[libcamera-devel] IMX258 and speckled images

[Dave Stevenson]

Hi Kieran

Thanks for responding

On Thu, 23 Jun 2022 at 17:40, Kieran Bingham
<kieran.bingham at ideasonboard.com> wrote:
>
> Hi Dave,
>
> Quoting Dave Stevenson via libcamera-devel (2022-06-22 11:34:16)
> > Hi All.
> >
> > A quick query as I'm aware that others on here are using the Sony IMX258 sensor.
> > Are you using the sensor driver as-is from mainline[1], or does it need changes?
>
> I think the Nautilus uses the IMX258. But as a ChromeOS device, the
> driver should be expected to be as-is from mainline.
>
>
> > A third party manufacturer has made an IMX258 module for the Pi[2],
> > but we're getting dark speckles in the raw image, which we're guessing
> > is down to PDAF pixel correction. The module manufacturer also
> > believes that there may be versions of the module with and without
> > PDAF pixels, which may well explain the issue.
>
> I guess this might look like noise on the resulting image? or is it
> fixed spots that look like broken pixels? (An image might help if
> feasible).

Sorry, I should have thought to send images with my first email :-/

https://drive.google.com/file/d/1i4T3CuB-zJU8MT0XX_XC-UsKRitTxdTF/view?usp=sharing
JPGs and DNGs for two scenes in each mode.
Register 0x4e21 as 0x14 is the current driver's behaviour. Setting it
to 0x10 clears the odd pixels. What does bit 2 do?!

> > It seems that register 0x4e21 being set to 0x14 by the driver is the
> > main cause, and the default of 0x10 gives good images. Of course
> > register 0x4e21 isn't documented in any of the datasheets and appnotes
> > I have from Sony. I have an enquiry in with them, but I wondered if
> > others had seen similar issues, or might have documentation about that
> > register.
>
> Umang, can you confirm if this is the driver in use by the Nautilus?

I have received some information back from Sony over different values
in the OTP for different variants of the sensor. If someone does have
a device which uses it, then I can give them some I2C commands to read
the relevant bits and see what those platforms are using. It's the
normal NDA faff with sensor manufacturers so I can't really send it to
an open list.

Part of the issue is knowing which variant the existing driver is
meant to work with to ensure we don't break that.

  Dave