Yes Michal Nazarewicz has seen this on a S3C UDC, 

Good to confirm that.  As I mentioned, other folk have seen much the same bug on other hardware; we
know it's neither a new nor a HW-specific issue.

Correction to an earlier comment of mine: 

https://patchwork.kernel.org/patch/195562/
does correctly identify the commit which
caused this bug in multiple drivers.

However, that patch won't apply to the latest
tree from Linus.

Michal, can you update and resubmit, so I can
at least ack your fix and xpedite its merge?

It'd be good  to have RNDIS work again, a fair
number of folk rely on it, and this seems to be
the main obstacle to continuing to do that.


- Dave