Re: Testing endpoint halt support for raw-gadget

[Alan Stern <stern@rowland.harvard.edu>]

On Fri, 10 Apr 2020, Andrey Konovalov wrote:

> On Fri, Apr 10, 2020 at 2:29 AM Alan Stern <stern@rowland.harvard.edu> wrote:

> > Have you implemented wedge as well as halt?  Wedge is needed for the
> > mass-storage protocol; as far as I know it isn't used anywhere else.
> 
> No, I didn't know about "wedge" at all :) Looks like the API for it is
> really simple, just usb_ep_set_wedge(). I'll need to figure out what
> it is and how it works, and I'll send a patch that adds halt/wedge
> support then.

usb_ep_set_wedge(ep) does almost the same thing as 
usb_ep_set_halt(ep).  The difference is that a Clear-Feature(halt) 
request from the host will un-halt an endpoint if it is merely halted, 
but it won't un-halt a wedged endpoint.  (I don't think this is 
documented anywhere, unfortunately.)

> > And have you given any thought to suspend/resume support?  It's a bit
> > tricky because you have to consider both gadget suspend and USB bus
> > suspend.
> 
> Hm, no. Is there something specific I need to consider to support it?
> I guess I'll need to read about how it works as well, before I can
> understand what it would require and ask meaningful questions.

The really tricky part involves a gadget that is in system suspend.  If 
the USB bus isn't also suspended, the gadget won't work properly -- it 
won't be able to respond to requests from the host.  Basically, when a 
UDC driver sees that the system is going into suspend, it has no choice 
but to disconnect from the USB bus.  This probably isn't implemented 
very well in a lot of UDC drivers.

USB bus suspend, on the other hand, _should_ be implemented.

> > Nothing else springs to mind.
> 
> Something else: I've been testing raw-gadget with various UDCs that I
> have [1] and everything seems to work, except for emulating SuperSpeed
> devices with net2280. I've just found it out yesterday night, and
> haven't had a chance to debug that yet, but if you know about some
> potential issues I could encounter with SuperSpeed/USB3+, please let
> me know.

Well, USB-3 has streams, unlike USB-2.  You may want to think about
supporting them.

Also, bear in mind that dummy-hcd doesn't support isochronous transfers 
(although all real UDCs do support them).  So perhaps you haven't given 
them as much testing.  usbtest can help a little with that.

Alan Stern