feat(camera): support e-con e-CAM81_CUONX (AR0821) on JAJ

[dakejahl]

Summary

Adds e-con e-CAM81_CUONX (8MP AR0821 with on-module ISP/MCU, UYVY out) support on the JAJ target: the vendored and ported driver, two JAJ overlays (dual 2-lane and CAM1-only 4-lane for 4K30), and a usage doc at docs/cameras/econ-cam81/README.md. Compile- and DT-apply-validated; bench validation on hardware is still pending.

Problem

e-con only ships this camera for L4T 35.x (kernel 5.10, in-tree kernel/nvidia drivers) as full replacement DTBs for the NVIDIA dev kit. Nothing existed for our R36.5.0 nvidia-oot tree or ARK carriers, and the camera can't ride the existing Bayer/argus path at all: it is an MCU-fronted YUV device with its own V4L2 control set and I2C-flashed firmware.

Solution

Port the ar0821 driver to kernel 5.15 / nvidia-oot and vendor it via kernel_overlay/ with a fail-loud build.sh registration next to the IMX708 one. The port is small: pad ops take v4l2_subdev_state, and e-con's subdev g_parm/s_parm become g/s_frame_interval — R36's VI channel already routes VIDIOC_[GS]_PARM there via v4l2_[gs]_parm_cap, so none of e-con's kernel/VI framework patches are needed (the driver references none of those symbols). Their dev-kit DTBs are converted into two jetson-io overlays wired per ARK's flash-validated JAJ CSI topology (CAM0 = serial_a/port-index 0 + lane_polarity 6 rather than e-con's serial_b; the adapter's PCA6408 expander binds through the stock ti,tca6408 compatible with the BOOT0/RESET pins left unhogged so the driver can request them). The module builds warning-free under the oot -Werror against the staged R36.5 tree, and both dtbos compile and fdtoverlay-apply onto the built JAJ base DTB. CAM0-on-2-lane has no working prior art from e-con and is flagged as pending bench validation in the docs.

[dakejahl]

Bench bring-up notes (paused)

Software on this PR is largely in place and on the test JAJ (driver loaded, dual overlay selected via jetson-io, DT nodes okay). The camera still never probes: adapter LED never lights, i2cdetect on cam mux buses 9/10 is blank, pca953x @ 0x20 fails with -121, and ar0821 @ 0x42 stays deferred waiting on that expander. IMX219/477/708 still work on the same CSI ports, so the carrier rail/I2C path is fine.

What we already tried

• FFC matrix exhausted (same-side / opposite-side, both contact orientations) on both CAM0 and CAM1.
• Overlay experiments on the bench JAJ: drop carrier CAM*_PWDN hog, PCA hog pin 7 only, add vcc-supply (used &hdr40_vdd_3v3 — p3768_vdd_3v3_sys is not a symbol on the JAJ base DTB). Redeployed + rebooted; still no LED / no ACK.

Likely next steps

1. Hard power check on the ACC-RB-WTB-ADP: meter CN3 pin 1 (VCC_3P3) to pin 4 (GND) with Jetson on and FFC seated (pin 1 is marked; JST SM04B-SRSS — don’t probe the 0.5 mm FFC). ~3.3V → adapter rail alive, chase CN2/module/LED; 0V → no power into the ACC board (cable mapping or dead input rail).
2. If 0V after a known-good cable: treat the adapter/module as suspect (e-con docs warn reverse FPC can damage the kit; stock e-con FPC is aimed at NVIDIA’s mirrored CSI pinout vs JAJ’s RPi-style numbering).
3. If 3.3V + 0x20/0x42 appear: finish probe, then revisit CSI wiring — JAJ dual currently uses CAM0 serial_a/port-index 0; e-con stock + working JAJ IMX477 use serial_b/port-index 1 for CAM0; CAM1 4-lane serial_c/2 is the safer first port.
4. Optionally land the DT tweaks from this session (PWDN hog removed, PCA hog pin 7, vcc-supply) if they aren’t already on the branch tip, and document the CN3 meter check + NVIDIA-vs-JAJ FPC pinout caveat in docs/cameras/econ-cam81/README.md.