# 90-mirror-wall-watchdog.conf — PID 1 must kick the hardware watchdog.
#
# Installed by install-kiosk.sh to /etc/systemd/system.conf.d/, but ONLY when the
# box actually has a watchdog device (/dev/watchdog). Takes effect on
# `systemctl daemon-reexec`, which the installer runs; a reboot also picks it up.
#
# This is the layer that catches what the sysctls cannot. The soft/hard lockup
# detectors need kernel work to keep happening; a hardware watchdog needs the
# opposite — if PID 1 stops writing /dev/watchdog0 for one whole timeout, the
# chipset resets the machine no matter what the kernel thinks it is doing. That
# includes the case we hit on 2026-11-02, where the box answered ARP and TCP
# handshakes from kernel state while no userspace process could be started.
#
# 30s is not an arbitrary number: it is the timeout iTCO_wdt initialised itself
# with ("heartbeat=30 sec"). systemd kicks at a third of the configured timeout,
# so 30s means a kick every 10s. Asking for more than the device supports would
# make systemd kick less often than the hardware expires — a self-inflicted
# reboot loop — so install-kiosk.sh keeps this at or below the driver's value.
#
# Cost: a PID-1 stall longer than 30s (a genuinely stuck disk, say) reboots the
# wall. On an unattended display that is the intended behaviour. To back off,
# raise the number to something under the device timeout, or set 0 and rely on
# the sysctls alone:
#
#   printf '[Manager]\nRuntimeWatchdogSec=0\n' | sudo tee /etc/systemd/system.conf.d/91-mirror-wall-watchdog-off.conf
#   sudo systemctl daemon-reexec

[Manager]
RuntimeWatchdogSec=30s