += FIELD_GET(TXQ_CNT_L, data[0]) + /* BK */
@@ -61,11 +61,11 @@ static struct sk_buff *mt7663s_build_rx_skb(void *data, int data_len,
 	return skb;
 }
 
-static int mt7663s_rx_run_queue(struct mt7615_dev *dev, enum mt76_rxq_id qid,
+static int mt7663s_rx_run_queue(struct mt76_dev *dev, enum mt76_rxq_id qid,
 				struct mt76s_intr *intr)
 {
-	struct mt76_queue *q = &dev->mt76.q_rx[qid];
-	struct mt76_sdio *sdio = &dev->mt76.sdio;
+	struct mt76_queue *q = &dev->q_rx[qid];
+	struct mt76_sdio *sdio = &dev->sdio;
 	int len = 0, err, i, order;
 	struct page *page;
 	u8 *buf;
@@ -86,9 +86,12 @@ static int mt7663s_rx_run_queue(struct mt7615_dev *dev, enum mt76_rxq_id qid,
 
 	buf = page_address(page);
 
+	sdio_claim_host(sdio->func);
 	err = sdio_readsb(sdio->func, buf, MCR_WRDR(qid), len);
+	sdio_release_host(sdio->func);
+
 	if (err < 0) {
-		dev_err(dev->mt76.dev, "sdio read data failed:%d\n", err);
+		dev_err(dev->dev, "sdio read data failed:%d\n", err);
 		__free_pages(page, order);
 		return err;
 	}
@@ -113,7 +116,7 @@ static int mt7663s_rx_run_queue(struct mt7615_dev *dev, enum mt76_rxq_id qid,
 	q->queued += i;
 	spin_unlock_bh(&q->lock);
 
-	return err;
+	return i;
 }
 
 static int mt7663s_tx_update_sched(struct mt76_dev *dev,
@@ -212,39 +215,60 @@ void mt7663s_tx_work(struct work_struct *work)
 	wake_up_process(sdio->kthread);
 }
 
-void mt7663s_sdio_irq(struct sdio_func *func)
+void mt7663s_rx_work(struct work_struct *work)
 {
-	struct mt7615_dev *dev = sdio_get_drvdata(func);
-	struct mt76_sdio *sdio = &dev->mt76.sdio;
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio, rx_work);
+	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
 	struct mt76s_intr intr;
+	int nframes = 0, ret;
 
 	/* disable interrupt */
-	sdio_writel(func, WHLPCR_INT_EN_CLR, MCR_WHLPCR, 0);
+	sdio_claim_host(sdio->func);
+	sdio_writel(sdio->func, WHLPCR_INT_EN_CLR, MCR_WHLPCR, 0);
+	sdio_readsb(sdio->func, &intr, MCR_WHISR, sizeof(struct mt76s_intr));
+	sdio_release_host(sdio->func);
 
-	do {
-		sdio_readsb(func, &intr, MCR_WHISR, sizeof(struct mt76s_intr));
-		trace_dev_irq(&dev->mt76, intr.isr, 0);
+	trace_dev_irq(dev, intr.isr, 0);
 
-		if (!test_bit(MT76_STATE_INITIALIZED, &dev->mt76.phy.state))
-			goto out;
-
-		if (intr.isr & WHIER_RX0_DONE_INT_EN) {
-			mt7663s_rx_run_queue(dev, 0, &intr);
+	if (intr.isr & WHIER_RX0_DONE_INT_EN) {
+		ret = mt7663s_rx_run_queue(dev, 0, &intr);
+		if (ret > 0) {
 			wake_up_process(sdio->kthread);
+			nframes += ret;
 		}
+	}
 
-		if (intr.isr & WHIER_RX1_DONE_INT_EN) {
-			mt7663s_rx_run_queue(dev, 1, &intr);
+	if (intr.isr & WHIER_RX1_DONE_INT_EN) {
+		ret = mt7663s_rx_run_queue(dev, 1, &intr);
+		if (ret > 0) {
 			wake_up_process(sdio->kthread);
+			nframes += ret;
 		}
+	}
+
+	if (intr.isr & WHIER_TX_DONE_INT_EN) {
+		mt7663s_refill_sched_quota(dev, intr.tx.wtqcr);
+		queue_work(sdio->txrx_wq, &sdio->tx_work);
+	}
+
+	if (nframes) {
+		queue_work(sdio->txrx_wq, &sdio->rx_work);
+		return;
+	}
 
-		if (intr.isr & WHIER_TX_DONE_INT_EN) {
-			mt7663s_refill_sched_quota(dev, intr.tx.wtqcr);
-			queue_work(sdio->txrx_wq, &sdio->tx_work);
-			wake_up_process(sdio->kthread);
-		}
-	} while (intr.isr);
-out:
 	/* enable interrupt */
-	sdio_writel(func, WHLPCR_INT_EN_SET, MCR_WHLPCR, 0);
+	sdio_claim_host(sdio->func);
+	sdio_writel(sdio->func, WHLPCR_INT_EN_SET, MCR_WHLPCR, 0);
+	sdio_release_host(sdio->func);
+}
+
+void mt7663s_sdio_irq(struct sdio_func *func)
+{
+	struct mt7615_dev *dev = sdio_get_drvdata(func);
+	struct mt76_sdio *sdio = &dev->mt76.sdio;
+
+	if (!test_bit(MT76_STATE_INITIALIZED, &dev->mt76.phy.state))
+		return;
+
+	queue_work(sdio->txrx_wq, &sdio->rx_work);
 }
diff --git a/drivers/net/wireless/mediatek/mt76/sdio.c b/drivers/net/wireless/mediatek/mt76/sdio.c
index 4a233e0e9d25..e9fa0ca8f9cd 100644
--- a/drivers/net/wireless/mediatek/mt76/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/sdio.c
@@ -69,6 +69,7 @@ void mt76s_stop_txrx(struct mt76_dev *dev)
 	struct mt76_sdio *sdio = &dev->sdio;
 
 	cancel_work_sync(&sdio->tx_work);
+	cancel_work_sync(&sdio->rx_work);
 	cancel_work_sync(&sdio->stat_work);
 	clear_bit(MT76_READING_STATS, &dev->phy.state);
 
-- 
2.26.2

[PATCH 3/4] mt76: mt76s: move status processing in txrx wq

[Lorenzo Bianconi]

As it has been done for tx and rx processing, move tx/rx status
processing into mt76s_txrx_wq workqueue

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/wireless/mediatek/mt76/mt76.h     |  6 +--
 .../wireless/mediatek/mt76/mt7615/sdio_txrx.c |  6 +--
 drivers/net/wireless/mediatek/mt76/sdio.c     | 54 +++++++------------
 3 files changed, 25 insertions(+), 41 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 6079f44287ea..17f424d78c2c 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -447,12 +447,12 @@ struct mt76_usb {
 };
 
 struct mt76_sdio {
-	struct task_struct *kthread;
-	struct work_struct stat_work;
-
 	struct workqueue_struct *txrx_wq;
 	struct work_struct tx_work;
 	struct work_struct rx_work;
+	struct work_struct work;
+
+	struct work_struct stat_work;
 
 	unsigned long state;
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
index c214960504bf..8872b145df64 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
@@ -212,7 +212,7 @@ void mt7663s_tx_work(struct work_struct *work)
 	if (nframes)
 		queue_work(sdio->txrx_wq, &sdio->tx_work);
 
-	wake_up_process(sdio->kthread);
+	queue_work(sdio->txrx_wq, &sdio->work);
 }
 
 void mt7663s_rx_work(struct work_struct *work)
@@ -233,7 +233,7 @@ void mt7663s_rx_work(struct work_struct *work)
 	if (intr.isr & WHIER_RX0_DONE_INT_EN) {
 		ret = mt7663s_rx_run_queue(dev, 0, &intr);
 		if (ret > 0) {
-			wake_up_process(sdio->kthread);
+			queue_work(sdio->txrx_wq, &sdio->work);
 			nframes += ret;
 		}
 	}
@@ -241,7 +241,7 @@ void mt7663s_rx_work(struct work_struct *work)
 	if (intr.isr & WHIER_RX1_DONE_INT_EN) {
 		ret = mt7663s_rx_run_queue(dev, 1, &intr);
 		if (ret > 0) {
-			wake_up_process(sdio->kthread);
+			queue_work(sdio->txrx_wq, &sdio->work);
 			nframes += ret;
 		}
 	}
diff --git a/drivers/net/wireless/mediatek/mt76/sdio.c b/drivers/net/wireless/mediatek/mt76/sdio.c
index e9fa0ca8f9cd..5432b388ceab 100644
--- a/drivers/net/wireless/mediatek/mt76/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/sdio.c
@@ -70,6 +70,7 @@ void mt76s_stop_txrx(struct mt76_dev *dev)
 
 	cancel_work_sync(&sdio->tx_work);
 	cancel_work_sync(&sdio->rx_work);
+	cancel_work_sync(&sdio->work);
 	cancel_work_sync(&sdio->stat_work);
 	clear_bit(MT76_READING_STATS, &dev->phy.state);
 
@@ -282,41 +283,29 @@ static const struct mt76_queue_ops sdio_queue_ops = {
 	.tx_queue_skb_raw = mt76s_tx_queue_skb_raw,
 };
 
-static int mt76s_kthread_run(void *data)
+static void mt76s_txrx_work(struct work_struct *work)
 {
-	struct mt76_dev *dev = data;
-	struct mt76_phy *mphy = &dev->phy;
-
-	while (!kthread_should_stop()) {
-		int i, nframes = 0;
-
-		cond_resched();
-
-		/* rx processing */
-		local_bh_disable();
-		rcu_read_lock();
-
-		mt76_for_each_q_rx(dev, i)
-			nframes += mt76s_process_rx_queue(dev, &dev->q_rx[i]);
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio, work);
+	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
+	int i;
 
-		rcu_read_unlock();
-		local_bh_enable();
+	/* rx processing */
+	local_bh_disable();
+	rcu_read_lock();
 
-		/* tx processing */
-		for (i = 0; i < MT_TXQ_MCU_WA; i++)
-			nframes += mt76s_process_tx_queue(dev, i);
+	mt76_for_each_q_rx(dev, i)
+		mt76s_process_rx_queue(dev, &dev->q_rx[i]);
 
-		if (dev->drv->tx_status_data &&
-		    !test_and_set_bit(MT76_READING_STATS, &mphy->state))
-			queue_work(dev->wq, &dev->sdio.stat_work);
+	rcu_read_unlock();
+	local_bh_enable();
 
-		if (!nframes || !test_bit(MT76_STATE_RUNNING, &mphy->state)) {
-			set_current_state(TASK_INTERRUPTIBLE);
-			schedule();
-		}
-	}
+	/* tx processing */
+	for (i = 0; i < MT_TXQ_MCU_WA; i++)
+		mt76s_process_tx_queue(dev, i);
 
-	return 0;
+	if (dev->drv->tx_status_data &&
+	    !test_and_set_bit(MT76_READING_STATS, &dev->phy.state))
+		queue_work(dev->wq, &dev->sdio.stat_work);
 }
 
 void mt76s_deinit(struct mt76_dev *dev)
@@ -324,9 +313,7 @@ void mt76s_deinit(struct mt76_dev *dev)
 	struct mt76_sdio *sdio = &dev->sdio;
 	int i;
 
-	kthread_stop(sdio->kthread);
 	mt76s_stop_txrx(dev);
-
 	if (sdio->txrx_wq) {
 		destroy_workqueue(sdio->txrx_wq);
 		sdio->txrx_wq = NULL;
@@ -364,11 +351,8 @@ int mt76s_init(struct mt76_dev *dev, struct sdio_func *func,
 	if (!sdio->txrx_wq)
 		return -ENOMEM;
 
-	sdio->kthread = kthread_create(mt76s_kthread_run, dev, "mt76s");
-	if (IS_ERR(sdio->kthread))
-		return PTR_ERR(sdio->kthread);
-
 	INIT_WORK(&sdio->stat_work, mt76s_tx_status_data);
+	INIT_WORK(&sdio->work, mt76s_txrx_work);
 
 	mutex_init(&sdio->sched.lock);
 	dev->queue_ops = &sdio_queue_ops;
-- 
2.26.2

[PATCH 4/4] mt76: mt76s: move tx/rx processing in 2 separate works

[Lorenzo Bianconi]

In order to maximize parallelism, split status work in tx status work
and rx net work

Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
---
 drivers/net/wireless/mediatek/mt76/mt76.h     | 11 +++--
 .../net/wireless/mediatek/mt76/mt7615/sdio.c  |  4 +-
 .../wireless/mediatek/mt76/mt7615/sdio_txrx.c | 20 +++++-----
 drivers/net/wireless/mediatek/mt76/sdio.c     | 40 ++++++++++++-------
 4 files changed, 46 insertions(+), 29 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76.h b/drivers/net/wireless/mediatek/mt76/mt76.h
index 17f424d78c2c..d44a35463197 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76.h
@@ -448,9 +448,14 @@ struct mt76_usb {
 
 struct mt76_sdio {
 	struct workqueue_struct *txrx_wq;
-	struct work_struct tx_work;
-	struct work_struct rx_work;
-	struct work_struct work;
+	struct {
+		struct work_struct xmit_work;
+		struct work_struct status_work;
+	} tx;
+	struct {
+		struct work_struct recv_work;
+		struct work_struct net_work;
+	} rx;
 
 	struct work_struct stat_work;
 
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
index 0cc3f0aca70b..ddf62dc05c00 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio.c
@@ -368,8 +368,8 @@ static int mt7663s_probe(struct sdio_func *func,
 	if (ret < 0)
 		goto err_free;
 
-	INIT_WORK(&mdev->sdio.tx_work, mt7663s_tx_work);
-	INIT_WORK(&mdev->sdio.rx_work, mt7663s_rx_work);
+	INIT_WORK(&mdev->sdio.tx.xmit_work, mt7663s_tx_work);
+	INIT_WORK(&mdev->sdio.rx.recv_work, mt7663s_rx_work);
 
 	ret = mt7663s_hw_init(dev, func);
 	if (ret)
diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
index 8872b145df64..c945b4e0320d 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/sdio_txrx.c
@@ -196,7 +196,8 @@ static int mt7663s_tx_run_queue(struct mt76_dev *dev, struct mt76_queue *q)
 
 void mt7663s_tx_work(struct work_struct *work)
 {
-	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio, tx_work);
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio,
+					      tx.xmit_work);
 	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
 	int i, nframes = 0;
 
@@ -210,14 +211,15 @@ void mt7663s_tx_work(struct work_struct *work)
 		nframes += ret;
 	}
 	if (nframes)
-		queue_work(sdio->txrx_wq, &sdio->tx_work);
+		queue_work(sdio->txrx_wq, &sdio->tx.xmit_work);
 
-	queue_work(sdio->txrx_wq, &sdio->work);
+	queue_work(sdio->txrx_wq, &sdio->tx.status_work);
 }
 
 void mt7663s_rx_work(struct work_struct *work)
 {
-	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio, rx_work);
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio,
+					      rx.recv_work);
 	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
 	struct mt76s_intr intr;
 	int nframes = 0, ret;
@@ -233,7 +235,7 @@ void mt7663s_rx_work(struct work_struct *work)
 	if (intr.isr & WHIER_RX0_DONE_INT_EN) {
 		ret = mt7663s_rx_run_queue(dev, 0, &intr);
 		if (ret > 0) {
-			queue_work(sdio->txrx_wq, &sdio->work);
+			queue_work(sdio->txrx_wq, &sdio->rx.net_work);
 			nframes += ret;
 		}
 	}
@@ -241,18 +243,18 @@ void mt7663s_rx_work(struct work_struct *work)
 	if (intr.isr & WHIER_RX1_DONE_INT_EN) {
 		ret = mt7663s_rx_run_queue(dev, 1, &intr);
 		if (ret > 0) {
-			queue_work(sdio->txrx_wq, &sdio->work);
+			queue_work(sdio->txrx_wq, &sdio->rx.net_work);
 			nframes += ret;
 		}
 	}
 
 	if (intr.isr & WHIER_TX_DONE_INT_EN) {
 		mt7663s_refill_sched_quota(dev, intr.tx.wtqcr);
-		queue_work(sdio->txrx_wq, &sdio->tx_work);
+		queue_work(sdio->txrx_wq, &sdio->tx.xmit_work);
 	}
 
 	if (nframes) {
-		queue_work(sdio->txrx_wq, &sdio->rx_work);
+		queue_work(sdio->txrx_wq, &sdio->rx.recv_work);
 		return;
 	}
 
@@ -270,5 +272,5 @@ void mt7663s_sdio_irq(struct sdio_func *func)
 	if (!test_bit(MT76_STATE_INITIALIZED, &dev->mt76.phy.state))
 		return;
 
-	queue_work(sdio->txrx_wq, &sdio->rx_work);
+	queue_work(sdio->txrx_wq, &sdio->rx.recv_work);
 }
diff --git a/drivers/net/wireless/mediatek/mt76/sdio.c b/drivers/net/wireless/mediatek/mt76/sdio.c
index 5432b388ceab..40fd752d1234 100644
--- a/drivers/net/wireless/mediatek/mt76/sdio.c
+++ b/drivers/net/wireless/mediatek/mt76/sdio.c
@@ -68,9 +68,10 @@ void mt76s_stop_txrx(struct mt76_dev *dev)
 {
 	struct mt76_sdio *sdio = &dev->sdio;
 
-	cancel_work_sync(&sdio->tx_work);
-	cancel_work_sync(&sdio->rx_work);
-	cancel_work_sync(&sdio->work);
+	cancel_work_sync(&sdio->tx.xmit_work);
+	cancel_work_sync(&sdio->tx.status_work);
+	cancel_work_sync(&sdio->rx.recv_work);
+	cancel_work_sync(&sdio->rx.net_work);
 	cancel_work_sync(&sdio->stat_work);
 	clear_bit(MT76_READING_STATS, &dev->phy.state);
 
@@ -274,7 +275,7 @@ static void mt76s_tx_kick(struct mt76_dev *dev, struct mt76_queue *q)
 {
 	struct mt76_sdio *sdio = &dev->sdio;
 
-	queue_work(sdio->txrx_wq, &sdio->tx_work);
+	queue_work(sdio->txrx_wq, &sdio->tx.xmit_work);
 }
 
 static const struct mt76_queue_ops sdio_queue_ops = {
@@ -283,9 +284,25 @@ static const struct mt76_queue_ops sdio_queue_ops = {
 	.tx_queue_skb_raw = mt76s_tx_queue_skb_raw,
 };
 
-static void mt76s_txrx_work(struct work_struct *work)
+static void mt76s_tx_work(struct work_struct *work)
 {
-	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio, work);
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio,
+					      tx.status_work);
+	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
+	int i;
+
+	for (i = 0; i < MT_TXQ_MCU_WA; i++)
+		mt76s_process_tx_queue(dev, i);
+
+	if (dev->drv->tx_status_data &&
+	    !test_and_set_bit(MT76_READING_STATS, &dev->phy.state))
+		queue_work(dev->wq, &dev->sdio.stat_work);
+}
+
+static void mt76s_rx_work(struct work_struct *work)
+{
+	struct mt76_sdio *sdio = container_of(work, struct mt76_sdio,
+					      rx.net_work);
 	struct mt76_dev *dev = container_of(sdio, struct mt76_dev, sdio);
 	int i;
 
@@ -298,14 +315,6 @@ static void mt76s_txrx_work(struct work_struct *work)
 
 	rcu_read_unlock();
 	local_bh_enable();
-
-	/* tx processing */
-	for (i = 0; i < MT_TXQ_MCU_WA; i++)
-		mt76s_process_tx_queue(dev, i);
-
-	if (dev->drv->tx_status_data &&
-	    !test_and_set_bit(MT76_READING_STATS, &dev->phy.state))
-		queue_work(dev->wq, &dev->sdio.stat_work);
 }
 
 void mt76s_deinit(struct mt76_dev *dev)
@@ -352,7 +361,8 @@ int mt76s_init(struct mt76_dev *dev, struct sdio_func *func,
 		return -ENOMEM;
 
 	INIT_WORK(&sdio->stat_work, mt76s_tx_status_data);
-	INIT_WORK(&sdio->work, mt76s_txrx_work);
+	INIT_WORK(&sdio->tx.status_work, mt76s_tx_work);
+	INIT_WORK(&sdio->rx.net_work, mt76s_rx_work);
 
 	mutex_init(&sdio->sched.lock);
 	dev->queue_ops = &sdio_queue_ops;
-- 
2.26.2