Comment # 13 on bug 1042933 from Borislav Petkov

(In reply to Thomas Blume from comment #12)
> Boris, I'm wondering when exactly the watchdog timer starts counting down.
> Is it when opening /dev/watchdog or when doing the ping (ioctl(watchdog_fd,
> WDIOC_KEEPALIVE, 0) or something else?

When you do WDIOC_SETTIMEOUT. When the kernel gets the timeout correctly
from userspace, it does the ping which goes and reprograms the timer.

It also reprograms the timer when you do watchdog_ping()

 [ That thing does WDIOC_KEEPALIVE which does the reprograming too. ]

and since nothing changes the timeout, it should simply "extend" the
timeout to the 10s interval and thus not fire during the test usleeps
for t/2 seconds.

Do you see anything in dmesg from the watchdog while the test runs, some
failure messages or so?

If not, you could simply go and add pr_err() calls to
drivers/watchdog/hpwdt.c, more specifically hpwdt_ping() and dump
the reload variable there, hpwdt_change_timer() and a couple more
interesting.

And since it is a module, you don't need to reboot the machine - simply
rmmod/insmod it.

And in order to avoid the panicking, change hpwdt_pretimeout() to do:

        if (allow_kdump)
                hpwdt_stop();

    return NMI_HANDLED;

so that you don't panic the box.

This way you'll have a better idea what is happening.

HTH.