[media] vb2: fix a regression in poll() behavior for output,streams

[ Upstream commit 4623e59 ]

In the 3.17 kernel the poll() behavior changed for output streams:
as long as not all buffers were queued up poll() would return that
userspace can write. This is fine for the write() call, but when
using stream I/O this changed the behavior since the expectation
was that it would wait for buffers to become available for dequeuing.

This patch only enables the check whether you can queue buffers
for file I/O only, and skips it for stream I/O.

Signed-off-by: Hans Verkuil <[email protected]>
Acked-by: Laurent Pinchart <[email protected]>
Cc: <[email protected]>      # for v3.17 and up
Signed-off-by: Hans Verkuil <[email protected]>
Signed-off-by: Mauro Carvalho Chehab <[email protected]>
Signed-off-by: Sasha Levin <[email protected]>

Author: sashalevin

drivers/media/v4l2-core/videobuf2-core.c

@@ -2662,10 +2662,10 @@ unsigned int vb2_poll(struct vb2_queue *q, struct file *file, poll_table *wait)
 		return res | POLLERR;
 
 	/*
-	 * For output streams you can write as long as there are fewer buffers
-	 * queued than there are buffers available.
+	 * For output streams you can call write() as long as there are fewer
+	 * buffers queued than there are buffers available.
 	 */
-	if (V4L2_TYPE_IS_OUTPUT(q->type) && q->queued_count < q->num_buffers)
+	if (V4L2_TYPE_IS_OUTPUT(q->type) && q->fileio && q->queued_count < q->num_buffers)
 		return res | POLLOUT | POLLWRNORM;
 
 	if (list_empty(&q->done_list))