Re: [PATCH 2/4] media: i2c: Add ST VD55G0 global shutter sensor driver

[George Neacsu <geocausa@gmail.com>]

On Mon, Oct 19, 2026 at 07:34:10PM +0000, Sergey Lebedev wrote:
> From 19.2 MHz, 403.2 MHz (806.4 Mbps, 42 times the clock) and
> 384 MHz (768 Mbps, 40 times) are exact, so the change is one value in
> your board's DT, in the sensor endpoint:
>
> 	link-frequencies = /bits/ 64 <403200000>;
>
> If it still stops after one frame, I would be glad to compare what the
> two drivers write at stream start.

That took the clock mismatch out of the experiment. With 403.2 MHz in
the DT and your v1 otherwise as posted, the sensor probes cleanly and
reports link_frequency=403200000 and pixel_rate=134400000, but capture
still delivers exactly one 644x604 Y10 frame and then times out waiting
for frame 2.

I then went back to the earlier VD55G0 A/B tests on this same Snapdragon
machine. One result was particularly useful: the 1200-pixel line length
alone still stalled, and omitting the 0x0329 write with the old line
length still stalled, but 1200 plus leaving 0x0329 untouched streamed
continuously. Your driver already has the 1200-pixel line length, but it
also has:

	vd55g0_write(sensor, VD55G0_REG_ISL_ENABLE, 0, &ret);

I rebuilt v1 with only that single write omitted, keeping the 403.2 MHz
DT setting and everything else unchanged. That fixes continuous capture
here:

- 5/5 frames in the initial Y10 644x604 capture
- 120/120 frames in a continuous run at about 60.2 fps
- immediate stream restart: 10/10 frames
- v4l2-compliance on the sensor subdev: 46/46, 0 failures, 0 warnings

There were no new CAMSS/CSID/VFE errors during the runs.

So the link-frequency correction is needed, but the one-frame stall is a
separate issue caused here by clearing VD55G0_REG_ISL_ENABLE (0x0329).
I have not tried to infer yet why clearing it prevents subsequent frames
on this sensor, but the A/B is repeatable and agrees with the older
VD55G0 results on this machine.

Would you be comfortable dropping that write in the next revision? I can
retest the revision on the Snapdragon model and add a Tested-by once it
is clean.

Best regards,
George