[libcamera-devel,v2,2/6] libcamera: v4l2_device: Adjust colorspace if pixel format is RGB

17196 diff mbox series

The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
Rec 601 Y'CbCr encoding and limited range) in the kernel.

The RGB pixel formats should not use any Y'CbCr encoding and is always
full range. Adjust the colorspace before reporting back to the
userspace in such a situation.

Moving forwards, the ColorSpace::Srgb will be defined in the true sense
for RGB pixel formats.

Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
---
 include/libcamera/internal/v4l2_device.h |  5 ++++-
 src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
 src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
 src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
 4 files changed, 35 insertions(+), 11 deletions(-)

Hi Umang,

Thank you for the patch.

On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
> Rec 601 Y'CbCr encoding and limited range) in the kernel.
> 
> The RGB pixel formats should not use any Y'CbCr encoding and is always
> full range. Adjust the colorspace before reporting back to the
> userspace in such a situation.
> 
> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
> for RGB pixel formats.

Mentioning sRGB here is a bit confusing I think. Let's focus on what
this patch does (and this will also help making sure I understood this
correctly):

V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
formats, including RGB and raw formats. This causes the libcamera
ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
overriding the encoding reported by the kernel to YCbCrEncoding::None
for non-YUV pixel formats and media bus formats.

Similarly, override the quantization range of non-YUV formats to full
range, as limited range isn't used for RGB and raw formats.

> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
> ---
>  include/libcamera/internal/v4l2_device.h |  5 ++++-
>  src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
>  src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
>  src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
>  4 files changed, 35 insertions(+), 11 deletions(-)
> 
> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
> index a52a5f2c..5ae2ef8a 100644
> --- a/include/libcamera/internal/v4l2_device.h
> +++ b/include/libcamera/internal/v4l2_device.h
> @@ -22,6 +22,8 @@
>  #include <libcamera/color_space.h>
>  #include <libcamera/controls.h>
>  
> +#include "libcamera/internal/formats.h"
> +
>  namespace libcamera {
>  
>  class EventNotifier;
> @@ -59,7 +61,8 @@ protected:
>  	int fd() const { return fd_.get(); }
>  
>  	template<typename T>
> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);

						      PixelFormatInfo::ColourEncoding colourEncoding);
Same below.

>  
>  	template<typename T>
>  	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
> index b22a981f..1fb08b9d 100644
> --- a/src/libcamera/v4l2_device.cpp
> +++ b/src/libcamera/v4l2_device.cpp
> @@ -24,6 +24,7 @@
>  #include <libcamera/base/log.h>
>  #include <libcamera/base/utils.h>
>  
> +#include "libcamera/internal/formats.h"
>  #include "libcamera/internal/sysfs.h"
>  
>  /**
> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
>   * \retval std::nullopt One or more V4L2 color space fields were not recognised
>   */
>  template<typename T>
> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
>  {
>  	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
>  	if (itColor == v4l2ToColorSpace.end())
> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>  			return std::nullopt;
>  
>  		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
> +

I'd add a comment here, or we'll forget why.

		/*
		 * V4L2 has no "none" encoding, override the value returned by
		 * the kernel for non-YUV formats as YCbCr encoding isn't
		 * applicable in that case..
		 */

> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)

Shouldn't this be

		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)

? Raw formats have no YCbCr encoding either. The subject line would need
to be updated accordingly, I'd write "Adjust colorspace based on pixel
format".

> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
>  	}
>  
>  	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>  			return std::nullopt;
>  
>  		colorSpace.range = itRange->second;
> +
> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)

Same here.

> +			colorSpace.range = ColorSpace::Range::Full;
>  	}
>  
>  	return colorSpace;
>  }
>  
> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
> +							    const PixelFormatInfo::ColourEncoding &);
> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
> +							    const PixelFormatInfo::ColourEncoding &);
> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
> +							    const PixelFormatInfo::ColourEncoding &);
>  
>  /**
>   * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> index ab74b9f0..a52c414f 100644
> --- a/src/libcamera/v4l2_subdevice.cpp
> +++ b/src/libcamera/v4l2_subdevice.cpp
> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>  	format->size.width = subdevFmt.format.width;
>  	format->size.height = subdevFmt.format.height;
>  	format->mbus_code = subdevFmt.format.code;
> -	format->colorSpace = toColorSpace(subdevFmt.format);
> +	auto iter = formatInfoMap.find(format->mbus_code);
> +	if (iter == formatInfoMap.end())
> +		return -EINVAL;

That seems a bit harsh. I'm thinking about the simple pipeline handler
in particular, which will propagate media bus formats through the
pipeline without caring much about the format itself. Would it better to
instead pick a default ?

> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>  
>  	return 0;
>  }
> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>  	format->size.width = subdevFmt.format.width;
>  	format->size.height = subdevFmt.format.height;
>  	format->mbus_code = subdevFmt.format.code;
> -	format->colorSpace = toColorSpace(subdevFmt.format);
> +	auto iter = formatInfoMap.find(format->mbus_code);
> +	if (iter == formatInfoMap.end())
> +		return -EINVAL;

Same here.

> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>  
>  	return 0;
>  }
> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
> index 5a2d0e5b..0e3f5436 100644
> --- a/src/libcamera/v4l2_videodevice.cpp
> +++ b/src/libcamera/v4l2_videodevice.cpp
> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)

I'm tempted to add a helper just above this function:

namespace {

std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
{
	V4L2PixelFormat fourcc{ pix.pixelformat };
	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
}

} /* namespace */

to make the color below a bit nicer (it would actually not need to be
changed at all). Up to you.

>  	format->size.height = pix->height;
>  	format->fourcc = V4L2PixelFormat(pix->pixelformat);
>  	format->planesCount = pix->num_planes;
> -	format->colorSpace = toColorSpace(*pix);
> +	format->colorSpace =
> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>  
>  	for (unsigned int i = 0; i < format->planesCount; ++i) {
>  		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
>  		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>  		format->planes[i].size = pix->plane_fmt[i].sizeimage;
>  	}
> -	format->colorSpace = toColorSpace(*pix);
> +	format->colorSpace =
> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>  
>  	return 0;
>  }
> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
>  	format->planesCount = 1;
>  	format->planes[0].bpl = pix->bytesperline;
>  	format->planes[0].size = pix->sizeimage;
> -	format->colorSpace = toColorSpace(*pix);
> +	format->colorSpace =
> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>  
>  	return 0;
>  }
> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
>  	format->planesCount = 1;
>  	format->planes[0].bpl = pix->bytesperline;
>  	format->planes[0].size = pix->sizeimage;
> -	format->colorSpace = toColorSpace(*pix);
> +	format->colorSpace =
> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>  
>  	return 0;
>  }

Another comment.

On Thu, Aug 25, 2022 at 02:05:57AM +0300, Laurent Pinchart via libcamera-devel wrote:
> Hi Umang,
> 
> Thank you for the patch.
> 
> On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
> > The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
> > This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
> > Rec 601 Y'CbCr encoding and limited range) in the kernel.
> > 
> > The RGB pixel formats should not use any Y'CbCr encoding and is always
> > full range. Adjust the colorspace before reporting back to the
> > userspace in such a situation.
> > 
> > Moving forwards, the ColorSpace::Srgb will be defined in the true sense
> > for RGB pixel formats.
> 
> Mentioning sRGB here is a bit confusing I think. Let's focus on what
> this patch does (and this will also help making sure I understood this
> correctly):
> 
> V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
> formats, including RGB and raw formats. This causes the libcamera
> ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
> overriding the encoding reported by the kernel to YCbCrEncoding::None
> for non-YUV pixel formats and media bus formats.
> 
> Similarly, override the quantization range of non-YUV formats to full
> range, as limited range isn't used for RGB and raw formats.
> 
> > Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
> > ---
> >  include/libcamera/internal/v4l2_device.h |  5 ++++-
> >  src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
> >  src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
> >  src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
> >  4 files changed, 35 insertions(+), 11 deletions(-)
> > 
> > diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
> > index a52a5f2c..5ae2ef8a 100644
> > --- a/include/libcamera/internal/v4l2_device.h
> > +++ b/include/libcamera/internal/v4l2_device.h
> > @@ -22,6 +22,8 @@
> >  #include <libcamera/color_space.h>
> >  #include <libcamera/controls.h>
> >  
> > +#include "libcamera/internal/formats.h"
> > +
> >  namespace libcamera {
> >  
> >  class EventNotifier;
> > @@ -59,7 +61,8 @@ protected:
> >  	int fd() const { return fd_.get(); }
> >  
> >  	template<typename T>
> > -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
> > +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
> > +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
> 
> 						      PixelFormatInfo::ColourEncoding colourEncoding);
> Same below.
> 
> >  
> >  	template<typename T>
> >  	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
> > diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
> > index b22a981f..1fb08b9d 100644
> > --- a/src/libcamera/v4l2_device.cpp
> > +++ b/src/libcamera/v4l2_device.cpp
> > @@ -24,6 +24,7 @@
> >  #include <libcamera/base/log.h>
> >  #include <libcamera/base/utils.h>
> >  
> > +#include "libcamera/internal/formats.h"
> >  #include "libcamera/internal/sysfs.h"
> >  
> >  /**
> > @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
> >   * \retval std::nullopt One or more V4L2 color space fields were not recognised
> >   */

The documentation needs to be updated with the new parameter.

> >  template<typename T>
> > -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> > +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
> > +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
> >  {
> >  	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
> >  	if (itColor == v4l2ToColorSpace.end())
> > @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >  			return std::nullopt;
> >  
> >  		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
> > +
> 
> I'd add a comment here, or we'll forget why.
> 
> 		/*
> 		 * V4L2 has no "none" encoding, override the value returned by
> 		 * the kernel for non-YUV formats as YCbCr encoding isn't
> 		 * applicable in that case..
> 		 */
> 
> > +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> 
> Shouldn't this be
> 
> 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
> 
> ? Raw formats have no YCbCr encoding either. The subject line would need
> to be updated accordingly, I'd write "Adjust colorspace based on pixel
> format".
> 
> > +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
> >  	}
> >  
> >  	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
> > @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >  			return std::nullopt;
> >  
> >  		colorSpace.range = itRange->second;
> > +
> > +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> 
> Same here.
> 
> > +			colorSpace.range = ColorSpace::Range::Full;
> >  	}
> >  
> >  	return colorSpace;
> >  }
> >  
> > -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
> > -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
> > -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
> > +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
> > +							    const PixelFormatInfo::ColourEncoding &);
> > +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
> > +							    const PixelFormatInfo::ColourEncoding &);
> > +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
> > +							    const PixelFormatInfo::ColourEncoding &);
> >  
> >  /**
> >   * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
> > diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> > index ab74b9f0..a52c414f 100644
> > --- a/src/libcamera/v4l2_subdevice.cpp
> > +++ b/src/libcamera/v4l2_subdevice.cpp
> > @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >  	format->size.width = subdevFmt.format.width;
> >  	format->size.height = subdevFmt.format.height;
> >  	format->mbus_code = subdevFmt.format.code;
> > -	format->colorSpace = toColorSpace(subdevFmt.format);
> > +	auto iter = formatInfoMap.find(format->mbus_code);
> > +	if (iter == formatInfoMap.end())
> > +		return -EINVAL;
> 
> That seems a bit harsh. I'm thinking about the simple pipeline handler
> in particular, which will propagate media bus formats through the
> pipeline without caring much about the format itself. Would it better to
> instead pick a default ?
> 
> > +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >  
> >  	return 0;
> >  }
> > @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >  	format->size.width = subdevFmt.format.width;
> >  	format->size.height = subdevFmt.format.height;
> >  	format->mbus_code = subdevFmt.format.code;
> > -	format->colorSpace = toColorSpace(subdevFmt.format);
> > +	auto iter = formatInfoMap.find(format->mbus_code);
> > +	if (iter == formatInfoMap.end())
> > +		return -EINVAL;
> 
> Same here.
> 
> > +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >  
> >  	return 0;
> >  }
> > diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
> > index 5a2d0e5b..0e3f5436 100644
> > --- a/src/libcamera/v4l2_videodevice.cpp
> > +++ b/src/libcamera/v4l2_videodevice.cpp
> > @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
> 
> I'm tempted to add a helper just above this function:
> 
> namespace {
> 
> std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
> {
> 	V4L2PixelFormat fourcc{ pix.pixelformat };
> 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
> }
> 
> } /* namespace */
> 
> to make the color below a bit nicer (it would actually not need to be
> changed at all). Up to you.
> 
> >  	format->size.height = pix->height;
> >  	format->fourcc = V4L2PixelFormat(pix->pixelformat);
> >  	format->planesCount = pix->num_planes;
> > -	format->colorSpace = toColorSpace(*pix);
> > +	format->colorSpace =
> > +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >  
> >  	for (unsigned int i = 0; i < format->planesCount; ++i) {
> >  		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> > @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
> >  		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >  		format->planes[i].size = pix->plane_fmt[i].sizeimage;
> >  	}
> > -	format->colorSpace = toColorSpace(*pix);
> > +	format->colorSpace =
> > +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >  
> >  	return 0;
> >  }
> > @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
> >  	format->planesCount = 1;
> >  	format->planes[0].bpl = pix->bytesperline;
> >  	format->planes[0].size = pix->sizeimage;
> > -	format->colorSpace = toColorSpace(*pix);
> > +	format->colorSpace =
> > +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >  
> >  	return 0;
> >  }
> > @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
> >  	format->planesCount = 1;
> >  	format->planes[0].bpl = pix->bytesperline;
> >  	format->planes[0].size = pix->sizeimage;
> > -	format->colorSpace = toColorSpace(*pix);
> > +	format->colorSpace =
> > +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >  
> >  	return 0;
> >  }

Hi Laurent,

Thank you for the review.

On 8/25/22 4:35 AM, Laurent Pinchart wrote:
> Hi Umang,
>
> Thank you for the patch.
>
> On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
>> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
>> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
>> Rec 601 Y'CbCr encoding and limited range) in the kernel.
>>
>> The RGB pixel formats should not use any Y'CbCr encoding and is always
>> full range. Adjust the colorspace before reporting back to the
>> userspace in such a situation.
>>
>> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
>> for RGB pixel formats.
> Mentioning sRGB here is a bit confusing I think. Let's focus on what
> this patch does (and this will also help making sure I understood this
> correctly):
>
> V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
> formats, including RGB and raw formats. This causes the libcamera
> ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
> overriding the encoding reported by the kernel to YCbCrEncoding::None
> for non-YUV pixel formats and media bus formats.

Correct.
>
> Similarly, override the quantization range of non-YUV formats to full
> range, as limited range isn't used for RGB and raw formats.

Seems correct too, I agree my commit message was not at it's best.
>
>> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
>> ---
>>   include/libcamera/internal/v4l2_device.h |  5 ++++-
>>   src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
>>   src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
>>   src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
>>   4 files changed, 35 insertions(+), 11 deletions(-)
>>
>> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
>> index a52a5f2c..5ae2ef8a 100644
>> --- a/include/libcamera/internal/v4l2_device.h
>> +++ b/include/libcamera/internal/v4l2_device.h
>> @@ -22,6 +22,8 @@
>>   #include <libcamera/color_space.h>
>>   #include <libcamera/controls.h>
>>   
>> +#include "libcamera/internal/formats.h"
>> +
>>   namespace libcamera {
>>   
>>   class EventNotifier;
>> @@ -59,7 +61,8 @@ protected:
>>   	int fd() const { return fd_.get(); }
>>   
>>   	template<typename T>
>> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
>> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
>> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
> 						      PixelFormatInfo::ColourEncoding colourEncoding);
> Same below.
>
>>   
>>   	template<typename T>
>>   	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
>> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
>> index b22a981f..1fb08b9d 100644
>> --- a/src/libcamera/v4l2_device.cpp
>> +++ b/src/libcamera/v4l2_device.cpp
>> @@ -24,6 +24,7 @@
>>   #include <libcamera/base/log.h>
>>   #include <libcamera/base/utils.h>
>>   
>> +#include "libcamera/internal/formats.h"
>>   #include "libcamera/internal/sysfs.h"
>>   
>>   /**
>> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
>>    * \retval std::nullopt One or more V4L2 color space fields were not recognised
>>    */
>>   template<typename T>
>> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
>> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
>>   {
>>   	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
>>   	if (itColor == v4l2ToColorSpace.end())
>> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>   			return std::nullopt;
>>   
>>   		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
>> +
> I'd add a comment here, or we'll forget why.
>
> 		/*
> 		 * V4L2 has no "none" encoding, override the value returned by
> 		 * the kernel for non-YUV formats as YCbCr encoding isn't
> 		 * applicable in that case..
> 		 */
>
>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> Shouldn't this be
>
> 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
>
> ? Raw formats have no YCbCr encoding either. The subject line would need
> to be updated accordingly, I'd write "Adjust colorspace based on pixel
> format".
>
>> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
>>   	}
>>   
>>   	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
>> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>   			return std::nullopt;
>>   
>>   		colorSpace.range = itRange->second;
>> +
>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> Same here.
>
>> +			colorSpace.range = ColorSpace::Range::Full;
>>   	}
>>   
>>   	return colorSpace;
>>   }
>>   
>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
>> +							    const PixelFormatInfo::ColourEncoding &);
>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
>> +							    const PixelFormatInfo::ColourEncoding &);
>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
>> +							    const PixelFormatInfo::ColourEncoding &);
>>   
>>   /**
>>    * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
>> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
>> index ab74b9f0..a52c414f 100644
>> --- a/src/libcamera/v4l2_subdevice.cpp
>> +++ b/src/libcamera/v4l2_subdevice.cpp
>> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>   	format->size.width = subdevFmt.format.width;
>>   	format->size.height = subdevFmt.format.height;
>>   	format->mbus_code = subdevFmt.format.code;
>> -	format->colorSpace = toColorSpace(subdevFmt.format);
>> +	auto iter = formatInfoMap.find(format->mbus_code);
>> +	if (iter == formatInfoMap.end())
>> +		return -EINVAL;
> That seems a bit harsh. I'm thinking about the simple pipeline handler
> in particular, which will propagate media bus formats through the
> pipeline without caring much about the format itself. Would it better to
> instead pick a default ?

I wasn't aware of that. I assumed (formatInfoMap.find(format->mbus_code) 
== formatInfoMap.end()) would be very hard to trigger and if it does at 
all, we want to fail loudly.

I am not much aware of simple pipeline handler format's passing - I'll 
take a look to answer the question - "Which default to pick"
>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>   
>>   	return 0;
>>   }
>> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>   	format->size.width = subdevFmt.format.width;
>>   	format->size.height = subdevFmt.format.height;
>>   	format->mbus_code = subdevFmt.format.code;
>> -	format->colorSpace = toColorSpace(subdevFmt.format);
>> +	auto iter = formatInfoMap.find(format->mbus_code);
>> +	if (iter == formatInfoMap.end())
>> +		return -EINVAL;
> Same here.
>
>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>   
>>   	return 0;
>>   }
>> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
>> index 5a2d0e5b..0e3f5436 100644
>> --- a/src/libcamera/v4l2_videodevice.cpp
>> +++ b/src/libcamera/v4l2_videodevice.cpp
>> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
> I'm tempted to add a helper just above this function:
>
> namespace {
>
> std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
> {
> 	V4L2PixelFormat fourcc{ pix.pixelformat };
> 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
> }
>
> } /* namespace */
>
> to make the color below a bit nicer (it would actually not need to be
> changed at all). Up to you.

Ack
>
>>   	format->size.height = pix->height;
>>   	format->fourcc = V4L2PixelFormat(pix->pixelformat);
>>   	format->planesCount = pix->num_planes;
>> -	format->colorSpace = toColorSpace(*pix);
>> +	format->colorSpace =
>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>   
>>   	for (unsigned int i = 0; i < format->planesCount; ++i) {
>>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
>>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>>   		format->planes[i].size = pix->plane_fmt[i].sizeimage;
>>   	}
>> -	format->colorSpace = toColorSpace(*pix);
>> +	format->colorSpace =
>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>   
>>   	return 0;
>>   }
>> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
>>   	format->planesCount = 1;
>>   	format->planes[0].bpl = pix->bytesperline;
>>   	format->planes[0].size = pix->sizeimage;
>> -	format->colorSpace = toColorSpace(*pix);
>> +	format->colorSpace =
>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>   
>>   	return 0;
>>   }
>> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
>>   	format->planesCount = 1;
>>   	format->planes[0].bpl = pix->bytesperline;
>>   	format->planes[0].size = pix->sizeimage;
>> -	format->colorSpace = toColorSpace(*pix);
>> +	format->colorSpace =
>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>   
>>   	return 0;
>>   }

Hi Laurent,

On 8/25/22 8:39 PM, Laurent Pinchart wrote:
> Another comment.
>
> On Thu, Aug 25, 2022 at 02:05:57AM +0300, Laurent Pinchart via libcamera-devel wrote:
>> Hi Umang,
>>
>> Thank you for the patch.
>>
>> On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
>>> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
>>> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
>>> Rec 601 Y'CbCr encoding and limited range) in the kernel.
>>>
>>> The RGB pixel formats should not use any Y'CbCr encoding and is always
>>> full range. Adjust the colorspace before reporting back to the
>>> userspace in such a situation.
>>>
>>> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
>>> for RGB pixel formats.
>> Mentioning sRGB here is a bit confusing I think. Let's focus on what
>> this patch does (and this will also help making sure I understood this
>> correctly):
>>
>> V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
>> formats, including RGB and raw formats. This causes the libcamera
>> ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
>> overriding the encoding reported by the kernel to YCbCrEncoding::None
>> for non-YUV pixel formats and media bus formats.
>>
>> Similarly, override the quantization range of non-YUV formats to full
>> range, as limited range isn't used for RGB and raw formats.
>>
>>> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
>>> ---
>>>   include/libcamera/internal/v4l2_device.h |  5 ++++-
>>>   src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
>>>   src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
>>>   src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
>>>   4 files changed, 35 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
>>> index a52a5f2c..5ae2ef8a 100644
>>> --- a/include/libcamera/internal/v4l2_device.h
>>> +++ b/include/libcamera/internal/v4l2_device.h
>>> @@ -22,6 +22,8 @@
>>>   #include <libcamera/color_space.h>
>>>   #include <libcamera/controls.h>
>>>   
>>> +#include "libcamera/internal/formats.h"
>>> +
>>>   namespace libcamera {
>>>   
>>>   class EventNotifier;
>>> @@ -59,7 +61,8 @@ protected:
>>>   	int fd() const { return fd_.get(); }
>>>   
>>>   	template<typename T>
>>> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
>>> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
>>> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
>> 						      PixelFormatInfo::ColourEncoding colourEncoding);
>> Same below.
>>
>>>   
>>>   	template<typename T>
>>>   	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
>>> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
>>> index b22a981f..1fb08b9d 100644
>>> --- a/src/libcamera/v4l2_device.cpp
>>> +++ b/src/libcamera/v4l2_device.cpp
>>> @@ -24,6 +24,7 @@
>>>   #include <libcamera/base/log.h>
>>>   #include <libcamera/base/utils.h>
>>>   
>>> +#include "libcamera/internal/formats.h"
>>>   #include "libcamera/internal/sysfs.h"
>>>   
>>>   /**
>>> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
>>>    * \retval std::nullopt One or more V4L2 color space fields were not recognised
>>>    */
> The documentation needs to be updated with the new parameter.
>
>>>   template<typename T>
>>> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
>>> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
>>>   {
>>>   	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
>>>   	if (itColor == v4l2ToColorSpace.end())
>>> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>>   			return std::nullopt;
>>>   
>>>   		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
>>> +
>> I'd add a comment here, or we'll forget why.
>>
>> 		/*
>> 		 * V4L2 has no "none" encoding, override the value returned by
>> 		 * the kernel for non-YUV formats as YCbCr encoding isn't
>> 		 * applicable in that case..
>> 		 */
>>
>>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
>> Shouldn't this be
>>
>> 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
>>
>> ? Raw formats have no YCbCr encoding either. The subject line would need
>> to be updated accordingly, I'd write "Adjust colorspace based on pixel
>> format".
>>
>>> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
>>>   	}
>>>   
>>>   	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
>>> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>>   			return std::nullopt;
>>>   
>>>   		colorSpace.range = itRange->second;
>>> +
>>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
>> Same here.
>>
>>> +			colorSpace.range = ColorSpace::Range::Full;
>>>   	}
>>>   
>>>   	return colorSpace;
>>>   }
>>>   
>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
>>> +							    const PixelFormatInfo::ColourEncoding &);
>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
>>> +							    const PixelFormatInfo::ColourEncoding &);
>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
>>> +							    const PixelFormatInfo::ColourEncoding &);
>>>   
>>>   /**
>>>    * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
>>> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
>>> index ab74b9f0..a52c414f 100644
>>> --- a/src/libcamera/v4l2_subdevice.cpp
>>> +++ b/src/libcamera/v4l2_subdevice.cpp
>>> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>>   	format->size.width = subdevFmt.format.width;
>>>   	format->size.height = subdevFmt.format.height;
>>>   	format->mbus_code = subdevFmt.format.code;
>>> -	format->colorSpace = toColorSpace(subdevFmt.format);
>>> +	auto iter = formatInfoMap.find(format->mbus_code);
>>> +	if (iter == formatInfoMap.end())
>>> +		return -EINVAL;
>> That seems a bit harsh. I'm thinking about the simple pipeline handler
>> in particular, which will propagate media bus formats through the
>> pipeline without caring much about the format itself. Would it better to
>> instead pick a default ?
>>
>>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>>   
>>>   	return 0;
>>>   }
>>> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>>   	format->size.width = subdevFmt.format.width;
>>>   	format->size.height = subdevFmt.format.height;
>>>   	format->mbus_code = subdevFmt.format.code;
>>> -	format->colorSpace = toColorSpace(subdevFmt.format);
>>> +	auto iter = formatInfoMap.find(format->mbus_code);
>>> +	if (iter == formatInfoMap.end())
>>> +		return -EINVAL;
>> Same here.

Esp in V4l2Subdevice::setFormat() , picking a default and reporting that 
back makes me a bit unconformtable ....

should we chose to fail hard here?

diff --git a/src/libcamera/v4l2_subdevice.cpp 
b/src/libcamera/v4l2_subdevice.cpp
index 70f179c2..d415a738 100644
--- a/src/libcamera/v4l2_subdevice.cpp
+++ b/src/libcamera/v4l2_subdevice.cpp
@@ -502,10 +502,19 @@ int V4L2Subdevice::getFormat(unsigned int pad, 
V4L2SubdeviceFormat *format,

         format->size.width = subdevFmt.format.width;
         format->size.height = subdevFmt.format.height;
+
+       auto iter = formatInfoMap.find(subdevFmt.format.code);
+       if (iter == formatInfoMap.end()) {
+               /* Pick a default format to continue */
+               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
+               iter = formatInfoMap.find(defCode);
+               LOG(V4L2, Error)
+                        << "Mapping for subdev format " << 
subdevFmt.format.code
+                        << " is missing, falling back to " << 
iter->second.name;
+
+               subdevFmt.format.code = defCode;
+       }
         format->mbus_code = subdevFmt.format.code;
-       auto iter = formatInfoMap.find(format->mbus_code);
-       if (iter == formatInfoMap.end())
-               return -EINVAL;
         format->colorSpace = toColorSpace(subdevFmt.format, 
iter->second.colourEncoding);

         return 0;
@@ -551,10 +560,20 @@ int V4L2Subdevice::setFormat(unsigned int pad, 
V4L2SubdeviceFormat *format,

         format->size.width = subdevFmt.format.width;
         format->size.height = subdevFmt.format.height;
+
+       auto iter = formatInfoMap.find(subdevFmt.format.code);
+       if (iter == formatInfoMap.end()) {
+               /* Pick a default format to continue */
+               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
+               iter = formatInfoMap.find(defCode);
+
+               LOG(V4L2, Error)
+                       << "Mapping for subdev format " << 
subdevFmt.format.code
+                       << " is missing, falling back to " << 
iter->second.name;
+
+               subdevFmt.format.code = defCode;
+       }
         format->mbus_code = subdevFmt.format.code;
-       auto iter = formatInfoMap.find(format->mbus_code);
-       if (iter == formatInfoMap.end())
-               return -EINVAL;
         format->colorSpace = toColorSpace(subdevFmt.format, 
iter->second.colourEncoding);

         return 0;

>>
>>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>>   
>>>   	return 0;
>>>   }
>>> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
>>> index 5a2d0e5b..0e3f5436 100644
>>> --- a/src/libcamera/v4l2_videodevice.cpp
>>> +++ b/src/libcamera/v4l2_videodevice.cpp
>>> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
>> I'm tempted to add a helper just above this function:
>>
>> namespace {
>>
>> std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
>> {
>> 	V4L2PixelFormat fourcc{ pix.pixelformat };
>> 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
>> }
>>
>> } /* namespace */
>>
>> to make the color below a bit nicer (it would actually not need to be
>> changed at all). Up to you.
>>
>>>   	format->size.height = pix->height;
>>>   	format->fourcc = V4L2PixelFormat(pix->pixelformat);
>>>   	format->planesCount = pix->num_planes;
>>> -	format->colorSpace = toColorSpace(*pix);
>>> +	format->colorSpace =
>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>   
>>>   	for (unsigned int i = 0; i < format->planesCount; ++i) {
>>>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>>> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
>>>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>>>   		format->planes[i].size = pix->plane_fmt[i].sizeimage;
>>>   	}
>>> -	format->colorSpace = toColorSpace(*pix);
>>> +	format->colorSpace =
>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>   
>>>   	return 0;
>>>   }
>>> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
>>>   	format->planesCount = 1;
>>>   	format->planes[0].bpl = pix->bytesperline;
>>>   	format->planes[0].size = pix->sizeimage;
>>> -	format->colorSpace = toColorSpace(*pix);
>>> +	format->colorSpace =
>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>   
>>>   	return 0;
>>>   }
>>> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
>>>   	format->planesCount = 1;
>>>   	format->planes[0].bpl = pix->bytesperline;
>>>   	format->planes[0].size = pix->sizeimage;
>>> -	format->colorSpace = toColorSpace(*pix);
>>> +	format->colorSpace =
>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>   
>>>   	return 0;
>>>   }

Hi Umang,

On Fri, Aug 26, 2022 at 11:46:58AM +0530, Umang Jain wrote:
> On 8/25/22 4:35 AM, Laurent Pinchart wrote:
> > On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
> >> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
> >> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
> >> Rec 601 Y'CbCr encoding and limited range) in the kernel.
> >>
> >> The RGB pixel formats should not use any Y'CbCr encoding and is always
> >> full range. Adjust the colorspace before reporting back to the
> >> userspace in such a situation.
> >>
> >> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
> >> for RGB pixel formats.
> > 
> > Mentioning sRGB here is a bit confusing I think. Let's focus on what
> > this patch does (and this will also help making sure I understood this
> > correctly):
> >
> > V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
> > formats, including RGB and raw formats. This causes the libcamera
> > ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
> > overriding the encoding reported by the kernel to YCbCrEncoding::None
> > for non-YUV pixel formats and media bus formats.
> 
> Correct.
> 
> > Similarly, override the quantization range of non-YUV formats to full
> > range, as limited range isn't used for RGB and raw formats.
> 
> Seems correct too, I agree my commit message was not at it's best.
> 
> >> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
> >> ---
> >>   include/libcamera/internal/v4l2_device.h |  5 ++++-
> >>   src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
> >>   src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
> >>   src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
> >>   4 files changed, 35 insertions(+), 11 deletions(-)
> >>
> >> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
> >> index a52a5f2c..5ae2ef8a 100644
> >> --- a/include/libcamera/internal/v4l2_device.h
> >> +++ b/include/libcamera/internal/v4l2_device.h
> >> @@ -22,6 +22,8 @@
> >>   #include <libcamera/color_space.h>
> >>   #include <libcamera/controls.h>
> >>   
> >> +#include "libcamera/internal/formats.h"
> >> +
> >>   namespace libcamera {
> >>   
> >>   class EventNotifier;
> >> @@ -59,7 +61,8 @@ protected:
> >>   	int fd() const { return fd_.get(); }
> >>   
> >>   	template<typename T>
> >> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
> >> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
> >> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
> > 						      PixelFormatInfo::ColourEncoding colourEncoding);
> > Same below.
> >
> >>   
> >>   	template<typename T>
> >>   	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
> >> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
> >> index b22a981f..1fb08b9d 100644
> >> --- a/src/libcamera/v4l2_device.cpp
> >> +++ b/src/libcamera/v4l2_device.cpp
> >> @@ -24,6 +24,7 @@
> >>   #include <libcamera/base/log.h>
> >>   #include <libcamera/base/utils.h>
> >>   
> >> +#include "libcamera/internal/formats.h"
> >>   #include "libcamera/internal/sysfs.h"
> >>   
> >>   /**
> >> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
> >>    * \retval std::nullopt One or more V4L2 color space fields were not recognised
> >>    */
> >>   template<typename T>
> >> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
> >> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
> >>   {
> >>   	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
> >>   	if (itColor == v4l2ToColorSpace.end())
> >> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>   			return std::nullopt;
> >>   
> >>   		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
> >> +
> > 
> > I'd add a comment here, or we'll forget why.
> >
> > 		/*
> > 		 * V4L2 has no "none" encoding, override the value returned by
> > 		 * the kernel for non-YUV formats as YCbCr encoding isn't
> > 		 * applicable in that case..
> > 		 */
> >
> >> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> > 
> > Shouldn't this be
> >
> > 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
> >
> > ? Raw formats have no YCbCr encoding either. The subject line would need
> > to be updated accordingly, I'd write "Adjust colorspace based on pixel
> > format".
> >
> >> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
> >>   	}
> >>   
> >>   	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
> >> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>   			return std::nullopt;
> >>   
> >>   		colorSpace.range = itRange->second;
> >> +
> >> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> > 
> > Same here.
> >
> >> +			colorSpace.range = ColorSpace::Range::Full;
> >>   	}
> >>   
> >>   	return colorSpace;
> >>   }
> >>   
> >> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
> >> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
> >> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
> >> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
> >> +							    const PixelFormatInfo::ColourEncoding &);
> >> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
> >> +							    const PixelFormatInfo::ColourEncoding &);
> >> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
> >> +							    const PixelFormatInfo::ColourEncoding &);
> >>   
> >>   /**
> >>    * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
> >> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> >> index ab74b9f0..a52c414f 100644
> >> --- a/src/libcamera/v4l2_subdevice.cpp
> >> +++ b/src/libcamera/v4l2_subdevice.cpp
> >> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>   	format->size.width = subdevFmt.format.width;
> >>   	format->size.height = subdevFmt.format.height;
> >>   	format->mbus_code = subdevFmt.format.code;
> >> -	format->colorSpace = toColorSpace(subdevFmt.format);
> >> +	auto iter = formatInfoMap.find(format->mbus_code);
> >> +	if (iter == formatInfoMap.end())
> >> +		return -EINVAL;
> > 
> > That seems a bit harsh. I'm thinking about the simple pipeline handler
> > in particular, which will propagate media bus formats through the
> > pipeline without caring much about the format itself. Would it better to
> > instead pick a default ?
> 
> I wasn't aware of that. I assumed (formatInfoMap.find(format->mbus_code) 
> == formatInfoMap.end()) would be very hard to trigger and if it does at 
> all, we want to fail loudly.

Currently, we only use formatInfoMap in two places:

- in V4L2SubdeviceFormat::bitsPerPixel()
- in the << operator

The former will complain with an error message and return 0 if the media
bus code is unknown, the latter will print the hex value.

bitsPerPixel() is used in CameraSensor::sensorInfo() only. We thus
complain already if a sensor has an unknown output format. We however
don't complain about unknown formats anywhere else, and there may thus
be places where we already handle them silently. I'm thinking in
particular about format propagation in the simple pipeline handler,
where we just get the format on the output of a subdevice and set it on
the input of the connected subdevice, without caring about the exact
format. We may also be use MEDIA_BUS_FMT_METADATA_FIXED without knowing
it, with the change here an attempt to call getFormat() on a subdev pad
that carries metadata will return in -EINVAL.

I would prefer keeping the current behaviour of selecting a reasonable
default in this patch series, and then add a stricter check on top if we
want to go that way.

> I am not much aware of simple pipeline handler format's passing - I'll 
> take a look to answer the question - "Which default to pick"
> 
> >> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>   
> >>   	return 0;
> >>   }
> >> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>   	format->size.width = subdevFmt.format.width;
> >>   	format->size.height = subdevFmt.format.height;
> >>   	format->mbus_code = subdevFmt.format.code;
> >> -	format->colorSpace = toColorSpace(subdevFmt.format);
> >> +	auto iter = formatInfoMap.find(format->mbus_code);
> >> +	if (iter == formatInfoMap.end())
> >> +		return -EINVAL;
> > 
> > Same here.
> >
> >> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>   
> >>   	return 0;
> >>   }
> >> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
> >> index 5a2d0e5b..0e3f5436 100644
> >> --- a/src/libcamera/v4l2_videodevice.cpp
> >> +++ b/src/libcamera/v4l2_videodevice.cpp
> >> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
> > 
> > I'm tempted to add a helper just above this function:
> >
> > namespace {
> >
> > std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
> > {
> > 	V4L2PixelFormat fourcc{ pix.pixelformat };
> > 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
> > }
> >
> > } /* namespace */
> >
> > to make the color below a bit nicer (it would actually not need to be
> > changed at all). Up to you.
> 
> Ack
> 
> >>   	format->size.height = pix->height;
> >>   	format->fourcc = V4L2PixelFormat(pix->pixelformat);
> >>   	format->planesCount = pix->num_planes;
> >> -	format->colorSpace = toColorSpace(*pix);
> >> +	format->colorSpace =
> >> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>   
> >>   	for (unsigned int i = 0; i < format->planesCount; ++i) {
> >>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
> >>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >>   		format->planes[i].size = pix->plane_fmt[i].sizeimage;
> >>   	}
> >> -	format->colorSpace = toColorSpace(*pix);
> >> +	format->colorSpace =
> >> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>   
> >>   	return 0;
> >>   }
> >> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
> >>   	format->planesCount = 1;
> >>   	format->planes[0].bpl = pix->bytesperline;
> >>   	format->planes[0].size = pix->sizeimage;
> >> -	format->colorSpace = toColorSpace(*pix);
> >> +	format->colorSpace =
> >> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>   
> >>   	return 0;
> >>   }
> >> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
> >>   	format->planesCount = 1;
> >>   	format->planes[0].bpl = pix->bytesperline;
> >>   	format->planes[0].size = pix->sizeimage;
> >> -	format->colorSpace = toColorSpace(*pix);
> >> +	format->colorSpace =
> >> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>   
> >>   	return 0;
> >>   }

Hi Umang,

On Fri, Aug 26, 2022 at 01:56:20PM +0530, Umang Jain wrote:
> On 8/25/22 8:39 PM, Laurent Pinchart wrote:
> > On Thu, Aug 25, 2022 at 02:05:57AM +0300, Laurent Pinchart via libcamera-devel wrote:
> >> On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
> >>> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
> >>> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
> >>> Rec 601 Y'CbCr encoding and limited range) in the kernel.
> >>>
> >>> The RGB pixel formats should not use any Y'CbCr encoding and is always
> >>> full range. Adjust the colorspace before reporting back to the
> >>> userspace in such a situation.
> >>>
> >>> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
> >>> for RGB pixel formats.
> >>
> >> Mentioning sRGB here is a bit confusing I think. Let's focus on what
> >> this patch does (and this will also help making sure I understood this
> >> correctly):
> >>
> >> V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
> >> formats, including RGB and raw formats. This causes the libcamera
> >> ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
> >> overriding the encoding reported by the kernel to YCbCrEncoding::None
> >> for non-YUV pixel formats and media bus formats.
> >>
> >> Similarly, override the quantization range of non-YUV formats to full
> >> range, as limited range isn't used for RGB and raw formats.
> >>
> >>> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
> >>> ---
> >>>   include/libcamera/internal/v4l2_device.h |  5 ++++-
> >>>   src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
> >>>   src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
> >>>   src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
> >>>   4 files changed, 35 insertions(+), 11 deletions(-)
> >>>
> >>> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
> >>> index a52a5f2c..5ae2ef8a 100644
> >>> --- a/include/libcamera/internal/v4l2_device.h
> >>> +++ b/include/libcamera/internal/v4l2_device.h
> >>> @@ -22,6 +22,8 @@
> >>>   #include <libcamera/color_space.h>
> >>>   #include <libcamera/controls.h>
> >>>   
> >>> +#include "libcamera/internal/formats.h"
> >>> +
> >>>   namespace libcamera {
> >>>   
> >>>   class EventNotifier;
> >>> @@ -59,7 +61,8 @@ protected:
> >>>   	int fd() const { return fd_.get(); }
> >>>   
> >>>   	template<typename T>
> >>> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
> >>> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
> >>> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
> >>
> >> 						      PixelFormatInfo::ColourEncoding colourEncoding);
> >> Same below.
> >>
> >>>   
> >>>   	template<typename T>
> >>>   	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
> >>> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
> >>> index b22a981f..1fb08b9d 100644
> >>> --- a/src/libcamera/v4l2_device.cpp
> >>> +++ b/src/libcamera/v4l2_device.cpp
> >>> @@ -24,6 +24,7 @@
> >>>   #include <libcamera/base/log.h>
> >>>   #include <libcamera/base/utils.h>
> >>>   
> >>> +#include "libcamera/internal/formats.h"
> >>>   #include "libcamera/internal/sysfs.h"
> >>>   
> >>>   /**
> >>> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
> >>>    * \retval std::nullopt One or more V4L2 color space fields were not recognised
> >>>    */
> > The documentation needs to be updated with the new parameter.
> >
> >>>   template<typename T>
> >>> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
> >>> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
> >>>   {
> >>>   	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
> >>>   	if (itColor == v4l2ToColorSpace.end())
> >>> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>>   			return std::nullopt;
> >>>   
> >>>   		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
> >>> +
> >>
> >> I'd add a comment here, or we'll forget why.
> >>
> >> 		/*
> >> 		 * V4L2 has no "none" encoding, override the value returned by
> >> 		 * the kernel for non-YUV formats as YCbCr encoding isn't
> >> 		 * applicable in that case..
> >> 		 */
> >>
> >>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> >>
> >> Shouldn't this be
> >>
> >> 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
> >>
> >> ? Raw formats have no YCbCr encoding either. The subject line would need
> >> to be updated accordingly, I'd write "Adjust colorspace based on pixel
> >> format".
> >>
> >>> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
> >>>   	}
> >>>   
> >>>   	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
> >>> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>>   			return std::nullopt;
> >>>   
> >>>   		colorSpace.range = itRange->second;
> >>> +
> >>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> >>
> >> Same here.
> >>
> >>> +			colorSpace.range = ColorSpace::Range::Full;
> >>>   	}
> >>>   
> >>>   	return colorSpace;
> >>>   }
> >>>   
> >>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
> >>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
> >>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
> >>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
> >>> +							    const PixelFormatInfo::ColourEncoding &);
> >>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
> >>> +							    const PixelFormatInfo::ColourEncoding &);
> >>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
> >>> +							    const PixelFormatInfo::ColourEncoding &);
> >>>   
> >>>   /**
> >>>    * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
> >>> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> >>> index ab74b9f0..a52c414f 100644
> >>> --- a/src/libcamera/v4l2_subdevice.cpp
> >>> +++ b/src/libcamera/v4l2_subdevice.cpp
> >>> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>>   	format->size.width = subdevFmt.format.width;
> >>>   	format->size.height = subdevFmt.format.height;
> >>>   	format->mbus_code = subdevFmt.format.code;
> >>> -	format->colorSpace = toColorSpace(subdevFmt.format);
> >>> +	auto iter = formatInfoMap.find(format->mbus_code);
> >>> +	if (iter == formatInfoMap.end())
> >>> +		return -EINVAL;
> >>
> >> That seems a bit harsh. I'm thinking about the simple pipeline handler
> >> in particular, which will propagate media bus formats through the
> >> pipeline without caring much about the format itself. Would it better to
> >> instead pick a default ?
> >>
> >>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>>   
> >>>   	return 0;
> >>>   }
> >>> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>>   	format->size.width = subdevFmt.format.width;
> >>>   	format->size.height = subdevFmt.format.height;
> >>>   	format->mbus_code = subdevFmt.format.code;
> >>> -	format->colorSpace = toColorSpace(subdevFmt.format);
> >>> +	auto iter = formatInfoMap.find(format->mbus_code);
> >>> +	if (iter == formatInfoMap.end())
> >>> +		return -EINVAL;
> >>
> >> Same here.
> 
> Esp in V4l2Subdevice::setFormat() , picking a default and reporting that 
> back makes me a bit unconformtable ....
> 
> should we chose to fail hard here?

For the reasons explained in another reply in this thread, I'd rather
add hard failures separately, if we decide it's the way to go. We would
then need to consider the impact on different parts of libcamera (simple
pipeline handler, metadata formats, ...). Even an error/warning message
is something I would introduce in a separate patch on top, to be able to
test it thoroughly without blocking the rest of the series.

> diff --git a/src/libcamera/v4l2_subdevice.cpp 
> b/src/libcamera/v4l2_subdevice.cpp
> index 70f179c2..d415a738 100644
> --- a/src/libcamera/v4l2_subdevice.cpp
> +++ b/src/libcamera/v4l2_subdevice.cpp
> @@ -502,10 +502,19 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> 
>          format->size.width = subdevFmt.format.width;
>          format->size.height = subdevFmt.format.height;
> +
> +       auto iter = formatInfoMap.find(subdevFmt.format.code);
> +       if (iter == formatInfoMap.end()) {
> +               /* Pick a default format to continue */
> +               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
> +               iter = formatInfoMap.find(defCode);

You can avoid the lookup there and just use the RGB encoding:

	PixelFormatInfo::ColourEncoding colourEncoding;
	auto iter = formatInfoMap.find(subdevFmt.format.code);
	if (iter != formatInfoMap.end()) {
		colourEncoding = iter->second.colourEncoding;
	} else {
		LOG(V4L2, Warning)
			<< "Unknown subdev format "
			<< utils::hex(subdevFmt.format.code, 4)
			<< ", defaulting to RGB encoding";

		colourEncoding = PixelFormatInfo::ColourEncodingRGB;
	}

> +               LOG(V4L2, Error)

Make it a warning.

> +                        << "Mapping for subdev format " << subdevFmt.format.code

Print the code in hex.

> +                        << " is missing, falling back to " << iter->second.name;
> +
> +               subdevFmt.format.code = defCode;

Why do you override the media bus code returned by the driver ?

All these comments apply below.

> +       }
>          format->mbus_code = subdevFmt.format.code;
> -       auto iter = formatInfoMap.find(format->mbus_code);
> -       if (iter == formatInfoMap.end())
> -               return -EINVAL;
>          format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> 
>          return 0;
> @@ -551,10 +560,20 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> 
>          format->size.width = subdevFmt.format.width;
>          format->size.height = subdevFmt.format.height;
> +
> +       auto iter = formatInfoMap.find(subdevFmt.format.code);
> +       if (iter == formatInfoMap.end()) {
> +               /* Pick a default format to continue */
> +               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
> +               iter = formatInfoMap.find(defCode);
> +
> +               LOG(V4L2, Error)
> +                       << "Mapping for subdev format " << subdevFmt.format.code
> +                       << " is missing, falling back to " << iter->second.name;
> +
> +               subdevFmt.format.code = defCode;
> +       }
>          format->mbus_code = subdevFmt.format.code;
> -       auto iter = formatInfoMap.find(format->mbus_code);
> -       if (iter == formatInfoMap.end())
> -               return -EINVAL;
>          format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> 
>          return 0;
> 
> >>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>>   
> >>>   	return 0;
> >>>   }
> >>> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
> >>> index 5a2d0e5b..0e3f5436 100644
> >>> --- a/src/libcamera/v4l2_videodevice.cpp
> >>> +++ b/src/libcamera/v4l2_videodevice.cpp
> >>> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
> >>
> >> I'm tempted to add a helper just above this function:
> >>
> >> namespace {
> >>
> >> std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
> >> {
> >> 	V4L2PixelFormat fourcc{ pix.pixelformat };
> >> 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
> >> }
> >>
> >> } /* namespace */
> >>
> >> to make the color below a bit nicer (it would actually not need to be
> >> changed at all). Up to you.
> >>
> >>>   	format->size.height = pix->height;
> >>>   	format->fourcc = V4L2PixelFormat(pix->pixelformat);
> >>>   	format->planesCount = pix->num_planes;
> >>> -	format->colorSpace = toColorSpace(*pix);
> >>> +	format->colorSpace =
> >>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>   
> >>>   	for (unsigned int i = 0; i < format->planesCount; ++i) {
> >>>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >>> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
> >>>   		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >>>   		format->planes[i].size = pix->plane_fmt[i].sizeimage;
> >>>   	}
> >>> -	format->colorSpace = toColorSpace(*pix);
> >>> +	format->colorSpace =
> >>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>   
> >>>   	return 0;
> >>>   }
> >>> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
> >>>   	format->planesCount = 1;
> >>>   	format->planes[0].bpl = pix->bytesperline;
> >>>   	format->planes[0].size = pix->sizeimage;
> >>> -	format->colorSpace = toColorSpace(*pix);
> >>> +	format->colorSpace =
> >>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>   
> >>>   	return 0;
> >>>   }
> >>> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
> >>>   	format->planesCount = 1;
> >>>   	format->planes[0].bpl = pix->bytesperline;
> >>>   	format->planes[0].size = pix->sizeimage;
> >>> -	format->colorSpace = toColorSpace(*pix);
> >>> +	format->colorSpace =
> >>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>   
> >>>   	return 0;
> >>>   }

Hi Laurent,

On 8/26/22 7:06 PM, Laurent Pinchart wrote:
> Hi Umang,
>
> On Fri, Aug 26, 2022 at 01:56:20PM +0530, Umang Jain wrote:
>> On 8/25/22 8:39 PM, Laurent Pinchart wrote:
>>> On Thu, Aug 25, 2022 at 02:05:57AM +0300, Laurent Pinchart via libcamera-devel wrote:
>>>> On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
>>>>> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
>>>>> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
>>>>> Rec 601 Y'CbCr encoding and limited range) in the kernel.
>>>>>
>>>>> The RGB pixel formats should not use any Y'CbCr encoding and is always
>>>>> full range. Adjust the colorspace before reporting back to the
>>>>> userspace in such a situation.
>>>>>
>>>>> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
>>>>> for RGB pixel formats.
>>>> Mentioning sRGB here is a bit confusing I think. Let's focus on what
>>>> this patch does (and this will also help making sure I understood this
>>>> correctly):
>>>>
>>>> V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
>>>> formats, including RGB and raw formats. This causes the libcamera
>>>> ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
>>>> overriding the encoding reported by the kernel to YCbCrEncoding::None
>>>> for non-YUV pixel formats and media bus formats.
>>>>
>>>> Similarly, override the quantization range of non-YUV formats to full
>>>> range, as limited range isn't used for RGB and raw formats.
>>>>
>>>>> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
>>>>> ---
>>>>>    include/libcamera/internal/v4l2_device.h |  5 ++++-
>>>>>    src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
>>>>>    src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
>>>>>    src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
>>>>>    4 files changed, 35 insertions(+), 11 deletions(-)
>>>>>
>>>>> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
>>>>> index a52a5f2c..5ae2ef8a 100644
>>>>> --- a/include/libcamera/internal/v4l2_device.h
>>>>> +++ b/include/libcamera/internal/v4l2_device.h
>>>>> @@ -22,6 +22,8 @@
>>>>>    #include <libcamera/color_space.h>
>>>>>    #include <libcamera/controls.h>
>>>>>    
>>>>> +#include "libcamera/internal/formats.h"
>>>>> +
>>>>>    namespace libcamera {
>>>>>    
>>>>>    class EventNotifier;
>>>>> @@ -59,7 +61,8 @@ protected:
>>>>>    	int fd() const { return fd_.get(); }
>>>>>    
>>>>>    	template<typename T>
>>>>> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
>>>>> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
>>>>> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
>>>> 						      PixelFormatInfo::ColourEncoding colourEncoding);
>>>> Same below.
>>>>
>>>>>    
>>>>>    	template<typename T>
>>>>>    	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
>>>>> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
>>>>> index b22a981f..1fb08b9d 100644
>>>>> --- a/src/libcamera/v4l2_device.cpp
>>>>> +++ b/src/libcamera/v4l2_device.cpp
>>>>> @@ -24,6 +24,7 @@
>>>>>    #include <libcamera/base/log.h>
>>>>>    #include <libcamera/base/utils.h>
>>>>>    
>>>>> +#include "libcamera/internal/formats.h"
>>>>>    #include "libcamera/internal/sysfs.h"
>>>>>    
>>>>>    /**
>>>>> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
>>>>>     * \retval std::nullopt One or more V4L2 color space fields were not recognised
>>>>>     */
>>> The documentation needs to be updated with the new parameter.
>>>
>>>>>    template<typename T>
>>>>> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>>>> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
>>>>> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
>>>>>    {
>>>>>    	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
>>>>>    	if (itColor == v4l2ToColorSpace.end())
>>>>> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>>>>    			return std::nullopt;
>>>>>    
>>>>>    		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
>>>>> +
>>>> I'd add a comment here, or we'll forget why.
>>>>
>>>> 		/*
>>>> 		 * V4L2 has no "none" encoding, override the value returned by
>>>> 		 * the kernel for non-YUV formats as YCbCr encoding isn't
>>>> 		 * applicable in that case..
>>>> 		 */
>>>>
>>>>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
>>>> Shouldn't this be
>>>>
>>>> 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
>>>>
>>>> ? Raw formats have no YCbCr encoding either. The subject line would need
>>>> to be updated accordingly, I'd write "Adjust colorspace based on pixel
>>>> format".
>>>>
>>>>> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
>>>>>    	}
>>>>>    
>>>>>    	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
>>>>> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
>>>>>    			return std::nullopt;
>>>>>    
>>>>>    		colorSpace.range = itRange->second;
>>>>> +
>>>>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
>>>> Same here.
>>>>
>>>>> +			colorSpace.range = ColorSpace::Range::Full;
>>>>>    	}
>>>>>    
>>>>>    	return colorSpace;
>>>>>    }
>>>>>    
>>>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
>>>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
>>>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
>>>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
>>>>> +							    const PixelFormatInfo::ColourEncoding &);
>>>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
>>>>> +							    const PixelFormatInfo::ColourEncoding &);
>>>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
>>>>> +							    const PixelFormatInfo::ColourEncoding &);
>>>>>    
>>>>>    /**
>>>>>     * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
>>>>> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
>>>>> index ab74b9f0..a52c414f 100644
>>>>> --- a/src/libcamera/v4l2_subdevice.cpp
>>>>> +++ b/src/libcamera/v4l2_subdevice.cpp
>>>>> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>>>>    	format->size.width = subdevFmt.format.width;
>>>>>    	format->size.height = subdevFmt.format.height;
>>>>>    	format->mbus_code = subdevFmt.format.code;
>>>>> -	format->colorSpace = toColorSpace(subdevFmt.format);
>>>>> +	auto iter = formatInfoMap.find(format->mbus_code);
>>>>> +	if (iter == formatInfoMap.end())
>>>>> +		return -EINVAL;
>>>> That seems a bit harsh. I'm thinking about the simple pipeline handler
>>>> in particular, which will propagate media bus formats through the
>>>> pipeline without caring much about the format itself. Would it better to
>>>> instead pick a default ?
>>>>
>>>>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>>>>    
>>>>>    	return 0;
>>>>>    }
>>>>> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>>>>    	format->size.width = subdevFmt.format.width;
>>>>>    	format->size.height = subdevFmt.format.height;
>>>>>    	format->mbus_code = subdevFmt.format.code;
>>>>> -	format->colorSpace = toColorSpace(subdevFmt.format);
>>>>> +	auto iter = formatInfoMap.find(format->mbus_code);
>>>>> +	if (iter == formatInfoMap.end())
>>>>> +		return -EINVAL;
>>>> Same here.
>> Esp in V4l2Subdevice::setFormat() , picking a default and reporting that
>> back makes me a bit unconformtable ....
>>
>> should we chose to fail hard here?
> For the reasons explained in another reply in this thread, I'd rather
> add hard failures separately, if we decide it's the way to go. We would
> then need to consider the impact on different parts of libcamera (simple
> pipeline handler, metadata formats, ...). Even an error/warning message
> is something I would introduce in a separate patch on top, to be able to
> test it thoroughly without blocking the rest of the series.

Ok.
>
>> diff --git a/src/libcamera/v4l2_subdevice.cpp
>> b/src/libcamera/v4l2_subdevice.cpp
>> index 70f179c2..d415a738 100644
>> --- a/src/libcamera/v4l2_subdevice.cpp
>> +++ b/src/libcamera/v4l2_subdevice.cpp
>> @@ -502,10 +502,19 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>
>>           format->size.width = subdevFmt.format.width;
>>           format->size.height = subdevFmt.format.height;
>> +
>> +       auto iter = formatInfoMap.find(subdevFmt.format.code);
>> +       if (iter == formatInfoMap.end()) {
>> +               /* Pick a default format to continue */
>> +               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
>> +               iter = formatInfoMap.find(defCode);
> You can avoid the lookup there and just use the RGB encoding:
>
> 	PixelFormatInfo::ColourEncoding colourEncoding;
> 	auto iter = formatInfoMap.find(subdevFmt.format.code);
> 	if (iter != formatInfoMap.end()) {
> 		colourEncoding = iter->second.colourEncoding;
> 	} else {
> 		LOG(V4L2, Warning)
> 			<< "Unknown subdev format "
> 			<< utils::hex(subdevFmt.format.code, 4)
> 			<< ", defaulting to RGB encoding";
>
> 		colourEncoding = PixelFormatInfo::ColourEncodingRGB;
> 	}
>
>> +               LOG(V4L2, Error)
> Make it a warning.
>
>> +                        << "Mapping for subdev format " << subdevFmt.format.code
> Print the code in hex.
>
>> +                        << " is missing, falling back to " << iter->second.name;
>> +
>> +               subdevFmt.format.code = defCode;
> Why do you override the media bus code returned by the driver ?

It wasn't explicit in the previous conversation that we need to pick a 
default colorspace /only/

My thought process went on the lines "we need to pick a default media 
bus format"  (if the returned code mapping is missing..)
>
> All these comments apply below.
>
>> +       }
>>           format->mbus_code = subdevFmt.format.code;
>> -       auto iter = formatInfoMap.find(format->mbus_code);
>> -       if (iter == formatInfoMap.end())
>> -               return -EINVAL;
>>           format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>
>>           return 0;
>> @@ -551,10 +560,20 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
>>
>>           format->size.width = subdevFmt.format.width;
>>           format->size.height = subdevFmt.format.height;
>> +
>> +       auto iter = formatInfoMap.find(subdevFmt.format.code);
>> +       if (iter == formatInfoMap.end()) {
>> +               /* Pick a default format to continue */
>> +               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
>> +               iter = formatInfoMap.find(defCode);
>> +
>> +               LOG(V4L2, Error)
>> +                       << "Mapping for subdev format " << subdevFmt.format.code
>> +                       << " is missing, falling back to " << iter->second.name;
>> +
>> +               subdevFmt.format.code = defCode;
>> +       }
>>           format->mbus_code = subdevFmt.format.code;
>> -       auto iter = formatInfoMap.find(format->mbus_code);
>> -       if (iter == formatInfoMap.end())
>> -               return -EINVAL;
>>           format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>
>>           return 0;
>>
>>>>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
>>>>>    
>>>>>    	return 0;
>>>>>    }
>>>>> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
>>>>> index 5a2d0e5b..0e3f5436 100644
>>>>> --- a/src/libcamera/v4l2_videodevice.cpp
>>>>> +++ b/src/libcamera/v4l2_videodevice.cpp
>>>>> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
>>>> I'm tempted to add a helper just above this function:
>>>>
>>>> namespace {
>>>>
>>>> std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
>>>> {
>>>> 	V4L2PixelFormat fourcc{ pix.pixelformat };
>>>> 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
>>>> }
>>>>
>>>> } /* namespace */
>>>>
>>>> to make the color below a bit nicer (it would actually not need to be
>>>> changed at all). Up to you.
>>>>
>>>>>    	format->size.height = pix->height;
>>>>>    	format->fourcc = V4L2PixelFormat(pix->pixelformat);
>>>>>    	format->planesCount = pix->num_planes;
>>>>> -	format->colorSpace = toColorSpace(*pix);
>>>>> +	format->colorSpace =
>>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>>>    
>>>>>    	for (unsigned int i = 0; i < format->planesCount; ++i) {
>>>>>    		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>>>>> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
>>>>>    		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
>>>>>    		format->planes[i].size = pix->plane_fmt[i].sizeimage;
>>>>>    	}
>>>>> -	format->colorSpace = toColorSpace(*pix);
>>>>> +	format->colorSpace =
>>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>>>    
>>>>>    	return 0;
>>>>>    }
>>>>> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
>>>>>    	format->planesCount = 1;
>>>>>    	format->planes[0].bpl = pix->bytesperline;
>>>>>    	format->planes[0].size = pix->sizeimage;
>>>>> -	format->colorSpace = toColorSpace(*pix);
>>>>> +	format->colorSpace =
>>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>>>    
>>>>>    	return 0;
>>>>>    }
>>>>> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
>>>>>    	format->planesCount = 1;
>>>>>    	format->planes[0].bpl = pix->bytesperline;
>>>>>    	format->planes[0].size = pix->sizeimage;
>>>>> -	format->colorSpace = toColorSpace(*pix);
>>>>> +	format->colorSpace =
>>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
>>>>>    
>>>>>    	return 0;
>>>>>    }

Hi Umang,

On Fri, Aug 26, 2022 at 11:03:03PM +0530, Umang Jain wrote:
> On 8/26/22 7:06 PM, Laurent Pinchart wrote:
> > On Fri, Aug 26, 2022 at 01:56:20PM +0530, Umang Jain wrote:
> >> On 8/25/22 8:39 PM, Laurent Pinchart wrote:
> >>> On Thu, Aug 25, 2022 at 02:05:57AM +0300, Laurent Pinchart via libcamera-devel wrote:
> >>>> On Wed, Aug 24, 2022 at 09:54:21PM +0530, Umang Jain via libcamera-devel wrote:
> >>>>> The V4L2_COLORSPACE_SRGB colorspace maps to ColorSpace::Srgb.
> >>>>> This is wrong as the V4L2_COLORSPACE_SRGB is ill-defined (defines
> >>>>> Rec 601 Y'CbCr encoding and limited range) in the kernel.
> >>>>>
> >>>>> The RGB pixel formats should not use any Y'CbCr encoding and is always
> >>>>> full range. Adjust the colorspace before reporting back to the
> >>>>> userspace in such a situation.
> >>>>>
> >>>>> Moving forwards, the ColorSpace::Srgb will be defined in the true sense
> >>>>> for RGB pixel formats.
> >>>> Mentioning sRGB here is a bit confusing I think. Let's focus on what
> >>>> this patch does (and this will also help making sure I understood this
> >>>> correctly):
> >>>>
> >>>> V4L2 has no "none" YCbCr encoding, and thus reports an encoding for all
> >>>> formats, including RGB and raw formats. This causes the libcamera
> >>>> ColorSpace to report incorrect encodings for non-YUV formats. Fix it by
> >>>> overriding the encoding reported by the kernel to YCbCrEncoding::None
> >>>> for non-YUV pixel formats and media bus formats.
> >>>>
> >>>> Similarly, override the quantization range of non-YUV formats to full
> >>>> range, as limited range isn't used for RGB and raw formats.
> >>>>
> >>>>> Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
> >>>>> ---
> >>>>>    include/libcamera/internal/v4l2_device.h |  5 ++++-
> >>>>>    src/libcamera/v4l2_device.cpp            | 19 +++++++++++++++----
> >>>>>    src/libcamera/v4l2_subdevice.cpp         | 10 ++++++++--
> >>>>>    src/libcamera/v4l2_videodevice.cpp       | 12 ++++++++----
> >>>>>    4 files changed, 35 insertions(+), 11 deletions(-)
> >>>>>
> >>>>> diff --git a/include/libcamera/internal/v4l2_device.h b/include/libcamera/internal/v4l2_device.h
> >>>>> index a52a5f2c..5ae2ef8a 100644
> >>>>> --- a/include/libcamera/internal/v4l2_device.h
> >>>>> +++ b/include/libcamera/internal/v4l2_device.h
> >>>>> @@ -22,6 +22,8 @@
> >>>>>    #include <libcamera/color_space.h>
> >>>>>    #include <libcamera/controls.h>
> >>>>>    
> >>>>> +#include "libcamera/internal/formats.h"
> >>>>> +
> >>>>>    namespace libcamera {
> >>>>>    
> >>>>>    class EventNotifier;
> >>>>> @@ -59,7 +61,8 @@ protected:
> >>>>>    	int fd() const { return fd_.get(); }
> >>>>>    
> >>>>>    	template<typename T>
> >>>>> -	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format);
> >>>>> +	static std::optional<ColorSpace> toColorSpace(const T &v4l2Format,
> >>>>> +						      const PixelFormatInfo::ColourEncoding &colourEncoding);
> >>>> 						      PixelFormatInfo::ColourEncoding colourEncoding);
> >>>> Same below.
> >>>>
> >>>>>    
> >>>>>    	template<typename T>
> >>>>>    	static int fromColorSpace(const std::optional<ColorSpace> &colorSpace, T &v4l2Format);
> >>>>> diff --git a/src/libcamera/v4l2_device.cpp b/src/libcamera/v4l2_device.cpp
> >>>>> index b22a981f..1fb08b9d 100644
> >>>>> --- a/src/libcamera/v4l2_device.cpp
> >>>>> +++ b/src/libcamera/v4l2_device.cpp
> >>>>> @@ -24,6 +24,7 @@
> >>>>>    #include <libcamera/base/log.h>
> >>>>>    #include <libcamera/base/utils.h>
> >>>>>    
> >>>>> +#include "libcamera/internal/formats.h"
> >>>>>    #include "libcamera/internal/sysfs.h"
> >>>>>    
> >>>>>    /**
> >>>>> @@ -816,7 +817,8 @@ static const std::map<ColorSpace::Range, v4l2_quantization> rangeToV4l2 = {
> >>>>>     * \retval std::nullopt One or more V4L2 color space fields were not recognised
> >>>>>     */
> >>> The documentation needs to be updated with the new parameter.
> >>>
> >>>>>    template<typename T>
> >>>>> -std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>>>> +std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format,
> >>>>> +						   const PixelFormatInfo::ColourEncoding &colourEncoding)
> >>>>>    {
> >>>>>    	auto itColor = v4l2ToColorSpace.find(v4l2Format.colorspace);
> >>>>>    	if (itColor == v4l2ToColorSpace.end())
> >>>>> @@ -839,6 +841,9 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>>>>    			return std::nullopt;
> >>>>>    
> >>>>>    		colorSpace.ycbcrEncoding = itYcbcrEncoding->second;
> >>>>> +
> >>>> I'd add a comment here, or we'll forget why.
> >>>>
> >>>> 		/*
> >>>> 		 * V4L2 has no "none" encoding, override the value returned by
> >>>> 		 * the kernel for non-YUV formats as YCbCr encoding isn't
> >>>> 		 * applicable in that case..
> >>>> 		 */
> >>>>
> >>>>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> >>>> Shouldn't this be
> >>>>
> >>>> 		if (colourEncoding != PixelFormatInfo::ColourEncodingYUV)
> >>>>
> >>>> ? Raw formats have no YCbCr encoding either. The subject line would need
> >>>> to be updated accordingly, I'd write "Adjust colorspace based on pixel
> >>>> format".
> >>>>
> >>>>> +			colorSpace.ycbcrEncoding = ColorSpace::YcbcrEncoding::None;
> >>>>>    	}
> >>>>>    
> >>>>>    	if (v4l2Format.quantization != V4L2_QUANTIZATION_DEFAULT) {
> >>>>> @@ -847,14 +852,20 @@ std::optional<ColorSpace> V4L2Device::toColorSpace(const T &v4l2Format)
> >>>>>    			return std::nullopt;
> >>>>>    
> >>>>>    		colorSpace.range = itRange->second;
> >>>>> +
> >>>>> +		if (colourEncoding == PixelFormatInfo::ColourEncodingRGB)
> >>>> Same here.
> >>>>
> >>>>> +			colorSpace.range = ColorSpace::Range::Full;
> >>>>>    	}
> >>>>>    
> >>>>>    	return colorSpace;
> >>>>>    }
> >>>>>    
> >>>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &);
> >>>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &);
> >>>>> -template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &);
> >>>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format &,
> >>>>> +							    const PixelFormatInfo::ColourEncoding &);
> >>>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_pix_format_mplane &,
> >>>>> +							    const PixelFormatInfo::ColourEncoding &);
> >>>>> +template std::optional<ColorSpace> V4L2Device::toColorSpace(const struct v4l2_mbus_framefmt &,
> >>>>> +							    const PixelFormatInfo::ColourEncoding &);
> >>>>>    
> >>>>>    /**
> >>>>>     * \brief Fill in the color space fields of a V4L2 format from a ColorSpace
> >>>>> diff --git a/src/libcamera/v4l2_subdevice.cpp b/src/libcamera/v4l2_subdevice.cpp
> >>>>> index ab74b9f0..a52c414f 100644
> >>>>> --- a/src/libcamera/v4l2_subdevice.cpp
> >>>>> +++ b/src/libcamera/v4l2_subdevice.cpp
> >>>>> @@ -502,7 +502,10 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>>>>    	format->size.width = subdevFmt.format.width;
> >>>>>    	format->size.height = subdevFmt.format.height;
> >>>>>    	format->mbus_code = subdevFmt.format.code;
> >>>>> -	format->colorSpace = toColorSpace(subdevFmt.format);
> >>>>> +	auto iter = formatInfoMap.find(format->mbus_code);
> >>>>> +	if (iter == formatInfoMap.end())
> >>>>> +		return -EINVAL;
> >>>> That seems a bit harsh. I'm thinking about the simple pipeline handler
> >>>> in particular, which will propagate media bus formats through the
> >>>> pipeline without caring much about the format itself. Would it better to
> >>>> instead pick a default ?
> >>>>
> >>>>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>>>>    
> >>>>>    	return 0;
> >>>>>    }
> >>>>> @@ -548,7 +551,10 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>>>>    	format->size.width = subdevFmt.format.width;
> >>>>>    	format->size.height = subdevFmt.format.height;
> >>>>>    	format->mbus_code = subdevFmt.format.code;
> >>>>> -	format->colorSpace = toColorSpace(subdevFmt.format);
> >>>>> +	auto iter = formatInfoMap.find(format->mbus_code);
> >>>>> +	if (iter == formatInfoMap.end())
> >>>>> +		return -EINVAL;
> >>>> Same here.
> >> Esp in V4l2Subdevice::setFormat() , picking a default and reporting that
> >> back makes me a bit unconformtable ....
> >>
> >> should we chose to fail hard here?
> > For the reasons explained in another reply in this thread, I'd rather
> > add hard failures separately, if we decide it's the way to go. We would
> > then need to consider the impact on different parts of libcamera (simple
> > pipeline handler, metadata formats, ...). Even an error/warning message
> > is something I would introduce in a separate patch on top, to be able to
> > test it thoroughly without blocking the rest of the series.
> 
> Ok.
> >
> >> diff --git a/src/libcamera/v4l2_subdevice.cpp
> >> b/src/libcamera/v4l2_subdevice.cpp
> >> index 70f179c2..d415a738 100644
> >> --- a/src/libcamera/v4l2_subdevice.cpp
> >> +++ b/src/libcamera/v4l2_subdevice.cpp
> >> @@ -502,10 +502,19 @@ int V4L2Subdevice::getFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>
> >>           format->size.width = subdevFmt.format.width;
> >>           format->size.height = subdevFmt.format.height;
> >> +
> >> +       auto iter = formatInfoMap.find(subdevFmt.format.code);
> >> +       if (iter == formatInfoMap.end()) {
> >> +               /* Pick a default format to continue */
> >> +               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
> >> +               iter = formatInfoMap.find(defCode);
> > You can avoid the lookup there and just use the RGB encoding:
> >
> > 	PixelFormatInfo::ColourEncoding colourEncoding;
> > 	auto iter = formatInfoMap.find(subdevFmt.format.code);
> > 	if (iter != formatInfoMap.end()) {
> > 		colourEncoding = iter->second.colourEncoding;
> > 	} else {
> > 		LOG(V4L2, Warning)
> > 			<< "Unknown subdev format "
> > 			<< utils::hex(subdevFmt.format.code, 4)
> > 			<< ", defaulting to RGB encoding";
> >
> > 		colourEncoding = PixelFormatInfo::ColourEncodingRGB;
> > 	}
> >
> >> +               LOG(V4L2, Error)
> > Make it a warning.
> >
> >> +                        << "Mapping for subdev format " << subdevFmt.format.code
> > Print the code in hex.
> >
> >> +                        << " is missing, falling back to " << iter->second.name;
> >> +
> >> +               subdevFmt.format.code = defCode;
> > Why do you override the media bus code returned by the driver ?
> 
> It wasn't explicit in the previous conversation that we need to pick a 
> default colorspace /only/
> 
> My thought process went on the lines "we need to pick a default media 
> bus format"  (if the returned code mapping is missing..)

I should have been clearer, sorry. A default color space should be fine
I think, because we need to do something, and we can't do better, but
overriding an unknown format will lead to possible problems.

> > All these comments apply below.
> >
> >> +       }
> >>           format->mbus_code = subdevFmt.format.code;
> >> -       auto iter = formatInfoMap.find(format->mbus_code);
> >> -       if (iter == formatInfoMap.end())
> >> -               return -EINVAL;
> >>           format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>
> >>           return 0;
> >> @@ -551,10 +560,20 @@ int V4L2Subdevice::setFormat(unsigned int pad, V4L2SubdeviceFormat *format,
> >>
> >>           format->size.width = subdevFmt.format.width;
> >>           format->size.height = subdevFmt.format.height;
> >> +
> >> +       auto iter = formatInfoMap.find(subdevFmt.format.code);
> >> +       if (iter == formatInfoMap.end()) {
> >> +               /* Pick a default format to continue */
> >> +               uint32_t defCode = MEDIA_BUS_FMT_RGB888_1X24;
> >> +               iter = formatInfoMap.find(defCode);
> >> +
> >> +               LOG(V4L2, Error)
> >> +                       << "Mapping for subdev format " << subdevFmt.format.code
> >> +                       << " is missing, falling back to " << iter->second.name;
> >> +
> >> +               subdevFmt.format.code = defCode;
> >> +       }
> >>           format->mbus_code = subdevFmt.format.code;
> >> -       auto iter = formatInfoMap.find(format->mbus_code);
> >> -       if (iter == formatInfoMap.end())
> >> -               return -EINVAL;
> >>           format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>
> >>           return 0;
> >>
> >>>>> +	format->colorSpace = toColorSpace(subdevFmt.format, iter->second.colourEncoding);
> >>>>>    
> >>>>>    	return 0;
> >>>>>    }
> >>>>> diff --git a/src/libcamera/v4l2_videodevice.cpp b/src/libcamera/v4l2_videodevice.cpp
> >>>>> index 5a2d0e5b..0e3f5436 100644
> >>>>> --- a/src/libcamera/v4l2_videodevice.cpp
> >>>>> +++ b/src/libcamera/v4l2_videodevice.cpp
> >>>>> @@ -931,7 +931,8 @@ int V4L2VideoDevice::getFormatMultiplane(V4L2DeviceFormat *format)
> >>>> I'm tempted to add a helper just above this function:
> >>>>
> >>>> namespace {
> >>>>
> >>>> std::optional<ColorSpace> toColorSpace(const struct v4l2_pix_format_mplane &pix)
> >>>> {
> >>>> 	V4L2PixelFormat fourcc{ pix.pixelformat };
> >>>> 	return toColorSpace(pix, PixelFormatInfo::info(fourcc).colourEncoding);
> >>>> }
> >>>>
> >>>> } /* namespace */
> >>>>
> >>>> to make the color below a bit nicer (it would actually not need to be
> >>>> changed at all). Up to you.
> >>>>
> >>>>>    	format->size.height = pix->height;
> >>>>>    	format->fourcc = V4L2PixelFormat(pix->pixelformat);
> >>>>>    	format->planesCount = pix->num_planes;
> >>>>> -	format->colorSpace = toColorSpace(*pix);
> >>>>> +	format->colorSpace =
> >>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>>>    
> >>>>>    	for (unsigned int i = 0; i < format->planesCount; ++i) {
> >>>>>    		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >>>>> @@ -987,7 +988,8 @@ int V4L2VideoDevice::trySetFormatMultiplane(V4L2DeviceFormat *format, bool set)
> >>>>>    		format->planes[i].bpl = pix->plane_fmt[i].bytesperline;
> >>>>>    		format->planes[i].size = pix->plane_fmt[i].sizeimage;
> >>>>>    	}
> >>>>> -	format->colorSpace = toColorSpace(*pix);
> >>>>> +	format->colorSpace =
> >>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>>>    
> >>>>>    	return 0;
> >>>>>    }
> >>>>> @@ -1011,7 +1013,8 @@ int V4L2VideoDevice::getFormatSingleplane(V4L2DeviceFormat *format)
> >>>>>    	format->planesCount = 1;
> >>>>>    	format->planes[0].bpl = pix->bytesperline;
> >>>>>    	format->planes[0].size = pix->sizeimage;
> >>>>> -	format->colorSpace = toColorSpace(*pix);
> >>>>> +	format->colorSpace =
> >>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>>>    
> >>>>>    	return 0;
> >>>>>    }
> >>>>> @@ -1053,7 +1056,8 @@ int V4L2VideoDevice::trySetFormatSingleplane(V4L2DeviceFormat *format, bool set)
> >>>>>    	format->planesCount = 1;
> >>>>>    	format->planes[0].bpl = pix->bytesperline;
> >>>>>    	format->planes[0].size = pix->sizeimage;
> >>>>> -	format->colorSpace = toColorSpace(*pix);
> >>>>> +	format->colorSpace =
> >>>>> +		toColorSpace(*pix, PixelFormatInfo::info(format->fourcc).colourEncoding);
> >>>>>    
> >>>>>    	return 0;
> >>>>>    }