Running into a weird crash on a Jetson Orin Nano Super and hoping someone’s seen this before.
Setup:
-
Realtek rtl88x2ce WiFi (PCIe M.2,
0001:01:00.0, bridge14100000.pcie) -
YOLOX-m inference at 12fps, FP16 PyTorch on CUDA
-
USB V4L2 camera running continuously
What happens:
When the AP drops the internet connection abruptly, the Jetson hard resets within 12 seconds. No logs survive — journald, pstore/ramoops, kern.log, tegra_cbb_err all empty at the crash timestamps. Looks like a BCCPLEX-level reset.
The weird part: if the internet is down from the moment the device boots, it runs fine indefinitely with CUDA and camera both going. It only crashes when it successfully connects first, CUDA loads, camera starts, then the link drops. From logs:
09:17:23 — connected to API
09:17:31 — YOLOX loaded on CUDA
09:17:34 — camera up, capture loop running
09:17:37 — DNS errors start (internet dropped)
09:17:48 — hard reset
So the consistent-down case is fine, the drops-while-running case kills it every time.
Started happening after the site got a new AP. Same workload was running on the old AP without issues — new AP seems to drop the link harder/faster.
Checked so far:
-
pstore/ramoops — empty, reset happens before kernel can write anything
-
tegra_cbb_err — empty
-
kern.log — nothing at crash timestamps
-
/proc/last_rst_reason — not available on this board
No serial console available unfortunately.
Is the rtl88x2ce driver known to cause PCIe fatal errors on abrupt link loss when CUDA is active? Any known patches, kernel params, or PCIe AER settings that help? Open to trying a USB WiFi adapter if that’s the general recommendation for this situation.