Re: [PATCH v1] Revert "Bluetooth: core: Fix missing power_on work cancel on HCI close"

["Mateusz Jończyk" <mat.jonczyk@o2.pl>]

W dniu 14.06.2022 o 20:17, Max Krummenacher pisze:
> From: Max Krummenacher <max.krummenacher@toradex.com>
>
> This reverts commit ff7f2926114d3a50f5ffe461a9bce8d761748da5.
>
> The commit ff7f2926114d ("Bluetooth: core: Fix missing power_on work
> cancel on HCI close") introduced between v5.18 and v5.19-rc1 makes
> going to suspend freeze. v5.19-rc2 is equally affected.
>
> This has been seen on a Colibri iMX6ULL WB which has a Marvell 8997
> based WiFi / Bluetooth module connected over SDIO.
> [...]

Hello,

commit ff7f2926114d ("Bluetooth: core: Fix missing power_on work cancel on HCI close")

causes problems also on my laptop (HP 17-by0001nw with a Realtek Bluetooth adapter).

I have Bluetooth disabled by default on startup (via systemd-rfkill.service ) and
vanilla kernel 5.19.0-rc5 fails to suspend (the screen turns black, but I am then able to
touch a trackpad and log in). Reverting that commit on top of 5.19.0-rc5 fixes the issue.

On bare 5.19.0-rc5, after startup, the kworker/u9:0+hci0 process hangs indefinitely
with this stacktrace (obtained through "cat /proc/163/stack" )

        [<0>] __flush_work+0x143/0x220
        [<0>] __cancel_work_timer+0x122/0x1a0
        [<0>] cancel_work_sync+0x10/0x20
        [<0>] hci_dev_close_sync+0x2a/0x550 [bluetooth]
        [<0>] hci_dev_do_close+0x2a/0x60 [bluetooth]
        [<0>] hci_power_on+0x91/0x200 [bluetooth]
        [<0>] process_one_work+0x21c/0x3c0
        [<0>] worker_thread+0x4a/0x3a0
        [<0>] kthread+0xcf/0xf0
        [<0>] ret_from_fork+0x22/0x30

It appears that the hci_power_on() function calls hci_dev_do_close() in this block:

    if (hci_dev_test_flag(hdev, HCI_RFKILLED) ||
         /* [...] */ ) {
        hci_dev_clear_flag(hdev, HCI_AUTO_OFF);
        hci_dev_do_close(hdev);
    } else /* [...] */

which then calls hci_dev_close_sync(). With the problematic commit, that function
calls

       cancel_work_sync(&hdev->power_on)

which tries to cancel the execution of the hci_power_on() function's itself, which leads to a deadlock.

When trying to suspend, the "/lib/systemd/systemd-sleep suspend" process hangs on

        [<0>] hci_suspend_dev+0x87/0xf0 [bluetooth]
        [<0>] hci_suspend_notifier+0x38/0x80 [bluetooth]
        [<0>] notifier_call_chain_robust+0x5e/0xc0
        [<0>] blocking_notifier_call_chain_robust+0x42/0x60
        [<0>] pm_notifier_call_chain_robust+0x1d/0x40
        [<0>] pm_suspend+0x116/0x5a0
        [<0>] state_store+0x82/0xe0
        [<0>] kobj_attr_store+0x12/0x20
        [<0>] sysfs_kf_write+0x3e/0x50
        [<0>] kernfs_fop_write_iter+0x138/0x1c0
        [<0>] new_sync_write+0x104/0x180
        [<0>] vfs_write+0x1d7/0x260
        [<0>] ksys_write+0x67/0xe0
        [<0>] __x64_sys_write+0x1a/0x20
        [<0>] do_syscall_64+0x3b/0x90
        [<0>] entry_SYSCALL_64_after_hwframe+0x46/0xb0

My device is:

Bus 001 Device 005: ID 0bda:b00b Realtek Semiconductor Corp. Bluetooth Radio

So probably the
commit ff7f2926114d ("Bluetooth: core: Fix missing power_on work cancel on HCI close")
should be reverted or corrected.

Greetings,

Mateusz Jończyk

> Signed-off-by: Max Krummenacher <max.krummenacher@toradex.com>
> ---
>
>  net/bluetooth/hci_core.c | 2 ++
>  net/bluetooth/hci_sync.c | 1 -
>  2 files changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> index 59a5c1341c26..19df3905c5f8 100644
> --- a/net/bluetooth/hci_core.c
> +++ b/net/bluetooth/hci_core.c
> @@ -2675,6 +2675,8 @@ void hci_unregister_dev(struct hci_dev *hdev)
>  	list_del(&hdev->list);
>  	write_unlock(&hci_dev_list_lock);
>  
> +	cancel_work_sync(&hdev->power_on);
> +
>  	hci_cmd_sync_clear(hdev);
>  
>  	if (!test_bit(HCI_QUIRK_NO_SUSPEND_NOTIFIER, &hdev->quirks))
> diff --git a/net/bluetooth/hci_sync.c b/net/bluetooth/hci_sync.c
> index 286d6767f017..1739e8cb3291 100644
> --- a/net/bluetooth/hci_sync.c
> +++ b/net/bluetooth/hci_sync.c
> @@ -4088,7 +4088,6 @@ int hci_dev_close_sync(struct hci_dev *hdev)
>  
>  	bt_dev_dbg(hdev, "");
>  
> -	cancel_work_sync(&hdev->power_on);
>  	cancel_delayed_work(&hdev->power_off);
>  	cancel_delayed_work(&hdev->ncmd_timer);
>