Re: [PATCH v3] media: i2c: ov13858: add horizontal and vertical flip controls

[Sakari Ailus <sakari.ailus@linux.intel.com>]

Hi Sergey,

On Sat, Sep 26, 2026 at 08:02:48AM +0000, Sergey Lebedev wrote:
> The driver programs OV13858_REG_FORMAT1 (0x3820) from its mode tables and
> never exposes the readout direction, so a module mounted rotated cannot be
> corrected.
> 
> The Microsoft Surface Pro 11 for Business (Intel) mounts this sensor upside
> down, and ipu-bridge now says so - commit b238116ccd4b ("media: ipu-bridge:
> Add upside-down quirk for Surface Pro 11"). libcamera reads that rotation
> and tries to compensate with sensor flips, finds neither control, and falls
> back to Rot0, so with the quirk alone the picture stays upside down in any
> application that does not rotate it itself.
> 
> In 0x3820, BIT(4) set flips vertically and BIT(3) cleared mirrors - the
> mirror is active low, which is ordinary for the family. There is no public
> datasheet, so the bits were found by experiment: every bit of 0x3820
> through 0x3823 written singly mid-stream, and exactly two move the image.
> Two existing drivers agree. Intel's out-of-tree ov13858 declares 0x3820 as
> a bitfield with "hflip: 0 enable, 1 disable" at BIT(3) and "vflip: 0
> disable, 1 enable" at BIT(4), and writes !ctrl->val for the mirror;
> ov13b10 clears BIT(3) and sets BIT(4) of the same register in tree.
> 
> Both controls were checked by eye, one static scene in all four states:
> unflipped the image is upside down, vflip alone stands it up, hflip alone
> mirrors it, and the two together give the view this machine needs. The
> Bayer order at the output is the same in all four, so unlike imx219 and
> imx258, whose flips select a different media bus code, these controls do
> not need V4L2_CTRL_FLAG_MODIFY_LAYOUT.
> 
> Every mode table already sets the mirror bit, so the defaults write back
> what the mode list just wrote. __v4l2_ctrl_handler_setup() runs after that
> list and before MODE_SELECT, so the read-modify-write here sees the value
> the mode just programmed.
> 
> The two are clustered, so a combined S_EXT_CTRLS costs one register
> read-modify-write rather than two, and grabbed while streaming: nothing
> here documents a mid-stream flip as safe, and imx290 and imx219 grab
> theirs because on those parts it is not.
> 
> Signed-off-by: Sergey Lebedev <lsa.uz@pm.me>
> Tested-by: German Pablo Lindo <germanpapulindez@gmail.com>

Is Documentation/process/coding-assistants.rst relevant for the patch?

Please clean up the above commit message.

> ---
> v3, commit message only:
> 
>  - Cites b238116ccd4b in the "commit <sha> (...)" form checkpatch asks
>    for; the Media CI flagged v2 for it.
>  - v2 said the quirk on its own "names a rotation nothing can undo".
>    Snapshot and Firefox do undo it - they rotate the picture themselves -
>    and qcam, which does not, shows it upside down. The sentence now says
>    that.
>  - Tested-by from German Pablo Lindo, on a second Surface Pro 11: qcam
>    upright with this patch, Snapshot and Firefox unchanged.
> 
> The code is v2's, byte for byte. Sakari's set_selection() point is
> answered in v2's notes, and nothing here changes it.
> 
> v2: https://lore.kernel.org/all/20260921124444.79396-1-lsa.uz@pm.me/
> v1: https://lore.kernel.org/all/20260921082609.30830-1-lsa.uz@pm.me/
> 
>  drivers/media/i2c/ov13858.c | 52 +++++++++++++++++++++++++++++++++++++
>  1 file changed, 52 insertions(+)
> 
> diff --git a/drivers/media/i2c/ov13858.c b/drivers/media/i2c/ov13858.c
> index de2b79a9a0e..d870d2edb4f 100644
> --- a/drivers/media/i2c/ov13858.c
> +++ b/drivers/media/i2c/ov13858.c
> @@ -76,6 +76,14 @@
>  #define OV13858_DGTL_GAIN_DEFAULT	1024	/* Default gain = 1 X */
>  #define OV13858_DGTL_GAIN_STEP		1	/* Each step = 1/1024 */
>  
> +/*
> + * Readout direction. The mirror bit is active low, and the value every mode
> + * table programs already has it set.
> + */

This only applies to horizontal flipping.

> +#define OV13858_REG_FORMAT1		0x3820
> +#define OV13858_FORMAT1_VFLIP		BIT(4)
> +#define OV13858_FORMAT1_HFLIP_N		BIT(3)
> +
>  /* Test Pattern Control */
>  #define OV13858_REG_TEST_PATTERN	0x4503
>  #define OV13858_TEST_PATTERN_ENABLE	BIT(7)
> @@ -1042,6 +1050,8 @@ struct ov13858 {
>  	struct v4l2_ctrl *vblank;
>  	struct v4l2_ctrl *hblank;
>  	struct v4l2_ctrl *exposure;
> +	struct v4l2_ctrl *hflip;
> +	struct v4l2_ctrl *vflip;
>  
>  	/* Current mode */
>  	const struct ov13858_mode *cur_mode;
> @@ -1208,6 +1218,31 @@ static int ov13858_enable_test_pattern(struct ov13858 *ov13858, u32 pattern)
>  				 OV13858_REG_VALUE_08BIT, val);
>  }
>  
> +static int ov13858_update_flips(struct ov13858 *ov13858)
> +{
> +	u32 val;
> +	int ret;
> +
> +	ret = ov13858_read_reg(ov13858, OV13858_REG_FORMAT1,
> +			       OV13858_REG_VALUE_08BIT, &val);
> +	if (ret)
> +		return ret;
> +
> +	if (ov13858->vflip->val)
> +		val |= OV13858_FORMAT1_VFLIP;
> +	else
> +		val &= ~OV13858_FORMAT1_VFLIP;
> +
> +	/* The mirror bit is active low, as it is on ov13b10. */
> +	if (ov13858->hflip->val)
> +		val &= ~OV13858_FORMAT1_HFLIP_N;
> +	else
> +		val |= OV13858_FORMAT1_HFLIP_N;
> +
> +	return ov13858_write_reg(ov13858, OV13858_REG_FORMAT1,
> +				 OV13858_REG_VALUE_08BIT, val);
> +}
> +
>  static int ov13858_set_ctrl(struct v4l2_ctrl *ctrl)
>  {
>  	struct ov13858 *ov13858 = container_of(ctrl->handler,
> @@ -1254,6 +1289,10 @@ static int ov13858_set_ctrl(struct v4l2_ctrl *ctrl)
>  					ov13858->cur_mode->height
>  					  + ctrl->val);
>  		break;
> +	case V4L2_CID_HFLIP:
> +	case V4L2_CID_VFLIP:
> +		ret = ov13858_update_flips(ov13858);
> +		break;
>  	case V4L2_CID_TEST_PATTERN:
>  		ret = ov13858_enable_test_pattern(ov13858, ctrl->val);
>  		break;
> @@ -1481,6 +1520,13 @@ static int ov13858_set_stream(struct v4l2_subdev *sd, int enable)
>  		pm_runtime_put(ov13858->dev);
>  	}
>  
> +	/*
> +	 * Do not let the flips change while streaming. ov13858->mutex is the
> +	 * control handler's own lock and is held here, hence the __ form.
> +	 */
> +	__v4l2_ctrl_grab(ov13858->hflip, enable);
> +	__v4l2_ctrl_grab(ov13858->vflip, enable);
> +
>  	mutex_unlock(&ov13858->mutex);
>  
>  	return ret;
> @@ -1619,6 +1665,12 @@ static int ov13858_init_controls(struct ov13858 *ov13858)
>  			  OV13858_DGTL_GAIN_MIN, OV13858_DGTL_GAIN_MAX,
>  			  OV13858_DGTL_GAIN_STEP, OV13858_DGTL_GAIN_DEFAULT);
>  
> +	ov13858->hflip = v4l2_ctrl_new_std(ctrl_hdlr, &ov13858_ctrl_ops,
> +					   V4L2_CID_HFLIP, 0, 1, 1, 0);
> +	ov13858->vflip = v4l2_ctrl_new_std(ctrl_hdlr, &ov13858_ctrl_ops,
> +					   V4L2_CID_VFLIP, 0, 1, 1, 0);
> +	v4l2_ctrl_cluster(2, &ov13858->hflip);
> +
>  	v4l2_ctrl_new_std_menu_items(ctrl_hdlr, &ov13858_ctrl_ops,
>  				     V4L2_CID_TEST_PATTERN,
>  				     ARRAY_SIZE(ov13858_test_pattern_menu) - 1,

-- 
Regards,

Sakari Ailus