Re: [PATCH 12/19] media: i2c: ov4689: Implement digital gain control

[Mikhail Rudenko <mike.rudenko@gmail.com>]

On 2023-12-18 at 20:04 +02, Laurent Pinchart <laurent.pinchart@ideasonboard.com> wrote:

> Hi Mikhail,
>
> On Tue, Dec 12, 2023 at 03:52:48PM +0300, Mikhail Rudenko wrote:
>> On 2023-12-12 at 00:15 +02, Laurent Pinchart wrote:
>> > On Mon, Dec 11, 2023 at 08:50:15PM +0300, Mikhail Rudenko wrote:
>> >> The OV4689 sensor supports digital gain up to 16x. Implement
>> >> corresponding control in the driver. Default digital gain value is not
>> >> modified by this patch.
>> >>
>> >> Signed-off-by: Mikhail Rudenko <mike.rudenko@gmail.com>
>> >> ---
>> >>  drivers/media/i2c/ov4689.c | 16 ++++++++++++++--
>> >>  1 file changed, 14 insertions(+), 2 deletions(-)
>> >>
>> >> diff --git a/drivers/media/i2c/ov4689.c b/drivers/media/i2c/ov4689.c
>> >> index 62aeae43d749..ed0ce1b9e55b 100644
>> >> --- a/drivers/media/i2c/ov4689.c
>> >> +++ b/drivers/media/i2c/ov4689.c
>> >> @@ -35,6 +35,12 @@
>> >>  #define OV4689_GAIN_STEP		1
>> >>  #define OV4689_GAIN_DEFAULT		0x80
>> >>
>> >> +#define OV4689_REG_DIG_GAIN		CCI_REG16(0x352A)
>> >
>> > Lowercase for hex constatns please.
>>
>> Ah, missed it somehow. Is this convention kernel-wide or media specific?
>> I think checkpatch could have detetected this..
>
> It's media-wide :-) Lower-case hex constants are the majority through
> the kernel, but there's no tree-wide ban on upper-case.
>
>> >> +#define OV4689_DIG_GAIN_MIN		1
>> >> +#define OV4689_DIG_GAIN_MAX		0x7fff
>> >> +#define OV4689_DIG_GAIN_STEP		1
>> >> +#define OV4689_DIG_GAIN_DEFAULT		0x800
>> >> +
>> >>  #define OV4689_REG_TEST_PATTERN		CCI_REG8(0x5040)
>> >>  #define OV4689_TEST_PATTERN_ENABLE	0x80
>> >>  #define OV4689_TEST_PATTERN_DISABLE	0x0
>> >> @@ -131,7 +137,6 @@ static const struct cci_reg_sequence ov4689_2688x1520_regs[] = {
>> >>
>> >>  	/* AEC PK */
>> >>  	{CCI_REG8(0x3503), 0x04}, /* AEC_MANUAL gain_input_as_sensor_gain_format = 1 */
>> >> -	{CCI_REG8(0x352a), 0x08}, /* DIG_GAIN_FRAC_LONG dig_gain_long[14:8] = 0x08 (2x) */
>> >
>> > Is the default value really x2 ? That's not very nice :-S
>> >
>> > It would be much nicer if the default value of the control mapped to x1,
>> > otherwise it's impossible for userspace to interpret the scale of the
>> > digital gain value in a generic way. I suppose that could break existing
>> > applications though, which isn't great.
>>
>> The datasheet does not explicitly say how register values are mapped to
>> the actual gain. 0x8 comes from the original register tables, and can
>> also be found in a few other drivers for this sensor, although they do
>> not implement digital gain control.
>>
>> OTOH, the power-on value of this register, and default value as found in
>> the datasheet, is 0x4. This was the motivation behind that "(2x)"
>> annotation.
>
> I wonder if the chip has a TPG that would be located before the digital
> gain. It would be a nice way to test the digital gain scale.

Thanks for the suggestion, just tested that. Unfortunately, all the
supported test patterns are not affected by digital gain at all :(

But what if we set the digital gain control's default value in
v4l2_ctrl_new_std to 0x400 (power-on default), right after that set
ctrl->cur.val to 0x800 (default value before this series), and explain
the situation in a comment? Thus we could keep the effective default
value, and make it clear that it is 2x at the same time.

What do you think?

>> So, I'm afraid that we cannot interpret the absolute scale of the
>> digital gain in any case, unless we have more documentation. I tend to
>> keep the default value of 0x8 for the reasons of not (possibly) breaking
>> userspace.
>>
>> > Out of curiosity, can you tell what SoC(s) you're using this sensor with
>> > ?
>>
>> It's Rockchip 3399. I run most of my tests with AGC and AWB off, to be
>> sure they do not hide some important details.
>>
>> >>
>> >>  	/* ADC and analog control*/
>> >>  	{CCI_REG8(0x3603), 0x40},
>> >> @@ -622,6 +627,9 @@ static int ov4689_set_ctrl(struct v4l2_ctrl *ctrl)
>> >>  				OV4689_TIMING_FLIP_MASK,
>> >>  				val ? 0 : OV4689_TIMING_FLIP_BOTH, &ret);
>> >>  		break;
>> >> +	case V4L2_CID_DIGITAL_GAIN:
>> >> +		cci_write(regmap, OV4689_REG_DIG_GAIN, val, &ret);
>> >> +		break;
>> >>  	default:
>> >>  		dev_warn(dev, "%s Unhandled id:0x%x, val:0x%x\n",
>> >>  			 __func__, ctrl->id, val);
>> >> @@ -650,7 +658,7 @@ static int ov4689_initialize_controls(struct ov4689 *ov4689)
>> >>
>> >>  	handler = &ov4689->ctrl_handler;
>> >>  	mode = ov4689->cur_mode;
>> >> -	ret = v4l2_ctrl_handler_init(handler, 13);
>> >> +	ret = v4l2_ctrl_handler_init(handler, 14);
>> >>  	if (ret)
>> >>  		return ret;
>> >>
>> >> @@ -693,6 +701,10 @@ static int ov4689_initialize_controls(struct ov4689 *ov4689)
>> >>  	v4l2_ctrl_new_std(handler, &ov4689_ctrl_ops, V4L2_CID_VFLIP, 0, 1, 1, 0);
>> >>  	v4l2_ctrl_new_std(handler, &ov4689_ctrl_ops, V4L2_CID_HFLIP, 0, 1, 1, 0);
>> >>
>> >> +	v4l2_ctrl_new_std(handler, &ov4689_ctrl_ops, V4L2_CID_DIGITAL_GAIN,
>> >> +			  OV4689_DIG_GAIN_MIN, OV4689_DIG_GAIN_MAX,
>> >> +			  OV4689_DIG_GAIN_STEP, OV4689_DIG_GAIN_DEFAULT);
>> >> +
>> >>  	if (handler->error) {
>> >>  		ret = handler->error;
>> >>  		dev_err(ov4689->dev, "Failed to init controls(%d)\n", ret);


--
Best regards,
Mikhail Rudenko