Merge tag 'dwc3-for-v3.7' of git://git.kernel.org/pub/scm/linux/kernel/git/balbi/usb into usb-next

usb: dwc3: patches for v3.7 merge window

Some much needed changes for our dwc3 driver. First there's a
rework on the ep0 handling due to some Silicon issue we uncovered
which affects all users of this IP core (there's a missing
XferNotReady(DATA) event in some conditions). This issue which
show up as a SETUP transfers which wouldn't complete ever and
we would fail TD 7.06 of the Link Layer Test from USB-IF and
Lecroy's USB3 Exerciser.

We also fix a long standing bug regarding EP0 enable sequencing
where we weren't setting a particular bit (Ignore Sequence
Number). Since we never saw any problems caused by that, it
didn't deserve being sent to stable tree.

On this pull request we also fix Burst Size initialization which
should be done only in SuperSpeed and we were mistakenly setting
Burst Size to the maximum value on non-SuperSpeed mode. Again,
since we never saw any problems caused by that, we're not sending
this patch to stable.

There's also a memory ordering fix regarding usage of bitmaps in
dwc3 driver.

You will also find some sparse warnings fix, a fix for missed
isochronous packets when the endpoint is already busy, and a
fix for synchronization delay on dwc3_stop_active_transfer().

4 files changed, 170 insertions, 155 deletions

diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
index 79a24fab13d..bed2c161546 100644
--- a/drivers/usb/dwc3/core.c
+++ b/drivers/usb/dwc3/core.c
@@ -100,6 +100,7 @@ void dwc3_put_device_id(int id)
 
 	ret = test_bit(id, dwc3_devs);
 	WARN(!ret, "dwc3: ID %d not in use\n", id);
+	smp_mb__before_clear_bit();
 	clear_bit(id, dwc3_devs);
 }
 EXPORT_SYMBOL_GPL(dwc3_put_device_id);
@@ -462,7 +463,7 @@ static int __devinit dwc3_probe(struct platform_device *pdev)
 		return -ENOMEM;
 	}
 
-	regs = devm_ioremap(dev, res->start, resource_size(res));
+	regs = devm_ioremap_nocache(dev, res->start, resource_size(res));
 	if (!regs) {
 		dev_err(dev, "ioremap failed\n");
 		return -ENOMEM;
diff --git a/drivers/usb/dwc3/core.h b/drivers/usb/dwc3/core.h
index dbc5713d84f..243affc9343 100644
--- a/drivers/usb/dwc3/core.h
+++ b/drivers/usb/dwc3/core.h
@@ -457,7 +457,6 @@ enum dwc3_phy {
 enum dwc3_ep0_next {
 	DWC3_EP0_UNKNOWN = 0,
 	DWC3_EP0_COMPLETE,
-	DWC3_EP0_NRDY_SETUP,
 	DWC3_EP0_NRDY_DATA,
 	DWC3_EP0_NRDY_STATUS,
 };
@@ -784,7 +783,6 @@ struct dwc3_event_depevt {
 #define DEPEVT_STREAMEVT_NOTFOUND	2
 
 /* Control-only Status */
-#define DEPEVT_STATUS_CONTROL_SETUP	0
 #define DEPEVT_STATUS_CONTROL_DATA	1
 #define DEPEVT_STATUS_CONTROL_STATUS	2
 
diff --git a/drivers/usb/dwc3/ep0.c b/drivers/usb/dwc3/ep0.c
index 9b94886b66e..1bba97ba218 100644
--- a/drivers/usb/dwc3/ep0.c
+++ b/drivers/usb/dwc3/ep0.c
@@ -125,7 +125,6 @@ static int __dwc3_gadget_ep0_queue(struct dwc3_ep *dep,
 		struct dwc3_request *req)
 {
 	struct dwc3		*dwc = dep->dwc;
-	int			ret = 0;
 
 	req->request.actual	= 0;
 	req->request.status	= -EINPROGRESS;
@@ -156,16 +155,72 @@ static int __dwc3_gadget_ep0_queue(struct dwc3_ep *dep,
 
 		dep->flags &= ~(DWC3_EP_PENDING_REQUEST |
 				DWC3_EP0_DIR_IN);
-	} else if (dwc->delayed_status) {
+
+		return 0;
+	}
+
+	/*
+	 * In case gadget driver asked us to delay the STATUS phase,
+	 * handle it here.
+	 */
+	if (dwc->delayed_status) {
+		unsigned	direction;
+
+		direction = !dwc->ep0_expect_in;
 		dwc->delayed_status = false;
 
 		if (dwc->ep0state == EP0_STATUS_PHASE)
-			__dwc3_ep0_do_control_status(dwc, dwc->eps[1]);
+			__dwc3_ep0_do_control_status(dwc, dwc->eps[direction]);
 		else
 			dev_dbg(dwc->dev, "too early for delayed status\n");
+
+		return 0;
 	}
 
-	return ret;
+	/*
+	 * Unfortunately we have uncovered a limitation wrt the Data Phase.
+	 *
+	 * Section 9.4 says we can wait for the XferNotReady(DATA) event to
+	 * come before issueing Start Transfer command, but if we do, we will
+	 * miss situations where the host starts another SETUP phase instead of
+	 * the DATA phase.  Such cases happen at least on TD.7.6 of the Link
+	 * Layer Compliance Suite.
+	 *
+	 * The problem surfaces due to the fact that in case of back-to-back
+	 * SETUP packets there will be no XferNotReady(DATA) generated and we
+	 * will be stuck waiting for XferNotReady(DATA) forever.
+	 *
+	 * By looking at tables 9-13 and 9-14 of the Databook, we can see that
+	 * it tells us to start Data Phase right away. It also mentions that if
+	 * we receive a SETUP phase instead of the DATA phase, core will issue
+	 * XferComplete for the DATA phase, before actually initiating it in
+	 * the wire, with the TRB's status set to "SETUP_PENDING". Such status
+	 * can only be used to print some debugging logs, as the core expects
+	 * us to go through to the STATUS phase and start a CONTROL_STATUS TRB,
+	 * just so it completes right away, without transferring anything and,
+	 * only then, we can go back to the SETUP phase.
+	 *
+	 * Because of this scenario, SNPS decided to change the programming
+	 * model of control transfers and support on-demand transfers only for
+	 * the STATUS phase. To fix the issue we have now, we will always wait
+	 * for gadget driver to queue the DATA phase's struct usb_request, then
+	 * start it right away.
+	 *
+	 * If we're actually in a 2-stage transfer, we will wait for
+	 * XferNotReady(STATUS).
+	 */
+	if (dwc->three_stage_setup) {
+		unsigned        direction;
+
+		direction = dwc->ep0_expect_in;
+		dwc->ep0state = EP0_DATA_PHASE;
+
+		__dwc3_ep0_do_control_data(dwc, dwc->eps[direction], req);
+
+		dep->flags &= ~DWC3_EP0_DIR_IN;
+	}
+
+	return 0;
 }
 
 int dwc3_gadget_ep0_queue(struct usb_ep *ep, struct usb_request *request,
@@ -207,9 +262,14 @@ out:
 
 static void dwc3_ep0_stall_and_restart(struct dwc3 *dwc)
 {
-	struct dwc3_ep		*dep = dwc->eps[0];
+	struct dwc3_ep		*dep;
+
+	/* reinitialize physical ep1 */
+	dep = dwc->eps[1];
+	dep->flags = DWC3_EP_ENABLED;
 
 	/* stall is always issued on EP0 */
+	dep = dwc->eps[0];
 	__dwc3_gadget_ep_set_halt(dep, 1);
 	dep->flags = DWC3_EP_ENABLED;
 	dwc->delayed_status = false;
@@ -698,6 +758,7 @@ static void dwc3_ep0_complete_data(struct dwc3 *dwc,
 	struct dwc3_trb		*trb;
 	struct dwc3_ep		*ep0;
 	u32			transferred;
+	u32			status;
 	u32			length;
 	u8			epnum;
 
@@ -710,6 +771,17 @@ static void dwc3_ep0_complete_data(struct dwc3 *dwc,
 	ur = &r->request;
 
 	trb = dwc->ep0_trb;
+
+	status = DWC3_TRB_SIZE_TRBSTS(trb->size);
+	if (status == DWC3_TRBSTS_SETUP_PENDING) {
+		dev_dbg(dwc->dev, "Setup Pending received\n");
+
+		if (r)
+			dwc3_gadget_giveback(ep0, r, -ECONNRESET);
+
+		return;
+	}
+
 	length = trb->size & DWC3_TRB_SIZE_MASK;
 
 	if (dwc->ep0_bounced) {
@@ -746,8 +818,11 @@ static void dwc3_ep0_complete_status(struct dwc3 *dwc,
 {
 	struct dwc3_request	*r;
 	struct dwc3_ep		*dep;
+	struct dwc3_trb		*trb;
+	u32			status;
 
 	dep = dwc->eps[0];
+	trb = dwc->ep0_trb;
 
 	if (!list_empty(&dep->request_list)) {
 		r = next_request(&dep->request_list);
@@ -767,6 +842,10 @@ static void dwc3_ep0_complete_status(struct dwc3 *dwc,
 		}
 	}
 
+	status = DWC3_TRB_SIZE_TRBSTS(trb->size);
+	if (status == DWC3_TRBSTS_SETUP_PENDING)
+		dev_dbg(dwc->dev, "Setup Pending received\n");
+
 	dwc->ep0state = EP0_SETUP_PHASE;
 	dwc3_ep0_out_start(dwc);
 }
@@ -800,12 +879,6 @@ static void dwc3_ep0_xfer_complete(struct dwc3 *dwc,
 	}
 }
 
-static void dwc3_ep0_do_control_setup(struct dwc3 *dwc,
-		const struct dwc3_event_depevt *event)
-{
-	dwc3_ep0_out_start(dwc);
-}
-
 static void __dwc3_ep0_do_control_data(struct dwc3 *dwc,
 		struct dwc3_ep *dep, struct dwc3_request *req)
 {
@@ -858,29 +931,6 @@ static void __dwc3_ep0_do_control_data(struct dwc3 *dwc,
 	WARN_ON(ret < 0);
 }
 
-static void dwc3_ep0_do_control_data(struct dwc3 *dwc,
-		const struct dwc3_event_depevt *event)
-{
-	struct dwc3_ep		*dep;
-	struct dwc3_request	*req;
-
-	dep = dwc->eps[0];
-
-	if (list_empty(&dep->request_list)) {
-		dev_vdbg(dwc->dev, "pending request for EP0 Data phase\n");
-		dep->flags |= DWC3_EP_PENDING_REQUEST;
-
-		if (event->endpoint_number)
-			dep->flags |= DWC3_EP0_DIR_IN;
-		return;
-	}
-
-	req = next_request(&dep->request_list);
-	dep = dwc->eps[event->endpoint_number];
-
-	__dwc3_ep0_do_control_data(dwc, dep, req);
-}
-
 static int dwc3_ep0_start_control_status(struct dwc3_ep *dep)
 {
 	struct dwc3		*dwc = dep->dwc;
@@ -912,100 +962,61 @@ static void dwc3_ep0_do_control_status(struct dwc3 *dwc,
 	__dwc3_ep0_do_control_status(dwc, dep);
 }
 
-static void dwc3_ep0_xfernotready(struct dwc3 *dwc,
-		const struct dwc3_event_depevt *event)
+static void dwc3_ep0_end_control_data(struct dwc3 *dwc, struct dwc3_ep *dep)
 {
-	dwc->setup_packet_pending = true;
-
-	/*
-	 * This part is very tricky: If we have just handled
-	 * XferNotReady(Setup) and we're now expecting a
-	 * XferComplete but, instead, we receive another
-	 * XferNotReady(Setup), we should STALL and restart
-	 * the state machine.
-	 *
-	 * In all other cases, we just continue waiting
-	 * for the XferComplete event.
-	 *
-	 * We are a little bit unsafe here because we're
-	 * not trying to ensure that last event was, indeed,
-	 * XferNotReady(Setup).
-	 *
-	 * Still, we don't expect any condition where that
-	 * should happen and, even if it does, it would be
-	 * another error condition.
-	 */
-	if (dwc->ep0_next_event == DWC3_EP0_COMPLETE) {
-		switch (event->status) {
-		case DEPEVT_STATUS_CONTROL_SETUP:
-			dev_vdbg(dwc->dev, "Unexpected XferNotReady(Setup)\n");
-			dwc3_ep0_stall_and_restart(dwc);
-			break;
-		case DEPEVT_STATUS_CONTROL_DATA:
-			/* FALLTHROUGH */
-		case DEPEVT_STATUS_CONTROL_STATUS:
-			/* FALLTHROUGH */
-		default:
-			dev_vdbg(dwc->dev, "waiting for XferComplete\n");
-		}
+	struct dwc3_gadget_ep_cmd_params params;
+	u32			cmd;
+	int			ret;
 
+	if (!dep->resource_index)
 		return;
-	}
-
-	switch (event->status) {
-	case DEPEVT_STATUS_CONTROL_SETUP:
-		dev_vdbg(dwc->dev, "Control Setup\n");
 
-		dwc->ep0state = EP0_SETUP_PHASE;
+	cmd = DWC3_DEPCMD_ENDTRANSFER;
+	cmd |= DWC3_DEPCMD_CMDIOC;
+	cmd |= DWC3_DEPCMD_PARAM(dep->resource_index);
+	memset(&params, 0, sizeof(params));
+	ret = dwc3_send_gadget_ep_cmd(dwc, dep->number, cmd, &params);
+	WARN_ON_ONCE(ret);
+	dep->resource_index = 0;
+}
 
-		dwc3_ep0_do_control_setup(dwc, event);
-		break;
+static void dwc3_ep0_xfernotready(struct dwc3 *dwc,
+		const struct dwc3_event_depevt *event)
+{
+	dwc->setup_packet_pending = true;
 
+	switch (event->status) {
 	case DEPEVT_STATUS_CONTROL_DATA:
 		dev_vdbg(dwc->dev, "Control Data\n");
 
-		dwc->ep0state = EP0_DATA_PHASE;
-
-		if (dwc->ep0_next_event != DWC3_EP0_NRDY_DATA) {
-			dev_vdbg(dwc->dev, "Expected %d got %d\n",
-					dwc->ep0_next_event,
-					DWC3_EP0_NRDY_DATA);
-
-			dwc3_ep0_stall_and_restart(dwc);
-			return;
-		}
-
 		/*
-		 * One of the possible error cases is when Host _does_
-		 * request for Data Phase, but it does so on the wrong
-		 * direction.
+		 * We already have a DATA transfer in the controller's cache,
+		 * if we receive a XferNotReady(DATA) we will ignore it, unless
+		 * it's for the wrong direction.
 		 *
-		 * Here, we already know ep0_next_event is DATA (see above),
-		 * so we only need to check for direction.
+		 * In that case, we must issue END_TRANSFER command to the Data
+		 * Phase we already have started and issue SetStall on the
+		 * control endpoint.
 		 */
 		if (dwc->ep0_expect_in != event->endpoint_number) {
+			struct dwc3_ep	*dep = dwc->eps[dwc->ep0_expect_in];
+
 			dev_vdbg(dwc->dev, "Wrong direction for Data phase\n");
+			dwc3_ep0_end_control_data(dwc, dep);
 			dwc3_ep0_stall_and_restart(dwc);
 			return;
 		}
 
-		dwc3_ep0_do_control_data(dwc, event);
 		break;
 
 	case DEPEVT_STATUS_CONTROL_STATUS:
+		if (dwc->ep0_next_event != DWC3_EP0_NRDY_STATUS)
+			return;
+
 		dev_vdbg(dwc->dev, "Control Status\n");
 
 		dwc->ep0state = EP0_STATUS_PHASE;
 
-		if (dwc->ep0_next_event != DWC3_EP0_NRDY_STATUS) {
-			dev_vdbg(dwc->dev, "Expected %d got %d\n",
-					dwc->ep0_next_event,
-					DWC3_EP0_NRDY_STATUS);
-
-			dwc3_ep0_stall_and_restart(dwc);
-			return;
-		}
-
 		if (dwc->delayed_status) {
 			WARN_ON_ONCE(event->endpoint_number != 1);
 			dev_vdbg(dwc->dev, "Mass Storage delayed status\n");
diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
index 58fdfad96b4..ba444e7f9c4 100644
--- a/drivers/usb/dwc3/gadget.c
+++ b/drivers/usb/dwc3/gadget.c
@@ -431,15 +431,25 @@ static int dwc3_gadget_start_config(struct dwc3 *dwc, struct dwc3_ep *dep)
 
 static int dwc3_gadget_set_ep_config(struct dwc3 *dwc, struct dwc3_ep *dep,
 		const struct usb_endpoint_descriptor *desc,
-		const struct usb_ss_ep_comp_descriptor *comp_desc)
+		const struct usb_ss_ep_comp_descriptor *comp_desc,
+		bool ignore)
 {
 	struct dwc3_gadget_ep_cmd_params params;
 
 	memset(&params, 0x00, sizeof(params));
 
 	params.param0 = DWC3_DEPCFG_EP_TYPE(usb_endpoint_type(desc))
-		| DWC3_DEPCFG_MAX_PACKET_SIZE(usb_endpoint_maxp(desc))
-		| DWC3_DEPCFG_BURST_SIZE(dep->endpoint.maxburst - 1);
+		| DWC3_DEPCFG_MAX_PACKET_SIZE(usb_endpoint_maxp(desc));
+
+	/* Burst size is only needed in SuperSpeed mode */
+	if (dwc->gadget.speed == USB_SPEED_SUPER) {
+		u32 burst = dep->endpoint.maxburst - 1;
+
+		params.param0 |= DWC3_DEPCFG_BURST_SIZE(burst);
+	}
+
+	if (ignore)
+		params.param0 |= DWC3_DEPCFG_IGN_SEQ_NUM;
 
 	params.param1 = DWC3_DEPCFG_XFER_COMPLETE_EN
 		| DWC3_DEPCFG_XFER_NOT_READY_EN;
@@ -498,7 +508,8 @@ static int dwc3_gadget_set_xfer_resource(struct dwc3 *dwc, struct dwc3_ep *dep)
  */
 static int __dwc3_gadget_ep_enable(struct dwc3_ep *dep,
 		const struct usb_endpoint_descriptor *desc,
-		const struct usb_ss_ep_comp_descriptor *comp_desc)
+		const struct usb_ss_ep_comp_descriptor *comp_desc,
+		bool ignore)
 {
 	struct dwc3		*dwc = dep->dwc;
 	u32			reg;
@@ -510,7 +521,7 @@ static int __dwc3_gadget_ep_enable(struct dwc3_ep *dep,
 			return ret;
 	}
 
-	ret = dwc3_gadget_set_ep_config(dwc, dep, desc, comp_desc);
+	ret = dwc3_gadget_set_ep_config(dwc, dep, desc, comp_desc, ignore);
 	if (ret)
 		return ret;
 
@@ -558,27 +569,7 @@ static void dwc3_remove_requests(struct dwc3 *dwc, struct dwc3_ep *dep)
 	if (!list_empty(&dep->req_queued)) {
 		dwc3_stop_active_transfer(dwc, dep->number);
 
-		/*
-		 * NOTICE: We are violating what the Databook says about the
-		 * EndTransfer command. Ideally we would _always_ wait for the
-		 * EndTransfer Command Completion IRQ, but that's causing too
-		 * much trouble synchronizing between us and gadget driver.
-		 *
-		 * We have discussed this with the IP Provider and it was
-		 * suggested to giveback all requests here, but give HW some
-		 * extra time to synchronize with the interconnect. We're using
-		 * an arbitraty 100us delay for that.
-		 *
-		 * Note also that a similar handling was tested by Synopsys
-		 * (thanks a lot Paul) and nothing bad has come out of it.
-		 * In short, what we're doing is:
-		 *
-		 * - Issue EndTransfer WITH CMDIOC bit set
-		 * - Wait 100us
-		 * - giveback all requests to gadget driver
-		 */
-		udelay(100);
-
+		/* - giveback all requests to gadget driver */
 		while (!list_empty(&dep->req_queued)) {
 			req = next_request(&dep->req_queued);
 
@@ -657,6 +648,12 @@ static int dwc3_gadget_ep_enable(struct usb_ep *ep,
 	dep = to_dwc3_ep(ep);
 	dwc = dep->dwc;
 
+	if (dep->flags & DWC3_EP_ENABLED) {
+		dev_WARN_ONCE(dwc->dev, true, "%s is already enabled\n",
+				dep->name);
+		return 0;
+	}
+
 	switch (usb_endpoint_type(desc)) {
 	case USB_ENDPOINT_XFER_CONTROL:
 		strlcat(dep->name, "-control", sizeof(dep->name));
@@ -674,16 +671,10 @@ static int dwc3_gadget_ep_enable(struct usb_ep *ep,
 		dev_err(dwc->dev, "invalid endpoint transfer type\n");
 	}
 
-	if (dep->flags & DWC3_EP_ENABLED) {
-		dev_WARN_ONCE(dwc->dev, true, "%s is already enabled\n",
-				dep->name);
-		return 0;
-	}
-
 	dev_vdbg(dwc->dev, "Enabling %s\n", dep->name);
 
 	spin_lock_irqsave(&dwc->lock, flags);
-	ret = __dwc3_gadget_ep_enable(dep, desc, ep->comp_desc);
+	ret = __dwc3_gadget_ep_enable(dep, desc, ep->comp_desc, false);
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
 	return ret;
@@ -1087,15 +1078,10 @@ static int __dwc3_gadget_ep_queue(struct dwc3_ep *dep, struct dwc3_request *req)
 	 *
 	 */
 	if (dep->flags & DWC3_EP_PENDING_REQUEST) {
-		int	ret;
-
 		ret = __dwc3_gadget_kick_transfer(dep, 0, true);
-		if (ret && ret != -EBUSY) {
-			struct dwc3	*dwc = dep->dwc;
-
+		if (ret && ret != -EBUSY)
 			dev_dbg(dwc->dev, "%s: failed to kick transfers\n",
 					dep->name);
-		}
 	}
 
 	/*
@@ -1104,16 +1090,14 @@ static int __dwc3_gadget_ep_queue(struct dwc3_ep *dep, struct dwc3_request *req)
 	 *    core may not see the modified TRB(s).
 	 */
 	if (usb_endpoint_xfer_isoc(dep->endpoint.desc) &&
-			(dep->flags & DWC3_EP_BUSY)) {
+			(dep->flags & DWC3_EP_BUSY) &&
+			!(dep->flags & DWC3_EP_MISSED_ISOC)) {
 		WARN_ON_ONCE(!dep->resource_index);
 		ret = __dwc3_gadget_kick_transfer(dep, dep->resource_index,
 				false);
-		if (ret && ret != -EBUSY) {
-			struct dwc3	*dwc = dep->dwc;
-
+		if (ret && ret != -EBUSY)
 			dev_dbg(dwc->dev, "%s: failed to kick transfers\n",
 					dep->name);
-		}
 	}
 
 	/*
@@ -1518,14 +1502,14 @@ static int dwc3_gadget_start(struct usb_gadget *g,
 	dwc3_gadget_ep0_desc.wMaxPacketSize = cpu_to_le16(512);
 
 	dep = dwc->eps[0];
-	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL);
+	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, false);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
 		goto err0;
 	}
 
 	dep = dwc->eps[1];
-	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL);
+	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, false);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
 		goto err1;
@@ -1750,7 +1734,7 @@ static void dwc3_endpoint_transfer_complete(struct dwc3 *dwc,
 		int		i;
 
 		for (i = 0; i < DWC3_ENDPOINTS_NUM; i++) {
-			struct dwc3_ep	*dep = dwc->eps[i];
+			dep = dwc->eps[i];
 
 			if (!(dep->flags & DWC3_EP_ENABLED))
 				continue;
@@ -1877,6 +1861,25 @@ static void dwc3_stop_active_transfer(struct dwc3 *dwc, u32 epnum)
 	if (!dep->resource_index)
 		return;
 
+	/*
+	 * NOTICE: We are violating what the Databook says about the
+	 * EndTransfer command. Ideally we would _always_ wait for the
+	 * EndTransfer Command Completion IRQ, but that's causing too
+	 * much trouble synchronizing between us and gadget driver.
+	 *
+	 * We have discussed this with the IP Provider and it was
+	 * suggested to giveback all requests here, but give HW some
+	 * extra time to synchronize with the interconnect. We're using
+	 * an arbitraty 100us delay for that.
+	 *
+	 * Note also that a similar handling was tested by Synopsys
+	 * (thanks a lot Paul) and nothing bad has come out of it.
+	 * In short, what we're doing is:
+	 *
+	 * - Issue EndTransfer WITH CMDIOC bit set
+	 * - Wait 100us
+	 */
+
 	cmd = DWC3_DEPCMD_ENDTRANSFER;
 	cmd |= DWC3_DEPCMD_HIPRI_FORCERM | DWC3_DEPCMD_CMDIOC;
 	cmd |= DWC3_DEPCMD_PARAM(dep->resource_index);
@@ -1884,6 +1887,8 @@ static void dwc3_stop_active_transfer(struct dwc3 *dwc, u32 epnum)
 	ret = dwc3_send_gadget_ep_cmd(dwc, dep->number, cmd, &params);
 	WARN_ON_ONCE(ret);
 	dep->resource_index = 0;
+
+	udelay(100);
 }
 
 static void dwc3_stop_active_transfers(struct dwc3 *dwc)
@@ -2141,14 +2146,14 @@ static void dwc3_gadget_conndone_interrupt(struct dwc3 *dwc)
 	}
 
 	dep = dwc->eps[0];
-	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL);
+	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, true);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
 		return;
 	}
 
 	dep = dwc->eps[1];
-	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL);
+	ret = __dwc3_gadget_ep_enable(dep, &dwc3_gadget_ep0_desc, NULL, true);
 	if (ret) {
 		dev_err(dwc->dev, "failed to enable %s\n", dep->name);
 		return;