From 2f1318a06d0a9703f0085522361a4bc66f61d808 Mon Sep 17 00:00:00 2001
From: Felix Fietkau <nbd@nbd.name>
Date: Fri, 24 Jul 2020 16:11:52 +0200
Subject: [PATCH] mt76: convert from tx tasklet to tx worker thread

This should improve performance by allowing the scheduler to move the tx
scheduling work to idle CPUs

Signed-off-by: Felix Fietkau <nbd@nbd.name>
---
 dma.c             |  1 +
 mac80211.c        | 14 +++++++++++---
 mt76.h            |  4 ++--
 mt7603/dma.c      |  3 +--
 mt7603/mac.c      |  4 ++--
 mt7615/dma.c      |  3 +--
 mt7615/mac.c      |  8 ++++----
 mt7615/main.c     |  2 +-
 mt7615/pci.c      |  3 ++-
 mt7615/usb.c      |  2 --
 mt76x0/pci.c      |  2 +-
 mt76x02_dma.h     |  1 -
 mt76x02_mmio.c    | 22 ++++++++--------------
 mt76x2/pci.c      |  3 ++-
 mt76x2/pci_init.c |  2 +-
 mt7915/dma.c      |  1 -
 mt7915/mac.c      |  6 +++---
 testmode.c        |  6 +++---
 tx.c              |  6 +++---
 usb.c             | 14 ++++++++------
 20 files changed, 54 insertions(+), 53 deletions(-)

diff --git a/dma.c b/dma.c
index 345162a7..67fd5072 100644
--- a/dma.c
+++ b/dma.c
@@ -606,6 +606,7 @@ void mt76_dma_cleanup(struct mt76_dev *dev)
 {
 	int i;
 
+	mt76_worker_disable(&dev->tx_worker);
 	netif_napi_del(&dev->tx_napi);
 	for (i = 0; i < ARRAY_SIZE(dev->q_tx); i++)
 		mt76_dma_tx_cleanup(dev, i, true);
diff --git a/mac80211.c b/mac80211.c
index 8b41558c..4fdecdc4 100644
--- a/mac80211.c
+++ b/mac80211.c
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2016 Felix Fietkau <nbd@nbd.name>
  */
+#include <uapi/linux/sched/types.h>
 #include <linux/of.h>
 #include "mt76.h"
 
@@ -442,8 +443,6 @@ mt76_alloc_device(struct device *pdev, unsigned int size,
 	for (i = 0; i < ARRAY_SIZE(dev->q_rx); i++)
 		skb_queue_head_init(&dev->rx_skb[i]);
 
-	tasklet_init(&dev->tx_tasklet, mt76_tx_tasklet, (unsigned long)dev);
-
 	dev->wq = alloc_ordered_workqueue("mt76", 0);
 	if (!dev->wq) {
 		ieee80211_free_hw(hw);
@@ -457,6 +456,7 @@ EXPORT_SYMBOL_GPL(mt76_alloc_device);
 int mt76_register_device(struct mt76_dev *dev, bool vht,
 			 struct ieee80211_rate *rates, int n_rates)
 {
+	struct sched_param sparam = {.sched_priority = 1};
 	struct ieee80211_hw *hw = dev->hw;
 	struct mt76_phy *phy = &dev->phy;
 	int ret;
@@ -486,7 +486,14 @@ int mt76_register_device(struct mt76_dev *dev, bool vht,
 			return ret;
 	}
 
-	return ieee80211_register_hw(hw);
+	ret = ieee80211_register_hw(hw);
+	if (ret)
+		return ret;
+
+	WARN_ON(mt76_worker_setup(hw, &dev->tx_worker, mt76_tx_worker, "tx"));
+	sched_setscheduler(dev->tx_worker.task, SCHED_FIFO, &sparam);
+
+	return 0;
 }
 EXPORT_SYMBOL_GPL(mt76_register_device);
 
@@ -503,6 +510,7 @@ EXPORT_SYMBOL_GPL(mt76_unregister_device);
 
 void mt76_free_device(struct mt76_dev *dev)
 {
+	mt76_worker_teardown(&dev->tx_worker);
 	if (dev->wq) {
 		destroy_workqueue(dev->wq);
 		dev->wq = NULL;
diff --git a/mt76.h b/mt76.h
index 64a75393..f82df8ca 100644
--- a/mt76.h
+++ b/mt76.h
@@ -610,7 +610,7 @@ struct mt76_dev {
 	const struct mt76_queue_ops *queue_ops;
 	int tx_dma_idx[4];
 
-	struct tasklet_struct tx_tasklet;
+	struct mt76_worker tx_worker;
 	struct napi_struct tx_napi;
 	struct delayed_work mac_work;
 
@@ -920,7 +920,7 @@ void mt76_stop_tx_queues(struct mt76_dev *dev, struct ieee80211_sta *sta,
 void mt76_tx_check_agg_ssn(struct ieee80211_sta *sta, struct sk_buff *skb);
 void mt76_txq_schedule(struct mt76_phy *phy, enum mt76_txq_id qid);
 void mt76_txq_schedule_all(struct mt76_phy *phy);
-void mt76_tx_tasklet(unsigned long data);
+void mt76_tx_worker(struct mt76_worker *w);
 void mt76_release_buffered_frames(struct ieee80211_hw *hw,
 				  struct ieee80211_sta *sta,
 				  u16 tids, int nframes,
diff --git a/mt7603/dma.c b/mt7603/dma.c
index a08b8528..9ec86da9 100644
--- a/mt7603/dma.c
+++ b/mt7603/dma.c
@@ -165,7 +165,7 @@ static int mt7603_poll_tx(struct napi_struct *napi, int budget)
 
 	mt7603_mac_sta_poll(dev);
 
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 
 	return 0;
 }
@@ -249,6 +249,5 @@ void mt7603_dma_cleanup(struct mt7603_dev *dev)
 		   MT_WPDMA_GLO_CFG_RX_DMA_EN |
 		   MT_WPDMA_GLO_CFG_TX_WRITEBACK_DONE);
 
-	tasklet_kill(&dev->mt76.tx_tasklet);
 	mt76_dma_cleanup(&dev->mt76);
 }
diff --git a/mt7603/mac.c b/mt7603/mac.c
index 625dc6d5..cbf21f40 100644
--- a/mt7603/mac.c
+++ b/mt7603/mac.c
@@ -1405,7 +1405,7 @@ static void mt7603_mac_watchdog_reset(struct mt7603_dev *dev)
 	/* lock/unlock all queues to ensure that no tx is pending */
 	mt76_txq_schedule_all(&dev->mphy);
 
-	tasklet_disable(&dev->mt76.tx_tasklet);
+	mt76_worker_disable(&dev->mt76.tx_worker);
 	tasklet_disable(&dev->mt76.pre_tbtt_tasklet);
 	napi_disable(&dev->mt76.napi[0]);
 	napi_disable(&dev->mt76.napi[1]);
@@ -1454,7 +1454,7 @@ skip_dma_reset:
 	clear_bit(MT76_RESET, &dev->mphy.state);
 	mutex_unlock(&dev->mt76.mutex);
 
-	tasklet_enable(&dev->mt76.tx_tasklet);
+	mt76_worker_enable(&dev->mt76.tx_worker);
 	napi_enable(&dev->mt76.tx_napi);
 	napi_schedule(&dev->mt76.tx_napi);
 
diff --git a/mt7615/dma.c b/mt7615/dma.c
index 3b375fdf..09504b0b 100644
--- a/mt7615/dma.c
+++ b/mt7615/dma.c
@@ -118,7 +118,7 @@ static int mt7615_poll_tx(struct napi_struct *napi, int budget)
 	mt7615_tx_cleanup(dev);
 
 	mt7615_pm_power_save_sched(dev);
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 
 	if (napi_complete_done(napi, 0))
 		mt7615_irq_enable(dev, MT_INT_TX_DONE_ALL);
@@ -324,6 +324,5 @@ void mt7615_dma_cleanup(struct mt7615_dev *dev)
 		   MT_WPDMA_GLO_CFG_RX_DMA_EN);
 	mt76_set(dev, MT_WPDMA_GLO_CFG, MT_WPDMA_GLO_CFG_SW_RESET);
 
-	tasklet_kill(&dev->mt76.tx_tasklet);
 	mt76_dma_cleanup(&dev->mt76);
 }
diff --git a/mt7615/mac.c b/mt7615/mac.c
index e3663c7c..6306e714 100644
--- a/mt7615/mac.c
+++ b/mt7615/mac.c
@@ -1447,7 +1447,7 @@ static void mt7615_mac_tx_free(struct mt7615_dev *dev, struct sk_buff *skb)
 	rcu_read_unlock();
 
 	mt7615_pm_power_save_sched(dev);
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 }
 
 void mt7615_queue_rx_skb(struct mt76_dev *mdev, enum mt76_rxq_id q,
@@ -1872,7 +1872,7 @@ void mt7615_pm_wake_work(struct work_struct *work)
 	}
 	spin_unlock_bh(&dev->pm.txq_lock);
 
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 
 out:
 	ieee80211_wake_queues(mphy->hw);
@@ -2114,7 +2114,7 @@ void mt7615_mac_reset_work(struct work_struct *work)
 	if (ext_phy)
 		mt76_txq_schedule_all(ext_phy);
 
-	tasklet_disable(&dev->mt76.tx_tasklet);
+	mt76_worker_disable(&dev->mt76.tx_worker);
 	napi_disable(&dev->mt76.napi[0]);
 	napi_disable(&dev->mt76.napi[1]);
 	napi_disable(&dev->mt76.tx_napi);
@@ -2135,7 +2135,7 @@ void mt7615_mac_reset_work(struct work_struct *work)
 	clear_bit(MT76_MCU_RESET, &dev->mphy.state);
 	clear_bit(MT76_RESET, &dev->mphy.state);
 
-	tasklet_enable(&dev->mt76.tx_tasklet);
+	mt76_worker_enable(&dev->mt76.tx_worker);
 	napi_enable(&dev->mt76.tx_napi);
 	napi_schedule(&dev->mt76.tx_napi);
 
diff --git a/mt7615/main.c b/mt7615/main.c
index ff2a9489..267faba3 100644
--- a/mt7615/main.c
+++ b/mt7615/main.c
@@ -707,7 +707,7 @@ mt7615_wake_tx_queue(struct ieee80211_hw *hw, struct ieee80211_txq *txq)
 	}
 
 	dev->pm.last_activity = jiffies;
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 }
 
 static void mt7615_tx(struct ieee80211_hw *hw,
diff --git a/mt7615/pci.c b/mt7615/pci.c
index b9794f8a..dbd29d89 100644
--- a/mt7615/pci.c
+++ b/mt7615/pci.c
@@ -88,7 +88,7 @@ static int mt7615_pci_suspend(struct pci_dev *pdev, pm_message_t state)
 	}
 
 	napi_disable(&mdev->tx_napi);
-	tasklet_kill(&mdev->tx_tasklet);
+	mt76_worker_disable(&mdev->tx_worker);
 
 	mt76_for_each_q_rx(mdev, i) {
 		napi_disable(&mdev->napi[i]);
@@ -162,6 +162,7 @@ static int mt7615_pci_resume(struct pci_dev *pdev)
 	if (pdma_reset)
 		dev_err(mdev->dev, "PDMA engine must be reinitialized\n");
 
+	mt76_worker_enable(&mdev->tx_worker);
 	mt76_for_each_q_rx(mdev, i) {
 		napi_enable(&mdev->napi[i]);
 		napi_schedule(&mdev->napi[i]);
diff --git a/mt7615/usb.c b/mt7615/usb.c
index f1b4a631..7e83c350 100644
--- a/mt7615/usb.c
+++ b/mt7615/usb.c
@@ -179,9 +179,7 @@ static int mt7663u_suspend(struct usb_interface *intf, pm_message_t state)
 	}
 
 	mt76u_stop_rx(&dev->mt76);
-
 	mt76u_stop_tx(&dev->mt76);
-	tasklet_kill(&dev->mt76.tx_tasklet);
 
 	return 0;
 }
diff --git a/mt76x0/pci.c b/mt76x0/pci.c
index 5e567483..dda11c70 100644
--- a/mt76x0/pci.c
+++ b/mt76x0/pci.c
@@ -204,7 +204,7 @@ static void mt76x0e_cleanup(struct mt76x02_dev *dev)
 	tasklet_disable(&dev->mt76.pre_tbtt_tasklet);
 	mt76x0_chip_onoff(dev, false, false);
 	mt76x0e_stop_hw(dev);
-	mt76x02_dma_cleanup(dev);
+	mt76_dma_cleanup(&dev->mt76);
 	mt76x02_mcu_cleanup(dev);
 }
 
diff --git a/mt76x02_dma.h b/mt76x02_dma.h
index 4aff4f8e..23b0e7d1 100644
--- a/mt76x02_dma.h
+++ b/mt76x02_dma.h
@@ -61,6 +61,5 @@ mt76x02_wait_for_wpdma(struct mt76_dev *dev, int timeout)
 
 int mt76x02_dma_init(struct mt76x02_dev *dev);
 void mt76x02_dma_disable(struct mt76x02_dev *dev);
-void mt76x02_dma_cleanup(struct mt76x02_dev *dev);
 
 #endif /* __MT76x02_DMA_H */
diff --git a/mt76x02_mmio.c b/mt76x02_mmio.c
index 1389ba1f..156fed61 100644
--- a/mt76x02_mmio.c
+++ b/mt76x02_mmio.c
@@ -150,9 +150,11 @@ static void mt76x02_process_tx_status_fifo(struct mt76x02_dev *dev)
 		mt76x02_send_tx_status(dev, &stat, &update);
 }
 
-static void mt76x02_tx_tasklet(unsigned long data)
+static void mt76x02_tx_worker(struct mt76_worker *w)
 {
-	struct mt76x02_dev *dev = (struct mt76x02_dev *)data;
+	struct mt76x02_dev *dev;
+
+	dev = container_of(w, struct mt76x02_dev, mt76.tx_worker);
 
 	mt76x02_mac_poll_tx_status(dev, false);
 	mt76x02_process_tx_status_fifo(dev);
@@ -177,7 +179,7 @@ static int mt76x02_poll_tx(struct napi_struct *napi, int budget)
 	for (i = MT_TXQ_MCU; i >= 0; i--)
 		mt76_queue_tx_cleanup(dev, i, false);
 
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 
 	return 0;
 }
@@ -196,8 +198,7 @@ int mt76x02_dma_init(struct mt76x02_dev *dev)
 	if (!status_fifo)
 		return -ENOMEM;
 
-	tasklet_init(&dev->mt76.tx_tasklet, mt76x02_tx_tasklet,
-		     (unsigned long)dev);
+	dev->mt76.tx_worker.fn = mt76x02_tx_worker;
 	tasklet_init(&dev->mt76.pre_tbtt_tasklet, mt76x02_pre_tbtt_tasklet,
 		     (unsigned long)dev);
 
@@ -325,13 +326,6 @@ static void mt76x02_dma_enable(struct mt76x02_dev *dev)
 		   MT_WPDMA_GLO_CFG_TX_WRITEBACK_DONE);
 }
 
-void mt76x02_dma_cleanup(struct mt76x02_dev *dev)
-{
-	tasklet_kill(&dev->mt76.tx_tasklet);
-	mt76_dma_cleanup(&dev->mt76);
-}
-EXPORT_SYMBOL_GPL(mt76x02_dma_cleanup);
-
 void mt76x02_dma_disable(struct mt76x02_dev *dev)
 {
 	u32 val = mt76_rr(dev, MT_WPDMA_GLO_CFG);
@@ -449,7 +443,7 @@ static void mt76x02_watchdog_reset(struct mt76x02_dev *dev)
 	set_bit(MT76_RESET, &dev->mphy.state);
 
 	tasklet_disable(&dev->mt76.pre_tbtt_tasklet);
-	tasklet_disable(&dev->mt76.tx_tasklet);
+	mt76_worker_disable(&dev->mt76.tx_worker);
 	napi_disable(&dev->mt76.tx_napi);
 
 	mt76_for_each_q_rx(&dev->mt76, i) {
@@ -506,7 +500,7 @@ static void mt76x02_watchdog_reset(struct mt76x02_dev *dev)
 
 	clear_bit(MT76_RESET, &dev->mphy.state);
 
-	tasklet_enable(&dev->mt76.tx_tasklet);
+	mt76_worker_enable(&dev->mt76.tx_worker);
 	napi_enable(&dev->mt76.tx_napi);
 	napi_schedule(&dev->mt76.tx_napi);
 
diff --git a/mt76x2/pci.c b/mt76x2/pci.c
index 377cf9c0..4d50dad2 100644
--- a/mt76x2/pci.c
+++ b/mt76x2/pci.c
@@ -113,7 +113,7 @@ mt76x2e_suspend(struct pci_dev *pdev, pm_message_t state)
 
 	napi_disable(&mdev->tx_napi);
 	tasklet_kill(&mdev->pre_tbtt_tasklet);
-	tasklet_kill(&mdev->tx_tasklet);
+	mt76_worker_disable(&mdev->tx_worker);
 
 	mt76_for_each_q_rx(mdev, i)
 		napi_disable(&mdev->napi[i]);
@@ -147,6 +147,7 @@ mt76x2e_resume(struct pci_dev *pdev)
 
 	pci_restore_state(pdev);
 
+	mt76_worker_enable(&mdev->tx_worker);
 	mt76_for_each_q_rx(mdev, i) {
 		napi_enable(&mdev->napi[i]);
 		napi_schedule(&mdev->napi[i]);
diff --git a/mt76x2/pci_init.c b/mt76x2/pci_init.c
index 101a0fe0..48a3ebc9 100644
--- a/mt76x2/pci_init.c
+++ b/mt76x2/pci_init.c
@@ -283,7 +283,7 @@ void mt76x2_cleanup(struct mt76x02_dev *dev)
 	tasklet_disable(&dev->dfs_pd.dfs_tasklet);
 	tasklet_disable(&dev->mt76.pre_tbtt_tasklet);
 	mt76x2_stop_hardware(dev);
-	mt76x02_dma_cleanup(dev);
+	mt76_dma_cleanup(&dev->mt76);
 	mt76x02_mcu_cleanup(dev);
 }
 
diff --git a/mt7915/dma.c b/mt7915/dma.c
index bdc69460..c8829987 100644
--- a/mt7915/dma.c
+++ b/mt7915/dma.c
@@ -273,6 +273,5 @@ void mt7915_dma_cleanup(struct mt7915_dev *dev)
 		 MT_WFDMA0_RST_DMASHDL_ALL_RST |
 		 MT_WFDMA0_RST_LOGIC_RST);
 
-	tasklet_kill(&dev->mt76.tx_tasklet);
 	mt76_dma_cleanup(&dev->mt76);
 }
diff --git a/mt7915/mac.c b/mt7915/mac.c
index efdef23c..353f7edc 100644
--- a/mt7915/mac.c
+++ b/mt7915/mac.c
@@ -1010,7 +1010,7 @@ void mt7915_mac_tx_free(struct mt7915_dev *dev, struct sk_buff *skb)
 	dev_kfree_skb(skb);
 
 	mt7915_mac_sta_poll(dev);
-	tasklet_schedule(&dev->mt76.tx_tasklet);
+	mt76_worker_schedule(&dev->mt76.tx_worker);
 }
 
 void mt7915_tx_complete_skb(struct mt76_dev *mdev, enum mt76_txq_id qid,
@@ -1283,7 +1283,7 @@ void mt7915_mac_reset_work(struct work_struct *work)
 	if (ext_phy)
 		mt76_txq_schedule_all(ext_phy);
 
-	tasklet_disable(&dev->mt76.tx_tasklet);
+	mt76_worker_disable(&dev->mt76.tx_worker);
 	napi_disable(&dev->mt76.napi[0]);
 	napi_disable(&dev->mt76.napi[1]);
 	napi_disable(&dev->mt76.napi[2]);
@@ -1303,7 +1303,7 @@ void mt7915_mac_reset_work(struct work_struct *work)
 	clear_bit(MT76_MCU_RESET, &dev->mphy.state);
 	clear_bit(MT76_RESET, &dev->mphy.state);
 
-	tasklet_enable(&dev->mt76.tx_tasklet);
+	mt76_worker_enable(&dev->mt76.tx_worker);
 	napi_enable(&dev->mt76.tx_napi);
 	napi_schedule(&dev->mt76.tx_napi);
 
diff --git a/testmode.c b/testmode.c
index 75bb02cd..3be4e2ff 100644
--- a/testmode.c
+++ b/testmode.c
@@ -160,7 +160,7 @@ mt76_testmode_tx_start(struct mt76_dev *dev)
 	td->tx_queued = 0;
 	td->tx_done = 0;
 	td->tx_pending = td->tx_count;
-	tasklet_schedule(&dev->tx_tasklet);
+	mt76_worker_schedule(&dev->tx_worker);
 }
 
 static void
@@ -168,11 +168,11 @@ mt76_testmode_tx_stop(struct mt76_dev *dev)
 {
 	struct mt76_testmode_data *td = &dev->test;
 
-	tasklet_disable(&dev->tx_tasklet);
+	mt76_worker_disable(&dev->tx_worker);
 
 	td->tx_pending = 0;
 
-	tasklet_enable(&dev->tx_tasklet);
+	mt76_worker_enable(&dev->tx_worker);
 
 	wait_event_timeout(dev->tx_wait, td->tx_done == td->tx_queued, 10 * HZ);
 
diff --git a/tx.c b/tx.c
index 85bd8c6b..af82915b 100644
--- a/tx.c
+++ b/tx.c
@@ -576,9 +576,9 @@ void mt76_txq_schedule_all(struct mt76_phy *phy)
 }
 EXPORT_SYMBOL_GPL(mt76_txq_schedule_all);
 
-void mt76_tx_tasklet(unsigned long data)
+void mt76_tx_worker(struct mt76_worker *w)
 {
-	struct mt76_dev *dev = (struct mt76_dev *)data;
+	struct mt76_dev *dev = container_of(w, struct mt76_dev, tx_worker);
 
 	mt76_txq_schedule_all(&dev->phy);
 	if (dev->phy2)
@@ -621,7 +621,7 @@ void mt76_wake_tx_queue(struct ieee80211_hw *hw, struct ieee80211_txq *txq)
 	if (!test_bit(MT76_STATE_RUNNING, &phy->state))
 		return;
 
-	tasklet_schedule(&dev->tx_tasklet);
+	mt76_worker_schedule(&dev->tx_worker);
 }
 EXPORT_SYMBOL_GPL(mt76_wake_tx_queue);
 
diff --git a/usb.c b/usb.c
index 5f19f9e5..e0450f5a 100644
--- a/usb.c
+++ b/usb.c
@@ -792,9 +792,9 @@ int mt76u_resume_rx(struct mt76_dev *dev)
 }
 EXPORT_SYMBOL_GPL(mt76u_resume_rx);
 
-static void mt76u_tx_tasklet(unsigned long data)
+static void mt76u_tx_worker(struct mt76_worker *w)
 {
-	struct mt76_dev *dev = (struct mt76_dev *)data;
+	struct mt76_dev *dev = container_of(w, struct mt76_dev, tx_worker);
 	struct mt76_queue_entry entry;
 	struct mt76_sw_queue *sq;
 	struct mt76_queue *q;
@@ -882,7 +882,7 @@ static void mt76u_complete_tx(struct urb *urb)
 		dev_err(dev->dev, "tx urb failed: %d\n", urb->status);
 	e->done = true;
 
-	tasklet_schedule(&dev->tx_tasklet);
+	mt76_worker_schedule(&dev->tx_worker);
 }
 
 static int
@@ -1037,6 +1037,8 @@ void mt76u_stop_tx(struct mt76_dev *dev)
 	struct mt76_queue *q;
 	int i, j, ret;
 
+	mt76_worker_disable(&dev->tx_worker);
+
 	ret = wait_event_timeout(dev->tx_wait, !mt76_has_tx_pending(&dev->phy),
 				 HZ / 5);
 	if (!ret) {
@@ -1048,8 +1050,6 @@ void mt76u_stop_tx(struct mt76_dev *dev)
 				usb_kill_urb(q->entry[j].urb);
 		}
 
-		tasklet_kill(&dev->tx_tasklet);
-
 		/* On device removal we maight queue skb's, but mt76u_tx_kick()
 		 * will fail to submit urb, cleanup those skb's manually.
 		 */
@@ -1072,6 +1072,8 @@ void mt76u_stop_tx(struct mt76_dev *dev)
 	cancel_work_sync(&dev->usb.stat_work);
 	clear_bit(MT76_READING_STATS, &dev->phy.state);
 
+	mt76_worker_enable(&dev->tx_worker);
+
 	mt76_tx_status_check(dev, NULL, true);
 }
 EXPORT_SYMBOL_GPL(mt76u_stop_tx);
@@ -1121,8 +1123,8 @@ int mt76u_init(struct mt76_dev *dev,
 	mt76u_ops.rmw = ext ? mt76u_rmw_ext : mt76u_rmw;
 	mt76u_ops.write_copy = ext ? mt76u_copy_ext : mt76u_copy;
 
+	dev->tx_worker.fn = mt76u_tx_worker;
 	tasklet_init(&usb->rx_tasklet, mt76u_rx_tasklet, (unsigned long)dev);
-	tasklet_init(&dev->tx_tasklet, mt76u_tx_tasklet, (unsigned long)dev);
 	INIT_WORK(&usb->stat_work, mt76u_tx_status_data);
 
 	usb->data_len = usb_maxpacket(udev, usb_sndctrlpipe(udev, 0), 1);