On Sat, 2021-02-27 at 11:23 +0100, Lorenzo Bianconi wrote:

quoted

Stop Tx burst for ext_phy after mac reset.

Signed-off-by: Ryder Lee <ryder.lee@mediatek.com>
---
 drivers/net/wireless/mediatek/mt76/mt7615/mac.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/net/wireless/mediatek/mt76/mt7615/mac.c b/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
index df2dc77a323a..0f613063c1ff 100644
--- a/drivers/net/wireless/mediatek/mt76/mt7615/mac.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7615/mac.c

@@ -2085,6 +2085,7 @@ void mt7615_mac_reset_work(struct work_struct *work)
 	del_timer_sync(&dev->phy.roc_timer);
 	cancel_work_sync(&dev->phy.roc_work);
 	if (phy2) {
+		set_bit(MT76_RESET, &phy2->mt76->state);

Hi Ryder,

since we run mt76_txq_schedule_all then and mt76_worker_disable few lines below, what is the point
of setting the state to RESET?

Just in case mt76_txq_schedule_* still has the possibility to be called
by some other place at the time reset happens. Also, I think MT76_RESET
is more reliable to break Tx scheduling (from inside of the burst loop).

Ryder

quoted

 		cancel_delayed_work_sync(&phy2->mt76->mac_work);
 		del_timer_sync(&phy2->roc_timer);
 		cancel_work_sync(&phy2->roc_work);

@@ -2118,6 +2119,8 @@ void mt7615_mac_reset_work(struct work_struct *work)
 
 	clear_bit(MT76_MCU_RESET, &dev->mphy.state);
 	clear_bit(MT76_RESET, &dev->mphy.state);
+	if (phy2)
+		clear_bit(MT76_RESET, &phy2->mt76->state);
 
 	mt76_worker_enable(&dev->mt76.tx_worker);
 	napi_enable(&dev->mt76.tx_napi);

-- 
2.18.0