Skip to content

The hardware watchdog armed by install.sh. Tried, 2026-11-17

The appliance's hardware watchdog is on by default (XGS: Settings, Hardware watchdog, since 2026-10-17). Until this, a first install left the timer unarmed until the next boot. Now install/install.sh arms it at once, unless a watchdogd already holds it, and says which of these happened under "== hardware watchdog ==" (#1183).

How it was tried

On the XGS 136 (OPNsense 26.7.9), at 13:08 UTC, from a copy of the pull request's tree - 38a8cc94; its commits after it change documentation only - install.sh was run whole in each step, as XGS: Updates runs it. Before each step the timer was put in the state the step names:

  • LT1 an update: the timer armed since boot;
  • LT2 a first install: the timer stopped first;
  • LT3 descriptor 9 held, as XGS: Updates holds it while it runs the installer;
  • LT4 the setting off and Applied, then on and Applied;
  • LT5 a watchdogd started by hand holds the timer;
  • LT9 os-watchdog's watchdogd, started as that plugin starts it, holds the timer;
  • LT10 a dry run (watchdogd -n) beside it;
  • LT11 a watchdogd on the kernel's software timer only.

What came of it

  • LT1 "armed already": the watchdogd that held the timer since boot ran on.
  • LT2 "armed: AMD FCH Rev 41h+ Watchdog Timer, 128 s": the timer armed at once, before any reboot.
  • LT3 armed, and the watchdogd install.sh started holds neither descriptor 9 nor the lock, so XGS: Updates does not wait on it.
  • LT4 with the setting off: "the hardware watchdog is switched off in the settings", "not armed"; Apply disarmed the timer, and Apply with the setting on armed it again.
  • LT5 "the timer is held by another watchdogd, ...; left to it"; once that one stopped, watchdog.sh start armed the timer.
  • LT9 os-watchdog's watchdogd was stopped ("while this script is installed it arms the timer"), and the timer armed.
  • LT10 armed; the dry run beside it was not counted as holding the timer.
  • LT11 "the timer is held by another watchdogd, ..., on the kernel's software timer only - the hardware timer is not armed; left to it". One more watchdogd by hand was refused by the kernel (setting WDIOC_SETSOFT 0: Invalid argument); once the first stopped, watchdog.sh start armed the timer.
  • Afterwards the timer was armed, 128 s. install.sh ended 0 in every step, after 18 to 20 s each.

The steps on the XGS 136, each with what install.sh said and what holds the timer after

The same steps ran on the second XGS 3300 that morning, 09:43-09:46 UTC, with the same outcome. After a reboot of each appliance the stop hook had disarmed the timer and the start hook armed it again. Those runs are in the pull request's notes, not photographed.

Discussion

The discussion of this page is held in the project's GitHub Discussions. Showing it here loads giscus from giscus.app and GitHub, and writing in it needs a GitHub account - what that means.