uvcvideo: Move decode processing to process context

Newer high definition cameras, and cameras with multiple lenses such as
the range of stereovision cameras now available have ever increasing
data rates.

The inclusion of a variable length packet header in URB packets mean
that we must memcpy the frame data out to our destination 'manually'.
This can result in data rates of up to 2 gigabits per second being
processed.

To improve efficiency, and maximise throughput, handle the URB decode
processing through a work queue to move it from interrupt context, and
allow multiple processors to work on URBs in parallel.

Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>

2 files changed, 90 insertions, 18 deletions

diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
index c9d35bcab79a44..4e0c37885d241c 100644
--- a/drivers/media/usb/uvc/uvc_video.c
+++ b/drivers/media/usb/uvc/uvc_video.c
@@ -1058,27 +1058,64 @@ static int uvc_video_decode_start(struct uvc_streaming *stream,
 	return data[0];
 }
 
-static void uvc_video_decode_data(struct uvc_streaming *stream,
-		struct uvc_buffer *buf, const __u8 *data, int len)
+/**
+ * uvc_video_decode_data_work: Asynchronous memcpy processing
+ *
+ * Perform memcpy tasks in process context, with completion handlers
+ * to return the URB, and buffer handles.
+ *
+ * The work submitter must pre-determine that the work is safe
+ */
+static void uvc_video_decode_data_work(struct work_struct *work)
 {
-	unsigned int maxlen, nbytes;
-	void *mem;
+	struct uvc_urb *uvc_urb = container_of(work, struct uvc_urb, work);
+	unsigned int i;
+	int ret;
+
+	for (i = 0; i < uvc_urb->packets; i++) {
+		struct uvc_decode_op *op = &uvc_urb->decodes[i];
+
+		memcpy(op->dst, op->src, op->len);
+
+		/* Release reference taken on this buffer */
+		uvc_queue_buffer_release(op->buf);
+	}
+
+	ret = usb_submit_urb(uvc_urb->urb, GFP_ATOMIC);
+	if (ret  < 0) {
+		uvc_printk(KERN_ERR, "Failed to resubmit video URB (%d).\n",
+			ret);
+	}
+}
+
+static void uvc_video_decode_data(struct uvc_decode_op *decode,
+		struct uvc_urb *uvc_urb, struct uvc_buffer *buf,
+		const __u8 *data, int len)
+{
+	unsigned int maxlen;
 
 	if (len <= 0)
 		return;
 
-	/* Copy the video data to the buffer. */
 	maxlen = buf->length - buf->bytesused;
-	mem = buf->mem + buf->bytesused;
-	nbytes = min((unsigned int)len, maxlen);
-	memcpy(mem, data, nbytes);
-	buf->bytesused += nbytes;
+
+	/* Take a buffer reference for async work */
+	kref_get(&buf->ref);
+
+	decode->buf = buf;
+	decode->src = data;
+	decode->dst = buf->mem + buf->bytesused;
+	decode->len = min_t(unsigned int, len, maxlen);
+
+	buf->bytesused += decode->len;
 
 	/* Complete the current frame if the buffer size was exceeded. */
 	if (len > maxlen) {
 		uvc_trace(UVC_TRACE_FRAME, "Frame complete (overflow).\n");
 		buf->state = UVC_BUF_STATE_READY;
 	}
+
+	uvc_urb->packets++;
 }
 
 static void uvc_video_decode_end(struct uvc_streaming *stream,
@@ -1161,6 +1198,8 @@ static void uvc_video_decode_isoc(struct uvc_urb *uvc_urb,
 	int ret, i;
 
 	for (i = 0; i < urb->number_of_packets; ++i) {
+		struct uvc_decode_op *op = &uvc_urb->decodes[uvc_urb->packets];
+
 		if (urb->iso_frame_desc[i].status < 0) {
 			uvc_trace(UVC_TRACE_FRAME, "USB isochronous frame "
 				"lost (%d).\n", urb->iso_frame_desc[i].status);
@@ -1186,7 +1225,7 @@ static void uvc_video_decode_isoc(struct uvc_urb *uvc_urb,
 			continue;
 
 		/* Decode the payload data. */
-		uvc_video_decode_data(stream, buf, mem + ret,
+		uvc_video_decode_data(op, uvc_urb, buf, mem + ret,
 			urb->iso_frame_desc[i].actual_length - ret);
 
 		/* Process the header again. */
@@ -1247,9 +1286,12 @@ static void uvc_video_decode_bulk(struct uvc_urb *uvc_urb,
 	 * sure buf is never dereferenced if NULL.
 	 */
 
-	/* Process video data. */
-	if (!stream->bulk.skip_payload && buf != NULL)
-		uvc_video_decode_data(stream, buf, mem, len);
+	/* Prepare video data for processing. */
+	if (!stream->bulk.skip_payload && buf != NULL) {
+		struct uvc_decode_op *op = &uvc_urb->decodes[0];
+
+		uvc_video_decode_data(op, uvc_urb, buf, mem, len);
+	}
 
 	/* Detect the payload end by a URB smaller than the maximum size (or
 	 * a payload size equal to the maximum) and process the header again.
@@ -1322,7 +1364,6 @@ static void uvc_video_complete(struct urb *urb)
 	struct uvc_streaming *stream = uvc_urb->stream;
 	struct uvc_video_queue *queue = &stream->queue;
 	struct uvc_buffer *buf = NULL;
-	int ret;
 
 	switch (urb->status) {
 	case 0:
@@ -1344,12 +1385,15 @@ static void uvc_video_complete(struct urb *urb)
 
 	buf = uvc_queue_get_current_buffer(queue);
 
+	/* Re-initialise the URB packet work */
+	uvc_urb->packets = 0;
+
+	/* Process the URB headers, but work is deferred to a work queue */
 	stream->decode(uvc_urb, buf);
 
-	if ((ret = usb_submit_urb(urb, GFP_ATOMIC)) < 0) {
-		uvc_printk(KERN_ERR, "Failed to resubmit video URB (%d).\n",
-			ret);
-	}
+	/* Handle any heavy lifting required */
+	INIT_WORK(&uvc_urb->work, uvc_video_decode_data_work);
+	queue_work(stream->async_wq, &uvc_urb->work);
 }
 
 /*
@@ -1618,6 +1662,11 @@ static int uvc_init_video(struct uvc_streaming *stream, gfp_t gfp_flags)
 
 	uvc_video_stats_start(stream);
 
+	stream->async_wq = alloc_workqueue("uvcvideo", WQ_UNBOUND | WQ_HIGHPRI,
+			0);
+	if (!stream->async_wq)
+		return -ENOMEM;
+
 	if (intf->num_altsetting > 1) {
 		struct usb_host_endpoint *best_ep = NULL;
 		unsigned int best_psize = UINT_MAX;
diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
index 0aa9bdc2ed4f38..eed22cfe4aaa16 100644
--- a/drivers/media/usb/uvc/uvcvideo.h
+++ b/drivers/media/usb/uvc/uvcvideo.h
@@ -456,12 +456,30 @@ struct uvc_stats_stream {
 };
 
 /**
+ * struct uvc_decode_op: Context structure to schedule asynchronous memcpy
+ *
+ * @buf: active buf object for this decode
+ * @dst: copy destination address
+ * @src: copy source address
+ * @len: copy length
+ */
+struct uvc_decode_op {
+	struct uvc_buffer *buf;
+	void *dst;
+	const __u8 *src;
+	int len;
+};
+
+/**
  * struct uvc_urb - URB context management structure
  *
  * @urb: described URB. Must be allocated with usb_alloc_urb()
  * @stream: UVC streaming context
  * @urb_buffer: memory storage for the URB
  * @urb_dma: DMA coherent addressing for the urb_buffer
+ * @packets: counter to indicate the number of copy operations
+ * @decodes: work descriptors for asynchronous copy operations
+ * @work: work queue entry for asynchronous decode
  */
 struct uvc_urb {
 	struct urb *urb;
@@ -469,6 +487,10 @@ struct uvc_urb {
 
 	char *urb_buffer;
 	dma_addr_t urb_dma;
+
+	unsigned int packets;
+	struct uvc_decode_op decodes[UVC_MAX_PACKETS];
+	struct work_struct work;
 };
 
 struct uvc_streaming {
@@ -501,6 +523,7 @@ struct uvc_streaming {
 	/* Buffers queue. */
 	unsigned int frozen : 1;
 	struct uvc_video_queue queue;
+	struct workqueue_struct *async_wq;
 	void (*decode) (struct uvc_urb *uvc_urb, struct uvc_buffer *buf);
 
 	/* Context data used by the bulk completion handler. */