Ed L Cashin <ecashin@xxxxxxxxxx> writes:
> Hi. I'm preparing to use the NMI watchdog feature to do some
> debugging. To get ready, I've created a module like this:
>
> /* ... */
> int deadlock_init(void)
> {
> local_irq_disable();
> while ("I want to loop!")
> ;
>
> return 0;
> }
>
> module_init(deadlock_init);
> /* ... */
>
> I boot with these parameters:
>
> lapic nmi_watchdog=2 console=ttyS0,9600n8 console=tty0
>
> I see nmi.c:check_nmi_watchdog is satisfied:
>
> testing NMI watchdog ... OK.
>
> ... and I can see the nmi count in /proc/interrupts increasing every
> couple minutes or so. I have a serial console going to another
> computer. When I load this module, though, I don't ever see any
> debugging information on the screen or on the serial console.
>
> Is there anything else I need to do? I'm assuming that I should see
> an NMI-watchdog-generated oops because of the infinite loop with
> interrupts disabled.
For anyone who's interested in this, I found out a couple more
things. The machine that I was using for the test described above has
a Pentium 4, and I couldn't see the nmi interrupt count increasing in
/proc/interrupts when I booted with nmi_interrupt=1. (2.6.4-52 Suse
kernel.)
I did the same thing (infinite loop module) on a machine with an
Athlon XP 2200+ booting a gfs-patched 2.6.8.1 kernel with
nmi_watchdog=1, and I got a nice oops only a few seconds after loading
my misnamed "deadlock" module. The stack trace wasn't very helpful,
but the EIP was in deadlock_init, which is where the infinite loop
is. Neat!
--
Ed L Cashin <ecashin@xxxxxxxxxx>
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
