To implement a watchdog in software or hardware, only a single function needs to be written and registered on the global watchdog_list.
The function must examine the cmd argument and act on it as follows:
If cmd is zero, the watchdog must be disabled and the error argument left untouched.
Else the watchdog should be reset and configured to a timeout of (1 <<(cmd & WD_INTERVAL)) nanoseconds or larger and the error argument be set to zero.
If the watchdog cannot be configured to the proposed timeout, it must be disabled and the error argument left untouched.
There is no specification of what the watchdog should do when it times out, but a hardware reset or similar ""drastic but certain"" behaviour is recommended.