On Thu, Jan 24, 2013 at 03:35:58PM +0100, Michael Grzeschik wrote: > On Thu, Jan 24, 2013 at 03:21:37PM +0100, Laurent Pinchart wrote: > > Hi Michael, > > > > On Thursday 24 January 2013 14:54:01 Michael Grzeschik wrote: > > > On Thu, Jan 24, 2013 at 12:02:12PM +0100, Laurent Pinchart wrote: > > > > On Thursday 24 January 2013 11:22:02 Michael Grzeschik wrote: > > > > > The current implementation of the composite framework connects the > > > > > gadget function to the device-controller at kernel boottime. > > > > > > > > > > It's possible that the userspace application, which handles the > > > > > enumeration requests of the gadget, starts with high latency when the > > > > > host did already gave up to detect the gadget. In that case the > > > > > gadget-application will not work in the first case. > > > > > > > > This shouldn't happen, as the UVC gadget driver should start disconnected > > > > (it calls usb_function_deactive() in uvc_function_bind()). If the host > > > > can enumerate the gadget before the application starts it would then > > > > likely be a device controller driver bug, not a UVC gadget driver bug. > > > > > > Unlikely this is not the case, as usb_gadget_probe_driver triggers an > > > usb_gadget_connect() right after driver->bind() was called. > > > > Indeed, for new-style UDC that's correct. > > > > Isn't that an issue with the framework ? I would expect gadget drivers to be > > able to start disconnected. > > That probably is. But it might only be a problem for gadgets with > userspace components which handle ep0 requests. For other gadgets, which > are not as dynamic as an webcam, it might be enough to have its > enumeration handled in kernel and start right on. > > That needs further discussion. Therefore my patch is an RFC and a simple > solution to ship arround that issue. Should we take it though? Framework will take care of it, but other controllers need to be fixed up first so they don't connect their pullup directly. We're getting close to that, though. Give me another merge window and gadget drivers will be required to manage pullup. -- balbi
Attachment:
signature.asc
Description: Digital signature
