A couple of bugs in the driver are preventing SETUP packets
with an OUT data phase from working properly.

Interestingly those are incredibly rare (RNDIS typically
uses them and thus is broken without this fix).

The main problem was an incorrect register offset being
applied for arming RX on EP0. The other problem relates
to stalling such a packet before the data phase, in which
case we don't get an ACK cycle, and get the next SETUP
packet directly, so we shouldn't reject it.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>