wifi: mt76: mt7925: Fix unregister deadlock
Subsystems (2)
PATCH 1/1 wifi: mt76: mt7925: Fix unregister deadlock
During device shutdown or removal, a deadlock can occur between the PCIe remove path and the driver's asynchronous reset work.
The unregistration path calls napi_disable() before cancelling the reset work. If the reset work runs concurrently, it may re-enable NAPI and schedule it. Because the device is being unregistered, this can lead to NAPI state corruption where NAPI is marked as scheduled but never polled, causing subsequent napi_disable() calls to hang forever.
Fix this by:
1. Moving cancel_work_sync(&dev->reset_work) to the very start of mt7925e_unregister_device(), ensuring it is stopped before NAPI is disabled.
2. Setting the MT76_REMOVED flag early in the PCI remove path to prevent new reset work from being queued.
3. Checking MT76_REMOVED in mt7925_mac_reset_work() and aborting the reset early if the device is being removed.
--- drivers/net/wireless/mediatek/mt76/mt7925/mac.c | 7 +++++++ drivers/net/wireless/mediatek/mt76/mt7925/pci.c | 4 ++-- 2 files changed, 9 insertions(+), 2 deletions(-)
drivers/net/wireless/mediatek/mt76/mt7925/mac.c +7 -0
@@ -1319,6 +1319,9 @@ void mt7925_mac_reset_work(struct work_struct *work)cancel_work_sync (&pm->wake_work);for (i =0 ; i <10 ; i++) { +if (test_bit (MT76_REMOVED , &dev->mt76.phy.state)) + goto out; +mutex_lock (&dev->mt76.mutex); ret =mt792x_dev_reset (dev);mutex_unlock (&dev->mt76.mutex);
@@ -1338,8 +1341,12 @@ void mt7925_mac_reset_work(struct work_struct *work)ieee80211_scan_completed (dev->mphy.hw, &info); } +out: dev->hw_full_reset = false; pm->suspended = false; +if (test_bit (MT76_REMOVED , &dev->mt76.phy.state)) +return ; +ieee80211_wake_queues (hw); ieee80211_iterate_active_interfaces(hw, IEEE80211_IFACE_ITER_RESUME_ALL,
drivers/net/wireless/mediatek/mt76/mt7925/pci.c +2 -2
@@ -43,13 +43,13 @@ static void mt7925e_unregister_device(struct mt792x_dev *dev)if (dev->phy.chip_cap & MT792x_CHIP_CAP_WF_RF_PIN_CTRL_EVT_EN)wiphy_rfkill_stop_polling (hw->wiphy); +cancel_work_sync (&dev->reset_work);cancel_work_sync (&dev->init_work);mt76_unregister_device (&dev->mt76);mt76_for_each_q_rx (&dev->mt76, i)napi_disable (&dev->mt76.napi[i]);cancel_delayed_work_sync (&pm->ps_work);cancel_work_sync (&pm->wake_work); -cancel_work_sync (&dev->reset_work);mt7925_tx_token_put (dev);__mt792x_mcu_drv_pmctrl (dev);
@@ -498,8 +498,8 @@ static void mt7925_pci_remove(struct pci_dev *pdev)struct mt76_dev *mdev =pci_get_drvdata (pdev); struct mt792x_dev *dev = container_of(mdev, struct mt792x_dev, mt76); -mt7925e_unregister_device (dev);set_bit (MT76_REMOVED , &mdev->phy.state); +mt7925e_unregister_device (dev);devm_free_irq (&pdev->dev, pdev->irq, dev);mt76_free_device (&dev->mt76);pci_free_irq_vectors (pdev);
Signature · 1 line
2.45.2
I was having issues with the MT7925 on the ASRock B850M with Bluetooth. I tested this and other patches in my investigation.
My issue was a platform bug causing a failure in USB enumeration before the driver path (unrelated to the patch).
I can confirm this patch does not introduce regressions on this platform. I tested system power cycling and suspension, BT/Wifi adapter enable/disable, and also modprobing in/out mt7925{e/u}
Patch applied on top of mailine (7.2 rc1).
Thanks,
Rafael Passos