Re: [PATCH v2] Bluetooth: Shutdown controller after workqueues are flushed or cancelled

[Mattijs Korpershoek]

Hi Kai-Heng,

Kai-Heng Feng <kai.heng.feng@xxxxxxxxxxxxx> writes:

> Hi Mattijs,
>
>> [snipped]
>
> Thanks for your testing. I think I finally got it:
> btmtksdio_shutdown()
> -> mtk_hci_wmt_sync()
>  -> __hci_cmd_send()
>   then waiting for BTMTKSDIO_TX_WAIT_VND_EVT, which is cleared in
> btmtksdio_recv_event():
> btmtksdio_recv_event()
> -> hci_recv_frame()
>  -> queue_work(hdev->workqueue, &hdev->rx_work);
>
> That means it has to be done before the following drain_workqueue() call.
> Can you please see if moving the ->shutdown() part right before
> drain_workqueue() can fix the issue?
I've tested the following patch:

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 2560ed2f144d..131e69a9a66a 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -1757,6 +1757,14 @@ int hci_dev_do_close(struct hci_dev *hdev)
                        cancel_delayed_work_sync(&adv_instance->rpa_expired_cb);
        }
 
+       if (!hci_dev_test_flag(hdev, HCI_UNREGISTER) &&
+           !hci_dev_test_flag(hdev, HCI_USER_CHANNEL) &&
+           test_bit(HCI_UP, &hdev->flags)) {
+               /* Execute vendor specific shutdown routine */
+               if (hdev->shutdown)
+                       hdev->shutdown(hdev);
+       }
+
        /* Avoid potential lockdep warnings from the *_flush() calls by
         * ensuring the workqueue is empty up front.
         */
@@ -1798,14 +1806,6 @@ int hci_dev_do_close(struct hci_dev *hdev)
                clear_bit(HCI_INIT, &hdev->flags);
        }
 
-       if (!hci_dev_test_flag(hdev, HCI_UNREGISTER) &&
-           !hci_dev_test_flag(hdev, HCI_USER_CHANNEL) &&
-           test_bit(HCI_UP, &hdev->flags)) {
-               /* Execute vendor specific shutdown routine */
-               if (hdev->shutdown)
-                       hdev->shutdown(hdev);
-       }
-
        /* flush cmd  work */
        flush_work(&hdev->cmd_work);

It does not seem to fix the issue.
Adding the bits in btmtksdio_flush() does not change the result of the
above patch.

Here are the logs. These are just with the above patch (not with the
btmtksdio_flush() changes.

dmesg: https://pastebin.com/FZZBkqGC
btmtksdio_ftrace: https://pastebin.com/JQ0UWenY

Mattijs
>
> Kai-Heng
>
>>
>> Mattijs
>>
>> >
>> > Kai-Heng