usb: dwc3: gadget: handle request->zero

So far, dwc3 has always missed request->zero
handling for every endpoint. Let's implement
that so we can handle cases where transfer must
be finished with a ZLP.

Note that dwc3 is a little special. Even though
we're dealing with a ZLP, we still need a buffer
of wMaxPacketSize bytes; to hide that detail from
every gadget driver, we have a preallocated buffer
of 1024 bytes (biggest bulk size) to use (and
share) among all endpoints.
Reported-by: NRavi B <ravibabu@ti.com>
Signed-off-by: NFelipe Balbi <balbi@ti.com>

drivers/usb/dwc3/core.h

@@ -37,6 +37,7 @@
 #define DWC3_MSG_MAX	500
 
 /* Global constants */
+#define DWC3_ZLP_BUF_SIZE	1024	/* size of a superspeed bulk */
 #define DWC3_EP0_BOUNCE_SIZE	512
 #define DWC3_ENDPOINTS_NUM	32
 #define DWC3_XHCI_RESOURCES_NUM	2
@@ -647,6 +648,7 @@ struct dwc3_scratchpad_array {
  * @ctrl_req: usb control request which is used for ep0
  * @ep0_trb: trb which is used for the ctrl_req
  * @ep0_bounce: bounce buffer for ep0
+ * @zlp_buf: used when request->zero is set
  * @setup_buf: used while precessing STD USB requests
  * @ctrl_req_addr: dma address of ctrl_req
  * @ep0_trb: dma address of ep0_trb
@@ -734,6 +736,7 @@ struct dwc3 {
 	struct usb_ctrlrequest	*ctrl_req;
 	struct dwc3_trb		*ep0_trb;
 	void			*ep0_bounce;
+	void			*zlp_buf;
 	void			*scratchbuf;
 	u8			*setup_buf;
 	dma_addr_t		ctrl_req_addr;

drivers/usb/dwc3/gadget.c

@@ -1159,6 +1159,32 @@ static int __dwc3_gadget_ep_queue(struct dwc3_ep *dep, struct dwc3_request *req)
 	return ret;
 }
 
+static void __dwc3_gadget_ep_zlp_complete(struct usb_ep *ep,
+		struct usb_request *request)
+{
+	dwc3_gadget_ep_free_request(ep, request);
+}
+
+static int __dwc3_gadget_ep_queue_zlp(struct dwc3 *dwc, struct dwc3_ep *dep)
+{
+	struct dwc3_request		*req;
+	struct usb_request		*request;
+	struct usb_ep			*ep = &dep->endpoint;
+
+	dwc3_trace(trace_dwc3_gadget, "queueing ZLP\n");
+	request = dwc3_gadget_ep_alloc_request(ep, GFP_ATOMIC);
+	if (!request)
+		return -ENOMEM;
+
+	request->length = 0;
+	request->buf = dwc->zlp_buf;
+	request->complete = __dwc3_gadget_ep_zlp_complete;
+
+	req = to_dwc3_request(request);
+
+	return __dwc3_gadget_ep_queue(dep, req);
+}
+
 static int dwc3_gadget_ep_queue(struct usb_ep *ep, struct usb_request *request,
 	gfp_t gfp_flags)
 {
@@ -1172,6 +1198,16 @@ static int dwc3_gadget_ep_queue(struct usb_ep *ep, struct usb_request *request,
 
 	spin_lock_irqsave(&dwc->lock, flags);
 	ret = __dwc3_gadget_ep_queue(dep, req);
+
+	/*
+	 * Okay, here's the thing, if gadget driver has requested for a ZLP by
+	 * setting request->zero, instead of doing magic, we will just queue an
+	 * extra usb_request ourselves so that it gets handled the same way as
+	 * any other request.
+	 */
+	if (ret == 0 && request->zero && (request->length % ep->maxpacket == 0))
+		ret = __dwc3_gadget_ep_queue_zlp(dwc, dep);
+
 	spin_unlock_irqrestore(&dwc->lock, flags);
 
 	return ret;
@@ -2744,6 +2780,12 @@ int dwc3_gadget_init(struct dwc3 *dwc)
 		goto err3;
 	}
 
+	dwc->zlp_buf = kzalloc(DWC3_ZLP_BUF_SIZE, GFP_KERNEL);
+	if (!dwc->zlp_buf) {
+		ret = -ENOMEM;
+		goto err4;
+	}
+
 	dwc->gadget.ops			= &dwc3_gadget_ops;
 	dwc->gadget.speed		= USB_SPEED_UNKNOWN;
 	dwc->gadget.sg_supported	= true;
@@ -2785,16 +2827,19 @@ int dwc3_gadget_init(struct dwc3 *dwc)
 
 	ret = dwc3_gadget_init_endpoints(dwc);
 	if (ret)
-		goto err4;
+		goto err5;
 
 	ret = usb_add_gadget_udc(dwc->dev, &dwc->gadget);
 	if (ret) {
 		dev_err(dwc->dev, "failed to register udc\n");
-		goto err4;
+		goto err5;
 	}
 
 	return 0;
 
+err5:
+	kfree(dwc->zlp_buf);
+
 err4:
 	dwc3_gadget_free_endpoints(dwc);
 	dma_free_coherent(dwc->dev, DWC3_EP0_BOUNCE_SIZE,
@@ -2827,6 +2872,7 @@ void dwc3_gadget_exit(struct dwc3 *dwc)
 			dwc->ep0_bounce, dwc->ep0_bounce_addr);
 
 	kfree(dwc->setup_buf);
+	kfree(dwc->zlp_buf);
 
 	dma_free_coherent(dwc->dev, sizeof(*dwc->ep0_trb),
 			dwc->ep0_trb, dwc->ep0_trb_addr);