Re: [PATCH v5 2/2] media: i2c: Add Sony IMX908 image sensor driver

[Dave Stevenson]

Hi Sakari (and Laurent)

On Wed, 21 Oct 2026 at 13:05, Laurent Pinchart
<laurent.pinchart@xxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Oct 21, 2026 at 02:56:50PM +0300, Sakari Ailus wrote:
> > On Tue, Oct 20, 2026 at 03:50:09PM +0100, Dave Stevenson wrote:
> > > On Tue, 20 Oct 2026 at 13:56, Laurent Pinchart wrote:
> > > > On Tue, Oct 20, 2026 at 01:47:11PM +0300, Sakari Ailus wrote:
> > > > > On Mon, Oct 19, 2026 at 10:47:20AM +0900, Lachlan Michael wrote:
> > > > > > The IMX908 is an 8.39 MP (3856x2176) CMOS image sensor with a MIPI CSI-2
> > > > > > output interface. The driver supports RAW10 and RAW12 output formats,
> > > > > > exposure and analogue gain controls, image flipping, blanking controls
> > > > > > and test pattern generation.
> > > > >
> > > > > The imx908 appears to be a Starvis 3 sensor. How close is it to the Starvis
> > > > > 2, for which Dave (Cc'd) is working on a driver (see
> > > > > <URL:https://lore.kernel.org/linux-media/atdAsDGw_jvhaydM@kekkonen.localdomain/T/#m0c50ea017ef9ae0d3d76a36b840db9b6f7ac296c>)?
> > > > > The register defines in the two drivers appear to be quite well aligned.
> > >
> > > Asked and answered at v1
> > > https://lore.kernel.org/linux-media/TY6PR01MB17839FF48671E744E7E23A93E92D42@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> > >
> > > > The current version of the driver implements very few features of the
> > > > IMX908 sensor, as crop support has been removed pending the raw camera
> > > > sensor model, and Clear HDR3 is being developed. I expect more
> > > > differences to appear when those features will be merged, so I believe
> > > > it makes sense to have a separate driver here.
> > > >
> > > > In the future, I expect the raw camera sensor model to open the door to
> > > > more code sharing and more drivers supporting a wider variety of
> > > > devices. We can decide then if the imx908 should stay a standalone
> > > > driver, support other sensors as well, possibly across the Starvis 2 and
> > > > 3 families. I'm willing to help with the refactoring at that point if
> > > > needed, the maintenance burden won't be on you (and I'm sure that
> > > > Lachlan and his team will of course maintain the code base).
> > >
> > > Agreed.
> > > I have compared the datasheet for imx908 against the Starvis2 ones,
> > > and when we start looking at HDR modes they diverge a reasonable
> > > amount.
> > >
> > > Even the later Starvis2 sensors (IMX585 and IMX676) have minor changes
> > > to the earlier ones, so adding Starvis3 would need some care.
> >
> > There's quite a bit of this in the CCS driver, due to the variance it
> > allows for implementors. I don't see this being a problem as long as the
> > foundations are reasonably similar, albeit we can see whether this could be
> > further extended later on to other sensors.
> >
> > I'm fine with having a separate imx908 driver for now. Dave's driver also
> > exposes the variant information through the media entity name so that
> > should be done by the imx908, too (at least if there can be multiple ones).

No it doesn't. My driver passes .name from the variant to
v4l2_i2c_subdev_set_name, so the media entity name is common for both
mono and colour variants.
It's the same with the imx290 driver supporting the different variants
of 290/327/462.

libcamera uses the media bus codes to determine whether the sensor is
mono or colour, not the entity name.
Having different entity names would mean duplicate entries in
libcamera's sensor_properties, or otherwise manipulate the name.

  Dave

> As far as I understand, the IMX908 does not have a mono version.
> Lachlan, could you please confirm that ?
>
> > I'll review the driver properly soonish but I reckon it's in a pretty good
> > shape already.
>
> --
> Regards,
>
> Laurent Pinchart