Hi, looking at this code, which fixed a deadlock,it looks to me like it has introduced another deadlock. [assume a storage device on a hub being suspended] hub_suspend() -> hub_quiesce() -> flush_work(&hub->tt.clear_work) -> /* note that tt.clear_work is not on its own queue, it uses simple schedule_work(). Hence the work lands on system_wq, shared with arbitrary works */ kmalloc(... , GFP_KERNEL) -> usb_storage or uas -> usb_autopm_get_interface() -> DEADLOCK I think you need to use a dedicated workqueue for the hub driver. Regards Oliver
