See above. Until we can rely on HCD being finished with unlinks, I'd make disconnect() methods be sure everything they submitted has been completed.
(One requirement for disconnect methods has always been that the urbs must have completed; that hasn't changed.)
Hi Dave, thanks for your answer. Why this last requirement? Doesn't it make "fire and forget" style urb submission awkward?
The reason to have usbcore clean up the urbs was primarily to make sure that the HC/HCD state didn't get corrupted, as happened previously. And to create synchronization points so that config change events could be handled properly on hardware (like OHCI and EHCI, but not UHCI) that maintains endpoint-specific state.
The fact that, with OHCI and EHCI, it happened to also be sufficient to enable "fire and forget" policies in the USB device drivers was (for some) a happy side effect.
But as you noted, since UHCI doesn't disable() endpoints, that requirement is still present ... although I seem to recall that someone recently post a patch to teach UHCI how to implement disable(), and that'd likely be handy.
- Dave
------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
