>> Can this be related due to the fact that we use a signle-threaded >> workqueue for delete/reset/reconnect? (delete cancel_sync the active >> reconnect work...) >> >> Does this untested patch help? > > That seems to do it! Is this a formal tested-by?