The driver was using reverse engineered pixel rates, and had different
values for different link frequencies.
In common with the other Starvis sensors (eg imx290), this isn't
actually the situation, and the pixel array runs off a different
PLL to the MIPI block, and there is a FIFO between the two.

It's not quite so simple as HMAX is said to be in units of INCK,
not pixel clocks. Working through the various modes that appears that
even that is slightly wrong as it is units of a scaled INCK of
either 72MHz (for INCK = 24 or 72MHz) or 74.25MHz (for INCK = 27,
37.125, or 74.25MHz).

Rework the configuration so that the pixel rate is fixed, with the
different link frequencies adjusting the minimum hmax value so that
the FIFO doesn't overflow.

Switching between 2 and 4 data lanes also only changes the MIPI data
rate, so also can be handled by altering the minimum HMAX.

Once you've got HMAX / V4L2_CID_HBLANK represented sensibly, then
frame rate control through V4L2_CID_HBLANK and V4L2_CID_VBLANK is
trivial, so add in frame rate control too.
Exposure is set in lines (as usual), so amending VBLANK changes the
range for the exposure control.

This has been tested on a Pi5 with Waveshare "IMX415-98 IR-CUT Camera"
module (SKU 28524). That uses a 24MHz clock, so only the 360MHz and
720MHz link frequencies have been tested (other frequencies require
a 27, 37.125, or 74.25MHz clock input.

It'd be nice to add support for 12bit readout and windowed mode readout,
but those are a job for another day.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
---
Changes in v2:
- Fixed VBLANK fall-through for setting the exposure correctly
- Sorted the checkpatch complaint over using fallthrough; vs commenting
  why we're falling through.
- Corrected HMAX for 891Mbit/s mode. I now have a module with 37.125MHz
  clock, and have validated that.
- Link to v1: https://lore.kernel.org/r/20250109-media-imx415-v1-0-366ba6a234ab@raspberrypi.com

---
Dave Stevenson (3):
      media: i2c: imx415: Add read/write control of VBLANK
      media: i2c: imx415: Make HBLANK controllable and in consistent units
      media: i2c: imx415: Link frequencies are not exclusive to num lanes

 drivers/media/i2c/imx415.c | 182 +++++++++++++++++++++++----------------------
 1 file changed, 93 insertions(+), 89 deletions(-)
---
base-commit: 94794b5ce4d90ab134b0b101a02fddf6e74c437d
change-id: 20250107-media-imx415-2df67ae21107

Best regards,
-- 
Dave Stevenson <dave.stevenson@raspberrypi.com>

Hi Dave,

Thanks for your patches!

On 1/16/25 18:43, Dave Stevenson wrote:
> The driver was using reverse engineered pixel rates, and had different
> values for different link frequencies.
> In common with the other Starvis sensors (eg imx290), this isn't
> actually the situation, and the pixel array runs off a different
> PLL to the MIPI block, and there is a FIFO between the two.
> 
> It's not quite so simple as HMAX is said to be in units of INCK,
> not pixel clocks. Working through the various modes that appears that
> even that is slightly wrong as it is units of a scaled INCK of
> either 72MHz (for INCK = 24 or 72MHz) or 74.25MHz (for INCK = 27,
> 37.125, or 74.25MHz).
> 
> Rework the configuration so that the pixel rate is fixed, with the
> different link frequencies adjusting the minimum hmax value so that
> the FIFO doesn't overflow.
> 
> Switching between 2 and 4 data lanes also only changes the MIPI data
> rate, so also can be handled by altering the minimum HMAX.
> 
> Once you've got HMAX / V4L2_CID_HBLANK represented sensibly, then
> frame rate control through V4L2_CID_HBLANK and V4L2_CID_VBLANK is
> trivial, so add in frame rate control too.
> Exposure is set in lines (as usual), so amending VBLANK changes the
> range for the exposure control.
> 
> This has been tested on a Pi5 with Waveshare "IMX415-98 IR-CUT Camera"
> module (SKU 28524). That uses a 24MHz clock, so only the 360MHz and
> 720MHz link frequencies have been tested (other frequencies require
> a 27, 37.125, or 74.25MHz clock input.
> 
> It'd be nice to add support for 12bit readout and windowed mode readout,
> but those are a job for another day.
> 
> Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>

Reviewed-by: Michael Riesch <michael.riesch@wolfvision.net>

Regards,
Michael

> ---
> Changes in v2:
> - Fixed VBLANK fall-through for setting the exposure correctly
> - Sorted the checkpatch complaint over using fallthrough; vs commenting
>   why we're falling through.
> - Corrected HMAX for 891Mbit/s mode. I now have a module with 37.125MHz
>   clock, and have validated that.
> - Link to v1: https://lore.kernel.org/r/20250109-media-imx415-v1-0-366ba6a234ab@raspberrypi.com
> 
> ---
> Dave Stevenson (3):
>       media: i2c: imx415: Add read/write control of VBLANK
>       media: i2c: imx415: Make HBLANK controllable and in consistent units
>       media: i2c: imx415: Link frequencies are not exclusive to num lanes
> 
>  drivers/media/i2c/imx415.c | 182 +++++++++++++++++++++++----------------------
>  1 file changed, 93 insertions(+), 89 deletions(-)
> ---
> base-commit: 94794b5ce4d90ab134b0b101a02fddf6e74c437d
> change-id: 20250107-media-imx415-2df67ae21107
> 
> Best regards,