All of lore.kernel.org
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Jiaqi Shi" <shijiaqi_develop@163.com>, imx@lists.linux.dev
Cc: Frank.Li@kernel.org, media-ci@linuxtv.org
Subject: Re: [PATCH 1/2] media: imx: remove legacy video node control and format handling
Date: Sat, 21 Nov 2026 17:26:10 +0000	[thread overview]
Message-ID: <sashiko-outbox-227206@kernel.org> (raw)
In-Reply-To: <20261121170232.414316-2-shijiaqi_develop@163.com>

Thank you for your contribution! Sashiko AI review found 6 potential
issues to consider:
- [High] media: imx: broken pixel format validation in
  capture_find_format()
- [High] media: imx: 32-bit planar sizeimage overflow in
  __capture_try_fmt()
- [Medium] media: imx: __capture_try_fmt() fails with -EINVAL on large
  dimensions
- [Medium] media: imx: unsanitized field handling in __capture_try_fmt()
- [Medium] media: imx: wrong default RGB quantization in
  __capture_try_fmt()
- [Medium] media: imx: callers of imx_media_capture_device_init() set
  legacy_api
--

commit 110cb63254b68ca17643e4ce4ba642ee6210e734
Author: Jiaqi Shi <shijiaqi_develop@163.com>

media: imx: remove legacy video node control and format handling

This commit removes control inheritance from pipeline subdevices to capture
video nodes and switches capture format handling to source-independent format
ioctls initialized to a default UYVY 640x480 format.

> diff --git a/drivers/staging/media/imx/imx-ic-prpencvf.c b/drivers/staging/media/imx/imx-ic-prpencvf.c
> index 3430ddeac3569..372d25c791d35 100644
> --- a/drivers/staging/media/imx/imx-ic-prpencvf.c
> +++ b/drivers/staging/media/imx/imx-ic-prpencvf.c
> @@ -1280,13 +1280,13 @@ static int prp_registered(struct v4l2_subdev *sd)
>  
>  	priv->vdev = imx_media_capture_device_init(ic_priv->ipu_dev,
>  						   &ic_priv->sd,
> -						   PRPENCVF_SRC_PAD, true);
> +						   PRPENCVF_SRC_PAD, true, true);

[Severity: Medium]
Should prp_registered() and csi_registered() pass legacy_api = false (or
remove the legacy_api parameter from imx_media_capture_device_init()) now
that legacy format and control handling is being removed?

Because both callers still pass legacy_api = true, imx_media_capture_device_init()
never sets V4L2_CAP_IO_MC on any capture video node, which causes
v4l_enum_fmt() in v4l2-core to zero f->mbus_code before calling
capture_enum_fmt_vid_cap().  It also leaves capture_legacy_ioctl_ops active
(while capture_ioctl_ops is unused), so VIDIOC_ENUM_FRAMESIZES still calls
capture_legacy_enum_framesizes(), which queries the source subdev's active
format and fails with -EINVAL on IC capture nodes for any pixel format whose
cc->codes[0] is not an IPU-internal media bus code.

[ ... ]

> diff --git a/drivers/staging/media/imx/imx-media-capture.c b/drivers/staging/media/imx/imx-media-capture.c
> index 9216a221b72f4..adc524038ed81 100644
> --- a/drivers/staging/media/imx/imx-media-capture.c
> +++ b/drivers/staging/media/imx/imx-media-capture.c

[ ... ]

> @@ -74,7 +70,11 @@ static const struct imx_media_pixfmt *capture_find_format(u32 code, u32 fourcc)
>  		return cc;
>  	}
>  
> -	return imx_media_find_mbus_format(code, PIXFMT_SEL_ANY);
> +	cc = imx_media_find_mbus_format(code, PIXFMT_SEL_ANY);
> +	if (cc && fourcc && cc->fourcc != fourcc)
> +		return NULL;
> +
> +	return cc;
>  }

[Severity: High]
Does capture_find_format() misvalidate both passthrough and IPU-internal
formats when called from capture_validate_fmt() at stream start?

In the non-IPU branch above, imx_media_find_mbus_format(code, PIXFMT_SEL_ANY)
returns only the first entry in pixel_formats[] matching code:

drivers/staging/media/imx/imx-media-utils.c:pixel_formats[] {
    ...
	}, {
		.fourcc = V4L2_PIX_FMT_GREY,
		.codes = IMX_BUS_FMTS(
			MEDIA_BUS_FMT_Y8_1X8,
			MEDIA_BUS_FMT_Y10_1X10,
			MEDIA_BUS_FMT_Y12_1X12
		),
		.cs     = IPUV3_COLORSPACE_RGB,
		.bpp    = 8,
		.bayer  = true,
	}, {
		.fourcc = V4L2_PIX_FMT_Y10,
		.codes = IMX_BUS_FMTS(MEDIA_BUS_FMT_Y10_1X10),
		.cs     = IPUV3_COLORSPACE_RGB,
		.bpp    = 16,
		.bayer  = true,
    ...
}

Because V4L2_PIX_FMT_GREY precedes V4L2_PIX_FMT_Y10 and V4L2_PIX_FMT_Y12 and
lists MEDIA_BUS_FMT_Y10_1X10 and MEDIA_BUS_FMT_Y12_1X12,
imx_media_find_mbus_format() returns V4L2_PIX_FMT_GREY for those codes.
Checking cc->fourcc != fourcc then rejects valid V4L2_PIX_FMT_Y10 and
V4L2_PIX_FMT_Y12 capture with -EPIPE while accepting 8-bpp V4L2_PIX_FMT_GREY
buffers for 16-bit passthrough DMA.

In the IPU branch of capture_find_format(), when fourcc is not found in
fmt_sel, it falls back to the first format in fmt_sel instead of returning
NULL:

drivers/staging/media/imx/imx-media-capture.c:capture_find_format() {
	cc = imx_media_find_ipu_format(code, PIXFMT_SEL_YUV_RGB);
	if (cc) {
		enum imx_pixfmt_sel fmt_sel = cc->cs == IPUV3_COLORSPACE_YUV
					    ? PIXFMT_SEL_YUV : PIXFMT_SEL_RGB;

		cc = imx_media_find_pixel_format(fourcc, fmt_sel);
		if (!cc) {
			imx_media_enum_pixel_formats(&fourcc, 0, fmt_sel, 0);
			cc = imx_media_find_pixel_format(fourcc, fmt_sel);
		}

		return cc;
	}
    ...
}

Because all Bayer and greyscale entries in pixel_formats[] have
.cs = IPUV3_COLORSPACE_RGB, configuring a Bayer or greyscale pixel format on
a video node fed by MEDIA_BUS_FMT_ARGB8888_1X32 falls back to a
PIXFMT_SEL_RGB descriptor with cc->cs == IPUV3_COLORSPACE_RGB, passes
priv->vdev.cc->cs != cc->cs in capture_validate_fmt(), and reaches
ipu_cpmem_set_image(), triggering WARN_ON(1) for V4L2_PIX_FMT_Y10/Y12 or
leaving the CPMEM channel at 32 bpp after an unchecked ipu_cpmem_set_fmt()
-EINVAL error and overflowing 8-bpp or 16-bpp DMA buffers.

> @@ -136,10 +136,12 @@ static int capture_g_fmt_vid_cap(struct file *file, void *fh,
>  }
>  
>  static const struct imx_media_pixfmt *
> -__capture_try_fmt(struct v4l2_pix_format *pixfmt, struct v4l2_rect *compose)
> +__capture_try_fmt(struct capture_priv *priv, struct v4l2_pix_format *pixfmt,
> +		  struct v4l2_rect *compose)
>  {
>  	struct v4l2_mbus_framefmt fmt_src;
>  	const struct imx_media_pixfmt *cc;
> +	u64 sizeimage;
>  
>  	/*
>  	 * Find the pixel format, default to the first supported format if not
> @@ -167,9 +169,21 @@ __capture_try_fmt(struct v4l2_pix_format *pixfmt, struct v4l2_rect *compose)
>  		}
>  	}

[Severity: Medium]
Now that all capture nodes use __capture_try_fmt(), does this leave
pixfmt->field unsanitized and unvalidated against the source subdev?

In __capture_try_fmt(), the V4L2_FIELD_SEQ_TB and V4L2_FIELD_SEQ_BT cases sit
inside if (V4L2_FIELD_IS_INTERLACED(pixfmt->field)), which is false for
sequential fields, making that switch unreachable.  No other check sanitizes
pixfmt->field, so V4L2_FIELD_ANY or invalid field values are returned as-is
from VIDIOC_TRY_FMT and VIDIOC_S_FMT, and capture_validate_fmt() does not
verify priv->vdev.fmt.field against fmt_src.format.field before configuring
IDMAC interweaving.

> -	v4l2_fill_mbus_format(&fmt_src, pixfmt, 0);
> +	/* Clamp dimensions to prevent zero-sized buffers. */
> +	pixfmt->width = clamp(pixfmt->width, 1, 65535);
> +	pixfmt->height = clamp(pixfmt->height, 1, 65535);
> +
> +	v4l2_fill_mbus_format(&fmt_src, pixfmt,
> +			      cc->codes ? cc->codes[0] : 0);
> +	imx_media_try_colorimetry(&fmt_src, priv->ic_route);

[Severity: Medium]
Does passing cc->codes ? cc->codes[0] : 0 give in-memory-only RGB formats the
wrong default quantization?

V4L2_PIX_FMT_BGR24, V4L2_PIX_FMT_XBGR32, V4L2_PIX_FMT_BGRX32 and
V4L2_PIX_FMT_RGBX32 have .codes = NULL in pixel_formats[].  When fmt_src.code
is 0, imx_media_try_colorimetry() cannot find a format descriptor, leaves
is_rgb = false, and maps V4L2_QUANTIZATION_DEFAULT to
V4L2_QUANTIZATION_LIM_RANGE instead of V4L2_QUANTIZATION_FULL_RANGE.

>  	imx_media_mbus_fmt_to_pix_fmt(pixfmt, &fmt_src, cc);
>  
> +	sizeimage = (u64)pixfmt->bytesperline * pixfmt->height;
> +	if (cc->planar)
> +		sizeimage = sizeimage * cc->bpp / 8;
> +	if (sizeimage > U32_MAX)
> +		return NULL;

[Severity: High]
Can pixfmt->sizeimage wrap to 0 for planar formats even when the 64-bit
sizeimage check here succeeds?

imx_media_mbus_fmt_to_pix_fmt() computes pix->sizeimage in 32-bit arithmetic
before shifting right by 3:

drivers/staging/media/imx/imx-media-utils.c:imx_media_mbus_fmt_to_pix_fmt() {
    ...
	pix->sizeimage = cc->planar ? ((stride * pix->height * cc->bpp) >> 3) :
			 stride * pix->height;
}

For a 16-bpp planar format such as V4L2_PIX_FMT_NV16 or V4L2_PIX_FMT_YUV422P
at 16384x16384, stride * pix->height * cc->bpp is 2^32, so pixfmt->sizeimage
wraps to 0 in u32 while the 64-bit sizeimage in __capture_try_fmt() is
512 MiB (<= U32_MAX) and is never written back to pixfmt->sizeimage.  A
subsequent VIDIOC_REQBUFS or VIDIOC_CREATE_BUFS then divides by zero in
capture_queue_setup() at VID_MEM_LIMIT / pix->sizeimage.

> @@ -181,7 +195,11 @@ __capture_try_fmt(struct v4l2_pix_format *pixfmt, struct v4l2_rect *compose)
>  static int capture_try_fmt_vid_cap(struct file *file, void *fh,
>  				   struct v4l2_format *f)
>  {
> -	__capture_try_fmt(&f->fmt.pix, NULL);
> +	struct capture_priv *priv = video_drvdata(file);
> +
> +	if (!__capture_try_fmt(priv, &f->fmt.pix, NULL))
> +		return -EINVAL;
> +
>  	return 0;
>  }

[Severity: Medium]
Should VIDIOC_TRY_FMT and VIDIOC_S_FMT clamp width and height to dimensions
that fit within sizeimage rather than returning -EINVAL when sizeimage
exceeds U32_MAX?

capture_enum_framesizes() advertises dimensions up to 65535x65535, and
__capture_try_fmt() clamps width and height to 65535 (rounded up to 65536 in
imx_media_mbus_fmt_to_pix_fmt()).  For any format with bpp >= 12 (including
the default UYVY), requesting 65535x65535 or probing maximum dimensions with
~0U x ~0U yields sizeimage > U32_MAX and causes VIDIOC_TRY_FMT and
VIDIOC_S_FMT to fail with -EINVAL instead of adjusting the dimensions into a
valid range.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261121170232.414316-1-shijiaqi_develop@163.com?part=1

  reply	other threads:[~2026-11-21 17:26 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-21 17:02 [PATCH v2 0/2] media: imx: staging cleanup and FIM export to v4l2-core Jiaqi Shi
2026-11-21 17:02 ` [PATCH 1/2] media: imx: remove legacy video node control and format handling Jiaqi Shi
2026-11-21 17:26   ` sashiko-bot [this message]
2026-11-21 17:02 ` [PATCH 2/2] media: v4l2-core: add frame interval monitor helper Jiaqi Shi
2026-11-21 17:13   ` sashiko-bot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=sashiko-outbox-227206@kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=Frank.Li@kernel.org \
    --cc=imx@lists.linux.dev \
    --cc=media-ci@linuxtv.org \
    --cc=sashiko-reviews@lists.linux.dev \
    --cc=shijiaqi_develop@163.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.