On Tue, 30 Jul 2013, Julius Werner wrote: > > Wait a moment. Why does each of these attempts lead to a 5-second > > timeout? Why don't they fail immediately? > > Now that you mention it, that's a very good question. I have brought this up with Sarah on more than one occasion, but we never found a good answer. The effects are quite visible when somebody unplugs a USB-3 disk drive in the middle of a data transfer. > The kernel > enqueues a control transfer to the now disconnected address because > it's internal bookkeeping is not yet updated, but I guess that should > usually fail very fast from an xHC-internal transaction timeout. I > have now tried to debug the cause of this: I see the transfer being > enqueued and the doorbell triggered, but never get a transfer event > back from it (until the software timeout calls usb_kill_urb() which > stops the endpoint). With the same setup on a PantherPoint system I > get the same U1/U2 disable control messages on unplugging, but they > fail within <5ms with a transaction error... so I guess this must be a > LynxPoint hardware bug. An odd sort of bug. You'd think that not getting a response back would be one of the first types of error the hardware designers would check for. > Regardless, calling usb_reset_device() is still wrong and will at > least lead to pointless transfer attempts and error messages, so I > think my patch still has merit. > > > What will happen here if udev is NULL? Maybe you should change the > > test to (!udev || !(portstatus & ...)). > > Right... I'm not sure if that can happen in practice, but I'll change > it just in case. Somebody said that in theory, ports can put themselves in the Disabled state at any time, spontaneously. If this happened just after a device was attached, you would end up with udev being NULL and the connect status being set. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html