Re: [PATCH/RFC v2 1/4] v4l: Add metadata buffer type and format

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Hans,

Thank you for the review.

On Monday 23 May 2016 12:09:16 Hans Verkuil wrote:
> On 05/12/2016 02:18 AM, Laurent Pinchart wrote:
> > The metadata buffer type is used to transfer metadata between userspace
> > and kernelspace through a V4L2 buffers queue. It comes with a new
> > metadata capture capability and format description.
> 
> Thanks for the patch! I have some comments/suggestions below:
> > Signed-off-by: Laurent Pinchart
> > <laurent.pinchart+renesas@ideasonboard.com>
> > ---
> > 
> >  Documentation/DocBook/media/v4l/dev-meta.xml  | 93 ++++++++++++++++++++++
> >  Documentation/DocBook/media/v4l/v4l2.xml      |  1 +
> >  drivers/media/v4l2-core/v4l2-compat-ioctl32.c | 19 ++++++
> >  drivers/media/v4l2-core/v4l2-dev.c            | 16 +++--
> >  drivers/media/v4l2-core/v4l2-ioctl.c          | 34 ++++++++++
> >  drivers/media/v4l2-core/videobuf2-v4l2.c      |  3 +
> >  include/media/v4l2-ioctl.h                    |  8 +++
> >  include/uapi/linux/videodev2.h                | 14 ++++
> >  8 files changed, 182 insertions(+), 6 deletions(-)
> >  create mode 100644 Documentation/DocBook/media/v4l/dev-meta.xml
> > 
> > diff --git a/Documentation/DocBook/media/v4l/dev-meta.xml
> > b/Documentation/DocBook/media/v4l/dev-meta.xml new file mode 100644
> > index 000000000000..9b5b1fba2007
> > --- /dev/null
> > +++ b/Documentation/DocBook/media/v4l/dev-meta.xml
> > @@ -0,0 +1,93 @@
> > +  <title>Metadata Interface</title>
> > +
> > +  <note>
> > +    <title>Experimental</title>
> > +    <para>This is an <link linkend="experimental"> experimental </link>
> 
> No space before/after 'experimental'. It will look ugly in the docbook (I
> tried it :-) ).

Will be fixed in the next version.

> > +    interface and may change in the future.</para>
> > +  </note>
> > +
> > +  <para>
> > +Metadata refers to any non-image data that supplements video frames with
> > +additional information. This may include statistics computed over the
> > +image or frame capture parameters supplied by the image source. This
> > +interface is intended for transfer of metadata to userspace and control
> > +of that operation.
>
> I think it can also be in the other direction. I'm thinking of metadata
> needed by codecs. I'm not sure if it should be mentioned that this is not
> supported at the moment and that the ML should be contacted for more info
> if someone wants this.

Metadata for codecs make sense, but given that the topic hasn't been 
researched yet, I've kept metadata support focused on the capture side only 
for now. I propose revisiting the topic if (or rather when) someone needs 
metadata support on output video nodes.

> > +  </para>
> > +
> > +  <para>
> > +The metadata interface is implemented on video capture devices. The
> > device can
>
> s/on/for/?

I meant "on video capture device nodes", I'll fix that.
 
> > +be dedicated to metadata or can implement both video and metadata capture
> > +as specified in its reported capabilities.
> > +  </para>
> > +
> > +  <section>
> > +    <title>Querying Capabilities</title>
> > +
> > +    <para>
> > +Devices supporting the metadata interface set the
> > +<constant>V4L2_CAP_META_CAPTURE</constant> flag in the
> > +<structfield>capabilities</structfield> field of &v4l2-capability;
> > +returned by the &VIDIOC-QUERYCAP; ioctl. That flag means the device can
> > +capture metadata to memory.
> > +    </para>
> 
> I know this is a copy and paste from the other interface descriptions, but
> it is rather outdated. I would write this instead:
> 
>  <para>
> Device nodes supporting the metadata interface set the
> <constant>V4L2_CAP_META_CAPTURE</constant> flag in the
> <structfield>device_caps</structfield> field of &v4l2-capability;
> returned by the &VIDIOC-QUERYCAP; ioctl. That flag means the device node can
> capture metadata to memory.
>  </para>

Will be fixed in the next version.
 
> Any driver using this will be recent and always have a valid device_caps
> field (which, as you know, didn't exist in old kernels).
> 
> I find the capabilities field fairly useless in most cases due to the fact
> that it contains the capabilities of all device nodes created by the device
> driver.

I agree.

> > +    <para>
> > +At least one of the read/write or streaming I/O methods must be
> > supported.
> > +    </para>
> > +  </section>
> > +
> > +  <section>
> > +    <title>Data Format Negotiation</title>
> > +
> > +    <para>
> > +The metadata device uses the <link linkend="format">format</link> ioctls
> > to +select the capture format. The metadata buffer content format is
> > bound to that +selectable format. In addition to the basic
> > +<link linkend="format">format</link> ioctls, the &VIDIOC-ENUM-FMT; ioctl
> > +must be supported as well.
> > +    </para>
> > +
> > +    <para>
> > +To use the <link linkend="format">format</link> ioctls applications set
> > the +<structfield>type</structfield> field of a &v4l2-format; to
> > +<constant>V4L2_BUF_TYPE_META_CAPTURE</constant> and use the
> > &v4l2-meta-format; +<structfield>meta</structfield> member of the
> > <structfield>fmt</structfield> +union as needed per the desired
> > operation.
> > +Currently there are two fields, <structfield>dataformat</structfield> and
> > +<structfield>buffersize</structfield>, of struct &v4l2-meta-format; that
> > are +used. Content of the <structfield>dataformat</structfield> is the
> > V4L2 FourCC +code of the data format. The
> > <structfield>buffersize</structfield> field is the +maximum buffer size
> > in bytes required for data transfer, set by the driver in +order to
> > inform applications.
> 
> Should it be mentioned here that changing the video format might change the
> buffersize? In case the buffersize is always a multiple of the width?

I'll reply to Sakari's mail further down in this thread about this.

> > +    </para>
> > +
> > +    <table pgwide="1" frame="none" id="v4l2-meta-format">
> > +      <title>struct <structname>v4l2_meta_format</structname></title>
> > +      <tgroup cols="3">
> > +        &cs-str;
> > +        <tbody valign="top">
> > +          <row>
> > +            <entry>__u32</entry>
> > +            <entry><structfield>dataformat</structfield></entry>
> > +            <entry>
> > +The data format, set by the application. This is a little endian
> > +<link linkend="v4l2-fourcc">four character code</link>.
> > +V4L2 defines metadata formats in <xref linkend="meta-formats" />.
> > +           </entry>
> > +          </row>
> > +          <row>
> > +            <entry>__u32</entry>
> > +            <entry><structfield>buffersize</structfield></entry>
> > +            <entry>
> > +Maximum size in bytes required for data. Value is set by the driver.
> > +           </entry>
> > +          </row>
> > +          <row>
> > +            <entry>__u8</entry>
> > +            <entry><structfield>reserved[24]</structfield></entry>
> > +            <entry>This array is reserved for future extensions.
> > +Drivers and applications must set it to zero.</entry>
> > +          </row>
> > +        </tbody>
> > +      </tgroup>
> > +    </table>
> > +
> > +  </section>

[snip]

> > diff --git a/drivers/media/v4l2-core/v4l2-ioctl.c
> > b/drivers/media/v4l2-core/v4l2-ioctl.c index 28e5be2c2eef..5d003152ff68
> > 100644
> > --- a/drivers/media/v4l2-core/v4l2-ioctl.c
> > +++ b/drivers/media/v4l2-core/v4l2-ioctl.c

[snip]

> > @@ -1534,6 +1558,11 @@ static int v4l_s_fmt(const struct v4l2_ioctl_ops
> > *ops,
> >  			break;
> >  		CLEAR_AFTER_FIELD(p, fmt.sdr);
> >  		return ops->vidioc_s_fmt_sdr_out(file, fh, arg);
> > +	case V4L2_BUF_TYPE_META_CAPTURE:
> > +		if (unlikely(!is_rx || !is_vid ||
> > !ops->vidioc_s_fmt_meta_cap))
> > +			break;
> > +		CLEAR_AFTER_FIELD(p, fmt.meta);
> 
> I would suggest using "CLEAR_FROM_FIELD(p, fmt.meta.reserved)" here. Of
> course, you'd need to add a new CLEAR_FROM_FIELD macro for this.
> 
> I think this also should be used for the SDR_CAPTURE/OUTPUT types and
> possibly for other types as well (can be done later of course).
> 
> This would also zero the reserved field, so drivers don't need to care about
> it.

Will be fixed in the next version (in v4l_g_fmt() too).

> > +		return ops->vidioc_s_fmt_meta_cap(file, fh, arg);
> >  	}
> >  	return -EINVAL;
> >  }

-- 
Regards,

Laurent Pinchart