On Tue, 21 Jan 2014, Sarah Sharp wrote: > On Mon, Jan 20, 2014 at 02:45:11PM -0600, Felipe Balbi wrote: > > If the HW reports the endpoint as Halted, there's > > no need to start any URB for that endpoint, we can > > simply return -EPIPE and this will tell upper layers > > that the endpoint is, indeed, halted. > > > > This patch fixes usbtest's set/clear halt test #13 > > on xHCI-based hosts. > > It looks like that test assumes that the xHCI driver should not allow > URBs to be enqueued to the endpoint when the driver needs to call > usb_clear_halt(). Or is it expecting that the host will attempt the > transfer, and the device will stall the transfer? It expects that the computer will attempt the transfer and the device will reply with a STALL. > The reason the driver still queues URBs is because the xHC halts the > endpoint for events that are outside of stalls. For instance, it will > halt the endpoint on a babble error, a transfer error, or a split > transaction error. The upper layers don't call usb_clear_halt() for > those cases, so the xHCI driver handles the halted endpoint internally. > Look for calls to xhci_requires_manual_halt_cleanup(). > > The xHC also halts the endpoint on a control transfer stall. The upper > layers don't ever call usb_clear_halt on the control endpoint, because > the bus spec says the next control transfer will clear the halted > condition. Therefore the xHCI driver has to handle the halt manually. It sounds like there's a little confusion here. To say that the endpoint is halted is somewhat ambiguous; it could refer to a halt on the host side or a halt on the device side. usb_clear_halt() is meant to clear a halt on the device side. It also tells the HCD that it has done so, in case the HCD is keeping track, but it is not meant for clearing a halt on the host side. usb_reset_endpoint() does that (among other things). > You could try to differentiate between an endpoint halt that requires a > manual cleanup, and ones that the upper layer should handle, perhaps by > adding a new EP_HALTED_MANUAL flag. The driver could accept URBs for > the manual cleanup case, and reject URBs with -EPIPE for the ones the > upper layers should handle. That should not be necessary. The HCD should always accept URB submissions. Nothing will get sent to the device until the HCD clears any host-side halts, but the upper layers don't worry about that because the HCD takes care of it automatically. Contrariwise, the upper layers are responsible for clearing device-side halts. The HCD should ignore that end of things as much as it can. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html