VOXL2 M0154 hangs (no self-recovery) with FLIR Boson connected directly to internal USB2 port

[JohnnyS]

Hardware:
VOXL2, M0154 (2023 revision), SKU TF-M0054-4-V1-C11
Software:
voxl-suite 1.6.5, system-image 1.8.08-M0054-14.1a-perf, kernel 4.19.125

Camera: FLIR Boson (FLIR vendor ID 09cb), connected via UVC/USB2 directly to VOXL2's internal USB2 host port (not via M0153 MIPI adapter)

Comparison unit: an older VOXL2 M0054 (2021) with the same setup DOES NOT show this problem.

SYMPTOM
Board hard-freezes (not a clean reboot) within ~30-100s of the camera actively streaming, when the camera is connected directly to the internal USB2 port (no hub in between). Freeze is total: kernel logging, journald, and an independent on-device background process (unrelated to camera/USB) all stop simultaneously with zero warning in dmesg/journal beforehand. No panic, no watchdog trace, no lockup/stall message — just silence. The board does not self-recover; only a manual power cycle brings it back. Confirmed via app-level logs too: buffered log writes already visible on a live tail -f are lost from disk after recovery, showing the whole system (including the kernel's own writeback) stopped instantly with no chance to flush.

Observed at the moment of freeze: power LED stays solid on; camera's own LED goes dark — suggesting the camera's specific USB port power rail is what actually loses power, not the whole board.

This problem happens repeatedly on two different M0154 boards, but does not happen on M0054 boards.

REPRODUCTION

1. Connect Boson camera directly to VOXL2 internal USB2 port (no hub).
2. Stream from the camera (e.g. continuous V4L2 capture).
3. With no USB-C cable to a host connected: board freezes within roughly 30-100 seconds, unrecoverable without power cycle.
   With USB-C cable connected then disconnected mid-stream: also reproduces (sometimes as a clean PMIC SMPL/UVLO reset with automatic reboot, sometimes as the hard freeze above — inconsistent).

WORKAROUND FOUND
Placing a plain bus-powered USB2 hub (Genesys Logic 4-port, 05e3:0610) between the camera and the internal port fixes it completely — board survives indefinitely, with or without a USB-C cable connected. This points to the internal USB2 rail feeding that port not having enough local current/transient capability for the Boson's direct USB draw; the hub's own local bulk capacitance/current-limited power switch appears to absorb what the internal rail can't.

ANOTHER WORKAROUND
Connect any device to the USB-C output - it can be a laptop, a USB storage device or even a USB hub with nothing connected to it. In such a case, there will be no resets.

ADDITIONAL NOTES

• No working hardware watchdog was found on this unit: /dev/watchdog does not exist, and msm_watchdog logs "wdog absent resource not present" at boot — so any future hang, from this or any other cause, also won't self-recover.
• Docs note M0154 is the revision where the USB-C port's host-mode support was brought to parity with VOXL2 Mini (M0054 didn't support USB-C host the same way) — asking in case that host-mode rework touched this same internal USB2 power path.
• Separately, after extended testing, this unit's USB-C debug port has stopped enumerating on a host entirely (no lsusb/adb activity at all, even after full power cycle) — possibly unrelated connector/PHY damage from repeated stress, flagging in case it's relevant to the same root cause.

QUESTION:
Is there a known current/power budget limitation on VOXL2's internal USB2 port when powering the Boson directly (vs. via the M0153 MIPI adapter path), and did anything on that specific power rail change between M0054 and M0154?