A hard hang writes nothing to disk — capture on the host, before you trigger
scope: generic · severity: trap · confidence: proven · subsystem: diagnosis
A log file on the device will be empty after a hard hang. The write never reached storage.
Capture all pre-test state on the host before you trigger anything.
Your options for evidence that survives, in order of dependability:
- Netconsole — the dependable channel. Transmits over usb0 and wlan0, including from NMI context. Prefer usb0: wlan0 WARNs from hardirq.
- ramoops / pstore — a warm reboot preserves
/sys/fs/pstore/; a power-button reset destroys it. And verify pstore works on your device before relying on it: taimen’s bootloader is recorded as scrubbing the ramoops region anyway. - A file on the device — useless for hard hangs. Fine for anything that survives to flush.
Before any unattended hang-producing work, know which of these you actually have. Discovering you have none after the hang costs the whole experiment.
Related: watchdog-out-of-range-disarms-instead-of-clamping, a-journal-grep-matches-your-own-command-line.