Re: NMI watchdog question

[Robert Hancock <hancockr@shaw.ca>]

Pallai Roland wrote:
> Hi,
>  I'm playing with the NMI watchdog (nmi_watchdog=1) on a reproductable
> hard lockup (no keyboard, etc) but seems like it doesn't works and I
> can't understand why, please explain to me the possible causes.. I
> belive it should work in this situation..
> 
> environment:
>  P4C800 motherboard, P4-2.4 cpu (APIC 2.0 on)
>  Promise 20378 SATA controller on the motherboard (sil_promise driver)
>  Maxtor diamondmax plus 9 200G sata disk
>  (and an empty PCI expander plus some more other under-testing hardware
> which doesn't matter in the experiment)
> 
>  mainline kernel 2.6.11
>  serial and VGA console, root on NFS
> 
> 
> steps to the lockup:
>  1. booting the machine with sata drive on the promise controller
>  2. dd if=/dev/sda of=/dev/null bs=4k
>  3. unplug the power from drive
>  4. waiting about 2 seconds
>  5. plug the power back
> 
>  dd stucked in 'D' here for 10-15 seconds and than the kernel say:
>   ata1: command timeout
> 
>  and voila, the box is dead, but without any message from the NMI
> watchdog :(

The NMI watchdog only triggers if something is blocking interrupts from 
getting through - if timer interrupts are still happening it won't 
activate. You can try Alt-Sysrq-T to get a traceback of where the 
current process is stuck..