The hci_req_sync_cancel() is just as much related to the request
cleanup as hci_request_cancel_all() is. Just move the former into the
latter and do the cleanup from a single place in hci_dev_do_close().
The important thing is to avoid deadlocks by holding the req_sync
lock: previously hci_request_cancel_all was done right after releasing
the lock and with this patch it's right before taking it.
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
cancel_delayed_work(&hdev->power_off);
- hci_req_sync_cancel(hdev, ENODEV);
+ hci_request_cancel_all(hdev);
hci_req_sync_lock(hdev);
if (!test_and_clear_bit(HCI_UP, &hdev->flags)) {
hci_req_sync_unlock(hdev);
- hci_request_cancel_all(hdev);
-
hci_dev_put(hdev);
return 0;
}
void hci_request_cancel_all(struct hci_dev *hdev)
{
+ hci_req_sync_cancel(hdev, ENODEV);
+
cancel_work_sync(&hdev->discov_update);
cancel_work_sync(&hdev->bg_scan_update);
cancel_delayed_work_sync(&hdev->le_scan_disable);