On Tue, Jul 18, 2023 at 07:26:00PM +0800, Xu Yang wrote: > In current design, the ehci driver will not unlink itd/sitds from the > hardware list when dequeue isochronous urbs. Rather just wait until they > complete normally or their time slot expires. However, this will cause > issues if the controller has stopped periodic schedule before finished > all periodic schedule. The urb will not be done forever in this case and > then usb_kill/poison_urb() will always wait there. > > The ChipIdea IP exactly has a bug: if frame babble occurs during periodic > transfer, PE (PORTSC.bit2) will be cleared and the controller will stop > periodic schedule immediately. So if the user tries to kill or poison > related urb, it will wait there since the urb can't be done forever. > > This patch will check if this issue occurs, then it will unlink itd/sitds > from the hardware list depends on the result. This is not the right approach. There already is code in the driver for unlinking itds/sitds when the periodic schedule isn't running: See how the "live" variable is used in scan_isoc(). You don't need to add new code to do the same thing, you simply have to make sure that live is set to false if the controller has stopped the periodic schedule unexpectedly. (Be very careful about handling the case where CMD_PSE is set and STS_PSS is clear. This can happen when the controller has been told to start the periodic schedule but it hasn't done so yet.) Alan Stern
