Re: [PATCH v5 0/2] media: Add Himax HM1092 mono NIR sensor driver

[Ramshouriesh R]

Hi Jake,

I have v6 ready now. I went back to the ASUS Windows files rather than
deriving the link frequency from an assumed 30 fps measurement.

The Qualcomm power sequence explicitly requests a 24 MHz EXTCLK. The
sensor initialization programs:

0x030d = 0x0c = 12 (PLL pre-divider)
0x030f = 0x5a = 90 (PLL multiplier)

The resulting CSI-2 DDR link frequency is:

link_freq
= EXTCLK * multiplier / pre-divider
= 24,000,000 * 90 / 12
= 180,000,000 Hz

Because CSI-2 D-PHY transfers on both clock edges, the per-lane bit rate is:

lane_bit_rate
= 2 * link_freq
= 2 * 180,000,000
= 360,000,000 bit/s

The mode uses one data lane and RAW10, so the corresponding pixel rate is:

pixel_rate
= lane_bit_rate * number_of_lanes / bits_per_pixel
= 360,000,000 * 1 / 10
= 36,000,000 pixels/s

The initialization table programs:

HTS = 0x0650 = 1616 pixel clocks per line
VTS = 0x02ee = 750 lines per frame

Therefore, the calculated frame rate is:

frame_rate
= pixel_rate / (HTS * VTS)
= 36,000,000 / (1616 * 750)
= 36,000,000 / 1,212,000
= 29.702970... fps

So v6 now reports:

link frequency: 180 MHz
lane bit rate: 360 Mbit/s
pixel rate: 36 MHz
frame rate: approximately 29.7 fps

The binding and driver both use the 180 MHz link frequency.

I folded exposure, analogue gain and digital gain into v6 using your
register map and notes:

exposure uses separate writes to 0x0202 and 0x0203
analogue gain uses 0x0205
digital gain uses separate writes to 0x020e and 0x020f
group hold uses 0x0104 = 1 before an update and 0x0104 = 0 afterward

The exposure maximum is derived from the frame length:

exposure_max
= VTS - margin
= 750 - 21
= 729 lines

The exposure range is therefore 2..729 lines. I selected 500 lines as
the default from unstretched illuminated captures on the ASUS system:

500 decimal = 0x01f4

The init sequence and control setup both program it as:

0x0202 = 0x01
0x0203 = 0xf4

This keeps the sensor state and V4L2 control default consistent.

Analogue gain has range 0..255 and default 0.

Digital gain has:

minimum: 0x0100 = 256 = 1.0x
maximum: 0x0fff = 4095
default: 0x0100

I tested exposure 2, 500 and 729 and analogue gain 0 and 255 on the
560x360 mode. Those settings were accepted and changed real frame
levels in the expected direction. I also exercised the digital-gain
endpoints while streaming and captured valid frames. Runtime PM
returned the sensor to suspend after capture.

Here is the ordered Qualcomm table you requested: 185 (address, value)
entries regenerated directly from com.qti.sensormodule.hm1092.bin:

https://gist.github.com/sppidy/8b6b57f047343c88d4bb6c8ef22a910d

Please post the comparison against your Dell hm1092.sys table when you have it.

For ipu-bridge, I would prefer a follow-up patch on top of v6. That
keeps the x86/IPU7 ownership and testing explicit, since I cannot
exercise that path on the A14.

I currently credit your findings in the commit body, with Reported-by
and links to both messages, and in the cover letter. If you think
Suggested-by is more appropriate for the control implementation, or
want Co-developed-by with your Signed-off-by, let me know before I
send v6.

Regards,
Ramshouriesh