Re: [syzbot] WARNING in call_timer_fn

[Luiz Augusto von Dentz <luiz.dentz@gmail.com>]

Hi Hillf,

On Thu, Nov 17, 2022 at 5:28 PM Hillf Danton <hdanton@sina.com> wrote:
>
> On Thu, Nov 17, 2022 at 1:04 PM Luiz Augusto von Dentz <luiz.dentz@gmail.com> wrote:
> > >
> > > I thought we would have something similar to shutdown_timer (e.g.
> > > shutdown_delayed_work) so we can safely free its object/struct, at
> > > least that was the impression I got when discussing with Steven.
>
> Because of the cough in Bluetooth's throat does not mean it makes sense
> to ask workqueue to take a flu jab. Why not cure it directly locally,
> given the cases of workqueue under the drivers dir. And timer?

Like Thomas said we can only resolve this partially with the likes of
cancel_workqueue(_sync), though we can use HCI_UNREGISTER to stop
rescheduling cmd_timer, so imo having something similar to
shutdown_timer makes things simpler for subsystems/drivers, anyway I'm
fine fixing it directly since it is probably going to be simpler to
backport, perhaps something like the following is all we need:

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 0540555b3704..977684f5fb57 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -4090,6 +4090,7 @@ static void hci_cmd_work(struct work_struct *work)

                        rcu_read_lock();
                        if (test_bit(HCI_RESET, &hdev->flags) ||
+                           test_bit(HCI_UNREGISTER, &hdev->flags) ||
                            hci_dev_test_flag(hdev, HCI_CMD_DRAIN_WORKQUEUE))
                                cancel_delayed_work(&hdev->cmd_timer);
                        else

-- 
Luiz Augusto von Dentz