On Wed, Jun 14, 2017 at 10:26:29AM -0400, Alan Stern wrote:
> On Wed, 14 Jun 2017, Peter Chen wrote:
>
> > On Tue, Jun 13, 2017 at 10:22:26AM -0400, Alan Stern wrote:
> > > On Tue, 13 Jun 2017, Felipe Balbi wrote:
> > >
> > > >
> > > > Hi Alan,
> > > >
> > > > Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> writes:
> > > > > Felipe:
> > > > >
> > > > > A UDC driver will invoke the gadget driver's ->disconnect callback
> > > > > whenever the D+ pullup is turned off. During gadget driver unbinding,
> > > > > this happens automatically when usb_gadget_remove_driver() calls
> > > > > usb_gadget_disconnect().
> > > >
> > > > usb_gadget_disconnect() only calls UDC's ->pullup(), not gadget driver's
> > > > ->disconnect()
> > > >
> > > > int usb_gadget_disconnect(struct usb_gadget *gadget)
> > > > {
> > > > int ret = 0;
> > > >
> > > > if (!gadget->ops->pullup) {
> > > > ret = -EOPNOTSUPP;
> > > > goto out;
> > > > }
> > > >
> > > > if (gadget->deactivated) {
> > > > /*
> > > > * If gadget is deactivated we only save new state.
> > > > * Gadget will stay disconnected after activation.
> > > > */
> > > > gadget->connected = false;
> > > > goto out;
> > > > }
> > > >
> > > > ret = gadget->ops->pullup(gadget, 0);
> > > > if (!ret)
> > > > gadget->connected = 0;
> > > >
> > > > out:
> > > > trace_usb_gadget_disconnect(gadget, ret);
> > > >
> > > > return ret;
> > > > }
> > > > EXPORT_SYMBOL_GPL(usb_gadget_disconnect);
> > >
> > > Yes, but as I mentioned above, the pullup routine calls the gadget
> > > driver's disconnect method.
> > >
> > > > > But immediately thereafter, usb_gadget_remove_driver() calls the gadget
> > > > > driver's ->disconnect callback directly! Do we have any reason for
> > > > > doing this? I don't see point to it. Should that call be removed?
> > > >
> > > > Unless I'm missing something, that call is necessary :-) Have you faced
> > > > any issues with it? Are there any UDCs calling gadget driver's
> > > > ->disconnect() from ->pullup() ?
> > >
> > > I haven't faced any issues because gadget drivers don't seem to mind
> > > ->disconnect getting called twice in a row. :-) (And note that the
> > > API doesn't have a corresponding ->connect callback... Although to
> > > tell the truth, it's not clear what a gadget driver needs to do in
> > > either callback. Can we simply remove ->disconnect altogether?)
> > >
> > > Both dummy-hcd and net2280 call ->disconnect from ->pullup. I haven't
> > > checked other UDC drivers, but it seems like something they should all
> > > do. Except perhaps in the case where the UDC driver doesn't have a
> > > pullup method.
> > >
> >
> > No, ->pullup is expected to be called from UDC core, and the UDC core
> > makes sure the coming ->disconnect at kinds of situations.
>
> I think you are wrong. The UDC core calls ->pullup from the
> usb_gadget_disconnect() routine, but it does not call ->disconnect from
> that routine.
>
You may misunderstand my point, I mean there are always ->disconnect
call after ->pullup at UDC core, eg, usb_gadget_remove_driver and
usb_udc_softconn_store. So individual UDC driver does not need
->disconnect at its ->pullup implementation.
> Furthermore, there is an advantage to having the UDC driver call
> ->disconnect from within its ->pullup method: It can retain its private
> spinlock across the ->disconnect call. This will help prevent races.
> For example, what would happen if ->disconnect was called while a
> ->setup callback was in progress? Or vice versa?
>
You mean the private UDC driver spinlock can help prevent races well for
->setup and ->pullup in UDC driver? Yes, it is the good point. I am
wondering why composite_disconnect has spinlock and without for
composite_setup.
--
Best Regards,
Peter Chen
--
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
