[libcamera-devel] [PATCH v3 04/13] libcamera: v4l2_subdevice: Implement get selection

[Jacopo Mondi]

Hi Laurent,

On Sat, Apr 25, 2020 at 03:42:43PM +0300, Laurent Pinchart wrote:
> Hi Jacopo,
>
> Thank you for the patch.
>
> On Fri, Apr 24, 2020 at 11:52:55PM +0200, Jacopo Mondi wrote:
> > Implement get selection for the V4L2Subdevice class to support
> > retrieving three targets: the subdevice native size, the analog crop and
> > the active pixel area.
> >
> > Signed-off-by: Jacopo Mondi <jacopo at jmondi.org>
> > ---
> >  src/libcamera/include/v4l2_subdevice.h |  6 ++
> >  src/libcamera/v4l2_subdevice.cpp       | 92 ++++++++++++++++++++++++++
> >  2 files changed, 98 insertions(+)
> >
> > diff --git a/src/libcamera/include/v4l2_subdevice.h b/src/libcamera/include/v4l2_subdevice.h
> > index 4a04eadfb1f9..763ffadb61fb 100644
> > --- a/src/libcamera/include/v4l2_subdevice.h
> > +++ b/src/libcamera/include/v4l2_subdevice.h
> > @@ -46,6 +46,10 @@ public:
> >
> >  	const MediaEntity *entity() const { return entity_; }
> >
> > +	int getNativeSize(unsigned int pad, Size *size);
> > +	int getActiveArea(unsigned int pad, Rectangle *rect);
>
> I think we're mixing two layers here. The V4L2 API defines selection
> targets. The fact the crop bounds or crop default targets map to the
> full size and active area is specific to camera sensors. I would thus
> move these two functions to the CameraSensor class. As this will require
> accessing different selection targets, we could add a target argument to
> getCrop(), but I think it may just be simpler to drop setCrop() and
> setCompose() and make getSelection() and setSelection() public.
>

That' a thin distinction imho. My prefere here was to minimize the
amount of V4L2 dependencies in CameraSensor, so that it will still use
a 'getNativeSize' if we have to support an API != V4L2.

I get your point though, and as you can read from function
documentation
"if the sub-device represents a camera sensor then..." I agree the
v4l2 subdev should not expose sensor-specific concepts maybe.

Although, it might require more work later to change this. I'm fine
changin this, but be aware that it implies polluting the library with
the V4L2_SEL_TGT_* flags.


> > +	int getCropRectangle(unsigned int pad, Rectangle *rect);
>
> As the counterpart function is called setCrop(), shouldn't this be
> called getCrop() ?
>

ack

> > +
> >  	int setCrop(unsigned int pad, Rectangle *rect);
> >  	int setCompose(unsigned int pad, Rectangle *rect);
> >
> > @@ -67,6 +71,8 @@ private:
> >  	std::vector<SizeRange> enumPadSizes(unsigned int pad,
> >  					    unsigned int code);
> >
> > +	int getSelection(unsigned int pad, unsigned int target,
> > +			 Rectangle *rect);
> >  	int setSelection(unsigned int pad, unsigned int target,
> >  			 Rectangle *rect);
> >
> > diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> > index 5a479a96b795..4dcf8ce48754 100644
> > --- a/src/libcamera/v4l2_subdevice.cpp
> > +++ b/src/libcamera/v4l2_subdevice.cpp
> > @@ -133,6 +133,68 @@ int V4L2Subdevice::open()
> >   * \return The subdevice's associated media entity.
> >   */
> >
> > +/**
> > + * \brief Get the sub-device native size
> > + * \param[in] pad The 0-indexed pad number the rectangle is to be applied to
> > + * \param[out] size The sub-device native area size
> > + *
> > + * Retrieve the size of the sub-device native area.
> > + * If the sub-device represent an image sensor, the native area describes
> > + * the pixel array dimensions, including inactive and calibration pixels.
> > + *
> > + * \return 0 on success or a negative error code otherwise
> > + * \retval -ENOTTY The native size information is not available
> > + */
> > +int V4L2Subdevice::getNativeSize(unsigned int pad, Size *size)
> > +{
> > +	Rectangle rect = {};
> > +	int ret = getSelection(pad, V4L2_SEL_TGT_NATIVE_SIZE, &rect);
> > +	if (ret)
> > +		return ret;
> > +
> > +	size->width = rect.width;
> > +	size->height = rect.height;
> > +
> > +	return 0;
> > +}
> > +
> > +/**
> > + * \brief Get the active area rectangle
> > + * \param[in] pad The 0-indexed pad number the rectangle is to be applied to
> > + * \param[out] rect The rectangle describing the sub-device active area
> > + *
> > + * Retrieve the rectangle describing the sub-device active area.
> > + * If the sub-device represent an image sensor, the active area describes
> > + * the pixel array active portion.
> > + *
> > + * The returned \a rect 'x' and 'y' fields represent the displacement from the
> > + * native size rectangle top-left corner.
> > + *
> > + * \return 0 on success or a negative error code otherwise
> > + * \retval -ENOTTY The active areas size information is not available
> > + */
> > +int V4L2Subdevice::getActiveArea(unsigned int pad, Rectangle *rect)
> > +{
> > +	return getSelection(pad, V4L2_SEL_TGT_CROP_DEFAULT, rect);
> > +}
> > +
> > +/**
> > + * \brief Get the crop rectangle
> > + * \param[in] pad The 0-indexed pad number the rectangle is to be applied to
> > + * \param[out] rect The rectangle describing the cropped area rectangle
> > + *
> > + * Retrieve the rectangle representing the cropped area. If the sub-device
> > + * represents an image sensor, the cropped area describes the pixel array
> > + * portion from which the final image is produced from.
> > + *
> > + * \return 0 on success or a negative error code otherwise
> > + * \retval -ENOTTY The crop rectangle size information is not available
> > + */
> > +int V4L2Subdevice::getCropRectangle(unsigned int pad, Rectangle *rect)
> > +{
> > +	return getSelection(pad, V4L2_SEL_TGT_CROP, rect);
> > +}
> > +
> >  /**
> >   * \brief Set a crop rectangle on one of the V4L2 subdevice pads
> >   * \param[in] pad The 0-indexed pad number the rectangle is to be applied to
> > @@ -343,6 +405,36 @@ std::vector<SizeRange> V4L2Subdevice::enumPadSizes(unsigned int pad,
> >  	return sizes;
> >  }
> >
> > +int V4L2Subdevice::getSelection(unsigned int pad, unsigned int target,
> > +				Rectangle *rect)
> > +{
> > +	struct v4l2_subdev_selection sel = {};
> > +
> > +	sel.which = V4L2_SUBDEV_FORMAT_ACTIVE;
> > +	sel.pad = pad;
> > +	sel.target = target;
> > +	sel.flags = 0;
> > +
> > +	int ret = ioctl(VIDIOC_SUBDEV_G_SELECTION, &sel);
> > +	if (ret < 0 && ret != -ENOTTY) {
> > +		LOG(V4L2, Error)
> > +			<< "Unable to get rectangle " << target << " on pad "
> > +			<< pad << ": " << strerror(-ret);
> > +		return ret;
> > +	}
> > +	if (ret == -ENOTTY) {
> > +		LOG(V4L2, Debug) << "Selection API not available";
> > +		return ret;
> > +	}
>
> Do you expect valid users of this function when the subdev doesn't
> implement the selection API ? I can imagine that a caller would check
> for -ENOTTY and use some fallback mechanism to make this error
> non-fatal, in which case a debug message could be better than an error
> message, but do we have any practical use case for that ? Is it only
> meant to cover a transition period until sensor drivers get converted ?
>

I see you have found that out by yourself

> > +
> > +	rect->x = sel.r.left;
> > +	rect->y = sel.r.top;
> > +	rect->width = sel.r.width;
> > +	rect->height = sel.r.height;
> > +
> > +	return 0;
> > +}
> > +
> >  int V4L2Subdevice::setSelection(unsigned int pad, unsigned int target,
> >  				Rectangle *rect)
> >  {
>
> --
> Regards,
>
> Laurent Pinchart