Message-ID: <aYIb8ZIZDfCJZEZ9@zed>
Date: Tue, 3 Feb 2026 17:15:44 +0100
From: Jacopo Mondi <jacopo.mondi@...asonboard.com>
To: Antoine Bouyer <antoine.bouyer@....com>
Cc: julien.vuillaumier@....com, alexi.birlinger@....com, 
	daniel.baluta@....com, peng.fan@....com, frank.li@....com, 
	jacopo.mondi@...asonboard.com, laurent.pinchart@...asonboard.com, mchehab@...nel.org, 
	robh@...nel.org, krzk+dt@...nel.org, conor+dt@...nel.org, shawnguo@...nel.org, 
	s.hauer@...gutronix.de, kernel@...gutronix.de, festevam@...il.com, 
	linux-kernel@...r.kernel.org, linux-media@...r.kernel.org, devicetree@...r.kernel.org, 
	linux-arm-kernel@...ts.infradead.org
Subject: Re: [RFC v1 01/11] media: uapi: v4l2-isp: Add v4l2 ISP extensible
 statistics definitions

Hi Antoine
  thanks a lot for extendable stats

On Fri, Jan 23, 2026 at 09:09:28AM +0100, Antoine Bouyer wrote:
> Extend the v4l2-isp extensible format introduced for isp parameters buffer
> to the statistics buffer as well.
>
> Like for ISP configuration purpose, that will help supporting various ISP
> hardware versions reporting different statistics data with less impact on
> userspace.
>
> The `v4l2_isp_stats_buffer` reuses the `v4l2_isp_params_buffer` container
> definitions, with similar header, versions and flags. V0 and V1 versions

Why do you need two flags ?

Params had to introduce two because we had two drivers already
mainlined using the pre-v4l2-isp version of extensible params which
had defined their version identifier as 1 and 0 and we didn't want to
break existing userspace using those identifiers. So we had to accept
both V0 and V1 as "first version of the v4l2-isp extensible parameters
format".

For stats we don't have users, so I guess we can start with V1 == 0 ?

> are provided to match with params versions. On the other side, ENABLE and
> DISABLE flags are not really meaningfull for statistics purpose. So VALID
> and INVALID flags are introduced. Purpose is to force ISP driver to
> validate a statistics buffer, before it is consumed by userspace.

Interesting. What do you mean with "validate a statistics buffer" ?
And if a driver has to do validation, why would it send upstream a
non-validated buffer ?

>
> Signed-off-by: Antoine Bouyer <antoine.bouyer@....com>
> ---
>  include/uapi/linux/media/v4l2-isp.h | 85 +++++++++++++++++++++++++++++
>  1 file changed, 85 insertions(+)
>
> diff --git a/include/uapi/linux/media/v4l2-isp.h b/include/uapi/linux/media/v4l2-isp.h
> index 779168f9058e..ed1279b86694 100644
> --- a/include/uapi/linux/media/v4l2-isp.h
> +++ b/include/uapi/linux/media/v4l2-isp.h
> @@ -99,4 +99,89 @@ struct v4l2_isp_params_buffer {
>  	__u8 data[] __counted_by(data_size);
>  };
>
> +/**
> + * enum v4l2_isp_stats_version - V4L2 ISP statistics versioning
> + *
> + * @V4L2_ISP_STATS_VERSION_V0: First version of the V4L2 ISP statistics format
> + *			       (for compatibility)
> + * @V4L2_ISP_STATS_VERSION_V1: First version of the V4L2 ISP statistics format
> + *
> + * V0 and V1 are identical, and comply with V4l2 ISP parameters versions. So
> + * both V0 and V1 refers to the first version of the V4L2 ISP statistics
> + * format.
> + *
> + * Future revisions of the V4L2 ISP statistics format should start from the
> + * value of 2.
> + */
> +enum v4l2_isp_stats_version {
> +	V4L2_ISP_STATS_VERSION_V0 = 0,
> +	V4L2_ISP_STATS_VERSION_V1,

As suggested I would make V1 == 0

> +};
> +
> +#define V4L2_ISP_PARAMS_FL_BLOCK_VALID		(1U << 0)
> +#define V4L2_ISP_PARAMS_FL_BLOCK_INVALID	(1U << 1)
> +
> +/*
> + * Reserve the first 8 bits for V4L2_ISP_STATS_FL_* flag.
> + *
> + * Driver-specific flags should be defined as:
> + * #define DRIVER_SPECIFIC_FLAG0     ((1U << V4L2_ISP_STATS_FL_DRIVER_FLAGS(0))
> + * #define DRIVER_SPECIFIC_FLAG1     ((1U << V4L2_ISP_STATS_FL_DRIVER_FLAGS(1))
> + */
> +#define V4L2_ISP_STATS_FL_DRIVER_FLAGS(n)       ((n) + 8)

Currently we have no users of V4L2_ISP_PARAMS_FL_DRIVER_FLAGS so we
could even consider making it a V4L2_ISP_FL_DRIVER_FLAGS

Or do you think it is worth creating a new symbol ?

> +
> +/**
> + * struct v4l2_isp_stats_block_header - V4L2 extensible statistics block header
> + * @type: The statistics block type (driver-specific)
> + * @flags: A bitmask of block flags (driver-specific)
> + * @size: Size (in bytes) of the statistics block, including this header
> + *
> + * This structure represents the common part of all the ISP statistics blocks.
> + * Each statistics block shall embed an instance of this structure type as its
> + * first member, followed by the block-specific statistics data.
> + *
> + * The @type field is an ISP driver-specific value that identifies the block
> + * type. The @size field specifies the size of the parameters block.
> + *
> + * The @flags field is a bitmask of per-block flags V4L2_STATS_ISP_FL_* and
> + * driver-specific flags specified by the driver header.
> + */
> +struct v4l2_isp_stats_block_header {
> +	__u16 type;
> +	__u16 flags;
> +	__u32 size;
> +} __attribute__((aligned(8)));
> +

This is currently identical to v4l2_isp_params_block_header.

Can we create a single header for both stats and params and provide a

#define v4l2_isp_params_block_header v4l2_isp_block_header

for maintaining compatibility with existing users ?

Or do you expect stats and params to eventually need different headers ?

> +/**
> + * struct v4l2_isp_stats_buffer - V4L2 extensible statistics data
> + * @version: The statistics buffer version (driver-specific)
> + * @data_size: The statistics data effective size, excluding this header
> + * @data: The statistics data
> + *
> + * This structure contains the statistics information of the ISP hardware,
> + * serialized for userspace into a data buffer. Each statistics block is
> + * represented by a block-specific structure which contains a
> + * :c:type:`v4l2_isp_stats_block_header` entry as first member. Driver
> + * populates the @data buffer with statistics information of the ISP blocks it
> + * intends to share to userspace. As a consequence, the data buffer effective
> + * size changes according to the number of ISP blocks that driver intends to
> + * provide and is set by the driver in the @data_size field.
> + *
> + * The statistics buffer is versioned by the @version field to allow modifying
> + * and extending its definition. Driver shall populate the @version field to
> + * inform the userpsace about the version it intends to use. The userspace will
> + * parse and handle the @data buffer according to the data layout specific to
> + * the indicated version.
> + *
> + * For each ISP block that driver wants to report, a block-specific structure
> + * is appended to the @data buffer, one after the other without gaps in
> + * between. Driver shall populate the @data_size field with the effective
> + * size, in bytes, of the @data buffer.
> + */
> +struct v4l2_isp_stats_buffer {
> +	__u32 version;
> +	__u32 data_size;
> +	__u8 data[] __counted_by(data_size);
> +};
> +

Same question. Should we introduce a struct v4l2_isp_buffer ?

Thanks!

>  #endif /* _UAPI_V4L2_ISP_H_ */
> --
> 2.52.0
>
>

Powered by blists - more mailing lists