[PATCH] UVC: Remove extra commit on resume()

[PATCH] UVC: Remove extra commit on resume()

[Guennadi Liakhovetski]

From: Aviv Greenberg <aviv.d.greenberg@intel.com>

The UVC spec is a bit vague wrt devices using bulk endpoints,
specifically, how to signal to a device to start streaming.

For devices using isoc endpoints, the sequence for start streaming is:
1) The host sends PROBE_CONTROL(SET_CUR) PROBE_CONTROL(GET_CUR)
2) Host selects desired config and calls COMMIT_CONTROL(SET_CUR)
3) Host selects an alt interface other then zero - e.g SELECT_ALTERNATE_INTERFACE(1)
4) The device starts streaming

However for devices using bulk endpoints, there must be *no* alt interface
other than setting zero. From the UVC spec:
"A VideoStreaming interface containing a bulk endpoint for streaming shall
support only alternate setting zero. Additional alternate settings containing
bulk endpoints are not permitted in a device that is compliant with the Video
Class specification."

So for devices using bulk endpoints, step #3 above is irrelevant, and thus
cannot be used as an indication for the device to start streaming.
So in practice, such devices start streaming immediately after a
COMMIT_CONTROL(SET_CUR).

In the uvc resume() handler, an unsolicited commit is sent, which causes
devices using bulk endpoints to start streaming unintentionally.

This patch modifies resume() handler to send a commit only if streaming
needs to be reestablished, i.e if the device was actually streaming before is
was suspended.

Signed-off-by: Aviv Greenberg <aviv.d.greenberg@intel.com>
Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
---
 drivers/media/usb/uvc/uvc_video.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
index 3394c34..c111de2 100644
--- a/drivers/media/usb/uvc/uvc_video.c
+++ b/drivers/media/usb/uvc/uvc_video.c
@@ -1709,15 +1709,15 @@ int uvc_video_resume(struct uvc_streaming *stream, int reset)
 
 	uvc_video_clock_reset(stream);
 
+	if (!uvc_queue_streaming(&stream->queue))
+		return 0;
+
 	ret = uvc_commit_video(stream, &stream->ctrl);
 	if (ret < 0) {
 		uvc_queue_enable(&stream->queue, 0);
 		return ret;
 	}
 
-	if (!uvc_queue_streaming(&stream->queue))
-		return 0;
-
 	ret = uvc_init_video(stream, GFP_NOIO);
 	if (ret < 0)
 		uvc_queue_enable(&stream->queue, 0);
-- 
1.7.9.5

Re: [PATCH] UVC: Remove extra commit on resume()

[Laurent Pinchart]

Hi Guennadi,

Thank you for the patch, and sorry for the late reply.

On Tuesday 02 September 2014 08:16:28 Guennadi Liakhovetski wrote:
> From: Aviv Greenberg <aviv.d.greenberg@intel.com>
> 
> The UVC spec is a bit vague wrt devices using bulk endpoints,
> specifically, how to signal to a device to start streaming.
> 
> For devices using isoc endpoints, the sequence for start streaming is:
> 1) The host sends PROBE_CONTROL(SET_CUR) PROBE_CONTROL(GET_CUR)
> 2) Host selects desired config and calls COMMIT_CONTROL(SET_CUR)
> 3) Host selects an alt interface other then zero - e.g
> SELECT_ALTERNATE_INTERFACE(1) 4) The device starts streaming
> 
> However for devices using bulk endpoints, there must be *no* alt interface
> other than setting zero. From the UVC spec:
> "A VideoStreaming interface containing a bulk endpoint for streaming shall
> support only alternate setting zero. Additional alternate settings
> containing bulk endpoints are not permitted in a device that is compliant
> with the Video Class specification."
> 
> So for devices using bulk endpoints, step #3 above is irrelevant, and thus
> cannot be used as an indication for the device to start streaming.
> So in practice, such devices start streaming immediately after a
> COMMIT_CONTROL(SET_CUR).
> 
> In the uvc resume() handler, an unsolicited commit is sent, which causes
> devices using bulk endpoints to start streaming unintentionally.
> 
> This patch modifies resume() handler to send a commit only if streaming
> needs to be reestablished, i.e if the device was actually streaming before
> is was suspended.

Speaking of bulk devices, based on your experience, how do devices detect a 
stream stop condition in practice ?

> Signed-off-by: Aviv Greenberg <aviv.d.greenberg@intel.com>
> Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>

Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>

I've taken the patch in my tree and will send a pull request for v3.20.

> ---
>  drivers/media/usb/uvc/uvc_video.c |    6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/media/usb/uvc/uvc_video.c
> b/drivers/media/usb/uvc/uvc_video.c index 3394c34..c111de2 100644
> --- a/drivers/media/usb/uvc/uvc_video.c
> +++ b/drivers/media/usb/uvc/uvc_video.c
> @@ -1709,15 +1709,15 @@ int uvc_video_resume(struct uvc_streaming *stream,
> int reset)
> 
>  	uvc_video_clock_reset(stream);
> 
> +	if (!uvc_queue_streaming(&stream->queue))
> +		return 0;
> +
>  	ret = uvc_commit_video(stream, &stream->ctrl);
>  	if (ret < 0) {
>  		uvc_queue_enable(&stream->queue, 0);
>  		return ret;
>  	}
> 
> -	if (!uvc_queue_streaming(&stream->queue))
> -		return 0;
> -
>  	ret = uvc_init_video(stream, GFP_NOIO);
>  	if (ret < 0)
>  		uvc_queue_enable(&stream->queue, 0);

-- 
Regards,

Laurent Pinchart