usb: dwc3: gadget: Improve END_TRANSFER handling

index 1c8b349379af..da0af11fbc1a 100644
@@ -688,6 +688,7 @@ struct dwc3_ep {
 #define DWC3_EP_STALL		BIT(1)
 #define DWC3_EP_WEDGE		BIT(2)
 #define DWC3_EP_TRANSFER_STARTED BIT(3)
+#define DWC3_EP_END_TRANSFER_PENDING BIT(4)
 #define DWC3_EP_PENDING_REQUEST	BIT(5)
 
 	/* This last one is specific to EP0 */

index fd1b100d2927..6dee4dabc0a4 100644
@@ -1136,8 +1136,10 @@ void dwc3_ep0_interrupt(struct dwc3 *dwc,
 	case DWC3_DEPEVT_EPCMDCMPLT:
 		cmd = DEPEVT_PARAMETER_CMD(event->parameters);
 
-		if (cmd == DWC3_DEPCMD_ENDTRANSFER)
+		if (cmd == DWC3_DEPCMD_ENDTRANSFER) {
+			dep->flags &= ~DWC3_EP_END_TRANSFER_PENDING;
 			dep->flags &= ~DWC3_EP_TRANSFER_STARTED;
+		}
 		break;
 	}
 }

index b3f8514d1f27..218022c261bc 100644
@@ -2621,6 +2621,7 @@ static void dwc3_endpoint_interrupt(struct dwc3 *dwc,
 		cmd = DEPEVT_PARAMETER_CMD(event->parameters);
 
 		if (cmd == DWC3_DEPCMD_ENDTRANSFER) {
+			dep->flags &= ~DWC3_EP_END_TRANSFER_PENDING;
 			dep->flags &= ~DWC3_EP_TRANSFER_STARTED;
 			dwc3_gadget_ep_cleanup_cancelled_requests(dep);
 		}

@@ -2679,7 +2680,8 @@ static void dwc3_stop_active_transfer(struct dwc3_ep *dep, bool force,
 	u32 cmd;
 	int ret;
 
-	if (!(dep->flags & DWC3_EP_TRANSFER_STARTED))
+	if (!(dep->flags & DWC3_EP_TRANSFER_STARTED) ||
+	    (dep->flags & DWC3_EP_END_TRANSFER_PENDING))
 		return;
 
 	/*

@@ -2724,6 +2726,8 @@ static void dwc3_stop_active_transfer(struct dwc3_ep *dep, bool force,
 
 	if (!interrupt)
 		dep->flags &= ~DWC3_EP_TRANSFER_STARTED;
+	else
+		dep->flags |= DWC3_EP_END_TRANSFER_PENDING;
 
 	if (dwc3_is_usb31(dwc) || dwc->revision < DWC3_REVISION_310A)
 		udelay(100);

index da0af11fbc1a..77c4a9abe365 100644
@@ -690,6 +690,7 @@ struct dwc3_ep {
 #define DWC3_EP_TRANSFER_STARTED BIT(3)
 #define DWC3_EP_END_TRANSFER_PENDING BIT(4)
 #define DWC3_EP_PENDING_REQUEST	BIT(5)
+#define DWC3_EP_DELAY_START	BIT(6)
 
 	/* This last one is specific to EP0 */
 #define DWC3_EP0_DIR_IN		BIT(31)

index 218022c261bc..61b7bef98cf9 100644
@@ -1450,6 +1450,12 @@ static int __dwc3_gadget_ep_queue(struct dwc3_ep *dep, struct dwc3_request *req)
 	list_add_tail(&req->list, &dep->pending_list);
 	req->status = DWC3_REQUEST_STATUS_QUEUED;
 
+	/* Start the transfer only after the END_TRANSFER is completed */
+	if (dep->flags & DWC3_EP_END_TRANSFER_PENDING) {
+		dep->flags |= DWC3_EP_DELAY_START;
+		return 0;
+	}
+
 	/*
 	 * NOTICE: Isochronous endpoints should NEVER be prestarted. We must
 	 * wait for a XferNotReady event so we will know what's the current

@@ -2624,6 +2630,11 @@ static void dwc3_endpoint_interrupt(struct dwc3 *dwc,
 			dep->flags &= ~DWC3_EP_END_TRANSFER_PENDING;
 			dep->flags &= ~DWC3_EP_TRANSFER_STARTED;
 			dwc3_gadget_ep_cleanup_cancelled_requests(dep);
+			if ((dep->flags & DWC3_EP_DELAY_START) &&
+			    !usb_endpoint_xfer_isoc(dep->endpoint.desc))
+				__dwc3_gadget_kick_transfer(dep);
+
+			dep->flags &= ~DWC3_EP_DELAY_START;
 		}
 		break;
 	case DWC3_DEPEVT_STREAMEVT:

index 61b7bef98cf9..fe9a7e37725c 100644
@@ -2702,16 +2702,13 @@ static void dwc3_stop_active_transfer(struct dwc3_ep *dep, bool force,
 	 * 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 arbitrary 100us delay for that.
+	 * suggested to giveback all requests here.
 	 *
 	 * 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
+	 * In short, what we're doing is issuing EndTransfer with
+	 * CMDIOC bit set and delay kicking transfer until the
+	 * EndTransfer command had completed.
 	 *
 	 * As of IP version 3.10a of the DWC_usb3 IP, the controller
 	 * supports a mode to work around the above limitation. The

@@ -2720,8 +2717,7 @@ static void dwc3_stop_active_transfer(struct dwc3_ep *dep, bool force,
 	 * by writing GUCTL2[14]. This polling is already done in the
 	 * dwc3_send_gadget_ep_cmd() function so if the mode is
 	 * enabled, the EndTransfer command will have completed upon
-	 * returning from this function and we don't need to delay for
-	 * 100us.
+	 * returning from this function.
 	 *
 	 * This mode is NOT available on the DWC_usb31 IP.
 	 */

@@ -2739,9 +2735,6 @@ static void dwc3_stop_active_transfer(struct dwc3_ep *dep, bool force,
 		dep->flags &= ~DWC3_EP_TRANSFER_STARTED;
 	else
 		dep->flags |= DWC3_EP_END_TRANSFER_PENDING;
-
-	if (dwc3_is_usb31(dwc) || dwc->revision < DWC3_REVISION_310A)
-		udelay(100);
 }
 
 static void dwc3_clear_stall_all_ep(struct dwc3 *dwc)