Internals: the watchdog subsystem

kernel/core/core_watchdog.{c,h} - two watchdog worlds, one header.

Policy watchdogs (monitored slots)

A task registers a slot with a timeout:

slot = watchdog_register("mydaemon", 5000, policy);
watchdog_start(slot);
... watchdog_feed(slot);   /* at least every 5 s */

Timeout fires -> the TASK is killed, the reason lands in the kernel log (test log ERROR in the E2E), the KERNEL lives on. This is the right tool for "my background worker may hang".

Hardware watchdogs

• Intel TCO (LPC): PMBASE discovered from the chipset, TCOBASE derived, the timer armed; expiry hard-resets the machine. Live on QEMU q35 - the E2E S12 prints the discovery line, arms, starves, and asserts the REAL second boot.
• ACPI WDAT: the table's entries program the hardware; same verify-by-reboot cycle (E2E S13).

watchdog shows both worlds: hardware support + slot table + policy tests.

The E2E scenarios that gate this

S7 (status/slots), S8 (policy kill + log), S12 (TCO real reset + second boot), S13 (WDAT real reset + second boot). Fourteen total in tools/wp12_watchdog_logging_e2e.py - all fourteen must PASS at release.