This is a requirement which has always existed but, somehow, wasn't
reflected in the documentation and problems weren't found until now
when Tuba Yavuz found a possible deadlock happening between dwc3 and
f_hid. She described the situation as follows:

spin_lock_irqsave(&hidg->write_spinlock, flags); // first acquire
/* we our function has been disabled by host */
if (!hidg->req) {
	free_ep_req(hidg->in_ep, hidg->req);
	goto try_again;
}

[...]

status = usb_ep_queue(hidg->in_ep, hidg->req, GFP_ATOMIC);
=>
	[...]
	=> usb_gadget_giveback_request
		=>
		f_hidg_req_complete
			=>
			spin_lock_irqsave(&hidg->write_spinlock, flags); // second acquire

Note that this happens because dwc3 would call ->complete() on a
failed usb_ep_queue() due to failed Start Transfer command. This is,
anyway, a theoretical situation because dwc3 currently uses "No
Response Update Transfer" command for Bulk and Interrupt endpoints.

It's still good to make this case impossible to happen even if the "No
Reponse Update Transfer" command is changed.
Reported-by: Tuba Yavuz <tuba@ece.ufl.edu>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>