[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CABBYNZJZeP1eFZXrVPMAhEtdNVSWkuAkyopLcrUG6wbQpvgsww@mail.gmail.com>
Date: Fri, 31 Oct 2025 10:35:22 -0400
From: Luiz Augusto von Dentz <luiz.dentz@...il.com>
To: Linmao Li <lilinmao@...inos.cn>
Cc: linux-bluetooth@...r.kernel.org, marcel@...tmann.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] Bluetooth: btusb: Prevent autosuspend when
le_scan_disable work is pending
Hi,
On Fri, Oct 31, 2025 at 4:15 AM Linmao Li <lilinmao@...inos.cn> wrote:
>
> When USB autosuspend occurs while le_scan_disable work is scheduled,
> HCI commands sent by the work fail with timeout, leaving LE scan in
> an inconsistent state.
>
> Scenario:
> T=0: LE scan starts, le_scan_disable work queued (+10240ms)
> T=8s: Autosuspend check: tx_in_flight=0, suspend proceeds
> T=10s: le_scan_disable work executes on suspended device
> → HCI command 0x2042 times out
>
> The tx_in_flight check only protects actively transmitted URBs, missing
> the window where work is queued but hasn't submitted its URB yet.
>
> Fix by checking delayed_work_pending(&hdev->le_scan_disable) during
> autosuspend. Return -EBUSY if pending to block suspend until work
> completes. Only set BTUSB_SUSPENDING after all checks pass to avoid
> leaving the flag set if suspend is aborted.
Hmm, we could also just cancel the work then, in face we do have
hci_suspend_sync->hci_stop_discovery_sync->cancel_delayed_work(&hdev->le_scan_disable);
but perhaps it is not being called always because HCI_LE_SCAN is not
enabled, so I wonder if we should do something like:
diff --git a/net/bluetooth/hci_sync.c b/net/bluetooth/hci_sync.c
index 1cbdd2ce03f2..59618fde7bcb 100644
--- a/net/bluetooth/hci_sync.c
+++ b/net/bluetooth/hci_sync.c
@@ -5464,6 +5464,11 @@ int hci_stop_discovery_sync(struct hci_dev *hdev)
bt_dev_dbg(hdev, "state %u", hdev->discovery.state);
+ /* Always stop le_scan_disable since that works as discovery timer for
+ * the rounds of discovery irrespective of the discovery type.
+ */
+ cancel_delayed_work(&hdev->le_scan_disable);
+
if (d->state == DISCOVERY_FINDING || d->state == DISCOVERY_STOPPING) {
if (test_bit(HCI_INQUIRY, &hdev->flags)) {
err = __hci_cmd_sync_status(hdev, HCI_OP_INQUIRY_CANCEL,
@@ -5472,14 +5477,9 @@ int hci_stop_discovery_sync(struct hci_dev *hdev)
return err;
}
- if (hci_dev_test_flag(hdev, HCI_LE_SCAN)) {
- cancel_delayed_work(&hdev->le_scan_disable);
-
- err = hci_scan_disable_sync(hdev);
- if (err)
- return err;
- }
-
+ err = hci_scan_disable_sync(hdev);
+ if (err)
+ return err;
} else {
err = hci_scan_disable_sync(hdev);
if (err)
We may as well rename/reworkd le_scan_disable to discovery.work since
that acts as I commented above it works as discovery timer not just
le_scan_disable.
> Signed-off-by: Linmao Li <lilinmao@...inos.cn>
> ---
> drivers/bluetooth/btusb.c | 14 ++++++++++----
> 1 file changed, 10 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/bluetooth/btusb.c b/drivers/bluetooth/btusb.c
> index 5e9ebf0c5312..a344ea1dc466 100644
> --- a/drivers/bluetooth/btusb.c
> +++ b/drivers/bluetooth/btusb.c
> @@ -4389,6 +4389,7 @@ static void btusb_disconnect(struct usb_interface *intf)
> static int btusb_suspend(struct usb_interface *intf, pm_message_t message)
> {
> struct btusb_data *data = usb_get_intfdata(intf);
> + struct hci_dev *hdev = data->hdev;
>
> BT_DBG("intf %p", intf);
>
> @@ -4402,14 +4403,19 @@ static int btusb_suspend(struct usb_interface *intf, pm_message_t message)
> return 0;
>
> spin_lock_irq(&data->txlock);
> - if (!(PMSG_IS_AUTO(message) && data->tx_in_flight)) {
> - set_bit(BTUSB_SUSPENDING, &data->flags);
> - spin_unlock_irq(&data->txlock);
> - } else {
> + if (PMSG_IS_AUTO(message) && data->tx_in_flight) {
> spin_unlock_irq(&data->txlock);
> data->suspend_count--;
> return -EBUSY;
> }
> + spin_unlock_irq(&data->txlock);
> +
> + if (PMSG_IS_AUTO(message) && delayed_work_pending(&hdev->le_scan_disable)) {
> + data->suspend_count--;
> + return -EBUSY;
> + }
> +
> + set_bit(BTUSB_SUSPENDING, &data->flags);
>
> cancel_work_sync(&data->work);
>
> --
> 2.25.1
>
--
Luiz Augusto von Dentz
Powered by blists - more mailing lists