usb: gadget: uvc: rework to enqueue in pump worker from encoded queue

We install an kthread with pfifo priority that is iterating over all
prepared requests and keeps the isoc queue busy. This way it will be
scheduled with the same priority as the interrupt handler.

As the kthread is triggered with video_enable it will immediately
queue some zero length requests into the hw if there is no buffer data
available. It also watches the level of needed zero length requests in
the hardware not to fall under the UVCG_REQ_MAX_ZERO_COUNT threshold.
This way we can drop the function uvc_video_ep_queue_initial_requests
entirely.

By using the kthread to do the actual request handling the interrupt
handler will not be running into the time consuming and eventually
locking work of actually enqueueing the requests back into its own
pipeline. This work can now even be scheduled on another cpu.

Signed-off-by: Michael Grzeschik <[email protected]>
Link: https://lore.kernel.org/r/20240403-uvc_request_length_by_interval-v7-3-e224bb1035f0@pengutronix.de
Signed-off-by: Greg Kroah-Hartman <[email protected]>

drivers/usb/gadget/function/f_uvc.c

@@ -991,6 +991,8 @@ static void uvc_function_unbind(struct usb_configuration *c,
 
 	uvcg_info(f, "%s()\n", __func__);
 
+	kthread_cancel_work_sync(&video->hw_submit);
+
 	if (video->async_wq)
 		destroy_workqueue(video->async_wq);
 

drivers/usb/gadget/function/uvc.h

@@ -94,6 +94,9 @@ struct uvc_video {
 	struct work_struct pump;
 	struct workqueue_struct *async_wq;
 
+	struct kthread_worker   *kworker;
+	struct kthread_work     hw_submit;
+
 	atomic_t queued;
 
 	/* Frame parameters */

drivers/usb/gadget/function/uvc_video.c

@@ -324,61 +324,14 @@ static int uvcg_video_usb_req_queue(struct uvc_video *video,
 	return 0;
 }
 
-/*
- * Must only be called from uvcg_video_enable - since after that we only want to
- * queue requests to the endpoint from the uvc_video_complete complete handler.
- * This function is needed in order to 'kick start' the flow of requests from
- * gadget driver to the usb controller.
- */
-static void uvc_video_ep_queue_initial_requests(struct uvc_video *video)
-{
-	struct usb_request *req = NULL;
-	unsigned long flags = 0;
-	unsigned int count = 0;
-	int ret = 0;
-
-	/*
-	 * We only queue half of the free list since we still want to have
-	 * some free usb_requests in the free list for the video_pump async_wq
-	 * thread to encode uvc buffers into. Otherwise we could get into a
-	 * situation where the free list does not have any usb requests to
-	 * encode into - we always end up queueing 0 length requests to the
-	 * end point.
-	 */
-	unsigned int half_list_size = video->uvc_num_requests / 2;
-
-	spin_lock_irqsave(&video->req_lock, flags);
-	/*
-	 * Take these requests off the free list and queue them all to the
-	 * endpoint. Since we queue 0 length requests with the req_lock held,
-	 * there isn't any 'data' race involved here with the complete handler.
-	 */
-	while (count < half_list_size) {
-		req = list_first_entry(&video->req_free, struct usb_request,
-					list);
-		list_del(&req->list);
-		req->length = 0;
-		ret = uvcg_video_ep_queue(video, req);
-		if (ret < 0) {
-			uvcg_queue_cancel(&video->queue, 0);
-			break;
-		}
-		count++;
-	}
-	spin_unlock_irqrestore(&video->req_lock, flags);
-}
-
 static void
 uvc_video_complete(struct usb_ep *ep, struct usb_request *req)
 {
 	struct uvc_request *ureq = req->context;
 	struct uvc_video *video = ureq->video;
 	struct uvc_video_queue *queue = &video->queue;
 	struct uvc_buffer *last_buf;
-	struct usb_request *to_queue = req;
 	unsigned long flags;
-	bool is_bulk = video->max_payload_size;
-	int ret = 0;
 
 	spin_lock_irqsave(&video->req_lock, flags);
 	atomic_dec(&video->queued);
@@ -441,65 +394,85 @@ uvc_video_complete(struct usb_ep *ep, struct usb_request *req)
 		return;
 	}
 
+	list_add_tail(&req->list, &video->req_free);
 	/*
-	 * Here we check whether any request is available in the ready
-	 * list. If it is, queue it to the ep and add the current
-	 * usb_request to the req_free list - for video_pump to fill in.
-	 * Otherwise, just use the current usb_request to queue a 0
-	 * length request to the ep. Since we always add to the req_free
-	 * list if we dequeue from the ready list, there will never
-	 * be a situation where the req_free list is completely out of
-	 * requests and cannot recover.
+	 * Queue work to the wq as well since it is possible that a
+	 * buffer may not have been completely encoded with the set of
+	 * in-flight usb requests for whih the complete callbacks are
+	 * firing.
+	 * In that case, if we do not queue work to the worker thread,
+	 * the buffer will never be marked as complete - and therefore
+	 * not be returned to userpsace. As a result,
+	 * dequeue -> queue -> dequeue flow of uvc buffers will not
+	 * happen. Since there are is a new free request wake up the pump.
 	 */
-	to_queue->length = 0;
-	if (!list_empty(&video->req_ready)) {
-		to_queue = list_first_entry(&video->req_ready,
-			struct usb_request, list);
-		list_del(&to_queue->list);
-		list_add_tail(&req->list, &video->req_free);
-		/*
-		 * Queue work to the wq as well since it is possible that a
-		 * buffer may not have been completely encoded with the set of
-		 * in-flight usb requests for whih the complete callbacks are
-		 * firing.
-		 * In that case, if we do not queue work to the worker thread,
-		 * the buffer will never be marked as complete - and therefore
-		 * not be returned to userpsace. As a result,
-		 * dequeue -> queue -> dequeue flow of uvc buffers will not
-		 * happen.
-		 */
-		queue_work(video->async_wq, &video->pump);
-	} else if (atomic_read(&video->queued) > UVCG_REQ_MAX_ZERO_COUNT) {
-		list_add_tail(&to_queue->list, &video->req_free);
-		/*
-		 * There is a new free request - wake up the pump.
-		 */
-		queue_work(video->async_wq, &video->pump);
+	queue_work(video->async_wq, &video->pump);
 
-		spin_unlock_irqrestore(&video->req_lock, flags);
+	spin_unlock_irqrestore(&video->req_lock, flags);
 
-		return;
-	}
-	/*
-	 * Queue to the endpoint. The actual queueing to ep will
-	 * only happen on one thread - the async_wq for bulk endpoints
-	 * and this thread for isoc endpoints.
-	 */
-	ret = uvcg_video_usb_req_queue(video, to_queue, !is_bulk);
-	if (ret < 0) {
+	kthread_queue_work(video->kworker, &video->hw_submit);
+}
+
+static void uvcg_video_hw_submit(struct kthread_work *work)
+{
+	struct uvc_video *video = container_of(work, struct uvc_video, hw_submit);
+	bool is_bulk = video->max_payload_size;
+	unsigned long flags;
+	struct usb_request *req;
+	int ret = 0;
+
+	while (true) {
+		if (!video->ep->enabled)
+			return;
+		spin_lock_irqsave(&video->req_lock, flags);
 		/*
-		 * Endpoint error, but the stream is still enabled.
-		 * Put request back in req_free for it to be cleaned
-		 * up later.
+		 * Here we check whether any request is available in the ready
+		 * list. If it is, queue it to the ep and add the current
+		 * usb_request to the req_free list - for video_pump to fill in.
+		 * Otherwise, just use the current usb_request to queue a 0
+		 * length request to the ep. Since we always add to the req_free
+		 * list if we dequeue from the ready list, there will never
+		 * be a situation where the req_free list is completely out of
+		 * requests and cannot recover.
 		 */
-		list_add_tail(&to_queue->list, &video->req_free);
+		if (!list_empty(&video->req_ready)) {
+			req = list_first_entry(&video->req_ready,
+					       struct usb_request, list);
+		} else {
+			if (list_empty(&video->req_free) ||
+			    (atomic_read(&video->queued) > UVCG_REQ_MAX_ZERO_COUNT)) {
+				spin_unlock_irqrestore(&video->req_lock, flags);
+
+				return;
+			}
+			req = list_first_entry(&video->req_free, struct usb_request,
+					       list);
+			req->length = 0;
+		}
+		list_del(&req->list);
+
 		/*
-		 * There is a new free request - wake up the pump.
+		 * Queue to the endpoint. The actual queueing to ep will
+		 * only happen on one thread - the async_wq for bulk endpoints
+		 * and this thread for isoc endpoints.
 		 */
-		queue_work(video->async_wq, &video->pump);
-	}
+		ret = uvcg_video_usb_req_queue(video, req, !is_bulk);
+		if (ret < 0) {
+			/*
+			 * Endpoint error, but the stream is still enabled.
+			 * Put request back in req_free for it to be cleaned
+			 * up later.
+			 */
+			list_add_tail(&req->list, &video->req_free);
+			/*
+			 * There is a new free request - wake up the pump.
+			 */
+			queue_work(video->async_wq, &video->pump);
 
-	spin_unlock_irqrestore(&video->req_lock, flags);
+		}
+
+		spin_unlock_irqrestore(&video->req_lock, flags);
+	}
 }
 
 static int
@@ -771,7 +744,7 @@ int uvcg_video_enable(struct uvc_video *video)
 
 	atomic_set(&video->queued, 0);
 
-	uvc_video_ep_queue_initial_requests(video);
+	kthread_queue_work(video->kworker, &video->hw_submit);
 	queue_work(video->async_wq, &video->pump);
 
 	return ret;
@@ -794,6 +767,17 @@ int uvcg_video_init(struct uvc_video *video, struct uvc_device *uvc)
 	if (!video->async_wq)
 		return -EINVAL;
 
+	/* Allocate a kthread for asynchronous hw submit handler. */
+	video->kworker = kthread_create_worker(0, "UVCG");
+	if (IS_ERR(video->kworker)) {
+		uvcg_err(&video->uvc->func, "failed to create UVCG kworker\n");
+		return PTR_ERR(video->kworker);
+	}
+
+	kthread_init_work(&video->hw_submit, uvcg_video_hw_submit);
+
+	sched_set_fifo(video->kworker->task);
+
 	video->uvc = uvc;
 	video->fcc = V4L2_PIX_FMT_YUYV;
 	video->bpp = 16;