usb: chipidea: udc: don't stall endpoint if request list is empty in …

…isr_tr_complete_low

When attaching an imx28 or imx53 in USB gadget mode to a Windows host and
starting a rndis connection we see this message every 4-10 seconds:

    g_ether gadget: high speed config #2: RNDIS

Analysis shows that each time this message is printed, the rndis connection is
re-establish due to a reset because of a stalled endpoint (ep 0, dir 1). The
endpoint is stalled because the reqeust complete bit on that endpoint is set,
but in isr_tr_complete_low() the endpoint request list (mEp->qh.queue) is
empty.

This patch removed this check, because the code doesn't take the following
situation into account:

The loop over all endpoints in isr_tr_complete_handler() will call ep_nuke() on
both ep0/dir0 and ep/dir1 in the first loop. Pending reqeusts will be flushed
and completed here. There seems to be a race condition, the request is nuked,
but the request complete bit will be set, too. The subsequent check (in
ep0/dir1's loop cycle) for endpoint request list (mEp->qh.queue) empty will
fail.

Both other mainline chipidea drivers (mv_udc_core.c and fsl_udc_core.c) don't
have this check.

Cc: stable <stable@vger.kernel.org>
Signed-off-by: Michael Grzeschik <m.grzeschik@pengutronix.de>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/usb/chipidea/udc.c

@@ -771,10 +771,7 @@ __acquires(mEp->lock)
 {
 	struct ci13xxx_req *mReq, *mReqTemp;
 	struct ci13xxx_ep *mEpTemp = mEp;
-	int uninitialized_var(retval);
-
-	if (list_empty(&mEp->qh.queue))
-		return -EINVAL;
+	int retval = 0;
 
 	list_for_each_entry_safe(mReq, mReqTemp, &mEp->qh.queue,
 			queue) {